1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net-sysfs.c - network device class and attributes
5 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
8 #include <linux/capability.h>
9 #include <linux/kernel.h>
10 #include <linux/netdevice.h>
11 #include <linux/if_arp.h>
12 #include <linux/slab.h>
13 #include <linux/sched/signal.h>
14 #include <linux/sched/isolation.h>
15 #include <linux/nsproxy.h>
17 #include <net/net_namespace.h>
18 #include <linux/rtnetlink.h>
19 #include <linux/vmalloc.h>
20 #include <linux/export.h>
21 #include <linux/jiffies.h>
22 #include <linux/pm_runtime.h>
24 #include <linux/of_net.h>
25 #include <linux/cpu.h>
27 #include "net-sysfs.h"
30 static const char fmt_hex[] = "%#x\n";
31 static const char fmt_dec[] = "%d\n";
32 static const char fmt_ulong[] = "%lu\n";
33 static const char fmt_u64[] = "%llu\n";
35 static inline int dev_isalive(const struct net_device *dev)
37 return dev->reg_state <= NETREG_REGISTERED;
40 /* use same locking rules as GIF* ioctl's */
41 static ssize_t netdev_show(const struct device *dev,
42 struct device_attribute *attr, char *buf,
43 ssize_t (*format)(const struct net_device *, char *))
45 struct net_device *ndev = to_net_dev(dev);
46 ssize_t ret = -EINVAL;
48 read_lock(&dev_base_lock);
49 if (dev_isalive(ndev))
50 ret = (*format)(ndev, buf);
51 read_unlock(&dev_base_lock);
56 /* generate a show function for simple field */
57 #define NETDEVICE_SHOW(field, format_string) \
58 static ssize_t format_##field(const struct net_device *dev, char *buf) \
60 return sprintf(buf, format_string, dev->field); \
62 static ssize_t field##_show(struct device *dev, \
63 struct device_attribute *attr, char *buf) \
65 return netdev_show(dev, attr, buf, format_##field); \
68 #define NETDEVICE_SHOW_RO(field, format_string) \
69 NETDEVICE_SHOW(field, format_string); \
70 static DEVICE_ATTR_RO(field)
72 #define NETDEVICE_SHOW_RW(field, format_string) \
73 NETDEVICE_SHOW(field, format_string); \
74 static DEVICE_ATTR_RW(field)
76 /* use same locking and permission rules as SIF* ioctl's */
77 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
78 const char *buf, size_t len,
79 int (*set)(struct net_device *, unsigned long))
81 struct net_device *netdev = to_net_dev(dev);
82 struct net *net = dev_net(netdev);
86 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
89 ret = kstrtoul(buf, 0, &new);
94 return restart_syscall();
96 if (dev_isalive(netdev)) {
97 ret = (*set)(netdev, new);
106 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
107 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
108 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
109 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
110 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
111 NETDEVICE_SHOW_RO(type, fmt_dec);
112 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
114 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
117 struct net_device *ndev = to_net_dev(dev);
119 return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
121 static DEVICE_ATTR_RO(iflink);
123 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
125 return sprintf(buf, fmt_dec, dev->name_assign_type);
128 static ssize_t name_assign_type_show(struct device *dev,
129 struct device_attribute *attr,
132 struct net_device *ndev = to_net_dev(dev);
133 ssize_t ret = -EINVAL;
135 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
136 ret = netdev_show(dev, attr, buf, format_name_assign_type);
140 static DEVICE_ATTR_RO(name_assign_type);
142 /* use same locking rules as GIFHWADDR ioctl's */
143 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
146 struct net_device *ndev = to_net_dev(dev);
147 ssize_t ret = -EINVAL;
149 read_lock(&dev_base_lock);
150 if (dev_isalive(ndev))
151 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
152 read_unlock(&dev_base_lock);
155 static DEVICE_ATTR_RO(address);
157 static ssize_t broadcast_show(struct device *dev,
158 struct device_attribute *attr, char *buf)
160 struct net_device *ndev = to_net_dev(dev);
162 if (dev_isalive(ndev))
163 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
166 static DEVICE_ATTR_RO(broadcast);
168 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
170 if (!netif_running(dev))
172 return dev_change_carrier(dev, (bool)new_carrier);
175 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
176 const char *buf, size_t len)
178 struct net_device *netdev = to_net_dev(dev);
180 /* The check is also done in change_carrier; this helps returning early
181 * without hitting the trylock/restart in netdev_store.
183 if (!netdev->netdev_ops->ndo_change_carrier)
186 return netdev_store(dev, attr, buf, len, change_carrier);
189 static ssize_t carrier_show(struct device *dev,
190 struct device_attribute *attr, char *buf)
192 struct net_device *netdev = to_net_dev(dev);
194 if (netif_running(netdev))
195 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
199 static DEVICE_ATTR_RW(carrier);
201 static ssize_t speed_show(struct device *dev,
202 struct device_attribute *attr, char *buf)
204 struct net_device *netdev = to_net_dev(dev);
207 /* The check is also done in __ethtool_get_link_ksettings; this helps
208 * returning early without hitting the trylock/restart below.
210 if (!netdev->ethtool_ops->get_link_ksettings)
214 return restart_syscall();
216 if (netif_running(netdev) && netif_device_present(netdev)) {
217 struct ethtool_link_ksettings cmd;
219 if (!__ethtool_get_link_ksettings(netdev, &cmd))
220 ret = sprintf(buf, fmt_dec, cmd.base.speed);
225 static DEVICE_ATTR_RO(speed);
227 static ssize_t duplex_show(struct device *dev,
228 struct device_attribute *attr, char *buf)
230 struct net_device *netdev = to_net_dev(dev);
233 /* The check is also done in __ethtool_get_link_ksettings; this helps
234 * returning early without hitting the trylock/restart below.
236 if (!netdev->ethtool_ops->get_link_ksettings)
240 return restart_syscall();
242 if (netif_running(netdev)) {
243 struct ethtool_link_ksettings cmd;
245 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
248 switch (cmd.base.duplex) {
259 ret = sprintf(buf, "%s\n", duplex);
265 static DEVICE_ATTR_RO(duplex);
267 static ssize_t testing_show(struct device *dev,
268 struct device_attribute *attr, char *buf)
270 struct net_device *netdev = to_net_dev(dev);
272 if (netif_running(netdev))
273 return sprintf(buf, fmt_dec, !!netif_testing(netdev));
277 static DEVICE_ATTR_RO(testing);
279 static ssize_t dormant_show(struct device *dev,
280 struct device_attribute *attr, char *buf)
282 struct net_device *netdev = to_net_dev(dev);
284 if (netif_running(netdev))
285 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
289 static DEVICE_ATTR_RO(dormant);
291 static const char *const operstates[] = {
293 "notpresent", /* currently unused */
301 static ssize_t operstate_show(struct device *dev,
302 struct device_attribute *attr, char *buf)
304 const struct net_device *netdev = to_net_dev(dev);
305 unsigned char operstate;
307 read_lock(&dev_base_lock);
308 operstate = netdev->operstate;
309 if (!netif_running(netdev))
310 operstate = IF_OPER_DOWN;
311 read_unlock(&dev_base_lock);
313 if (operstate >= ARRAY_SIZE(operstates))
314 return -EINVAL; /* should not happen */
316 return sprintf(buf, "%s\n", operstates[operstate]);
318 static DEVICE_ATTR_RO(operstate);
320 static ssize_t carrier_changes_show(struct device *dev,
321 struct device_attribute *attr,
324 struct net_device *netdev = to_net_dev(dev);
326 return sprintf(buf, fmt_dec,
327 atomic_read(&netdev->carrier_up_count) +
328 atomic_read(&netdev->carrier_down_count));
330 static DEVICE_ATTR_RO(carrier_changes);
332 static ssize_t carrier_up_count_show(struct device *dev,
333 struct device_attribute *attr,
336 struct net_device *netdev = to_net_dev(dev);
338 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
340 static DEVICE_ATTR_RO(carrier_up_count);
342 static ssize_t carrier_down_count_show(struct device *dev,
343 struct device_attribute *attr,
346 struct net_device *netdev = to_net_dev(dev);
348 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
350 static DEVICE_ATTR_RO(carrier_down_count);
352 /* read-write attributes */
354 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
356 return dev_set_mtu(dev, (int)new_mtu);
359 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
360 const char *buf, size_t len)
362 return netdev_store(dev, attr, buf, len, change_mtu);
364 NETDEVICE_SHOW_RW(mtu, fmt_dec);
366 static int change_flags(struct net_device *dev, unsigned long new_flags)
368 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
371 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
372 const char *buf, size_t len)
374 return netdev_store(dev, attr, buf, len, change_flags);
376 NETDEVICE_SHOW_RW(flags, fmt_hex);
378 static ssize_t tx_queue_len_store(struct device *dev,
379 struct device_attribute *attr,
380 const char *buf, size_t len)
382 if (!capable(CAP_NET_ADMIN))
385 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
387 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
389 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
391 WRITE_ONCE(dev->gro_flush_timeout, val);
395 static ssize_t gro_flush_timeout_store(struct device *dev,
396 struct device_attribute *attr,
397 const char *buf, size_t len)
399 if (!capable(CAP_NET_ADMIN))
402 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
404 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
406 static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
408 WRITE_ONCE(dev->napi_defer_hard_irqs, val);
412 static ssize_t napi_defer_hard_irqs_store(struct device *dev,
413 struct device_attribute *attr,
414 const char *buf, size_t len)
416 if (!capable(CAP_NET_ADMIN))
419 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
421 NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
423 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
424 const char *buf, size_t len)
426 struct net_device *netdev = to_net_dev(dev);
427 struct net *net = dev_net(netdev);
431 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
434 /* ignore trailing newline */
435 if (len > 0 && buf[len - 1] == '\n')
439 return restart_syscall();
441 if (dev_isalive(netdev)) {
442 ret = dev_set_alias(netdev, buf, count);
446 netdev_state_change(netdev);
454 static ssize_t ifalias_show(struct device *dev,
455 struct device_attribute *attr, char *buf)
457 const struct net_device *netdev = to_net_dev(dev);
461 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
463 ret = sprintf(buf, "%s\n", tmp);
466 static DEVICE_ATTR_RW(ifalias);
468 static int change_group(struct net_device *dev, unsigned long new_group)
470 dev_set_group(dev, (int)new_group);
474 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
475 const char *buf, size_t len)
477 return netdev_store(dev, attr, buf, len, change_group);
479 NETDEVICE_SHOW(group, fmt_dec);
480 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
482 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
484 return dev_change_proto_down(dev, (bool)proto_down);
487 static ssize_t proto_down_store(struct device *dev,
488 struct device_attribute *attr,
489 const char *buf, size_t len)
491 struct net_device *netdev = to_net_dev(dev);
493 /* The check is also done in change_proto_down; this helps returning
494 * early without hitting the trylock/restart in netdev_store.
496 if (!netdev->netdev_ops->ndo_change_proto_down)
499 return netdev_store(dev, attr, buf, len, change_proto_down);
501 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
503 static ssize_t phys_port_id_show(struct device *dev,
504 struct device_attribute *attr, char *buf)
506 struct net_device *netdev = to_net_dev(dev);
507 ssize_t ret = -EINVAL;
509 /* The check is also done in dev_get_phys_port_id; this helps returning
510 * early without hitting the trylock/restart below.
512 if (!netdev->netdev_ops->ndo_get_phys_port_id)
516 return restart_syscall();
518 if (dev_isalive(netdev)) {
519 struct netdev_phys_item_id ppid;
521 ret = dev_get_phys_port_id(netdev, &ppid);
523 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
529 static DEVICE_ATTR_RO(phys_port_id);
531 static ssize_t phys_port_name_show(struct device *dev,
532 struct device_attribute *attr, char *buf)
534 struct net_device *netdev = to_net_dev(dev);
535 ssize_t ret = -EINVAL;
537 /* The checks are also done in dev_get_phys_port_name; this helps
538 * returning early without hitting the trylock/restart below.
540 if (!netdev->netdev_ops->ndo_get_phys_port_name &&
541 !netdev->netdev_ops->ndo_get_devlink_port)
545 return restart_syscall();
547 if (dev_isalive(netdev)) {
550 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
552 ret = sprintf(buf, "%s\n", name);
558 static DEVICE_ATTR_RO(phys_port_name);
560 static ssize_t phys_switch_id_show(struct device *dev,
561 struct device_attribute *attr, char *buf)
563 struct net_device *netdev = to_net_dev(dev);
564 ssize_t ret = -EINVAL;
566 /* The checks are also done in dev_get_phys_port_name; this helps
567 * returning early without hitting the trylock/restart below. This works
568 * because recurse is false when calling dev_get_port_parent_id.
570 if (!netdev->netdev_ops->ndo_get_port_parent_id &&
571 !netdev->netdev_ops->ndo_get_devlink_port)
575 return restart_syscall();
577 if (dev_isalive(netdev)) {
578 struct netdev_phys_item_id ppid = { };
580 ret = dev_get_port_parent_id(netdev, &ppid, false);
582 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
588 static DEVICE_ATTR_RO(phys_switch_id);
590 static struct attribute *net_class_attrs[] __ro_after_init = {
591 &dev_attr_netdev_group.attr,
593 &dev_attr_dev_id.attr,
594 &dev_attr_dev_port.attr,
595 &dev_attr_iflink.attr,
596 &dev_attr_ifindex.attr,
597 &dev_attr_name_assign_type.attr,
598 &dev_attr_addr_assign_type.attr,
599 &dev_attr_addr_len.attr,
600 &dev_attr_link_mode.attr,
601 &dev_attr_address.attr,
602 &dev_attr_broadcast.attr,
603 &dev_attr_speed.attr,
604 &dev_attr_duplex.attr,
605 &dev_attr_dormant.attr,
606 &dev_attr_testing.attr,
607 &dev_attr_operstate.attr,
608 &dev_attr_carrier_changes.attr,
609 &dev_attr_ifalias.attr,
610 &dev_attr_carrier.attr,
612 &dev_attr_flags.attr,
613 &dev_attr_tx_queue_len.attr,
614 &dev_attr_gro_flush_timeout.attr,
615 &dev_attr_napi_defer_hard_irqs.attr,
616 &dev_attr_phys_port_id.attr,
617 &dev_attr_phys_port_name.attr,
618 &dev_attr_phys_switch_id.attr,
619 &dev_attr_proto_down.attr,
620 &dev_attr_carrier_up_count.attr,
621 &dev_attr_carrier_down_count.attr,
624 ATTRIBUTE_GROUPS(net_class);
626 /* Show a given an attribute in the statistics group */
627 static ssize_t netstat_show(const struct device *d,
628 struct device_attribute *attr, char *buf,
629 unsigned long offset)
631 struct net_device *dev = to_net_dev(d);
632 ssize_t ret = -EINVAL;
634 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
635 offset % sizeof(u64) != 0);
637 read_lock(&dev_base_lock);
638 if (dev_isalive(dev)) {
639 struct rtnl_link_stats64 temp;
640 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
642 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
644 read_unlock(&dev_base_lock);
648 /* generate a read-only statistics attribute */
649 #define NETSTAT_ENTRY(name) \
650 static ssize_t name##_show(struct device *d, \
651 struct device_attribute *attr, char *buf) \
653 return netstat_show(d, attr, buf, \
654 offsetof(struct rtnl_link_stats64, name)); \
656 static DEVICE_ATTR_RO(name)
658 NETSTAT_ENTRY(rx_packets);
659 NETSTAT_ENTRY(tx_packets);
660 NETSTAT_ENTRY(rx_bytes);
661 NETSTAT_ENTRY(tx_bytes);
662 NETSTAT_ENTRY(rx_errors);
663 NETSTAT_ENTRY(tx_errors);
664 NETSTAT_ENTRY(rx_dropped);
665 NETSTAT_ENTRY(tx_dropped);
666 NETSTAT_ENTRY(multicast);
667 NETSTAT_ENTRY(collisions);
668 NETSTAT_ENTRY(rx_length_errors);
669 NETSTAT_ENTRY(rx_over_errors);
670 NETSTAT_ENTRY(rx_crc_errors);
671 NETSTAT_ENTRY(rx_frame_errors);
672 NETSTAT_ENTRY(rx_fifo_errors);
673 NETSTAT_ENTRY(rx_missed_errors);
674 NETSTAT_ENTRY(tx_aborted_errors);
675 NETSTAT_ENTRY(tx_carrier_errors);
676 NETSTAT_ENTRY(tx_fifo_errors);
677 NETSTAT_ENTRY(tx_heartbeat_errors);
678 NETSTAT_ENTRY(tx_window_errors);
679 NETSTAT_ENTRY(rx_compressed);
680 NETSTAT_ENTRY(tx_compressed);
681 NETSTAT_ENTRY(rx_nohandler);
683 static struct attribute *netstat_attrs[] __ro_after_init = {
684 &dev_attr_rx_packets.attr,
685 &dev_attr_tx_packets.attr,
686 &dev_attr_rx_bytes.attr,
687 &dev_attr_tx_bytes.attr,
688 &dev_attr_rx_errors.attr,
689 &dev_attr_tx_errors.attr,
690 &dev_attr_rx_dropped.attr,
691 &dev_attr_tx_dropped.attr,
692 &dev_attr_multicast.attr,
693 &dev_attr_collisions.attr,
694 &dev_attr_rx_length_errors.attr,
695 &dev_attr_rx_over_errors.attr,
696 &dev_attr_rx_crc_errors.attr,
697 &dev_attr_rx_frame_errors.attr,
698 &dev_attr_rx_fifo_errors.attr,
699 &dev_attr_rx_missed_errors.attr,
700 &dev_attr_tx_aborted_errors.attr,
701 &dev_attr_tx_carrier_errors.attr,
702 &dev_attr_tx_fifo_errors.attr,
703 &dev_attr_tx_heartbeat_errors.attr,
704 &dev_attr_tx_window_errors.attr,
705 &dev_attr_rx_compressed.attr,
706 &dev_attr_tx_compressed.attr,
707 &dev_attr_rx_nohandler.attr,
711 static const struct attribute_group netstat_group = {
712 .name = "statistics",
713 .attrs = netstat_attrs,
716 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
717 static struct attribute *wireless_attrs[] = {
721 static const struct attribute_group wireless_group = {
723 .attrs = wireless_attrs,
727 #else /* CONFIG_SYSFS */
728 #define net_class_groups NULL
729 #endif /* CONFIG_SYSFS */
732 #define to_rx_queue_attr(_attr) \
733 container_of(_attr, struct rx_queue_attribute, attr)
735 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
737 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
740 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
741 struct netdev_rx_queue *queue = to_rx_queue(kobj);
743 if (!attribute->show)
746 return attribute->show(queue, buf);
749 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
750 const char *buf, size_t count)
752 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
753 struct netdev_rx_queue *queue = to_rx_queue(kobj);
755 if (!attribute->store)
758 return attribute->store(queue, buf, count);
761 static const struct sysfs_ops rx_queue_sysfs_ops = {
762 .show = rx_queue_attr_show,
763 .store = rx_queue_attr_store,
767 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
773 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
777 map = rcu_dereference(queue->rps_map);
779 for (i = 0; i < map->len; i++)
780 cpumask_set_cpu(map->cpus[i], mask);
782 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
784 free_cpumask_var(mask);
786 return len < PAGE_SIZE ? len : -EINVAL;
789 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
790 const char *buf, size_t len)
792 struct rps_map *old_map, *map;
794 int err, cpu, i, hk_flags;
795 static DEFINE_MUTEX(rps_map_mutex);
797 if (!capable(CAP_NET_ADMIN))
800 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
803 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
805 free_cpumask_var(mask);
809 if (!cpumask_empty(mask)) {
810 hk_flags = HK_FLAG_DOMAIN | HK_FLAG_WQ;
811 cpumask_and(mask, mask, housekeeping_cpumask(hk_flags));
812 if (cpumask_empty(mask)) {
813 free_cpumask_var(mask);
818 map = kzalloc(max_t(unsigned int,
819 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
822 free_cpumask_var(mask);
827 for_each_cpu_and(cpu, mask, cpu_online_mask)
828 map->cpus[i++] = cpu;
837 mutex_lock(&rps_map_mutex);
838 old_map = rcu_dereference_protected(queue->rps_map,
839 mutex_is_locked(&rps_map_mutex));
840 rcu_assign_pointer(queue->rps_map, map);
843 static_branch_inc(&rps_needed);
845 static_branch_dec(&rps_needed);
847 mutex_unlock(&rps_map_mutex);
850 kfree_rcu(old_map, rcu);
852 free_cpumask_var(mask);
856 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
859 struct rps_dev_flow_table *flow_table;
860 unsigned long val = 0;
863 flow_table = rcu_dereference(queue->rps_flow_table);
865 val = (unsigned long)flow_table->mask + 1;
868 return sprintf(buf, "%lu\n", val);
871 static void rps_dev_flow_table_release(struct rcu_head *rcu)
873 struct rps_dev_flow_table *table = container_of(rcu,
874 struct rps_dev_flow_table, rcu);
878 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
879 const char *buf, size_t len)
881 unsigned long mask, count;
882 struct rps_dev_flow_table *table, *old_table;
883 static DEFINE_SPINLOCK(rps_dev_flow_lock);
886 if (!capable(CAP_NET_ADMIN))
889 rc = kstrtoul(buf, 0, &count);
895 /* mask = roundup_pow_of_two(count) - 1;
896 * without overflows...
898 while ((mask | (mask >> 1)) != mask)
900 /* On 64 bit arches, must check mask fits in table->mask (u32),
901 * and on 32bit arches, must check
902 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
904 #if BITS_PER_LONG > 32
905 if (mask > (unsigned long)(u32)mask)
908 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
909 / sizeof(struct rps_dev_flow)) {
910 /* Enforce a limit to prevent overflow */
914 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
919 for (count = 0; count <= mask; count++)
920 table->flows[count].cpu = RPS_NO_CPU;
925 spin_lock(&rps_dev_flow_lock);
926 old_table = rcu_dereference_protected(queue->rps_flow_table,
927 lockdep_is_held(&rps_dev_flow_lock));
928 rcu_assign_pointer(queue->rps_flow_table, table);
929 spin_unlock(&rps_dev_flow_lock);
932 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
937 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
938 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
940 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
941 = __ATTR(rps_flow_cnt, 0644,
942 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
943 #endif /* CONFIG_RPS */
945 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
947 &rps_cpus_attribute.attr,
948 &rps_dev_flow_table_cnt_attribute.attr,
952 ATTRIBUTE_GROUPS(rx_queue_default);
954 static void rx_queue_release(struct kobject *kobj)
956 struct netdev_rx_queue *queue = to_rx_queue(kobj);
959 struct rps_dev_flow_table *flow_table;
961 map = rcu_dereference_protected(queue->rps_map, 1);
963 RCU_INIT_POINTER(queue->rps_map, NULL);
967 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
969 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
970 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
974 memset(kobj, 0, sizeof(*kobj));
978 static const void *rx_queue_namespace(struct kobject *kobj)
980 struct netdev_rx_queue *queue = to_rx_queue(kobj);
981 struct device *dev = &queue->dev->dev;
982 const void *ns = NULL;
984 if (dev->class && dev->class->ns_type)
985 ns = dev->class->namespace(dev);
990 static void rx_queue_get_ownership(struct kobject *kobj,
991 kuid_t *uid, kgid_t *gid)
993 const struct net *net = rx_queue_namespace(kobj);
995 net_ns_get_ownership(net, uid, gid);
998 static struct kobj_type rx_queue_ktype __ro_after_init = {
999 .sysfs_ops = &rx_queue_sysfs_ops,
1000 .release = rx_queue_release,
1001 .default_groups = rx_queue_default_groups,
1002 .namespace = rx_queue_namespace,
1003 .get_ownership = rx_queue_get_ownership,
1006 static int rx_queue_add_kobject(struct net_device *dev, int index)
1008 struct netdev_rx_queue *queue = dev->_rx + index;
1009 struct kobject *kobj = &queue->kobj;
1012 /* Kobject_put later will trigger rx_queue_release call which
1013 * decreases dev refcount: Take that reference here
1015 dev_hold(queue->dev);
1017 kobj->kset = dev->queues_kset;
1018 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1023 if (dev->sysfs_rx_queue_group) {
1024 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1029 kobject_uevent(kobj, KOBJ_ADD);
1038 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1041 struct netdev_rx_queue *queue = dev->_rx + index;
1042 struct kobject *kobj = &queue->kobj;
1045 error = sysfs_change_owner(kobj, kuid, kgid);
1049 if (dev->sysfs_rx_queue_group)
1050 error = sysfs_group_change_owner(
1051 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1055 #endif /* CONFIG_SYSFS */
1058 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1065 if (!dev->sysfs_rx_queue_group)
1068 for (i = old_num; i < new_num; i++) {
1069 error = rx_queue_add_kobject(dev, i);
1076 while (--i >= new_num) {
1077 struct kobject *kobj = &dev->_rx[i].kobj;
1079 if (!refcount_read(&dev_net(dev)->count))
1080 kobj->uevent_suppress = 1;
1081 if (dev->sysfs_rx_queue_group)
1082 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1092 static int net_rx_queue_change_owner(struct net_device *dev, int num,
1093 kuid_t kuid, kgid_t kgid)
1100 if (!dev->sysfs_rx_queue_group)
1103 for (i = 0; i < num; i++) {
1104 error = rx_queue_change_owner(dev, i, kuid, kgid);
1117 * netdev_queue sysfs structures and functions.
1119 struct netdev_queue_attribute {
1120 struct attribute attr;
1121 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1122 ssize_t (*store)(struct netdev_queue *queue,
1123 const char *buf, size_t len);
1125 #define to_netdev_queue_attr(_attr) \
1126 container_of(_attr, struct netdev_queue_attribute, attr)
1128 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1130 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1131 struct attribute *attr, char *buf)
1133 const struct netdev_queue_attribute *attribute
1134 = to_netdev_queue_attr(attr);
1135 struct netdev_queue *queue = to_netdev_queue(kobj);
1137 if (!attribute->show)
1140 return attribute->show(queue, buf);
1143 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1144 struct attribute *attr,
1145 const char *buf, size_t count)
1147 const struct netdev_queue_attribute *attribute
1148 = to_netdev_queue_attr(attr);
1149 struct netdev_queue *queue = to_netdev_queue(kobj);
1151 if (!attribute->store)
1154 return attribute->store(queue, buf, count);
1157 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1158 .show = netdev_queue_attr_show,
1159 .store = netdev_queue_attr_store,
1162 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1164 unsigned long trans_timeout;
1166 spin_lock_irq(&queue->_xmit_lock);
1167 trans_timeout = queue->trans_timeout;
1168 spin_unlock_irq(&queue->_xmit_lock);
1170 return sprintf(buf, fmt_ulong, trans_timeout);
1173 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1175 struct net_device *dev = queue->dev;
1178 i = queue - dev->_tx;
1179 BUG_ON(i >= dev->num_tx_queues);
1184 static ssize_t traffic_class_show(struct netdev_queue *queue,
1187 struct net_device *dev = queue->dev;
1191 if (!netif_is_multiqueue(dev))
1194 index = get_netdev_queue_index(queue);
1196 /* If queue belongs to subordinate dev use its TC mapping */
1197 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1199 tc = netdev_txq_to_tc(dev, index);
1203 /* We can report the traffic class one of two ways:
1204 * Subordinate device traffic classes are reported with the traffic
1205 * class first, and then the subordinate class so for example TC0 on
1206 * subordinate device 2 will be reported as "0-2". If the queue
1207 * belongs to the root device it will be reported with just the
1208 * traffic class, so just "0" for TC 0 for example.
1210 return dev->num_tc < 0 ? sprintf(buf, "%d%d\n", tc, dev->num_tc) :
1211 sprintf(buf, "%d\n", tc);
1215 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1218 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1221 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1222 const char *buf, size_t len)
1224 struct net_device *dev = queue->dev;
1225 int err, index = get_netdev_queue_index(queue);
1228 if (!capable(CAP_NET_ADMIN))
1231 /* The check is also done later; this helps returning early without
1232 * hitting the trylock/restart below.
1234 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1237 err = kstrtou32(buf, 10, &rate);
1241 if (!rtnl_trylock())
1242 return restart_syscall();
1245 if (dev->netdev_ops->ndo_set_tx_maxrate)
1246 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1250 queue->tx_maxrate = rate;
1256 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1257 = __ATTR_RW(tx_maxrate);
1260 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1261 = __ATTR_RO(tx_timeout);
1263 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1264 = __ATTR_RO(traffic_class);
1268 * Byte queue limits sysfs structures and functions.
1270 static ssize_t bql_show(char *buf, unsigned int value)
1272 return sprintf(buf, "%u\n", value);
1275 static ssize_t bql_set(const char *buf, const size_t count,
1276 unsigned int *pvalue)
1281 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1282 value = DQL_MAX_LIMIT;
1284 err = kstrtouint(buf, 10, &value);
1287 if (value > DQL_MAX_LIMIT)
1296 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1299 struct dql *dql = &queue->dql;
1301 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1304 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1305 const char *buf, size_t len)
1307 struct dql *dql = &queue->dql;
1311 err = kstrtouint(buf, 10, &value);
1315 dql->slack_hold_time = msecs_to_jiffies(value);
1320 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1321 = __ATTR(hold_time, 0644,
1322 bql_show_hold_time, bql_set_hold_time);
1324 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1327 struct dql *dql = &queue->dql;
1329 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1332 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1333 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1335 #define BQL_ATTR(NAME, FIELD) \
1336 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1339 return bql_show(buf, queue->dql.FIELD); \
1342 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1343 const char *buf, size_t len) \
1345 return bql_set(buf, len, &queue->dql.FIELD); \
1348 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1349 = __ATTR(NAME, 0644, \
1350 bql_show_ ## NAME, bql_set_ ## NAME)
1352 BQL_ATTR(limit, limit);
1353 BQL_ATTR(limit_max, max_limit);
1354 BQL_ATTR(limit_min, min_limit);
1356 static struct attribute *dql_attrs[] __ro_after_init = {
1357 &bql_limit_attribute.attr,
1358 &bql_limit_max_attribute.attr,
1359 &bql_limit_min_attribute.attr,
1360 &bql_hold_time_attribute.attr,
1361 &bql_inflight_attribute.attr,
1365 static const struct attribute_group dql_group = {
1366 .name = "byte_queue_limits",
1369 #endif /* CONFIG_BQL */
1372 static ssize_t xps_cpus_show(struct netdev_queue *queue,
1375 int cpu, len, ret, num_tc = 1, tc = 0;
1376 struct net_device *dev = queue->dev;
1377 struct xps_dev_maps *dev_maps;
1379 unsigned long index;
1381 if (!netif_is_multiqueue(dev))
1384 index = get_netdev_queue_index(queue);
1386 if (!rtnl_trylock())
1387 return restart_syscall();
1390 /* Do not allow XPS on subordinate device directly */
1391 num_tc = dev->num_tc;
1394 goto err_rtnl_unlock;
1397 /* If queue belongs to subordinate dev use its map */
1398 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1400 tc = netdev_txq_to_tc(dev, index);
1403 goto err_rtnl_unlock;
1407 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
1409 goto err_rtnl_unlock;
1413 dev_maps = rcu_dereference(dev->xps_cpus_map);
1415 for_each_possible_cpu(cpu) {
1416 int i, tci = cpu * num_tc + tc;
1417 struct xps_map *map;
1419 map = rcu_dereference(dev_maps->attr_map[tci]);
1423 for (i = map->len; i--;) {
1424 if (map->queues[i] == index) {
1425 cpumask_set_cpu(cpu, mask);
1435 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1436 free_cpumask_var(mask);
1437 return len < PAGE_SIZE ? len : -EINVAL;
1444 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1445 const char *buf, size_t len)
1447 struct net_device *dev = queue->dev;
1448 unsigned long index;
1452 if (!netif_is_multiqueue(dev))
1455 if (!capable(CAP_NET_ADMIN))
1458 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1461 index = get_netdev_queue_index(queue);
1463 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1465 free_cpumask_var(mask);
1469 if (!rtnl_trylock()) {
1470 free_cpumask_var(mask);
1471 return restart_syscall();
1474 err = netif_set_xps_queue(dev, mask, index);
1477 free_cpumask_var(mask);
1482 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1483 = __ATTR_RW(xps_cpus);
1485 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1487 int j, len, ret, num_tc = 1, tc = 0;
1488 struct net_device *dev = queue->dev;
1489 struct xps_dev_maps *dev_maps;
1490 unsigned long *mask, index;
1492 index = get_netdev_queue_index(queue);
1494 if (!rtnl_trylock())
1495 return restart_syscall();
1498 num_tc = dev->num_tc;
1499 tc = netdev_txq_to_tc(dev, index);
1502 goto err_rtnl_unlock;
1505 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1508 goto err_rtnl_unlock;
1512 dev_maps = rcu_dereference(dev->xps_rxqs_map);
1516 for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues),
1517 j < dev->num_rx_queues;) {
1518 int i, tci = j * num_tc + tc;
1519 struct xps_map *map;
1521 map = rcu_dereference(dev_maps->attr_map[tci]);
1525 for (i = map->len; i--;) {
1526 if (map->queues[i] == index) {
1537 len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues);
1540 return len < PAGE_SIZE ? len : -EINVAL;
1547 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1550 struct net_device *dev = queue->dev;
1551 struct net *net = dev_net(dev);
1552 unsigned long *mask, index;
1555 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1558 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1562 index = get_netdev_queue_index(queue);
1564 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1570 if (!rtnl_trylock()) {
1572 return restart_syscall();
1576 err = __netif_set_xps_queue(dev, mask, index, true);
1585 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1586 = __ATTR_RW(xps_rxqs);
1587 #endif /* CONFIG_XPS */
1589 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1590 &queue_trans_timeout.attr,
1591 &queue_traffic_class.attr,
1593 &xps_cpus_attribute.attr,
1594 &xps_rxqs_attribute.attr,
1595 &queue_tx_maxrate.attr,
1599 ATTRIBUTE_GROUPS(netdev_queue_default);
1601 static void netdev_queue_release(struct kobject *kobj)
1603 struct netdev_queue *queue = to_netdev_queue(kobj);
1605 memset(kobj, 0, sizeof(*kobj));
1606 dev_put(queue->dev);
1609 static const void *netdev_queue_namespace(struct kobject *kobj)
1611 struct netdev_queue *queue = to_netdev_queue(kobj);
1612 struct device *dev = &queue->dev->dev;
1613 const void *ns = NULL;
1615 if (dev->class && dev->class->ns_type)
1616 ns = dev->class->namespace(dev);
1621 static void netdev_queue_get_ownership(struct kobject *kobj,
1622 kuid_t *uid, kgid_t *gid)
1624 const struct net *net = netdev_queue_namespace(kobj);
1626 net_ns_get_ownership(net, uid, gid);
1629 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1630 .sysfs_ops = &netdev_queue_sysfs_ops,
1631 .release = netdev_queue_release,
1632 .default_groups = netdev_queue_default_groups,
1633 .namespace = netdev_queue_namespace,
1634 .get_ownership = netdev_queue_get_ownership,
1637 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1639 struct netdev_queue *queue = dev->_tx + index;
1640 struct kobject *kobj = &queue->kobj;
1643 /* Kobject_put later will trigger netdev_queue_release call
1644 * which decreases dev refcount: Take that reference here
1646 dev_hold(queue->dev);
1648 kobj->kset = dev->queues_kset;
1649 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1655 error = sysfs_create_group(kobj, &dql_group);
1660 kobject_uevent(kobj, KOBJ_ADD);
1668 static int tx_queue_change_owner(struct net_device *ndev, int index,
1669 kuid_t kuid, kgid_t kgid)
1671 struct netdev_queue *queue = ndev->_tx + index;
1672 struct kobject *kobj = &queue->kobj;
1675 error = sysfs_change_owner(kobj, kuid, kgid);
1680 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1684 #endif /* CONFIG_SYSFS */
1687 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1693 for (i = old_num; i < new_num; i++) {
1694 error = netdev_queue_add_kobject(dev, i);
1701 while (--i >= new_num) {
1702 struct netdev_queue *queue = dev->_tx + i;
1704 if (!refcount_read(&dev_net(dev)->count))
1705 queue->kobj.uevent_suppress = 1;
1707 sysfs_remove_group(&queue->kobj, &dql_group);
1709 kobject_put(&queue->kobj);
1715 #endif /* CONFIG_SYSFS */
1718 static int net_tx_queue_change_owner(struct net_device *dev, int num,
1719 kuid_t kuid, kgid_t kgid)
1725 for (i = 0; i < num; i++) {
1726 error = tx_queue_change_owner(dev, i, kuid, kgid);
1734 #endif /* CONFIG_SYSFS */
1737 static int register_queue_kobjects(struct net_device *dev)
1739 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1742 dev->queues_kset = kset_create_and_add("queues",
1743 NULL, &dev->dev.kobj);
1744 if (!dev->queues_kset)
1746 real_rx = dev->real_num_rx_queues;
1748 real_tx = dev->real_num_tx_queues;
1750 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1755 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1763 netdev_queue_update_kobjects(dev, txq, 0);
1764 net_rx_queue_update_kobjects(dev, rxq, 0);
1766 kset_unregister(dev->queues_kset);
1771 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1773 int error = 0, real_rx = 0, real_tx = 0;
1776 if (ndev->queues_kset) {
1777 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1781 real_rx = ndev->real_num_rx_queues;
1783 real_tx = ndev->real_num_tx_queues;
1785 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1789 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1796 static void remove_queue_kobjects(struct net_device *dev)
1798 int real_rx = 0, real_tx = 0;
1801 real_rx = dev->real_num_rx_queues;
1803 real_tx = dev->real_num_tx_queues;
1805 net_rx_queue_update_kobjects(dev, real_rx, 0);
1806 netdev_queue_update_kobjects(dev, real_tx, 0);
1808 dev->real_num_rx_queues = 0;
1809 dev->real_num_tx_queues = 0;
1811 kset_unregister(dev->queues_kset);
1815 static bool net_current_may_mount(void)
1817 struct net *net = current->nsproxy->net_ns;
1819 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1822 static void *net_grab_current_ns(void)
1824 struct net *ns = current->nsproxy->net_ns;
1825 #ifdef CONFIG_NET_NS
1827 refcount_inc(&ns->passive);
1832 static const void *net_initial_ns(void)
1837 static const void *net_netlink_ns(struct sock *sk)
1839 return sock_net(sk);
1842 const struct kobj_ns_type_operations net_ns_type_operations = {
1843 .type = KOBJ_NS_TYPE_NET,
1844 .current_may_mount = net_current_may_mount,
1845 .grab_current_ns = net_grab_current_ns,
1846 .netlink_ns = net_netlink_ns,
1847 .initial_ns = net_initial_ns,
1848 .drop_ns = net_drop_ns,
1850 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1852 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1854 struct net_device *dev = to_net_dev(d);
1857 /* pass interface to uevent. */
1858 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1862 /* pass ifindex to uevent.
1863 * ifindex is useful as it won't change (interface name may change)
1864 * and is what RtNetlink uses natively.
1866 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1873 * netdev_release -- destroy and free a dead device.
1874 * Called when last reference to device kobject is gone.
1876 static void netdev_release(struct device *d)
1878 struct net_device *dev = to_net_dev(d);
1880 BUG_ON(dev->reg_state != NETREG_RELEASED);
1882 /* no need to wait for rcu grace period:
1883 * device is dead and about to be freed.
1885 kfree(rcu_access_pointer(dev->ifalias));
1886 netdev_freemem(dev);
1889 static const void *net_namespace(struct device *d)
1891 struct net_device *dev = to_net_dev(d);
1893 return dev_net(dev);
1896 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1898 struct net_device *dev = to_net_dev(d);
1899 const struct net *net = dev_net(dev);
1901 net_ns_get_ownership(net, uid, gid);
1904 static struct class net_class __ro_after_init = {
1906 .dev_release = netdev_release,
1907 .dev_groups = net_class_groups,
1908 .dev_uevent = netdev_uevent,
1909 .ns_type = &net_ns_type_operations,
1910 .namespace = net_namespace,
1911 .get_ownership = net_get_ownership,
1914 #ifdef CONFIG_OF_NET
1915 static int of_dev_node_match(struct device *dev, const void *data)
1917 for (; dev; dev = dev->parent) {
1918 if (dev->of_node == data)
1926 * of_find_net_device_by_node - lookup the net device for the device node
1927 * @np: OF device node
1929 * Looks up the net_device structure corresponding with the device node.
1930 * If successful, returns a pointer to the net_device with the embedded
1931 * struct device refcount incremented by one, or NULL on failure. The
1932 * refcount must be dropped when done with the net_device.
1934 struct net_device *of_find_net_device_by_node(struct device_node *np)
1938 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1942 return to_net_dev(dev);
1944 EXPORT_SYMBOL(of_find_net_device_by_node);
1947 /* Delete sysfs entries but hold kobject reference until after all
1948 * netdev references are gone.
1950 void netdev_unregister_kobject(struct net_device *ndev)
1952 struct device *dev = &ndev->dev;
1954 if (!refcount_read(&dev_net(ndev)->count))
1955 dev_set_uevent_suppress(dev, 1);
1957 kobject_get(&dev->kobj);
1959 remove_queue_kobjects(ndev);
1961 pm_runtime_set_memalloc_noio(dev, false);
1966 /* Create sysfs entries for network device. */
1967 int netdev_register_kobject(struct net_device *ndev)
1969 struct device *dev = &ndev->dev;
1970 const struct attribute_group **groups = ndev->sysfs_groups;
1973 device_initialize(dev);
1974 dev->class = &net_class;
1975 dev->platform_data = ndev;
1976 dev->groups = groups;
1978 dev_set_name(dev, "%s", ndev->name);
1981 /* Allow for a device specific group */
1985 *groups++ = &netstat_group;
1987 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1988 if (ndev->ieee80211_ptr)
1989 *groups++ = &wireless_group;
1990 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1991 else if (ndev->wireless_handlers)
1992 *groups++ = &wireless_group;
1995 #endif /* CONFIG_SYSFS */
1997 error = device_add(dev);
2001 error = register_queue_kobjects(ndev);
2007 pm_runtime_set_memalloc_noio(dev, true);
2012 /* Change owner for sysfs entries when moving network devices across network
2013 * namespaces owned by different user namespaces.
2015 int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2016 const struct net *net_new)
2018 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2019 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2020 struct device *dev = &ndev->dev;
2023 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2024 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2026 /* The network namespace was changed but the owning user namespace is
2027 * identical so there's no need to change the owner of sysfs entries.
2029 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2032 error = device_change_owner(dev, new_uid, new_gid);
2036 error = queue_change_owner(ndev, new_uid, new_gid);
2043 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2046 return class_create_file_ns(&net_class, class_attr, ns);
2048 EXPORT_SYMBOL(netdev_class_create_file_ns);
2050 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2053 class_remove_file_ns(&net_class, class_attr, ns);
2055 EXPORT_SYMBOL(netdev_class_remove_file_ns);
2057 int __init netdev_kobject_init(void)
2059 kobj_ns_type_register(&net_ns_type_operations);
2060 return class_register(&net_class);