2 * net-sysfs.c - network device class and attributes
4 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
6 * This program is free software; you can redistribute it and/or
7 * modify it under the terms of the GNU General Public License
8 * as published by the Free Software Foundation; either version
9 * 2 of the License, or (at your option) any later version.
12 #include <linux/capability.h>
13 #include <linux/kernel.h>
14 #include <linux/netdevice.h>
15 #include <linux/if_arp.h>
16 #include <linux/slab.h>
17 #include <linux/sched/signal.h>
18 #include <linux/nsproxy.h>
20 #include <net/net_namespace.h>
21 #include <linux/rtnetlink.h>
22 #include <linux/vmalloc.h>
23 #include <linux/export.h>
24 #include <linux/jiffies.h>
25 #include <linux/pm_runtime.h>
27 #include <linux/of_net.h>
28 #include <linux/cpu.h>
30 #include "net-sysfs.h"
33 static const char fmt_hex[] = "%#x\n";
34 static const char fmt_dec[] = "%d\n";
35 static const char fmt_ulong[] = "%lu\n";
36 static const char fmt_u64[] = "%llu\n";
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 sprintf(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 sprintf(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 sprintf(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 return netdev_store(dev, attr, buf, len, change_carrier);
184 static ssize_t carrier_show(struct device *dev,
185 struct device_attribute *attr, char *buf)
187 struct net_device *netdev = to_net_dev(dev);
189 if (netif_running(netdev))
190 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
194 static DEVICE_ATTR_RW(carrier);
196 static ssize_t speed_show(struct device *dev,
197 struct device_attribute *attr, char *buf)
199 struct net_device *netdev = to_net_dev(dev);
203 return restart_syscall();
205 if (netif_running(netdev)) {
206 struct ethtool_link_ksettings cmd;
208 if (!__ethtool_get_link_ksettings(netdev, &cmd))
209 ret = sprintf(buf, fmt_dec, cmd.base.speed);
214 static DEVICE_ATTR_RO(speed);
216 static ssize_t duplex_show(struct device *dev,
217 struct device_attribute *attr, char *buf)
219 struct net_device *netdev = to_net_dev(dev);
223 return restart_syscall();
225 if (netif_running(netdev)) {
226 struct ethtool_link_ksettings cmd;
228 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
231 switch (cmd.base.duplex) {
242 ret = sprintf(buf, "%s\n", duplex);
248 static DEVICE_ATTR_RO(duplex);
250 static ssize_t dormant_show(struct device *dev,
251 struct device_attribute *attr, char *buf)
253 struct net_device *netdev = to_net_dev(dev);
255 if (netif_running(netdev))
256 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
260 static DEVICE_ATTR_RO(dormant);
262 static const char *const operstates[] = {
264 "notpresent", /* currently unused */
267 "testing", /* currently unused */
272 static ssize_t operstate_show(struct device *dev,
273 struct device_attribute *attr, char *buf)
275 const struct net_device *netdev = to_net_dev(dev);
276 unsigned char operstate;
278 read_lock(&dev_base_lock);
279 operstate = netdev->operstate;
280 if (!netif_running(netdev))
281 operstate = IF_OPER_DOWN;
282 read_unlock(&dev_base_lock);
284 if (operstate >= ARRAY_SIZE(operstates))
285 return -EINVAL; /* should not happen */
287 return sprintf(buf, "%s\n", operstates[operstate]);
289 static DEVICE_ATTR_RO(operstate);
291 static ssize_t carrier_changes_show(struct device *dev,
292 struct device_attribute *attr,
295 struct net_device *netdev = to_net_dev(dev);
297 return sprintf(buf, fmt_dec,
298 atomic_read(&netdev->carrier_up_count) +
299 atomic_read(&netdev->carrier_down_count));
301 static DEVICE_ATTR_RO(carrier_changes);
303 static ssize_t carrier_up_count_show(struct device *dev,
304 struct device_attribute *attr,
307 struct net_device *netdev = to_net_dev(dev);
309 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
311 static DEVICE_ATTR_RO(carrier_up_count);
313 static ssize_t carrier_down_count_show(struct device *dev,
314 struct device_attribute *attr,
317 struct net_device *netdev = to_net_dev(dev);
319 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
321 static DEVICE_ATTR_RO(carrier_down_count);
323 /* read-write attributes */
325 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
327 return dev_set_mtu(dev, (int)new_mtu);
330 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
331 const char *buf, size_t len)
333 return netdev_store(dev, attr, buf, len, change_mtu);
335 NETDEVICE_SHOW_RW(mtu, fmt_dec);
337 static int change_flags(struct net_device *dev, unsigned long new_flags)
339 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
342 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
343 const char *buf, size_t len)
345 return netdev_store(dev, attr, buf, len, change_flags);
347 NETDEVICE_SHOW_RW(flags, fmt_hex);
349 static ssize_t tx_queue_len_store(struct device *dev,
350 struct device_attribute *attr,
351 const char *buf, size_t len)
353 if (!capable(CAP_NET_ADMIN))
356 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
358 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
360 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
362 dev->gro_flush_timeout = val;
366 static ssize_t gro_flush_timeout_store(struct device *dev,
367 struct device_attribute *attr,
368 const char *buf, size_t len)
370 if (!capable(CAP_NET_ADMIN))
373 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
375 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
377 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
378 const char *buf, size_t len)
380 struct net_device *netdev = to_net_dev(dev);
381 struct net *net = dev_net(netdev);
385 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
388 /* ignore trailing newline */
389 if (len > 0 && buf[len - 1] == '\n')
393 return restart_syscall();
395 if (dev_isalive(netdev)) {
396 ret = dev_set_alias(netdev, buf, count);
400 netdev_state_change(netdev);
408 static ssize_t ifalias_show(struct device *dev,
409 struct device_attribute *attr, char *buf)
411 const struct net_device *netdev = to_net_dev(dev);
415 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
417 ret = sprintf(buf, "%s\n", tmp);
420 static DEVICE_ATTR_RW(ifalias);
422 static int change_group(struct net_device *dev, unsigned long new_group)
424 dev_set_group(dev, (int)new_group);
428 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
429 const char *buf, size_t len)
431 return netdev_store(dev, attr, buf, len, change_group);
433 NETDEVICE_SHOW(group, fmt_dec);
434 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
436 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
438 return dev_change_proto_down(dev, (bool)proto_down);
441 static ssize_t proto_down_store(struct device *dev,
442 struct device_attribute *attr,
443 const char *buf, size_t len)
445 return netdev_store(dev, attr, buf, len, change_proto_down);
447 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
449 static ssize_t phys_port_id_show(struct device *dev,
450 struct device_attribute *attr, char *buf)
452 struct net_device *netdev = to_net_dev(dev);
453 ssize_t ret = -EINVAL;
456 return restart_syscall();
458 if (dev_isalive(netdev)) {
459 struct netdev_phys_item_id ppid;
461 ret = dev_get_phys_port_id(netdev, &ppid);
463 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
469 static DEVICE_ATTR_RO(phys_port_id);
471 static ssize_t phys_port_name_show(struct device *dev,
472 struct device_attribute *attr, char *buf)
474 struct net_device *netdev = to_net_dev(dev);
475 ssize_t ret = -EINVAL;
478 return restart_syscall();
480 if (dev_isalive(netdev)) {
483 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
485 ret = sprintf(buf, "%s\n", name);
491 static DEVICE_ATTR_RO(phys_port_name);
493 static ssize_t phys_switch_id_show(struct device *dev,
494 struct device_attribute *attr, char *buf)
496 struct net_device *netdev = to_net_dev(dev);
497 ssize_t ret = -EINVAL;
500 return restart_syscall();
502 if (dev_isalive(netdev)) {
503 struct netdev_phys_item_id ppid = { };
505 ret = dev_get_port_parent_id(netdev, &ppid, false);
507 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
513 static DEVICE_ATTR_RO(phys_switch_id);
515 static struct attribute *net_class_attrs[] __ro_after_init = {
516 &dev_attr_netdev_group.attr,
518 &dev_attr_dev_id.attr,
519 &dev_attr_dev_port.attr,
520 &dev_attr_iflink.attr,
521 &dev_attr_ifindex.attr,
522 &dev_attr_name_assign_type.attr,
523 &dev_attr_addr_assign_type.attr,
524 &dev_attr_addr_len.attr,
525 &dev_attr_link_mode.attr,
526 &dev_attr_address.attr,
527 &dev_attr_broadcast.attr,
528 &dev_attr_speed.attr,
529 &dev_attr_duplex.attr,
530 &dev_attr_dormant.attr,
531 &dev_attr_operstate.attr,
532 &dev_attr_carrier_changes.attr,
533 &dev_attr_ifalias.attr,
534 &dev_attr_carrier.attr,
536 &dev_attr_flags.attr,
537 &dev_attr_tx_queue_len.attr,
538 &dev_attr_gro_flush_timeout.attr,
539 &dev_attr_phys_port_id.attr,
540 &dev_attr_phys_port_name.attr,
541 &dev_attr_phys_switch_id.attr,
542 &dev_attr_proto_down.attr,
543 &dev_attr_carrier_up_count.attr,
544 &dev_attr_carrier_down_count.attr,
547 ATTRIBUTE_GROUPS(net_class);
549 /* Show a given an attribute in the statistics group */
550 static ssize_t netstat_show(const struct device *d,
551 struct device_attribute *attr, char *buf,
552 unsigned long offset)
554 struct net_device *dev = to_net_dev(d);
555 ssize_t ret = -EINVAL;
557 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
558 offset % sizeof(u64) != 0);
560 read_lock(&dev_base_lock);
561 if (dev_isalive(dev)) {
562 struct rtnl_link_stats64 temp;
563 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
565 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
567 read_unlock(&dev_base_lock);
571 /* generate a read-only statistics attribute */
572 #define NETSTAT_ENTRY(name) \
573 static ssize_t name##_show(struct device *d, \
574 struct device_attribute *attr, char *buf) \
576 return netstat_show(d, attr, buf, \
577 offsetof(struct rtnl_link_stats64, name)); \
579 static DEVICE_ATTR_RO(name)
581 NETSTAT_ENTRY(rx_packets);
582 NETSTAT_ENTRY(tx_packets);
583 NETSTAT_ENTRY(rx_bytes);
584 NETSTAT_ENTRY(tx_bytes);
585 NETSTAT_ENTRY(rx_errors);
586 NETSTAT_ENTRY(tx_errors);
587 NETSTAT_ENTRY(rx_dropped);
588 NETSTAT_ENTRY(tx_dropped);
589 NETSTAT_ENTRY(multicast);
590 NETSTAT_ENTRY(collisions);
591 NETSTAT_ENTRY(rx_length_errors);
592 NETSTAT_ENTRY(rx_over_errors);
593 NETSTAT_ENTRY(rx_crc_errors);
594 NETSTAT_ENTRY(rx_frame_errors);
595 NETSTAT_ENTRY(rx_fifo_errors);
596 NETSTAT_ENTRY(rx_missed_errors);
597 NETSTAT_ENTRY(tx_aborted_errors);
598 NETSTAT_ENTRY(tx_carrier_errors);
599 NETSTAT_ENTRY(tx_fifo_errors);
600 NETSTAT_ENTRY(tx_heartbeat_errors);
601 NETSTAT_ENTRY(tx_window_errors);
602 NETSTAT_ENTRY(rx_compressed);
603 NETSTAT_ENTRY(tx_compressed);
604 NETSTAT_ENTRY(rx_nohandler);
606 static struct attribute *netstat_attrs[] __ro_after_init = {
607 &dev_attr_rx_packets.attr,
608 &dev_attr_tx_packets.attr,
609 &dev_attr_rx_bytes.attr,
610 &dev_attr_tx_bytes.attr,
611 &dev_attr_rx_errors.attr,
612 &dev_attr_tx_errors.attr,
613 &dev_attr_rx_dropped.attr,
614 &dev_attr_tx_dropped.attr,
615 &dev_attr_multicast.attr,
616 &dev_attr_collisions.attr,
617 &dev_attr_rx_length_errors.attr,
618 &dev_attr_rx_over_errors.attr,
619 &dev_attr_rx_crc_errors.attr,
620 &dev_attr_rx_frame_errors.attr,
621 &dev_attr_rx_fifo_errors.attr,
622 &dev_attr_rx_missed_errors.attr,
623 &dev_attr_tx_aborted_errors.attr,
624 &dev_attr_tx_carrier_errors.attr,
625 &dev_attr_tx_fifo_errors.attr,
626 &dev_attr_tx_heartbeat_errors.attr,
627 &dev_attr_tx_window_errors.attr,
628 &dev_attr_rx_compressed.attr,
629 &dev_attr_tx_compressed.attr,
630 &dev_attr_rx_nohandler.attr,
634 static const struct attribute_group netstat_group = {
635 .name = "statistics",
636 .attrs = netstat_attrs,
639 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
640 static struct attribute *wireless_attrs[] = {
644 static const struct attribute_group wireless_group = {
646 .attrs = wireless_attrs,
650 #else /* CONFIG_SYSFS */
651 #define net_class_groups NULL
652 #endif /* CONFIG_SYSFS */
655 #define to_rx_queue_attr(_attr) \
656 container_of(_attr, struct rx_queue_attribute, attr)
658 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
660 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
663 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
664 struct netdev_rx_queue *queue = to_rx_queue(kobj);
666 if (!attribute->show)
669 return attribute->show(queue, buf);
672 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
673 const char *buf, size_t count)
675 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
676 struct netdev_rx_queue *queue = to_rx_queue(kobj);
678 if (!attribute->store)
681 return attribute->store(queue, buf, count);
684 static const struct sysfs_ops rx_queue_sysfs_ops = {
685 .show = rx_queue_attr_show,
686 .store = rx_queue_attr_store,
690 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
696 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
700 map = rcu_dereference(queue->rps_map);
702 for (i = 0; i < map->len; i++)
703 cpumask_set_cpu(map->cpus[i], mask);
705 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
707 free_cpumask_var(mask);
709 return len < PAGE_SIZE ? len : -EINVAL;
712 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
713 const char *buf, size_t len)
715 struct rps_map *old_map, *map;
718 static DEFINE_MUTEX(rps_map_mutex);
720 if (!capable(CAP_NET_ADMIN))
723 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
726 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
728 free_cpumask_var(mask);
732 map = kzalloc(max_t(unsigned int,
733 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
736 free_cpumask_var(mask);
741 for_each_cpu_and(cpu, mask, cpu_online_mask)
742 map->cpus[i++] = cpu;
751 mutex_lock(&rps_map_mutex);
752 old_map = rcu_dereference_protected(queue->rps_map,
753 mutex_is_locked(&rps_map_mutex));
754 rcu_assign_pointer(queue->rps_map, map);
757 static_branch_inc(&rps_needed);
759 static_branch_dec(&rps_needed);
761 mutex_unlock(&rps_map_mutex);
764 kfree_rcu(old_map, rcu);
766 free_cpumask_var(mask);
770 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
773 struct rps_dev_flow_table *flow_table;
774 unsigned long val = 0;
777 flow_table = rcu_dereference(queue->rps_flow_table);
779 val = (unsigned long)flow_table->mask + 1;
782 return sprintf(buf, "%lu\n", val);
785 static void rps_dev_flow_table_release(struct rcu_head *rcu)
787 struct rps_dev_flow_table *table = container_of(rcu,
788 struct rps_dev_flow_table, rcu);
792 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
793 const char *buf, size_t len)
795 unsigned long mask, count;
796 struct rps_dev_flow_table *table, *old_table;
797 static DEFINE_SPINLOCK(rps_dev_flow_lock);
800 if (!capable(CAP_NET_ADMIN))
803 rc = kstrtoul(buf, 0, &count);
809 /* mask = roundup_pow_of_two(count) - 1;
810 * without overflows...
812 while ((mask | (mask >> 1)) != mask)
814 /* On 64 bit arches, must check mask fits in table->mask (u32),
815 * and on 32bit arches, must check
816 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
818 #if BITS_PER_LONG > 32
819 if (mask > (unsigned long)(u32)mask)
822 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
823 / sizeof(struct rps_dev_flow)) {
824 /* Enforce a limit to prevent overflow */
828 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
833 for (count = 0; count <= mask; count++)
834 table->flows[count].cpu = RPS_NO_CPU;
839 spin_lock(&rps_dev_flow_lock);
840 old_table = rcu_dereference_protected(queue->rps_flow_table,
841 lockdep_is_held(&rps_dev_flow_lock));
842 rcu_assign_pointer(queue->rps_flow_table, table);
843 spin_unlock(&rps_dev_flow_lock);
846 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
851 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
852 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
854 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
855 = __ATTR(rps_flow_cnt, 0644,
856 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
857 #endif /* CONFIG_RPS */
859 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
861 &rps_cpus_attribute.attr,
862 &rps_dev_flow_table_cnt_attribute.attr,
866 ATTRIBUTE_GROUPS(rx_queue_default);
868 static void rx_queue_release(struct kobject *kobj)
870 struct netdev_rx_queue *queue = to_rx_queue(kobj);
873 struct rps_dev_flow_table *flow_table;
875 map = rcu_dereference_protected(queue->rps_map, 1);
877 RCU_INIT_POINTER(queue->rps_map, NULL);
881 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
883 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
884 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
888 memset(kobj, 0, sizeof(*kobj));
892 static const void *rx_queue_namespace(struct kobject *kobj)
894 struct netdev_rx_queue *queue = to_rx_queue(kobj);
895 struct device *dev = &queue->dev->dev;
896 const void *ns = NULL;
898 if (dev->class && dev->class->ns_type)
899 ns = dev->class->namespace(dev);
904 static void rx_queue_get_ownership(struct kobject *kobj,
905 kuid_t *uid, kgid_t *gid)
907 const struct net *net = rx_queue_namespace(kobj);
909 net_ns_get_ownership(net, uid, gid);
912 static struct kobj_type rx_queue_ktype __ro_after_init = {
913 .sysfs_ops = &rx_queue_sysfs_ops,
914 .release = rx_queue_release,
915 .default_groups = rx_queue_default_groups,
916 .namespace = rx_queue_namespace,
917 .get_ownership = rx_queue_get_ownership,
920 static int rx_queue_add_kobject(struct net_device *dev, int index)
922 struct netdev_rx_queue *queue = dev->_rx + index;
923 struct kobject *kobj = &queue->kobj;
926 kobj->kset = dev->queues_kset;
927 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
932 dev_hold(queue->dev);
934 if (dev->sysfs_rx_queue_group) {
935 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
942 kobject_uevent(kobj, KOBJ_ADD);
946 #endif /* CONFIG_SYSFS */
949 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
956 if (!dev->sysfs_rx_queue_group)
959 for (i = old_num; i < new_num; i++) {
960 error = rx_queue_add_kobject(dev, i);
967 while (--i >= new_num) {
968 struct kobject *kobj = &dev->_rx[i].kobj;
970 if (!refcount_read(&dev_net(dev)->count))
971 kobj->uevent_suppress = 1;
972 if (dev->sysfs_rx_queue_group)
973 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
985 * netdev_queue sysfs structures and functions.
987 struct netdev_queue_attribute {
988 struct attribute attr;
989 ssize_t (*show)(struct netdev_queue *queue, char *buf);
990 ssize_t (*store)(struct netdev_queue *queue,
991 const char *buf, size_t len);
993 #define to_netdev_queue_attr(_attr) \
994 container_of(_attr, struct netdev_queue_attribute, attr)
996 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
998 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
999 struct attribute *attr, char *buf)
1001 const struct netdev_queue_attribute *attribute
1002 = to_netdev_queue_attr(attr);
1003 struct netdev_queue *queue = to_netdev_queue(kobj);
1005 if (!attribute->show)
1008 return attribute->show(queue, buf);
1011 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1012 struct attribute *attr,
1013 const char *buf, size_t count)
1015 const struct netdev_queue_attribute *attribute
1016 = to_netdev_queue_attr(attr);
1017 struct netdev_queue *queue = to_netdev_queue(kobj);
1019 if (!attribute->store)
1022 return attribute->store(queue, buf, count);
1025 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1026 .show = netdev_queue_attr_show,
1027 .store = netdev_queue_attr_store,
1030 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1032 unsigned long trans_timeout;
1034 spin_lock_irq(&queue->_xmit_lock);
1035 trans_timeout = queue->trans_timeout;
1036 spin_unlock_irq(&queue->_xmit_lock);
1038 return sprintf(buf, "%lu", trans_timeout);
1041 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1043 struct net_device *dev = queue->dev;
1046 i = queue - dev->_tx;
1047 BUG_ON(i >= dev->num_tx_queues);
1052 static ssize_t traffic_class_show(struct netdev_queue *queue,
1055 struct net_device *dev = queue->dev;
1059 if (!netif_is_multiqueue(dev))
1062 index = get_netdev_queue_index(queue);
1064 /* If queue belongs to subordinate dev use its TC mapping */
1065 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1067 tc = netdev_txq_to_tc(dev, index);
1071 /* We can report the traffic class one of two ways:
1072 * Subordinate device traffic classes are reported with the traffic
1073 * class first, and then the subordinate class so for example TC0 on
1074 * subordinate device 2 will be reported as "0-2". If the queue
1075 * belongs to the root device it will be reported with just the
1076 * traffic class, so just "0" for TC 0 for example.
1078 return dev->num_tc < 0 ? sprintf(buf, "%u%d\n", tc, dev->num_tc) :
1079 sprintf(buf, "%u\n", tc);
1083 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1086 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1089 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1090 const char *buf, size_t len)
1092 struct net_device *dev = queue->dev;
1093 int err, index = get_netdev_queue_index(queue);
1096 if (!capable(CAP_NET_ADMIN))
1099 err = kstrtou32(buf, 10, &rate);
1103 if (!rtnl_trylock())
1104 return restart_syscall();
1107 if (dev->netdev_ops->ndo_set_tx_maxrate)
1108 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1112 queue->tx_maxrate = rate;
1118 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1119 = __ATTR_RW(tx_maxrate);
1122 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1123 = __ATTR_RO(tx_timeout);
1125 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1126 = __ATTR_RO(traffic_class);
1130 * Byte queue limits sysfs structures and functions.
1132 static ssize_t bql_show(char *buf, unsigned int value)
1134 return sprintf(buf, "%u\n", value);
1137 static ssize_t bql_set(const char *buf, const size_t count,
1138 unsigned int *pvalue)
1143 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1144 value = DQL_MAX_LIMIT;
1146 err = kstrtouint(buf, 10, &value);
1149 if (value > DQL_MAX_LIMIT)
1158 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1161 struct dql *dql = &queue->dql;
1163 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1166 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1167 const char *buf, size_t len)
1169 struct dql *dql = &queue->dql;
1173 err = kstrtouint(buf, 10, &value);
1177 dql->slack_hold_time = msecs_to_jiffies(value);
1182 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1183 = __ATTR(hold_time, 0644,
1184 bql_show_hold_time, bql_set_hold_time);
1186 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1189 struct dql *dql = &queue->dql;
1191 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1194 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1195 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1197 #define BQL_ATTR(NAME, FIELD) \
1198 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1201 return bql_show(buf, queue->dql.FIELD); \
1204 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1205 const char *buf, size_t len) \
1207 return bql_set(buf, len, &queue->dql.FIELD); \
1210 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1211 = __ATTR(NAME, 0644, \
1212 bql_show_ ## NAME, bql_set_ ## NAME)
1214 BQL_ATTR(limit, limit);
1215 BQL_ATTR(limit_max, max_limit);
1216 BQL_ATTR(limit_min, min_limit);
1218 static struct attribute *dql_attrs[] __ro_after_init = {
1219 &bql_limit_attribute.attr,
1220 &bql_limit_max_attribute.attr,
1221 &bql_limit_min_attribute.attr,
1222 &bql_hold_time_attribute.attr,
1223 &bql_inflight_attribute.attr,
1227 static const struct attribute_group dql_group = {
1228 .name = "byte_queue_limits",
1231 #endif /* CONFIG_BQL */
1234 static ssize_t xps_cpus_show(struct netdev_queue *queue,
1237 struct net_device *dev = queue->dev;
1238 int cpu, len, num_tc = 1, tc = 0;
1239 struct xps_dev_maps *dev_maps;
1241 unsigned long index;
1243 if (!netif_is_multiqueue(dev))
1246 index = get_netdev_queue_index(queue);
1249 /* Do not allow XPS on subordinate device directly */
1250 num_tc = dev->num_tc;
1254 /* If queue belongs to subordinate dev use its map */
1255 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1257 tc = netdev_txq_to_tc(dev, index);
1262 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1266 dev_maps = rcu_dereference(dev->xps_cpus_map);
1268 for_each_possible_cpu(cpu) {
1269 int i, tci = cpu * num_tc + tc;
1270 struct xps_map *map;
1272 map = rcu_dereference(dev_maps->attr_map[tci]);
1276 for (i = map->len; i--;) {
1277 if (map->queues[i] == index) {
1278 cpumask_set_cpu(cpu, mask);
1286 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1287 free_cpumask_var(mask);
1288 return len < PAGE_SIZE ? len : -EINVAL;
1291 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1292 const char *buf, size_t len)
1294 struct net_device *dev = queue->dev;
1295 unsigned long index;
1299 if (!netif_is_multiqueue(dev))
1302 if (!capable(CAP_NET_ADMIN))
1305 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1308 index = get_netdev_queue_index(queue);
1310 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1312 free_cpumask_var(mask);
1316 err = netif_set_xps_queue(dev, mask, index);
1318 free_cpumask_var(mask);
1323 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1324 = __ATTR_RW(xps_cpus);
1326 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1328 struct net_device *dev = queue->dev;
1329 struct xps_dev_maps *dev_maps;
1330 unsigned long *mask, index;
1331 int j, len, num_tc = 1, tc = 0;
1333 index = get_netdev_queue_index(queue);
1336 num_tc = dev->num_tc;
1337 tc = netdev_txq_to_tc(dev, index);
1341 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1346 dev_maps = rcu_dereference(dev->xps_rxqs_map);
1350 for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues),
1351 j < dev->num_rx_queues;) {
1352 int i, tci = j * num_tc + tc;
1353 struct xps_map *map;
1355 map = rcu_dereference(dev_maps->attr_map[tci]);
1359 for (i = map->len; i--;) {
1360 if (map->queues[i] == index) {
1369 len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues);
1372 return len < PAGE_SIZE ? len : -EINVAL;
1375 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1378 struct net_device *dev = queue->dev;
1379 struct net *net = dev_net(dev);
1380 unsigned long *mask, index;
1383 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1386 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1390 index = get_netdev_queue_index(queue);
1392 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1399 err = __netif_set_xps_queue(dev, mask, index, true);
1406 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1407 = __ATTR_RW(xps_rxqs);
1408 #endif /* CONFIG_XPS */
1410 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1411 &queue_trans_timeout.attr,
1412 &queue_traffic_class.attr,
1414 &xps_cpus_attribute.attr,
1415 &xps_rxqs_attribute.attr,
1416 &queue_tx_maxrate.attr,
1420 ATTRIBUTE_GROUPS(netdev_queue_default);
1422 static void netdev_queue_release(struct kobject *kobj)
1424 struct netdev_queue *queue = to_netdev_queue(kobj);
1426 memset(kobj, 0, sizeof(*kobj));
1427 dev_put(queue->dev);
1430 static const void *netdev_queue_namespace(struct kobject *kobj)
1432 struct netdev_queue *queue = to_netdev_queue(kobj);
1433 struct device *dev = &queue->dev->dev;
1434 const void *ns = NULL;
1436 if (dev->class && dev->class->ns_type)
1437 ns = dev->class->namespace(dev);
1442 static void netdev_queue_get_ownership(struct kobject *kobj,
1443 kuid_t *uid, kgid_t *gid)
1445 const struct net *net = netdev_queue_namespace(kobj);
1447 net_ns_get_ownership(net, uid, gid);
1450 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1451 .sysfs_ops = &netdev_queue_sysfs_ops,
1452 .release = netdev_queue_release,
1453 .default_groups = netdev_queue_default_groups,
1454 .namespace = netdev_queue_namespace,
1455 .get_ownership = netdev_queue_get_ownership,
1458 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1460 struct netdev_queue *queue = dev->_tx + index;
1461 struct kobject *kobj = &queue->kobj;
1464 kobj->kset = dev->queues_kset;
1465 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1470 dev_hold(queue->dev);
1473 error = sysfs_create_group(kobj, &dql_group);
1480 kobject_uevent(kobj, KOBJ_ADD);
1484 #endif /* CONFIG_SYSFS */
1487 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1493 for (i = old_num; i < new_num; i++) {
1494 error = netdev_queue_add_kobject(dev, i);
1501 while (--i >= new_num) {
1502 struct netdev_queue *queue = dev->_tx + i;
1504 if (!refcount_read(&dev_net(dev)->count))
1505 queue->kobj.uevent_suppress = 1;
1507 sysfs_remove_group(&queue->kobj, &dql_group);
1509 kobject_put(&queue->kobj);
1515 #endif /* CONFIG_SYSFS */
1518 static int register_queue_kobjects(struct net_device *dev)
1520 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1523 dev->queues_kset = kset_create_and_add("queues",
1524 NULL, &dev->dev.kobj);
1525 if (!dev->queues_kset)
1527 real_rx = dev->real_num_rx_queues;
1529 real_tx = dev->real_num_tx_queues;
1531 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1536 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1544 netdev_queue_update_kobjects(dev, txq, 0);
1545 net_rx_queue_update_kobjects(dev, rxq, 0);
1547 kset_unregister(dev->queues_kset);
1552 static void remove_queue_kobjects(struct net_device *dev)
1554 int real_rx = 0, real_tx = 0;
1557 real_rx = dev->real_num_rx_queues;
1559 real_tx = dev->real_num_tx_queues;
1561 net_rx_queue_update_kobjects(dev, real_rx, 0);
1562 netdev_queue_update_kobjects(dev, real_tx, 0);
1564 kset_unregister(dev->queues_kset);
1568 static bool net_current_may_mount(void)
1570 struct net *net = current->nsproxy->net_ns;
1572 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1575 static void *net_grab_current_ns(void)
1577 struct net *ns = current->nsproxy->net_ns;
1578 #ifdef CONFIG_NET_NS
1580 refcount_inc(&ns->passive);
1585 static const void *net_initial_ns(void)
1590 static const void *net_netlink_ns(struct sock *sk)
1592 return sock_net(sk);
1595 const struct kobj_ns_type_operations net_ns_type_operations = {
1596 .type = KOBJ_NS_TYPE_NET,
1597 .current_may_mount = net_current_may_mount,
1598 .grab_current_ns = net_grab_current_ns,
1599 .netlink_ns = net_netlink_ns,
1600 .initial_ns = net_initial_ns,
1601 .drop_ns = net_drop_ns,
1603 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1605 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1607 struct net_device *dev = to_net_dev(d);
1610 /* pass interface to uevent. */
1611 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1615 /* pass ifindex to uevent.
1616 * ifindex is useful as it won't change (interface name may change)
1617 * and is what RtNetlink uses natively.
1619 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1626 * netdev_release -- destroy and free a dead device.
1627 * Called when last reference to device kobject is gone.
1629 static void netdev_release(struct device *d)
1631 struct net_device *dev = to_net_dev(d);
1633 BUG_ON(dev->reg_state != NETREG_RELEASED);
1635 /* no need to wait for rcu grace period:
1636 * device is dead and about to be freed.
1638 kfree(rcu_access_pointer(dev->ifalias));
1639 netdev_freemem(dev);
1642 static const void *net_namespace(struct device *d)
1644 struct net_device *dev = to_net_dev(d);
1646 return dev_net(dev);
1649 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1651 struct net_device *dev = to_net_dev(d);
1652 const struct net *net = dev_net(dev);
1654 net_ns_get_ownership(net, uid, gid);
1657 static struct class net_class __ro_after_init = {
1659 .dev_release = netdev_release,
1660 .dev_groups = net_class_groups,
1661 .dev_uevent = netdev_uevent,
1662 .ns_type = &net_ns_type_operations,
1663 .namespace = net_namespace,
1664 .get_ownership = net_get_ownership,
1667 #ifdef CONFIG_OF_NET
1668 static int of_dev_node_match(struct device *dev, const void *data)
1673 ret = dev->parent->of_node == data;
1675 return ret == 0 ? dev->of_node == data : ret;
1679 * of_find_net_device_by_node - lookup the net device for the device node
1680 * @np: OF device node
1682 * Looks up the net_device structure corresponding with the device node.
1683 * If successful, returns a pointer to the net_device with the embedded
1684 * struct device refcount incremented by one, or NULL on failure. The
1685 * refcount must be dropped when done with the net_device.
1687 struct net_device *of_find_net_device_by_node(struct device_node *np)
1691 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1695 return to_net_dev(dev);
1697 EXPORT_SYMBOL(of_find_net_device_by_node);
1700 /* Delete sysfs entries but hold kobject reference until after all
1701 * netdev references are gone.
1703 void netdev_unregister_kobject(struct net_device *ndev)
1705 struct device *dev = &ndev->dev;
1707 if (!refcount_read(&dev_net(ndev)->count))
1708 dev_set_uevent_suppress(dev, 1);
1710 kobject_get(&dev->kobj);
1712 remove_queue_kobjects(ndev);
1714 pm_runtime_set_memalloc_noio(dev, false);
1719 /* Create sysfs entries for network device. */
1720 int netdev_register_kobject(struct net_device *ndev)
1722 struct device *dev = &ndev->dev;
1723 const struct attribute_group **groups = ndev->sysfs_groups;
1726 device_initialize(dev);
1727 dev->class = &net_class;
1728 dev->platform_data = ndev;
1729 dev->groups = groups;
1731 dev_set_name(dev, "%s", ndev->name);
1734 /* Allow for a device specific group */
1738 *groups++ = &netstat_group;
1740 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1741 if (ndev->ieee80211_ptr)
1742 *groups++ = &wireless_group;
1743 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1744 else if (ndev->wireless_handlers)
1745 *groups++ = &wireless_group;
1748 #endif /* CONFIG_SYSFS */
1750 error = device_add(dev);
1754 error = register_queue_kobjects(ndev);
1760 pm_runtime_set_memalloc_noio(dev, true);
1765 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
1768 return class_create_file_ns(&net_class, class_attr, ns);
1770 EXPORT_SYMBOL(netdev_class_create_file_ns);
1772 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
1775 class_remove_file_ns(&net_class, class_attr, ns);
1777 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1779 int __init netdev_kobject_init(void)
1781 kobj_ns_type_register(&net_ns_type_operations);
1782 return class_register(&net_class);