xref: /openbmc/linux/net/smc/smc_pnet.c (revision 6726d552)
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 
32 static struct net_device *__pnet_find_base_ndev(struct net_device *ndev);
33 static struct net_device *pnet_find_base_ndev(struct net_device *ndev);
34 
35 static 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 
51 static struct genl_family smc_pnet_nl_family;
52 
53 enum smc_pnet_nametype {
54 	SMC_PNET_ETH	= 1,
55 	SMC_PNET_IB	= 2,
56 };
57 
58 /* pnet entry stored in pnet table */
59 struct 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 */
77 bool 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 */
85 static 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  */
101 static 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_dev;
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_dev, &smcd_dev_list.list, list) {
166 		if (smcd_dev->pnetid_by_user &&
167 		    (!pnet_name ||
168 		     smc_pnet_match(pnet_name, smcd_dev->pnetid))) {
169 			pr_warn_ratelimited("smc: smcd device %s "
170 					    "erased user defined pnetid "
171 					    "%.16s\n", dev_name(&smcd_dev->dev),
172 					    smcd_dev->pnetid);
173 			memset(smcd_dev->pnetid, 0, SMC_MAX_PNETID_LEN);
174 			smcd_dev->pnetid_by_user = false;
175 			rc = 0;
176 		}
177 	}
178 	mutex_unlock(&smcd_dev_list.mutex);
179 	return rc;
180 }
181 
182 /* Add the reference to a given network device to the pnet table.
183  */
184 static int smc_pnet_add_by_ndev(struct net_device *ndev)
185 {
186 	struct smc_pnetentry *pnetelem, *tmp_pe;
187 	struct smc_pnettable *pnettable;
188 	struct net *net = dev_net(ndev);
189 	struct smc_net *sn;
190 	int rc = -ENOENT;
191 
192 	/* get pnettable for namespace */
193 	sn = net_generic(net, smc_net_id);
194 	pnettable = &sn->pnettable;
195 
196 	mutex_lock(&pnettable->lock);
197 	list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, list) {
198 		if (pnetelem->type == SMC_PNET_ETH && !pnetelem->ndev &&
199 		    !strncmp(pnetelem->eth_name, ndev->name, IFNAMSIZ)) {
200 			netdev_hold(ndev, &pnetelem->dev_tracker, GFP_ATOMIC);
201 			pnetelem->ndev = ndev;
202 			rc = 0;
203 			pr_warn_ratelimited("smc: adding net device %s with "
204 					    "user defined pnetid %.16s\n",
205 					    pnetelem->eth_name,
206 					    pnetelem->pnet_name);
207 			break;
208 		}
209 	}
210 	mutex_unlock(&pnettable->lock);
211 	return rc;
212 }
213 
214 /* Remove the reference to a given network device from the pnet table.
215  */
216 static int smc_pnet_remove_by_ndev(struct net_device *ndev)
217 {
218 	struct smc_pnetentry *pnetelem, *tmp_pe;
219 	struct smc_pnettable *pnettable;
220 	struct net *net = dev_net(ndev);
221 	struct smc_net *sn;
222 	int rc = -ENOENT;
223 
224 	/* get pnettable for namespace */
225 	sn = net_generic(net, smc_net_id);
226 	pnettable = &sn->pnettable;
227 
228 	mutex_lock(&pnettable->lock);
229 	list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, list) {
230 		if (pnetelem->type == SMC_PNET_ETH && pnetelem->ndev == ndev) {
231 			netdev_put(pnetelem->ndev, &pnetelem->dev_tracker);
232 			pnetelem->ndev = NULL;
233 			rc = 0;
234 			pr_warn_ratelimited("smc: removing net device %s with "
235 					    "user defined pnetid %.16s\n",
236 					    pnetelem->eth_name,
237 					    pnetelem->pnet_name);
238 			break;
239 		}
240 	}
241 	mutex_unlock(&pnettable->lock);
242 	return rc;
243 }
244 
245 /* Apply pnetid to ib device when no pnetid is set.
246  */
247 static bool smc_pnet_apply_ib(struct smc_ib_device *ib_dev, u8 ib_port,
248 			      char *pnet_name)
249 {
250 	bool applied = false;
251 
252 	mutex_lock(&smc_ib_devices.mutex);
253 	if (!smc_pnet_is_pnetid_set(ib_dev->pnetid[ib_port - 1])) {
254 		memcpy(ib_dev->pnetid[ib_port - 1], pnet_name,
255 		       SMC_MAX_PNETID_LEN);
256 		ib_dev->pnetid_by_user[ib_port - 1] = true;
257 		applied = true;
258 	}
259 	mutex_unlock(&smc_ib_devices.mutex);
260 	return applied;
261 }
262 
263 /* Apply pnetid to smcd device when no pnetid is set.
264  */
265 static bool smc_pnet_apply_smcd(struct smcd_dev *smcd_dev, char *pnet_name)
266 {
267 	bool applied = false;
268 
269 	mutex_lock(&smcd_dev_list.mutex);
270 	if (!smc_pnet_is_pnetid_set(smcd_dev->pnetid)) {
271 		memcpy(smcd_dev->pnetid, pnet_name, SMC_MAX_PNETID_LEN);
272 		smcd_dev->pnetid_by_user = true;
273 		applied = true;
274 	}
275 	mutex_unlock(&smcd_dev_list.mutex);
276 	return applied;
277 }
278 
279 /* The limit for pnetid is 16 characters.
280  * Valid characters should be (single-byte character set) a-z, A-Z, 0-9.
281  * Lower case letters are converted to upper case.
282  * Interior blanks should not be used.
283  */
284 static bool smc_pnetid_valid(const char *pnet_name, char *pnetid)
285 {
286 	char *bf = skip_spaces(pnet_name);
287 	size_t len = strlen(bf);
288 	char *end = bf + len;
289 
290 	if (!len)
291 		return false;
292 	while (--end >= bf && isspace(*end))
293 		;
294 	if (end - bf >= SMC_MAX_PNETID_LEN)
295 		return false;
296 	while (bf <= end) {
297 		if (!isalnum(*bf))
298 			return false;
299 		*pnetid++ = islower(*bf) ? toupper(*bf) : *bf;
300 		bf++;
301 	}
302 	*pnetid = '\0';
303 	return true;
304 }
305 
306 /* Find an infiniband device by a given name. The device might not exist. */
307 static struct smc_ib_device *smc_pnet_find_ib(char *ib_name)
308 {
309 	struct smc_ib_device *ibdev;
310 
311 	mutex_lock(&smc_ib_devices.mutex);
312 	list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
313 		if (!strncmp(ibdev->ibdev->name, ib_name,
314 			     sizeof(ibdev->ibdev->name)) ||
315 		    (ibdev->ibdev->dev.parent &&
316 		     !strncmp(dev_name(ibdev->ibdev->dev.parent), ib_name,
317 			     IB_DEVICE_NAME_MAX - 1))) {
318 			goto out;
319 		}
320 	}
321 	ibdev = NULL;
322 out:
323 	mutex_unlock(&smc_ib_devices.mutex);
324 	return ibdev;
325 }
326 
327 /* Find an smcd device by a given name. The device might not exist. */
328 static struct smcd_dev *smc_pnet_find_smcd(char *smcd_name)
329 {
330 	struct smcd_dev *smcd_dev;
331 
332 	mutex_lock(&smcd_dev_list.mutex);
333 	list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) {
334 		if (!strncmp(dev_name(&smcd_dev->dev), smcd_name,
335 			     IB_DEVICE_NAME_MAX - 1))
336 			goto out;
337 	}
338 	smcd_dev = NULL;
339 out:
340 	mutex_unlock(&smcd_dev_list.mutex);
341 	return smcd_dev;
342 }
343 
344 static int smc_pnet_add_eth(struct smc_pnettable *pnettable, struct net *net,
345 			    char *eth_name, char *pnet_name)
346 {
347 	struct smc_pnetentry *tmp_pe, *new_pe;
348 	struct net_device *ndev, *base_ndev;
349 	u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
350 	bool new_netdev;
351 	int rc;
352 
353 	/* check if (base) netdev already has a pnetid. If there is one, we do
354 	 * not want to add a pnet table entry
355 	 */
356 	rc = -EEXIST;
357 	ndev = dev_get_by_name(net, eth_name);	/* dev_hold() */
358 	if (ndev) {
359 		base_ndev = pnet_find_base_ndev(ndev);
360 		if (!smc_pnetid_by_dev_port(base_ndev->dev.parent,
361 					    base_ndev->dev_port, ndev_pnetid))
362 			goto out_put;
363 	}
364 
365 	/* add a new netdev entry to the pnet table if there isn't one */
366 	rc = -ENOMEM;
367 	new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL);
368 	if (!new_pe)
369 		goto out_put;
370 	new_pe->type = SMC_PNET_ETH;
371 	memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN);
372 	strncpy(new_pe->eth_name, eth_name, IFNAMSIZ);
373 	rc = -EEXIST;
374 	new_netdev = true;
375 	mutex_lock(&pnettable->lock);
376 	list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
377 		if (tmp_pe->type == SMC_PNET_ETH &&
378 		    !strncmp(tmp_pe->eth_name, eth_name, IFNAMSIZ)) {
379 			new_netdev = false;
380 			break;
381 		}
382 	}
383 	if (new_netdev) {
384 		if (ndev) {
385 			new_pe->ndev = ndev;
386 			netdev_tracker_alloc(ndev, &new_pe->dev_tracker,
387 					     GFP_ATOMIC);
388 		}
389 		list_add_tail(&new_pe->list, &pnettable->pnetlist);
390 		mutex_unlock(&pnettable->lock);
391 	} else {
392 		mutex_unlock(&pnettable->lock);
393 		kfree(new_pe);
394 		goto out_put;
395 	}
396 	if (ndev)
397 		pr_warn_ratelimited("smc: net device %s "
398 				    "applied user defined pnetid %.16s\n",
399 				    new_pe->eth_name, new_pe->pnet_name);
400 	return 0;
401 
402 out_put:
403 	dev_put(ndev);
404 	return rc;
405 }
406 
407 static int smc_pnet_add_ib(struct smc_pnettable *pnettable, char *ib_name,
408 			   u8 ib_port, char *pnet_name)
409 {
410 	struct smc_pnetentry *tmp_pe, *new_pe;
411 	struct smc_ib_device *ib_dev;
412 	bool smcddev_applied = true;
413 	bool ibdev_applied = true;
414 	struct smcd_dev *smcd_dev;
415 	bool new_ibdev;
416 
417 	/* try to apply the pnetid to active devices */
418 	ib_dev = smc_pnet_find_ib(ib_name);
419 	if (ib_dev) {
420 		ibdev_applied = smc_pnet_apply_ib(ib_dev, ib_port, pnet_name);
421 		if (ibdev_applied)
422 			pr_warn_ratelimited("smc: ib device %s ibport %d "
423 					    "applied user defined pnetid "
424 					    "%.16s\n", ib_dev->ibdev->name,
425 					    ib_port,
426 					    ib_dev->pnetid[ib_port - 1]);
427 	}
428 	smcd_dev = smc_pnet_find_smcd(ib_name);
429 	if (smcd_dev) {
430 		smcddev_applied = smc_pnet_apply_smcd(smcd_dev, pnet_name);
431 		if (smcddev_applied)
432 			pr_warn_ratelimited("smc: smcd device %s "
433 					    "applied user defined pnetid "
434 					    "%.16s\n", dev_name(&smcd_dev->dev),
435 					    smcd_dev->pnetid);
436 	}
437 	/* Apply fails when a device has a hardware-defined pnetid set, do not
438 	 * add a pnet table entry in that case.
439 	 */
440 	if (!ibdev_applied || !smcddev_applied)
441 		return -EEXIST;
442 
443 	/* add a new ib entry to the pnet table if there isn't one */
444 	new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL);
445 	if (!new_pe)
446 		return -ENOMEM;
447 	new_pe->type = SMC_PNET_IB;
448 	memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN);
449 	strncpy(new_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX);
450 	new_pe->ib_port = ib_port;
451 
452 	new_ibdev = true;
453 	mutex_lock(&pnettable->lock);
454 	list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
455 		if (tmp_pe->type == SMC_PNET_IB &&
456 		    !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) {
457 			new_ibdev = false;
458 			break;
459 		}
460 	}
461 	if (new_ibdev) {
462 		list_add_tail(&new_pe->list, &pnettable->pnetlist);
463 		mutex_unlock(&pnettable->lock);
464 	} else {
465 		mutex_unlock(&pnettable->lock);
466 		kfree(new_pe);
467 	}
468 	return (new_ibdev) ? 0 : -EEXIST;
469 }
470 
471 /* Append a pnetid to the end of the pnet table if not already on this list.
472  */
473 static int smc_pnet_enter(struct net *net, struct nlattr *tb[])
474 {
475 	char pnet_name[SMC_MAX_PNETID_LEN + 1];
476 	struct smc_pnettable *pnettable;
477 	bool new_netdev = false;
478 	bool new_ibdev = false;
479 	struct smc_net *sn;
480 	u8 ibport = 1;
481 	char *string;
482 	int rc;
483 
484 	/* get pnettable for namespace */
485 	sn = net_generic(net, smc_net_id);
486 	pnettable = &sn->pnettable;
487 
488 	rc = -EINVAL;
489 	if (!tb[SMC_PNETID_NAME])
490 		goto error;
491 	string = (char *)nla_data(tb[SMC_PNETID_NAME]);
492 	if (!smc_pnetid_valid(string, pnet_name))
493 		goto error;
494 
495 	if (tb[SMC_PNETID_ETHNAME]) {
496 		string = (char *)nla_data(tb[SMC_PNETID_ETHNAME]);
497 		rc = smc_pnet_add_eth(pnettable, net, string, pnet_name);
498 		if (!rc)
499 			new_netdev = true;
500 		else if (rc != -EEXIST)
501 			goto error;
502 	}
503 
504 	/* if this is not the initial namespace, stop here */
505 	if (net != &init_net)
506 		return new_netdev ? 0 : -EEXIST;
507 
508 	rc = -EINVAL;
509 	if (tb[SMC_PNETID_IBNAME]) {
510 		string = (char *)nla_data(tb[SMC_PNETID_IBNAME]);
511 		string = strim(string);
512 		if (tb[SMC_PNETID_IBPORT]) {
513 			ibport = nla_get_u8(tb[SMC_PNETID_IBPORT]);
514 			if (ibport < 1 || ibport > SMC_MAX_PORTS)
515 				goto error;
516 		}
517 		rc = smc_pnet_add_ib(pnettable, string, ibport, pnet_name);
518 		if (!rc)
519 			new_ibdev = true;
520 		else if (rc != -EEXIST)
521 			goto error;
522 	}
523 	return (new_netdev || new_ibdev) ? 0 : -EEXIST;
524 
525 error:
526 	return rc;
527 }
528 
529 /* Convert an smc_pnetentry to a netlink attribute sequence */
530 static int smc_pnet_set_nla(struct sk_buff *msg,
531 			    struct smc_pnetentry *pnetelem)
532 {
533 	if (nla_put_string(msg, SMC_PNETID_NAME, pnetelem->pnet_name))
534 		return -1;
535 	if (pnetelem->type == SMC_PNET_ETH) {
536 		if (nla_put_string(msg, SMC_PNETID_ETHNAME,
537 				   pnetelem->eth_name))
538 			return -1;
539 	} else {
540 		if (nla_put_string(msg, SMC_PNETID_ETHNAME, "n/a"))
541 			return -1;
542 	}
543 	if (pnetelem->type == SMC_PNET_IB) {
544 		if (nla_put_string(msg, SMC_PNETID_IBNAME, pnetelem->ib_name) ||
545 		    nla_put_u8(msg, SMC_PNETID_IBPORT, pnetelem->ib_port))
546 			return -1;
547 	} else {
548 		if (nla_put_string(msg, SMC_PNETID_IBNAME, "n/a") ||
549 		    nla_put_u8(msg, SMC_PNETID_IBPORT, 0xff))
550 			return -1;
551 	}
552 
553 	return 0;
554 }
555 
556 static int smc_pnet_add(struct sk_buff *skb, struct genl_info *info)
557 {
558 	struct net *net = genl_info_net(info);
559 
560 	return smc_pnet_enter(net, info->attrs);
561 }
562 
563 static int smc_pnet_del(struct sk_buff *skb, struct genl_info *info)
564 {
565 	struct net *net = genl_info_net(info);
566 
567 	if (!info->attrs[SMC_PNETID_NAME])
568 		return -EINVAL;
569 	return smc_pnet_remove_by_pnetid(net,
570 				(char *)nla_data(info->attrs[SMC_PNETID_NAME]));
571 }
572 
573 static int smc_pnet_dump_start(struct netlink_callback *cb)
574 {
575 	cb->args[0] = 0;
576 	return 0;
577 }
578 
579 static int smc_pnet_dumpinfo(struct sk_buff *skb,
580 			     u32 portid, u32 seq, u32 flags,
581 			     struct smc_pnetentry *pnetelem)
582 {
583 	void *hdr;
584 
585 	hdr = genlmsg_put(skb, portid, seq, &smc_pnet_nl_family,
586 			  flags, SMC_PNETID_GET);
587 	if (!hdr)
588 		return -ENOMEM;
589 	if (smc_pnet_set_nla(skb, pnetelem) < 0) {
590 		genlmsg_cancel(skb, hdr);
591 		return -EMSGSIZE;
592 	}
593 	genlmsg_end(skb, hdr);
594 	return 0;
595 }
596 
597 static int _smc_pnet_dump(struct net *net, struct sk_buff *skb, u32 portid,
598 			  u32 seq, u8 *pnetid, int start_idx)
599 {
600 	struct smc_pnettable *pnettable;
601 	struct smc_pnetentry *pnetelem;
602 	struct smc_net *sn;
603 	int idx = 0;
604 
605 	/* get pnettable for namespace */
606 	sn = net_generic(net, smc_net_id);
607 	pnettable = &sn->pnettable;
608 
609 	/* dump pnettable entries */
610 	mutex_lock(&pnettable->lock);
611 	list_for_each_entry(pnetelem, &pnettable->pnetlist, list) {
612 		if (pnetid && !smc_pnet_match(pnetelem->pnet_name, pnetid))
613 			continue;
614 		if (idx++ < start_idx)
615 			continue;
616 		/* if this is not the initial namespace, dump only netdev */
617 		if (net != &init_net && pnetelem->type != SMC_PNET_ETH)
618 			continue;
619 		if (smc_pnet_dumpinfo(skb, portid, seq, NLM_F_MULTI,
620 				      pnetelem)) {
621 			--idx;
622 			break;
623 		}
624 	}
625 	mutex_unlock(&pnettable->lock);
626 	return idx;
627 }
628 
629 static int smc_pnet_dump(struct sk_buff *skb, struct netlink_callback *cb)
630 {
631 	struct net *net = sock_net(skb->sk);
632 	int idx;
633 
634 	idx = _smc_pnet_dump(net, skb, NETLINK_CB(cb->skb).portid,
635 			     cb->nlh->nlmsg_seq, NULL, cb->args[0]);
636 
637 	cb->args[0] = idx;
638 	return skb->len;
639 }
640 
641 /* Retrieve one PNETID entry */
642 static int smc_pnet_get(struct sk_buff *skb, struct genl_info *info)
643 {
644 	struct net *net = genl_info_net(info);
645 	struct sk_buff *msg;
646 	void *hdr;
647 
648 	if (!info->attrs[SMC_PNETID_NAME])
649 		return -EINVAL;
650 
651 	msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
652 	if (!msg)
653 		return -ENOMEM;
654 
655 	_smc_pnet_dump(net, msg, info->snd_portid, info->snd_seq,
656 		       nla_data(info->attrs[SMC_PNETID_NAME]), 0);
657 
658 	/* finish multi part message and send it */
659 	hdr = nlmsg_put(msg, info->snd_portid, info->snd_seq, NLMSG_DONE, 0,
660 			NLM_F_MULTI);
661 	if (!hdr) {
662 		nlmsg_free(msg);
663 		return -EMSGSIZE;
664 	}
665 	return genlmsg_reply(msg, info);
666 }
667 
668 /* Remove and delete all pnetids from pnet table.
669  */
670 static int smc_pnet_flush(struct sk_buff *skb, struct genl_info *info)
671 {
672 	struct net *net = genl_info_net(info);
673 
674 	smc_pnet_remove_by_pnetid(net, NULL);
675 	return 0;
676 }
677 
678 /* SMC_PNETID generic netlink operation definition */
679 static const struct genl_ops smc_pnet_ops[] = {
680 	{
681 		.cmd = SMC_PNETID_GET,
682 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
683 		/* can be retrieved by unprivileged users */
684 		.doit = smc_pnet_get,
685 		.dumpit = smc_pnet_dump,
686 		.start = smc_pnet_dump_start
687 	},
688 	{
689 		.cmd = SMC_PNETID_ADD,
690 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
691 		.flags = GENL_ADMIN_PERM,
692 		.doit = smc_pnet_add
693 	},
694 	{
695 		.cmd = SMC_PNETID_DEL,
696 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
697 		.flags = GENL_ADMIN_PERM,
698 		.doit = smc_pnet_del
699 	},
700 	{
701 		.cmd = SMC_PNETID_FLUSH,
702 		.validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
703 		.flags = GENL_ADMIN_PERM,
704 		.doit = smc_pnet_flush
705 	}
706 };
707 
708 /* SMC_PNETID family definition */
709 static struct genl_family smc_pnet_nl_family __ro_after_init = {
710 	.hdrsize = 0,
711 	.name = SMCR_GENL_FAMILY_NAME,
712 	.version = SMCR_GENL_FAMILY_VERSION,
713 	.maxattr = SMC_PNETID_MAX,
714 	.policy = smc_pnet_policy,
715 	.netnsok = true,
716 	.module = THIS_MODULE,
717 	.ops = smc_pnet_ops,
718 	.n_ops =  ARRAY_SIZE(smc_pnet_ops)
719 };
720 
721 bool smc_pnet_is_ndev_pnetid(struct net *net, u8 *pnetid)
722 {
723 	struct smc_net *sn = net_generic(net, smc_net_id);
724 	struct smc_pnetids_ndev_entry *pe;
725 	bool rc = false;
726 
727 	read_lock(&sn->pnetids_ndev.lock);
728 	list_for_each_entry(pe, &sn->pnetids_ndev.list, list) {
729 		if (smc_pnet_match(pnetid, pe->pnetid)) {
730 			rc = true;
731 			goto unlock;
732 		}
733 	}
734 
735 unlock:
736 	read_unlock(&sn->pnetids_ndev.lock);
737 	return rc;
738 }
739 
740 static int smc_pnet_add_pnetid(struct net *net, u8 *pnetid)
741 {
742 	struct smc_net *sn = net_generic(net, smc_net_id);
743 	struct smc_pnetids_ndev_entry *pe, *pi;
744 
745 	pe = kzalloc(sizeof(*pe), GFP_KERNEL);
746 	if (!pe)
747 		return -ENOMEM;
748 
749 	write_lock(&sn->pnetids_ndev.lock);
750 	list_for_each_entry(pi, &sn->pnetids_ndev.list, list) {
751 		if (smc_pnet_match(pnetid, pe->pnetid)) {
752 			refcount_inc(&pi->refcnt);
753 			kfree(pe);
754 			goto unlock;
755 		}
756 	}
757 	refcount_set(&pe->refcnt, 1);
758 	memcpy(pe->pnetid, pnetid, SMC_MAX_PNETID_LEN);
759 	list_add_tail(&pe->list, &sn->pnetids_ndev.list);
760 
761 unlock:
762 	write_unlock(&sn->pnetids_ndev.lock);
763 	return 0;
764 }
765 
766 static void smc_pnet_remove_pnetid(struct net *net, u8 *pnetid)
767 {
768 	struct smc_net *sn = net_generic(net, smc_net_id);
769 	struct smc_pnetids_ndev_entry *pe, *pe2;
770 
771 	write_lock(&sn->pnetids_ndev.lock);
772 	list_for_each_entry_safe(pe, pe2, &sn->pnetids_ndev.list, list) {
773 		if (smc_pnet_match(pnetid, pe->pnetid)) {
774 			if (refcount_dec_and_test(&pe->refcnt)) {
775 				list_del(&pe->list);
776 				kfree(pe);
777 			}
778 			break;
779 		}
780 	}
781 	write_unlock(&sn->pnetids_ndev.lock);
782 }
783 
784 static void smc_pnet_add_base_pnetid(struct net *net, struct net_device *dev,
785 				     u8 *ndev_pnetid)
786 {
787 	struct net_device *base_dev;
788 
789 	base_dev = __pnet_find_base_ndev(dev);
790 	if (base_dev->flags & IFF_UP &&
791 	    !smc_pnetid_by_dev_port(base_dev->dev.parent, base_dev->dev_port,
792 				    ndev_pnetid)) {
793 		/* add to PNETIDs list */
794 		smc_pnet_add_pnetid(net, ndev_pnetid);
795 	}
796 }
797 
798 /* create initial list of netdevice pnetids */
799 static void smc_pnet_create_pnetids_list(struct net *net)
800 {
801 	u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
802 	struct net_device *dev;
803 
804 	rtnl_lock();
805 	for_each_netdev(net, dev)
806 		smc_pnet_add_base_pnetid(net, dev, ndev_pnetid);
807 	rtnl_unlock();
808 }
809 
810 /* clean up list of netdevice pnetids */
811 static void smc_pnet_destroy_pnetids_list(struct net *net)
812 {
813 	struct smc_net *sn = net_generic(net, smc_net_id);
814 	struct smc_pnetids_ndev_entry *pe, *temp_pe;
815 
816 	write_lock(&sn->pnetids_ndev.lock);
817 	list_for_each_entry_safe(pe, temp_pe, &sn->pnetids_ndev.list, list) {
818 		list_del(&pe->list);
819 		kfree(pe);
820 	}
821 	write_unlock(&sn->pnetids_ndev.lock);
822 }
823 
824 static int smc_pnet_netdev_event(struct notifier_block *this,
825 				 unsigned long event, void *ptr)
826 {
827 	struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
828 	struct net *net = dev_net(event_dev);
829 	u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
830 
831 	switch (event) {
832 	case NETDEV_REBOOT:
833 	case NETDEV_UNREGISTER:
834 		smc_pnet_remove_by_ndev(event_dev);
835 		smc_ib_ndev_change(event_dev, event);
836 		return NOTIFY_OK;
837 	case NETDEV_REGISTER:
838 		smc_pnet_add_by_ndev(event_dev);
839 		smc_ib_ndev_change(event_dev, event);
840 		return NOTIFY_OK;
841 	case NETDEV_UP:
842 		smc_pnet_add_base_pnetid(net, event_dev, ndev_pnetid);
843 		return NOTIFY_OK;
844 	case NETDEV_DOWN:
845 		event_dev = __pnet_find_base_ndev(event_dev);
846 		if (!smc_pnetid_by_dev_port(event_dev->dev.parent,
847 					    event_dev->dev_port, ndev_pnetid)) {
848 			/* remove from PNETIDs list */
849 			smc_pnet_remove_pnetid(net, ndev_pnetid);
850 		}
851 		return NOTIFY_OK;
852 	default:
853 		return NOTIFY_DONE;
854 	}
855 }
856 
857 static struct notifier_block smc_netdev_notifier = {
858 	.notifier_call = smc_pnet_netdev_event
859 };
860 
861 /* init network namespace */
862 int smc_pnet_net_init(struct net *net)
863 {
864 	struct smc_net *sn = net_generic(net, smc_net_id);
865 	struct smc_pnettable *pnettable = &sn->pnettable;
866 	struct smc_pnetids_ndev *pnetids_ndev = &sn->pnetids_ndev;
867 
868 	INIT_LIST_HEAD(&pnettable->pnetlist);
869 	mutex_init(&pnettable->lock);
870 	INIT_LIST_HEAD(&pnetids_ndev->list);
871 	rwlock_init(&pnetids_ndev->lock);
872 
873 	smc_pnet_create_pnetids_list(net);
874 
875 	/* disable handshake limitation by default */
876 	net->smc.limit_smc_hs = 0;
877 
878 	return 0;
879 }
880 
881 int __init smc_pnet_init(void)
882 {
883 	int rc;
884 
885 	rc = genl_register_family(&smc_pnet_nl_family);
886 	if (rc)
887 		return rc;
888 	rc = register_netdevice_notifier(&smc_netdev_notifier);
889 	if (rc)
890 		genl_unregister_family(&smc_pnet_nl_family);
891 
892 	return rc;
893 }
894 
895 /* exit network namespace */
896 void smc_pnet_net_exit(struct net *net)
897 {
898 	/* flush pnet table */
899 	smc_pnet_remove_by_pnetid(net, NULL);
900 	smc_pnet_destroy_pnetids_list(net);
901 }
902 
903 void smc_pnet_exit(void)
904 {
905 	unregister_netdevice_notifier(&smc_netdev_notifier);
906 	genl_unregister_family(&smc_pnet_nl_family);
907 }
908 
909 static struct net_device *__pnet_find_base_ndev(struct net_device *ndev)
910 {
911 	int i, nest_lvl;
912 
913 	ASSERT_RTNL();
914 	nest_lvl = ndev->lower_level;
915 	for (i = 0; i < nest_lvl; i++) {
916 		struct list_head *lower = &ndev->adj_list.lower;
917 
918 		if (list_empty(lower))
919 			break;
920 		lower = lower->next;
921 		ndev = netdev_lower_get_next(ndev, &lower);
922 	}
923 	return ndev;
924 }
925 
926 /* Determine one base device for stacked net devices.
927  * If the lower device level contains more than one devices
928  * (for instance with bonding slaves), just the first device
929  * is used to reach a base device.
930  */
931 static struct net_device *pnet_find_base_ndev(struct net_device *ndev)
932 {
933 	rtnl_lock();
934 	ndev = __pnet_find_base_ndev(ndev);
935 	rtnl_unlock();
936 	return ndev;
937 }
938 
939 static int smc_pnet_find_ndev_pnetid_by_table(struct net_device *ndev,
940 					      u8 *pnetid)
941 {
942 	struct smc_pnettable *pnettable;
943 	struct net *net = dev_net(ndev);
944 	struct smc_pnetentry *pnetelem;
945 	struct smc_net *sn;
946 	int rc = -ENOENT;
947 
948 	/* get pnettable for namespace */
949 	sn = net_generic(net, smc_net_id);
950 	pnettable = &sn->pnettable;
951 
952 	mutex_lock(&pnettable->lock);
953 	list_for_each_entry(pnetelem, &pnettable->pnetlist, list) {
954 		if (pnetelem->type == SMC_PNET_ETH && ndev == pnetelem->ndev) {
955 			/* get pnetid of netdev device */
956 			memcpy(pnetid, pnetelem->pnet_name, SMC_MAX_PNETID_LEN);
957 			rc = 0;
958 			break;
959 		}
960 	}
961 	mutex_unlock(&pnettable->lock);
962 	return rc;
963 }
964 
965 static int smc_pnet_determine_gid(struct smc_ib_device *ibdev, int i,
966 				  struct smc_init_info *ini)
967 {
968 	if (!ini->check_smcrv2 &&
969 	    !smc_ib_determine_gid(ibdev, i, ini->vlan_id, ini->ib_gid, NULL,
970 				  NULL)) {
971 		ini->ib_dev = ibdev;
972 		ini->ib_port = i;
973 		return 0;
974 	}
975 	if (ini->check_smcrv2 &&
976 	    !smc_ib_determine_gid(ibdev, i, ini->vlan_id, ini->smcrv2.ib_gid_v2,
977 				  NULL, &ini->smcrv2)) {
978 		ini->smcrv2.ib_dev_v2 = ibdev;
979 		ini->smcrv2.ib_port_v2 = i;
980 		return 0;
981 	}
982 	return -ENODEV;
983 }
984 
985 /* find a roce device for the given pnetid */
986 static void _smc_pnet_find_roce_by_pnetid(u8 *pnet_id,
987 					  struct smc_init_info *ini,
988 					  struct smc_ib_device *known_dev,
989 					  struct net *net)
990 {
991 	struct smc_ib_device *ibdev;
992 	int i;
993 
994 	mutex_lock(&smc_ib_devices.mutex);
995 	list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
996 		if (ibdev == known_dev ||
997 		    !rdma_dev_access_netns(ibdev->ibdev, net))
998 			continue;
999 		for (i = 1; i <= SMC_MAX_PORTS; i++) {
1000 			if (!rdma_is_port_valid(ibdev->ibdev, i))
1001 				continue;
1002 			if (smc_pnet_match(ibdev->pnetid[i - 1], pnet_id) &&
1003 			    smc_ib_port_active(ibdev, i) &&
1004 			    !test_bit(i - 1, ibdev->ports_going_away)) {
1005 				if (!smc_pnet_determine_gid(ibdev, i, ini))
1006 					goto out;
1007 			}
1008 		}
1009 	}
1010 out:
1011 	mutex_unlock(&smc_ib_devices.mutex);
1012 }
1013 
1014 /* find alternate roce device with same pnet_id, vlan_id and net namespace */
1015 void smc_pnet_find_alt_roce(struct smc_link_group *lgr,
1016 			    struct smc_init_info *ini,
1017 			    struct smc_ib_device *known_dev)
1018 {
1019 	struct net *net = lgr->net;
1020 
1021 	_smc_pnet_find_roce_by_pnetid(lgr->pnet_id, ini, known_dev, net);
1022 }
1023 
1024 /* if handshake network device belongs to a roce device, return its
1025  * IB device and port
1026  */
1027 static void smc_pnet_find_rdma_dev(struct net_device *netdev,
1028 				   struct smc_init_info *ini)
1029 {
1030 	struct net *net = dev_net(netdev);
1031 	struct smc_ib_device *ibdev;
1032 
1033 	mutex_lock(&smc_ib_devices.mutex);
1034 	list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
1035 		struct net_device *ndev;
1036 		int i;
1037 
1038 		/* check rdma net namespace */
1039 		if (!rdma_dev_access_netns(ibdev->ibdev, net))
1040 			continue;
1041 
1042 		for (i = 1; i <= SMC_MAX_PORTS; i++) {
1043 			if (!rdma_is_port_valid(ibdev->ibdev, i))
1044 				continue;
1045 			if (!ibdev->ibdev->ops.get_netdev)
1046 				continue;
1047 			ndev = ibdev->ibdev->ops.get_netdev(ibdev->ibdev, i);
1048 			if (!ndev)
1049 				continue;
1050 			dev_put(ndev);
1051 			if (netdev == ndev &&
1052 			    smc_ib_port_active(ibdev, i) &&
1053 			    !test_bit(i - 1, ibdev->ports_going_away)) {
1054 				if (!smc_pnet_determine_gid(ibdev, i, ini))
1055 					break;
1056 			}
1057 		}
1058 	}
1059 	mutex_unlock(&smc_ib_devices.mutex);
1060 }
1061 
1062 /* Determine the corresponding IB device port based on the hardware PNETID.
1063  * Searching stops at the first matching active IB device port with vlan_id
1064  * configured.
1065  * If nothing found, check pnetid table.
1066  * If nothing found, try to use handshake device
1067  */
1068 static void smc_pnet_find_roce_by_pnetid(struct net_device *ndev,
1069 					 struct smc_init_info *ini)
1070 {
1071 	u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
1072 	struct net *net;
1073 
1074 	ndev = pnet_find_base_ndev(ndev);
1075 	net = dev_net(ndev);
1076 	if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port,
1077 				   ndev_pnetid) &&
1078 	    smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid)) {
1079 		smc_pnet_find_rdma_dev(ndev, ini);
1080 		return; /* pnetid could not be determined */
1081 	}
1082 	_smc_pnet_find_roce_by_pnetid(ndev_pnetid, ini, NULL, net);
1083 }
1084 
1085 static void smc_pnet_find_ism_by_pnetid(struct net_device *ndev,
1086 					struct smc_init_info *ini)
1087 {
1088 	u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
1089 	struct smcd_dev *ismdev;
1090 
1091 	ndev = pnet_find_base_ndev(ndev);
1092 	if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port,
1093 				   ndev_pnetid) &&
1094 	    smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid))
1095 		return; /* pnetid could not be determined */
1096 
1097 	mutex_lock(&smcd_dev_list.mutex);
1098 	list_for_each_entry(ismdev, &smcd_dev_list.list, list) {
1099 		if (smc_pnet_match(ismdev->pnetid, ndev_pnetid) &&
1100 		    !ismdev->going_away &&
1101 		    (!ini->ism_peer_gid[0] ||
1102 		     !smc_ism_cantalk(ini->ism_peer_gid[0], ini->vlan_id,
1103 				      ismdev))) {
1104 			ini->ism_dev[0] = ismdev;
1105 			break;
1106 		}
1107 	}
1108 	mutex_unlock(&smcd_dev_list.mutex);
1109 }
1110 
1111 /* PNET table analysis for a given sock:
1112  * determine ib_device and port belonging to used internal TCP socket
1113  * ethernet interface.
1114  */
1115 void smc_pnet_find_roce_resource(struct sock *sk, struct smc_init_info *ini)
1116 {
1117 	struct dst_entry *dst = sk_dst_get(sk);
1118 
1119 	if (!dst)
1120 		goto out;
1121 	if (!dst->dev)
1122 		goto out_rel;
1123 
1124 	smc_pnet_find_roce_by_pnetid(dst->dev, ini);
1125 
1126 out_rel:
1127 	dst_release(dst);
1128 out:
1129 	return;
1130 }
1131 
1132 void smc_pnet_find_ism_resource(struct sock *sk, struct smc_init_info *ini)
1133 {
1134 	struct dst_entry *dst = sk_dst_get(sk);
1135 
1136 	ini->ism_dev[0] = NULL;
1137 	if (!dst)
1138 		goto out;
1139 	if (!dst->dev)
1140 		goto out_rel;
1141 
1142 	smc_pnet_find_ism_by_pnetid(dst->dev, ini);
1143 
1144 out_rel:
1145 	dst_release(dst);
1146 out:
1147 	return;
1148 }
1149 
1150 /* Lookup and apply a pnet table entry to the given ib device.
1151  */
1152 int smc_pnetid_by_table_ib(struct smc_ib_device *smcibdev, u8 ib_port)
1153 {
1154 	char *ib_name = smcibdev->ibdev->name;
1155 	struct smc_pnettable *pnettable;
1156 	struct smc_pnetentry *tmp_pe;
1157 	struct smc_net *sn;
1158 	int rc = -ENOENT;
1159 
1160 	/* get pnettable for init namespace */
1161 	sn = net_generic(&init_net, smc_net_id);
1162 	pnettable = &sn->pnettable;
1163 
1164 	mutex_lock(&pnettable->lock);
1165 	list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
1166 		if (tmp_pe->type == SMC_PNET_IB &&
1167 		    !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX) &&
1168 		    tmp_pe->ib_port == ib_port) {
1169 			smc_pnet_apply_ib(smcibdev, ib_port, tmp_pe->pnet_name);
1170 			rc = 0;
1171 			break;
1172 		}
1173 	}
1174 	mutex_unlock(&pnettable->lock);
1175 
1176 	return rc;
1177 }
1178 
1179 /* Lookup and apply a pnet table entry to the given smcd device.
1180  */
1181 int smc_pnetid_by_table_smcd(struct smcd_dev *smcddev)
1182 {
1183 	const char *ib_name = dev_name(&smcddev->dev);
1184 	struct smc_pnettable *pnettable;
1185 	struct smc_pnetentry *tmp_pe;
1186 	struct smc_net *sn;
1187 	int rc = -ENOENT;
1188 
1189 	/* get pnettable for init namespace */
1190 	sn = net_generic(&init_net, smc_net_id);
1191 	pnettable = &sn->pnettable;
1192 
1193 	mutex_lock(&pnettable->lock);
1194 	list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
1195 		if (tmp_pe->type == SMC_PNET_IB &&
1196 		    !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) {
1197 			smc_pnet_apply_smcd(smcddev, tmp_pe->pnet_name);
1198 			rc = 0;
1199 			break;
1200 		}
1201 	}
1202 	mutex_unlock(&pnettable->lock);
1203 
1204 	return rc;
1205 }
1206