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 /* Caller holds RTNL or dev_base_lock */
36 static inline int dev_isalive(const struct net_device *dev)
38 return dev->reg_state <= NETREG_REGISTERED;
41 /* use same locking rules as GIF* ioctl's */
42 static ssize_t netdev_show(const struct device *dev,
43 struct device_attribute *attr, char *buf,
44 ssize_t (*format)(const struct net_device *, char *))
46 struct net_device *ndev = to_net_dev(dev);
47 ssize_t ret = -EINVAL;
49 read_lock(&dev_base_lock);
50 if (dev_isalive(ndev))
51 ret = (*format)(ndev, buf);
52 read_unlock(&dev_base_lock);
57 /* generate a show function for simple field */
58 #define NETDEVICE_SHOW(field, format_string) \
59 static ssize_t format_##field(const struct net_device *dev, char *buf) \
61 return sprintf(buf, format_string, dev->field); \
63 static ssize_t field##_show(struct device *dev, \
64 struct device_attribute *attr, char *buf) \
66 return netdev_show(dev, attr, buf, format_##field); \
69 #define NETDEVICE_SHOW_RO(field, format_string) \
70 NETDEVICE_SHOW(field, format_string); \
71 static DEVICE_ATTR_RO(field)
73 #define NETDEVICE_SHOW_RW(field, format_string) \
74 NETDEVICE_SHOW(field, format_string); \
75 static DEVICE_ATTR_RW(field)
77 /* use same locking and permission rules as SIF* ioctl's */
78 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
79 const char *buf, size_t len,
80 int (*set)(struct net_device *, unsigned long))
82 struct net_device *netdev = to_net_dev(dev);
83 struct net *net = dev_net(netdev);
87 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
90 ret = kstrtoul(buf, 0, &new);
95 return restart_syscall();
97 if (dev_isalive(netdev)) {
98 ret = (*set)(netdev, new);
107 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
108 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
109 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
110 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
111 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
112 NETDEVICE_SHOW_RO(type, fmt_dec);
113 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
115 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
118 struct net_device *ndev = to_net_dev(dev);
120 return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
122 static DEVICE_ATTR_RO(iflink);
124 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
126 return sprintf(buf, fmt_dec, dev->name_assign_type);
129 static ssize_t name_assign_type_show(struct device *dev,
130 struct device_attribute *attr,
133 struct net_device *ndev = to_net_dev(dev);
134 ssize_t ret = -EINVAL;
136 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
137 ret = netdev_show(dev, attr, buf, format_name_assign_type);
141 static DEVICE_ATTR_RO(name_assign_type);
143 /* use same locking rules as GIFHWADDR ioctl's */
144 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
147 struct net_device *ndev = to_net_dev(dev);
148 ssize_t ret = -EINVAL;
150 read_lock(&dev_base_lock);
151 if (dev_isalive(ndev))
152 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
153 read_unlock(&dev_base_lock);
156 static DEVICE_ATTR_RO(address);
158 static ssize_t broadcast_show(struct device *dev,
159 struct device_attribute *attr, char *buf)
161 struct net_device *ndev = to_net_dev(dev);
163 if (dev_isalive(ndev))
164 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
167 static DEVICE_ATTR_RO(broadcast);
169 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
171 if (!netif_running(dev))
173 return dev_change_carrier(dev, (bool)new_carrier);
176 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
177 const char *buf, size_t len)
179 struct net_device *netdev = to_net_dev(dev);
181 /* The check is also done in change_carrier; this helps returning early
182 * without hitting the trylock/restart in netdev_store.
184 if (!netdev->netdev_ops->ndo_change_carrier)
187 return netdev_store(dev, attr, buf, len, change_carrier);
190 static ssize_t carrier_show(struct device *dev,
191 struct device_attribute *attr, char *buf)
193 struct net_device *netdev = to_net_dev(dev);
195 if (netif_running(netdev))
196 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
200 static DEVICE_ATTR_RW(carrier);
202 static ssize_t speed_show(struct device *dev,
203 struct device_attribute *attr, char *buf)
205 struct net_device *netdev = to_net_dev(dev);
208 /* The check is also done in __ethtool_get_link_ksettings; this helps
209 * returning early without hitting the trylock/restart below.
211 if (!netdev->ethtool_ops->get_link_ksettings)
215 return restart_syscall();
217 if (netif_running(netdev) && netif_device_present(netdev)) {
218 struct ethtool_link_ksettings cmd;
220 if (!__ethtool_get_link_ksettings(netdev, &cmd))
221 ret = sprintf(buf, fmt_dec, cmd.base.speed);
226 static DEVICE_ATTR_RO(speed);
228 static ssize_t duplex_show(struct device *dev,
229 struct device_attribute *attr, char *buf)
231 struct net_device *netdev = to_net_dev(dev);
234 /* The check is also done in __ethtool_get_link_ksettings; this helps
235 * returning early without hitting the trylock/restart below.
237 if (!netdev->ethtool_ops->get_link_ksettings)
241 return restart_syscall();
243 if (netif_running(netdev)) {
244 struct ethtool_link_ksettings cmd;
246 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
249 switch (cmd.base.duplex) {
260 ret = sprintf(buf, "%s\n", duplex);
266 static DEVICE_ATTR_RO(duplex);
268 static ssize_t testing_show(struct device *dev,
269 struct device_attribute *attr, char *buf)
271 struct net_device *netdev = to_net_dev(dev);
273 if (netif_running(netdev))
274 return sprintf(buf, fmt_dec, !!netif_testing(netdev));
278 static DEVICE_ATTR_RO(testing);
280 static ssize_t dormant_show(struct device *dev,
281 struct device_attribute *attr, char *buf)
283 struct net_device *netdev = to_net_dev(dev);
285 if (netif_running(netdev))
286 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
290 static DEVICE_ATTR_RO(dormant);
292 static const char *const operstates[] = {
294 "notpresent", /* currently unused */
302 static ssize_t operstate_show(struct device *dev,
303 struct device_attribute *attr, char *buf)
305 const struct net_device *netdev = to_net_dev(dev);
306 unsigned char operstate;
308 read_lock(&dev_base_lock);
309 operstate = netdev->operstate;
310 if (!netif_running(netdev))
311 operstate = IF_OPER_DOWN;
312 read_unlock(&dev_base_lock);
314 if (operstate >= ARRAY_SIZE(operstates))
315 return -EINVAL; /* should not happen */
317 return sprintf(buf, "%s\n", operstates[operstate]);
319 static DEVICE_ATTR_RO(operstate);
321 static ssize_t carrier_changes_show(struct device *dev,
322 struct device_attribute *attr,
325 struct net_device *netdev = to_net_dev(dev);
327 return sprintf(buf, fmt_dec,
328 atomic_read(&netdev->carrier_up_count) +
329 atomic_read(&netdev->carrier_down_count));
331 static DEVICE_ATTR_RO(carrier_changes);
333 static ssize_t carrier_up_count_show(struct device *dev,
334 struct device_attribute *attr,
337 struct net_device *netdev = to_net_dev(dev);
339 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
341 static DEVICE_ATTR_RO(carrier_up_count);
343 static ssize_t carrier_down_count_show(struct device *dev,
344 struct device_attribute *attr,
347 struct net_device *netdev = to_net_dev(dev);
349 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
351 static DEVICE_ATTR_RO(carrier_down_count);
353 /* read-write attributes */
355 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
357 return dev_set_mtu(dev, (int)new_mtu);
360 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
361 const char *buf, size_t len)
363 return netdev_store(dev, attr, buf, len, change_mtu);
365 NETDEVICE_SHOW_RW(mtu, fmt_dec);
367 static int change_flags(struct net_device *dev, unsigned long new_flags)
369 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
372 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
373 const char *buf, size_t len)
375 return netdev_store(dev, attr, buf, len, change_flags);
377 NETDEVICE_SHOW_RW(flags, fmt_hex);
379 static ssize_t tx_queue_len_store(struct device *dev,
380 struct device_attribute *attr,
381 const char *buf, size_t len)
383 if (!capable(CAP_NET_ADMIN))
386 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
388 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
390 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
392 WRITE_ONCE(dev->gro_flush_timeout, val);
396 static ssize_t gro_flush_timeout_store(struct device *dev,
397 struct device_attribute *attr,
398 const char *buf, size_t len)
400 if (!capable(CAP_NET_ADMIN))
403 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
405 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
407 static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
409 WRITE_ONCE(dev->napi_defer_hard_irqs, val);
413 static ssize_t napi_defer_hard_irqs_store(struct device *dev,
414 struct device_attribute *attr,
415 const char *buf, size_t len)
417 if (!capable(CAP_NET_ADMIN))
420 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
422 NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
424 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
425 const char *buf, size_t len)
427 struct net_device *netdev = to_net_dev(dev);
428 struct net *net = dev_net(netdev);
432 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
435 /* ignore trailing newline */
436 if (len > 0 && buf[len - 1] == '\n')
440 return restart_syscall();
442 if (dev_isalive(netdev)) {
443 ret = dev_set_alias(netdev, buf, count);
447 netdev_state_change(netdev);
455 static ssize_t ifalias_show(struct device *dev,
456 struct device_attribute *attr, char *buf)
458 const struct net_device *netdev = to_net_dev(dev);
462 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
464 ret = sprintf(buf, "%s\n", tmp);
467 static DEVICE_ATTR_RW(ifalias);
469 static int change_group(struct net_device *dev, unsigned long new_group)
471 dev_set_group(dev, (int)new_group);
475 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
476 const char *buf, size_t len)
478 return netdev_store(dev, attr, buf, len, change_group);
480 NETDEVICE_SHOW(group, fmt_dec);
481 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
483 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
485 return dev_change_proto_down(dev, (bool)proto_down);
488 static ssize_t proto_down_store(struct device *dev,
489 struct device_attribute *attr,
490 const char *buf, size_t len)
492 struct net_device *netdev = to_net_dev(dev);
494 /* The check is also done in change_proto_down; this helps returning
495 * early without hitting the trylock/restart in netdev_store.
497 if (!netdev->netdev_ops->ndo_change_proto_down)
500 return netdev_store(dev, attr, buf, len, change_proto_down);
502 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
504 static ssize_t phys_port_id_show(struct device *dev,
505 struct device_attribute *attr, char *buf)
507 struct net_device *netdev = to_net_dev(dev);
508 ssize_t ret = -EINVAL;
510 /* The check is also done in dev_get_phys_port_id; this helps returning
511 * early without hitting the trylock/restart below.
513 if (!netdev->netdev_ops->ndo_get_phys_port_id)
517 return restart_syscall();
519 if (dev_isalive(netdev)) {
520 struct netdev_phys_item_id ppid;
522 ret = dev_get_phys_port_id(netdev, &ppid);
524 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
530 static DEVICE_ATTR_RO(phys_port_id);
532 static ssize_t phys_port_name_show(struct device *dev,
533 struct device_attribute *attr, char *buf)
535 struct net_device *netdev = to_net_dev(dev);
536 ssize_t ret = -EINVAL;
538 /* The checks are also done in dev_get_phys_port_name; this helps
539 * returning early without hitting the trylock/restart below.
541 if (!netdev->netdev_ops->ndo_get_phys_port_name &&
542 !netdev->netdev_ops->ndo_get_devlink_port)
546 return restart_syscall();
548 if (dev_isalive(netdev)) {
551 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
553 ret = sprintf(buf, "%s\n", name);
559 static DEVICE_ATTR_RO(phys_port_name);
561 static ssize_t phys_switch_id_show(struct device *dev,
562 struct device_attribute *attr, char *buf)
564 struct net_device *netdev = to_net_dev(dev);
565 ssize_t ret = -EINVAL;
567 /* The checks are also done in dev_get_phys_port_name; this helps
568 * returning early without hitting the trylock/restart below. This works
569 * because recurse is false when calling dev_get_port_parent_id.
571 if (!netdev->netdev_ops->ndo_get_port_parent_id &&
572 !netdev->netdev_ops->ndo_get_devlink_port)
576 return restart_syscall();
578 if (dev_isalive(netdev)) {
579 struct netdev_phys_item_id ppid = { };
581 ret = dev_get_port_parent_id(netdev, &ppid, false);
583 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
589 static DEVICE_ATTR_RO(phys_switch_id);
591 static ssize_t threaded_show(struct device *dev,
592 struct device_attribute *attr, char *buf)
594 struct net_device *netdev = to_net_dev(dev);
595 ssize_t ret = -EINVAL;
598 return restart_syscall();
600 if (dev_isalive(netdev))
601 ret = sprintf(buf, fmt_dec, netdev->threaded);
607 static int modify_napi_threaded(struct net_device *dev, unsigned long val)
611 if (list_empty(&dev->napi_list))
614 if (val != 0 && val != 1)
617 ret = dev_set_threaded(dev, val);
622 static ssize_t threaded_store(struct device *dev,
623 struct device_attribute *attr,
624 const char *buf, size_t len)
626 return netdev_store(dev, attr, buf, len, modify_napi_threaded);
628 static DEVICE_ATTR_RW(threaded);
630 static struct attribute *net_class_attrs[] __ro_after_init = {
631 &dev_attr_netdev_group.attr,
633 &dev_attr_dev_id.attr,
634 &dev_attr_dev_port.attr,
635 &dev_attr_iflink.attr,
636 &dev_attr_ifindex.attr,
637 &dev_attr_name_assign_type.attr,
638 &dev_attr_addr_assign_type.attr,
639 &dev_attr_addr_len.attr,
640 &dev_attr_link_mode.attr,
641 &dev_attr_address.attr,
642 &dev_attr_broadcast.attr,
643 &dev_attr_speed.attr,
644 &dev_attr_duplex.attr,
645 &dev_attr_dormant.attr,
646 &dev_attr_testing.attr,
647 &dev_attr_operstate.attr,
648 &dev_attr_carrier_changes.attr,
649 &dev_attr_ifalias.attr,
650 &dev_attr_carrier.attr,
652 &dev_attr_flags.attr,
653 &dev_attr_tx_queue_len.attr,
654 &dev_attr_gro_flush_timeout.attr,
655 &dev_attr_napi_defer_hard_irqs.attr,
656 &dev_attr_phys_port_id.attr,
657 &dev_attr_phys_port_name.attr,
658 &dev_attr_phys_switch_id.attr,
659 &dev_attr_proto_down.attr,
660 &dev_attr_carrier_up_count.attr,
661 &dev_attr_carrier_down_count.attr,
662 &dev_attr_threaded.attr,
665 ATTRIBUTE_GROUPS(net_class);
667 /* Show a given an attribute in the statistics group */
668 static ssize_t netstat_show(const struct device *d,
669 struct device_attribute *attr, char *buf,
670 unsigned long offset)
672 struct net_device *dev = to_net_dev(d);
673 ssize_t ret = -EINVAL;
675 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
676 offset % sizeof(u64) != 0);
678 read_lock(&dev_base_lock);
679 if (dev_isalive(dev)) {
680 struct rtnl_link_stats64 temp;
681 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
683 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
685 read_unlock(&dev_base_lock);
689 /* generate a read-only statistics attribute */
690 #define NETSTAT_ENTRY(name) \
691 static ssize_t name##_show(struct device *d, \
692 struct device_attribute *attr, char *buf) \
694 return netstat_show(d, attr, buf, \
695 offsetof(struct rtnl_link_stats64, name)); \
697 static DEVICE_ATTR_RO(name)
699 NETSTAT_ENTRY(rx_packets);
700 NETSTAT_ENTRY(tx_packets);
701 NETSTAT_ENTRY(rx_bytes);
702 NETSTAT_ENTRY(tx_bytes);
703 NETSTAT_ENTRY(rx_errors);
704 NETSTAT_ENTRY(tx_errors);
705 NETSTAT_ENTRY(rx_dropped);
706 NETSTAT_ENTRY(tx_dropped);
707 NETSTAT_ENTRY(multicast);
708 NETSTAT_ENTRY(collisions);
709 NETSTAT_ENTRY(rx_length_errors);
710 NETSTAT_ENTRY(rx_over_errors);
711 NETSTAT_ENTRY(rx_crc_errors);
712 NETSTAT_ENTRY(rx_frame_errors);
713 NETSTAT_ENTRY(rx_fifo_errors);
714 NETSTAT_ENTRY(rx_missed_errors);
715 NETSTAT_ENTRY(tx_aborted_errors);
716 NETSTAT_ENTRY(tx_carrier_errors);
717 NETSTAT_ENTRY(tx_fifo_errors);
718 NETSTAT_ENTRY(tx_heartbeat_errors);
719 NETSTAT_ENTRY(tx_window_errors);
720 NETSTAT_ENTRY(rx_compressed);
721 NETSTAT_ENTRY(tx_compressed);
722 NETSTAT_ENTRY(rx_nohandler);
724 static struct attribute *netstat_attrs[] __ro_after_init = {
725 &dev_attr_rx_packets.attr,
726 &dev_attr_tx_packets.attr,
727 &dev_attr_rx_bytes.attr,
728 &dev_attr_tx_bytes.attr,
729 &dev_attr_rx_errors.attr,
730 &dev_attr_tx_errors.attr,
731 &dev_attr_rx_dropped.attr,
732 &dev_attr_tx_dropped.attr,
733 &dev_attr_multicast.attr,
734 &dev_attr_collisions.attr,
735 &dev_attr_rx_length_errors.attr,
736 &dev_attr_rx_over_errors.attr,
737 &dev_attr_rx_crc_errors.attr,
738 &dev_attr_rx_frame_errors.attr,
739 &dev_attr_rx_fifo_errors.attr,
740 &dev_attr_rx_missed_errors.attr,
741 &dev_attr_tx_aborted_errors.attr,
742 &dev_attr_tx_carrier_errors.attr,
743 &dev_attr_tx_fifo_errors.attr,
744 &dev_attr_tx_heartbeat_errors.attr,
745 &dev_attr_tx_window_errors.attr,
746 &dev_attr_rx_compressed.attr,
747 &dev_attr_tx_compressed.attr,
748 &dev_attr_rx_nohandler.attr,
752 static const struct attribute_group netstat_group = {
753 .name = "statistics",
754 .attrs = netstat_attrs,
757 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
758 static struct attribute *wireless_attrs[] = {
762 static const struct attribute_group wireless_group = {
764 .attrs = wireless_attrs,
768 #else /* CONFIG_SYSFS */
769 #define net_class_groups NULL
770 #endif /* CONFIG_SYSFS */
773 #define to_rx_queue_attr(_attr) \
774 container_of(_attr, struct rx_queue_attribute, attr)
776 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
778 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
781 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
782 struct netdev_rx_queue *queue = to_rx_queue(kobj);
784 if (!attribute->show)
787 return attribute->show(queue, buf);
790 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
791 const char *buf, size_t count)
793 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
794 struct netdev_rx_queue *queue = to_rx_queue(kobj);
796 if (!attribute->store)
799 return attribute->store(queue, buf, count);
802 static const struct sysfs_ops rx_queue_sysfs_ops = {
803 .show = rx_queue_attr_show,
804 .store = rx_queue_attr_store,
808 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
814 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
818 map = rcu_dereference(queue->rps_map);
820 for (i = 0; i < map->len; i++)
821 cpumask_set_cpu(map->cpus[i], mask);
823 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
825 free_cpumask_var(mask);
827 return len < PAGE_SIZE ? len : -EINVAL;
830 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
831 const char *buf, size_t len)
833 struct rps_map *old_map, *map;
835 int err, cpu, i, hk_flags;
836 static DEFINE_MUTEX(rps_map_mutex);
838 if (!capable(CAP_NET_ADMIN))
841 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
844 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
846 free_cpumask_var(mask);
850 if (!cpumask_empty(mask)) {
851 hk_flags = HK_FLAG_DOMAIN | HK_FLAG_WQ;
852 cpumask_and(mask, mask, housekeeping_cpumask(hk_flags));
853 if (cpumask_empty(mask)) {
854 free_cpumask_var(mask);
859 map = kzalloc(max_t(unsigned int,
860 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
863 free_cpumask_var(mask);
868 for_each_cpu_and(cpu, mask, cpu_online_mask)
869 map->cpus[i++] = cpu;
878 mutex_lock(&rps_map_mutex);
879 old_map = rcu_dereference_protected(queue->rps_map,
880 mutex_is_locked(&rps_map_mutex));
881 rcu_assign_pointer(queue->rps_map, map);
884 static_branch_inc(&rps_needed);
886 static_branch_dec(&rps_needed);
888 mutex_unlock(&rps_map_mutex);
891 kfree_rcu(old_map, rcu);
893 free_cpumask_var(mask);
897 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
900 struct rps_dev_flow_table *flow_table;
901 unsigned long val = 0;
904 flow_table = rcu_dereference(queue->rps_flow_table);
906 val = (unsigned long)flow_table->mask + 1;
909 return sprintf(buf, "%lu\n", val);
912 static void rps_dev_flow_table_release(struct rcu_head *rcu)
914 struct rps_dev_flow_table *table = container_of(rcu,
915 struct rps_dev_flow_table, rcu);
919 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
920 const char *buf, size_t len)
922 unsigned long mask, count;
923 struct rps_dev_flow_table *table, *old_table;
924 static DEFINE_SPINLOCK(rps_dev_flow_lock);
927 if (!capable(CAP_NET_ADMIN))
930 rc = kstrtoul(buf, 0, &count);
936 /* mask = roundup_pow_of_two(count) - 1;
937 * without overflows...
939 while ((mask | (mask >> 1)) != mask)
941 /* On 64 bit arches, must check mask fits in table->mask (u32),
942 * and on 32bit arches, must check
943 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
945 #if BITS_PER_LONG > 32
946 if (mask > (unsigned long)(u32)mask)
949 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
950 / sizeof(struct rps_dev_flow)) {
951 /* Enforce a limit to prevent overflow */
955 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
960 for (count = 0; count <= mask; count++)
961 table->flows[count].cpu = RPS_NO_CPU;
966 spin_lock(&rps_dev_flow_lock);
967 old_table = rcu_dereference_protected(queue->rps_flow_table,
968 lockdep_is_held(&rps_dev_flow_lock));
969 rcu_assign_pointer(queue->rps_flow_table, table);
970 spin_unlock(&rps_dev_flow_lock);
973 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
978 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
979 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
981 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
982 = __ATTR(rps_flow_cnt, 0644,
983 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
984 #endif /* CONFIG_RPS */
986 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
988 &rps_cpus_attribute.attr,
989 &rps_dev_flow_table_cnt_attribute.attr,
993 ATTRIBUTE_GROUPS(rx_queue_default);
995 static void rx_queue_release(struct kobject *kobj)
997 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1000 struct rps_dev_flow_table *flow_table;
1002 map = rcu_dereference_protected(queue->rps_map, 1);
1004 RCU_INIT_POINTER(queue->rps_map, NULL);
1005 kfree_rcu(map, rcu);
1008 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1010 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1011 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1015 memset(kobj, 0, sizeof(*kobj));
1016 dev_put(queue->dev);
1019 static const void *rx_queue_namespace(struct kobject *kobj)
1021 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1022 struct device *dev = &queue->dev->dev;
1023 const void *ns = NULL;
1025 if (dev->class && dev->class->ns_type)
1026 ns = dev->class->namespace(dev);
1031 static void rx_queue_get_ownership(struct kobject *kobj,
1032 kuid_t *uid, kgid_t *gid)
1034 const struct net *net = rx_queue_namespace(kobj);
1036 net_ns_get_ownership(net, uid, gid);
1039 static struct kobj_type rx_queue_ktype __ro_after_init = {
1040 .sysfs_ops = &rx_queue_sysfs_ops,
1041 .release = rx_queue_release,
1042 .default_groups = rx_queue_default_groups,
1043 .namespace = rx_queue_namespace,
1044 .get_ownership = rx_queue_get_ownership,
1047 static int rx_queue_add_kobject(struct net_device *dev, int index)
1049 struct netdev_rx_queue *queue = dev->_rx + index;
1050 struct kobject *kobj = &queue->kobj;
1053 /* Kobject_put later will trigger rx_queue_release call which
1054 * decreases dev refcount: Take that reference here
1056 dev_hold(queue->dev);
1058 kobj->kset = dev->queues_kset;
1059 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1064 if (dev->sysfs_rx_queue_group) {
1065 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1070 kobject_uevent(kobj, KOBJ_ADD);
1079 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1082 struct netdev_rx_queue *queue = dev->_rx + index;
1083 struct kobject *kobj = &queue->kobj;
1086 error = sysfs_change_owner(kobj, kuid, kgid);
1090 if (dev->sysfs_rx_queue_group)
1091 error = sysfs_group_change_owner(
1092 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1096 #endif /* CONFIG_SYSFS */
1099 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1106 if (!dev->sysfs_rx_queue_group)
1109 for (i = old_num; i < new_num; i++) {
1110 error = rx_queue_add_kobject(dev, i);
1117 while (--i >= new_num) {
1118 struct kobject *kobj = &dev->_rx[i].kobj;
1120 if (!refcount_read(&dev_net(dev)->ns.count))
1121 kobj->uevent_suppress = 1;
1122 if (dev->sysfs_rx_queue_group)
1123 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1133 static int net_rx_queue_change_owner(struct net_device *dev, int num,
1134 kuid_t kuid, kgid_t kgid)
1141 if (!dev->sysfs_rx_queue_group)
1144 for (i = 0; i < num; i++) {
1145 error = rx_queue_change_owner(dev, i, kuid, kgid);
1158 * netdev_queue sysfs structures and functions.
1160 struct netdev_queue_attribute {
1161 struct attribute attr;
1162 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1163 ssize_t (*store)(struct netdev_queue *queue,
1164 const char *buf, size_t len);
1166 #define to_netdev_queue_attr(_attr) \
1167 container_of(_attr, struct netdev_queue_attribute, attr)
1169 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1171 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1172 struct attribute *attr, char *buf)
1174 const struct netdev_queue_attribute *attribute
1175 = to_netdev_queue_attr(attr);
1176 struct netdev_queue *queue = to_netdev_queue(kobj);
1178 if (!attribute->show)
1181 return attribute->show(queue, buf);
1184 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1185 struct attribute *attr,
1186 const char *buf, size_t count)
1188 const struct netdev_queue_attribute *attribute
1189 = to_netdev_queue_attr(attr);
1190 struct netdev_queue *queue = to_netdev_queue(kobj);
1192 if (!attribute->store)
1195 return attribute->store(queue, buf, count);
1198 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1199 .show = netdev_queue_attr_show,
1200 .store = netdev_queue_attr_store,
1203 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1205 unsigned long trans_timeout;
1207 spin_lock_irq(&queue->_xmit_lock);
1208 trans_timeout = queue->trans_timeout;
1209 spin_unlock_irq(&queue->_xmit_lock);
1211 return sprintf(buf, fmt_ulong, trans_timeout);
1214 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1216 struct net_device *dev = queue->dev;
1219 i = queue - dev->_tx;
1220 BUG_ON(i >= dev->num_tx_queues);
1225 static ssize_t traffic_class_show(struct netdev_queue *queue,
1228 struct net_device *dev = queue->dev;
1232 if (!netif_is_multiqueue(dev))
1235 if (!rtnl_trylock())
1236 return restart_syscall();
1238 index = get_netdev_queue_index(queue);
1240 /* If queue belongs to subordinate dev use its TC mapping */
1241 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1243 num_tc = dev->num_tc;
1244 tc = netdev_txq_to_tc(dev, index);
1251 /* We can report the traffic class one of two ways:
1252 * Subordinate device traffic classes are reported with the traffic
1253 * class first, and then the subordinate class so for example TC0 on
1254 * subordinate device 2 will be reported as "0-2". If the queue
1255 * belongs to the root device it will be reported with just the
1256 * traffic class, so just "0" for TC 0 for example.
1258 return num_tc < 0 ? sprintf(buf, "%d%d\n", tc, num_tc) :
1259 sprintf(buf, "%d\n", tc);
1263 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1266 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1269 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1270 const char *buf, size_t len)
1272 struct net_device *dev = queue->dev;
1273 int err, index = get_netdev_queue_index(queue);
1276 if (!capable(CAP_NET_ADMIN))
1279 /* The check is also done later; this helps returning early without
1280 * hitting the trylock/restart below.
1282 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1285 err = kstrtou32(buf, 10, &rate);
1289 if (!rtnl_trylock())
1290 return restart_syscall();
1293 if (dev->netdev_ops->ndo_set_tx_maxrate)
1294 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1298 queue->tx_maxrate = rate;
1304 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1305 = __ATTR_RW(tx_maxrate);
1308 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1309 = __ATTR_RO(tx_timeout);
1311 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1312 = __ATTR_RO(traffic_class);
1316 * Byte queue limits sysfs structures and functions.
1318 static ssize_t bql_show(char *buf, unsigned int value)
1320 return sprintf(buf, "%u\n", value);
1323 static ssize_t bql_set(const char *buf, const size_t count,
1324 unsigned int *pvalue)
1329 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1330 value = DQL_MAX_LIMIT;
1332 err = kstrtouint(buf, 10, &value);
1335 if (value > DQL_MAX_LIMIT)
1344 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1347 struct dql *dql = &queue->dql;
1349 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1352 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1353 const char *buf, size_t len)
1355 struct dql *dql = &queue->dql;
1359 err = kstrtouint(buf, 10, &value);
1363 dql->slack_hold_time = msecs_to_jiffies(value);
1368 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1369 = __ATTR(hold_time, 0644,
1370 bql_show_hold_time, bql_set_hold_time);
1372 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1375 struct dql *dql = &queue->dql;
1377 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1380 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1381 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1383 #define BQL_ATTR(NAME, FIELD) \
1384 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1387 return bql_show(buf, queue->dql.FIELD); \
1390 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1391 const char *buf, size_t len) \
1393 return bql_set(buf, len, &queue->dql.FIELD); \
1396 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1397 = __ATTR(NAME, 0644, \
1398 bql_show_ ## NAME, bql_set_ ## NAME)
1400 BQL_ATTR(limit, limit);
1401 BQL_ATTR(limit_max, max_limit);
1402 BQL_ATTR(limit_min, min_limit);
1404 static struct attribute *dql_attrs[] __ro_after_init = {
1405 &bql_limit_attribute.attr,
1406 &bql_limit_max_attribute.attr,
1407 &bql_limit_min_attribute.attr,
1408 &bql_hold_time_attribute.attr,
1409 &bql_inflight_attribute.attr,
1413 static const struct attribute_group dql_group = {
1414 .name = "byte_queue_limits",
1417 #endif /* CONFIG_BQL */
1420 static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1421 int tc, char *buf, enum xps_map_type type)
1423 struct xps_dev_maps *dev_maps;
1424 unsigned long *mask;
1425 unsigned int nr_ids;
1429 dev_maps = rcu_dereference(dev->xps_maps[type]);
1431 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1432 * when dev_maps hasn't been allocated yet, to be backward compatible.
1434 nr_ids = dev_maps ? dev_maps->nr_ids :
1435 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1437 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1443 if (!dev_maps || tc >= dev_maps->num_tc)
1446 for (j = 0; j < nr_ids; j++) {
1447 int i, tci = j * dev_maps->num_tc + tc;
1448 struct xps_map *map;
1450 map = rcu_dereference(dev_maps->attr_map[tci]);
1454 for (i = map->len; i--;) {
1455 if (map->queues[i] == index) {
1464 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1467 return len < PAGE_SIZE ? len : -EINVAL;
1470 static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1472 struct net_device *dev = queue->dev;
1476 if (!netif_is_multiqueue(dev))
1479 index = get_netdev_queue_index(queue);
1481 if (!rtnl_trylock())
1482 return restart_syscall();
1484 /* If queue belongs to subordinate dev use its map */
1485 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1487 tc = netdev_txq_to_tc(dev, index);
1493 /* Make sure the subordinate device can't be freed */
1494 get_device(&dev->dev);
1497 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1499 put_device(&dev->dev);
1503 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1504 const char *buf, size_t len)
1506 struct net_device *dev = queue->dev;
1511 if (!netif_is_multiqueue(dev))
1514 if (!capable(CAP_NET_ADMIN))
1517 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1520 index = get_netdev_queue_index(queue);
1522 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1524 free_cpumask_var(mask);
1528 if (!rtnl_trylock()) {
1529 free_cpumask_var(mask);
1530 return restart_syscall();
1533 err = netif_set_xps_queue(dev, mask, index);
1536 free_cpumask_var(mask);
1541 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1542 = __ATTR_RW(xps_cpus);
1544 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1546 struct net_device *dev = queue->dev;
1550 index = get_netdev_queue_index(queue);
1552 if (!rtnl_trylock())
1553 return restart_syscall();
1555 tc = netdev_txq_to_tc(dev, index);
1560 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1563 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1566 struct net_device *dev = queue->dev;
1567 struct net *net = dev_net(dev);
1568 unsigned long *mask;
1572 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1575 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1579 index = get_netdev_queue_index(queue);
1581 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1587 if (!rtnl_trylock()) {
1589 return restart_syscall();
1593 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1602 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1603 = __ATTR_RW(xps_rxqs);
1604 #endif /* CONFIG_XPS */
1606 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1607 &queue_trans_timeout.attr,
1608 &queue_traffic_class.attr,
1610 &xps_cpus_attribute.attr,
1611 &xps_rxqs_attribute.attr,
1612 &queue_tx_maxrate.attr,
1616 ATTRIBUTE_GROUPS(netdev_queue_default);
1618 static void netdev_queue_release(struct kobject *kobj)
1620 struct netdev_queue *queue = to_netdev_queue(kobj);
1622 memset(kobj, 0, sizeof(*kobj));
1623 dev_put(queue->dev);
1626 static const void *netdev_queue_namespace(struct kobject *kobj)
1628 struct netdev_queue *queue = to_netdev_queue(kobj);
1629 struct device *dev = &queue->dev->dev;
1630 const void *ns = NULL;
1632 if (dev->class && dev->class->ns_type)
1633 ns = dev->class->namespace(dev);
1638 static void netdev_queue_get_ownership(struct kobject *kobj,
1639 kuid_t *uid, kgid_t *gid)
1641 const struct net *net = netdev_queue_namespace(kobj);
1643 net_ns_get_ownership(net, uid, gid);
1646 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1647 .sysfs_ops = &netdev_queue_sysfs_ops,
1648 .release = netdev_queue_release,
1649 .default_groups = netdev_queue_default_groups,
1650 .namespace = netdev_queue_namespace,
1651 .get_ownership = netdev_queue_get_ownership,
1654 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1656 struct netdev_queue *queue = dev->_tx + index;
1657 struct kobject *kobj = &queue->kobj;
1660 /* Kobject_put later will trigger netdev_queue_release call
1661 * which decreases dev refcount: Take that reference here
1663 dev_hold(queue->dev);
1665 kobj->kset = dev->queues_kset;
1666 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1672 error = sysfs_create_group(kobj, &dql_group);
1677 kobject_uevent(kobj, KOBJ_ADD);
1685 static int tx_queue_change_owner(struct net_device *ndev, int index,
1686 kuid_t kuid, kgid_t kgid)
1688 struct netdev_queue *queue = ndev->_tx + index;
1689 struct kobject *kobj = &queue->kobj;
1692 error = sysfs_change_owner(kobj, kuid, kgid);
1697 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1701 #endif /* CONFIG_SYSFS */
1704 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1710 for (i = old_num; i < new_num; i++) {
1711 error = netdev_queue_add_kobject(dev, i);
1718 while (--i >= new_num) {
1719 struct netdev_queue *queue = dev->_tx + i;
1721 if (!refcount_read(&dev_net(dev)->ns.count))
1722 queue->kobj.uevent_suppress = 1;
1724 sysfs_remove_group(&queue->kobj, &dql_group);
1726 kobject_put(&queue->kobj);
1732 #endif /* CONFIG_SYSFS */
1735 static int net_tx_queue_change_owner(struct net_device *dev, int num,
1736 kuid_t kuid, kgid_t kgid)
1742 for (i = 0; i < num; i++) {
1743 error = tx_queue_change_owner(dev, i, kuid, kgid);
1751 #endif /* CONFIG_SYSFS */
1754 static int register_queue_kobjects(struct net_device *dev)
1756 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1759 dev->queues_kset = kset_create_and_add("queues",
1760 NULL, &dev->dev.kobj);
1761 if (!dev->queues_kset)
1763 real_rx = dev->real_num_rx_queues;
1765 real_tx = dev->real_num_tx_queues;
1767 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1772 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1780 netdev_queue_update_kobjects(dev, txq, 0);
1781 net_rx_queue_update_kobjects(dev, rxq, 0);
1783 kset_unregister(dev->queues_kset);
1788 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1790 int error = 0, real_rx = 0, real_tx = 0;
1793 if (ndev->queues_kset) {
1794 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1798 real_rx = ndev->real_num_rx_queues;
1800 real_tx = ndev->real_num_tx_queues;
1802 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1806 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1813 static void remove_queue_kobjects(struct net_device *dev)
1815 int real_rx = 0, real_tx = 0;
1818 real_rx = dev->real_num_rx_queues;
1820 real_tx = dev->real_num_tx_queues;
1822 net_rx_queue_update_kobjects(dev, real_rx, 0);
1823 netdev_queue_update_kobjects(dev, real_tx, 0);
1825 dev->real_num_rx_queues = 0;
1826 dev->real_num_tx_queues = 0;
1828 kset_unregister(dev->queues_kset);
1832 static bool net_current_may_mount(void)
1834 struct net *net = current->nsproxy->net_ns;
1836 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1839 static void *net_grab_current_ns(void)
1841 struct net *ns = current->nsproxy->net_ns;
1842 #ifdef CONFIG_NET_NS
1844 refcount_inc(&ns->passive);
1849 static const void *net_initial_ns(void)
1854 static const void *net_netlink_ns(struct sock *sk)
1856 return sock_net(sk);
1859 const struct kobj_ns_type_operations net_ns_type_operations = {
1860 .type = KOBJ_NS_TYPE_NET,
1861 .current_may_mount = net_current_may_mount,
1862 .grab_current_ns = net_grab_current_ns,
1863 .netlink_ns = net_netlink_ns,
1864 .initial_ns = net_initial_ns,
1865 .drop_ns = net_drop_ns,
1867 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1869 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1871 struct net_device *dev = to_net_dev(d);
1874 /* pass interface to uevent. */
1875 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1879 /* pass ifindex to uevent.
1880 * ifindex is useful as it won't change (interface name may change)
1881 * and is what RtNetlink uses natively.
1883 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1890 * netdev_release -- destroy and free a dead device.
1891 * Called when last reference to device kobject is gone.
1893 static void netdev_release(struct device *d)
1895 struct net_device *dev = to_net_dev(d);
1897 BUG_ON(dev->reg_state != NETREG_RELEASED);
1899 /* no need to wait for rcu grace period:
1900 * device is dead and about to be freed.
1902 kfree(rcu_access_pointer(dev->ifalias));
1903 netdev_freemem(dev);
1906 static const void *net_namespace(struct device *d)
1908 struct net_device *dev = to_net_dev(d);
1910 return dev_net(dev);
1913 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1915 struct net_device *dev = to_net_dev(d);
1916 const struct net *net = dev_net(dev);
1918 net_ns_get_ownership(net, uid, gid);
1921 static struct class net_class __ro_after_init = {
1923 .dev_release = netdev_release,
1924 .dev_groups = net_class_groups,
1925 .dev_uevent = netdev_uevent,
1926 .ns_type = &net_ns_type_operations,
1927 .namespace = net_namespace,
1928 .get_ownership = net_get_ownership,
1932 static int of_dev_node_match(struct device *dev, const void *data)
1934 for (; dev; dev = dev->parent) {
1935 if (dev->of_node == data)
1943 * of_find_net_device_by_node - lookup the net device for the device node
1944 * @np: OF device node
1946 * Looks up the net_device structure corresponding with the device node.
1947 * If successful, returns a pointer to the net_device with the embedded
1948 * struct device refcount incremented by one, or NULL on failure. The
1949 * refcount must be dropped when done with the net_device.
1951 struct net_device *of_find_net_device_by_node(struct device_node *np)
1955 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1959 return to_net_dev(dev);
1961 EXPORT_SYMBOL(of_find_net_device_by_node);
1964 /* Delete sysfs entries but hold kobject reference until after all
1965 * netdev references are gone.
1967 void netdev_unregister_kobject(struct net_device *ndev)
1969 struct device *dev = &ndev->dev;
1971 if (!refcount_read(&dev_net(ndev)->ns.count))
1972 dev_set_uevent_suppress(dev, 1);
1974 kobject_get(&dev->kobj);
1976 remove_queue_kobjects(ndev);
1978 pm_runtime_set_memalloc_noio(dev, false);
1983 /* Create sysfs entries for network device. */
1984 int netdev_register_kobject(struct net_device *ndev)
1986 struct device *dev = &ndev->dev;
1987 const struct attribute_group **groups = ndev->sysfs_groups;
1990 device_initialize(dev);
1991 dev->class = &net_class;
1992 dev->platform_data = ndev;
1993 dev->groups = groups;
1995 dev_set_name(dev, "%s", ndev->name);
1998 /* Allow for a device specific group */
2002 *groups++ = &netstat_group;
2004 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
2005 if (ndev->ieee80211_ptr)
2006 *groups++ = &wireless_group;
2007 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
2008 else if (ndev->wireless_handlers)
2009 *groups++ = &wireless_group;
2012 #endif /* CONFIG_SYSFS */
2014 error = device_add(dev);
2018 error = register_queue_kobjects(ndev);
2024 pm_runtime_set_memalloc_noio(dev, true);
2029 /* Change owner for sysfs entries when moving network devices across network
2030 * namespaces owned by different user namespaces.
2032 int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2033 const struct net *net_new)
2035 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2036 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2037 struct device *dev = &ndev->dev;
2040 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2041 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2043 /* The network namespace was changed but the owning user namespace is
2044 * identical so there's no need to change the owner of sysfs entries.
2046 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2049 error = device_change_owner(dev, new_uid, new_gid);
2053 error = queue_change_owner(ndev, new_uid, new_gid);
2060 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2063 return class_create_file_ns(&net_class, class_attr, ns);
2065 EXPORT_SYMBOL(netdev_class_create_file_ns);
2067 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2070 class_remove_file_ns(&net_class, class_attr, ns);
2072 EXPORT_SYMBOL(netdev_class_remove_file_ns);
2074 int __init netdev_kobject_init(void)
2076 kobj_ns_type_register(&net_ns_type_operations);
2077 return class_register(&net_class);