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 return netdev_store(dev, attr, buf, len, change_proto_down);
493 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
495 static ssize_t phys_port_id_show(struct device *dev,
496 struct device_attribute *attr, char *buf)
498 struct net_device *netdev = to_net_dev(dev);
499 ssize_t ret = -EINVAL;
501 /* The check is also done in dev_get_phys_port_id; this helps returning
502 * early without hitting the trylock/restart below.
504 if (!netdev->netdev_ops->ndo_get_phys_port_id)
508 return restart_syscall();
510 if (dev_isalive(netdev)) {
511 struct netdev_phys_item_id ppid;
513 ret = dev_get_phys_port_id(netdev, &ppid);
515 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
521 static DEVICE_ATTR_RO(phys_port_id);
523 static ssize_t phys_port_name_show(struct device *dev,
524 struct device_attribute *attr, char *buf)
526 struct net_device *netdev = to_net_dev(dev);
527 ssize_t ret = -EINVAL;
529 /* The checks are also done in dev_get_phys_port_name; this helps
530 * returning early without hitting the trylock/restart below.
532 if (!netdev->netdev_ops->ndo_get_phys_port_name &&
533 !netdev->netdev_ops->ndo_get_devlink_port)
537 return restart_syscall();
539 if (dev_isalive(netdev)) {
542 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
544 ret = sprintf(buf, "%s\n", name);
550 static DEVICE_ATTR_RO(phys_port_name);
552 static ssize_t phys_switch_id_show(struct device *dev,
553 struct device_attribute *attr, char *buf)
555 struct net_device *netdev = to_net_dev(dev);
556 ssize_t ret = -EINVAL;
558 /* The checks are also done in dev_get_phys_port_name; this helps
559 * returning early without hitting the trylock/restart below. This works
560 * because recurse is false when calling dev_get_port_parent_id.
562 if (!netdev->netdev_ops->ndo_get_port_parent_id &&
563 !netdev->netdev_ops->ndo_get_devlink_port)
567 return restart_syscall();
569 if (dev_isalive(netdev)) {
570 struct netdev_phys_item_id ppid = { };
572 ret = dev_get_port_parent_id(netdev, &ppid, false);
574 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
580 static DEVICE_ATTR_RO(phys_switch_id);
582 static ssize_t threaded_show(struct device *dev,
583 struct device_attribute *attr, char *buf)
585 struct net_device *netdev = to_net_dev(dev);
586 ssize_t ret = -EINVAL;
589 return restart_syscall();
591 if (dev_isalive(netdev))
592 ret = sprintf(buf, fmt_dec, netdev->threaded);
598 static int modify_napi_threaded(struct net_device *dev, unsigned long val)
602 if (list_empty(&dev->napi_list))
605 if (val != 0 && val != 1)
608 ret = dev_set_threaded(dev, val);
613 static ssize_t threaded_store(struct device *dev,
614 struct device_attribute *attr,
615 const char *buf, size_t len)
617 return netdev_store(dev, attr, buf, len, modify_napi_threaded);
619 static DEVICE_ATTR_RW(threaded);
621 static struct attribute *net_class_attrs[] __ro_after_init = {
622 &dev_attr_netdev_group.attr,
624 &dev_attr_dev_id.attr,
625 &dev_attr_dev_port.attr,
626 &dev_attr_iflink.attr,
627 &dev_attr_ifindex.attr,
628 &dev_attr_name_assign_type.attr,
629 &dev_attr_addr_assign_type.attr,
630 &dev_attr_addr_len.attr,
631 &dev_attr_link_mode.attr,
632 &dev_attr_address.attr,
633 &dev_attr_broadcast.attr,
634 &dev_attr_speed.attr,
635 &dev_attr_duplex.attr,
636 &dev_attr_dormant.attr,
637 &dev_attr_testing.attr,
638 &dev_attr_operstate.attr,
639 &dev_attr_carrier_changes.attr,
640 &dev_attr_ifalias.attr,
641 &dev_attr_carrier.attr,
643 &dev_attr_flags.attr,
644 &dev_attr_tx_queue_len.attr,
645 &dev_attr_gro_flush_timeout.attr,
646 &dev_attr_napi_defer_hard_irqs.attr,
647 &dev_attr_phys_port_id.attr,
648 &dev_attr_phys_port_name.attr,
649 &dev_attr_phys_switch_id.attr,
650 &dev_attr_proto_down.attr,
651 &dev_attr_carrier_up_count.attr,
652 &dev_attr_carrier_down_count.attr,
653 &dev_attr_threaded.attr,
656 ATTRIBUTE_GROUPS(net_class);
658 /* Show a given an attribute in the statistics group */
659 static ssize_t netstat_show(const struct device *d,
660 struct device_attribute *attr, char *buf,
661 unsigned long offset)
663 struct net_device *dev = to_net_dev(d);
664 ssize_t ret = -EINVAL;
666 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
667 offset % sizeof(u64) != 0);
669 read_lock(&dev_base_lock);
670 if (dev_isalive(dev)) {
671 struct rtnl_link_stats64 temp;
672 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
674 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
676 read_unlock(&dev_base_lock);
680 /* generate a read-only statistics attribute */
681 #define NETSTAT_ENTRY(name) \
682 static ssize_t name##_show(struct device *d, \
683 struct device_attribute *attr, char *buf) \
685 return netstat_show(d, attr, buf, \
686 offsetof(struct rtnl_link_stats64, name)); \
688 static DEVICE_ATTR_RO(name)
690 NETSTAT_ENTRY(rx_packets);
691 NETSTAT_ENTRY(tx_packets);
692 NETSTAT_ENTRY(rx_bytes);
693 NETSTAT_ENTRY(tx_bytes);
694 NETSTAT_ENTRY(rx_errors);
695 NETSTAT_ENTRY(tx_errors);
696 NETSTAT_ENTRY(rx_dropped);
697 NETSTAT_ENTRY(tx_dropped);
698 NETSTAT_ENTRY(multicast);
699 NETSTAT_ENTRY(collisions);
700 NETSTAT_ENTRY(rx_length_errors);
701 NETSTAT_ENTRY(rx_over_errors);
702 NETSTAT_ENTRY(rx_crc_errors);
703 NETSTAT_ENTRY(rx_frame_errors);
704 NETSTAT_ENTRY(rx_fifo_errors);
705 NETSTAT_ENTRY(rx_missed_errors);
706 NETSTAT_ENTRY(tx_aborted_errors);
707 NETSTAT_ENTRY(tx_carrier_errors);
708 NETSTAT_ENTRY(tx_fifo_errors);
709 NETSTAT_ENTRY(tx_heartbeat_errors);
710 NETSTAT_ENTRY(tx_window_errors);
711 NETSTAT_ENTRY(rx_compressed);
712 NETSTAT_ENTRY(tx_compressed);
713 NETSTAT_ENTRY(rx_nohandler);
715 static struct attribute *netstat_attrs[] __ro_after_init = {
716 &dev_attr_rx_packets.attr,
717 &dev_attr_tx_packets.attr,
718 &dev_attr_rx_bytes.attr,
719 &dev_attr_tx_bytes.attr,
720 &dev_attr_rx_errors.attr,
721 &dev_attr_tx_errors.attr,
722 &dev_attr_rx_dropped.attr,
723 &dev_attr_tx_dropped.attr,
724 &dev_attr_multicast.attr,
725 &dev_attr_collisions.attr,
726 &dev_attr_rx_length_errors.attr,
727 &dev_attr_rx_over_errors.attr,
728 &dev_attr_rx_crc_errors.attr,
729 &dev_attr_rx_frame_errors.attr,
730 &dev_attr_rx_fifo_errors.attr,
731 &dev_attr_rx_missed_errors.attr,
732 &dev_attr_tx_aborted_errors.attr,
733 &dev_attr_tx_carrier_errors.attr,
734 &dev_attr_tx_fifo_errors.attr,
735 &dev_attr_tx_heartbeat_errors.attr,
736 &dev_attr_tx_window_errors.attr,
737 &dev_attr_rx_compressed.attr,
738 &dev_attr_tx_compressed.attr,
739 &dev_attr_rx_nohandler.attr,
743 static const struct attribute_group netstat_group = {
744 .name = "statistics",
745 .attrs = netstat_attrs,
748 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
749 static struct attribute *wireless_attrs[] = {
753 static const struct attribute_group wireless_group = {
755 .attrs = wireless_attrs,
759 #else /* CONFIG_SYSFS */
760 #define net_class_groups NULL
761 #endif /* CONFIG_SYSFS */
764 #define to_rx_queue_attr(_attr) \
765 container_of(_attr, struct rx_queue_attribute, attr)
767 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
769 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
772 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
773 struct netdev_rx_queue *queue = to_rx_queue(kobj);
775 if (!attribute->show)
778 return attribute->show(queue, buf);
781 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
782 const char *buf, size_t count)
784 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
785 struct netdev_rx_queue *queue = to_rx_queue(kobj);
787 if (!attribute->store)
790 return attribute->store(queue, buf, count);
793 static const struct sysfs_ops rx_queue_sysfs_ops = {
794 .show = rx_queue_attr_show,
795 .store = rx_queue_attr_store,
799 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
805 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
809 map = rcu_dereference(queue->rps_map);
811 for (i = 0; i < map->len; i++)
812 cpumask_set_cpu(map->cpus[i], mask);
814 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
816 free_cpumask_var(mask);
818 return len < PAGE_SIZE ? len : -EINVAL;
821 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
822 const char *buf, size_t len)
824 struct rps_map *old_map, *map;
827 static DEFINE_MUTEX(rps_map_mutex);
829 if (!capable(CAP_NET_ADMIN))
832 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
835 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
837 free_cpumask_var(mask);
841 if (!cpumask_empty(mask)) {
842 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
843 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
844 if (cpumask_empty(mask)) {
845 free_cpumask_var(mask);
850 map = kzalloc(max_t(unsigned int,
851 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
854 free_cpumask_var(mask);
859 for_each_cpu_and(cpu, mask, cpu_online_mask)
860 map->cpus[i++] = cpu;
869 mutex_lock(&rps_map_mutex);
870 old_map = rcu_dereference_protected(queue->rps_map,
871 mutex_is_locked(&rps_map_mutex));
872 rcu_assign_pointer(queue->rps_map, map);
875 static_branch_inc(&rps_needed);
877 static_branch_dec(&rps_needed);
879 mutex_unlock(&rps_map_mutex);
882 kfree_rcu(old_map, rcu);
884 free_cpumask_var(mask);
888 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
891 struct rps_dev_flow_table *flow_table;
892 unsigned long val = 0;
895 flow_table = rcu_dereference(queue->rps_flow_table);
897 val = (unsigned long)flow_table->mask + 1;
900 return sprintf(buf, "%lu\n", val);
903 static void rps_dev_flow_table_release(struct rcu_head *rcu)
905 struct rps_dev_flow_table *table = container_of(rcu,
906 struct rps_dev_flow_table, rcu);
910 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
911 const char *buf, size_t len)
913 unsigned long mask, count;
914 struct rps_dev_flow_table *table, *old_table;
915 static DEFINE_SPINLOCK(rps_dev_flow_lock);
918 if (!capable(CAP_NET_ADMIN))
921 rc = kstrtoul(buf, 0, &count);
927 /* mask = roundup_pow_of_two(count) - 1;
928 * without overflows...
930 while ((mask | (mask >> 1)) != mask)
932 /* On 64 bit arches, must check mask fits in table->mask (u32),
933 * and on 32bit arches, must check
934 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
936 #if BITS_PER_LONG > 32
937 if (mask > (unsigned long)(u32)mask)
940 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
941 / sizeof(struct rps_dev_flow)) {
942 /* Enforce a limit to prevent overflow */
946 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
951 for (count = 0; count <= mask; count++)
952 table->flows[count].cpu = RPS_NO_CPU;
957 spin_lock(&rps_dev_flow_lock);
958 old_table = rcu_dereference_protected(queue->rps_flow_table,
959 lockdep_is_held(&rps_dev_flow_lock));
960 rcu_assign_pointer(queue->rps_flow_table, table);
961 spin_unlock(&rps_dev_flow_lock);
964 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
969 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
970 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
972 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
973 = __ATTR(rps_flow_cnt, 0644,
974 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
975 #endif /* CONFIG_RPS */
977 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
979 &rps_cpus_attribute.attr,
980 &rps_dev_flow_table_cnt_attribute.attr,
984 ATTRIBUTE_GROUPS(rx_queue_default);
986 static void rx_queue_release(struct kobject *kobj)
988 struct netdev_rx_queue *queue = to_rx_queue(kobj);
991 struct rps_dev_flow_table *flow_table;
993 map = rcu_dereference_protected(queue->rps_map, 1);
995 RCU_INIT_POINTER(queue->rps_map, NULL);
999 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1001 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1002 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1006 memset(kobj, 0, sizeof(*kobj));
1007 dev_put_track(queue->dev, &queue->dev_tracker);
1010 static const void *rx_queue_namespace(struct kobject *kobj)
1012 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1013 struct device *dev = &queue->dev->dev;
1014 const void *ns = NULL;
1016 if (dev->class && dev->class->ns_type)
1017 ns = dev->class->namespace(dev);
1022 static void rx_queue_get_ownership(struct kobject *kobj,
1023 kuid_t *uid, kgid_t *gid)
1025 const struct net *net = rx_queue_namespace(kobj);
1027 net_ns_get_ownership(net, uid, gid);
1030 static struct kobj_type rx_queue_ktype __ro_after_init = {
1031 .sysfs_ops = &rx_queue_sysfs_ops,
1032 .release = rx_queue_release,
1033 .default_groups = rx_queue_default_groups,
1034 .namespace = rx_queue_namespace,
1035 .get_ownership = rx_queue_get_ownership,
1038 static int rx_queue_add_kobject(struct net_device *dev, int index)
1040 struct netdev_rx_queue *queue = dev->_rx + index;
1041 struct kobject *kobj = &queue->kobj;
1044 /* Kobject_put later will trigger rx_queue_release call which
1045 * decreases dev refcount: Take that reference here
1047 dev_hold_track(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1049 kobj->kset = dev->queues_kset;
1050 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1055 if (dev->sysfs_rx_queue_group) {
1056 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1061 kobject_uevent(kobj, KOBJ_ADD);
1070 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1073 struct netdev_rx_queue *queue = dev->_rx + index;
1074 struct kobject *kobj = &queue->kobj;
1077 error = sysfs_change_owner(kobj, kuid, kgid);
1081 if (dev->sysfs_rx_queue_group)
1082 error = sysfs_group_change_owner(
1083 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1087 #endif /* CONFIG_SYSFS */
1090 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1097 if (!dev->sysfs_rx_queue_group)
1100 for (i = old_num; i < new_num; i++) {
1101 error = rx_queue_add_kobject(dev, i);
1108 while (--i >= new_num) {
1109 struct kobject *kobj = &dev->_rx[i].kobj;
1111 if (!refcount_read(&dev_net(dev)->ns.count))
1112 kobj->uevent_suppress = 1;
1113 if (dev->sysfs_rx_queue_group)
1114 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1124 static int net_rx_queue_change_owner(struct net_device *dev, int num,
1125 kuid_t kuid, kgid_t kgid)
1132 if (!dev->sysfs_rx_queue_group)
1135 for (i = 0; i < num; i++) {
1136 error = rx_queue_change_owner(dev, i, kuid, kgid);
1149 * netdev_queue sysfs structures and functions.
1151 struct netdev_queue_attribute {
1152 struct attribute attr;
1153 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1154 ssize_t (*store)(struct netdev_queue *queue,
1155 const char *buf, size_t len);
1157 #define to_netdev_queue_attr(_attr) \
1158 container_of(_attr, struct netdev_queue_attribute, attr)
1160 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1162 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1163 struct attribute *attr, char *buf)
1165 const struct netdev_queue_attribute *attribute
1166 = to_netdev_queue_attr(attr);
1167 struct netdev_queue *queue = to_netdev_queue(kobj);
1169 if (!attribute->show)
1172 return attribute->show(queue, buf);
1175 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1176 struct attribute *attr,
1177 const char *buf, size_t count)
1179 const struct netdev_queue_attribute *attribute
1180 = to_netdev_queue_attr(attr);
1181 struct netdev_queue *queue = to_netdev_queue(kobj);
1183 if (!attribute->store)
1186 return attribute->store(queue, buf, count);
1189 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1190 .show = netdev_queue_attr_show,
1191 .store = netdev_queue_attr_store,
1194 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1196 unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1198 return sprintf(buf, fmt_ulong, trans_timeout);
1201 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1203 struct net_device *dev = queue->dev;
1206 i = queue - dev->_tx;
1207 BUG_ON(i >= dev->num_tx_queues);
1212 static ssize_t traffic_class_show(struct netdev_queue *queue,
1215 struct net_device *dev = queue->dev;
1219 if (!netif_is_multiqueue(dev))
1222 if (!rtnl_trylock())
1223 return restart_syscall();
1225 index = get_netdev_queue_index(queue);
1227 /* If queue belongs to subordinate dev use its TC mapping */
1228 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1230 num_tc = dev->num_tc;
1231 tc = netdev_txq_to_tc(dev, index);
1238 /* We can report the traffic class one of two ways:
1239 * Subordinate device traffic classes are reported with the traffic
1240 * class first, and then the subordinate class so for example TC0 on
1241 * subordinate device 2 will be reported as "0-2". If the queue
1242 * belongs to the root device it will be reported with just the
1243 * traffic class, so just "0" for TC 0 for example.
1245 return num_tc < 0 ? sprintf(buf, "%d%d\n", tc, num_tc) :
1246 sprintf(buf, "%d\n", tc);
1250 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1253 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1256 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1257 const char *buf, size_t len)
1259 struct net_device *dev = queue->dev;
1260 int err, index = get_netdev_queue_index(queue);
1263 if (!capable(CAP_NET_ADMIN))
1266 /* The check is also done later; this helps returning early without
1267 * hitting the trylock/restart below.
1269 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1272 err = kstrtou32(buf, 10, &rate);
1276 if (!rtnl_trylock())
1277 return restart_syscall();
1280 if (dev->netdev_ops->ndo_set_tx_maxrate)
1281 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1285 queue->tx_maxrate = rate;
1291 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1292 = __ATTR_RW(tx_maxrate);
1295 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1296 = __ATTR_RO(tx_timeout);
1298 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1299 = __ATTR_RO(traffic_class);
1303 * Byte queue limits sysfs structures and functions.
1305 static ssize_t bql_show(char *buf, unsigned int value)
1307 return sprintf(buf, "%u\n", value);
1310 static ssize_t bql_set(const char *buf, const size_t count,
1311 unsigned int *pvalue)
1316 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1317 value = DQL_MAX_LIMIT;
1319 err = kstrtouint(buf, 10, &value);
1322 if (value > DQL_MAX_LIMIT)
1331 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1334 struct dql *dql = &queue->dql;
1336 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1339 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1340 const char *buf, size_t len)
1342 struct dql *dql = &queue->dql;
1346 err = kstrtouint(buf, 10, &value);
1350 dql->slack_hold_time = msecs_to_jiffies(value);
1355 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1356 = __ATTR(hold_time, 0644,
1357 bql_show_hold_time, bql_set_hold_time);
1359 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1362 struct dql *dql = &queue->dql;
1364 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1367 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1368 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1370 #define BQL_ATTR(NAME, FIELD) \
1371 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1374 return bql_show(buf, queue->dql.FIELD); \
1377 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1378 const char *buf, size_t len) \
1380 return bql_set(buf, len, &queue->dql.FIELD); \
1383 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1384 = __ATTR(NAME, 0644, \
1385 bql_show_ ## NAME, bql_set_ ## NAME)
1387 BQL_ATTR(limit, limit);
1388 BQL_ATTR(limit_max, max_limit);
1389 BQL_ATTR(limit_min, min_limit);
1391 static struct attribute *dql_attrs[] __ro_after_init = {
1392 &bql_limit_attribute.attr,
1393 &bql_limit_max_attribute.attr,
1394 &bql_limit_min_attribute.attr,
1395 &bql_hold_time_attribute.attr,
1396 &bql_inflight_attribute.attr,
1400 static const struct attribute_group dql_group = {
1401 .name = "byte_queue_limits",
1404 #endif /* CONFIG_BQL */
1407 static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1408 int tc, char *buf, enum xps_map_type type)
1410 struct xps_dev_maps *dev_maps;
1411 unsigned long *mask;
1412 unsigned int nr_ids;
1416 dev_maps = rcu_dereference(dev->xps_maps[type]);
1418 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1419 * when dev_maps hasn't been allocated yet, to be backward compatible.
1421 nr_ids = dev_maps ? dev_maps->nr_ids :
1422 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1424 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1430 if (!dev_maps || tc >= dev_maps->num_tc)
1433 for (j = 0; j < nr_ids; j++) {
1434 int i, tci = j * dev_maps->num_tc + tc;
1435 struct xps_map *map;
1437 map = rcu_dereference(dev_maps->attr_map[tci]);
1441 for (i = map->len; i--;) {
1442 if (map->queues[i] == index) {
1451 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1454 return len < PAGE_SIZE ? len : -EINVAL;
1457 static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1459 struct net_device *dev = queue->dev;
1463 if (!netif_is_multiqueue(dev))
1466 index = get_netdev_queue_index(queue);
1468 if (!rtnl_trylock())
1469 return restart_syscall();
1471 /* If queue belongs to subordinate dev use its map */
1472 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1474 tc = netdev_txq_to_tc(dev, index);
1480 /* Make sure the subordinate device can't be freed */
1481 get_device(&dev->dev);
1484 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1486 put_device(&dev->dev);
1490 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1491 const char *buf, size_t len)
1493 struct net_device *dev = queue->dev;
1498 if (!netif_is_multiqueue(dev))
1501 if (!capable(CAP_NET_ADMIN))
1504 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1507 index = get_netdev_queue_index(queue);
1509 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1511 free_cpumask_var(mask);
1515 if (!rtnl_trylock()) {
1516 free_cpumask_var(mask);
1517 return restart_syscall();
1520 err = netif_set_xps_queue(dev, mask, index);
1523 free_cpumask_var(mask);
1528 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1529 = __ATTR_RW(xps_cpus);
1531 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1533 struct net_device *dev = queue->dev;
1537 index = get_netdev_queue_index(queue);
1539 if (!rtnl_trylock())
1540 return restart_syscall();
1542 tc = netdev_txq_to_tc(dev, index);
1547 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1550 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1553 struct net_device *dev = queue->dev;
1554 struct net *net = dev_net(dev);
1555 unsigned long *mask;
1559 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1562 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1566 index = get_netdev_queue_index(queue);
1568 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1574 if (!rtnl_trylock()) {
1576 return restart_syscall();
1580 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1589 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1590 = __ATTR_RW(xps_rxqs);
1591 #endif /* CONFIG_XPS */
1593 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1594 &queue_trans_timeout.attr,
1595 &queue_traffic_class.attr,
1597 &xps_cpus_attribute.attr,
1598 &xps_rxqs_attribute.attr,
1599 &queue_tx_maxrate.attr,
1603 ATTRIBUTE_GROUPS(netdev_queue_default);
1605 static void netdev_queue_release(struct kobject *kobj)
1607 struct netdev_queue *queue = to_netdev_queue(kobj);
1609 memset(kobj, 0, sizeof(*kobj));
1610 dev_put_track(queue->dev, &queue->dev_tracker);
1613 static const void *netdev_queue_namespace(struct kobject *kobj)
1615 struct netdev_queue *queue = to_netdev_queue(kobj);
1616 struct device *dev = &queue->dev->dev;
1617 const void *ns = NULL;
1619 if (dev->class && dev->class->ns_type)
1620 ns = dev->class->namespace(dev);
1625 static void netdev_queue_get_ownership(struct kobject *kobj,
1626 kuid_t *uid, kgid_t *gid)
1628 const struct net *net = netdev_queue_namespace(kobj);
1630 net_ns_get_ownership(net, uid, gid);
1633 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1634 .sysfs_ops = &netdev_queue_sysfs_ops,
1635 .release = netdev_queue_release,
1636 .default_groups = netdev_queue_default_groups,
1637 .namespace = netdev_queue_namespace,
1638 .get_ownership = netdev_queue_get_ownership,
1641 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1643 struct netdev_queue *queue = dev->_tx + index;
1644 struct kobject *kobj = &queue->kobj;
1647 /* Kobject_put later will trigger netdev_queue_release call
1648 * which decreases dev refcount: Take that reference here
1650 dev_hold_track(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1652 kobj->kset = dev->queues_kset;
1653 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1659 error = sysfs_create_group(kobj, &dql_group);
1664 kobject_uevent(kobj, KOBJ_ADD);
1672 static int tx_queue_change_owner(struct net_device *ndev, int index,
1673 kuid_t kuid, kgid_t kgid)
1675 struct netdev_queue *queue = ndev->_tx + index;
1676 struct kobject *kobj = &queue->kobj;
1679 error = sysfs_change_owner(kobj, kuid, kgid);
1684 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1688 #endif /* CONFIG_SYSFS */
1691 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1697 /* Tx queue kobjects are allowed to be updated when a device is being
1698 * unregistered, but solely to remove queues from qdiscs. Any path
1699 * adding queues should be fixed.
1701 WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1702 "New queues can't be registered after device unregistration.");
1704 for (i = old_num; i < new_num; i++) {
1705 error = netdev_queue_add_kobject(dev, i);
1712 while (--i >= new_num) {
1713 struct netdev_queue *queue = dev->_tx + i;
1715 if (!refcount_read(&dev_net(dev)->ns.count))
1716 queue->kobj.uevent_suppress = 1;
1718 sysfs_remove_group(&queue->kobj, &dql_group);
1720 kobject_put(&queue->kobj);
1726 #endif /* CONFIG_SYSFS */
1729 static int net_tx_queue_change_owner(struct net_device *dev, int num,
1730 kuid_t kuid, kgid_t kgid)
1736 for (i = 0; i < num; i++) {
1737 error = tx_queue_change_owner(dev, i, kuid, kgid);
1745 #endif /* CONFIG_SYSFS */
1748 static int register_queue_kobjects(struct net_device *dev)
1750 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1753 dev->queues_kset = kset_create_and_add("queues",
1754 NULL, &dev->dev.kobj);
1755 if (!dev->queues_kset)
1757 real_rx = dev->real_num_rx_queues;
1759 real_tx = dev->real_num_tx_queues;
1761 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1766 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1774 netdev_queue_update_kobjects(dev, txq, 0);
1775 net_rx_queue_update_kobjects(dev, rxq, 0);
1777 kset_unregister(dev->queues_kset);
1782 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1784 int error = 0, real_rx = 0, real_tx = 0;
1787 if (ndev->queues_kset) {
1788 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1792 real_rx = ndev->real_num_rx_queues;
1794 real_tx = ndev->real_num_tx_queues;
1796 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1800 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1807 static void remove_queue_kobjects(struct net_device *dev)
1809 int real_rx = 0, real_tx = 0;
1812 real_rx = dev->real_num_rx_queues;
1814 real_tx = dev->real_num_tx_queues;
1816 net_rx_queue_update_kobjects(dev, real_rx, 0);
1817 netdev_queue_update_kobjects(dev, real_tx, 0);
1819 dev->real_num_rx_queues = 0;
1820 dev->real_num_tx_queues = 0;
1822 kset_unregister(dev->queues_kset);
1826 static bool net_current_may_mount(void)
1828 struct net *net = current->nsproxy->net_ns;
1830 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1833 static void *net_grab_current_ns(void)
1835 struct net *ns = current->nsproxy->net_ns;
1836 #ifdef CONFIG_NET_NS
1838 refcount_inc(&ns->passive);
1843 static const void *net_initial_ns(void)
1848 static const void *net_netlink_ns(struct sock *sk)
1850 return sock_net(sk);
1853 const struct kobj_ns_type_operations net_ns_type_operations = {
1854 .type = KOBJ_NS_TYPE_NET,
1855 .current_may_mount = net_current_may_mount,
1856 .grab_current_ns = net_grab_current_ns,
1857 .netlink_ns = net_netlink_ns,
1858 .initial_ns = net_initial_ns,
1859 .drop_ns = net_drop_ns,
1861 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1863 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1865 struct net_device *dev = to_net_dev(d);
1868 /* pass interface to uevent. */
1869 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1873 /* pass ifindex to uevent.
1874 * ifindex is useful as it won't change (interface name may change)
1875 * and is what RtNetlink uses natively.
1877 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1884 * netdev_release -- destroy and free a dead device.
1885 * Called when last reference to device kobject is gone.
1887 static void netdev_release(struct device *d)
1889 struct net_device *dev = to_net_dev(d);
1891 BUG_ON(dev->reg_state != NETREG_RELEASED);
1893 /* no need to wait for rcu grace period:
1894 * device is dead and about to be freed.
1896 kfree(rcu_access_pointer(dev->ifalias));
1897 netdev_freemem(dev);
1900 static const void *net_namespace(struct device *d)
1902 struct net_device *dev = to_net_dev(d);
1904 return dev_net(dev);
1907 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1909 struct net_device *dev = to_net_dev(d);
1910 const struct net *net = dev_net(dev);
1912 net_ns_get_ownership(net, uid, gid);
1915 static struct class net_class __ro_after_init = {
1917 .dev_release = netdev_release,
1918 .dev_groups = net_class_groups,
1919 .dev_uevent = netdev_uevent,
1920 .ns_type = &net_ns_type_operations,
1921 .namespace = net_namespace,
1922 .get_ownership = net_get_ownership,
1926 static int of_dev_node_match(struct device *dev, const void *data)
1928 for (; dev; dev = dev->parent) {
1929 if (dev->of_node == data)
1937 * of_find_net_device_by_node - lookup the net device for the device node
1938 * @np: OF device node
1940 * Looks up the net_device structure corresponding with the device node.
1941 * If successful, returns a pointer to the net_device with the embedded
1942 * struct device refcount incremented by one, or NULL on failure. The
1943 * refcount must be dropped when done with the net_device.
1945 struct net_device *of_find_net_device_by_node(struct device_node *np)
1949 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1953 return to_net_dev(dev);
1955 EXPORT_SYMBOL(of_find_net_device_by_node);
1958 /* Delete sysfs entries but hold kobject reference until after all
1959 * netdev references are gone.
1961 void netdev_unregister_kobject(struct net_device *ndev)
1963 struct device *dev = &ndev->dev;
1965 if (!refcount_read(&dev_net(ndev)->ns.count))
1966 dev_set_uevent_suppress(dev, 1);
1968 kobject_get(&dev->kobj);
1970 remove_queue_kobjects(ndev);
1972 pm_runtime_set_memalloc_noio(dev, false);
1977 /* Create sysfs entries for network device. */
1978 int netdev_register_kobject(struct net_device *ndev)
1980 struct device *dev = &ndev->dev;
1981 const struct attribute_group **groups = ndev->sysfs_groups;
1984 device_initialize(dev);
1985 dev->class = &net_class;
1986 dev->platform_data = ndev;
1987 dev->groups = groups;
1989 dev_set_name(dev, "%s", ndev->name);
1992 /* Allow for a device specific group */
1996 *groups++ = &netstat_group;
1998 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1999 if (ndev->ieee80211_ptr)
2000 *groups++ = &wireless_group;
2001 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
2002 else if (ndev->wireless_handlers)
2003 *groups++ = &wireless_group;
2006 #endif /* CONFIG_SYSFS */
2008 error = device_add(dev);
2012 error = register_queue_kobjects(ndev);
2018 pm_runtime_set_memalloc_noio(dev, true);
2023 /* Change owner for sysfs entries when moving network devices across network
2024 * namespaces owned by different user namespaces.
2026 int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2027 const struct net *net_new)
2029 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2030 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2031 struct device *dev = &ndev->dev;
2034 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2035 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2037 /* The network namespace was changed but the owning user namespace is
2038 * identical so there's no need to change the owner of sysfs entries.
2040 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2043 error = device_change_owner(dev, new_uid, new_gid);
2047 error = queue_change_owner(ndev, new_uid, new_gid);
2054 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2057 return class_create_file_ns(&net_class, class_attr, ns);
2059 EXPORT_SYMBOL(netdev_class_create_file_ns);
2061 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2064 class_remove_file_ns(&net_class, class_attr, ns);
2066 EXPORT_SYMBOL(netdev_class_remove_file_ns);
2068 int __init netdev_kobject_init(void)
2070 kobj_ns_type_register(&net_ns_type_operations);
2071 return class_register(&net_class);