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>
28 #include "net-sysfs.h"
31 static const char fmt_hex[] = "%#x\n";
32 static const char fmt_dec[] = "%d\n";
33 static const char fmt_ulong[] = "%lu\n";
34 static const char fmt_u64[] = "%llu\n";
36 /* Caller holds RTNL or dev_base_lock */
37 static inline int dev_isalive(const struct net_device *dev)
39 return dev->reg_state <= NETREG_REGISTERED;
42 /* use same locking rules as GIF* ioctl's */
43 static ssize_t netdev_show(const struct device *dev,
44 struct device_attribute *attr, char *buf,
45 ssize_t (*format)(const struct net_device *, char *))
47 struct net_device *ndev = to_net_dev(dev);
48 ssize_t ret = -EINVAL;
50 read_lock(&dev_base_lock);
51 if (dev_isalive(ndev))
52 ret = (*format)(ndev, buf);
53 read_unlock(&dev_base_lock);
58 /* generate a show function for simple field */
59 #define NETDEVICE_SHOW(field, format_string) \
60 static ssize_t format_##field(const struct net_device *dev, char *buf) \
62 return sysfs_emit(buf, format_string, dev->field); \
64 static ssize_t field##_show(struct device *dev, \
65 struct device_attribute *attr, char *buf) \
67 return netdev_show(dev, attr, buf, format_##field); \
70 #define NETDEVICE_SHOW_RO(field, format_string) \
71 NETDEVICE_SHOW(field, format_string); \
72 static DEVICE_ATTR_RO(field)
74 #define NETDEVICE_SHOW_RW(field, format_string) \
75 NETDEVICE_SHOW(field, format_string); \
76 static DEVICE_ATTR_RW(field)
78 /* use same locking and permission rules as SIF* ioctl's */
79 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
80 const char *buf, size_t len,
81 int (*set)(struct net_device *, unsigned long))
83 struct net_device *netdev = to_net_dev(dev);
84 struct net *net = dev_net(netdev);
88 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
91 ret = kstrtoul(buf, 0, &new);
96 return restart_syscall();
98 if (dev_isalive(netdev)) {
99 ret = (*set)(netdev, new);
108 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
109 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
110 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
111 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
112 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
113 NETDEVICE_SHOW_RO(type, fmt_dec);
114 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
116 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
119 struct net_device *ndev = to_net_dev(dev);
121 return sysfs_emit(buf, fmt_dec, dev_get_iflink(ndev));
123 static DEVICE_ATTR_RO(iflink);
125 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
127 return sysfs_emit(buf, fmt_dec, dev->name_assign_type);
130 static ssize_t name_assign_type_show(struct device *dev,
131 struct device_attribute *attr,
134 struct net_device *ndev = to_net_dev(dev);
135 ssize_t ret = -EINVAL;
137 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
138 ret = netdev_show(dev, attr, buf, format_name_assign_type);
142 static DEVICE_ATTR_RO(name_assign_type);
144 /* use same locking rules as GIFHWADDR ioctl's */
145 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
148 struct net_device *ndev = to_net_dev(dev);
149 ssize_t ret = -EINVAL;
151 read_lock(&dev_base_lock);
152 if (dev_isalive(ndev))
153 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
154 read_unlock(&dev_base_lock);
157 static DEVICE_ATTR_RO(address);
159 static ssize_t broadcast_show(struct device *dev,
160 struct device_attribute *attr, char *buf)
162 struct net_device *ndev = to_net_dev(dev);
164 if (dev_isalive(ndev))
165 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
168 static DEVICE_ATTR_RO(broadcast);
170 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
172 if (!netif_running(dev))
174 return dev_change_carrier(dev, (bool)new_carrier);
177 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
178 const char *buf, size_t len)
180 struct net_device *netdev = to_net_dev(dev);
182 /* The check is also done in change_carrier; this helps returning early
183 * without hitting the trylock/restart in netdev_store.
185 if (!netdev->netdev_ops->ndo_change_carrier)
188 return netdev_store(dev, attr, buf, len, change_carrier);
191 static ssize_t carrier_show(struct device *dev,
192 struct device_attribute *attr, char *buf)
194 struct net_device *netdev = to_net_dev(dev);
196 if (netif_running(netdev))
197 return sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev));
201 static DEVICE_ATTR_RW(carrier);
203 static ssize_t speed_show(struct device *dev,
204 struct device_attribute *attr, char *buf)
206 struct net_device *netdev = to_net_dev(dev);
209 /* The check is also done in __ethtool_get_link_ksettings; this helps
210 * returning early without hitting the trylock/restart below.
212 if (!netdev->ethtool_ops->get_link_ksettings)
216 return restart_syscall();
218 if (netif_running(netdev) && netif_device_present(netdev)) {
219 struct ethtool_link_ksettings cmd;
221 if (!__ethtool_get_link_ksettings(netdev, &cmd))
222 ret = sysfs_emit(buf, fmt_dec, cmd.base.speed);
227 static DEVICE_ATTR_RO(speed);
229 static ssize_t duplex_show(struct device *dev,
230 struct device_attribute *attr, char *buf)
232 struct net_device *netdev = to_net_dev(dev);
235 /* The check is also done in __ethtool_get_link_ksettings; this helps
236 * returning early without hitting the trylock/restart below.
238 if (!netdev->ethtool_ops->get_link_ksettings)
242 return restart_syscall();
244 if (netif_running(netdev)) {
245 struct ethtool_link_ksettings cmd;
247 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
250 switch (cmd.base.duplex) {
261 ret = sysfs_emit(buf, "%s\n", duplex);
267 static DEVICE_ATTR_RO(duplex);
269 static ssize_t testing_show(struct device *dev,
270 struct device_attribute *attr, char *buf)
272 struct net_device *netdev = to_net_dev(dev);
274 if (netif_running(netdev))
275 return sysfs_emit(buf, fmt_dec, !!netif_testing(netdev));
279 static DEVICE_ATTR_RO(testing);
281 static ssize_t dormant_show(struct device *dev,
282 struct device_attribute *attr, char *buf)
284 struct net_device *netdev = to_net_dev(dev);
286 if (netif_running(netdev))
287 return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev));
291 static DEVICE_ATTR_RO(dormant);
293 static const char *const operstates[] = {
295 "notpresent", /* currently unused */
303 static ssize_t operstate_show(struct device *dev,
304 struct device_attribute *attr, char *buf)
306 const struct net_device *netdev = to_net_dev(dev);
307 unsigned char operstate;
309 operstate = READ_ONCE(netdev->operstate);
310 if (!netif_running(netdev))
311 operstate = IF_OPER_DOWN;
313 if (operstate >= ARRAY_SIZE(operstates))
314 return -EINVAL; /* should not happen */
316 return sysfs_emit(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 sysfs_emit(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 sysfs_emit(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 sysfs_emit(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 = sysfs_emit(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 = sysfs_emit(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 = sysfs_emit(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 = sysfs_emit(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 = sysfs_emit(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 = sysfs_emit(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 static struct attribute *wireless_attrs[] = {
752 static const struct attribute_group wireless_group = {
754 .attrs = wireless_attrs,
757 static bool wireless_group_needed(struct net_device *ndev)
759 #if IS_ENABLED(CONFIG_CFG80211)
760 if (ndev->ieee80211_ptr)
763 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
764 if (ndev->wireless_handlers)
770 #else /* CONFIG_SYSFS */
771 #define net_class_groups NULL
772 #endif /* CONFIG_SYSFS */
775 #define to_rx_queue_attr(_attr) \
776 container_of(_attr, struct rx_queue_attribute, attr)
778 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
780 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
783 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
784 struct netdev_rx_queue *queue = to_rx_queue(kobj);
786 if (!attribute->show)
789 return attribute->show(queue, buf);
792 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
793 const char *buf, size_t count)
795 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
796 struct netdev_rx_queue *queue = to_rx_queue(kobj);
798 if (!attribute->store)
801 return attribute->store(queue, buf, count);
804 static const struct sysfs_ops rx_queue_sysfs_ops = {
805 .show = rx_queue_attr_show,
806 .store = rx_queue_attr_store,
810 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
816 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
820 map = rcu_dereference(queue->rps_map);
822 for (i = 0; i < map->len; i++)
823 cpumask_set_cpu(map->cpus[i], mask);
825 len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask));
827 free_cpumask_var(mask);
829 return len < PAGE_SIZE ? len : -EINVAL;
832 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
833 const char *buf, size_t len)
835 struct rps_map *old_map, *map;
838 static DEFINE_MUTEX(rps_map_mutex);
840 if (!capable(CAP_NET_ADMIN))
843 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
846 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
848 free_cpumask_var(mask);
852 if (!cpumask_empty(mask)) {
853 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
854 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
855 if (cpumask_empty(mask)) {
856 free_cpumask_var(mask);
861 map = kzalloc(max_t(unsigned int,
862 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
865 free_cpumask_var(mask);
870 for_each_cpu_and(cpu, mask, cpu_online_mask)
871 map->cpus[i++] = cpu;
880 mutex_lock(&rps_map_mutex);
881 old_map = rcu_dereference_protected(queue->rps_map,
882 mutex_is_locked(&rps_map_mutex));
883 rcu_assign_pointer(queue->rps_map, map);
886 static_branch_inc(&rps_needed);
888 static_branch_dec(&rps_needed);
890 mutex_unlock(&rps_map_mutex);
893 kfree_rcu(old_map, rcu);
895 free_cpumask_var(mask);
899 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
902 struct rps_dev_flow_table *flow_table;
903 unsigned long val = 0;
906 flow_table = rcu_dereference(queue->rps_flow_table);
908 val = (unsigned long)flow_table->mask + 1;
911 return sysfs_emit(buf, "%lu\n", val);
914 static void rps_dev_flow_table_release(struct rcu_head *rcu)
916 struct rps_dev_flow_table *table = container_of(rcu,
917 struct rps_dev_flow_table, rcu);
921 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
922 const char *buf, size_t len)
924 unsigned long mask, count;
925 struct rps_dev_flow_table *table, *old_table;
926 static DEFINE_SPINLOCK(rps_dev_flow_lock);
929 if (!capable(CAP_NET_ADMIN))
932 rc = kstrtoul(buf, 0, &count);
938 /* mask = roundup_pow_of_two(count) - 1;
939 * without overflows...
941 while ((mask | (mask >> 1)) != mask)
943 /* On 64 bit arches, must check mask fits in table->mask (u32),
944 * and on 32bit arches, must check
945 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
947 #if BITS_PER_LONG > 32
948 if (mask > (unsigned long)(u32)mask)
951 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
952 / sizeof(struct rps_dev_flow)) {
953 /* Enforce a limit to prevent overflow */
957 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
962 for (count = 0; count <= mask; count++)
963 table->flows[count].cpu = RPS_NO_CPU;
968 spin_lock(&rps_dev_flow_lock);
969 old_table = rcu_dereference_protected(queue->rps_flow_table,
970 lockdep_is_held(&rps_dev_flow_lock));
971 rcu_assign_pointer(queue->rps_flow_table, table);
972 spin_unlock(&rps_dev_flow_lock);
975 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
980 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
981 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
983 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
984 = __ATTR(rps_flow_cnt, 0644,
985 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
986 #endif /* CONFIG_RPS */
988 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
990 &rps_cpus_attribute.attr,
991 &rps_dev_flow_table_cnt_attribute.attr,
995 ATTRIBUTE_GROUPS(rx_queue_default);
997 static void rx_queue_release(struct kobject *kobj)
999 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1001 struct rps_map *map;
1002 struct rps_dev_flow_table *flow_table;
1004 map = rcu_dereference_protected(queue->rps_map, 1);
1006 RCU_INIT_POINTER(queue->rps_map, NULL);
1007 kfree_rcu(map, rcu);
1010 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1012 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1013 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1017 memset(kobj, 0, sizeof(*kobj));
1018 netdev_put(queue->dev, &queue->dev_tracker);
1021 static const void *rx_queue_namespace(struct kobject *kobj)
1023 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1024 struct device *dev = &queue->dev->dev;
1025 const void *ns = NULL;
1027 if (dev->class && dev->class->ns_type)
1028 ns = dev->class->namespace(dev);
1033 static void rx_queue_get_ownership(struct kobject *kobj,
1034 kuid_t *uid, kgid_t *gid)
1036 const struct net *net = rx_queue_namespace(kobj);
1038 net_ns_get_ownership(net, uid, gid);
1041 static struct kobj_type rx_queue_ktype __ro_after_init = {
1042 .sysfs_ops = &rx_queue_sysfs_ops,
1043 .release = rx_queue_release,
1044 .default_groups = rx_queue_default_groups,
1045 .namespace = rx_queue_namespace,
1046 .get_ownership = rx_queue_get_ownership,
1049 static int rx_queue_add_kobject(struct net_device *dev, int index)
1051 struct netdev_rx_queue *queue = dev->_rx + index;
1052 struct kobject *kobj = &queue->kobj;
1055 /* Kobject_put later will trigger rx_queue_release call which
1056 * decreases dev refcount: Take that reference here
1058 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1060 kobj->kset = dev->queues_kset;
1061 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1066 if (dev->sysfs_rx_queue_group) {
1067 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1072 kobject_uevent(kobj, KOBJ_ADD);
1081 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1084 struct netdev_rx_queue *queue = dev->_rx + index;
1085 struct kobject *kobj = &queue->kobj;
1088 error = sysfs_change_owner(kobj, kuid, kgid);
1092 if (dev->sysfs_rx_queue_group)
1093 error = sysfs_group_change_owner(
1094 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1098 #endif /* CONFIG_SYSFS */
1101 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1108 if (!dev->sysfs_rx_queue_group)
1111 for (i = old_num; i < new_num; i++) {
1112 error = rx_queue_add_kobject(dev, i);
1119 while (--i >= new_num) {
1120 struct kobject *kobj = &dev->_rx[i].kobj;
1122 if (!refcount_read(&dev_net(dev)->ns.count))
1123 kobj->uevent_suppress = 1;
1124 if (dev->sysfs_rx_queue_group)
1125 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1135 static int net_rx_queue_change_owner(struct net_device *dev, int num,
1136 kuid_t kuid, kgid_t kgid)
1143 if (!dev->sysfs_rx_queue_group)
1146 for (i = 0; i < num; i++) {
1147 error = rx_queue_change_owner(dev, i, kuid, kgid);
1160 * netdev_queue sysfs structures and functions.
1162 struct netdev_queue_attribute {
1163 struct attribute attr;
1164 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1165 ssize_t (*store)(struct netdev_queue *queue,
1166 const char *buf, size_t len);
1168 #define to_netdev_queue_attr(_attr) \
1169 container_of(_attr, struct netdev_queue_attribute, attr)
1171 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1173 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1174 struct attribute *attr, char *buf)
1176 const struct netdev_queue_attribute *attribute
1177 = to_netdev_queue_attr(attr);
1178 struct netdev_queue *queue = to_netdev_queue(kobj);
1180 if (!attribute->show)
1183 return attribute->show(queue, buf);
1186 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1187 struct attribute *attr,
1188 const char *buf, size_t count)
1190 const struct netdev_queue_attribute *attribute
1191 = to_netdev_queue_attr(attr);
1192 struct netdev_queue *queue = to_netdev_queue(kobj);
1194 if (!attribute->store)
1197 return attribute->store(queue, buf, count);
1200 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1201 .show = netdev_queue_attr_show,
1202 .store = netdev_queue_attr_store,
1205 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1207 unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1209 return sysfs_emit(buf, fmt_ulong, trans_timeout);
1212 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1214 struct net_device *dev = queue->dev;
1217 i = queue - dev->_tx;
1218 BUG_ON(i >= dev->num_tx_queues);
1223 static ssize_t traffic_class_show(struct netdev_queue *queue,
1226 struct net_device *dev = queue->dev;
1230 if (!netif_is_multiqueue(dev))
1233 if (!rtnl_trylock())
1234 return restart_syscall();
1236 index = get_netdev_queue_index(queue);
1238 /* If queue belongs to subordinate dev use its TC mapping */
1239 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1241 num_tc = dev->num_tc;
1242 tc = netdev_txq_to_tc(dev, index);
1249 /* We can report the traffic class one of two ways:
1250 * Subordinate device traffic classes are reported with the traffic
1251 * class first, and then the subordinate class so for example TC0 on
1252 * subordinate device 2 will be reported as "0-2". If the queue
1253 * belongs to the root device it will be reported with just the
1254 * traffic class, so just "0" for TC 0 for example.
1256 return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) :
1257 sysfs_emit(buf, "%d\n", tc);
1261 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1264 return sysfs_emit(buf, "%lu\n", queue->tx_maxrate);
1267 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1268 const char *buf, size_t len)
1270 struct net_device *dev = queue->dev;
1271 int err, index = get_netdev_queue_index(queue);
1274 if (!capable(CAP_NET_ADMIN))
1277 /* The check is also done later; this helps returning early without
1278 * hitting the trylock/restart below.
1280 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1283 err = kstrtou32(buf, 10, &rate);
1287 if (!rtnl_trylock())
1288 return restart_syscall();
1291 if (dev->netdev_ops->ndo_set_tx_maxrate)
1292 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1296 queue->tx_maxrate = rate;
1302 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1303 = __ATTR_RW(tx_maxrate);
1306 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1307 = __ATTR_RO(tx_timeout);
1309 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1310 = __ATTR_RO(traffic_class);
1314 * Byte queue limits sysfs structures and functions.
1316 static ssize_t bql_show(char *buf, unsigned int value)
1318 return sysfs_emit(buf, "%u\n", value);
1321 static ssize_t bql_set(const char *buf, const size_t count,
1322 unsigned int *pvalue)
1327 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1328 value = DQL_MAX_LIMIT;
1330 err = kstrtouint(buf, 10, &value);
1333 if (value > DQL_MAX_LIMIT)
1342 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1345 struct dql *dql = &queue->dql;
1347 return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1350 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1351 const char *buf, size_t len)
1353 struct dql *dql = &queue->dql;
1357 err = kstrtouint(buf, 10, &value);
1361 dql->slack_hold_time = msecs_to_jiffies(value);
1366 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1367 = __ATTR(hold_time, 0644,
1368 bql_show_hold_time, bql_set_hold_time);
1370 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1373 struct dql *dql = &queue->dql;
1375 return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed);
1378 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1379 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1381 #define BQL_ATTR(NAME, FIELD) \
1382 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1385 return bql_show(buf, queue->dql.FIELD); \
1388 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1389 const char *buf, size_t len) \
1391 return bql_set(buf, len, &queue->dql.FIELD); \
1394 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1395 = __ATTR(NAME, 0644, \
1396 bql_show_ ## NAME, bql_set_ ## NAME)
1398 BQL_ATTR(limit, limit);
1399 BQL_ATTR(limit_max, max_limit);
1400 BQL_ATTR(limit_min, min_limit);
1402 static struct attribute *dql_attrs[] __ro_after_init = {
1403 &bql_limit_attribute.attr,
1404 &bql_limit_max_attribute.attr,
1405 &bql_limit_min_attribute.attr,
1406 &bql_hold_time_attribute.attr,
1407 &bql_inflight_attribute.attr,
1411 static const struct attribute_group dql_group = {
1412 .name = "byte_queue_limits",
1415 #endif /* CONFIG_BQL */
1418 static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1419 int tc, char *buf, enum xps_map_type type)
1421 struct xps_dev_maps *dev_maps;
1422 unsigned long *mask;
1423 unsigned int nr_ids;
1427 dev_maps = rcu_dereference(dev->xps_maps[type]);
1429 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1430 * when dev_maps hasn't been allocated yet, to be backward compatible.
1432 nr_ids = dev_maps ? dev_maps->nr_ids :
1433 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1435 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1441 if (!dev_maps || tc >= dev_maps->num_tc)
1444 for (j = 0; j < nr_ids; j++) {
1445 int i, tci = j * dev_maps->num_tc + tc;
1446 struct xps_map *map;
1448 map = rcu_dereference(dev_maps->attr_map[tci]);
1452 for (i = map->len; i--;) {
1453 if (map->queues[i] == index) {
1462 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1465 return len < PAGE_SIZE ? len : -EINVAL;
1468 static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1470 struct net_device *dev = queue->dev;
1474 if (!netif_is_multiqueue(dev))
1477 index = get_netdev_queue_index(queue);
1479 if (!rtnl_trylock())
1480 return restart_syscall();
1482 /* If queue belongs to subordinate dev use its map */
1483 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1485 tc = netdev_txq_to_tc(dev, index);
1491 /* Make sure the subordinate device can't be freed */
1492 get_device(&dev->dev);
1495 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1497 put_device(&dev->dev);
1501 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1502 const char *buf, size_t len)
1504 struct net_device *dev = queue->dev;
1509 if (!netif_is_multiqueue(dev))
1512 if (!capable(CAP_NET_ADMIN))
1515 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1518 index = get_netdev_queue_index(queue);
1520 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1522 free_cpumask_var(mask);
1526 if (!rtnl_trylock()) {
1527 free_cpumask_var(mask);
1528 return restart_syscall();
1531 err = netif_set_xps_queue(dev, mask, index);
1534 free_cpumask_var(mask);
1539 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1540 = __ATTR_RW(xps_cpus);
1542 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1544 struct net_device *dev = queue->dev;
1548 index = get_netdev_queue_index(queue);
1550 if (!rtnl_trylock())
1551 return restart_syscall();
1553 tc = netdev_txq_to_tc(dev, index);
1558 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1561 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1564 struct net_device *dev = queue->dev;
1565 struct net *net = dev_net(dev);
1566 unsigned long *mask;
1570 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1573 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1577 index = get_netdev_queue_index(queue);
1579 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1585 if (!rtnl_trylock()) {
1587 return restart_syscall();
1591 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1600 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1601 = __ATTR_RW(xps_rxqs);
1602 #endif /* CONFIG_XPS */
1604 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1605 &queue_trans_timeout.attr,
1606 &queue_traffic_class.attr,
1608 &xps_cpus_attribute.attr,
1609 &xps_rxqs_attribute.attr,
1610 &queue_tx_maxrate.attr,
1614 ATTRIBUTE_GROUPS(netdev_queue_default);
1616 static void netdev_queue_release(struct kobject *kobj)
1618 struct netdev_queue *queue = to_netdev_queue(kobj);
1620 memset(kobj, 0, sizeof(*kobj));
1621 netdev_put(queue->dev, &queue->dev_tracker);
1624 static const void *netdev_queue_namespace(struct kobject *kobj)
1626 struct netdev_queue *queue = to_netdev_queue(kobj);
1627 struct device *dev = &queue->dev->dev;
1628 const void *ns = NULL;
1630 if (dev->class && dev->class->ns_type)
1631 ns = dev->class->namespace(dev);
1636 static void netdev_queue_get_ownership(struct kobject *kobj,
1637 kuid_t *uid, kgid_t *gid)
1639 const struct net *net = netdev_queue_namespace(kobj);
1641 net_ns_get_ownership(net, uid, gid);
1644 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1645 .sysfs_ops = &netdev_queue_sysfs_ops,
1646 .release = netdev_queue_release,
1647 .default_groups = netdev_queue_default_groups,
1648 .namespace = netdev_queue_namespace,
1649 .get_ownership = netdev_queue_get_ownership,
1652 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1654 struct netdev_queue *queue = dev->_tx + index;
1655 struct kobject *kobj = &queue->kobj;
1658 /* Kobject_put later will trigger netdev_queue_release call
1659 * which decreases dev refcount: Take that reference here
1661 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1663 kobj->kset = dev->queues_kset;
1664 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1670 error = sysfs_create_group(kobj, &dql_group);
1675 kobject_uevent(kobj, KOBJ_ADD);
1683 static int tx_queue_change_owner(struct net_device *ndev, int index,
1684 kuid_t kuid, kgid_t kgid)
1686 struct netdev_queue *queue = ndev->_tx + index;
1687 struct kobject *kobj = &queue->kobj;
1690 error = sysfs_change_owner(kobj, kuid, kgid);
1695 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1699 #endif /* CONFIG_SYSFS */
1702 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1708 /* Tx queue kobjects are allowed to be updated when a device is being
1709 * unregistered, but solely to remove queues from qdiscs. Any path
1710 * adding queues should be fixed.
1712 WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1713 "New queues can't be registered after device unregistration.");
1715 for (i = old_num; i < new_num; i++) {
1716 error = netdev_queue_add_kobject(dev, i);
1723 while (--i >= new_num) {
1724 struct netdev_queue *queue = dev->_tx + i;
1726 if (!refcount_read(&dev_net(dev)->ns.count))
1727 queue->kobj.uevent_suppress = 1;
1729 sysfs_remove_group(&queue->kobj, &dql_group);
1731 kobject_put(&queue->kobj);
1737 #endif /* CONFIG_SYSFS */
1740 static int net_tx_queue_change_owner(struct net_device *dev, int num,
1741 kuid_t kuid, kgid_t kgid)
1747 for (i = 0; i < num; i++) {
1748 error = tx_queue_change_owner(dev, i, kuid, kgid);
1756 #endif /* CONFIG_SYSFS */
1759 static int register_queue_kobjects(struct net_device *dev)
1761 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1764 dev->queues_kset = kset_create_and_add("queues",
1765 NULL, &dev->dev.kobj);
1766 if (!dev->queues_kset)
1768 real_rx = dev->real_num_rx_queues;
1770 real_tx = dev->real_num_tx_queues;
1772 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1777 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1785 netdev_queue_update_kobjects(dev, txq, 0);
1786 net_rx_queue_update_kobjects(dev, rxq, 0);
1788 kset_unregister(dev->queues_kset);
1793 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1795 int error = 0, real_rx = 0, real_tx = 0;
1798 if (ndev->queues_kset) {
1799 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1803 real_rx = ndev->real_num_rx_queues;
1805 real_tx = ndev->real_num_tx_queues;
1807 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1811 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1818 static void remove_queue_kobjects(struct net_device *dev)
1820 int real_rx = 0, real_tx = 0;
1823 real_rx = dev->real_num_rx_queues;
1825 real_tx = dev->real_num_tx_queues;
1827 net_rx_queue_update_kobjects(dev, real_rx, 0);
1828 netdev_queue_update_kobjects(dev, real_tx, 0);
1830 dev->real_num_rx_queues = 0;
1831 dev->real_num_tx_queues = 0;
1833 kset_unregister(dev->queues_kset);
1837 static bool net_current_may_mount(void)
1839 struct net *net = current->nsproxy->net_ns;
1841 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1844 static void *net_grab_current_ns(void)
1846 struct net *ns = current->nsproxy->net_ns;
1847 #ifdef CONFIG_NET_NS
1849 refcount_inc(&ns->passive);
1854 static const void *net_initial_ns(void)
1859 static const void *net_netlink_ns(struct sock *sk)
1861 return sock_net(sk);
1864 const struct kobj_ns_type_operations net_ns_type_operations = {
1865 .type = KOBJ_NS_TYPE_NET,
1866 .current_may_mount = net_current_may_mount,
1867 .grab_current_ns = net_grab_current_ns,
1868 .netlink_ns = net_netlink_ns,
1869 .initial_ns = net_initial_ns,
1870 .drop_ns = net_drop_ns,
1872 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1874 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1876 struct net_device *dev = to_net_dev(d);
1879 /* pass interface to uevent. */
1880 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1884 /* pass ifindex to uevent.
1885 * ifindex is useful as it won't change (interface name may change)
1886 * and is what RtNetlink uses natively.
1888 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1895 * netdev_release -- destroy and free a dead device.
1896 * Called when last reference to device kobject is gone.
1898 static void netdev_release(struct device *d)
1900 struct net_device *dev = to_net_dev(d);
1902 BUG_ON(dev->reg_state != NETREG_RELEASED);
1904 /* no need to wait for rcu grace period:
1905 * device is dead and about to be freed.
1907 kfree(rcu_access_pointer(dev->ifalias));
1908 netdev_freemem(dev);
1911 static const void *net_namespace(struct device *d)
1913 struct net_device *dev = to_net_dev(d);
1915 return dev_net(dev);
1918 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1920 struct net_device *dev = to_net_dev(d);
1921 const struct net *net = dev_net(dev);
1923 net_ns_get_ownership(net, uid, gid);
1926 static struct class net_class __ro_after_init = {
1928 .dev_release = netdev_release,
1929 .dev_groups = net_class_groups,
1930 .dev_uevent = netdev_uevent,
1931 .ns_type = &net_ns_type_operations,
1932 .namespace = net_namespace,
1933 .get_ownership = net_get_ownership,
1937 static int of_dev_node_match(struct device *dev, const void *data)
1939 for (; dev; dev = dev->parent) {
1940 if (dev->of_node == data)
1948 * of_find_net_device_by_node - lookup the net device for the device node
1949 * @np: OF device node
1951 * Looks up the net_device structure corresponding with the device node.
1952 * If successful, returns a pointer to the net_device with the embedded
1953 * struct device refcount incremented by one, or NULL on failure. The
1954 * refcount must be dropped when done with the net_device.
1956 struct net_device *of_find_net_device_by_node(struct device_node *np)
1960 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1964 return to_net_dev(dev);
1966 EXPORT_SYMBOL(of_find_net_device_by_node);
1969 /* Delete sysfs entries but hold kobject reference until after all
1970 * netdev references are gone.
1972 void netdev_unregister_kobject(struct net_device *ndev)
1974 struct device *dev = &ndev->dev;
1976 if (!refcount_read(&dev_net(ndev)->ns.count))
1977 dev_set_uevent_suppress(dev, 1);
1979 kobject_get(&dev->kobj);
1981 remove_queue_kobjects(ndev);
1983 pm_runtime_set_memalloc_noio(dev, false);
1988 /* Create sysfs entries for network device. */
1989 int netdev_register_kobject(struct net_device *ndev)
1991 struct device *dev = &ndev->dev;
1992 const struct attribute_group **groups = ndev->sysfs_groups;
1995 device_initialize(dev);
1996 dev->class = &net_class;
1997 dev->platform_data = ndev;
1998 dev->groups = groups;
2000 dev_set_name(dev, "%s", ndev->name);
2003 /* Allow for a device specific group */
2007 *groups++ = &netstat_group;
2009 if (wireless_group_needed(ndev))
2010 *groups++ = &wireless_group;
2011 #endif /* CONFIG_SYSFS */
2013 error = device_add(dev);
2017 error = register_queue_kobjects(ndev);
2023 pm_runtime_set_memalloc_noio(dev, true);
2028 /* Change owner for sysfs entries when moving network devices across network
2029 * namespaces owned by different user namespaces.
2031 int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2032 const struct net *net_new)
2034 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2035 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2036 struct device *dev = &ndev->dev;
2039 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2040 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2042 /* The network namespace was changed but the owning user namespace is
2043 * identical so there's no need to change the owner of sysfs entries.
2045 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2048 error = device_change_owner(dev, new_uid, new_gid);
2052 error = queue_change_owner(ndev, new_uid, new_gid);
2059 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2062 return class_create_file_ns(&net_class, class_attr, ns);
2064 EXPORT_SYMBOL(netdev_class_create_file_ns);
2066 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2069 class_remove_file_ns(&net_class, class_attr, ns);
2071 EXPORT_SYMBOL(netdev_class_remove_file_ns);
2073 int __init netdev_kobject_init(void)
2075 kobj_ns_type_register(&net_ns_type_operations);
2076 return class_register(&net_class);