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>
26 #include <net/netdev_rx_queue.h>
29 #include "net-sysfs.h"
32 static const char fmt_hex[] = "%#x\n";
33 static const char fmt_dec[] = "%d\n";
34 static const char fmt_ulong[] = "%lu\n";
35 static const char fmt_u64[] = "%llu\n";
37 /* Caller holds RTNL or dev_base_lock */
38 static inline int dev_isalive(const struct net_device *dev)
40 return dev->reg_state <= NETREG_REGISTERED;
43 /* use same locking rules as GIF* ioctl's */
44 static ssize_t netdev_show(const struct device *dev,
45 struct device_attribute *attr, char *buf,
46 ssize_t (*format)(const struct net_device *, char *))
48 struct net_device *ndev = to_net_dev(dev);
49 ssize_t ret = -EINVAL;
51 read_lock(&dev_base_lock);
52 if (dev_isalive(ndev))
53 ret = (*format)(ndev, buf);
54 read_unlock(&dev_base_lock);
59 /* generate a show function for simple field */
60 #define NETDEVICE_SHOW(field, format_string) \
61 static ssize_t format_##field(const struct net_device *dev, char *buf) \
63 return sysfs_emit(buf, format_string, dev->field); \
65 static ssize_t field##_show(struct device *dev, \
66 struct device_attribute *attr, char *buf) \
68 return netdev_show(dev, attr, buf, format_##field); \
71 #define NETDEVICE_SHOW_RO(field, format_string) \
72 NETDEVICE_SHOW(field, format_string); \
73 static DEVICE_ATTR_RO(field)
75 #define NETDEVICE_SHOW_RW(field, format_string) \
76 NETDEVICE_SHOW(field, format_string); \
77 static DEVICE_ATTR_RW(field)
79 /* use same locking and permission rules as SIF* ioctl's */
80 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
81 const char *buf, size_t len,
82 int (*set)(struct net_device *, unsigned long))
84 struct net_device *netdev = to_net_dev(dev);
85 struct net *net = dev_net(netdev);
89 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
92 ret = kstrtoul(buf, 0, &new);
97 return restart_syscall();
99 if (dev_isalive(netdev)) {
100 ret = (*set)(netdev, new);
109 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
110 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
111 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
112 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
113 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
114 NETDEVICE_SHOW_RO(type, fmt_dec);
115 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
117 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
120 struct net_device *ndev = to_net_dev(dev);
122 return sysfs_emit(buf, fmt_dec, dev_get_iflink(ndev));
124 static DEVICE_ATTR_RO(iflink);
126 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
128 return sysfs_emit(buf, fmt_dec, dev->name_assign_type);
131 static ssize_t name_assign_type_show(struct device *dev,
132 struct device_attribute *attr,
135 struct net_device *ndev = to_net_dev(dev);
136 ssize_t ret = -EINVAL;
138 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
139 ret = netdev_show(dev, attr, buf, format_name_assign_type);
143 static DEVICE_ATTR_RO(name_assign_type);
145 /* use same locking rules as GIFHWADDR ioctl's */
146 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
149 struct net_device *ndev = to_net_dev(dev);
150 ssize_t ret = -EINVAL;
152 read_lock(&dev_base_lock);
153 if (dev_isalive(ndev))
154 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
155 read_unlock(&dev_base_lock);
158 static DEVICE_ATTR_RO(address);
160 static ssize_t broadcast_show(struct device *dev,
161 struct device_attribute *attr, char *buf)
163 struct net_device *ndev = to_net_dev(dev);
165 if (dev_isalive(ndev))
166 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
169 static DEVICE_ATTR_RO(broadcast);
171 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
173 if (!netif_running(dev))
175 return dev_change_carrier(dev, (bool)new_carrier);
178 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
179 const char *buf, size_t len)
181 struct net_device *netdev = to_net_dev(dev);
183 /* The check is also done in change_carrier; this helps returning early
184 * without hitting the trylock/restart in netdev_store.
186 if (!netdev->netdev_ops->ndo_change_carrier)
189 return netdev_store(dev, attr, buf, len, change_carrier);
192 static ssize_t carrier_show(struct device *dev,
193 struct device_attribute *attr, char *buf)
195 struct net_device *netdev = to_net_dev(dev);
199 return restart_syscall();
201 if (netif_running(netdev)) {
202 /* Synchronize carrier state with link watch,
203 * see also rtnl_getlink().
205 linkwatch_sync_dev(netdev);
207 ret = sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev));
213 static DEVICE_ATTR_RW(carrier);
215 static ssize_t speed_show(struct device *dev,
216 struct device_attribute *attr, char *buf)
218 struct net_device *netdev = to_net_dev(dev);
221 /* The check is also done in __ethtool_get_link_ksettings; this helps
222 * returning early without hitting the trylock/restart below.
224 if (!netdev->ethtool_ops->get_link_ksettings)
228 return restart_syscall();
230 if (netif_running(netdev) && netif_device_present(netdev)) {
231 struct ethtool_link_ksettings cmd;
233 if (!__ethtool_get_link_ksettings(netdev, &cmd))
234 ret = sysfs_emit(buf, fmt_dec, cmd.base.speed);
239 static DEVICE_ATTR_RO(speed);
241 static ssize_t duplex_show(struct device *dev,
242 struct device_attribute *attr, char *buf)
244 struct net_device *netdev = to_net_dev(dev);
247 /* The check is also done in __ethtool_get_link_ksettings; this helps
248 * returning early without hitting the trylock/restart below.
250 if (!netdev->ethtool_ops->get_link_ksettings)
254 return restart_syscall();
256 if (netif_running(netdev)) {
257 struct ethtool_link_ksettings cmd;
259 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
262 switch (cmd.base.duplex) {
273 ret = sysfs_emit(buf, "%s\n", duplex);
279 static DEVICE_ATTR_RO(duplex);
281 static ssize_t testing_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_testing(netdev));
291 static DEVICE_ATTR_RO(testing);
293 static ssize_t dormant_show(struct device *dev,
294 struct device_attribute *attr, char *buf)
296 struct net_device *netdev = to_net_dev(dev);
298 if (netif_running(netdev))
299 return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev));
303 static DEVICE_ATTR_RO(dormant);
305 static const char *const operstates[] = {
307 "notpresent", /* currently unused */
315 static ssize_t operstate_show(struct device *dev,
316 struct device_attribute *attr, char *buf)
318 const struct net_device *netdev = to_net_dev(dev);
319 unsigned char operstate;
321 read_lock(&dev_base_lock);
322 operstate = netdev->operstate;
323 if (!netif_running(netdev))
324 operstate = IF_OPER_DOWN;
325 read_unlock(&dev_base_lock);
327 if (operstate >= ARRAY_SIZE(operstates))
328 return -EINVAL; /* should not happen */
330 return sysfs_emit(buf, "%s\n", operstates[operstate]);
332 static DEVICE_ATTR_RO(operstate);
334 static ssize_t carrier_changes_show(struct device *dev,
335 struct device_attribute *attr,
338 struct net_device *netdev = to_net_dev(dev);
340 return sysfs_emit(buf, fmt_dec,
341 atomic_read(&netdev->carrier_up_count) +
342 atomic_read(&netdev->carrier_down_count));
344 static DEVICE_ATTR_RO(carrier_changes);
346 static ssize_t carrier_up_count_show(struct device *dev,
347 struct device_attribute *attr,
350 struct net_device *netdev = to_net_dev(dev);
352 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
354 static DEVICE_ATTR_RO(carrier_up_count);
356 static ssize_t carrier_down_count_show(struct device *dev,
357 struct device_attribute *attr,
360 struct net_device *netdev = to_net_dev(dev);
362 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
364 static DEVICE_ATTR_RO(carrier_down_count);
366 /* read-write attributes */
368 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
370 return dev_set_mtu(dev, (int)new_mtu);
373 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
374 const char *buf, size_t len)
376 return netdev_store(dev, attr, buf, len, change_mtu);
378 NETDEVICE_SHOW_RW(mtu, fmt_dec);
380 static int change_flags(struct net_device *dev, unsigned long new_flags)
382 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
385 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
386 const char *buf, size_t len)
388 return netdev_store(dev, attr, buf, len, change_flags);
390 NETDEVICE_SHOW_RW(flags, fmt_hex);
392 static ssize_t tx_queue_len_store(struct device *dev,
393 struct device_attribute *attr,
394 const char *buf, size_t len)
396 if (!capable(CAP_NET_ADMIN))
399 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
401 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
403 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
405 WRITE_ONCE(dev->gro_flush_timeout, val);
409 static ssize_t gro_flush_timeout_store(struct device *dev,
410 struct device_attribute *attr,
411 const char *buf, size_t len)
413 if (!capable(CAP_NET_ADMIN))
416 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
418 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
420 static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
422 WRITE_ONCE(dev->napi_defer_hard_irqs, val);
426 static ssize_t napi_defer_hard_irqs_store(struct device *dev,
427 struct device_attribute *attr,
428 const char *buf, size_t len)
430 if (!capable(CAP_NET_ADMIN))
433 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
435 NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
437 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
438 const char *buf, size_t len)
440 struct net_device *netdev = to_net_dev(dev);
441 struct net *net = dev_net(netdev);
445 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
448 /* ignore trailing newline */
449 if (len > 0 && buf[len - 1] == '\n')
453 return restart_syscall();
455 if (dev_isalive(netdev)) {
456 ret = dev_set_alias(netdev, buf, count);
460 netdev_state_change(netdev);
468 static ssize_t ifalias_show(struct device *dev,
469 struct device_attribute *attr, char *buf)
471 const struct net_device *netdev = to_net_dev(dev);
475 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
477 ret = sysfs_emit(buf, "%s\n", tmp);
480 static DEVICE_ATTR_RW(ifalias);
482 static int change_group(struct net_device *dev, unsigned long new_group)
484 dev_set_group(dev, (int)new_group);
488 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
489 const char *buf, size_t len)
491 return netdev_store(dev, attr, buf, len, change_group);
493 NETDEVICE_SHOW(group, fmt_dec);
494 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
496 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
498 return dev_change_proto_down(dev, (bool)proto_down);
501 static ssize_t proto_down_store(struct device *dev,
502 struct device_attribute *attr,
503 const char *buf, size_t len)
505 return netdev_store(dev, attr, buf, len, change_proto_down);
507 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
509 static ssize_t phys_port_id_show(struct device *dev,
510 struct device_attribute *attr, char *buf)
512 struct net_device *netdev = to_net_dev(dev);
513 ssize_t ret = -EINVAL;
515 /* The check is also done in dev_get_phys_port_id; this helps returning
516 * early without hitting the trylock/restart below.
518 if (!netdev->netdev_ops->ndo_get_phys_port_id)
522 return restart_syscall();
524 if (dev_isalive(netdev)) {
525 struct netdev_phys_item_id ppid;
527 ret = dev_get_phys_port_id(netdev, &ppid);
529 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
535 static DEVICE_ATTR_RO(phys_port_id);
537 static ssize_t phys_port_name_show(struct device *dev,
538 struct device_attribute *attr, char *buf)
540 struct net_device *netdev = to_net_dev(dev);
541 ssize_t ret = -EINVAL;
543 /* The checks are also done in dev_get_phys_port_name; this helps
544 * returning early without hitting the trylock/restart below.
546 if (!netdev->netdev_ops->ndo_get_phys_port_name &&
547 !netdev->devlink_port)
551 return restart_syscall();
553 if (dev_isalive(netdev)) {
556 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
558 ret = sysfs_emit(buf, "%s\n", name);
564 static DEVICE_ATTR_RO(phys_port_name);
566 static ssize_t phys_switch_id_show(struct device *dev,
567 struct device_attribute *attr, char *buf)
569 struct net_device *netdev = to_net_dev(dev);
570 ssize_t ret = -EINVAL;
572 /* The checks are also done in dev_get_phys_port_name; this helps
573 * returning early without hitting the trylock/restart below. This works
574 * because recurse is false when calling dev_get_port_parent_id.
576 if (!netdev->netdev_ops->ndo_get_port_parent_id &&
577 !netdev->devlink_port)
581 return restart_syscall();
583 if (dev_isalive(netdev)) {
584 struct netdev_phys_item_id ppid = { };
586 ret = dev_get_port_parent_id(netdev, &ppid, false);
588 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
594 static DEVICE_ATTR_RO(phys_switch_id);
596 static ssize_t threaded_show(struct device *dev,
597 struct device_attribute *attr, char *buf)
599 struct net_device *netdev = to_net_dev(dev);
600 ssize_t ret = -EINVAL;
603 return restart_syscall();
605 if (dev_isalive(netdev))
606 ret = sysfs_emit(buf, fmt_dec, netdev->threaded);
612 static int modify_napi_threaded(struct net_device *dev, unsigned long val)
616 if (list_empty(&dev->napi_list))
619 if (val != 0 && val != 1)
622 ret = dev_set_threaded(dev, val);
627 static ssize_t threaded_store(struct device *dev,
628 struct device_attribute *attr,
629 const char *buf, size_t len)
631 return netdev_store(dev, attr, buf, len, modify_napi_threaded);
633 static DEVICE_ATTR_RW(threaded);
635 static struct attribute *net_class_attrs[] __ro_after_init = {
636 &dev_attr_netdev_group.attr,
638 &dev_attr_dev_id.attr,
639 &dev_attr_dev_port.attr,
640 &dev_attr_iflink.attr,
641 &dev_attr_ifindex.attr,
642 &dev_attr_name_assign_type.attr,
643 &dev_attr_addr_assign_type.attr,
644 &dev_attr_addr_len.attr,
645 &dev_attr_link_mode.attr,
646 &dev_attr_address.attr,
647 &dev_attr_broadcast.attr,
648 &dev_attr_speed.attr,
649 &dev_attr_duplex.attr,
650 &dev_attr_dormant.attr,
651 &dev_attr_testing.attr,
652 &dev_attr_operstate.attr,
653 &dev_attr_carrier_changes.attr,
654 &dev_attr_ifalias.attr,
655 &dev_attr_carrier.attr,
657 &dev_attr_flags.attr,
658 &dev_attr_tx_queue_len.attr,
659 &dev_attr_gro_flush_timeout.attr,
660 &dev_attr_napi_defer_hard_irqs.attr,
661 &dev_attr_phys_port_id.attr,
662 &dev_attr_phys_port_name.attr,
663 &dev_attr_phys_switch_id.attr,
664 &dev_attr_proto_down.attr,
665 &dev_attr_carrier_up_count.attr,
666 &dev_attr_carrier_down_count.attr,
667 &dev_attr_threaded.attr,
670 ATTRIBUTE_GROUPS(net_class);
672 /* Show a given an attribute in the statistics group */
673 static ssize_t netstat_show(const struct device *d,
674 struct device_attribute *attr, char *buf,
675 unsigned long offset)
677 struct net_device *dev = to_net_dev(d);
678 ssize_t ret = -EINVAL;
680 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
681 offset % sizeof(u64) != 0);
683 read_lock(&dev_base_lock);
684 if (dev_isalive(dev)) {
685 struct rtnl_link_stats64 temp;
686 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
688 ret = sysfs_emit(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
690 read_unlock(&dev_base_lock);
694 /* generate a read-only statistics attribute */
695 #define NETSTAT_ENTRY(name) \
696 static ssize_t name##_show(struct device *d, \
697 struct device_attribute *attr, char *buf) \
699 return netstat_show(d, attr, buf, \
700 offsetof(struct rtnl_link_stats64, name)); \
702 static DEVICE_ATTR_RO(name)
704 NETSTAT_ENTRY(rx_packets);
705 NETSTAT_ENTRY(tx_packets);
706 NETSTAT_ENTRY(rx_bytes);
707 NETSTAT_ENTRY(tx_bytes);
708 NETSTAT_ENTRY(rx_errors);
709 NETSTAT_ENTRY(tx_errors);
710 NETSTAT_ENTRY(rx_dropped);
711 NETSTAT_ENTRY(tx_dropped);
712 NETSTAT_ENTRY(multicast);
713 NETSTAT_ENTRY(collisions);
714 NETSTAT_ENTRY(rx_length_errors);
715 NETSTAT_ENTRY(rx_over_errors);
716 NETSTAT_ENTRY(rx_crc_errors);
717 NETSTAT_ENTRY(rx_frame_errors);
718 NETSTAT_ENTRY(rx_fifo_errors);
719 NETSTAT_ENTRY(rx_missed_errors);
720 NETSTAT_ENTRY(tx_aborted_errors);
721 NETSTAT_ENTRY(tx_carrier_errors);
722 NETSTAT_ENTRY(tx_fifo_errors);
723 NETSTAT_ENTRY(tx_heartbeat_errors);
724 NETSTAT_ENTRY(tx_window_errors);
725 NETSTAT_ENTRY(rx_compressed);
726 NETSTAT_ENTRY(tx_compressed);
727 NETSTAT_ENTRY(rx_nohandler);
729 static struct attribute *netstat_attrs[] __ro_after_init = {
730 &dev_attr_rx_packets.attr,
731 &dev_attr_tx_packets.attr,
732 &dev_attr_rx_bytes.attr,
733 &dev_attr_tx_bytes.attr,
734 &dev_attr_rx_errors.attr,
735 &dev_attr_tx_errors.attr,
736 &dev_attr_rx_dropped.attr,
737 &dev_attr_tx_dropped.attr,
738 &dev_attr_multicast.attr,
739 &dev_attr_collisions.attr,
740 &dev_attr_rx_length_errors.attr,
741 &dev_attr_rx_over_errors.attr,
742 &dev_attr_rx_crc_errors.attr,
743 &dev_attr_rx_frame_errors.attr,
744 &dev_attr_rx_fifo_errors.attr,
745 &dev_attr_rx_missed_errors.attr,
746 &dev_attr_tx_aborted_errors.attr,
747 &dev_attr_tx_carrier_errors.attr,
748 &dev_attr_tx_fifo_errors.attr,
749 &dev_attr_tx_heartbeat_errors.attr,
750 &dev_attr_tx_window_errors.attr,
751 &dev_attr_rx_compressed.attr,
752 &dev_attr_tx_compressed.attr,
753 &dev_attr_rx_nohandler.attr,
757 static const struct attribute_group netstat_group = {
758 .name = "statistics",
759 .attrs = netstat_attrs,
762 static struct attribute *wireless_attrs[] = {
766 static const struct attribute_group wireless_group = {
768 .attrs = wireless_attrs,
771 static bool wireless_group_needed(struct net_device *ndev)
773 #if IS_ENABLED(CONFIG_CFG80211)
774 if (ndev->ieee80211_ptr)
777 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
778 if (ndev->wireless_handlers)
784 #else /* CONFIG_SYSFS */
785 #define net_class_groups NULL
786 #endif /* CONFIG_SYSFS */
789 #define to_rx_queue_attr(_attr) \
790 container_of(_attr, struct rx_queue_attribute, attr)
792 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
794 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
797 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
798 struct netdev_rx_queue *queue = to_rx_queue(kobj);
800 if (!attribute->show)
803 return attribute->show(queue, buf);
806 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
807 const char *buf, size_t count)
809 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
810 struct netdev_rx_queue *queue = to_rx_queue(kobj);
812 if (!attribute->store)
815 return attribute->store(queue, buf, count);
818 static const struct sysfs_ops rx_queue_sysfs_ops = {
819 .show = rx_queue_attr_show,
820 .store = rx_queue_attr_store,
824 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
830 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
834 map = rcu_dereference(queue->rps_map);
836 for (i = 0; i < map->len; i++)
837 cpumask_set_cpu(map->cpus[i], mask);
839 len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask));
841 free_cpumask_var(mask);
843 return len < PAGE_SIZE ? len : -EINVAL;
846 static int netdev_rx_queue_set_rps_mask(struct netdev_rx_queue *queue,
849 static DEFINE_MUTEX(rps_map_mutex);
850 struct rps_map *old_map, *map;
853 map = kzalloc(max_t(unsigned int,
854 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
860 for_each_cpu_and(cpu, mask, cpu_online_mask)
861 map->cpus[i++] = cpu;
870 mutex_lock(&rps_map_mutex);
871 old_map = rcu_dereference_protected(queue->rps_map,
872 mutex_is_locked(&rps_map_mutex));
873 rcu_assign_pointer(queue->rps_map, map);
876 static_branch_inc(&rps_needed);
878 static_branch_dec(&rps_needed);
880 mutex_unlock(&rps_map_mutex);
883 kfree_rcu(old_map, rcu);
887 int rps_cpumask_housekeeping(struct cpumask *mask)
889 if (!cpumask_empty(mask)) {
890 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
891 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
892 if (cpumask_empty(mask))
898 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
899 const char *buf, size_t len)
904 if (!capable(CAP_NET_ADMIN))
907 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
910 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
914 err = rps_cpumask_housekeeping(mask);
918 err = netdev_rx_queue_set_rps_mask(queue, mask);
921 free_cpumask_var(mask);
925 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
928 struct rps_dev_flow_table *flow_table;
929 unsigned long val = 0;
932 flow_table = rcu_dereference(queue->rps_flow_table);
934 val = (unsigned long)flow_table->mask + 1;
937 return sysfs_emit(buf, "%lu\n", val);
940 static void rps_dev_flow_table_release(struct rcu_head *rcu)
942 struct rps_dev_flow_table *table = container_of(rcu,
943 struct rps_dev_flow_table, rcu);
947 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
948 const char *buf, size_t len)
950 unsigned long mask, count;
951 struct rps_dev_flow_table *table, *old_table;
952 static DEFINE_SPINLOCK(rps_dev_flow_lock);
955 if (!capable(CAP_NET_ADMIN))
958 rc = kstrtoul(buf, 0, &count);
964 /* mask = roundup_pow_of_two(count) - 1;
965 * without overflows...
967 while ((mask | (mask >> 1)) != mask)
969 /* On 64 bit arches, must check mask fits in table->mask (u32),
970 * and on 32bit arches, must check
971 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
973 #if BITS_PER_LONG > 32
974 if (mask > (unsigned long)(u32)mask)
977 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
978 / sizeof(struct rps_dev_flow)) {
979 /* Enforce a limit to prevent overflow */
983 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
988 for (count = 0; count <= mask; count++)
989 table->flows[count].cpu = RPS_NO_CPU;
994 spin_lock(&rps_dev_flow_lock);
995 old_table = rcu_dereference_protected(queue->rps_flow_table,
996 lockdep_is_held(&rps_dev_flow_lock));
997 rcu_assign_pointer(queue->rps_flow_table, table);
998 spin_unlock(&rps_dev_flow_lock);
1001 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
1006 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
1007 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
1009 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
1010 = __ATTR(rps_flow_cnt, 0644,
1011 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
1012 #endif /* CONFIG_RPS */
1014 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
1016 &rps_cpus_attribute.attr,
1017 &rps_dev_flow_table_cnt_attribute.attr,
1021 ATTRIBUTE_GROUPS(rx_queue_default);
1023 static void rx_queue_release(struct kobject *kobj)
1025 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1027 struct rps_map *map;
1028 struct rps_dev_flow_table *flow_table;
1030 map = rcu_dereference_protected(queue->rps_map, 1);
1032 RCU_INIT_POINTER(queue->rps_map, NULL);
1033 kfree_rcu(map, rcu);
1036 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1038 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1039 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1043 memset(kobj, 0, sizeof(*kobj));
1044 netdev_put(queue->dev, &queue->dev_tracker);
1047 static const void *rx_queue_namespace(const struct kobject *kobj)
1049 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1050 struct device *dev = &queue->dev->dev;
1051 const void *ns = NULL;
1053 if (dev->class && dev->class->ns_type)
1054 ns = dev->class->namespace(dev);
1059 static void rx_queue_get_ownership(const struct kobject *kobj,
1060 kuid_t *uid, kgid_t *gid)
1062 const struct net *net = rx_queue_namespace(kobj);
1064 net_ns_get_ownership(net, uid, gid);
1067 static const struct kobj_type rx_queue_ktype = {
1068 .sysfs_ops = &rx_queue_sysfs_ops,
1069 .release = rx_queue_release,
1070 .default_groups = rx_queue_default_groups,
1071 .namespace = rx_queue_namespace,
1072 .get_ownership = rx_queue_get_ownership,
1075 static int rx_queue_default_mask(struct net_device *dev,
1076 struct netdev_rx_queue *queue)
1078 #if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL)
1079 struct cpumask *rps_default_mask = READ_ONCE(dev_net(dev)->core.rps_default_mask);
1081 if (rps_default_mask && !cpumask_empty(rps_default_mask))
1082 return netdev_rx_queue_set_rps_mask(queue, rps_default_mask);
1087 static int rx_queue_add_kobject(struct net_device *dev, int index)
1089 struct netdev_rx_queue *queue = dev->_rx + index;
1090 struct kobject *kobj = &queue->kobj;
1093 /* Kobject_put later will trigger rx_queue_release call which
1094 * decreases dev refcount: Take that reference here
1096 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1098 kobj->kset = dev->queues_kset;
1099 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1104 if (dev->sysfs_rx_queue_group) {
1105 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1110 error = rx_queue_default_mask(dev, queue);
1114 kobject_uevent(kobj, KOBJ_ADD);
1123 static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1126 struct netdev_rx_queue *queue = dev->_rx + index;
1127 struct kobject *kobj = &queue->kobj;
1130 error = sysfs_change_owner(kobj, kuid, kgid);
1134 if (dev->sysfs_rx_queue_group)
1135 error = sysfs_group_change_owner(
1136 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1140 #endif /* CONFIG_SYSFS */
1143 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1150 if (!dev->sysfs_rx_queue_group)
1153 for (i = old_num; i < new_num; i++) {
1154 error = rx_queue_add_kobject(dev, i);
1161 while (--i >= new_num) {
1162 struct kobject *kobj = &dev->_rx[i].kobj;
1164 if (!refcount_read(&dev_net(dev)->ns.count))
1165 kobj->uevent_suppress = 1;
1166 if (dev->sysfs_rx_queue_group)
1167 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1177 static int net_rx_queue_change_owner(struct net_device *dev, int num,
1178 kuid_t kuid, kgid_t kgid)
1185 if (!dev->sysfs_rx_queue_group)
1188 for (i = 0; i < num; i++) {
1189 error = rx_queue_change_owner(dev, i, kuid, kgid);
1202 * netdev_queue sysfs structures and functions.
1204 struct netdev_queue_attribute {
1205 struct attribute attr;
1206 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1207 ssize_t (*store)(struct netdev_queue *queue,
1208 const char *buf, size_t len);
1210 #define to_netdev_queue_attr(_attr) \
1211 container_of(_attr, struct netdev_queue_attribute, attr)
1213 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1215 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1216 struct attribute *attr, char *buf)
1218 const struct netdev_queue_attribute *attribute
1219 = to_netdev_queue_attr(attr);
1220 struct netdev_queue *queue = to_netdev_queue(kobj);
1222 if (!attribute->show)
1225 return attribute->show(queue, buf);
1228 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1229 struct attribute *attr,
1230 const char *buf, size_t count)
1232 const struct netdev_queue_attribute *attribute
1233 = to_netdev_queue_attr(attr);
1234 struct netdev_queue *queue = to_netdev_queue(kobj);
1236 if (!attribute->store)
1239 return attribute->store(queue, buf, count);
1242 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1243 .show = netdev_queue_attr_show,
1244 .store = netdev_queue_attr_store,
1247 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1249 unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1251 return sysfs_emit(buf, fmt_ulong, trans_timeout);
1254 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1256 struct net_device *dev = queue->dev;
1259 i = queue - dev->_tx;
1260 BUG_ON(i >= dev->num_tx_queues);
1265 static ssize_t traffic_class_show(struct netdev_queue *queue,
1268 struct net_device *dev = queue->dev;
1272 if (!netif_is_multiqueue(dev))
1275 if (!rtnl_trylock())
1276 return restart_syscall();
1278 index = get_netdev_queue_index(queue);
1280 /* If queue belongs to subordinate dev use its TC mapping */
1281 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1283 num_tc = dev->num_tc;
1284 tc = netdev_txq_to_tc(dev, index);
1291 /* We can report the traffic class one of two ways:
1292 * Subordinate device traffic classes are reported with the traffic
1293 * class first, and then the subordinate class so for example TC0 on
1294 * subordinate device 2 will be reported as "0-2". If the queue
1295 * belongs to the root device it will be reported with just the
1296 * traffic class, so just "0" for TC 0 for example.
1298 return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) :
1299 sysfs_emit(buf, "%d\n", tc);
1303 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1306 return sysfs_emit(buf, "%lu\n", queue->tx_maxrate);
1309 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1310 const char *buf, size_t len)
1312 struct net_device *dev = queue->dev;
1313 int err, index = get_netdev_queue_index(queue);
1316 if (!capable(CAP_NET_ADMIN))
1319 /* The check is also done later; this helps returning early without
1320 * hitting the trylock/restart below.
1322 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1325 err = kstrtou32(buf, 10, &rate);
1329 if (!rtnl_trylock())
1330 return restart_syscall();
1333 if (dev->netdev_ops->ndo_set_tx_maxrate)
1334 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1338 queue->tx_maxrate = rate;
1344 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1345 = __ATTR_RW(tx_maxrate);
1348 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1349 = __ATTR_RO(tx_timeout);
1351 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1352 = __ATTR_RO(traffic_class);
1356 * Byte queue limits sysfs structures and functions.
1358 static ssize_t bql_show(char *buf, unsigned int value)
1360 return sysfs_emit(buf, "%u\n", value);
1363 static ssize_t bql_set(const char *buf, const size_t count,
1364 unsigned int *pvalue)
1369 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1370 value = DQL_MAX_LIMIT;
1372 err = kstrtouint(buf, 10, &value);
1375 if (value > DQL_MAX_LIMIT)
1384 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1387 struct dql *dql = &queue->dql;
1389 return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1392 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1393 const char *buf, size_t len)
1395 struct dql *dql = &queue->dql;
1399 err = kstrtouint(buf, 10, &value);
1403 dql->slack_hold_time = msecs_to_jiffies(value);
1408 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1409 = __ATTR(hold_time, 0644,
1410 bql_show_hold_time, bql_set_hold_time);
1412 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1415 struct dql *dql = &queue->dql;
1417 return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed);
1420 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1421 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1423 #define BQL_ATTR(NAME, FIELD) \
1424 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1427 return bql_show(buf, queue->dql.FIELD); \
1430 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1431 const char *buf, size_t len) \
1433 return bql_set(buf, len, &queue->dql.FIELD); \
1436 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1437 = __ATTR(NAME, 0644, \
1438 bql_show_ ## NAME, bql_set_ ## NAME)
1440 BQL_ATTR(limit, limit);
1441 BQL_ATTR(limit_max, max_limit);
1442 BQL_ATTR(limit_min, min_limit);
1444 static struct attribute *dql_attrs[] __ro_after_init = {
1445 &bql_limit_attribute.attr,
1446 &bql_limit_max_attribute.attr,
1447 &bql_limit_min_attribute.attr,
1448 &bql_hold_time_attribute.attr,
1449 &bql_inflight_attribute.attr,
1453 static const struct attribute_group dql_group = {
1454 .name = "byte_queue_limits",
1457 #endif /* CONFIG_BQL */
1460 static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1461 int tc, char *buf, enum xps_map_type type)
1463 struct xps_dev_maps *dev_maps;
1464 unsigned long *mask;
1465 unsigned int nr_ids;
1469 dev_maps = rcu_dereference(dev->xps_maps[type]);
1471 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1472 * when dev_maps hasn't been allocated yet, to be backward compatible.
1474 nr_ids = dev_maps ? dev_maps->nr_ids :
1475 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1477 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1483 if (!dev_maps || tc >= dev_maps->num_tc)
1486 for (j = 0; j < nr_ids; j++) {
1487 int i, tci = j * dev_maps->num_tc + tc;
1488 struct xps_map *map;
1490 map = rcu_dereference(dev_maps->attr_map[tci]);
1494 for (i = map->len; i--;) {
1495 if (map->queues[i] == index) {
1504 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1507 return len < PAGE_SIZE ? len : -EINVAL;
1510 static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1512 struct net_device *dev = queue->dev;
1516 if (!netif_is_multiqueue(dev))
1519 index = get_netdev_queue_index(queue);
1521 if (!rtnl_trylock())
1522 return restart_syscall();
1524 /* If queue belongs to subordinate dev use its map */
1525 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1527 tc = netdev_txq_to_tc(dev, index);
1533 /* Make sure the subordinate device can't be freed */
1534 get_device(&dev->dev);
1537 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1539 put_device(&dev->dev);
1543 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1544 const char *buf, size_t len)
1546 struct net_device *dev = queue->dev;
1551 if (!netif_is_multiqueue(dev))
1554 if (!capable(CAP_NET_ADMIN))
1557 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1560 index = get_netdev_queue_index(queue);
1562 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1564 free_cpumask_var(mask);
1568 if (!rtnl_trylock()) {
1569 free_cpumask_var(mask);
1570 return restart_syscall();
1573 err = netif_set_xps_queue(dev, mask, index);
1576 free_cpumask_var(mask);
1581 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1582 = __ATTR_RW(xps_cpus);
1584 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1586 struct net_device *dev = queue->dev;
1590 index = get_netdev_queue_index(queue);
1592 if (!rtnl_trylock())
1593 return restart_syscall();
1595 tc = netdev_txq_to_tc(dev, index);
1600 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1603 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1606 struct net_device *dev = queue->dev;
1607 struct net *net = dev_net(dev);
1608 unsigned long *mask;
1612 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1615 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1619 index = get_netdev_queue_index(queue);
1621 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1627 if (!rtnl_trylock()) {
1629 return restart_syscall();
1633 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1642 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1643 = __ATTR_RW(xps_rxqs);
1644 #endif /* CONFIG_XPS */
1646 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1647 &queue_trans_timeout.attr,
1648 &queue_traffic_class.attr,
1650 &xps_cpus_attribute.attr,
1651 &xps_rxqs_attribute.attr,
1652 &queue_tx_maxrate.attr,
1656 ATTRIBUTE_GROUPS(netdev_queue_default);
1658 static void netdev_queue_release(struct kobject *kobj)
1660 struct netdev_queue *queue = to_netdev_queue(kobj);
1662 memset(kobj, 0, sizeof(*kobj));
1663 netdev_put(queue->dev, &queue->dev_tracker);
1666 static const void *netdev_queue_namespace(const struct kobject *kobj)
1668 struct netdev_queue *queue = to_netdev_queue(kobj);
1669 struct device *dev = &queue->dev->dev;
1670 const void *ns = NULL;
1672 if (dev->class && dev->class->ns_type)
1673 ns = dev->class->namespace(dev);
1678 static void netdev_queue_get_ownership(const struct kobject *kobj,
1679 kuid_t *uid, kgid_t *gid)
1681 const struct net *net = netdev_queue_namespace(kobj);
1683 net_ns_get_ownership(net, uid, gid);
1686 static const struct kobj_type netdev_queue_ktype = {
1687 .sysfs_ops = &netdev_queue_sysfs_ops,
1688 .release = netdev_queue_release,
1689 .default_groups = netdev_queue_default_groups,
1690 .namespace = netdev_queue_namespace,
1691 .get_ownership = netdev_queue_get_ownership,
1694 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1696 struct netdev_queue *queue = dev->_tx + index;
1697 struct kobject *kobj = &queue->kobj;
1700 /* Kobject_put later will trigger netdev_queue_release call
1701 * which decreases dev refcount: Take that reference here
1703 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1705 kobj->kset = dev->queues_kset;
1706 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1712 error = sysfs_create_group(kobj, &dql_group);
1717 kobject_uevent(kobj, KOBJ_ADD);
1725 static int tx_queue_change_owner(struct net_device *ndev, int index,
1726 kuid_t kuid, kgid_t kgid)
1728 struct netdev_queue *queue = ndev->_tx + index;
1729 struct kobject *kobj = &queue->kobj;
1732 error = sysfs_change_owner(kobj, kuid, kgid);
1737 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1741 #endif /* CONFIG_SYSFS */
1744 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1750 /* Tx queue kobjects are allowed to be updated when a device is being
1751 * unregistered, but solely to remove queues from qdiscs. Any path
1752 * adding queues should be fixed.
1754 WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1755 "New queues can't be registered after device unregistration.");
1757 for (i = old_num; i < new_num; i++) {
1758 error = netdev_queue_add_kobject(dev, i);
1765 while (--i >= new_num) {
1766 struct netdev_queue *queue = dev->_tx + i;
1768 if (!refcount_read(&dev_net(dev)->ns.count))
1769 queue->kobj.uevent_suppress = 1;
1771 sysfs_remove_group(&queue->kobj, &dql_group);
1773 kobject_put(&queue->kobj);
1779 #endif /* CONFIG_SYSFS */
1782 static int net_tx_queue_change_owner(struct net_device *dev, int num,
1783 kuid_t kuid, kgid_t kgid)
1789 for (i = 0; i < num; i++) {
1790 error = tx_queue_change_owner(dev, i, kuid, kgid);
1798 #endif /* CONFIG_SYSFS */
1801 static int register_queue_kobjects(struct net_device *dev)
1803 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1806 dev->queues_kset = kset_create_and_add("queues",
1807 NULL, &dev->dev.kobj);
1808 if (!dev->queues_kset)
1810 real_rx = dev->real_num_rx_queues;
1812 real_tx = dev->real_num_tx_queues;
1814 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1819 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1827 netdev_queue_update_kobjects(dev, txq, 0);
1828 net_rx_queue_update_kobjects(dev, rxq, 0);
1830 kset_unregister(dev->queues_kset);
1835 static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1837 int error = 0, real_rx = 0, real_tx = 0;
1840 if (ndev->queues_kset) {
1841 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1845 real_rx = ndev->real_num_rx_queues;
1847 real_tx = ndev->real_num_tx_queues;
1849 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1853 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1860 static void remove_queue_kobjects(struct net_device *dev)
1862 int real_rx = 0, real_tx = 0;
1865 real_rx = dev->real_num_rx_queues;
1867 real_tx = dev->real_num_tx_queues;
1869 net_rx_queue_update_kobjects(dev, real_rx, 0);
1870 netdev_queue_update_kobjects(dev, real_tx, 0);
1872 dev->real_num_rx_queues = 0;
1873 dev->real_num_tx_queues = 0;
1875 kset_unregister(dev->queues_kset);
1879 static bool net_current_may_mount(void)
1881 struct net *net = current->nsproxy->net_ns;
1883 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1886 static void *net_grab_current_ns(void)
1888 struct net *ns = current->nsproxy->net_ns;
1889 #ifdef CONFIG_NET_NS
1891 refcount_inc(&ns->passive);
1896 static const void *net_initial_ns(void)
1901 static const void *net_netlink_ns(struct sock *sk)
1903 return sock_net(sk);
1906 const struct kobj_ns_type_operations net_ns_type_operations = {
1907 .type = KOBJ_NS_TYPE_NET,
1908 .current_may_mount = net_current_may_mount,
1909 .grab_current_ns = net_grab_current_ns,
1910 .netlink_ns = net_netlink_ns,
1911 .initial_ns = net_initial_ns,
1912 .drop_ns = net_drop_ns,
1914 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1916 static int netdev_uevent(const struct device *d, struct kobj_uevent_env *env)
1918 const struct net_device *dev = to_net_dev(d);
1921 /* pass interface to uevent. */
1922 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1926 /* pass ifindex to uevent.
1927 * ifindex is useful as it won't change (interface name may change)
1928 * and is what RtNetlink uses natively.
1930 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1937 * netdev_release -- destroy and free a dead device.
1938 * Called when last reference to device kobject is gone.
1940 static void netdev_release(struct device *d)
1942 struct net_device *dev = to_net_dev(d);
1944 BUG_ON(dev->reg_state != NETREG_RELEASED);
1946 /* no need to wait for rcu grace period:
1947 * device is dead and about to be freed.
1949 kfree(rcu_access_pointer(dev->ifalias));
1950 netdev_freemem(dev);
1953 static const void *net_namespace(const struct device *d)
1955 const struct net_device *dev = to_net_dev(d);
1957 return dev_net(dev);
1960 static void net_get_ownership(const struct device *d, kuid_t *uid, kgid_t *gid)
1962 const struct net_device *dev = to_net_dev(d);
1963 const struct net *net = dev_net(dev);
1965 net_ns_get_ownership(net, uid, gid);
1968 static struct class net_class __ro_after_init = {
1970 .dev_release = netdev_release,
1971 .dev_groups = net_class_groups,
1972 .dev_uevent = netdev_uevent,
1973 .ns_type = &net_ns_type_operations,
1974 .namespace = net_namespace,
1975 .get_ownership = net_get_ownership,
1979 static int of_dev_node_match(struct device *dev, const void *data)
1981 for (; dev; dev = dev->parent) {
1982 if (dev->of_node == data)
1990 * of_find_net_device_by_node - lookup the net device for the device node
1991 * @np: OF device node
1993 * Looks up the net_device structure corresponding with the device node.
1994 * If successful, returns a pointer to the net_device with the embedded
1995 * struct device refcount incremented by one, or NULL on failure. The
1996 * refcount must be dropped when done with the net_device.
1998 struct net_device *of_find_net_device_by_node(struct device_node *np)
2002 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
2006 return to_net_dev(dev);
2008 EXPORT_SYMBOL(of_find_net_device_by_node);
2011 /* Delete sysfs entries but hold kobject reference until after all
2012 * netdev references are gone.
2014 void netdev_unregister_kobject(struct net_device *ndev)
2016 struct device *dev = &ndev->dev;
2018 if (!refcount_read(&dev_net(ndev)->ns.count))
2019 dev_set_uevent_suppress(dev, 1);
2021 kobject_get(&dev->kobj);
2023 remove_queue_kobjects(ndev);
2025 pm_runtime_set_memalloc_noio(dev, false);
2030 /* Create sysfs entries for network device. */
2031 int netdev_register_kobject(struct net_device *ndev)
2033 struct device *dev = &ndev->dev;
2034 const struct attribute_group **groups = ndev->sysfs_groups;
2037 device_initialize(dev);
2038 dev->class = &net_class;
2039 dev->platform_data = ndev;
2040 dev->groups = groups;
2042 dev_set_name(dev, "%s", ndev->name);
2045 /* Allow for a device specific group */
2049 *groups++ = &netstat_group;
2051 if (wireless_group_needed(ndev))
2052 *groups++ = &wireless_group;
2053 #endif /* CONFIG_SYSFS */
2055 error = device_add(dev);
2059 error = register_queue_kobjects(ndev);
2065 pm_runtime_set_memalloc_noio(dev, true);
2070 /* Change owner for sysfs entries when moving network devices across network
2071 * namespaces owned by different user namespaces.
2073 int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2074 const struct net *net_new)
2076 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2077 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2078 struct device *dev = &ndev->dev;
2081 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2082 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2084 /* The network namespace was changed but the owning user namespace is
2085 * identical so there's no need to change the owner of sysfs entries.
2087 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2090 error = device_change_owner(dev, new_uid, new_gid);
2094 error = queue_change_owner(ndev, new_uid, new_gid);
2101 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2104 return class_create_file_ns(&net_class, class_attr, ns);
2106 EXPORT_SYMBOL(netdev_class_create_file_ns);
2108 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2111 class_remove_file_ns(&net_class, class_attr, ns);
2113 EXPORT_SYMBOL(netdev_class_remove_file_ns);
2115 int __init netdev_kobject_init(void)
2117 kobj_ns_type_register(&net_ns_type_operations);
2118 return class_register(&net_class);