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)) {
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 ssize_t threaded_show(struct device *dev,
591 struct device_attribute *attr, char *buf)
593 struct net_device *netdev = to_net_dev(dev);
594 ssize_t ret = -EINVAL;
597 return restart_syscall();
599 if (dev_isalive(netdev))
600 ret = sprintf(buf, fmt_dec, netdev->threaded);
606 static int modify_napi_threaded(struct net_device *dev, unsigned long val)
610 if (list_empty(&dev->napi_list))
613 if (val != 0 && val != 1)
616 ret = dev_set_threaded(dev, val);
621 static ssize_t threaded_store(struct device *dev,
622 struct device_attribute *attr,
623 const char *buf, size_t len)
625 return netdev_store(dev, attr, buf, len, modify_napi_threaded);
627 static DEVICE_ATTR_RW(threaded);
629 static struct attribute *net_class_attrs[] __ro_after_init = {
630 &dev_attr_netdev_group.attr,
632 &dev_attr_dev_id.attr,
633 &dev_attr_dev_port.attr,
634 &dev_attr_iflink.attr,
635 &dev_attr_ifindex.attr,
636 &dev_attr_name_assign_type.attr,
637 &dev_attr_addr_assign_type.attr,
638 &dev_attr_addr_len.attr,
639 &dev_attr_link_mode.attr,
640 &dev_attr_address.attr,
641 &dev_attr_broadcast.attr,
642 &dev_attr_speed.attr,
643 &dev_attr_duplex.attr,
644 &dev_attr_dormant.attr,
645 &dev_attr_testing.attr,
646 &dev_attr_operstate.attr,
647 &dev_attr_carrier_changes.attr,
648 &dev_attr_ifalias.attr,
649 &dev_attr_carrier.attr,
651 &dev_attr_flags.attr,
652 &dev_attr_tx_queue_len.attr,
653 &dev_attr_gro_flush_timeout.attr,
654 &dev_attr_napi_defer_hard_irqs.attr,
655 &dev_attr_phys_port_id.attr,
656 &dev_attr_phys_port_name.attr,
657 &dev_attr_phys_switch_id.attr,
658 &dev_attr_proto_down.attr,
659 &dev_attr_carrier_up_count.attr,
660 &dev_attr_carrier_down_count.attr,
661 &dev_attr_threaded.attr,
664 ATTRIBUTE_GROUPS(net_class);
666 /* Show a given an attribute in the statistics group */
667 static ssize_t netstat_show(const struct device *d,
668 struct device_attribute *attr, char *buf,
669 unsigned long offset)
671 struct net_device *dev = to_net_dev(d);
672 ssize_t ret = -EINVAL;
674 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
675 offset % sizeof(u64) != 0);
677 read_lock(&dev_base_lock);
678 if (dev_isalive(dev)) {
679 struct rtnl_link_stats64 temp;
680 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
682 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
684 read_unlock(&dev_base_lock);
688 /* generate a read-only statistics attribute */
689 #define NETSTAT_ENTRY(name) \
690 static ssize_t name##_show(struct device *d, \
691 struct device_attribute *attr, char *buf) \
693 return netstat_show(d, attr, buf, \
694 offsetof(struct rtnl_link_stats64, name)); \
696 static DEVICE_ATTR_RO(name)
698 NETSTAT_ENTRY(rx_packets);
699 NETSTAT_ENTRY(tx_packets);
700 NETSTAT_ENTRY(rx_bytes);
701 NETSTAT_ENTRY(tx_bytes);
702 NETSTAT_ENTRY(rx_errors);
703 NETSTAT_ENTRY(tx_errors);
704 NETSTAT_ENTRY(rx_dropped);
705 NETSTAT_ENTRY(tx_dropped);
706 NETSTAT_ENTRY(multicast);
707 NETSTAT_ENTRY(collisions);
708 NETSTAT_ENTRY(rx_length_errors);
709 NETSTAT_ENTRY(rx_over_errors);
710 NETSTAT_ENTRY(rx_crc_errors);
711 NETSTAT_ENTRY(rx_frame_errors);
712 NETSTAT_ENTRY(rx_fifo_errors);
713 NETSTAT_ENTRY(rx_missed_errors);
714 NETSTAT_ENTRY(tx_aborted_errors);
715 NETSTAT_ENTRY(tx_carrier_errors);
716 NETSTAT_ENTRY(tx_fifo_errors);
717 NETSTAT_ENTRY(tx_heartbeat_errors);
718 NETSTAT_ENTRY(tx_window_errors);
719 NETSTAT_ENTRY(rx_compressed);
720 NETSTAT_ENTRY(tx_compressed);
721 NETSTAT_ENTRY(rx_nohandler);
723 static struct attribute *netstat_attrs[] __ro_after_init = {
724 &dev_attr_rx_packets.attr,
725 &dev_attr_tx_packets.attr,
726 &dev_attr_rx_bytes.attr,
727 &dev_attr_tx_bytes.attr,
728 &dev_attr_rx_errors.attr,
729 &dev_attr_tx_errors.attr,
730 &dev_attr_rx_dropped.attr,
731 &dev_attr_tx_dropped.attr,
732 &dev_attr_multicast.attr,
733 &dev_attr_collisions.attr,
734 &dev_attr_rx_length_errors.attr,
735 &dev_attr_rx_over_errors.attr,
736 &dev_attr_rx_crc_errors.attr,
737 &dev_attr_rx_frame_errors.attr,
738 &dev_attr_rx_fifo_errors.attr,
739 &dev_attr_rx_missed_errors.attr,
740 &dev_attr_tx_aborted_errors.attr,
741 &dev_attr_tx_carrier_errors.attr,
742 &dev_attr_tx_fifo_errors.attr,
743 &dev_attr_tx_heartbeat_errors.attr,
744 &dev_attr_tx_window_errors.attr,
745 &dev_attr_rx_compressed.attr,
746 &dev_attr_tx_compressed.attr,
747 &dev_attr_rx_nohandler.attr,
751 static const struct attribute_group netstat_group = {
752 .name = "statistics",
753 .attrs = netstat_attrs,
756 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
757 static struct attribute *wireless_attrs[] = {
761 static const struct attribute_group wireless_group = {
763 .attrs = wireless_attrs,
767 #else /* CONFIG_SYSFS */
768 #define net_class_groups NULL
769 #endif /* CONFIG_SYSFS */
772 #define to_rx_queue_attr(_attr) \
773 container_of(_attr, struct rx_queue_attribute, attr)
775 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
777 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
780 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
781 struct netdev_rx_queue *queue = to_rx_queue(kobj);
783 if (!attribute->show)
786 return attribute->show(queue, buf);
789 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
790 const char *buf, size_t count)
792 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
793 struct netdev_rx_queue *queue = to_rx_queue(kobj);
795 if (!attribute->store)
798 return attribute->store(queue, buf, count);
801 static const struct sysfs_ops rx_queue_sysfs_ops = {
802 .show = rx_queue_attr_show,
803 .store = rx_queue_attr_store,
807 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
813 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
817 map = rcu_dereference(queue->rps_map);
819 for (i = 0; i < map->len; i++)
820 cpumask_set_cpu(map->cpus[i], mask);
822 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
824 free_cpumask_var(mask);
826 return len < PAGE_SIZE ? len : -EINVAL;
829 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
830 const char *buf, size_t len)
832 struct rps_map *old_map, *map;
834 int err, cpu, i, hk_flags;
835 static DEFINE_MUTEX(rps_map_mutex);
837 if (!capable(CAP_NET_ADMIN))
840 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
843 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
845 free_cpumask_var(mask);
849 if (!cpumask_empty(mask)) {
850 hk_flags = HK_FLAG_DOMAIN | HK_FLAG_WQ;
851 cpumask_and(mask, mask, housekeeping_cpumask(hk_flags));
852 if (cpumask_empty(mask)) {
853 free_cpumask_var(mask);
858 map = kzalloc(max_t(unsigned int,
859 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
862 free_cpumask_var(mask);
867 for_each_cpu_and(cpu, mask, cpu_online_mask)
868 map->cpus[i++] = cpu;
877 mutex_lock(&rps_map_mutex);
878 old_map = rcu_dereference_protected(queue->rps_map,
879 mutex_is_locked(&rps_map_mutex));
880 rcu_assign_pointer(queue->rps_map, map);
883 static_branch_inc(&rps_needed);
885 static_branch_dec(&rps_needed);
887 mutex_unlock(&rps_map_mutex);
890 kfree_rcu(old_map, rcu);
892 free_cpumask_var(mask);
896 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
899 struct rps_dev_flow_table *flow_table;
900 unsigned long val = 0;
903 flow_table = rcu_dereference(queue->rps_flow_table);
905 val = (unsigned long)flow_table->mask + 1;
908 return sprintf(buf, "%lu\n", val);
911 static void rps_dev_flow_table_release(struct rcu_head *rcu)
913 struct rps_dev_flow_table *table = container_of(rcu,
914 struct rps_dev_flow_table, rcu);
918 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
919 const char *buf, size_t len)
921 unsigned long mask, count;
922 struct rps_dev_flow_table *table, *old_table;
923 static DEFINE_SPINLOCK(rps_dev_flow_lock);
926 if (!capable(CAP_NET_ADMIN))
929 rc = kstrtoul(buf, 0, &count);
935 /* mask = roundup_pow_of_two(count) - 1;
936 * without overflows...
938 while ((mask | (mask >> 1)) != mask)
940 /* On 64 bit arches, must check mask fits in table->mask (u32),
941 * and on 32bit arches, must check
942 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
944 #if BITS_PER_LONG > 32
945 if (mask > (unsigned long)(u32)mask)
948 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
949 / sizeof(struct rps_dev_flow)) {
950 /* Enforce a limit to prevent overflow */
954 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
959 for (count = 0; count <= mask; count++)
960 table->flows[count].cpu = RPS_NO_CPU;
965 spin_lock(&rps_dev_flow_lock);
966 old_table = rcu_dereference_protected(queue->rps_flow_table,
967 lockdep_is_held(&rps_dev_flow_lock));
968 rcu_assign_pointer(queue->rps_flow_table, table);
969 spin_unlock(&rps_dev_flow_lock);
972 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
977 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
978 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
980 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
981 = __ATTR(rps_flow_cnt, 0644,
982 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
983 #endif /* CONFIG_RPS */
985 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
987 &rps_cpus_attribute.attr,
988 &rps_dev_flow_table_cnt_attribute.attr,
992 ATTRIBUTE_GROUPS(rx_queue_default);
994 static void rx_queue_release(struct kobject *kobj)
996 struct netdev_rx_queue *queue = to_rx_queue(kobj);
999 struct rps_dev_flow_table *flow_table;
1001 map = rcu_dereference_protected(queue->rps_map, 1);
1003 RCU_INIT_POINTER(queue->rps_map, NULL);
1004 kfree_rcu(map, rcu);
1007 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1009 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1010 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1014 memset(kobj, 0, sizeof(*kobj));
1015 dev_put(queue->dev);
1018 static const void *rx_queue_namespace(struct kobject *kobj)
1020 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1021 struct device *dev = &queue->dev->dev;
1022 const void *ns = NULL;
1024 if (dev->class && dev->class->ns_type)
1025 ns = dev->class->namespace(dev);
1030 static void rx_queue_get_ownership(struct kobject *kobj,
1031 kuid_t *uid, kgid_t *gid)
1033 const struct net *net = rx_queue_namespace(kobj);
1035 net_ns_get_ownership(net, uid, gid);
1038 static struct kobj_type rx_queue_ktype __ro_after_init = {
1039 .sysfs_ops = &rx_queue_sysfs_ops,
1040 .release = rx_queue_release,
1041 .default_groups = rx_queue_default_groups,
1042 .namespace = rx_queue_namespace,
1043 .get_ownership = rx_queue_get_ownership,
1046 static int rx_queue_add_kobject(struct net_device *dev, int index)
1048 struct netdev_rx_queue *queue = dev->_rx + index;
1049 struct kobject *kobj = &queue->kobj;
1052 /* Kobject_put later will trigger rx_queue_release call which
1053 * decreases dev refcount: Take that reference here
1055 dev_hold(queue->dev);
1057 kobj->kset = dev->queues_kset;
1058 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1063 if (dev->sysfs_rx_queue_group) {
1064 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1069 kobject_uevent(kobj, KOBJ_ADD);
1078 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1081 struct netdev_rx_queue *queue = dev->_rx + index;
1082 struct kobject *kobj = &queue->kobj;
1085 error = sysfs_change_owner(kobj, kuid, kgid);
1089 if (dev->sysfs_rx_queue_group)
1090 error = sysfs_group_change_owner(
1091 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1095 #endif /* CONFIG_SYSFS */
1098 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1105 if (!dev->sysfs_rx_queue_group)
1108 for (i = old_num; i < new_num; i++) {
1109 error = rx_queue_add_kobject(dev, i);
1116 while (--i >= new_num) {
1117 struct kobject *kobj = &dev->_rx[i].kobj;
1119 if (!refcount_read(&dev_net(dev)->ns.count))
1120 kobj->uevent_suppress = 1;
1121 if (dev->sysfs_rx_queue_group)
1122 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1132 static int net_rx_queue_change_owner(struct net_device *dev, int num,
1133 kuid_t kuid, kgid_t kgid)
1140 if (!dev->sysfs_rx_queue_group)
1143 for (i = 0; i < num; i++) {
1144 error = rx_queue_change_owner(dev, i, kuid, kgid);
1157 * netdev_queue sysfs structures and functions.
1159 struct netdev_queue_attribute {
1160 struct attribute attr;
1161 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1162 ssize_t (*store)(struct netdev_queue *queue,
1163 const char *buf, size_t len);
1165 #define to_netdev_queue_attr(_attr) \
1166 container_of(_attr, struct netdev_queue_attribute, attr)
1168 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1170 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1171 struct attribute *attr, char *buf)
1173 const struct netdev_queue_attribute *attribute
1174 = to_netdev_queue_attr(attr);
1175 struct netdev_queue *queue = to_netdev_queue(kobj);
1177 if (!attribute->show)
1180 return attribute->show(queue, buf);
1183 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1184 struct attribute *attr,
1185 const char *buf, size_t count)
1187 const struct netdev_queue_attribute *attribute
1188 = to_netdev_queue_attr(attr);
1189 struct netdev_queue *queue = to_netdev_queue(kobj);
1191 if (!attribute->store)
1194 return attribute->store(queue, buf, count);
1197 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1198 .show = netdev_queue_attr_show,
1199 .store = netdev_queue_attr_store,
1202 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1204 unsigned long trans_timeout;
1206 spin_lock_irq(&queue->_xmit_lock);
1207 trans_timeout = queue->trans_timeout;
1208 spin_unlock_irq(&queue->_xmit_lock);
1210 return sprintf(buf, fmt_ulong, trans_timeout);
1213 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1215 struct net_device *dev = queue->dev;
1218 i = queue - dev->_tx;
1219 BUG_ON(i >= dev->num_tx_queues);
1224 static ssize_t traffic_class_show(struct netdev_queue *queue,
1227 struct net_device *dev = queue->dev;
1231 if (!netif_is_multiqueue(dev))
1234 if (!rtnl_trylock())
1235 return restart_syscall();
1237 index = get_netdev_queue_index(queue);
1239 /* If queue belongs to subordinate dev use its TC mapping */
1240 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1242 num_tc = dev->num_tc;
1243 tc = netdev_txq_to_tc(dev, index);
1250 /* We can report the traffic class one of two ways:
1251 * Subordinate device traffic classes are reported with the traffic
1252 * class first, and then the subordinate class so for example TC0 on
1253 * subordinate device 2 will be reported as "0-2". If the queue
1254 * belongs to the root device it will be reported with just the
1255 * traffic class, so just "0" for TC 0 for example.
1257 return num_tc < 0 ? sprintf(buf, "%d%d\n", tc, num_tc) :
1258 sprintf(buf, "%d\n", tc);
1262 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1265 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1268 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1269 const char *buf, size_t len)
1271 struct net_device *dev = queue->dev;
1272 int err, index = get_netdev_queue_index(queue);
1275 if (!capable(CAP_NET_ADMIN))
1278 /* The check is also done later; this helps returning early without
1279 * hitting the trylock/restart below.
1281 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1284 err = kstrtou32(buf, 10, &rate);
1288 if (!rtnl_trylock())
1289 return restart_syscall();
1292 if (dev->netdev_ops->ndo_set_tx_maxrate)
1293 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1297 queue->tx_maxrate = rate;
1303 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1304 = __ATTR_RW(tx_maxrate);
1307 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1308 = __ATTR_RO(tx_timeout);
1310 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1311 = __ATTR_RO(traffic_class);
1315 * Byte queue limits sysfs structures and functions.
1317 static ssize_t bql_show(char *buf, unsigned int value)
1319 return sprintf(buf, "%u\n", value);
1322 static ssize_t bql_set(const char *buf, const size_t count,
1323 unsigned int *pvalue)
1328 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1329 value = DQL_MAX_LIMIT;
1331 err = kstrtouint(buf, 10, &value);
1334 if (value > DQL_MAX_LIMIT)
1343 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1346 struct dql *dql = &queue->dql;
1348 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1351 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1352 const char *buf, size_t len)
1354 struct dql *dql = &queue->dql;
1358 err = kstrtouint(buf, 10, &value);
1362 dql->slack_hold_time = msecs_to_jiffies(value);
1367 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1368 = __ATTR(hold_time, 0644,
1369 bql_show_hold_time, bql_set_hold_time);
1371 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1374 struct dql *dql = &queue->dql;
1376 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1379 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1380 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1382 #define BQL_ATTR(NAME, FIELD) \
1383 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1386 return bql_show(buf, queue->dql.FIELD); \
1389 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1390 const char *buf, size_t len) \
1392 return bql_set(buf, len, &queue->dql.FIELD); \
1395 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1396 = __ATTR(NAME, 0644, \
1397 bql_show_ ## NAME, bql_set_ ## NAME)
1399 BQL_ATTR(limit, limit);
1400 BQL_ATTR(limit_max, max_limit);
1401 BQL_ATTR(limit_min, min_limit);
1403 static struct attribute *dql_attrs[] __ro_after_init = {
1404 &bql_limit_attribute.attr,
1405 &bql_limit_max_attribute.attr,
1406 &bql_limit_min_attribute.attr,
1407 &bql_hold_time_attribute.attr,
1408 &bql_inflight_attribute.attr,
1412 static const struct attribute_group dql_group = {
1413 .name = "byte_queue_limits",
1416 #endif /* CONFIG_BQL */
1419 static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1420 int tc, char *buf, enum xps_map_type type)
1422 struct xps_dev_maps *dev_maps;
1423 unsigned long *mask;
1424 unsigned int nr_ids;
1428 dev_maps = rcu_dereference(dev->xps_maps[type]);
1430 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1431 * when dev_maps hasn't been allocated yet, to be backward compatible.
1433 nr_ids = dev_maps ? dev_maps->nr_ids :
1434 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1436 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1442 if (!dev_maps || tc >= dev_maps->num_tc)
1445 for (j = 0; j < nr_ids; j++) {
1446 int i, tci = j * dev_maps->num_tc + tc;
1447 struct xps_map *map;
1449 map = rcu_dereference(dev_maps->attr_map[tci]);
1453 for (i = map->len; i--;) {
1454 if (map->queues[i] == index) {
1463 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1466 return len < PAGE_SIZE ? len : -EINVAL;
1469 static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1471 struct net_device *dev = queue->dev;
1475 if (!netif_is_multiqueue(dev))
1478 index = get_netdev_queue_index(queue);
1480 if (!rtnl_trylock())
1481 return restart_syscall();
1483 /* If queue belongs to subordinate dev use its map */
1484 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1486 tc = netdev_txq_to_tc(dev, index);
1492 /* Make sure the subordinate device can't be freed */
1493 get_device(&dev->dev);
1496 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1498 put_device(&dev->dev);
1502 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1503 const char *buf, size_t len)
1505 struct net_device *dev = queue->dev;
1510 if (!netif_is_multiqueue(dev))
1513 if (!capable(CAP_NET_ADMIN))
1516 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1519 index = get_netdev_queue_index(queue);
1521 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1523 free_cpumask_var(mask);
1527 if (!rtnl_trylock()) {
1528 free_cpumask_var(mask);
1529 return restart_syscall();
1532 err = netif_set_xps_queue(dev, mask, index);
1535 free_cpumask_var(mask);
1540 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1541 = __ATTR_RW(xps_cpus);
1543 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1545 struct net_device *dev = queue->dev;
1549 index = get_netdev_queue_index(queue);
1551 if (!rtnl_trylock())
1552 return restart_syscall();
1554 tc = netdev_txq_to_tc(dev, index);
1559 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1562 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1565 struct net_device *dev = queue->dev;
1566 struct net *net = dev_net(dev);
1567 unsigned long *mask;
1571 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1574 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1578 index = get_netdev_queue_index(queue);
1580 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1586 if (!rtnl_trylock()) {
1588 return restart_syscall();
1592 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1601 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1602 = __ATTR_RW(xps_rxqs);
1603 #endif /* CONFIG_XPS */
1605 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1606 &queue_trans_timeout.attr,
1607 &queue_traffic_class.attr,
1609 &xps_cpus_attribute.attr,
1610 &xps_rxqs_attribute.attr,
1611 &queue_tx_maxrate.attr,
1615 ATTRIBUTE_GROUPS(netdev_queue_default);
1617 static void netdev_queue_release(struct kobject *kobj)
1619 struct netdev_queue *queue = to_netdev_queue(kobj);
1621 memset(kobj, 0, sizeof(*kobj));
1622 dev_put(queue->dev);
1625 static const void *netdev_queue_namespace(struct kobject *kobj)
1627 struct netdev_queue *queue = to_netdev_queue(kobj);
1628 struct device *dev = &queue->dev->dev;
1629 const void *ns = NULL;
1631 if (dev->class && dev->class->ns_type)
1632 ns = dev->class->namespace(dev);
1637 static void netdev_queue_get_ownership(struct kobject *kobj,
1638 kuid_t *uid, kgid_t *gid)
1640 const struct net *net = netdev_queue_namespace(kobj);
1642 net_ns_get_ownership(net, uid, gid);
1645 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1646 .sysfs_ops = &netdev_queue_sysfs_ops,
1647 .release = netdev_queue_release,
1648 .default_groups = netdev_queue_default_groups,
1649 .namespace = netdev_queue_namespace,
1650 .get_ownership = netdev_queue_get_ownership,
1653 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1655 struct netdev_queue *queue = dev->_tx + index;
1656 struct kobject *kobj = &queue->kobj;
1659 /* Kobject_put later will trigger netdev_queue_release call
1660 * which decreases dev refcount: Take that reference here
1662 dev_hold(queue->dev);
1664 kobj->kset = dev->queues_kset;
1665 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1671 error = sysfs_create_group(kobj, &dql_group);
1676 kobject_uevent(kobj, KOBJ_ADD);
1684 static int tx_queue_change_owner(struct net_device *ndev, int index,
1685 kuid_t kuid, kgid_t kgid)
1687 struct netdev_queue *queue = ndev->_tx + index;
1688 struct kobject *kobj = &queue->kobj;
1691 error = sysfs_change_owner(kobj, kuid, kgid);
1696 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1700 #endif /* CONFIG_SYSFS */
1703 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1709 for (i = old_num; i < new_num; i++) {
1710 error = netdev_queue_add_kobject(dev, i);
1717 while (--i >= new_num) {
1718 struct netdev_queue *queue = dev->_tx + i;
1720 if (!refcount_read(&dev_net(dev)->ns.count))
1721 queue->kobj.uevent_suppress = 1;
1723 sysfs_remove_group(&queue->kobj, &dql_group);
1725 kobject_put(&queue->kobj);
1731 #endif /* CONFIG_SYSFS */
1734 static int net_tx_queue_change_owner(struct net_device *dev, int num,
1735 kuid_t kuid, kgid_t kgid)
1741 for (i = 0; i < num; i++) {
1742 error = tx_queue_change_owner(dev, i, kuid, kgid);
1750 #endif /* CONFIG_SYSFS */
1753 static int register_queue_kobjects(struct net_device *dev)
1755 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1758 dev->queues_kset = kset_create_and_add("queues",
1759 NULL, &dev->dev.kobj);
1760 if (!dev->queues_kset)
1762 real_rx = dev->real_num_rx_queues;
1764 real_tx = dev->real_num_tx_queues;
1766 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1771 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1779 netdev_queue_update_kobjects(dev, txq, 0);
1780 net_rx_queue_update_kobjects(dev, rxq, 0);
1782 kset_unregister(dev->queues_kset);
1787 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1789 int error = 0, real_rx = 0, real_tx = 0;
1792 if (ndev->queues_kset) {
1793 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1797 real_rx = ndev->real_num_rx_queues;
1799 real_tx = ndev->real_num_tx_queues;
1801 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1805 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1812 static void remove_queue_kobjects(struct net_device *dev)
1814 int real_rx = 0, real_tx = 0;
1817 real_rx = dev->real_num_rx_queues;
1819 real_tx = dev->real_num_tx_queues;
1821 net_rx_queue_update_kobjects(dev, real_rx, 0);
1822 netdev_queue_update_kobjects(dev, real_tx, 0);
1824 kset_unregister(dev->queues_kset);
1828 static bool net_current_may_mount(void)
1830 struct net *net = current->nsproxy->net_ns;
1832 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1835 static void *net_grab_current_ns(void)
1837 struct net *ns = current->nsproxy->net_ns;
1838 #ifdef CONFIG_NET_NS
1840 refcount_inc(&ns->passive);
1845 static const void *net_initial_ns(void)
1850 static const void *net_netlink_ns(struct sock *sk)
1852 return sock_net(sk);
1855 const struct kobj_ns_type_operations net_ns_type_operations = {
1856 .type = KOBJ_NS_TYPE_NET,
1857 .current_may_mount = net_current_may_mount,
1858 .grab_current_ns = net_grab_current_ns,
1859 .netlink_ns = net_netlink_ns,
1860 .initial_ns = net_initial_ns,
1861 .drop_ns = net_drop_ns,
1863 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1865 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1867 struct net_device *dev = to_net_dev(d);
1870 /* pass interface to uevent. */
1871 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1875 /* pass ifindex to uevent.
1876 * ifindex is useful as it won't change (interface name may change)
1877 * and is what RtNetlink uses natively.
1879 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1886 * netdev_release -- destroy and free a dead device.
1887 * Called when last reference to device kobject is gone.
1889 static void netdev_release(struct device *d)
1891 struct net_device *dev = to_net_dev(d);
1893 BUG_ON(dev->reg_state != NETREG_RELEASED);
1895 /* no need to wait for rcu grace period:
1896 * device is dead and about to be freed.
1898 kfree(rcu_access_pointer(dev->ifalias));
1899 netdev_freemem(dev);
1902 static const void *net_namespace(struct device *d)
1904 struct net_device *dev = to_net_dev(d);
1906 return dev_net(dev);
1909 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1911 struct net_device *dev = to_net_dev(d);
1912 const struct net *net = dev_net(dev);
1914 net_ns_get_ownership(net, uid, gid);
1917 static struct class net_class __ro_after_init = {
1919 .dev_release = netdev_release,
1920 .dev_groups = net_class_groups,
1921 .dev_uevent = netdev_uevent,
1922 .ns_type = &net_ns_type_operations,
1923 .namespace = net_namespace,
1924 .get_ownership = net_get_ownership,
1928 static int of_dev_node_match(struct device *dev, const void *data)
1930 for (; dev; dev = dev->parent) {
1931 if (dev->of_node == data)
1939 * of_find_net_device_by_node - lookup the net device for the device node
1940 * @np: OF device node
1942 * Looks up the net_device structure corresponding with the device node.
1943 * If successful, returns a pointer to the net_device with the embedded
1944 * struct device refcount incremented by one, or NULL on failure. The
1945 * refcount must be dropped when done with the net_device.
1947 struct net_device *of_find_net_device_by_node(struct device_node *np)
1951 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1955 return to_net_dev(dev);
1957 EXPORT_SYMBOL(of_find_net_device_by_node);
1960 /* Delete sysfs entries but hold kobject reference until after all
1961 * netdev references are gone.
1963 void netdev_unregister_kobject(struct net_device *ndev)
1965 struct device *dev = &ndev->dev;
1967 if (!refcount_read(&dev_net(ndev)->ns.count))
1968 dev_set_uevent_suppress(dev, 1);
1970 kobject_get(&dev->kobj);
1972 remove_queue_kobjects(ndev);
1974 pm_runtime_set_memalloc_noio(dev, false);
1979 /* Create sysfs entries for network device. */
1980 int netdev_register_kobject(struct net_device *ndev)
1982 struct device *dev = &ndev->dev;
1983 const struct attribute_group **groups = ndev->sysfs_groups;
1986 device_initialize(dev);
1987 dev->class = &net_class;
1988 dev->platform_data = ndev;
1989 dev->groups = groups;
1991 dev_set_name(dev, "%s", ndev->name);
1994 /* Allow for a device specific group */
1998 *groups++ = &netstat_group;
2000 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
2001 if (ndev->ieee80211_ptr)
2002 *groups++ = &wireless_group;
2003 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
2004 else if (ndev->wireless_handlers)
2005 *groups++ = &wireless_group;
2008 #endif /* CONFIG_SYSFS */
2010 error = device_add(dev);
2014 error = register_queue_kobjects(ndev);
2020 pm_runtime_set_memalloc_noio(dev, true);
2025 /* Change owner for sysfs entries when moving network devices across network
2026 * namespaces owned by different user namespaces.
2028 int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2029 const struct net *net_new)
2031 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2032 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2033 struct device *dev = &ndev->dev;
2036 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2037 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2039 /* The network namespace was changed but the owning user namespace is
2040 * identical so there's no need to change the owner of sysfs entries.
2042 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2045 error = device_change_owner(dev, new_uid, new_gid);
2049 error = queue_change_owner(ndev, new_uid, new_gid);
2056 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2059 return class_create_file_ns(&net_class, class_attr, ns);
2061 EXPORT_SYMBOL(netdev_class_create_file_ns);
2063 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2066 class_remove_file_ns(&net_class, class_attr, ns);
2068 EXPORT_SYMBOL(netdev_class_remove_file_ns);
2070 int __init netdev_kobject_init(void)
2072 kobj_ns_type_register(&net_ns_type_operations);
2073 return class_register(&net_class);