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 <net/switchdev.h>
16 #include <linux/if_arp.h>
17 #include <linux/slab.h>
18 #include <linux/sched/signal.h>
19 #include <linux/nsproxy.h>
21 #include <net/net_namespace.h>
22 #include <linux/rtnetlink.h>
23 #include <linux/vmalloc.h>
24 #include <linux/export.h>
25 #include <linux/jiffies.h>
26 #include <linux/pm_runtime.h>
28 #include <linux/of_net.h>
29 #include <linux/cpu.h>
31 #include "net-sysfs.h"
34 static const char fmt_hex[] = "%#x\n";
35 static const char fmt_dec[] = "%d\n";
36 static const char fmt_ulong[] = "%lu\n";
37 static const char fmt_u64[] = "%llu\n";
39 static inline int dev_isalive(const struct net_device *dev)
41 return dev->reg_state <= NETREG_REGISTERED;
44 /* use same locking rules as GIF* ioctl's */
45 static ssize_t netdev_show(const struct device *dev,
46 struct device_attribute *attr, char *buf,
47 ssize_t (*format)(const struct net_device *, char *))
49 struct net_device *ndev = to_net_dev(dev);
50 ssize_t ret = -EINVAL;
52 read_lock(&dev_base_lock);
53 if (dev_isalive(ndev))
54 ret = (*format)(ndev, buf);
55 read_unlock(&dev_base_lock);
60 /* generate a show function for simple field */
61 #define NETDEVICE_SHOW(field, format_string) \
62 static ssize_t format_##field(const struct net_device *dev, char *buf) \
64 return sprintf(buf, format_string, dev->field); \
66 static ssize_t field##_show(struct device *dev, \
67 struct device_attribute *attr, char *buf) \
69 return netdev_show(dev, attr, buf, format_##field); \
72 #define NETDEVICE_SHOW_RO(field, format_string) \
73 NETDEVICE_SHOW(field, format_string); \
74 static DEVICE_ATTR_RO(field)
76 #define NETDEVICE_SHOW_RW(field, format_string) \
77 NETDEVICE_SHOW(field, format_string); \
78 static DEVICE_ATTR_RW(field)
80 /* use same locking and permission rules as SIF* ioctl's */
81 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
82 const char *buf, size_t len,
83 int (*set)(struct net_device *, unsigned long))
85 struct net_device *netdev = to_net_dev(dev);
86 struct net *net = dev_net(netdev);
90 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
93 ret = kstrtoul(buf, 0, &new);
98 return restart_syscall();
100 if (dev_isalive(netdev)) {
101 ret = (*set)(netdev, new);
110 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
111 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
112 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
113 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
114 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
115 NETDEVICE_SHOW_RO(type, fmt_dec);
116 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
118 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
121 struct net_device *ndev = to_net_dev(dev);
123 return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
125 static DEVICE_ATTR_RO(iflink);
127 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
129 return sprintf(buf, fmt_dec, dev->name_assign_type);
132 static ssize_t name_assign_type_show(struct device *dev,
133 struct device_attribute *attr,
136 struct net_device *ndev = to_net_dev(dev);
137 ssize_t ret = -EINVAL;
139 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
140 ret = netdev_show(dev, attr, buf, format_name_assign_type);
144 static DEVICE_ATTR_RO(name_assign_type);
146 /* use same locking rules as GIFHWADDR ioctl's */
147 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
150 struct net_device *ndev = to_net_dev(dev);
151 ssize_t ret = -EINVAL;
153 read_lock(&dev_base_lock);
154 if (dev_isalive(ndev))
155 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
156 read_unlock(&dev_base_lock);
159 static DEVICE_ATTR_RO(address);
161 static ssize_t broadcast_show(struct device *dev,
162 struct device_attribute *attr, char *buf)
164 struct net_device *ndev = to_net_dev(dev);
166 if (dev_isalive(ndev))
167 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
170 static DEVICE_ATTR_RO(broadcast);
172 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
174 if (!netif_running(dev))
176 return dev_change_carrier(dev, (bool)new_carrier);
179 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
180 const char *buf, size_t len)
182 return netdev_store(dev, attr, buf, len, change_carrier);
185 static ssize_t carrier_show(struct device *dev,
186 struct device_attribute *attr, char *buf)
188 struct net_device *netdev = to_net_dev(dev);
190 if (netif_running(netdev))
191 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
195 static DEVICE_ATTR_RW(carrier);
197 static ssize_t speed_show(struct device *dev,
198 struct device_attribute *attr, char *buf)
200 struct net_device *netdev = to_net_dev(dev);
204 return restart_syscall();
206 if (netif_running(netdev) && netif_device_present(netdev)) {
207 struct ethtool_link_ksettings cmd;
209 if (!__ethtool_get_link_ksettings(netdev, &cmd))
210 ret = sprintf(buf, fmt_dec, cmd.base.speed);
215 static DEVICE_ATTR_RO(speed);
217 static ssize_t duplex_show(struct device *dev,
218 struct device_attribute *attr, char *buf)
220 struct net_device *netdev = to_net_dev(dev);
224 return restart_syscall();
226 if (netif_running(netdev)) {
227 struct ethtool_link_ksettings cmd;
229 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
232 switch (cmd.base.duplex) {
243 ret = sprintf(buf, "%s\n", duplex);
249 static DEVICE_ATTR_RO(duplex);
251 static ssize_t dormant_show(struct device *dev,
252 struct device_attribute *attr, char *buf)
254 struct net_device *netdev = to_net_dev(dev);
256 if (netif_running(netdev))
257 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
261 static DEVICE_ATTR_RO(dormant);
263 static const char *const operstates[] = {
265 "notpresent", /* currently unused */
268 "testing", /* currently unused */
273 static ssize_t operstate_show(struct device *dev,
274 struct device_attribute *attr, char *buf)
276 const struct net_device *netdev = to_net_dev(dev);
277 unsigned char operstate;
279 read_lock(&dev_base_lock);
280 operstate = netdev->operstate;
281 if (!netif_running(netdev))
282 operstate = IF_OPER_DOWN;
283 read_unlock(&dev_base_lock);
285 if (operstate >= ARRAY_SIZE(operstates))
286 return -EINVAL; /* should not happen */
288 return sprintf(buf, "%s\n", operstates[operstate]);
290 static DEVICE_ATTR_RO(operstate);
292 static ssize_t carrier_changes_show(struct device *dev,
293 struct device_attribute *attr,
296 struct net_device *netdev = to_net_dev(dev);
298 return sprintf(buf, fmt_dec,
299 atomic_read(&netdev->carrier_up_count) +
300 atomic_read(&netdev->carrier_down_count));
302 static DEVICE_ATTR_RO(carrier_changes);
304 static ssize_t carrier_up_count_show(struct device *dev,
305 struct device_attribute *attr,
308 struct net_device *netdev = to_net_dev(dev);
310 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
312 static DEVICE_ATTR_RO(carrier_up_count);
314 static ssize_t carrier_down_count_show(struct device *dev,
315 struct device_attribute *attr,
318 struct net_device *netdev = to_net_dev(dev);
320 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
322 static DEVICE_ATTR_RO(carrier_down_count);
324 /* read-write attributes */
326 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
328 return dev_set_mtu(dev, (int)new_mtu);
331 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
332 const char *buf, size_t len)
334 return netdev_store(dev, attr, buf, len, change_mtu);
336 NETDEVICE_SHOW_RW(mtu, fmt_dec);
338 static int change_flags(struct net_device *dev, unsigned long new_flags)
340 return dev_change_flags(dev, (unsigned int)new_flags);
343 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
344 const char *buf, size_t len)
346 return netdev_store(dev, attr, buf, len, change_flags);
348 NETDEVICE_SHOW_RW(flags, fmt_hex);
350 static ssize_t tx_queue_len_store(struct device *dev,
351 struct device_attribute *attr,
352 const char *buf, size_t len)
354 if (!capable(CAP_NET_ADMIN))
357 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
359 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
361 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
363 dev->gro_flush_timeout = val;
367 static ssize_t gro_flush_timeout_store(struct device *dev,
368 struct device_attribute *attr,
369 const char *buf, size_t len)
371 if (!capable(CAP_NET_ADMIN))
374 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
376 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
378 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
379 const char *buf, size_t len)
381 struct net_device *netdev = to_net_dev(dev);
382 struct net *net = dev_net(netdev);
386 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
389 /* ignore trailing newline */
390 if (len > 0 && buf[len - 1] == '\n')
394 return restart_syscall();
396 if (dev_isalive(netdev)) {
397 ret = dev_set_alias(netdev, buf, count);
401 netdev_state_change(netdev);
409 static ssize_t ifalias_show(struct device *dev,
410 struct device_attribute *attr, char *buf)
412 const struct net_device *netdev = to_net_dev(dev);
416 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
418 ret = sprintf(buf, "%s\n", tmp);
421 static DEVICE_ATTR_RW(ifalias);
423 static int change_group(struct net_device *dev, unsigned long new_group)
425 dev_set_group(dev, (int)new_group);
429 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
430 const char *buf, size_t len)
432 return netdev_store(dev, attr, buf, len, change_group);
434 NETDEVICE_SHOW(group, fmt_dec);
435 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
437 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
439 return dev_change_proto_down(dev, (bool)proto_down);
442 static ssize_t proto_down_store(struct device *dev,
443 struct device_attribute *attr,
444 const char *buf, size_t len)
446 return netdev_store(dev, attr, buf, len, change_proto_down);
448 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
450 static ssize_t phys_port_id_show(struct device *dev,
451 struct device_attribute *attr, char *buf)
453 struct net_device *netdev = to_net_dev(dev);
454 ssize_t ret = -EINVAL;
457 return restart_syscall();
459 if (dev_isalive(netdev)) {
460 struct netdev_phys_item_id ppid;
462 ret = dev_get_phys_port_id(netdev, &ppid);
464 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
470 static DEVICE_ATTR_RO(phys_port_id);
472 static ssize_t phys_port_name_show(struct device *dev,
473 struct device_attribute *attr, char *buf)
475 struct net_device *netdev = to_net_dev(dev);
476 ssize_t ret = -EINVAL;
479 return restart_syscall();
481 if (dev_isalive(netdev)) {
484 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
486 ret = sprintf(buf, "%s\n", name);
492 static DEVICE_ATTR_RO(phys_port_name);
494 static ssize_t phys_switch_id_show(struct device *dev,
495 struct device_attribute *attr, char *buf)
497 struct net_device *netdev = to_net_dev(dev);
498 ssize_t ret = -EINVAL;
501 return restart_syscall();
503 if (dev_isalive(netdev)) {
504 struct switchdev_attr attr = {
506 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
507 .flags = SWITCHDEV_F_NO_RECURSE,
510 ret = switchdev_port_attr_get(netdev, &attr);
512 ret = sprintf(buf, "%*phN\n", attr.u.ppid.id_len,
519 static DEVICE_ATTR_RO(phys_switch_id);
521 static struct attribute *net_class_attrs[] __ro_after_init = {
522 &dev_attr_netdev_group.attr,
524 &dev_attr_dev_id.attr,
525 &dev_attr_dev_port.attr,
526 &dev_attr_iflink.attr,
527 &dev_attr_ifindex.attr,
528 &dev_attr_name_assign_type.attr,
529 &dev_attr_addr_assign_type.attr,
530 &dev_attr_addr_len.attr,
531 &dev_attr_link_mode.attr,
532 &dev_attr_address.attr,
533 &dev_attr_broadcast.attr,
534 &dev_attr_speed.attr,
535 &dev_attr_duplex.attr,
536 &dev_attr_dormant.attr,
537 &dev_attr_operstate.attr,
538 &dev_attr_carrier_changes.attr,
539 &dev_attr_ifalias.attr,
540 &dev_attr_carrier.attr,
542 &dev_attr_flags.attr,
543 &dev_attr_tx_queue_len.attr,
544 &dev_attr_gro_flush_timeout.attr,
545 &dev_attr_phys_port_id.attr,
546 &dev_attr_phys_port_name.attr,
547 &dev_attr_phys_switch_id.attr,
548 &dev_attr_proto_down.attr,
549 &dev_attr_carrier_up_count.attr,
550 &dev_attr_carrier_down_count.attr,
553 ATTRIBUTE_GROUPS(net_class);
555 /* Show a given an attribute in the statistics group */
556 static ssize_t netstat_show(const struct device *d,
557 struct device_attribute *attr, char *buf,
558 unsigned long offset)
560 struct net_device *dev = to_net_dev(d);
561 ssize_t ret = -EINVAL;
563 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
564 offset % sizeof(u64) != 0);
566 read_lock(&dev_base_lock);
567 if (dev_isalive(dev)) {
568 struct rtnl_link_stats64 temp;
569 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
571 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
573 read_unlock(&dev_base_lock);
577 /* generate a read-only statistics attribute */
578 #define NETSTAT_ENTRY(name) \
579 static ssize_t name##_show(struct device *d, \
580 struct device_attribute *attr, char *buf) \
582 return netstat_show(d, attr, buf, \
583 offsetof(struct rtnl_link_stats64, name)); \
585 static DEVICE_ATTR_RO(name)
587 NETSTAT_ENTRY(rx_packets);
588 NETSTAT_ENTRY(tx_packets);
589 NETSTAT_ENTRY(rx_bytes);
590 NETSTAT_ENTRY(tx_bytes);
591 NETSTAT_ENTRY(rx_errors);
592 NETSTAT_ENTRY(tx_errors);
593 NETSTAT_ENTRY(rx_dropped);
594 NETSTAT_ENTRY(tx_dropped);
595 NETSTAT_ENTRY(multicast);
596 NETSTAT_ENTRY(collisions);
597 NETSTAT_ENTRY(rx_length_errors);
598 NETSTAT_ENTRY(rx_over_errors);
599 NETSTAT_ENTRY(rx_crc_errors);
600 NETSTAT_ENTRY(rx_frame_errors);
601 NETSTAT_ENTRY(rx_fifo_errors);
602 NETSTAT_ENTRY(rx_missed_errors);
603 NETSTAT_ENTRY(tx_aborted_errors);
604 NETSTAT_ENTRY(tx_carrier_errors);
605 NETSTAT_ENTRY(tx_fifo_errors);
606 NETSTAT_ENTRY(tx_heartbeat_errors);
607 NETSTAT_ENTRY(tx_window_errors);
608 NETSTAT_ENTRY(rx_compressed);
609 NETSTAT_ENTRY(tx_compressed);
610 NETSTAT_ENTRY(rx_nohandler);
612 static struct attribute *netstat_attrs[] __ro_after_init = {
613 &dev_attr_rx_packets.attr,
614 &dev_attr_tx_packets.attr,
615 &dev_attr_rx_bytes.attr,
616 &dev_attr_tx_bytes.attr,
617 &dev_attr_rx_errors.attr,
618 &dev_attr_tx_errors.attr,
619 &dev_attr_rx_dropped.attr,
620 &dev_attr_tx_dropped.attr,
621 &dev_attr_multicast.attr,
622 &dev_attr_collisions.attr,
623 &dev_attr_rx_length_errors.attr,
624 &dev_attr_rx_over_errors.attr,
625 &dev_attr_rx_crc_errors.attr,
626 &dev_attr_rx_frame_errors.attr,
627 &dev_attr_rx_fifo_errors.attr,
628 &dev_attr_rx_missed_errors.attr,
629 &dev_attr_tx_aborted_errors.attr,
630 &dev_attr_tx_carrier_errors.attr,
631 &dev_attr_tx_fifo_errors.attr,
632 &dev_attr_tx_heartbeat_errors.attr,
633 &dev_attr_tx_window_errors.attr,
634 &dev_attr_rx_compressed.attr,
635 &dev_attr_tx_compressed.attr,
636 &dev_attr_rx_nohandler.attr,
640 static const struct attribute_group netstat_group = {
641 .name = "statistics",
642 .attrs = netstat_attrs,
645 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
646 static struct attribute *wireless_attrs[] = {
650 static const struct attribute_group wireless_group = {
652 .attrs = wireless_attrs,
656 #else /* CONFIG_SYSFS */
657 #define net_class_groups NULL
658 #endif /* CONFIG_SYSFS */
661 #define to_rx_queue_attr(_attr) \
662 container_of(_attr, struct rx_queue_attribute, attr)
664 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
666 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
669 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
670 struct netdev_rx_queue *queue = to_rx_queue(kobj);
672 if (!attribute->show)
675 return attribute->show(queue, buf);
678 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
679 const char *buf, size_t count)
681 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
682 struct netdev_rx_queue *queue = to_rx_queue(kobj);
684 if (!attribute->store)
687 return attribute->store(queue, buf, count);
690 static const struct sysfs_ops rx_queue_sysfs_ops = {
691 .show = rx_queue_attr_show,
692 .store = rx_queue_attr_store,
696 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
702 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
706 map = rcu_dereference(queue->rps_map);
708 for (i = 0; i < map->len; i++)
709 cpumask_set_cpu(map->cpus[i], mask);
711 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
713 free_cpumask_var(mask);
715 return len < PAGE_SIZE ? len : -EINVAL;
718 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
719 const char *buf, size_t len)
721 struct rps_map *old_map, *map;
724 static DEFINE_MUTEX(rps_map_mutex);
726 if (!capable(CAP_NET_ADMIN))
729 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
732 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
734 free_cpumask_var(mask);
738 map = kzalloc(max_t(unsigned int,
739 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
742 free_cpumask_var(mask);
747 for_each_cpu_and(cpu, mask, cpu_online_mask)
748 map->cpus[i++] = cpu;
757 mutex_lock(&rps_map_mutex);
758 old_map = rcu_dereference_protected(queue->rps_map,
759 mutex_is_locked(&rps_map_mutex));
760 rcu_assign_pointer(queue->rps_map, map);
763 static_key_slow_inc(&rps_needed);
765 static_key_slow_dec(&rps_needed);
767 mutex_unlock(&rps_map_mutex);
770 kfree_rcu(old_map, rcu);
772 free_cpumask_var(mask);
776 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
779 struct rps_dev_flow_table *flow_table;
780 unsigned long val = 0;
783 flow_table = rcu_dereference(queue->rps_flow_table);
785 val = (unsigned long)flow_table->mask + 1;
788 return sprintf(buf, "%lu\n", val);
791 static void rps_dev_flow_table_release(struct rcu_head *rcu)
793 struct rps_dev_flow_table *table = container_of(rcu,
794 struct rps_dev_flow_table, rcu);
798 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
799 const char *buf, size_t len)
801 unsigned long mask, count;
802 struct rps_dev_flow_table *table, *old_table;
803 static DEFINE_SPINLOCK(rps_dev_flow_lock);
806 if (!capable(CAP_NET_ADMIN))
809 rc = kstrtoul(buf, 0, &count);
815 /* mask = roundup_pow_of_two(count) - 1;
816 * without overflows...
818 while ((mask | (mask >> 1)) != mask)
820 /* On 64 bit arches, must check mask fits in table->mask (u32),
821 * and on 32bit arches, must check
822 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
824 #if BITS_PER_LONG > 32
825 if (mask > (unsigned long)(u32)mask)
828 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
829 / sizeof(struct rps_dev_flow)) {
830 /* Enforce a limit to prevent overflow */
834 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
839 for (count = 0; count <= mask; count++)
840 table->flows[count].cpu = RPS_NO_CPU;
845 spin_lock(&rps_dev_flow_lock);
846 old_table = rcu_dereference_protected(queue->rps_flow_table,
847 lockdep_is_held(&rps_dev_flow_lock));
848 rcu_assign_pointer(queue->rps_flow_table, table);
849 spin_unlock(&rps_dev_flow_lock);
852 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
857 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
858 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
860 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
861 = __ATTR(rps_flow_cnt, 0644,
862 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
863 #endif /* CONFIG_RPS */
865 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
867 &rps_cpus_attribute.attr,
868 &rps_dev_flow_table_cnt_attribute.attr,
873 static void rx_queue_release(struct kobject *kobj)
875 struct netdev_rx_queue *queue = to_rx_queue(kobj);
878 struct rps_dev_flow_table *flow_table;
880 map = rcu_dereference_protected(queue->rps_map, 1);
882 RCU_INIT_POINTER(queue->rps_map, NULL);
886 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
888 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
889 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
893 memset(kobj, 0, sizeof(*kobj));
897 static const void *rx_queue_namespace(struct kobject *kobj)
899 struct netdev_rx_queue *queue = to_rx_queue(kobj);
900 struct device *dev = &queue->dev->dev;
901 const void *ns = NULL;
903 if (dev->class && dev->class->ns_type)
904 ns = dev->class->namespace(dev);
909 static void rx_queue_get_ownership(struct kobject *kobj,
910 kuid_t *uid, kgid_t *gid)
912 const struct net *net = rx_queue_namespace(kobj);
914 net_ns_get_ownership(net, uid, gid);
917 static struct kobj_type rx_queue_ktype __ro_after_init = {
918 .sysfs_ops = &rx_queue_sysfs_ops,
919 .release = rx_queue_release,
920 .default_attrs = rx_queue_default_attrs,
921 .namespace = rx_queue_namespace,
922 .get_ownership = rx_queue_get_ownership,
925 static int rx_queue_add_kobject(struct net_device *dev, int index)
927 struct netdev_rx_queue *queue = dev->_rx + index;
928 struct kobject *kobj = &queue->kobj;
931 /* Kobject_put later will trigger rx_queue_release call which
932 * decreases dev refcount: Take that reference here
934 dev_hold(queue->dev);
936 kobj->kset = dev->queues_kset;
937 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
942 if (dev->sysfs_rx_queue_group) {
943 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
948 kobject_uevent(kobj, KOBJ_ADD);
956 #endif /* CONFIG_SYSFS */
959 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
966 if (!dev->sysfs_rx_queue_group)
969 for (i = old_num; i < new_num; i++) {
970 error = rx_queue_add_kobject(dev, i);
977 while (--i >= new_num) {
978 struct kobject *kobj = &dev->_rx[i].kobj;
980 if (!refcount_read(&dev_net(dev)->count))
981 kobj->uevent_suppress = 1;
982 if (dev->sysfs_rx_queue_group)
983 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
995 * netdev_queue sysfs structures and functions.
997 struct netdev_queue_attribute {
998 struct attribute attr;
999 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1000 ssize_t (*store)(struct netdev_queue *queue,
1001 const char *buf, size_t len);
1003 #define to_netdev_queue_attr(_attr) \
1004 container_of(_attr, struct netdev_queue_attribute, attr)
1006 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1008 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1009 struct attribute *attr, char *buf)
1011 const struct netdev_queue_attribute *attribute
1012 = to_netdev_queue_attr(attr);
1013 struct netdev_queue *queue = to_netdev_queue(kobj);
1015 if (!attribute->show)
1018 return attribute->show(queue, buf);
1021 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1022 struct attribute *attr,
1023 const char *buf, size_t count)
1025 const struct netdev_queue_attribute *attribute
1026 = to_netdev_queue_attr(attr);
1027 struct netdev_queue *queue = to_netdev_queue(kobj);
1029 if (!attribute->store)
1032 return attribute->store(queue, buf, count);
1035 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1036 .show = netdev_queue_attr_show,
1037 .store = netdev_queue_attr_store,
1040 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1042 unsigned long trans_timeout;
1044 spin_lock_irq(&queue->_xmit_lock);
1045 trans_timeout = queue->trans_timeout;
1046 spin_unlock_irq(&queue->_xmit_lock);
1048 return sprintf(buf, fmt_ulong, trans_timeout);
1051 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1053 struct net_device *dev = queue->dev;
1056 i = queue - dev->_tx;
1057 BUG_ON(i >= dev->num_tx_queues);
1062 static ssize_t traffic_class_show(struct netdev_queue *queue,
1065 struct net_device *dev = queue->dev;
1069 if (!netif_is_multiqueue(dev))
1072 index = get_netdev_queue_index(queue);
1074 /* If queue belongs to subordinate dev use its TC mapping */
1075 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1077 tc = netdev_txq_to_tc(dev, index);
1081 /* We can report the traffic class one of two ways:
1082 * Subordinate device traffic classes are reported with the traffic
1083 * class first, and then the subordinate class so for example TC0 on
1084 * subordinate device 2 will be reported as "0-2". If the queue
1085 * belongs to the root device it will be reported with just the
1086 * traffic class, so just "0" for TC 0 for example.
1088 return dev->num_tc < 0 ? sprintf(buf, "%u%d\n", tc, dev->num_tc) :
1089 sprintf(buf, "%u\n", tc);
1093 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1096 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1099 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1100 const char *buf, size_t len)
1102 struct net_device *dev = queue->dev;
1103 int err, index = get_netdev_queue_index(queue);
1106 if (!capable(CAP_NET_ADMIN))
1109 err = kstrtou32(buf, 10, &rate);
1113 if (!rtnl_trylock())
1114 return restart_syscall();
1117 if (dev->netdev_ops->ndo_set_tx_maxrate)
1118 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1122 queue->tx_maxrate = rate;
1128 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1129 = __ATTR_RW(tx_maxrate);
1132 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1133 = __ATTR_RO(tx_timeout);
1135 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1136 = __ATTR_RO(traffic_class);
1140 * Byte queue limits sysfs structures and functions.
1142 static ssize_t bql_show(char *buf, unsigned int value)
1144 return sprintf(buf, "%u\n", value);
1147 static ssize_t bql_set(const char *buf, const size_t count,
1148 unsigned int *pvalue)
1153 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1154 value = DQL_MAX_LIMIT;
1156 err = kstrtouint(buf, 10, &value);
1159 if (value > DQL_MAX_LIMIT)
1168 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1171 struct dql *dql = &queue->dql;
1173 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1176 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1177 const char *buf, size_t len)
1179 struct dql *dql = &queue->dql;
1183 err = kstrtouint(buf, 10, &value);
1187 dql->slack_hold_time = msecs_to_jiffies(value);
1192 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1193 = __ATTR(hold_time, 0644,
1194 bql_show_hold_time, bql_set_hold_time);
1196 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1199 struct dql *dql = &queue->dql;
1201 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1204 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1205 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1207 #define BQL_ATTR(NAME, FIELD) \
1208 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1211 return bql_show(buf, queue->dql.FIELD); \
1214 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1215 const char *buf, size_t len) \
1217 return bql_set(buf, len, &queue->dql.FIELD); \
1220 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1221 = __ATTR(NAME, 0644, \
1222 bql_show_ ## NAME, bql_set_ ## NAME)
1224 BQL_ATTR(limit, limit);
1225 BQL_ATTR(limit_max, max_limit);
1226 BQL_ATTR(limit_min, min_limit);
1228 static struct attribute *dql_attrs[] __ro_after_init = {
1229 &bql_limit_attribute.attr,
1230 &bql_limit_max_attribute.attr,
1231 &bql_limit_min_attribute.attr,
1232 &bql_hold_time_attribute.attr,
1233 &bql_inflight_attribute.attr,
1237 static const struct attribute_group dql_group = {
1238 .name = "byte_queue_limits",
1241 #endif /* CONFIG_BQL */
1244 static ssize_t xps_cpus_show(struct netdev_queue *queue,
1247 int cpu, len, ret, num_tc = 1, tc = 0;
1248 struct net_device *dev = queue->dev;
1249 struct xps_dev_maps *dev_maps;
1251 unsigned long index;
1253 if (!netif_is_multiqueue(dev))
1256 index = get_netdev_queue_index(queue);
1258 if (!rtnl_trylock())
1259 return restart_syscall();
1262 /* Do not allow XPS on subordinate device directly */
1263 num_tc = dev->num_tc;
1266 goto err_rtnl_unlock;
1269 /* If queue belongs to subordinate dev use its map */
1270 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1272 tc = netdev_txq_to_tc(dev, index);
1275 goto err_rtnl_unlock;
1279 if (!zalloc_cpumask_var(&mask, GFP_KERNEL)) {
1281 goto err_rtnl_unlock;
1285 dev_maps = rcu_dereference(dev->xps_cpus_map);
1287 for_each_possible_cpu(cpu) {
1288 int i, tci = cpu * num_tc + tc;
1289 struct xps_map *map;
1291 map = rcu_dereference(dev_maps->attr_map[tci]);
1295 for (i = map->len; i--;) {
1296 if (map->queues[i] == index) {
1297 cpumask_set_cpu(cpu, mask);
1307 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1308 free_cpumask_var(mask);
1309 return len < PAGE_SIZE ? len : -EINVAL;
1316 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1317 const char *buf, size_t len)
1319 struct net_device *dev = queue->dev;
1320 unsigned long index;
1324 if (!netif_is_multiqueue(dev))
1327 if (!capable(CAP_NET_ADMIN))
1330 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1333 index = get_netdev_queue_index(queue);
1335 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1337 free_cpumask_var(mask);
1341 if (!rtnl_trylock()) {
1342 free_cpumask_var(mask);
1343 return restart_syscall();
1346 err = netif_set_xps_queue(dev, mask, index);
1349 free_cpumask_var(mask);
1354 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1355 = __ATTR_RW(xps_cpus);
1357 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1359 int j, len, ret, num_tc = 1, tc = 0;
1360 struct net_device *dev = queue->dev;
1361 struct xps_dev_maps *dev_maps;
1362 unsigned long *mask, index;
1364 index = get_netdev_queue_index(queue);
1366 if (!rtnl_trylock())
1367 return restart_syscall();
1370 num_tc = dev->num_tc;
1371 tc = netdev_txq_to_tc(dev, index);
1374 goto err_rtnl_unlock;
1377 mask = kcalloc(BITS_TO_LONGS(dev->num_rx_queues), sizeof(long),
1381 goto err_rtnl_unlock;
1385 dev_maps = rcu_dereference(dev->xps_rxqs_map);
1389 for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues),
1390 j < dev->num_rx_queues;) {
1391 int i, tci = j * num_tc + tc;
1392 struct xps_map *map;
1394 map = rcu_dereference(dev_maps->attr_map[tci]);
1398 for (i = map->len; i--;) {
1399 if (map->queues[i] == index) {
1410 len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues);
1413 return len < PAGE_SIZE ? len : -EINVAL;
1420 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1423 struct net_device *dev = queue->dev;
1424 struct net *net = dev_net(dev);
1425 unsigned long *mask, index;
1428 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1431 mask = kcalloc(BITS_TO_LONGS(dev->num_rx_queues), sizeof(long),
1436 index = get_netdev_queue_index(queue);
1438 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1444 if (!rtnl_trylock()) {
1446 return restart_syscall();
1450 err = __netif_set_xps_queue(dev, mask, index, true);
1459 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1460 = __ATTR_RW(xps_rxqs);
1461 #endif /* CONFIG_XPS */
1463 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1464 &queue_trans_timeout.attr,
1465 &queue_traffic_class.attr,
1467 &xps_cpus_attribute.attr,
1468 &xps_rxqs_attribute.attr,
1469 &queue_tx_maxrate.attr,
1474 static void netdev_queue_release(struct kobject *kobj)
1476 struct netdev_queue *queue = to_netdev_queue(kobj);
1478 memset(kobj, 0, sizeof(*kobj));
1479 dev_put(queue->dev);
1482 static const void *netdev_queue_namespace(struct kobject *kobj)
1484 struct netdev_queue *queue = to_netdev_queue(kobj);
1485 struct device *dev = &queue->dev->dev;
1486 const void *ns = NULL;
1488 if (dev->class && dev->class->ns_type)
1489 ns = dev->class->namespace(dev);
1494 static void netdev_queue_get_ownership(struct kobject *kobj,
1495 kuid_t *uid, kgid_t *gid)
1497 const struct net *net = netdev_queue_namespace(kobj);
1499 net_ns_get_ownership(net, uid, gid);
1502 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1503 .sysfs_ops = &netdev_queue_sysfs_ops,
1504 .release = netdev_queue_release,
1505 .default_attrs = netdev_queue_default_attrs,
1506 .namespace = netdev_queue_namespace,
1507 .get_ownership = netdev_queue_get_ownership,
1510 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1512 struct netdev_queue *queue = dev->_tx + index;
1513 struct kobject *kobj = &queue->kobj;
1516 /* Kobject_put later will trigger netdev_queue_release call
1517 * which decreases dev refcount: Take that reference here
1519 dev_hold(queue->dev);
1521 kobj->kset = dev->queues_kset;
1522 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1528 error = sysfs_create_group(kobj, &dql_group);
1533 kobject_uevent(kobj, KOBJ_ADD);
1540 #endif /* CONFIG_SYSFS */
1543 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1549 for (i = old_num; i < new_num; i++) {
1550 error = netdev_queue_add_kobject(dev, i);
1557 while (--i >= new_num) {
1558 struct netdev_queue *queue = dev->_tx + i;
1560 if (!refcount_read(&dev_net(dev)->count))
1561 queue->kobj.uevent_suppress = 1;
1563 sysfs_remove_group(&queue->kobj, &dql_group);
1565 kobject_put(&queue->kobj);
1571 #endif /* CONFIG_SYSFS */
1574 static int register_queue_kobjects(struct net_device *dev)
1576 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1579 dev->queues_kset = kset_create_and_add("queues",
1580 NULL, &dev->dev.kobj);
1581 if (!dev->queues_kset)
1583 real_rx = dev->real_num_rx_queues;
1585 real_tx = dev->real_num_tx_queues;
1587 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1592 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1600 netdev_queue_update_kobjects(dev, txq, 0);
1601 net_rx_queue_update_kobjects(dev, rxq, 0);
1603 kset_unregister(dev->queues_kset);
1608 static void remove_queue_kobjects(struct net_device *dev)
1610 int real_rx = 0, real_tx = 0;
1613 real_rx = dev->real_num_rx_queues;
1615 real_tx = dev->real_num_tx_queues;
1617 net_rx_queue_update_kobjects(dev, real_rx, 0);
1618 netdev_queue_update_kobjects(dev, real_tx, 0);
1620 dev->real_num_rx_queues = 0;
1621 dev->real_num_tx_queues = 0;
1623 kset_unregister(dev->queues_kset);
1627 static bool net_current_may_mount(void)
1629 struct net *net = current->nsproxy->net_ns;
1631 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1634 static void *net_grab_current_ns(void)
1636 struct net *ns = current->nsproxy->net_ns;
1637 #ifdef CONFIG_NET_NS
1639 refcount_inc(&ns->passive);
1644 static const void *net_initial_ns(void)
1649 static const void *net_netlink_ns(struct sock *sk)
1651 return sock_net(sk);
1654 const struct kobj_ns_type_operations net_ns_type_operations = {
1655 .type = KOBJ_NS_TYPE_NET,
1656 .current_may_mount = net_current_may_mount,
1657 .grab_current_ns = net_grab_current_ns,
1658 .netlink_ns = net_netlink_ns,
1659 .initial_ns = net_initial_ns,
1660 .drop_ns = net_drop_ns,
1662 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1664 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1666 struct net_device *dev = to_net_dev(d);
1669 /* pass interface to uevent. */
1670 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1674 /* pass ifindex to uevent.
1675 * ifindex is useful as it won't change (interface name may change)
1676 * and is what RtNetlink uses natively.
1678 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1685 * netdev_release -- destroy and free a dead device.
1686 * Called when last reference to device kobject is gone.
1688 static void netdev_release(struct device *d)
1690 struct net_device *dev = to_net_dev(d);
1692 BUG_ON(dev->reg_state != NETREG_RELEASED);
1694 /* no need to wait for rcu grace period:
1695 * device is dead and about to be freed.
1697 kfree(rcu_access_pointer(dev->ifalias));
1698 netdev_freemem(dev);
1701 static const void *net_namespace(struct device *d)
1703 struct net_device *dev = to_net_dev(d);
1705 return dev_net(dev);
1708 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1710 struct net_device *dev = to_net_dev(d);
1711 const struct net *net = dev_net(dev);
1713 net_ns_get_ownership(net, uid, gid);
1716 static struct class net_class __ro_after_init = {
1718 .dev_release = netdev_release,
1719 .dev_groups = net_class_groups,
1720 .dev_uevent = netdev_uevent,
1721 .ns_type = &net_ns_type_operations,
1722 .namespace = net_namespace,
1723 .get_ownership = net_get_ownership,
1726 #ifdef CONFIG_OF_NET
1727 static int of_dev_node_match(struct device *dev, const void *data)
1732 ret = dev->parent->of_node == data;
1734 return ret == 0 ? dev->of_node == data : ret;
1738 * of_find_net_device_by_node - lookup the net device for the device node
1739 * @np: OF device node
1741 * Looks up the net_device structure corresponding with the device node.
1742 * If successful, returns a pointer to the net_device with the embedded
1743 * struct device refcount incremented by one, or NULL on failure. The
1744 * refcount must be dropped when done with the net_device.
1746 struct net_device *of_find_net_device_by_node(struct device_node *np)
1750 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1754 return to_net_dev(dev);
1756 EXPORT_SYMBOL(of_find_net_device_by_node);
1759 /* Delete sysfs entries but hold kobject reference until after all
1760 * netdev references are gone.
1762 void netdev_unregister_kobject(struct net_device *ndev)
1764 struct device *dev = &ndev->dev;
1766 if (!refcount_read(&dev_net(ndev)->count))
1767 dev_set_uevent_suppress(dev, 1);
1769 kobject_get(&dev->kobj);
1771 remove_queue_kobjects(ndev);
1773 pm_runtime_set_memalloc_noio(dev, false);
1778 /* Create sysfs entries for network device. */
1779 int netdev_register_kobject(struct net_device *ndev)
1781 struct device *dev = &ndev->dev;
1782 const struct attribute_group **groups = ndev->sysfs_groups;
1785 device_initialize(dev);
1786 dev->class = &net_class;
1787 dev->platform_data = ndev;
1788 dev->groups = groups;
1790 dev_set_name(dev, "%s", ndev->name);
1793 /* Allow for a device specific group */
1797 *groups++ = &netstat_group;
1799 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1800 if (ndev->ieee80211_ptr)
1801 *groups++ = &wireless_group;
1802 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1803 else if (ndev->wireless_handlers)
1804 *groups++ = &wireless_group;
1807 #endif /* CONFIG_SYSFS */
1809 error = device_add(dev);
1813 error = register_queue_kobjects(ndev);
1819 pm_runtime_set_memalloc_noio(dev, true);
1824 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
1827 return class_create_file_ns(&net_class, class_attr, ns);
1829 EXPORT_SYMBOL(netdev_class_create_file_ns);
1831 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
1834 class_remove_file_ns(&net_class, class_attr, ns);
1836 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1838 int __init netdev_kobject_init(void)
1840 kobj_ns_type_register(&net_ns_type_operations);
1841 return class_register(&net_class);