1 // SPDX-License-Identifier: GPL-2.0
3 * Shared Memory Communications over RDMA (SMC-R) and RoCE
5 * Generic netlink support functions to configure an SMC-R PNET table
7 * Copyright IBM Corp. 2016
9 * Author(s): Thomas Richter <tmricht@linux.vnet.ibm.com>
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>
19 #include <uapi/linux/if.h>
20 #include <uapi/linux/smc.h>
22 #include <rdma/ib_verbs.h>
24 #include <net/netns/generic.h>
25 #include "smc_netns.h"
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);
35 static const struct nla_policy smc_pnet_policy[SMC_PNETID_MAX + 1] = {
37 .type = NLA_NUL_STRING,
38 .len = SMC_MAX_PNETID_LEN
40 [SMC_PNETID_ETHNAME] = {
41 .type = NLA_NUL_STRING,
44 [SMC_PNETID_IBNAME] = {
45 .type = NLA_NUL_STRING,
46 .len = IB_DEVICE_NAME_MAX - 1
48 [SMC_PNETID_IBPORT] = { .type = NLA_U8 }
51 static struct genl_family smc_pnet_nl_family;
53 enum smc_pnet_nametype {
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;
65 char eth_name[IFNAMSIZ + 1];
66 struct net_device *ndev;
69 char ib_name[IB_DEVICE_NAME_MAX + 1];
75 /* Check if the pnetid is set */
76 bool smc_pnet_is_pnetid_set(u8 *pnetid)
78 if (pnetid[0] == 0 || pnetid[0] == _S)
83 /* Check if two given pnetids match */
84 static bool smc_pnet_match(u8 *pnetid1, u8 *pnetid2)
88 for (i = 0; i < SMC_MAX_PNETID_LEN; i++) {
89 if ((pnetid1[i] == 0 || pnetid1[i] == _S) &&
90 (pnetid2[i] == 0 || pnetid2[i] == _S))
92 if (pnetid1[i] != pnetid2[i])
98 /* Remove a pnetid from the pnet table.
100 static int smc_pnet_remove_by_pnetid(struct net *net, char *pnet_name)
102 struct smc_pnetentry *pnetelem, *tmp_pe;
103 struct smc_pnettable *pnettable;
104 struct smc_ib_device *ibdev;
105 struct smcd_dev *smcd_dev;
110 /* get pnettable for namespace */
111 sn = net_generic(net, smc_net_id);
112 pnettable = &sn->pnettable;
114 /* remove table entry */
115 mutex_lock(&pnettable->lock);
116 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist,
119 smc_pnet_match(pnetelem->pnet_name, pnet_name)) {
120 list_del(&pnetelem->list);
121 if (pnetelem->type == SMC_PNET_ETH && pnetelem->ndev) {
122 dev_put(pnetelem->ndev);
123 pr_warn_ratelimited("smc: net device %s "
124 "erased user defined "
127 pnetelem->pnet_name);
133 mutex_unlock(&pnettable->lock);
135 /* if this is not the initial namespace, stop here */
136 if (net != &init_net)
139 /* remove ib devices */
140 mutex_lock(&smc_ib_devices.mutex);
141 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
142 for (ibport = 0; ibport < SMC_MAX_PORTS; ibport++) {
143 if (ibdev->pnetid_by_user[ibport] &&
145 smc_pnet_match(pnet_name,
146 ibdev->pnetid[ibport]))) {
147 pr_warn_ratelimited("smc: ib device %s ibport "
148 "%d erased user defined "
152 ibdev->pnetid[ibport]);
153 memset(ibdev->pnetid[ibport], 0,
155 ibdev->pnetid_by_user[ibport] = false;
160 mutex_unlock(&smc_ib_devices.mutex);
161 /* remove smcd devices */
162 mutex_lock(&smcd_dev_list.mutex);
163 list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) {
164 if (smcd_dev->pnetid_by_user &&
166 smc_pnet_match(pnet_name, smcd_dev->pnetid))) {
167 pr_warn_ratelimited("smc: smcd device %s "
168 "erased user defined pnetid "
169 "%.16s\n", dev_name(&smcd_dev->dev),
171 memset(smcd_dev->pnetid, 0, SMC_MAX_PNETID_LEN);
172 smcd_dev->pnetid_by_user = false;
176 mutex_unlock(&smcd_dev_list.mutex);
180 /* Add the reference to a given network device to the pnet table.
182 static int smc_pnet_add_by_ndev(struct net_device *ndev)
184 struct smc_pnetentry *pnetelem, *tmp_pe;
185 struct smc_pnettable *pnettable;
186 struct net *net = dev_net(ndev);
190 /* get pnettable for namespace */
191 sn = net_generic(net, smc_net_id);
192 pnettable = &sn->pnettable;
194 mutex_lock(&pnettable->lock);
195 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, list) {
196 if (pnetelem->type == SMC_PNET_ETH && !pnetelem->ndev &&
197 !strncmp(pnetelem->eth_name, ndev->name, IFNAMSIZ)) {
199 pnetelem->ndev = ndev;
201 pr_warn_ratelimited("smc: adding net device %s with "
202 "user defined pnetid %.16s\n",
204 pnetelem->pnet_name);
208 mutex_unlock(&pnettable->lock);
212 /* Remove the reference to a given network device from the pnet table.
214 static int smc_pnet_remove_by_ndev(struct net_device *ndev)
216 struct smc_pnetentry *pnetelem, *tmp_pe;
217 struct smc_pnettable *pnettable;
218 struct net *net = dev_net(ndev);
222 /* get pnettable for namespace */
223 sn = net_generic(net, smc_net_id);
224 pnettable = &sn->pnettable;
226 mutex_lock(&pnettable->lock);
227 list_for_each_entry_safe(pnetelem, tmp_pe, &pnettable->pnetlist, list) {
228 if (pnetelem->type == SMC_PNET_ETH && pnetelem->ndev == ndev) {
229 dev_put(pnetelem->ndev);
230 pnetelem->ndev = NULL;
232 pr_warn_ratelimited("smc: removing net device %s with "
233 "user defined pnetid %.16s\n",
235 pnetelem->pnet_name);
239 mutex_unlock(&pnettable->lock);
243 /* Apply pnetid to ib device when no pnetid is set.
245 static bool smc_pnet_apply_ib(struct smc_ib_device *ib_dev, u8 ib_port,
248 bool applied = false;
250 mutex_lock(&smc_ib_devices.mutex);
251 if (!smc_pnet_is_pnetid_set(ib_dev->pnetid[ib_port - 1])) {
252 memcpy(ib_dev->pnetid[ib_port - 1], pnet_name,
254 ib_dev->pnetid_by_user[ib_port - 1] = true;
257 mutex_unlock(&smc_ib_devices.mutex);
261 /* Apply pnetid to smcd device when no pnetid is set.
263 static bool smc_pnet_apply_smcd(struct smcd_dev *smcd_dev, char *pnet_name)
265 bool applied = false;
267 mutex_lock(&smcd_dev_list.mutex);
268 if (!smc_pnet_is_pnetid_set(smcd_dev->pnetid)) {
269 memcpy(smcd_dev->pnetid, pnet_name, SMC_MAX_PNETID_LEN);
270 smcd_dev->pnetid_by_user = true;
273 mutex_unlock(&smcd_dev_list.mutex);
277 /* The limit for pnetid is 16 characters.
278 * Valid characters should be (single-byte character set) a-z, A-Z, 0-9.
279 * Lower case letters are converted to upper case.
280 * Interior blanks should not be used.
282 static bool smc_pnetid_valid(const char *pnet_name, char *pnetid)
284 char *bf = skip_spaces(pnet_name);
285 size_t len = strlen(bf);
286 char *end = bf + len;
290 while (--end >= bf && isspace(*end))
292 if (end - bf >= SMC_MAX_PNETID_LEN)
297 *pnetid++ = islower(*bf) ? toupper(*bf) : *bf;
304 /* Find an infiniband device by a given name. The device might not exist. */
305 static struct smc_ib_device *smc_pnet_find_ib(char *ib_name)
307 struct smc_ib_device *ibdev;
309 mutex_lock(&smc_ib_devices.mutex);
310 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
311 if (!strncmp(ibdev->ibdev->name, ib_name,
312 sizeof(ibdev->ibdev->name)) ||
313 (ibdev->ibdev->dev.parent &&
314 !strncmp(dev_name(ibdev->ibdev->dev.parent), ib_name,
315 IB_DEVICE_NAME_MAX - 1))) {
321 mutex_unlock(&smc_ib_devices.mutex);
325 /* Find an smcd device by a given name. The device might not exist. */
326 static struct smcd_dev *smc_pnet_find_smcd(char *smcd_name)
328 struct smcd_dev *smcd_dev;
330 mutex_lock(&smcd_dev_list.mutex);
331 list_for_each_entry(smcd_dev, &smcd_dev_list.list, list) {
332 if (!strncmp(dev_name(&smcd_dev->dev), smcd_name,
333 IB_DEVICE_NAME_MAX - 1))
338 mutex_unlock(&smcd_dev_list.mutex);
342 static int smc_pnet_add_eth(struct smc_pnettable *pnettable, struct net *net,
343 char *eth_name, char *pnet_name)
345 struct smc_pnetentry *tmp_pe, *new_pe;
346 struct net_device *ndev, *base_ndev;
347 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
351 /* check if (base) netdev already has a pnetid. If there is one, we do
352 * not want to add a pnet table entry
355 ndev = dev_get_by_name(net, eth_name); /* dev_hold() */
357 base_ndev = pnet_find_base_ndev(ndev);
358 if (!smc_pnetid_by_dev_port(base_ndev->dev.parent,
359 base_ndev->dev_port, ndev_pnetid))
363 /* add a new netdev entry to the pnet table if there isn't one */
365 new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL);
368 new_pe->type = SMC_PNET_ETH;
369 memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN);
370 strncpy(new_pe->eth_name, eth_name, IFNAMSIZ);
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)) {
384 list_add_tail(&new_pe->list, &pnettable->pnetlist);
385 mutex_unlock(&pnettable->lock);
387 mutex_unlock(&pnettable->lock);
392 pr_warn_ratelimited("smc: net device %s "
393 "applied user defined pnetid %.16s\n",
394 new_pe->eth_name, new_pe->pnet_name);
403 static int smc_pnet_add_ib(struct smc_pnettable *pnettable, char *ib_name,
404 u8 ib_port, char *pnet_name)
406 struct smc_pnetentry *tmp_pe, *new_pe;
407 struct smc_ib_device *ib_dev;
408 bool smcddev_applied = true;
409 bool ibdev_applied = true;
410 struct smcd_dev *smcd_dev;
413 /* try to apply the pnetid to active devices */
414 ib_dev = smc_pnet_find_ib(ib_name);
416 ibdev_applied = smc_pnet_apply_ib(ib_dev, ib_port, pnet_name);
418 pr_warn_ratelimited("smc: ib device %s ibport %d "
419 "applied user defined pnetid "
420 "%.16s\n", ib_dev->ibdev->name,
422 ib_dev->pnetid[ib_port - 1]);
424 smcd_dev = smc_pnet_find_smcd(ib_name);
426 smcddev_applied = smc_pnet_apply_smcd(smcd_dev, pnet_name);
428 pr_warn_ratelimited("smc: smcd device %s "
429 "applied user defined pnetid "
430 "%.16s\n", dev_name(&smcd_dev->dev),
433 /* Apply fails when a device has a hardware-defined pnetid set, do not
434 * add a pnet table entry in that case.
436 if (!ibdev_applied || !smcddev_applied)
439 /* add a new ib entry to the pnet table if there isn't one */
440 new_pe = kzalloc(sizeof(*new_pe), GFP_KERNEL);
443 new_pe->type = SMC_PNET_IB;
444 memcpy(new_pe->pnet_name, pnet_name, SMC_MAX_PNETID_LEN);
445 strncpy(new_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX);
446 new_pe->ib_port = ib_port;
449 mutex_lock(&pnettable->lock);
450 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
451 if (tmp_pe->type == SMC_PNET_IB &&
452 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) {
458 list_add_tail(&new_pe->list, &pnettable->pnetlist);
459 mutex_unlock(&pnettable->lock);
461 mutex_unlock(&pnettable->lock);
464 return (new_ibdev) ? 0 : -EEXIST;
467 /* Append a pnetid to the end of the pnet table if not already on this list.
469 static int smc_pnet_enter(struct net *net, struct nlattr *tb[])
471 char pnet_name[SMC_MAX_PNETID_LEN + 1];
472 struct smc_pnettable *pnettable;
473 bool new_netdev = false;
474 bool new_ibdev = false;
480 /* get pnettable for namespace */
481 sn = net_generic(net, smc_net_id);
482 pnettable = &sn->pnettable;
485 if (!tb[SMC_PNETID_NAME])
487 string = (char *)nla_data(tb[SMC_PNETID_NAME]);
488 if (!smc_pnetid_valid(string, pnet_name))
491 if (tb[SMC_PNETID_ETHNAME]) {
492 string = (char *)nla_data(tb[SMC_PNETID_ETHNAME]);
493 rc = smc_pnet_add_eth(pnettable, net, string, pnet_name);
496 else if (rc != -EEXIST)
500 /* if this is not the initial namespace, stop here */
501 if (net != &init_net)
502 return new_netdev ? 0 : -EEXIST;
505 if (tb[SMC_PNETID_IBNAME]) {
506 string = (char *)nla_data(tb[SMC_PNETID_IBNAME]);
507 string = strim(string);
508 if (tb[SMC_PNETID_IBPORT]) {
509 ibport = nla_get_u8(tb[SMC_PNETID_IBPORT]);
510 if (ibport < 1 || ibport > SMC_MAX_PORTS)
513 rc = smc_pnet_add_ib(pnettable, string, ibport, pnet_name);
516 else if (rc != -EEXIST)
519 return (new_netdev || new_ibdev) ? 0 : -EEXIST;
525 /* Convert an smc_pnetentry to a netlink attribute sequence */
526 static int smc_pnet_set_nla(struct sk_buff *msg,
527 struct smc_pnetentry *pnetelem)
529 if (nla_put_string(msg, SMC_PNETID_NAME, pnetelem->pnet_name))
531 if (pnetelem->type == SMC_PNET_ETH) {
532 if (nla_put_string(msg, SMC_PNETID_ETHNAME,
536 if (nla_put_string(msg, SMC_PNETID_ETHNAME, "n/a"))
539 if (pnetelem->type == SMC_PNET_IB) {
540 if (nla_put_string(msg, SMC_PNETID_IBNAME, pnetelem->ib_name) ||
541 nla_put_u8(msg, SMC_PNETID_IBPORT, pnetelem->ib_port))
544 if (nla_put_string(msg, SMC_PNETID_IBNAME, "n/a") ||
545 nla_put_u8(msg, SMC_PNETID_IBPORT, 0xff))
552 static int smc_pnet_add(struct sk_buff *skb, struct genl_info *info)
554 struct net *net = genl_info_net(info);
556 return smc_pnet_enter(net, info->attrs);
559 static int smc_pnet_del(struct sk_buff *skb, struct genl_info *info)
561 struct net *net = genl_info_net(info);
563 if (!info->attrs[SMC_PNETID_NAME])
565 return smc_pnet_remove_by_pnetid(net,
566 (char *)nla_data(info->attrs[SMC_PNETID_NAME]));
569 static int smc_pnet_dump_start(struct netlink_callback *cb)
575 static int smc_pnet_dumpinfo(struct sk_buff *skb,
576 u32 portid, u32 seq, u32 flags,
577 struct smc_pnetentry *pnetelem)
581 hdr = genlmsg_put(skb, portid, seq, &smc_pnet_nl_family,
582 flags, SMC_PNETID_GET);
585 if (smc_pnet_set_nla(skb, pnetelem) < 0) {
586 genlmsg_cancel(skb, hdr);
589 genlmsg_end(skb, hdr);
593 static int _smc_pnet_dump(struct net *net, struct sk_buff *skb, u32 portid,
594 u32 seq, u8 *pnetid, int start_idx)
596 struct smc_pnettable *pnettable;
597 struct smc_pnetentry *pnetelem;
601 /* get pnettable for namespace */
602 sn = net_generic(net, smc_net_id);
603 pnettable = &sn->pnettable;
605 /* dump pnettable entries */
606 mutex_lock(&pnettable->lock);
607 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) {
608 if (pnetid && !smc_pnet_match(pnetelem->pnet_name, pnetid))
610 if (idx++ < start_idx)
612 /* if this is not the initial namespace, dump only netdev */
613 if (net != &init_net && pnetelem->type != SMC_PNET_ETH)
615 if (smc_pnet_dumpinfo(skb, portid, seq, NLM_F_MULTI,
621 mutex_unlock(&pnettable->lock);
625 static int smc_pnet_dump(struct sk_buff *skb, struct netlink_callback *cb)
627 struct net *net = sock_net(skb->sk);
630 idx = _smc_pnet_dump(net, skb, NETLINK_CB(cb->skb).portid,
631 cb->nlh->nlmsg_seq, NULL, cb->args[0]);
637 /* Retrieve one PNETID entry */
638 static int smc_pnet_get(struct sk_buff *skb, struct genl_info *info)
640 struct net *net = genl_info_net(info);
644 if (!info->attrs[SMC_PNETID_NAME])
647 msg = nlmsg_new(NLMSG_DEFAULT_SIZE, GFP_KERNEL);
651 _smc_pnet_dump(net, msg, info->snd_portid, info->snd_seq,
652 nla_data(info->attrs[SMC_PNETID_NAME]), 0);
654 /* finish multi part message and send it */
655 hdr = nlmsg_put(msg, info->snd_portid, info->snd_seq, NLMSG_DONE, 0,
661 return genlmsg_reply(msg, info);
664 /* Remove and delete all pnetids from pnet table.
666 static int smc_pnet_flush(struct sk_buff *skb, struct genl_info *info)
668 struct net *net = genl_info_net(info);
670 smc_pnet_remove_by_pnetid(net, NULL);
674 /* SMC_PNETID generic netlink operation definition */
675 static const struct genl_ops smc_pnet_ops[] = {
677 .cmd = SMC_PNETID_GET,
678 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
679 /* can be retrieved by unprivileged users */
680 .doit = smc_pnet_get,
681 .dumpit = smc_pnet_dump,
682 .start = smc_pnet_dump_start
685 .cmd = SMC_PNETID_ADD,
686 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
687 .flags = GENL_ADMIN_PERM,
691 .cmd = SMC_PNETID_DEL,
692 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
693 .flags = GENL_ADMIN_PERM,
697 .cmd = SMC_PNETID_FLUSH,
698 .validate = GENL_DONT_VALIDATE_STRICT | GENL_DONT_VALIDATE_DUMP,
699 .flags = GENL_ADMIN_PERM,
700 .doit = smc_pnet_flush
704 /* SMC_PNETID family definition */
705 static struct genl_family smc_pnet_nl_family __ro_after_init = {
707 .name = SMCR_GENL_FAMILY_NAME,
708 .version = SMCR_GENL_FAMILY_VERSION,
709 .maxattr = SMC_PNETID_MAX,
710 .policy = smc_pnet_policy,
712 .module = THIS_MODULE,
714 .n_ops = ARRAY_SIZE(smc_pnet_ops)
717 bool smc_pnet_is_ndev_pnetid(struct net *net, u8 *pnetid)
719 struct smc_net *sn = net_generic(net, smc_net_id);
720 struct smc_pnetids_ndev_entry *pe;
723 read_lock(&sn->pnetids_ndev.lock);
724 list_for_each_entry(pe, &sn->pnetids_ndev.list, list) {
725 if (smc_pnet_match(pnetid, pe->pnetid)) {
732 read_unlock(&sn->pnetids_ndev.lock);
736 static int smc_pnet_add_pnetid(struct net *net, u8 *pnetid)
738 struct smc_net *sn = net_generic(net, smc_net_id);
739 struct smc_pnetids_ndev_entry *pe, *pi;
741 pe = kzalloc(sizeof(*pe), GFP_KERNEL);
745 write_lock(&sn->pnetids_ndev.lock);
746 list_for_each_entry(pi, &sn->pnetids_ndev.list, list) {
747 if (smc_pnet_match(pnetid, pe->pnetid)) {
748 refcount_inc(&pi->refcnt);
753 refcount_set(&pe->refcnt, 1);
754 memcpy(pe->pnetid, pnetid, SMC_MAX_PNETID_LEN);
755 list_add_tail(&pe->list, &sn->pnetids_ndev.list);
758 write_unlock(&sn->pnetids_ndev.lock);
762 static void smc_pnet_remove_pnetid(struct net *net, u8 *pnetid)
764 struct smc_net *sn = net_generic(net, smc_net_id);
765 struct smc_pnetids_ndev_entry *pe, *pe2;
767 write_lock(&sn->pnetids_ndev.lock);
768 list_for_each_entry_safe(pe, pe2, &sn->pnetids_ndev.list, list) {
769 if (smc_pnet_match(pnetid, pe->pnetid)) {
770 if (refcount_dec_and_test(&pe->refcnt)) {
777 write_unlock(&sn->pnetids_ndev.lock);
780 static void smc_pnet_add_base_pnetid(struct net *net, struct net_device *dev,
783 struct net_device *base_dev;
785 base_dev = __pnet_find_base_ndev(dev);
786 if (base_dev->flags & IFF_UP &&
787 !smc_pnetid_by_dev_port(base_dev->dev.parent, base_dev->dev_port,
789 /* add to PNETIDs list */
790 smc_pnet_add_pnetid(net, ndev_pnetid);
794 /* create initial list of netdevice pnetids */
795 static void smc_pnet_create_pnetids_list(struct net *net)
797 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
798 struct net_device *dev;
800 /* Newly created netns do not have devices.
801 * Do not even acquire rtnl.
803 if (list_empty(&net->dev_base_head))
806 /* Note: This might not be needed, because smc_pnet_netdev_event()
807 * is also calling smc_pnet_add_base_pnetid() when handling
811 for_each_netdev(net, dev)
812 smc_pnet_add_base_pnetid(net, dev, ndev_pnetid);
816 /* clean up list of netdevice pnetids */
817 static void smc_pnet_destroy_pnetids_list(struct net *net)
819 struct smc_net *sn = net_generic(net, smc_net_id);
820 struct smc_pnetids_ndev_entry *pe, *temp_pe;
822 write_lock(&sn->pnetids_ndev.lock);
823 list_for_each_entry_safe(pe, temp_pe, &sn->pnetids_ndev.list, list) {
827 write_unlock(&sn->pnetids_ndev.lock);
830 static int smc_pnet_netdev_event(struct notifier_block *this,
831 unsigned long event, void *ptr)
833 struct net_device *event_dev = netdev_notifier_info_to_dev(ptr);
834 struct net *net = dev_net(event_dev);
835 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
839 case NETDEV_UNREGISTER:
840 smc_pnet_remove_by_ndev(event_dev);
842 case NETDEV_REGISTER:
843 smc_pnet_add_by_ndev(event_dev);
846 smc_pnet_add_base_pnetid(net, event_dev, ndev_pnetid);
849 event_dev = __pnet_find_base_ndev(event_dev);
850 if (!smc_pnetid_by_dev_port(event_dev->dev.parent,
851 event_dev->dev_port, ndev_pnetid)) {
852 /* remove from PNETIDs list */
853 smc_pnet_remove_pnetid(net, ndev_pnetid);
861 static struct notifier_block smc_netdev_notifier = {
862 .notifier_call = smc_pnet_netdev_event
865 /* init network namespace */
866 int smc_pnet_net_init(struct net *net)
868 struct smc_net *sn = net_generic(net, smc_net_id);
869 struct smc_pnettable *pnettable = &sn->pnettable;
870 struct smc_pnetids_ndev *pnetids_ndev = &sn->pnetids_ndev;
872 INIT_LIST_HEAD(&pnettable->pnetlist);
873 mutex_init(&pnettable->lock);
874 INIT_LIST_HEAD(&pnetids_ndev->list);
875 rwlock_init(&pnetids_ndev->lock);
877 smc_pnet_create_pnetids_list(net);
882 int __init smc_pnet_init(void)
886 rc = genl_register_family(&smc_pnet_nl_family);
889 rc = register_netdevice_notifier(&smc_netdev_notifier);
891 genl_unregister_family(&smc_pnet_nl_family);
896 /* exit network namespace */
897 void smc_pnet_net_exit(struct net *net)
899 /* flush pnet table */
900 smc_pnet_remove_by_pnetid(net, NULL);
901 smc_pnet_destroy_pnetids_list(net);
904 void smc_pnet_exit(void)
906 unregister_netdevice_notifier(&smc_netdev_notifier);
907 genl_unregister_family(&smc_pnet_nl_family);
910 static struct net_device *__pnet_find_base_ndev(struct net_device *ndev)
915 nest_lvl = ndev->lower_level;
916 for (i = 0; i < nest_lvl; i++) {
917 struct list_head *lower = &ndev->adj_list.lower;
919 if (list_empty(lower))
922 ndev = netdev_lower_get_next(ndev, &lower);
927 /* Determine one base device for stacked net devices.
928 * If the lower device level contains more than one devices
929 * (for instance with bonding slaves), just the first device
930 * is used to reach a base device.
932 static struct net_device *pnet_find_base_ndev(struct net_device *ndev)
935 ndev = __pnet_find_base_ndev(ndev);
940 static int smc_pnet_find_ndev_pnetid_by_table(struct net_device *ndev,
943 struct smc_pnettable *pnettable;
944 struct net *net = dev_net(ndev);
945 struct smc_pnetentry *pnetelem;
949 /* get pnettable for namespace */
950 sn = net_generic(net, smc_net_id);
951 pnettable = &sn->pnettable;
953 mutex_lock(&pnettable->lock);
954 list_for_each_entry(pnetelem, &pnettable->pnetlist, list) {
955 if (pnetelem->type == SMC_PNET_ETH && ndev == pnetelem->ndev) {
956 /* get pnetid of netdev device */
957 memcpy(pnetid, pnetelem->pnet_name, SMC_MAX_PNETID_LEN);
962 mutex_unlock(&pnettable->lock);
966 /* find a roce device for the given pnetid */
967 static void _smc_pnet_find_roce_by_pnetid(u8 *pnet_id,
968 struct smc_init_info *ini,
969 struct smc_ib_device *known_dev)
971 struct smc_ib_device *ibdev;
975 mutex_lock(&smc_ib_devices.mutex);
976 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
977 if (ibdev == known_dev)
979 for (i = 1; i <= SMC_MAX_PORTS; i++) {
980 if (!rdma_is_port_valid(ibdev->ibdev, i))
982 if (smc_pnet_match(ibdev->pnetid[i - 1], pnet_id) &&
983 smc_ib_port_active(ibdev, i) &&
984 !test_bit(i - 1, ibdev->ports_going_away) &&
985 !smc_ib_determine_gid(ibdev, i, ini->vlan_id,
986 ini->ib_gid, NULL)) {
994 mutex_unlock(&smc_ib_devices.mutex);
997 /* find alternate roce device with same pnet_id and vlan_id */
998 void smc_pnet_find_alt_roce(struct smc_link_group *lgr,
999 struct smc_init_info *ini,
1000 struct smc_ib_device *known_dev)
1002 _smc_pnet_find_roce_by_pnetid(lgr->pnet_id, ini, known_dev);
1005 /* if handshake network device belongs to a roce device, return its
1006 * IB device and port
1008 static void smc_pnet_find_rdma_dev(struct net_device *netdev,
1009 struct smc_init_info *ini)
1011 struct smc_ib_device *ibdev;
1013 mutex_lock(&smc_ib_devices.mutex);
1014 list_for_each_entry(ibdev, &smc_ib_devices.list, list) {
1015 struct net_device *ndev;
1018 for (i = 1; i <= SMC_MAX_PORTS; i++) {
1019 if (!rdma_is_port_valid(ibdev->ibdev, i))
1021 if (!ibdev->ibdev->ops.get_netdev)
1023 ndev = ibdev->ibdev->ops.get_netdev(ibdev->ibdev, i);
1027 if (netdev == ndev &&
1028 smc_ib_port_active(ibdev, i) &&
1029 !test_bit(i - 1, ibdev->ports_going_away) &&
1030 !smc_ib_determine_gid(ibdev, i, ini->vlan_id,
1031 ini->ib_gid, NULL)) {
1032 ini->ib_dev = ibdev;
1038 mutex_unlock(&smc_ib_devices.mutex);
1041 /* Determine the corresponding IB device port based on the hardware PNETID.
1042 * Searching stops at the first matching active IB device port with vlan_id
1044 * If nothing found, check pnetid table.
1045 * If nothing found, try to use handshake device
1047 static void smc_pnet_find_roce_by_pnetid(struct net_device *ndev,
1048 struct smc_init_info *ini)
1050 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
1052 ndev = pnet_find_base_ndev(ndev);
1053 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port,
1055 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid)) {
1056 smc_pnet_find_rdma_dev(ndev, ini);
1057 return; /* pnetid could not be determined */
1059 _smc_pnet_find_roce_by_pnetid(ndev_pnetid, ini, NULL);
1062 static void smc_pnet_find_ism_by_pnetid(struct net_device *ndev,
1063 struct smc_init_info *ini)
1065 u8 ndev_pnetid[SMC_MAX_PNETID_LEN];
1066 struct smcd_dev *ismdev;
1068 ndev = pnet_find_base_ndev(ndev);
1069 if (smc_pnetid_by_dev_port(ndev->dev.parent, ndev->dev_port,
1071 smc_pnet_find_ndev_pnetid_by_table(ndev, ndev_pnetid))
1072 return; /* pnetid could not be determined */
1074 mutex_lock(&smcd_dev_list.mutex);
1075 list_for_each_entry(ismdev, &smcd_dev_list.list, list) {
1076 if (smc_pnet_match(ismdev->pnetid, ndev_pnetid) &&
1077 !ismdev->going_away &&
1078 (!ini->ism_peer_gid[0] ||
1079 !smc_ism_cantalk(ini->ism_peer_gid[0], ini->vlan_id,
1081 ini->ism_dev[0] = ismdev;
1085 mutex_unlock(&smcd_dev_list.mutex);
1088 /* PNET table analysis for a given sock:
1089 * determine ib_device and port belonging to used internal TCP socket
1090 * ethernet interface.
1092 void smc_pnet_find_roce_resource(struct sock *sk, struct smc_init_info *ini)
1094 struct dst_entry *dst = sk_dst_get(sk);
1103 smc_pnet_find_roce_by_pnetid(dst->dev, ini);
1111 void smc_pnet_find_ism_resource(struct sock *sk, struct smc_init_info *ini)
1113 struct dst_entry *dst = sk_dst_get(sk);
1115 ini->ism_dev[0] = NULL;
1121 smc_pnet_find_ism_by_pnetid(dst->dev, ini);
1129 /* Lookup and apply a pnet table entry to the given ib device.
1131 int smc_pnetid_by_table_ib(struct smc_ib_device *smcibdev, u8 ib_port)
1133 char *ib_name = smcibdev->ibdev->name;
1134 struct smc_pnettable *pnettable;
1135 struct smc_pnetentry *tmp_pe;
1139 /* get pnettable for init namespace */
1140 sn = net_generic(&init_net, smc_net_id);
1141 pnettable = &sn->pnettable;
1143 mutex_lock(&pnettable->lock);
1144 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
1145 if (tmp_pe->type == SMC_PNET_IB &&
1146 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX) &&
1147 tmp_pe->ib_port == ib_port) {
1148 smc_pnet_apply_ib(smcibdev, ib_port, tmp_pe->pnet_name);
1153 mutex_unlock(&pnettable->lock);
1158 /* Lookup and apply a pnet table entry to the given smcd device.
1160 int smc_pnetid_by_table_smcd(struct smcd_dev *smcddev)
1162 const char *ib_name = dev_name(&smcddev->dev);
1163 struct smc_pnettable *pnettable;
1164 struct smc_pnetentry *tmp_pe;
1168 /* get pnettable for init namespace */
1169 sn = net_generic(&init_net, smc_net_id);
1170 pnettable = &sn->pnettable;
1172 mutex_lock(&pnettable->lock);
1173 list_for_each_entry(tmp_pe, &pnettable->pnetlist, list) {
1174 if (tmp_pe->type == SMC_PNET_IB &&
1175 !strncmp(tmp_pe->ib_name, ib_name, IB_DEVICE_NAME_MAX)) {
1176 smc_pnet_apply_smcd(smcddev, tmp_pe->pnet_name);
1181 mutex_unlock(&pnettable->lock);