1 // SPDX-License-Identifier: GPL-2.0-or-later
3 * net-sysfs.c - network device class and attributes
5 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
8 #include <linux/capability.h>
9 #include <linux/kernel.h>
10 #include <linux/netdevice.h>
11 #include <linux/if_arp.h>
12 #include <linux/slab.h>
13 #include <linux/sched/signal.h>
14 #include <linux/nsproxy.h>
16 #include <net/net_namespace.h>
17 #include <linux/rtnetlink.h>
18 #include <linux/vmalloc.h>
19 #include <linux/export.h>
20 #include <linux/jiffies.h>
21 #include <linux/pm_runtime.h>
23 #include <linux/of_net.h>
24 #include <linux/cpu.h>
26 #include "net-sysfs.h"
29 static const char fmt_hex[] = "%#x\n";
30 static const char fmt_dec[] = "%d\n";
31 static const char fmt_ulong[] = "%lu\n";
32 static const char fmt_u64[] = "%llu\n";
34 static inline int dev_isalive(const struct net_device *dev)
36 return dev->reg_state <= NETREG_REGISTERED;
39 /* use same locking rules as GIF* ioctl's */
40 static ssize_t netdev_show(const struct device *dev,
41 struct device_attribute *attr, char *buf,
42 ssize_t (*format)(const struct net_device *, char *))
44 struct net_device *ndev = to_net_dev(dev);
45 ssize_t ret = -EINVAL;
47 read_lock(&dev_base_lock);
48 if (dev_isalive(ndev))
49 ret = (*format)(ndev, buf);
50 read_unlock(&dev_base_lock);
55 /* generate a show function for simple field */
56 #define NETDEVICE_SHOW(field, format_string) \
57 static ssize_t format_##field(const struct net_device *dev, char *buf) \
59 return sprintf(buf, format_string, dev->field); \
61 static ssize_t field##_show(struct device *dev, \
62 struct device_attribute *attr, char *buf) \
64 return netdev_show(dev, attr, buf, format_##field); \
67 #define NETDEVICE_SHOW_RO(field, format_string) \
68 NETDEVICE_SHOW(field, format_string); \
69 static DEVICE_ATTR_RO(field)
71 #define NETDEVICE_SHOW_RW(field, format_string) \
72 NETDEVICE_SHOW(field, format_string); \
73 static DEVICE_ATTR_RW(field)
75 /* use same locking and permission rules as SIF* ioctl's */
76 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
77 const char *buf, size_t len,
78 int (*set)(struct net_device *, unsigned long))
80 struct net_device *netdev = to_net_dev(dev);
81 struct net *net = dev_net(netdev);
85 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
88 ret = kstrtoul(buf, 0, &new);
93 return restart_syscall();
95 if (dev_isalive(netdev)) {
96 ret = (*set)(netdev, new);
105 NETDEVICE_SHOW_RO(dev_id, fmt_hex);
106 NETDEVICE_SHOW_RO(dev_port, fmt_dec);
107 NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
108 NETDEVICE_SHOW_RO(addr_len, fmt_dec);
109 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
110 NETDEVICE_SHOW_RO(type, fmt_dec);
111 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
113 static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
116 struct net_device *ndev = to_net_dev(dev);
118 return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
120 static DEVICE_ATTR_RO(iflink);
122 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
124 return sprintf(buf, fmt_dec, dev->name_assign_type);
127 static ssize_t name_assign_type_show(struct device *dev,
128 struct device_attribute *attr,
131 struct net_device *ndev = to_net_dev(dev);
132 ssize_t ret = -EINVAL;
134 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
135 ret = netdev_show(dev, attr, buf, format_name_assign_type);
139 static DEVICE_ATTR_RO(name_assign_type);
141 /* use same locking rules as GIFHWADDR ioctl's */
142 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
145 struct net_device *ndev = to_net_dev(dev);
146 ssize_t ret = -EINVAL;
148 read_lock(&dev_base_lock);
149 if (dev_isalive(ndev))
150 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
151 read_unlock(&dev_base_lock);
154 static DEVICE_ATTR_RO(address);
156 static ssize_t broadcast_show(struct device *dev,
157 struct device_attribute *attr, char *buf)
159 struct net_device *ndev = to_net_dev(dev);
161 if (dev_isalive(ndev))
162 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
165 static DEVICE_ATTR_RO(broadcast);
167 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
169 if (!netif_running(dev))
171 return dev_change_carrier(dev, (bool)new_carrier);
174 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
175 const char *buf, size_t len)
177 return netdev_store(dev, attr, buf, len, change_carrier);
180 static ssize_t carrier_show(struct device *dev,
181 struct device_attribute *attr, char *buf)
183 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_link_ksettings cmd;
204 if (!__ethtool_get_link_ksettings(netdev, &cmd))
205 ret = sprintf(buf, fmt_dec, cmd.base.speed);
210 static DEVICE_ATTR_RO(speed);
212 static ssize_t duplex_show(struct device *dev,
213 struct device_attribute *attr, char *buf)
215 struct net_device *netdev = to_net_dev(dev);
219 return restart_syscall();
221 if (netif_running(netdev)) {
222 struct ethtool_link_ksettings cmd;
224 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
227 switch (cmd.base.duplex) {
238 ret = sprintf(buf, "%s\n", duplex);
244 static DEVICE_ATTR_RO(duplex);
246 static ssize_t dormant_show(struct device *dev,
247 struct device_attribute *attr, char *buf)
249 struct net_device *netdev = to_net_dev(dev);
251 if (netif_running(netdev))
252 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
256 static DEVICE_ATTR_RO(dormant);
258 static const char *const operstates[] = {
260 "notpresent", /* currently unused */
263 "testing", /* currently unused */
268 static ssize_t operstate_show(struct device *dev,
269 struct device_attribute *attr, char *buf)
271 const struct net_device *netdev = to_net_dev(dev);
272 unsigned char operstate;
274 read_lock(&dev_base_lock);
275 operstate = netdev->operstate;
276 if (!netif_running(netdev))
277 operstate = IF_OPER_DOWN;
278 read_unlock(&dev_base_lock);
280 if (operstate >= ARRAY_SIZE(operstates))
281 return -EINVAL; /* should not happen */
283 return sprintf(buf, "%s\n", operstates[operstate]);
285 static DEVICE_ATTR_RO(operstate);
287 static ssize_t carrier_changes_show(struct device *dev,
288 struct device_attribute *attr,
291 struct net_device *netdev = to_net_dev(dev);
293 return sprintf(buf, fmt_dec,
294 atomic_read(&netdev->carrier_up_count) +
295 atomic_read(&netdev->carrier_down_count));
297 static DEVICE_ATTR_RO(carrier_changes);
299 static ssize_t carrier_up_count_show(struct device *dev,
300 struct device_attribute *attr,
303 struct net_device *netdev = to_net_dev(dev);
305 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
307 static DEVICE_ATTR_RO(carrier_up_count);
309 static ssize_t carrier_down_count_show(struct device *dev,
310 struct device_attribute *attr,
313 struct net_device *netdev = to_net_dev(dev);
315 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
317 static DEVICE_ATTR_RO(carrier_down_count);
319 /* read-write attributes */
321 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
323 return dev_set_mtu(dev, (int)new_mtu);
326 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
327 const char *buf, size_t len)
329 return netdev_store(dev, attr, buf, len, change_mtu);
331 NETDEVICE_SHOW_RW(mtu, fmt_dec);
333 static int change_flags(struct net_device *dev, unsigned long new_flags)
335 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
338 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
339 const char *buf, size_t len)
341 return netdev_store(dev, attr, buf, len, change_flags);
343 NETDEVICE_SHOW_RW(flags, fmt_hex);
345 static ssize_t tx_queue_len_store(struct device *dev,
346 struct device_attribute *attr,
347 const char *buf, size_t len)
349 if (!capable(CAP_NET_ADMIN))
352 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
354 NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
356 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
358 dev->gro_flush_timeout = val;
362 static ssize_t gro_flush_timeout_store(struct device *dev,
363 struct device_attribute *attr,
364 const char *buf, size_t len)
366 if (!capable(CAP_NET_ADMIN))
369 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
371 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
373 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
374 const char *buf, size_t len)
376 struct net_device *netdev = to_net_dev(dev);
377 struct net *net = dev_net(netdev);
381 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
384 /* ignore trailing newline */
385 if (len > 0 && buf[len - 1] == '\n')
389 return restart_syscall();
391 if (dev_isalive(netdev)) {
392 ret = dev_set_alias(netdev, buf, count);
396 netdev_state_change(netdev);
404 static ssize_t ifalias_show(struct device *dev,
405 struct device_attribute *attr, char *buf)
407 const struct net_device *netdev = to_net_dev(dev);
411 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
413 ret = sprintf(buf, "%s\n", tmp);
416 static DEVICE_ATTR_RW(ifalias);
418 static int change_group(struct net_device *dev, unsigned long new_group)
420 dev_set_group(dev, (int)new_group);
424 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
425 const char *buf, size_t len)
427 return netdev_store(dev, attr, buf, len, change_group);
429 NETDEVICE_SHOW(group, fmt_dec);
430 static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
432 static int change_proto_down(struct net_device *dev, unsigned long proto_down)
434 return dev_change_proto_down(dev, (bool)proto_down);
437 static ssize_t proto_down_store(struct device *dev,
438 struct device_attribute *attr,
439 const char *buf, size_t len)
441 return netdev_store(dev, attr, buf, len, change_proto_down);
443 NETDEVICE_SHOW_RW(proto_down, fmt_dec);
445 static ssize_t phys_port_id_show(struct device *dev,
446 struct device_attribute *attr, char *buf)
448 struct net_device *netdev = to_net_dev(dev);
449 ssize_t ret = -EINVAL;
452 return restart_syscall();
454 if (dev_isalive(netdev)) {
455 struct netdev_phys_item_id ppid;
457 ret = dev_get_phys_port_id(netdev, &ppid);
459 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
465 static DEVICE_ATTR_RO(phys_port_id);
467 static ssize_t phys_port_name_show(struct device *dev,
468 struct device_attribute *attr, char *buf)
470 struct net_device *netdev = to_net_dev(dev);
471 ssize_t ret = -EINVAL;
474 return restart_syscall();
476 if (dev_isalive(netdev)) {
479 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
481 ret = sprintf(buf, "%s\n", name);
487 static DEVICE_ATTR_RO(phys_port_name);
489 static ssize_t phys_switch_id_show(struct device *dev,
490 struct device_attribute *attr, char *buf)
492 struct net_device *netdev = to_net_dev(dev);
493 ssize_t ret = -EINVAL;
496 return restart_syscall();
498 if (dev_isalive(netdev)) {
499 struct netdev_phys_item_id ppid = { };
501 ret = dev_get_port_parent_id(netdev, &ppid, false);
503 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
509 static DEVICE_ATTR_RO(phys_switch_id);
511 static struct attribute *net_class_attrs[] __ro_after_init = {
512 &dev_attr_netdev_group.attr,
514 &dev_attr_dev_id.attr,
515 &dev_attr_dev_port.attr,
516 &dev_attr_iflink.attr,
517 &dev_attr_ifindex.attr,
518 &dev_attr_name_assign_type.attr,
519 &dev_attr_addr_assign_type.attr,
520 &dev_attr_addr_len.attr,
521 &dev_attr_link_mode.attr,
522 &dev_attr_address.attr,
523 &dev_attr_broadcast.attr,
524 &dev_attr_speed.attr,
525 &dev_attr_duplex.attr,
526 &dev_attr_dormant.attr,
527 &dev_attr_operstate.attr,
528 &dev_attr_carrier_changes.attr,
529 &dev_attr_ifalias.attr,
530 &dev_attr_carrier.attr,
532 &dev_attr_flags.attr,
533 &dev_attr_tx_queue_len.attr,
534 &dev_attr_gro_flush_timeout.attr,
535 &dev_attr_phys_port_id.attr,
536 &dev_attr_phys_port_name.attr,
537 &dev_attr_phys_switch_id.attr,
538 &dev_attr_proto_down.attr,
539 &dev_attr_carrier_up_count.attr,
540 &dev_attr_carrier_down_count.attr,
543 ATTRIBUTE_GROUPS(net_class);
545 /* Show a given an attribute in the statistics group */
546 static ssize_t netstat_show(const struct device *d,
547 struct device_attribute *attr, char *buf,
548 unsigned long offset)
550 struct net_device *dev = to_net_dev(d);
551 ssize_t ret = -EINVAL;
553 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
554 offset % sizeof(u64) != 0);
556 read_lock(&dev_base_lock);
557 if (dev_isalive(dev)) {
558 struct rtnl_link_stats64 temp;
559 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
561 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
563 read_unlock(&dev_base_lock);
567 /* generate a read-only statistics attribute */
568 #define NETSTAT_ENTRY(name) \
569 static ssize_t name##_show(struct device *d, \
570 struct device_attribute *attr, char *buf) \
572 return netstat_show(d, attr, buf, \
573 offsetof(struct rtnl_link_stats64, name)); \
575 static DEVICE_ATTR_RO(name)
577 NETSTAT_ENTRY(rx_packets);
578 NETSTAT_ENTRY(tx_packets);
579 NETSTAT_ENTRY(rx_bytes);
580 NETSTAT_ENTRY(tx_bytes);
581 NETSTAT_ENTRY(rx_errors);
582 NETSTAT_ENTRY(tx_errors);
583 NETSTAT_ENTRY(rx_dropped);
584 NETSTAT_ENTRY(tx_dropped);
585 NETSTAT_ENTRY(multicast);
586 NETSTAT_ENTRY(collisions);
587 NETSTAT_ENTRY(rx_length_errors);
588 NETSTAT_ENTRY(rx_over_errors);
589 NETSTAT_ENTRY(rx_crc_errors);
590 NETSTAT_ENTRY(rx_frame_errors);
591 NETSTAT_ENTRY(rx_fifo_errors);
592 NETSTAT_ENTRY(rx_missed_errors);
593 NETSTAT_ENTRY(tx_aborted_errors);
594 NETSTAT_ENTRY(tx_carrier_errors);
595 NETSTAT_ENTRY(tx_fifo_errors);
596 NETSTAT_ENTRY(tx_heartbeat_errors);
597 NETSTAT_ENTRY(tx_window_errors);
598 NETSTAT_ENTRY(rx_compressed);
599 NETSTAT_ENTRY(tx_compressed);
600 NETSTAT_ENTRY(rx_nohandler);
602 static struct attribute *netstat_attrs[] __ro_after_init = {
603 &dev_attr_rx_packets.attr,
604 &dev_attr_tx_packets.attr,
605 &dev_attr_rx_bytes.attr,
606 &dev_attr_tx_bytes.attr,
607 &dev_attr_rx_errors.attr,
608 &dev_attr_tx_errors.attr,
609 &dev_attr_rx_dropped.attr,
610 &dev_attr_tx_dropped.attr,
611 &dev_attr_multicast.attr,
612 &dev_attr_collisions.attr,
613 &dev_attr_rx_length_errors.attr,
614 &dev_attr_rx_over_errors.attr,
615 &dev_attr_rx_crc_errors.attr,
616 &dev_attr_rx_frame_errors.attr,
617 &dev_attr_rx_fifo_errors.attr,
618 &dev_attr_rx_missed_errors.attr,
619 &dev_attr_tx_aborted_errors.attr,
620 &dev_attr_tx_carrier_errors.attr,
621 &dev_attr_tx_fifo_errors.attr,
622 &dev_attr_tx_heartbeat_errors.attr,
623 &dev_attr_tx_window_errors.attr,
624 &dev_attr_rx_compressed.attr,
625 &dev_attr_tx_compressed.attr,
626 &dev_attr_rx_nohandler.attr,
630 static const struct attribute_group netstat_group = {
631 .name = "statistics",
632 .attrs = netstat_attrs,
635 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
636 static struct attribute *wireless_attrs[] = {
640 static const struct attribute_group wireless_group = {
642 .attrs = wireless_attrs,
646 #else /* CONFIG_SYSFS */
647 #define net_class_groups NULL
648 #endif /* CONFIG_SYSFS */
651 #define to_rx_queue_attr(_attr) \
652 container_of(_attr, struct rx_queue_attribute, attr)
654 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
656 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
659 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
660 struct netdev_rx_queue *queue = to_rx_queue(kobj);
662 if (!attribute->show)
665 return attribute->show(queue, buf);
668 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
669 const char *buf, size_t count)
671 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
672 struct netdev_rx_queue *queue = to_rx_queue(kobj);
674 if (!attribute->store)
677 return attribute->store(queue, buf, count);
680 static const struct sysfs_ops rx_queue_sysfs_ops = {
681 .show = rx_queue_attr_show,
682 .store = rx_queue_attr_store,
686 static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
692 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
696 map = rcu_dereference(queue->rps_map);
698 for (i = 0; i < map->len; i++)
699 cpumask_set_cpu(map->cpus[i], mask);
701 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
703 free_cpumask_var(mask);
705 return len < PAGE_SIZE ? len : -EINVAL;
708 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
709 const char *buf, size_t len)
711 struct rps_map *old_map, *map;
714 static DEFINE_MUTEX(rps_map_mutex);
716 if (!capable(CAP_NET_ADMIN))
719 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
722 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
724 free_cpumask_var(mask);
728 map = kzalloc(max_t(unsigned int,
729 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
732 free_cpumask_var(mask);
737 for_each_cpu_and(cpu, mask, cpu_online_mask)
738 map->cpus[i++] = cpu;
747 mutex_lock(&rps_map_mutex);
748 old_map = rcu_dereference_protected(queue->rps_map,
749 mutex_is_locked(&rps_map_mutex));
750 rcu_assign_pointer(queue->rps_map, map);
753 static_branch_inc(&rps_needed);
755 static_branch_dec(&rps_needed);
757 mutex_unlock(&rps_map_mutex);
760 kfree_rcu(old_map, rcu);
762 free_cpumask_var(mask);
766 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
769 struct rps_dev_flow_table *flow_table;
770 unsigned long val = 0;
773 flow_table = rcu_dereference(queue->rps_flow_table);
775 val = (unsigned long)flow_table->mask + 1;
778 return sprintf(buf, "%lu\n", val);
781 static void rps_dev_flow_table_release(struct rcu_head *rcu)
783 struct rps_dev_flow_table *table = container_of(rcu,
784 struct rps_dev_flow_table, rcu);
788 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
789 const char *buf, size_t len)
791 unsigned long mask, count;
792 struct rps_dev_flow_table *table, *old_table;
793 static DEFINE_SPINLOCK(rps_dev_flow_lock);
796 if (!capable(CAP_NET_ADMIN))
799 rc = kstrtoul(buf, 0, &count);
805 /* mask = roundup_pow_of_two(count) - 1;
806 * without overflows...
808 while ((mask | (mask >> 1)) != mask)
810 /* On 64 bit arches, must check mask fits in table->mask (u32),
811 * and on 32bit arches, must check
812 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
814 #if BITS_PER_LONG > 32
815 if (mask > (unsigned long)(u32)mask)
818 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
819 / sizeof(struct rps_dev_flow)) {
820 /* Enforce a limit to prevent overflow */
824 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
829 for (count = 0; count <= mask; count++)
830 table->flows[count].cpu = RPS_NO_CPU;
835 spin_lock(&rps_dev_flow_lock);
836 old_table = rcu_dereference_protected(queue->rps_flow_table,
837 lockdep_is_held(&rps_dev_flow_lock));
838 rcu_assign_pointer(queue->rps_flow_table, table);
839 spin_unlock(&rps_dev_flow_lock);
842 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
847 static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
848 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
850 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
851 = __ATTR(rps_flow_cnt, 0644,
852 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
853 #endif /* CONFIG_RPS */
855 static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
857 &rps_cpus_attribute.attr,
858 &rps_dev_flow_table_cnt_attribute.attr,
862 ATTRIBUTE_GROUPS(rx_queue_default);
864 static void rx_queue_release(struct kobject *kobj)
866 struct netdev_rx_queue *queue = to_rx_queue(kobj);
869 struct rps_dev_flow_table *flow_table;
871 map = rcu_dereference_protected(queue->rps_map, 1);
873 RCU_INIT_POINTER(queue->rps_map, NULL);
877 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
879 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
880 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
884 memset(kobj, 0, sizeof(*kobj));
888 static const void *rx_queue_namespace(struct kobject *kobj)
890 struct netdev_rx_queue *queue = to_rx_queue(kobj);
891 struct device *dev = &queue->dev->dev;
892 const void *ns = NULL;
894 if (dev->class && dev->class->ns_type)
895 ns = dev->class->namespace(dev);
900 static void rx_queue_get_ownership(struct kobject *kobj,
901 kuid_t *uid, kgid_t *gid)
903 const struct net *net = rx_queue_namespace(kobj);
905 net_ns_get_ownership(net, uid, gid);
908 static struct kobj_type rx_queue_ktype __ro_after_init = {
909 .sysfs_ops = &rx_queue_sysfs_ops,
910 .release = rx_queue_release,
911 .default_groups = rx_queue_default_groups,
912 .namespace = rx_queue_namespace,
913 .get_ownership = rx_queue_get_ownership,
916 static int rx_queue_add_kobject(struct net_device *dev, int index)
918 struct netdev_rx_queue *queue = dev->_rx + index;
919 struct kobject *kobj = &queue->kobj;
922 /* Kobject_put later will trigger rx_queue_release call which
923 * decreases dev refcount: Take that reference here
925 dev_hold(queue->dev);
927 kobj->kset = dev->queues_kset;
928 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
933 if (dev->sysfs_rx_queue_group) {
934 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
939 kobject_uevent(kobj, KOBJ_ADD);
947 #endif /* CONFIG_SYSFS */
950 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
957 if (!dev->sysfs_rx_queue_group)
960 for (i = old_num; i < new_num; i++) {
961 error = rx_queue_add_kobject(dev, i);
968 while (--i >= new_num) {
969 struct kobject *kobj = &dev->_rx[i].kobj;
971 if (!refcount_read(&dev_net(dev)->count))
972 kobj->uevent_suppress = 1;
973 if (dev->sysfs_rx_queue_group)
974 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
986 * netdev_queue sysfs structures and functions.
988 struct netdev_queue_attribute {
989 struct attribute attr;
990 ssize_t (*show)(struct netdev_queue *queue, char *buf);
991 ssize_t (*store)(struct netdev_queue *queue,
992 const char *buf, size_t len);
994 #define to_netdev_queue_attr(_attr) \
995 container_of(_attr, struct netdev_queue_attribute, attr)
997 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
999 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1000 struct attribute *attr, char *buf)
1002 const struct netdev_queue_attribute *attribute
1003 = to_netdev_queue_attr(attr);
1004 struct netdev_queue *queue = to_netdev_queue(kobj);
1006 if (!attribute->show)
1009 return attribute->show(queue, buf);
1012 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1013 struct attribute *attr,
1014 const char *buf, size_t count)
1016 const struct netdev_queue_attribute *attribute
1017 = to_netdev_queue_attr(attr);
1018 struct netdev_queue *queue = to_netdev_queue(kobj);
1020 if (!attribute->store)
1023 return attribute->store(queue, buf, count);
1026 static const struct sysfs_ops netdev_queue_sysfs_ops = {
1027 .show = netdev_queue_attr_show,
1028 .store = netdev_queue_attr_store,
1031 static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1033 unsigned long trans_timeout;
1035 spin_lock_irq(&queue->_xmit_lock);
1036 trans_timeout = queue->trans_timeout;
1037 spin_unlock_irq(&queue->_xmit_lock);
1039 return sprintf(buf, "%lu", trans_timeout);
1042 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1044 struct net_device *dev = queue->dev;
1047 i = queue - dev->_tx;
1048 BUG_ON(i >= dev->num_tx_queues);
1053 static ssize_t traffic_class_show(struct netdev_queue *queue,
1056 struct net_device *dev = queue->dev;
1060 if (!netif_is_multiqueue(dev))
1063 index = get_netdev_queue_index(queue);
1065 /* If queue belongs to subordinate dev use its TC mapping */
1066 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1068 tc = netdev_txq_to_tc(dev, index);
1072 /* We can report the traffic class one of two ways:
1073 * Subordinate device traffic classes are reported with the traffic
1074 * class first, and then the subordinate class so for example TC0 on
1075 * subordinate device 2 will be reported as "0-2". If the queue
1076 * belongs to the root device it will be reported with just the
1077 * traffic class, so just "0" for TC 0 for example.
1079 return dev->num_tc < 0 ? sprintf(buf, "%u%d\n", tc, dev->num_tc) :
1080 sprintf(buf, "%u\n", tc);
1084 static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1087 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1090 static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1091 const char *buf, size_t len)
1093 struct net_device *dev = queue->dev;
1094 int err, index = get_netdev_queue_index(queue);
1097 if (!capable(CAP_NET_ADMIN))
1100 err = kstrtou32(buf, 10, &rate);
1104 if (!rtnl_trylock())
1105 return restart_syscall();
1108 if (dev->netdev_ops->ndo_set_tx_maxrate)
1109 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1113 queue->tx_maxrate = rate;
1119 static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1120 = __ATTR_RW(tx_maxrate);
1123 static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1124 = __ATTR_RO(tx_timeout);
1126 static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1127 = __ATTR_RO(traffic_class);
1131 * Byte queue limits sysfs structures and functions.
1133 static ssize_t bql_show(char *buf, unsigned int value)
1135 return sprintf(buf, "%u\n", value);
1138 static ssize_t bql_set(const char *buf, const size_t count,
1139 unsigned int *pvalue)
1144 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1145 value = DQL_MAX_LIMIT;
1147 err = kstrtouint(buf, 10, &value);
1150 if (value > DQL_MAX_LIMIT)
1159 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1162 struct dql *dql = &queue->dql;
1164 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1167 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1168 const char *buf, size_t len)
1170 struct dql *dql = &queue->dql;
1174 err = kstrtouint(buf, 10, &value);
1178 dql->slack_hold_time = msecs_to_jiffies(value);
1183 static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1184 = __ATTR(hold_time, 0644,
1185 bql_show_hold_time, bql_set_hold_time);
1187 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1190 struct dql *dql = &queue->dql;
1192 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1195 static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1196 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1198 #define BQL_ATTR(NAME, FIELD) \
1199 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1202 return bql_show(buf, queue->dql.FIELD); \
1205 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1206 const char *buf, size_t len) \
1208 return bql_set(buf, len, &queue->dql.FIELD); \
1211 static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1212 = __ATTR(NAME, 0644, \
1213 bql_show_ ## NAME, bql_set_ ## NAME)
1215 BQL_ATTR(limit, limit);
1216 BQL_ATTR(limit_max, max_limit);
1217 BQL_ATTR(limit_min, min_limit);
1219 static struct attribute *dql_attrs[] __ro_after_init = {
1220 &bql_limit_attribute.attr,
1221 &bql_limit_max_attribute.attr,
1222 &bql_limit_min_attribute.attr,
1223 &bql_hold_time_attribute.attr,
1224 &bql_inflight_attribute.attr,
1228 static const struct attribute_group dql_group = {
1229 .name = "byte_queue_limits",
1232 #endif /* CONFIG_BQL */
1235 static ssize_t xps_cpus_show(struct netdev_queue *queue,
1238 struct net_device *dev = queue->dev;
1239 int cpu, len, num_tc = 1, tc = 0;
1240 struct xps_dev_maps *dev_maps;
1242 unsigned long index;
1244 if (!netif_is_multiqueue(dev))
1247 index = get_netdev_queue_index(queue);
1250 /* Do not allow XPS on subordinate device directly */
1251 num_tc = dev->num_tc;
1255 /* If queue belongs to subordinate dev use its map */
1256 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1258 tc = netdev_txq_to_tc(dev, index);
1263 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1267 dev_maps = rcu_dereference(dev->xps_cpus_map);
1269 for_each_possible_cpu(cpu) {
1270 int i, tci = cpu * num_tc + tc;
1271 struct xps_map *map;
1273 map = rcu_dereference(dev_maps->attr_map[tci]);
1277 for (i = map->len; i--;) {
1278 if (map->queues[i] == index) {
1279 cpumask_set_cpu(cpu, mask);
1287 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
1288 free_cpumask_var(mask);
1289 return len < PAGE_SIZE ? len : -EINVAL;
1292 static ssize_t xps_cpus_store(struct netdev_queue *queue,
1293 const char *buf, size_t len)
1295 struct net_device *dev = queue->dev;
1296 unsigned long index;
1300 if (!netif_is_multiqueue(dev))
1303 if (!capable(CAP_NET_ADMIN))
1306 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1309 index = get_netdev_queue_index(queue);
1311 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1313 free_cpumask_var(mask);
1317 err = netif_set_xps_queue(dev, mask, index);
1319 free_cpumask_var(mask);
1324 static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1325 = __ATTR_RW(xps_cpus);
1327 static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1329 struct net_device *dev = queue->dev;
1330 struct xps_dev_maps *dev_maps;
1331 unsigned long *mask, index;
1332 int j, len, num_tc = 1, tc = 0;
1334 index = get_netdev_queue_index(queue);
1337 num_tc = dev->num_tc;
1338 tc = netdev_txq_to_tc(dev, index);
1342 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1347 dev_maps = rcu_dereference(dev->xps_rxqs_map);
1351 for (j = -1; j = netif_attrmask_next(j, NULL, dev->num_rx_queues),
1352 j < dev->num_rx_queues;) {
1353 int i, tci = j * num_tc + tc;
1354 struct xps_map *map;
1356 map = rcu_dereference(dev_maps->attr_map[tci]);
1360 for (i = map->len; i--;) {
1361 if (map->queues[i] == index) {
1370 len = bitmap_print_to_pagebuf(false, buf, mask, dev->num_rx_queues);
1373 return len < PAGE_SIZE ? len : -EINVAL;
1376 static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1379 struct net_device *dev = queue->dev;
1380 struct net *net = dev_net(dev);
1381 unsigned long *mask, index;
1384 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1387 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1391 index = get_netdev_queue_index(queue);
1393 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1400 err = __netif_set_xps_queue(dev, mask, index, true);
1407 static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1408 = __ATTR_RW(xps_rxqs);
1409 #endif /* CONFIG_XPS */
1411 static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1412 &queue_trans_timeout.attr,
1413 &queue_traffic_class.attr,
1415 &xps_cpus_attribute.attr,
1416 &xps_rxqs_attribute.attr,
1417 &queue_tx_maxrate.attr,
1421 ATTRIBUTE_GROUPS(netdev_queue_default);
1423 static void netdev_queue_release(struct kobject *kobj)
1425 struct netdev_queue *queue = to_netdev_queue(kobj);
1427 memset(kobj, 0, sizeof(*kobj));
1428 dev_put(queue->dev);
1431 static const void *netdev_queue_namespace(struct kobject *kobj)
1433 struct netdev_queue *queue = to_netdev_queue(kobj);
1434 struct device *dev = &queue->dev->dev;
1435 const void *ns = NULL;
1437 if (dev->class && dev->class->ns_type)
1438 ns = dev->class->namespace(dev);
1443 static void netdev_queue_get_ownership(struct kobject *kobj,
1444 kuid_t *uid, kgid_t *gid)
1446 const struct net *net = netdev_queue_namespace(kobj);
1448 net_ns_get_ownership(net, uid, gid);
1451 static struct kobj_type netdev_queue_ktype __ro_after_init = {
1452 .sysfs_ops = &netdev_queue_sysfs_ops,
1453 .release = netdev_queue_release,
1454 .default_groups = netdev_queue_default_groups,
1455 .namespace = netdev_queue_namespace,
1456 .get_ownership = netdev_queue_get_ownership,
1459 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1461 struct netdev_queue *queue = dev->_tx + index;
1462 struct kobject *kobj = &queue->kobj;
1465 /* Kobject_put later will trigger netdev_queue_release call
1466 * which decreases dev refcount: Take that reference here
1468 dev_hold(queue->dev);
1470 kobj->kset = dev->queues_kset;
1471 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1477 error = sysfs_create_group(kobj, &dql_group);
1482 kobject_uevent(kobj, KOBJ_ADD);
1489 #endif /* CONFIG_SYSFS */
1492 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1498 for (i = old_num; i < new_num; i++) {
1499 error = netdev_queue_add_kobject(dev, i);
1506 while (--i >= new_num) {
1507 struct netdev_queue *queue = dev->_tx + i;
1509 if (!refcount_read(&dev_net(dev)->count))
1510 queue->kobj.uevent_suppress = 1;
1512 sysfs_remove_group(&queue->kobj, &dql_group);
1514 kobject_put(&queue->kobj);
1520 #endif /* CONFIG_SYSFS */
1523 static int register_queue_kobjects(struct net_device *dev)
1525 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1528 dev->queues_kset = kset_create_and_add("queues",
1529 NULL, &dev->dev.kobj);
1530 if (!dev->queues_kset)
1532 real_rx = dev->real_num_rx_queues;
1534 real_tx = dev->real_num_tx_queues;
1536 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1541 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1549 netdev_queue_update_kobjects(dev, txq, 0);
1550 net_rx_queue_update_kobjects(dev, rxq, 0);
1552 kset_unregister(dev->queues_kset);
1557 static void remove_queue_kobjects(struct net_device *dev)
1559 int real_rx = 0, real_tx = 0;
1562 real_rx = dev->real_num_rx_queues;
1564 real_tx = dev->real_num_tx_queues;
1566 net_rx_queue_update_kobjects(dev, real_rx, 0);
1567 netdev_queue_update_kobjects(dev, real_tx, 0);
1569 kset_unregister(dev->queues_kset);
1573 static bool net_current_may_mount(void)
1575 struct net *net = current->nsproxy->net_ns;
1577 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1580 static void *net_grab_current_ns(void)
1582 struct net *ns = current->nsproxy->net_ns;
1583 #ifdef CONFIG_NET_NS
1585 refcount_inc(&ns->passive);
1590 static const void *net_initial_ns(void)
1595 static const void *net_netlink_ns(struct sock *sk)
1597 return sock_net(sk);
1600 const struct kobj_ns_type_operations net_ns_type_operations = {
1601 .type = KOBJ_NS_TYPE_NET,
1602 .current_may_mount = net_current_may_mount,
1603 .grab_current_ns = net_grab_current_ns,
1604 .netlink_ns = net_netlink_ns,
1605 .initial_ns = net_initial_ns,
1606 .drop_ns = net_drop_ns,
1608 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1610 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1612 struct net_device *dev = to_net_dev(d);
1615 /* pass interface to uevent. */
1616 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1620 /* pass ifindex to uevent.
1621 * ifindex is useful as it won't change (interface name may change)
1622 * and is what RtNetlink uses natively.
1624 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1631 * netdev_release -- destroy and free a dead device.
1632 * Called when last reference to device kobject is gone.
1634 static void netdev_release(struct device *d)
1636 struct net_device *dev = to_net_dev(d);
1638 BUG_ON(dev->reg_state != NETREG_RELEASED);
1640 /* no need to wait for rcu grace period:
1641 * device is dead and about to be freed.
1643 kfree(rcu_access_pointer(dev->ifalias));
1644 netdev_freemem(dev);
1647 static const void *net_namespace(struct device *d)
1649 struct net_device *dev = to_net_dev(d);
1651 return dev_net(dev);
1654 static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1656 struct net_device *dev = to_net_dev(d);
1657 const struct net *net = dev_net(dev);
1659 net_ns_get_ownership(net, uid, gid);
1662 static struct class net_class __ro_after_init = {
1664 .dev_release = netdev_release,
1665 .dev_groups = net_class_groups,
1666 .dev_uevent = netdev_uevent,
1667 .ns_type = &net_ns_type_operations,
1668 .namespace = net_namespace,
1669 .get_ownership = net_get_ownership,
1672 #ifdef CONFIG_OF_NET
1673 static int of_dev_node_match(struct device *dev, const void *data)
1678 ret = dev->parent->of_node == data;
1680 return ret == 0 ? dev->of_node == data : ret;
1684 * of_find_net_device_by_node - lookup the net device for the device node
1685 * @np: OF device node
1687 * Looks up the net_device structure corresponding with the device node.
1688 * If successful, returns a pointer to the net_device with the embedded
1689 * struct device refcount incremented by one, or NULL on failure. The
1690 * refcount must be dropped when done with the net_device.
1692 struct net_device *of_find_net_device_by_node(struct device_node *np)
1696 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1700 return to_net_dev(dev);
1702 EXPORT_SYMBOL(of_find_net_device_by_node);
1705 /* Delete sysfs entries but hold kobject reference until after all
1706 * netdev references are gone.
1708 void netdev_unregister_kobject(struct net_device *ndev)
1710 struct device *dev = &ndev->dev;
1712 if (!refcount_read(&dev_net(ndev)->count))
1713 dev_set_uevent_suppress(dev, 1);
1715 kobject_get(&dev->kobj);
1717 remove_queue_kobjects(ndev);
1719 pm_runtime_set_memalloc_noio(dev, false);
1724 /* Create sysfs entries for network device. */
1725 int netdev_register_kobject(struct net_device *ndev)
1727 struct device *dev = &ndev->dev;
1728 const struct attribute_group **groups = ndev->sysfs_groups;
1731 device_initialize(dev);
1732 dev->class = &net_class;
1733 dev->platform_data = ndev;
1734 dev->groups = groups;
1736 dev_set_name(dev, "%s", ndev->name);
1739 /* Allow for a device specific group */
1743 *groups++ = &netstat_group;
1745 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1746 if (ndev->ieee80211_ptr)
1747 *groups++ = &wireless_group;
1748 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1749 else if (ndev->wireless_handlers)
1750 *groups++ = &wireless_group;
1753 #endif /* CONFIG_SYSFS */
1755 error = device_add(dev);
1759 error = register_queue_kobjects(ndev);
1765 pm_runtime_set_memalloc_noio(dev, true);
1770 int netdev_class_create_file_ns(const struct class_attribute *class_attr,
1773 return class_create_file_ns(&net_class, class_attr, ns);
1775 EXPORT_SYMBOL(netdev_class_create_file_ns);
1777 void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
1780 class_remove_file_ns(&net_class, class_attr, ns);
1782 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1784 int __init netdev_kobject_init(void)
1786 kobj_ns_type_register(&net_ns_type_operations);
1787 return class_register(&net_class);