mm/zsmalloc: allocate exactly size of struct zs_pool
[linux/fpc-iii.git] / net / core / net-sysfs.c
blob999341244434309b6275fe6230a9f1bf7c57d2b2
1 /*
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>
19 #include <net/sock.h>
20 #include <net/net_namespace.h>
21 #include <linux/rtnetlink.h>
22 #include <linux/vmalloc.h>
23 #include <linux/export.h>
24 #include <linux/jiffies.h>
25 #include <linux/pm_runtime.h>
27 #include "net-sysfs.h"
29 #ifdef CONFIG_SYSFS
30 static const char fmt_hex[] = "%#x\n";
31 static const char fmt_long_hex[] = "%#lx\n";
32 static const char fmt_dec[] = "%d\n";
33 static const char fmt_udec[] = "%u\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);
55 return ret;
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) \
61 { \
62 return sprintf(buf, format_string, dev->field); \
63 } \
64 static ssize_t field##_show(struct device *dev, \
65 struct device_attribute *attr, char *buf) \
66 { \
67 return netdev_show(dev, attr, buf, format_##field); \
68 } \
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);
85 unsigned long new;
86 int ret = -EINVAL;
88 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
89 return -EPERM;
91 ret = kstrtoul(buf, 0, &new);
92 if (ret)
93 goto err;
95 if (!rtnl_trylock())
96 return restart_syscall();
98 if (dev_isalive(netdev)) {
99 if ((ret = (*set)(netdev, new)) == 0)
100 ret = len;
102 rtnl_unlock();
103 err:
104 return ret;
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(iflink, fmt_dec);
112 NETDEVICE_SHOW_RO(ifindex, fmt_dec);
113 NETDEVICE_SHOW_RO(type, fmt_dec);
114 NETDEVICE_SHOW_RO(link_mode, fmt_dec);
116 static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
118 return sprintf(buf, fmt_dec, dev->name_assign_type);
121 static ssize_t name_assign_type_show(struct device *dev,
122 struct device_attribute *attr,
123 char *buf)
125 struct net_device *ndev = to_net_dev(dev);
126 ssize_t ret = -EINVAL;
128 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
129 ret = netdev_show(dev, attr, buf, format_name_assign_type);
131 return ret;
133 static DEVICE_ATTR_RO(name_assign_type);
135 /* use same locking rules as GIFHWADDR ioctl's */
136 static ssize_t address_show(struct device *dev, struct device_attribute *attr,
137 char *buf)
139 struct net_device *ndev = to_net_dev(dev);
140 ssize_t ret = -EINVAL;
142 read_lock(&dev_base_lock);
143 if (dev_isalive(ndev))
144 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
145 read_unlock(&dev_base_lock);
146 return ret;
148 static DEVICE_ATTR_RO(address);
150 static ssize_t broadcast_show(struct device *dev,
151 struct device_attribute *attr, char *buf)
153 struct net_device *ndev = to_net_dev(dev);
154 if (dev_isalive(ndev))
155 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
156 return -EINVAL;
158 static DEVICE_ATTR_RO(broadcast);
160 static int change_carrier(struct net_device *dev, unsigned long new_carrier)
162 if (!netif_running(dev))
163 return -EINVAL;
164 return dev_change_carrier(dev, (bool) new_carrier);
167 static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
168 const char *buf, size_t len)
170 return netdev_store(dev, attr, buf, len, change_carrier);
173 static ssize_t carrier_show(struct device *dev,
174 struct device_attribute *attr, char *buf)
176 struct net_device *netdev = to_net_dev(dev);
177 if (netif_running(netdev)) {
178 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
180 return -EINVAL;
182 static DEVICE_ATTR_RW(carrier);
184 static ssize_t speed_show(struct device *dev,
185 struct device_attribute *attr, char *buf)
187 struct net_device *netdev = to_net_dev(dev);
188 int ret = -EINVAL;
190 if (!rtnl_trylock())
191 return restart_syscall();
193 if (netif_running(netdev)) {
194 struct ethtool_cmd cmd;
195 if (!__ethtool_get_settings(netdev, &cmd))
196 ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
198 rtnl_unlock();
199 return ret;
201 static DEVICE_ATTR_RO(speed);
203 static ssize_t duplex_show(struct device *dev,
204 struct device_attribute *attr, char *buf)
206 struct net_device *netdev = to_net_dev(dev);
207 int ret = -EINVAL;
209 if (!rtnl_trylock())
210 return restart_syscall();
212 if (netif_running(netdev)) {
213 struct ethtool_cmd cmd;
214 if (!__ethtool_get_settings(netdev, &cmd)) {
215 const char *duplex;
216 switch (cmd.duplex) {
217 case DUPLEX_HALF:
218 duplex = "half";
219 break;
220 case DUPLEX_FULL:
221 duplex = "full";
222 break;
223 default:
224 duplex = "unknown";
225 break;
227 ret = sprintf(buf, "%s\n", duplex);
230 rtnl_unlock();
231 return ret;
233 static DEVICE_ATTR_RO(duplex);
235 static ssize_t dormant_show(struct device *dev,
236 struct device_attribute *attr, char *buf)
238 struct net_device *netdev = to_net_dev(dev);
240 if (netif_running(netdev))
241 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
243 return -EINVAL;
245 static DEVICE_ATTR_RO(dormant);
247 static const char *const operstates[] = {
248 "unknown",
249 "notpresent", /* currently unused */
250 "down",
251 "lowerlayerdown",
252 "testing", /* currently unused */
253 "dormant",
254 "up"
257 static ssize_t operstate_show(struct device *dev,
258 struct device_attribute *attr, char *buf)
260 const struct net_device *netdev = to_net_dev(dev);
261 unsigned char operstate;
263 read_lock(&dev_base_lock);
264 operstate = netdev->operstate;
265 if (!netif_running(netdev))
266 operstate = IF_OPER_DOWN;
267 read_unlock(&dev_base_lock);
269 if (operstate >= ARRAY_SIZE(operstates))
270 return -EINVAL; /* should not happen */
272 return sprintf(buf, "%s\n", operstates[operstate]);
274 static DEVICE_ATTR_RO(operstate);
276 static ssize_t carrier_changes_show(struct device *dev,
277 struct device_attribute *attr,
278 char *buf)
280 struct net_device *netdev = to_net_dev(dev);
281 return sprintf(buf, fmt_dec,
282 atomic_read(&netdev->carrier_changes));
284 static DEVICE_ATTR_RO(carrier_changes);
286 /* read-write attributes */
288 static int change_mtu(struct net_device *dev, unsigned long new_mtu)
290 return dev_set_mtu(dev, (int) new_mtu);
293 static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
294 const char *buf, size_t len)
296 return netdev_store(dev, attr, buf, len, change_mtu);
298 NETDEVICE_SHOW_RW(mtu, fmt_dec);
300 static int change_flags(struct net_device *dev, unsigned long new_flags)
302 return dev_change_flags(dev, (unsigned int) new_flags);
305 static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
306 const char *buf, size_t len)
308 return netdev_store(dev, attr, buf, len, change_flags);
310 NETDEVICE_SHOW_RW(flags, fmt_hex);
312 static int change_tx_queue_len(struct net_device *dev, unsigned long new_len)
314 dev->tx_queue_len = new_len;
315 return 0;
318 static ssize_t tx_queue_len_store(struct device *dev,
319 struct device_attribute *attr,
320 const char *buf, size_t len)
322 if (!capable(CAP_NET_ADMIN))
323 return -EPERM;
325 return netdev_store(dev, attr, buf, len, change_tx_queue_len);
327 NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
329 static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
331 dev->gro_flush_timeout = val;
332 return 0;
335 static ssize_t gro_flush_timeout_store(struct device *dev,
336 struct device_attribute *attr,
337 const char *buf, size_t len)
339 if (!capable(CAP_NET_ADMIN))
340 return -EPERM;
342 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
344 NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
346 static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
347 const char *buf, size_t len)
349 struct net_device *netdev = to_net_dev(dev);
350 struct net *net = dev_net(netdev);
351 size_t count = len;
352 ssize_t ret;
354 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
355 return -EPERM;
357 /* ignore trailing newline */
358 if (len > 0 && buf[len - 1] == '\n')
359 --count;
361 if (!rtnl_trylock())
362 return restart_syscall();
363 ret = dev_set_alias(netdev, buf, count);
364 rtnl_unlock();
366 return ret < 0 ? ret : len;
369 static ssize_t ifalias_show(struct device *dev,
370 struct device_attribute *attr, char *buf)
372 const struct net_device *netdev = to_net_dev(dev);
373 ssize_t ret = 0;
375 if (!rtnl_trylock())
376 return restart_syscall();
377 if (netdev->ifalias)
378 ret = sprintf(buf, "%s\n", netdev->ifalias);
379 rtnl_unlock();
380 return ret;
382 static DEVICE_ATTR_RW(ifalias);
384 static int change_group(struct net_device *dev, unsigned long new_group)
386 dev_set_group(dev, (int) new_group);
387 return 0;
390 static ssize_t group_store(struct device *dev, struct device_attribute *attr,
391 const char *buf, size_t len)
393 return netdev_store(dev, attr, buf, len, change_group);
395 NETDEVICE_SHOW(group, fmt_dec);
396 static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
398 static ssize_t phys_port_id_show(struct device *dev,
399 struct device_attribute *attr, char *buf)
401 struct net_device *netdev = to_net_dev(dev);
402 ssize_t ret = -EINVAL;
404 if (!rtnl_trylock())
405 return restart_syscall();
407 if (dev_isalive(netdev)) {
408 struct netdev_phys_item_id ppid;
410 ret = dev_get_phys_port_id(netdev, &ppid);
411 if (!ret)
412 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
414 rtnl_unlock();
416 return ret;
418 static DEVICE_ATTR_RO(phys_port_id);
420 static ssize_t phys_switch_id_show(struct device *dev,
421 struct device_attribute *attr, char *buf)
423 struct net_device *netdev = to_net_dev(dev);
424 ssize_t ret = -EINVAL;
426 if (!rtnl_trylock())
427 return restart_syscall();
429 if (dev_isalive(netdev)) {
430 struct netdev_phys_item_id ppid;
432 ret = netdev_switch_parent_id_get(netdev, &ppid);
433 if (!ret)
434 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
436 rtnl_unlock();
438 return ret;
440 static DEVICE_ATTR_RO(phys_switch_id);
442 static struct attribute *net_class_attrs[] = {
443 &dev_attr_netdev_group.attr,
444 &dev_attr_type.attr,
445 &dev_attr_dev_id.attr,
446 &dev_attr_dev_port.attr,
447 &dev_attr_iflink.attr,
448 &dev_attr_ifindex.attr,
449 &dev_attr_name_assign_type.attr,
450 &dev_attr_addr_assign_type.attr,
451 &dev_attr_addr_len.attr,
452 &dev_attr_link_mode.attr,
453 &dev_attr_address.attr,
454 &dev_attr_broadcast.attr,
455 &dev_attr_speed.attr,
456 &dev_attr_duplex.attr,
457 &dev_attr_dormant.attr,
458 &dev_attr_operstate.attr,
459 &dev_attr_carrier_changes.attr,
460 &dev_attr_ifalias.attr,
461 &dev_attr_carrier.attr,
462 &dev_attr_mtu.attr,
463 &dev_attr_flags.attr,
464 &dev_attr_tx_queue_len.attr,
465 &dev_attr_gro_flush_timeout.attr,
466 &dev_attr_phys_port_id.attr,
467 &dev_attr_phys_switch_id.attr,
468 NULL,
470 ATTRIBUTE_GROUPS(net_class);
472 /* Show a given an attribute in the statistics group */
473 static ssize_t netstat_show(const struct device *d,
474 struct device_attribute *attr, char *buf,
475 unsigned long offset)
477 struct net_device *dev = to_net_dev(d);
478 ssize_t ret = -EINVAL;
480 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
481 offset % sizeof(u64) != 0);
483 read_lock(&dev_base_lock);
484 if (dev_isalive(dev)) {
485 struct rtnl_link_stats64 temp;
486 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
488 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
490 read_unlock(&dev_base_lock);
491 return ret;
494 /* generate a read-only statistics attribute */
495 #define NETSTAT_ENTRY(name) \
496 static ssize_t name##_show(struct device *d, \
497 struct device_attribute *attr, char *buf) \
499 return netstat_show(d, attr, buf, \
500 offsetof(struct rtnl_link_stats64, name)); \
502 static DEVICE_ATTR_RO(name)
504 NETSTAT_ENTRY(rx_packets);
505 NETSTAT_ENTRY(tx_packets);
506 NETSTAT_ENTRY(rx_bytes);
507 NETSTAT_ENTRY(tx_bytes);
508 NETSTAT_ENTRY(rx_errors);
509 NETSTAT_ENTRY(tx_errors);
510 NETSTAT_ENTRY(rx_dropped);
511 NETSTAT_ENTRY(tx_dropped);
512 NETSTAT_ENTRY(multicast);
513 NETSTAT_ENTRY(collisions);
514 NETSTAT_ENTRY(rx_length_errors);
515 NETSTAT_ENTRY(rx_over_errors);
516 NETSTAT_ENTRY(rx_crc_errors);
517 NETSTAT_ENTRY(rx_frame_errors);
518 NETSTAT_ENTRY(rx_fifo_errors);
519 NETSTAT_ENTRY(rx_missed_errors);
520 NETSTAT_ENTRY(tx_aborted_errors);
521 NETSTAT_ENTRY(tx_carrier_errors);
522 NETSTAT_ENTRY(tx_fifo_errors);
523 NETSTAT_ENTRY(tx_heartbeat_errors);
524 NETSTAT_ENTRY(tx_window_errors);
525 NETSTAT_ENTRY(rx_compressed);
526 NETSTAT_ENTRY(tx_compressed);
528 static struct attribute *netstat_attrs[] = {
529 &dev_attr_rx_packets.attr,
530 &dev_attr_tx_packets.attr,
531 &dev_attr_rx_bytes.attr,
532 &dev_attr_tx_bytes.attr,
533 &dev_attr_rx_errors.attr,
534 &dev_attr_tx_errors.attr,
535 &dev_attr_rx_dropped.attr,
536 &dev_attr_tx_dropped.attr,
537 &dev_attr_multicast.attr,
538 &dev_attr_collisions.attr,
539 &dev_attr_rx_length_errors.attr,
540 &dev_attr_rx_over_errors.attr,
541 &dev_attr_rx_crc_errors.attr,
542 &dev_attr_rx_frame_errors.attr,
543 &dev_attr_rx_fifo_errors.attr,
544 &dev_attr_rx_missed_errors.attr,
545 &dev_attr_tx_aborted_errors.attr,
546 &dev_attr_tx_carrier_errors.attr,
547 &dev_attr_tx_fifo_errors.attr,
548 &dev_attr_tx_heartbeat_errors.attr,
549 &dev_attr_tx_window_errors.attr,
550 &dev_attr_rx_compressed.attr,
551 &dev_attr_tx_compressed.attr,
552 NULL
556 static struct attribute_group netstat_group = {
557 .name = "statistics",
558 .attrs = netstat_attrs,
561 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
562 static struct attribute *wireless_attrs[] = {
563 NULL
566 static struct attribute_group wireless_group = {
567 .name = "wireless",
568 .attrs = wireless_attrs,
570 #endif
572 #else /* CONFIG_SYSFS */
573 #define net_class_groups NULL
574 #endif /* CONFIG_SYSFS */
576 #ifdef CONFIG_SYSFS
577 #define to_rx_queue_attr(_attr) container_of(_attr, \
578 struct rx_queue_attribute, attr)
580 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
582 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
583 char *buf)
585 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
586 struct netdev_rx_queue *queue = to_rx_queue(kobj);
588 if (!attribute->show)
589 return -EIO;
591 return attribute->show(queue, attribute, buf);
594 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
595 const char *buf, size_t count)
597 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
598 struct netdev_rx_queue *queue = to_rx_queue(kobj);
600 if (!attribute->store)
601 return -EIO;
603 return attribute->store(queue, attribute, buf, count);
606 static const struct sysfs_ops rx_queue_sysfs_ops = {
607 .show = rx_queue_attr_show,
608 .store = rx_queue_attr_store,
611 #ifdef CONFIG_RPS
612 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
613 struct rx_queue_attribute *attribute, char *buf)
615 struct rps_map *map;
616 cpumask_var_t mask;
617 size_t len = 0;
618 int i;
620 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
621 return -ENOMEM;
623 rcu_read_lock();
624 map = rcu_dereference(queue->rps_map);
625 if (map)
626 for (i = 0; i < map->len; i++)
627 cpumask_set_cpu(map->cpus[i], mask);
629 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
630 if (PAGE_SIZE - len < 3) {
631 rcu_read_unlock();
632 free_cpumask_var(mask);
633 return -EINVAL;
635 rcu_read_unlock();
637 free_cpumask_var(mask);
638 len += sprintf(buf + len, "\n");
639 return len;
642 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
643 struct rx_queue_attribute *attribute,
644 const char *buf, size_t len)
646 struct rps_map *old_map, *map;
647 cpumask_var_t mask;
648 int err, cpu, i;
649 static DEFINE_SPINLOCK(rps_map_lock);
651 if (!capable(CAP_NET_ADMIN))
652 return -EPERM;
654 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
655 return -ENOMEM;
657 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
658 if (err) {
659 free_cpumask_var(mask);
660 return err;
663 map = kzalloc(max_t(unsigned int,
664 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
665 GFP_KERNEL);
666 if (!map) {
667 free_cpumask_var(mask);
668 return -ENOMEM;
671 i = 0;
672 for_each_cpu_and(cpu, mask, cpu_online_mask)
673 map->cpus[i++] = cpu;
675 if (i)
676 map->len = i;
677 else {
678 kfree(map);
679 map = NULL;
682 spin_lock(&rps_map_lock);
683 old_map = rcu_dereference_protected(queue->rps_map,
684 lockdep_is_held(&rps_map_lock));
685 rcu_assign_pointer(queue->rps_map, map);
686 spin_unlock(&rps_map_lock);
688 if (map)
689 static_key_slow_inc(&rps_needed);
690 if (old_map) {
691 kfree_rcu(old_map, rcu);
692 static_key_slow_dec(&rps_needed);
694 free_cpumask_var(mask);
695 return len;
698 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
699 struct rx_queue_attribute *attr,
700 char *buf)
702 struct rps_dev_flow_table *flow_table;
703 unsigned long val = 0;
705 rcu_read_lock();
706 flow_table = rcu_dereference(queue->rps_flow_table);
707 if (flow_table)
708 val = (unsigned long)flow_table->mask + 1;
709 rcu_read_unlock();
711 return sprintf(buf, "%lu\n", val);
714 static void rps_dev_flow_table_release(struct rcu_head *rcu)
716 struct rps_dev_flow_table *table = container_of(rcu,
717 struct rps_dev_flow_table, rcu);
718 vfree(table);
721 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
722 struct rx_queue_attribute *attr,
723 const char *buf, size_t len)
725 unsigned long mask, count;
726 struct rps_dev_flow_table *table, *old_table;
727 static DEFINE_SPINLOCK(rps_dev_flow_lock);
728 int rc;
730 if (!capable(CAP_NET_ADMIN))
731 return -EPERM;
733 rc = kstrtoul(buf, 0, &count);
734 if (rc < 0)
735 return rc;
737 if (count) {
738 mask = count - 1;
739 /* mask = roundup_pow_of_two(count) - 1;
740 * without overflows...
742 while ((mask | (mask >> 1)) != mask)
743 mask |= (mask >> 1);
744 /* On 64 bit arches, must check mask fits in table->mask (u32),
745 * and on 32bit arches, must check
746 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
748 #if BITS_PER_LONG > 32
749 if (mask > (unsigned long)(u32)mask)
750 return -EINVAL;
751 #else
752 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
753 / sizeof(struct rps_dev_flow)) {
754 /* Enforce a limit to prevent overflow */
755 return -EINVAL;
757 #endif
758 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
759 if (!table)
760 return -ENOMEM;
762 table->mask = mask;
763 for (count = 0; count <= mask; count++)
764 table->flows[count].cpu = RPS_NO_CPU;
765 } else
766 table = NULL;
768 spin_lock(&rps_dev_flow_lock);
769 old_table = rcu_dereference_protected(queue->rps_flow_table,
770 lockdep_is_held(&rps_dev_flow_lock));
771 rcu_assign_pointer(queue->rps_flow_table, table);
772 spin_unlock(&rps_dev_flow_lock);
774 if (old_table)
775 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
777 return len;
780 static struct rx_queue_attribute rps_cpus_attribute =
781 __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
784 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
785 __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
786 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
787 #endif /* CONFIG_RPS */
789 static struct attribute *rx_queue_default_attrs[] = {
790 #ifdef CONFIG_RPS
791 &rps_cpus_attribute.attr,
792 &rps_dev_flow_table_cnt_attribute.attr,
793 #endif
794 NULL
797 static void rx_queue_release(struct kobject *kobj)
799 struct netdev_rx_queue *queue = to_rx_queue(kobj);
800 #ifdef CONFIG_RPS
801 struct rps_map *map;
802 struct rps_dev_flow_table *flow_table;
805 map = rcu_dereference_protected(queue->rps_map, 1);
806 if (map) {
807 RCU_INIT_POINTER(queue->rps_map, NULL);
808 kfree_rcu(map, rcu);
811 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
812 if (flow_table) {
813 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
814 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
816 #endif
818 memset(kobj, 0, sizeof(*kobj));
819 dev_put(queue->dev);
822 static const void *rx_queue_namespace(struct kobject *kobj)
824 struct netdev_rx_queue *queue = to_rx_queue(kobj);
825 struct device *dev = &queue->dev->dev;
826 const void *ns = NULL;
828 if (dev->class && dev->class->ns_type)
829 ns = dev->class->namespace(dev);
831 return ns;
834 static struct kobj_type rx_queue_ktype = {
835 .sysfs_ops = &rx_queue_sysfs_ops,
836 .release = rx_queue_release,
837 .default_attrs = rx_queue_default_attrs,
838 .namespace = rx_queue_namespace
841 static int rx_queue_add_kobject(struct net_device *dev, int index)
843 struct netdev_rx_queue *queue = dev->_rx + index;
844 struct kobject *kobj = &queue->kobj;
845 int error = 0;
847 kobj->kset = dev->queues_kset;
848 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
849 "rx-%u", index);
850 if (error)
851 goto exit;
853 if (dev->sysfs_rx_queue_group) {
854 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
855 if (error)
856 goto exit;
859 kobject_uevent(kobj, KOBJ_ADD);
860 dev_hold(queue->dev);
862 return error;
863 exit:
864 kobject_put(kobj);
865 return error;
867 #endif /* CONFIG_SYSFS */
870 net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
872 #ifdef CONFIG_SYSFS
873 int i;
874 int error = 0;
876 #ifndef CONFIG_RPS
877 if (!dev->sysfs_rx_queue_group)
878 return 0;
879 #endif
880 for (i = old_num; i < new_num; i++) {
881 error = rx_queue_add_kobject(dev, i);
882 if (error) {
883 new_num = old_num;
884 break;
888 while (--i >= new_num) {
889 if (dev->sysfs_rx_queue_group)
890 sysfs_remove_group(&dev->_rx[i].kobj,
891 dev->sysfs_rx_queue_group);
892 kobject_put(&dev->_rx[i].kobj);
895 return error;
896 #else
897 return 0;
898 #endif
901 #ifdef CONFIG_SYSFS
903 * netdev_queue sysfs structures and functions.
905 struct netdev_queue_attribute {
906 struct attribute attr;
907 ssize_t (*show)(struct netdev_queue *queue,
908 struct netdev_queue_attribute *attr, char *buf);
909 ssize_t (*store)(struct netdev_queue *queue,
910 struct netdev_queue_attribute *attr, const char *buf, size_t len);
912 #define to_netdev_queue_attr(_attr) container_of(_attr, \
913 struct netdev_queue_attribute, attr)
915 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
917 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
918 struct attribute *attr, char *buf)
920 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
921 struct netdev_queue *queue = to_netdev_queue(kobj);
923 if (!attribute->show)
924 return -EIO;
926 return attribute->show(queue, attribute, buf);
929 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
930 struct attribute *attr,
931 const char *buf, size_t count)
933 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
934 struct netdev_queue *queue = to_netdev_queue(kobj);
936 if (!attribute->store)
937 return -EIO;
939 return attribute->store(queue, attribute, buf, count);
942 static const struct sysfs_ops netdev_queue_sysfs_ops = {
943 .show = netdev_queue_attr_show,
944 .store = netdev_queue_attr_store,
947 static ssize_t show_trans_timeout(struct netdev_queue *queue,
948 struct netdev_queue_attribute *attribute,
949 char *buf)
951 unsigned long trans_timeout;
953 spin_lock_irq(&queue->_xmit_lock);
954 trans_timeout = queue->trans_timeout;
955 spin_unlock_irq(&queue->_xmit_lock);
957 return sprintf(buf, "%lu", trans_timeout);
960 static struct netdev_queue_attribute queue_trans_timeout =
961 __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
963 #ifdef CONFIG_BQL
965 * Byte queue limits sysfs structures and functions.
967 static ssize_t bql_show(char *buf, unsigned int value)
969 return sprintf(buf, "%u\n", value);
972 static ssize_t bql_set(const char *buf, const size_t count,
973 unsigned int *pvalue)
975 unsigned int value;
976 int err;
978 if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
979 value = DQL_MAX_LIMIT;
980 else {
981 err = kstrtouint(buf, 10, &value);
982 if (err < 0)
983 return err;
984 if (value > DQL_MAX_LIMIT)
985 return -EINVAL;
988 *pvalue = value;
990 return count;
993 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
994 struct netdev_queue_attribute *attr,
995 char *buf)
997 struct dql *dql = &queue->dql;
999 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1002 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1003 struct netdev_queue_attribute *attribute,
1004 const char *buf, size_t len)
1006 struct dql *dql = &queue->dql;
1007 unsigned int value;
1008 int err;
1010 err = kstrtouint(buf, 10, &value);
1011 if (err < 0)
1012 return err;
1014 dql->slack_hold_time = msecs_to_jiffies(value);
1016 return len;
1019 static struct netdev_queue_attribute bql_hold_time_attribute =
1020 __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
1021 bql_set_hold_time);
1023 static ssize_t bql_show_inflight(struct netdev_queue *queue,
1024 struct netdev_queue_attribute *attr,
1025 char *buf)
1027 struct dql *dql = &queue->dql;
1029 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1032 static struct netdev_queue_attribute bql_inflight_attribute =
1033 __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
1035 #define BQL_ATTR(NAME, FIELD) \
1036 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1037 struct netdev_queue_attribute *attr, \
1038 char *buf) \
1040 return bql_show(buf, queue->dql.FIELD); \
1043 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1044 struct netdev_queue_attribute *attr, \
1045 const char *buf, size_t len) \
1047 return bql_set(buf, len, &queue->dql.FIELD); \
1050 static struct netdev_queue_attribute bql_ ## NAME ## _attribute = \
1051 __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME, \
1052 bql_set_ ## NAME);
1054 BQL_ATTR(limit, limit)
1055 BQL_ATTR(limit_max, max_limit)
1056 BQL_ATTR(limit_min, min_limit)
1058 static struct attribute *dql_attrs[] = {
1059 &bql_limit_attribute.attr,
1060 &bql_limit_max_attribute.attr,
1061 &bql_limit_min_attribute.attr,
1062 &bql_hold_time_attribute.attr,
1063 &bql_inflight_attribute.attr,
1064 NULL
1067 static struct attribute_group dql_group = {
1068 .name = "byte_queue_limits",
1069 .attrs = dql_attrs,
1071 #endif /* CONFIG_BQL */
1073 #ifdef CONFIG_XPS
1074 static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1076 struct net_device *dev = queue->dev;
1077 unsigned int i;
1079 i = queue - dev->_tx;
1080 BUG_ON(i >= dev->num_tx_queues);
1082 return i;
1086 static ssize_t show_xps_map(struct netdev_queue *queue,
1087 struct netdev_queue_attribute *attribute, char *buf)
1089 struct net_device *dev = queue->dev;
1090 struct xps_dev_maps *dev_maps;
1091 cpumask_var_t mask;
1092 unsigned long index;
1093 size_t len = 0;
1094 int i;
1096 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1097 return -ENOMEM;
1099 index = get_netdev_queue_index(queue);
1101 rcu_read_lock();
1102 dev_maps = rcu_dereference(dev->xps_maps);
1103 if (dev_maps) {
1104 for_each_possible_cpu(i) {
1105 struct xps_map *map =
1106 rcu_dereference(dev_maps->cpu_map[i]);
1107 if (map) {
1108 int j;
1109 for (j = 0; j < map->len; j++) {
1110 if (map->queues[j] == index) {
1111 cpumask_set_cpu(i, mask);
1112 break;
1118 rcu_read_unlock();
1120 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
1121 if (PAGE_SIZE - len < 3) {
1122 free_cpumask_var(mask);
1123 return -EINVAL;
1126 free_cpumask_var(mask);
1127 len += sprintf(buf + len, "\n");
1128 return len;
1131 static ssize_t store_xps_map(struct netdev_queue *queue,
1132 struct netdev_queue_attribute *attribute,
1133 const char *buf, size_t len)
1135 struct net_device *dev = queue->dev;
1136 unsigned long index;
1137 cpumask_var_t mask;
1138 int err;
1140 if (!capable(CAP_NET_ADMIN))
1141 return -EPERM;
1143 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1144 return -ENOMEM;
1146 index = get_netdev_queue_index(queue);
1148 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1149 if (err) {
1150 free_cpumask_var(mask);
1151 return err;
1154 err = netif_set_xps_queue(dev, mask, index);
1156 free_cpumask_var(mask);
1158 return err ? : len;
1161 static struct netdev_queue_attribute xps_cpus_attribute =
1162 __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1163 #endif /* CONFIG_XPS */
1165 static struct attribute *netdev_queue_default_attrs[] = {
1166 &queue_trans_timeout.attr,
1167 #ifdef CONFIG_XPS
1168 &xps_cpus_attribute.attr,
1169 #endif
1170 NULL
1173 static void netdev_queue_release(struct kobject *kobj)
1175 struct netdev_queue *queue = to_netdev_queue(kobj);
1177 memset(kobj, 0, sizeof(*kobj));
1178 dev_put(queue->dev);
1181 static const void *netdev_queue_namespace(struct kobject *kobj)
1183 struct netdev_queue *queue = to_netdev_queue(kobj);
1184 struct device *dev = &queue->dev->dev;
1185 const void *ns = NULL;
1187 if (dev->class && dev->class->ns_type)
1188 ns = dev->class->namespace(dev);
1190 return ns;
1193 static struct kobj_type netdev_queue_ktype = {
1194 .sysfs_ops = &netdev_queue_sysfs_ops,
1195 .release = netdev_queue_release,
1196 .default_attrs = netdev_queue_default_attrs,
1197 .namespace = netdev_queue_namespace,
1200 static int netdev_queue_add_kobject(struct net_device *dev, int index)
1202 struct netdev_queue *queue = dev->_tx + index;
1203 struct kobject *kobj = &queue->kobj;
1204 int error = 0;
1206 kobj->kset = dev->queues_kset;
1207 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1208 "tx-%u", index);
1209 if (error)
1210 goto exit;
1212 #ifdef CONFIG_BQL
1213 error = sysfs_create_group(kobj, &dql_group);
1214 if (error)
1215 goto exit;
1216 #endif
1218 kobject_uevent(kobj, KOBJ_ADD);
1219 dev_hold(queue->dev);
1221 return 0;
1222 exit:
1223 kobject_put(kobj);
1224 return error;
1226 #endif /* CONFIG_SYSFS */
1229 netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1231 #ifdef CONFIG_SYSFS
1232 int i;
1233 int error = 0;
1235 for (i = old_num; i < new_num; i++) {
1236 error = netdev_queue_add_kobject(dev, i);
1237 if (error) {
1238 new_num = old_num;
1239 break;
1243 while (--i >= new_num) {
1244 struct netdev_queue *queue = dev->_tx + i;
1246 #ifdef CONFIG_BQL
1247 sysfs_remove_group(&queue->kobj, &dql_group);
1248 #endif
1249 kobject_put(&queue->kobj);
1252 return error;
1253 #else
1254 return 0;
1255 #endif /* CONFIG_SYSFS */
1258 static int register_queue_kobjects(struct net_device *dev)
1260 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1262 #ifdef CONFIG_SYSFS
1263 dev->queues_kset = kset_create_and_add("queues",
1264 NULL, &dev->dev.kobj);
1265 if (!dev->queues_kset)
1266 return -ENOMEM;
1267 real_rx = dev->real_num_rx_queues;
1268 #endif
1269 real_tx = dev->real_num_tx_queues;
1271 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1272 if (error)
1273 goto error;
1274 rxq = real_rx;
1276 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1277 if (error)
1278 goto error;
1279 txq = real_tx;
1281 return 0;
1283 error:
1284 netdev_queue_update_kobjects(dev, txq, 0);
1285 net_rx_queue_update_kobjects(dev, rxq, 0);
1286 return error;
1289 static void remove_queue_kobjects(struct net_device *dev)
1291 int real_rx = 0, real_tx = 0;
1293 #ifdef CONFIG_SYSFS
1294 real_rx = dev->real_num_rx_queues;
1295 #endif
1296 real_tx = dev->real_num_tx_queues;
1298 net_rx_queue_update_kobjects(dev, real_rx, 0);
1299 netdev_queue_update_kobjects(dev, real_tx, 0);
1300 #ifdef CONFIG_SYSFS
1301 kset_unregister(dev->queues_kset);
1302 #endif
1305 static bool net_current_may_mount(void)
1307 struct net *net = current->nsproxy->net_ns;
1309 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1312 static void *net_grab_current_ns(void)
1314 struct net *ns = current->nsproxy->net_ns;
1315 #ifdef CONFIG_NET_NS
1316 if (ns)
1317 atomic_inc(&ns->passive);
1318 #endif
1319 return ns;
1322 static const void *net_initial_ns(void)
1324 return &init_net;
1327 static const void *net_netlink_ns(struct sock *sk)
1329 return sock_net(sk);
1332 struct kobj_ns_type_operations net_ns_type_operations = {
1333 .type = KOBJ_NS_TYPE_NET,
1334 .current_may_mount = net_current_may_mount,
1335 .grab_current_ns = net_grab_current_ns,
1336 .netlink_ns = net_netlink_ns,
1337 .initial_ns = net_initial_ns,
1338 .drop_ns = net_drop_ns,
1340 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1342 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1344 struct net_device *dev = to_net_dev(d);
1345 int retval;
1347 /* pass interface to uevent. */
1348 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1349 if (retval)
1350 goto exit;
1352 /* pass ifindex to uevent.
1353 * ifindex is useful as it won't change (interface name may change)
1354 * and is what RtNetlink uses natively. */
1355 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1357 exit:
1358 return retval;
1362 * netdev_release -- destroy and free a dead device.
1363 * Called when last reference to device kobject is gone.
1365 static void netdev_release(struct device *d)
1367 struct net_device *dev = to_net_dev(d);
1369 BUG_ON(dev->reg_state != NETREG_RELEASED);
1371 kfree(dev->ifalias);
1372 netdev_freemem(dev);
1375 static const void *net_namespace(struct device *d)
1377 struct net_device *dev;
1378 dev = container_of(d, struct net_device, dev);
1379 return dev_net(dev);
1382 static struct class net_class = {
1383 .name = "net",
1384 .dev_release = netdev_release,
1385 .dev_groups = net_class_groups,
1386 .dev_uevent = netdev_uevent,
1387 .ns_type = &net_ns_type_operations,
1388 .namespace = net_namespace,
1391 /* Delete sysfs entries but hold kobject reference until after all
1392 * netdev references are gone.
1394 void netdev_unregister_kobject(struct net_device *ndev)
1396 struct device *dev = &(ndev->dev);
1398 kobject_get(&dev->kobj);
1400 remove_queue_kobjects(ndev);
1402 pm_runtime_set_memalloc_noio(dev, false);
1404 device_del(dev);
1407 /* Create sysfs entries for network device. */
1408 int netdev_register_kobject(struct net_device *ndev)
1410 struct device *dev = &(ndev->dev);
1411 const struct attribute_group **groups = ndev->sysfs_groups;
1412 int error = 0;
1414 device_initialize(dev);
1415 dev->class = &net_class;
1416 dev->platform_data = ndev;
1417 dev->groups = groups;
1419 dev_set_name(dev, "%s", ndev->name);
1421 #ifdef CONFIG_SYSFS
1422 /* Allow for a device specific group */
1423 if (*groups)
1424 groups++;
1426 *groups++ = &netstat_group;
1428 #if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1429 if (ndev->ieee80211_ptr)
1430 *groups++ = &wireless_group;
1431 #if IS_ENABLED(CONFIG_WIRELESS_EXT)
1432 else if (ndev->wireless_handlers)
1433 *groups++ = &wireless_group;
1434 #endif
1435 #endif
1436 #endif /* CONFIG_SYSFS */
1438 error = device_add(dev);
1439 if (error)
1440 return error;
1442 error = register_queue_kobjects(ndev);
1443 if (error) {
1444 device_del(dev);
1445 return error;
1448 pm_runtime_set_memalloc_noio(dev, true);
1450 return error;
1453 int netdev_class_create_file_ns(struct class_attribute *class_attr,
1454 const void *ns)
1456 return class_create_file_ns(&net_class, class_attr, ns);
1458 EXPORT_SYMBOL(netdev_class_create_file_ns);
1460 void netdev_class_remove_file_ns(struct class_attribute *class_attr,
1461 const void *ns)
1463 class_remove_file_ns(&net_class, class_attr, ns);
1465 EXPORT_SYMBOL(netdev_class_remove_file_ns);
1467 int __init netdev_kobject_init(void)
1469 kobj_ns_type_register(&net_ns_type_operations);
1470 return class_register(&net_class);