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/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>
28 #include "net-sysfs.h"
31 static const char fmt_hex[] = "%#x\n";
32 static const char fmt_long_hex[] = "%#lx\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 static inline int dev_isalive(const struct net_device *dev)
39 return dev->reg_state <= NETREG_REGISTERED;
42 /* use same locking rules as GIF* ioctl's */
43 static ssize_t netdev_show(const struct device *dev,
44 struct device_attribute *attr, char *buf,
45 ssize_t (*format)(const struct net_device *, char *))
47 struct net_device *ndev = to_net_dev(dev);
48 ssize_t ret = -EINVAL;
50 read_lock(&dev_base_lock);
51 if (dev_isalive(ndev))
52 ret = (*format)(ndev, buf);
53 read_unlock(&dev_base_lock);
58 /* generate a show function for simple field */
59 #define NETDEVICE_SHOW(field, format_string) \
60 static ssize_t format_##field(const struct net_device *dev, char *buf) \
62 return sprintf(buf, format_string, dev->field); \
64 static ssize_t field##_show(struct device *dev, \
65 struct device_attribute *attr, char *buf) \
67 return netdev_show(dev, attr, buf, format_##field); \
70 #define NETDEVICE_SHOW_RO(field, format_string) \
71 NETDEVICE_SHOW(field, format_string); \
72 static DEVICE_ATTR_RO(field)
74 #define NETDEVICE_SHOW_RW(field, format_string) \
75 NETDEVICE_SHOW(field, format_string); \
76 static DEVICE_ATTR_RW(field)
78 /* use same locking and permission rules as SIF* ioctl's */
79 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
80 const char *buf, size_t len,
81 int (*set)(struct net_device *, unsigned long))
83 struct net_device *netdev = to_net_dev(dev);
84 struct net *net = dev_net(netdev);
88 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
91 ret = kstrtoul(buf, 0, &new);
96 return restart_syscall();
98 if (dev_isalive(netdev)) {
99 if ((ret = (*set)(netdev, new)) == 0)
107 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
108 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
109 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
110 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
111 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
112 NETDEVICE_SHOW_RO(type, fmt_dec);
113 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
115 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
118 struct net_device *ndev = to_net_dev(dev);
120 return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
122 static DEVICE_ATTR_RO(iflink);
124 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
126 return sprintf(buf, fmt_dec, dev->name_assign_type);
129 static ssize_t name_assign_type_show(struct device *dev,
130 struct device_attribute *attr,
133 struct net_device *ndev = to_net_dev(dev);
134 ssize_t ret = -EINVAL;
136 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
137 ret = netdev_show(dev, attr, buf, format_name_assign_type);
141 static DEVICE_ATTR_RO(name_assign_type);
143 /* use same locking rules as GIFHWADDR ioctl's */
144 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
147 struct net_device *ndev = to_net_dev(dev);
148 ssize_t ret = -EINVAL;
150 read_lock(&dev_base_lock);
151 if (dev_isalive(ndev))
152 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
153 read_unlock(&dev_base_lock);
156 static DEVICE_ATTR_RO(address);
158 static ssize_t broadcast_show(struct device *dev,
159 struct device_attribute *attr, char *buf)
161 struct net_device *ndev = to_net_dev(dev);
162 if (dev_isalive(ndev))
163 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
166 static DEVICE_ATTR_RO(broadcast);
168 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
170 if (!netif_running(dev))
172 return dev_change_carrier(dev, (bool) new_carrier);
175 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
176 const char *buf, size_t len)
178 return netdev_store(dev, attr, buf, len, change_carrier);
181 static ssize_t carrier_show(struct device *dev,
182 struct device_attribute *attr, char *buf)
184 struct net_device *netdev = to_net_dev(dev);
185 if (netif_running(netdev)) {
186 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
190 static DEVICE_ATTR_RW(carrier);
192 static ssize_t speed_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 struct ethtool_cmd cmd;
203 if (!__ethtool_get_settings(netdev, &cmd))
204 ret = sprintf(buf, fmt_dec, ethtool_cmd_speed(&cmd));
209 static DEVICE_ATTR_RO(speed);
211 static ssize_t duplex_show(struct device *dev,
212 struct device_attribute *attr, char *buf)
214 struct net_device *netdev = to_net_dev(dev);
218 return restart_syscall();
220 if (netif_running(netdev)) {
221 struct ethtool_cmd cmd;
222 if (!__ethtool_get_settings(netdev, &cmd)) {
224 switch (cmd.duplex) {
235 ret = sprintf(buf, "%s\n", duplex);
241 static DEVICE_ATTR_RO(duplex);
243 static ssize_t dormant_show(struct device *dev,
244 struct device_attribute *attr, char *buf)
246 struct net_device *netdev = to_net_dev(dev);
248 if (netif_running(netdev))
249 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
253 static DEVICE_ATTR_RO(dormant);
255 static const char *const operstates[] = {
257 "notpresent", /* currently unused */
260 "testing", /* currently unused */
265 static ssize_t operstate_show(struct device *dev,
266 struct device_attribute *attr, char *buf)
268 const struct net_device *netdev = to_net_dev(dev);
269 unsigned char operstate;
271 read_lock(&dev_base_lock);
272 operstate = netdev->operstate;
273 if (!netif_running(netdev))
274 operstate = IF_OPER_DOWN;
275 read_unlock(&dev_base_lock);
277 if (operstate >= ARRAY_SIZE(operstates))
278 return -EINVAL; /* should not happen */
280 return sprintf(buf, "%s\n", operstates[operstate]);
282 static DEVICE_ATTR_RO(operstate);
284 static ssize_t carrier_changes_show(struct device *dev,
285 struct device_attribute *attr,
288 struct net_device *netdev = to_net_dev(dev);
289 return sprintf(buf, fmt_dec,
290 atomic_read(&netdev->carrier_changes));
292 static DEVICE_ATTR_RO(carrier_changes);
294 /* read-write attributes */
296 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
298 return dev_set_mtu(dev, (int) new_mtu);
301 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
302 const char *buf, size_t len)
304 return netdev_store(dev, attr, buf, len, change_mtu);
306 NETDEVICE_SHOW_RW(mtu, fmt_dec);
308 static int change_flags(struct net_device *dev, unsigned long new_flags)
310 return dev_change_flags(dev, (unsigned int) new_flags);
313 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
314 const char *buf, size_t len)
316 return netdev_store(dev, attr, buf, len, change_flags);
318 NETDEVICE_SHOW_RW(flags, fmt_hex);
320 static int change_tx_queue_len(struct net_device *dev, unsigned long new_len)
322 dev->tx_queue_len = new_len;
326 static ssize_t tx_queue_len_store(struct device *dev,
327 struct device_attribute *attr,
328 const char *buf, size_t len)
330 if (!capable(CAP_NET_ADMIN))
333 return netdev_store(dev, attr, buf, len, change_tx_queue_len);
335 NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
337 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
339 dev->gro_flush_timeout = val;
343 static ssize_t gro_flush_timeout_store(struct device *dev,
344 struct device_attribute *attr,
345 const char *buf, size_t len)
347 if (!capable(CAP_NET_ADMIN))
350 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
352 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
354 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
355 const char *buf, size_t len)
357 struct net_device *netdev = to_net_dev(dev);
358 struct net *net = dev_net(netdev);
362 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
365 /* ignore trailing newline */
366 if (len > 0 && buf[len - 1] == '\n')
370 return restart_syscall();
371 ret = dev_set_alias(netdev, buf, count);
374 return ret < 0 ? ret : len;
377 static ssize_t ifalias_show(struct device *dev,
378 struct device_attribute *attr, char *buf)
380 const struct net_device *netdev = to_net_dev(dev);
384 return restart_syscall();
386 ret = sprintf(buf, "%s\n", netdev->ifalias);
390 static DEVICE_ATTR_RW(ifalias);
392 static int change_group(struct net_device *dev, unsigned long new_group)
394 dev_set_group(dev, (int) new_group);
398 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
399 const char *buf, size_t len)
401 return netdev_store(dev, attr, buf, len, change_group);
403 NETDEVICE_SHOW(group, fmt_dec);
404 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
406 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
408 return dev_change_proto_down(dev, (bool) proto_down);
411 static ssize_t proto_down_store(struct device *dev,
412 struct device_attribute *attr,
413 const char *buf, size_t len)
415 return netdev_store(dev, attr, buf, len, change_proto_down);
417 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
419 static ssize_t phys_port_id_show(struct device *dev,
420 struct device_attribute *attr, char *buf)
422 struct net_device *netdev = to_net_dev(dev);
423 ssize_t ret = -EINVAL;
426 return restart_syscall();
428 if (dev_isalive(netdev)) {
429 struct netdev_phys_item_id ppid;
431 ret = dev_get_phys_port_id(netdev, &ppid);
433 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
439 static DEVICE_ATTR_RO(phys_port_id);
441 static ssize_t phys_port_name_show(struct device *dev,
442 struct device_attribute *attr, char *buf)
444 struct net_device *netdev = to_net_dev(dev);
445 ssize_t ret = -EINVAL;
448 return restart_syscall();
450 if (dev_isalive(netdev)) {
453 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
455 ret = sprintf(buf, "%s\n", name);
461 static DEVICE_ATTR_RO(phys_port_name);
463 static ssize_t phys_switch_id_show(struct device *dev,
464 struct device_attribute *attr, char *buf)
466 struct net_device *netdev = to_net_dev(dev);
467 ssize_t ret = -EINVAL;
470 return restart_syscall();
472 if (dev_isalive(netdev)) {
473 struct switchdev_attr attr = {
475 .id = SWITCHDEV_ATTR_ID_PORT_PARENT_ID,
476 .flags = SWITCHDEV_F_NO_RECURSE,
479 ret = switchdev_port_attr_get(netdev, &attr);
481 ret = sprintf(buf, "%*phN\n", attr.u.ppid.id_len,
488 static DEVICE_ATTR_RO(phys_switch_id);
490 static struct attribute *net_class_attrs[] = {
491 &dev_attr_netdev_group.attr,
493 &dev_attr_dev_id.attr,
494 &dev_attr_dev_port.attr,
495 &dev_attr_iflink.attr,
496 &dev_attr_ifindex.attr,
497 &dev_attr_name_assign_type.attr,
498 &dev_attr_addr_assign_type.attr,
499 &dev_attr_addr_len.attr,
500 &dev_attr_link_mode.attr,
501 &dev_attr_address.attr,
502 &dev_attr_broadcast.attr,
503 &dev_attr_speed.attr,
504 &dev_attr_duplex.attr,
505 &dev_attr_dormant.attr,
506 &dev_attr_operstate.attr,
507 &dev_attr_carrier_changes.attr,
508 &dev_attr_ifalias.attr,
509 &dev_attr_carrier.attr,
511 &dev_attr_flags.attr,
512 &dev_attr_tx_queue_len.attr,
513 &dev_attr_gro_flush_timeout.attr,
514 &dev_attr_phys_port_id.attr,
515 &dev_attr_phys_port_name.attr,
516 &dev_attr_phys_switch_id.attr,
517 &dev_attr_proto_down.attr,
520 ATTRIBUTE_GROUPS(net_class);
522 /* Show a given an attribute in the statistics group */
523 static ssize_t netstat_show(const struct device *d,
524 struct device_attribute *attr, char *buf,
525 unsigned long offset)
527 struct net_device *dev = to_net_dev(d);
528 ssize_t ret = -EINVAL;
530 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
531 offset % sizeof(u64) != 0);
533 read_lock(&dev_base_lock);
534 if (dev_isalive(dev)) {
535 struct rtnl_link_stats64 temp;
536 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
538 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
540 read_unlock(&dev_base_lock);
544 /* generate a read-only statistics attribute */
545 #define NETSTAT_ENTRY(name) \
546 static ssize_t name##_show(struct device *d, \
547 struct device_attribute *attr, char *buf) \
549 return netstat_show(d, attr, buf, \
550 offsetof(struct rtnl_link_stats64, name)); \
552 static DEVICE_ATTR_RO(name)
554 NETSTAT_ENTRY(rx_packets);
555 NETSTAT_ENTRY(tx_packets);
556 NETSTAT_ENTRY(rx_bytes);
557 NETSTAT_ENTRY(tx_bytes);
558 NETSTAT_ENTRY(rx_errors);
559 NETSTAT_ENTRY(tx_errors);
560 NETSTAT_ENTRY(rx_dropped);
561 NETSTAT_ENTRY(tx_dropped);
562 NETSTAT_ENTRY(multicast);
563 NETSTAT_ENTRY(collisions);
564 NETSTAT_ENTRY(rx_length_errors);
565 NETSTAT_ENTRY(rx_over_errors);
566 NETSTAT_ENTRY(rx_crc_errors);
567 NETSTAT_ENTRY(rx_frame_errors);
568 NETSTAT_ENTRY(rx_fifo_errors);
569 NETSTAT_ENTRY(rx_missed_errors);
570 NETSTAT_ENTRY(tx_aborted_errors);
571 NETSTAT_ENTRY(tx_carrier_errors);
572 NETSTAT_ENTRY(tx_fifo_errors);
573 NETSTAT_ENTRY(tx_heartbeat_errors);
574 NETSTAT_ENTRY(tx_window_errors);
575 NETSTAT_ENTRY(rx_compressed);
576 NETSTAT_ENTRY(tx_compressed);
577 NETSTAT_ENTRY(rx_nohandler);
579 static struct attribute *netstat_attrs[] = {
580 &dev_attr_rx_packets.attr,
581 &dev_attr_tx_packets.attr,
582 &dev_attr_rx_bytes.attr,
583 &dev_attr_tx_bytes.attr,
584 &dev_attr_rx_errors.attr,
585 &dev_attr_tx_errors.attr,
586 &dev_attr_rx_dropped.attr,
587 &dev_attr_tx_dropped.attr,
588 &dev_attr_multicast.attr,
589 &dev_attr_collisions.attr,
590 &dev_attr_rx_length_errors.attr,
591 &dev_attr_rx_over_errors.attr,
592 &dev_attr_rx_crc_errors.attr,
593 &dev_attr_rx_frame_errors.attr,
594 &dev_attr_rx_fifo_errors.attr,
595 &dev_attr_rx_missed_errors.attr,
596 &dev_attr_tx_aborted_errors.attr,
597 &dev_attr_tx_carrier_errors.attr,
598 &dev_attr_tx_fifo_errors.attr,
599 &dev_attr_tx_heartbeat_errors.attr,
600 &dev_attr_tx_window_errors.attr,
601 &dev_attr_rx_compressed.attr,
602 &dev_attr_tx_compressed.attr,
603 &dev_attr_rx_nohandler.attr,
608 static struct attribute_group netstat_group = {
609 .name = "statistics",
610 .attrs = netstat_attrs,
613 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
614 static struct attribute *wireless_attrs[] = {
618 static struct attribute_group wireless_group = {
620 .attrs = wireless_attrs,
624 #else /* CONFIG_SYSFS */
625 #define net_class_groups NULL
626 #endif /* CONFIG_SYSFS */
629 #define to_rx_queue_attr(_attr) container_of(_attr, \
630 struct rx_queue_attribute, attr)
632 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
634 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
637 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
638 struct netdev_rx_queue *queue = to_rx_queue(kobj);
640 if (!attribute->show)
643 return attribute->show(queue, attribute, buf);
646 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
647 const char *buf, size_t count)
649 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
650 struct netdev_rx_queue *queue = to_rx_queue(kobj);
652 if (!attribute->store)
655 return attribute->store(queue, attribute, buf, count);
658 static const struct sysfs_ops rx_queue_sysfs_ops = {
659 .show = rx_queue_attr_show,
660 .store = rx_queue_attr_store,
664 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
665 struct rx_queue_attribute *attribute, char *buf)
671 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
675 map = rcu_dereference(queue->rps_map);
677 for (i = 0; i < map->len; i++)
678 cpumask_set_cpu(map->cpus[i], mask);
680 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
682 free_cpumask_var(mask);
684 return len < PAGE_SIZE ? len : -EINVAL;
687 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
688 struct rx_queue_attribute *attribute,
689 const char *buf, size_t len)
691 struct rps_map *old_map, *map;
694 static DEFINE_MUTEX(rps_map_mutex);
696 if (!capable(CAP_NET_ADMIN))
699 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
702 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
704 free_cpumask_var(mask);
708 map = kzalloc(max_t(unsigned int,
709 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
712 free_cpumask_var(mask);
717 for_each_cpu_and(cpu, mask, cpu_online_mask)
718 map->cpus[i++] = cpu;
727 mutex_lock(&rps_map_mutex);
728 old_map = rcu_dereference_protected(queue->rps_map,
729 mutex_is_locked(&rps_map_mutex));
730 rcu_assign_pointer(queue->rps_map, map);
733 static_key_slow_inc(&rps_needed);
735 static_key_slow_dec(&rps_needed);
737 mutex_unlock(&rps_map_mutex);
740 kfree_rcu(old_map, rcu);
742 free_cpumask_var(mask);
746 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
747 struct rx_queue_attribute *attr,
750 struct rps_dev_flow_table *flow_table;
751 unsigned long val = 0;
754 flow_table = rcu_dereference(queue->rps_flow_table);
756 val = (unsigned long)flow_table->mask + 1;
759 return sprintf(buf, "%lu\n", val);
762 static void rps_dev_flow_table_release(struct rcu_head *rcu)
764 struct rps_dev_flow_table *table = container_of(rcu,
765 struct rps_dev_flow_table, rcu);
769 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
770 struct rx_queue_attribute *attr,
771 const char *buf, size_t len)
773 unsigned long mask, count;
774 struct rps_dev_flow_table *table, *old_table;
775 static DEFINE_SPINLOCK(rps_dev_flow_lock);
778 if (!capable(CAP_NET_ADMIN))
781 rc = kstrtoul(buf, 0, &count);
787 /* mask = roundup_pow_of_two(count) - 1;
788 * without overflows...
790 while ((mask | (mask >> 1)) != mask)
792 /* On 64 bit arches, must check mask fits in table->mask (u32),
793 * and on 32bit arches, must check
794 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
796 #if BITS_PER_LONG > 32
797 if (mask > (unsigned long)(u32)mask)
800 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
801 / sizeof(struct rps_dev_flow)) {
802 /* Enforce a limit to prevent overflow */
806 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
811 for (count = 0; count <= mask; count++)
812 table->flows[count].cpu = RPS_NO_CPU;
816 spin_lock(&rps_dev_flow_lock);
817 old_table = rcu_dereference_protected(queue->rps_flow_table,
818 lockdep_is_held(&rps_dev_flow_lock));
819 rcu_assign_pointer(queue->rps_flow_table, table);
820 spin_unlock(&rps_dev_flow_lock);
823 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
828 static struct rx_queue_attribute rps_cpus_attribute =
829 __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
832 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
833 __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
834 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
835 #endif /* CONFIG_RPS */
837 static struct attribute *rx_queue_default_attrs[] = {
839 &rps_cpus_attribute.attr,
840 &rps_dev_flow_table_cnt_attribute.attr,
845 static void rx_queue_release(struct kobject *kobj)
847 struct netdev_rx_queue *queue = to_rx_queue(kobj);
850 struct rps_dev_flow_table *flow_table;
853 map = rcu_dereference_protected(queue->rps_map, 1);
855 RCU_INIT_POINTER(queue->rps_map, NULL);
859 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
861 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
862 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
866 memset(kobj, 0, sizeof(*kobj));
870 static const void *rx_queue_namespace(struct kobject *kobj)
872 struct netdev_rx_queue *queue = to_rx_queue(kobj);
873 struct device *dev = &queue->dev->dev;
874 const void *ns = NULL;
876 if (dev->class && dev->class->ns_type)
877 ns = dev->class->namespace(dev);
882 static struct kobj_type rx_queue_ktype = {
883 .sysfs_ops = &rx_queue_sysfs_ops,
884 .release = rx_queue_release,
885 .default_attrs = rx_queue_default_attrs,
886 .namespace = rx_queue_namespace
889 static int rx_queue_add_kobject(struct net_device *dev, int index)
891 struct netdev_rx_queue *queue = dev->_rx + index;
892 struct kobject *kobj = &queue->kobj;
895 kobj->kset = dev->queues_kset;
896 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
901 if (dev->sysfs_rx_queue_group) {
902 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
907 kobject_uevent(kobj, KOBJ_ADD);
908 dev_hold(queue->dev);
915 #endif /* CONFIG_SYSFS */
918 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
925 if (!dev->sysfs_rx_queue_group)
928 for (i = old_num; i < new_num; i++) {
929 error = rx_queue_add_kobject(dev, i);
936 while (--i >= new_num) {
937 if (dev->sysfs_rx_queue_group)
938 sysfs_remove_group(&dev->_rx[i].kobj,
939 dev->sysfs_rx_queue_group);
940 kobject_put(&dev->_rx[i].kobj);
951 * netdev_queue sysfs structures and functions.
953 struct netdev_queue_attribute {
954 struct attribute attr;
955 ssize_t (*show)(struct netdev_queue *queue,
956 struct netdev_queue_attribute *attr, char *buf);
957 ssize_t (*store)(struct netdev_queue *queue,
958 struct netdev_queue_attribute *attr, const char *buf, size_t len);
960 #define to_netdev_queue_attr(_attr) container_of(_attr, \
961 struct netdev_queue_attribute, attr)
963 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
965 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
966 struct attribute *attr, char *buf)
968 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
969 struct netdev_queue *queue = to_netdev_queue(kobj);
971 if (!attribute->show)
974 return attribute->show(queue, attribute, buf);
977 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
978 struct attribute *attr,
979 const char *buf, size_t count)
981 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
982 struct netdev_queue *queue = to_netdev_queue(kobj);
984 if (!attribute->store)
987 return attribute->store(queue, attribute, buf, count);
990 static const struct sysfs_ops netdev_queue_sysfs_ops = {
991 .show = netdev_queue_attr_show,
992 .store = netdev_queue_attr_store,
995 static ssize_t show_trans_timeout(struct netdev_queue *queue,
996 struct netdev_queue_attribute *attribute,
999 unsigned long trans_timeout;
1001 spin_lock_irq(&queue->_xmit_lock);
1002 trans_timeout = queue->trans_timeout;
1003 spin_unlock_irq(&queue->_xmit_lock);
1005 return sprintf(buf, "%lu", trans_timeout);
1009 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1011 struct net_device *dev = queue->dev;
1014 i = queue - dev->_tx;
1015 BUG_ON(i >= dev->num_tx_queues);
1020 static ssize_t show_tx_maxrate(struct netdev_queue *queue,
1021 struct netdev_queue_attribute *attribute,
1024 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1027 static ssize_t set_tx_maxrate(struct netdev_queue *queue,
1028 struct netdev_queue_attribute *attribute,
1029 const char *buf, size_t len)
1031 struct net_device *dev = queue->dev;
1032 int err, index = get_netdev_queue_index(queue);
1035 err = kstrtou32(buf, 10, &rate);
1039 if (!rtnl_trylock())
1040 return restart_syscall();
1043 if (dev->netdev_ops->ndo_set_tx_maxrate)
1044 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1048 queue->tx_maxrate = rate;
1054 static struct netdev_queue_attribute queue_tx_maxrate =
1055 __ATTR(tx_maxrate, S_IRUGO | S_IWUSR,
1056 show_tx_maxrate, set_tx_maxrate);
1059 static struct netdev_queue_attribute queue_trans_timeout =
1060 __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
1064 * Byte queue limits sysfs structures and functions.
1066 static ssize_t bql_show(char *buf, unsigned int value)
1068 return sprintf(buf, "%u\n", value);
1071 static ssize_t bql_set(const char *buf, const size_t count,
1072 unsigned int *pvalue)
1077 if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
1078 value = DQL_MAX_LIMIT;
1080 err = kstrtouint(buf, 10, &value);
1083 if (value > DQL_MAX_LIMIT)
1092 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1093 struct netdev_queue_attribute *attr,
1096 struct dql *dql = &queue->dql;
1098 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1101 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1102 struct netdev_queue_attribute *attribute,
1103 const char *buf, size_t len)
1105 struct dql *dql = &queue->dql;
1109 err = kstrtouint(buf, 10, &value);
1113 dql->slack_hold_time = msecs_to_jiffies(value);
1118 static struct netdev_queue_attribute bql_hold_time_attribute =
1119 __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
1122 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1123 struct netdev_queue_attribute *attr,
1126 struct dql *dql = &queue->dql;
1128 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1131 static struct netdev_queue_attribute bql_inflight_attribute =
1132 __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
1134 #define BQL_ATTR(NAME, FIELD) \
1135 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1136 struct netdev_queue_attribute *attr, \
1139 return bql_show(buf, queue->dql.FIELD); \
1142 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1143 struct netdev_queue_attribute *attr, \
1144 const char *buf, size_t len) \
1146 return bql_set(buf, len, &queue->dql.FIELD); \
1149 static struct netdev_queue_attribute bql_ ## NAME ## _attribute = \
1150 __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME, \
1153 BQL_ATTR(limit, limit)
1154 BQL_ATTR(limit_max, max_limit)
1155 BQL_ATTR(limit_min, min_limit)
1157 static struct attribute *dql_attrs[] = {
1158 &bql_limit_attribute.attr,
1159 &bql_limit_max_attribute.attr,
1160 &bql_limit_min_attribute.attr,
1161 &bql_hold_time_attribute.attr,
1162 &bql_inflight_attribute.attr,
1166 static struct attribute_group dql_group = {
1167 .name = "byte_queue_limits",
1170 #endif /* CONFIG_BQL */
1173 static ssize_t show_xps_map(struct netdev_queue *queue,
1174 struct netdev_queue_attribute *attribute, char *buf)
1176 struct net_device *dev = queue->dev;
1177 struct xps_dev_maps *dev_maps;
1179 unsigned long index;
1182 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1185 index = get_netdev_queue_index(queue);
1188 dev_maps = rcu_dereference(dev->xps_maps);
1190 for_each_possible_cpu(i) {
1191 struct xps_map *map =
1192 rcu_dereference(dev_maps->cpu_map[i]);
1195 for (j = 0; j < map->len; j++) {
1196 if (map->queues[j] == index) {
1197 cpumask_set_cpu(i, mask);
1206 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1207 free_cpumask_var(mask);
1208 return len < PAGE_SIZE ? len : -EINVAL;
1211 static ssize_t store_xps_map(struct netdev_queue *queue,
1212 struct netdev_queue_attribute *attribute,
1213 const char *buf, size_t len)
1215 struct net_device *dev = queue->dev;
1216 unsigned long index;
1220 if (!capable(CAP_NET_ADMIN))
1223 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1226 index = get_netdev_queue_index(queue);
1228 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1230 free_cpumask_var(mask);
1234 err = netif_set_xps_queue(dev, mask, index);
1236 free_cpumask_var(mask);
1241 static struct netdev_queue_attribute xps_cpus_attribute =
1242 __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1243 #endif /* CONFIG_XPS */
1245 static struct attribute *netdev_queue_default_attrs[] = {
1246 &queue_trans_timeout.attr,
1248 &xps_cpus_attribute.attr,
1249 &queue_tx_maxrate.attr,
1254 static void netdev_queue_release(struct kobject *kobj)
1256 struct netdev_queue *queue = to_netdev_queue(kobj);
1258 memset(kobj, 0, sizeof(*kobj));
1259 dev_put(queue->dev);
1262 static const void *netdev_queue_namespace(struct kobject *kobj)
1264 struct netdev_queue *queue = to_netdev_queue(kobj);
1265 struct device *dev = &queue->dev->dev;
1266 const void *ns = NULL;
1268 if (dev->class && dev->class->ns_type)
1269 ns = dev->class->namespace(dev);
1274 static struct kobj_type netdev_queue_ktype = {
1275 .sysfs_ops = &netdev_queue_sysfs_ops,
1276 .release = netdev_queue_release,
1277 .default_attrs = netdev_queue_default_attrs,
1278 .namespace = netdev_queue_namespace,
1281 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1283 struct netdev_queue *queue = dev->_tx + index;
1284 struct kobject *kobj = &queue->kobj;
1287 kobj->kset = dev->queues_kset;
1288 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1294 error = sysfs_create_group(kobj, &dql_group);
1299 kobject_uevent(kobj, KOBJ_ADD);
1300 dev_hold(queue->dev);
1307 #endif /* CONFIG_SYSFS */
1310 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1316 for (i = old_num; i < new_num; i++) {
1317 error = netdev_queue_add_kobject(dev, i);
1324 while (--i >= new_num) {
1325 struct netdev_queue *queue = dev->_tx + i;
1328 sysfs_remove_group(&queue->kobj, &dql_group);
1330 kobject_put(&queue->kobj);
1336 #endif /* CONFIG_SYSFS */
1339 static int register_queue_kobjects(struct net_device *dev)
1341 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1344 dev->queues_kset = kset_create_and_add("queues",
1345 NULL, &dev->dev.kobj);
1346 if (!dev->queues_kset)
1348 real_rx = dev->real_num_rx_queues;
1350 real_tx = dev->real_num_tx_queues;
1352 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1357 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1365 netdev_queue_update_kobjects(dev, txq, 0);
1366 net_rx_queue_update_kobjects(dev, rxq, 0);
1370 static void remove_queue_kobjects(struct net_device *dev)
1372 int real_rx = 0, real_tx = 0;
1375 real_rx = dev->real_num_rx_queues;
1377 real_tx = dev->real_num_tx_queues;
1379 net_rx_queue_update_kobjects(dev, real_rx, 0);
1380 netdev_queue_update_kobjects(dev, real_tx, 0);
1382 kset_unregister(dev->queues_kset);
1386 static bool net_current_may_mount(void)
1388 struct net *net = current->nsproxy->net_ns;
1390 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1393 static void *net_grab_current_ns(void)
1395 struct net *ns = current->nsproxy->net_ns;
1396 #ifdef CONFIG_NET_NS
1398 atomic_inc(&ns->passive);
1403 static const void *net_initial_ns(void)
1408 static const void *net_netlink_ns(struct sock *sk)
1410 return sock_net(sk);
1413 struct kobj_ns_type_operations net_ns_type_operations = {
1414 .type = KOBJ_NS_TYPE_NET,
1415 .current_may_mount = net_current_may_mount,
1416 .grab_current_ns = net_grab_current_ns,
1417 .netlink_ns = net_netlink_ns,
1418 .initial_ns = net_initial_ns,
1419 .drop_ns = net_drop_ns,
1421 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1423 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1425 struct net_device *dev = to_net_dev(d);
1428 /* pass interface to uevent. */
1429 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1433 /* pass ifindex to uevent.
1434 * ifindex is useful as it won't change (interface name may change)
1435 * and is what RtNetlink uses natively. */
1436 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1443 * netdev_release -- destroy and free a dead device.
1444 * Called when last reference to device kobject is gone.
1446 static void netdev_release(struct device *d)
1448 struct net_device *dev = to_net_dev(d);
1450 BUG_ON(dev->reg_state != NETREG_RELEASED);
1452 kfree(dev->ifalias);
1453 netdev_freemem(dev);
1456 static const void *net_namespace(struct device *d)
1458 struct net_device *dev = to_net_dev(d);
1460 return dev_net(dev);
1463 static struct class net_class = {
1465 .dev_release = netdev_release,
1466 .dev_groups = net_class_groups,
1467 .dev_uevent = netdev_uevent,
1468 .ns_type = &net_ns_type_operations,
1469 .namespace = net_namespace,
1472 #ifdef CONFIG_OF_NET
1473 static int of_dev_node_match(struct device *dev, const void *data)
1478 ret = dev->parent->of_node == data;
1480 return ret == 0 ? dev->of_node == data : ret;
1484 * of_find_net_device_by_node - lookup the net device for the device node
1485 * @np: OF device node
1487 * Looks up the net_device structure corresponding with the device node.
1488 * If successful, returns a pointer to the net_device with the embedded
1489 * struct device refcount incremented by one, or NULL on failure. The
1490 * refcount must be dropped when done with the net_device.
1492 struct net_device *of_find_net_device_by_node(struct device_node *np)
1496 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1500 return to_net_dev(dev);
1502 EXPORT_SYMBOL(of_find_net_device_by_node);
1505 /* Delete sysfs entries but hold kobject reference until after all
1506 * netdev references are gone.
1508 void netdev_unregister_kobject(struct net_device *ndev)
1510 struct device *dev = &(ndev->dev);
1512 kobject_get(&dev->kobj);
1514 remove_queue_kobjects(ndev);
1516 pm_runtime_set_memalloc_noio(dev, false);
1521 /* Create sysfs entries for network device. */
1522 int netdev_register_kobject(struct net_device *ndev)
1524 struct device *dev = &(ndev->dev);
1525 const struct attribute_group **groups = ndev->sysfs_groups;
1528 device_initialize(dev);
1529 dev->class = &net_class;
1530 dev->platform_data = ndev;
1531 dev->groups = groups;
1533 dev_set_name(dev, "%s", ndev->name);
1536 /* Allow for a device specific group */
1540 *groups++ = &netstat_group;
1542 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1543 if (ndev->ieee80211_ptr)
1544 *groups++ = &wireless_group;
1545 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1546 else if (ndev->wireless_handlers)
1547 *groups++ = &wireless_group;
1550 #endif /* CONFIG_SYSFS */
1552 error = device_add(dev);
1556 error = register_queue_kobjects(ndev);
1562 pm_runtime_set_memalloc_noio(dev, true);
1567 int netdev_class_create_file_ns(struct class_attribute *class_attr,
1570 return class_create_file_ns(&net_class, class_attr, ns);
1572 EXPORT_SYMBOL(netdev_class_create_file_ns);
1574 void netdev_class_remove_file_ns(struct class_attribute *class_attr,
1577 class_remove_file_ns(&net_class, class_attr, ns);
1579 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1581 int __init netdev_kobject_init(void)
1583 kobj_ns_type_register(&net_ns_type_operations);
1584 return class_register(&net_class);