localmodconfig: Check if configs are already set for selects
[linux-2.6.git] / net / core / net-sysfs.c
blobfdf9e61d0651fab74e2836593a872586cd8efec3
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 <linux/if_arp.h>
16 #include <linux/slab.h>
17 #include <linux/nsproxy.h>
18 #include <net/sock.h>
19 #include <net/net_namespace.h>
20 #include <linux/rtnetlink.h>
21 #include <linux/wireless.h>
22 #include <linux/vmalloc.h>
23 #include <linux/export.h>
24 #include <linux/jiffies.h>
25 #include <net/wext.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 *net = to_net_dev(dev);
48 ssize_t ret = -EINVAL;
50 read_lock(&dev_base_lock);
51 if (dev_isalive(net))
52 ret = (*format)(net, 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 *net, char *buf) \
61 { \
62 return sprintf(buf, format_string, net->field); \
63 } \
64 static ssize_t show_##field(struct device *dev, \
65 struct device_attribute *attr, char *buf) \
66 { \
67 return netdev_show(dev, attr, buf, format_##field); \
71 /* use same locking and permission rules as SIF* ioctl's */
72 static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
73 const char *buf, size_t len,
74 int (*set)(struct net_device *, unsigned long))
76 struct net_device *net = to_net_dev(dev);
77 unsigned long new;
78 int ret = -EINVAL;
80 if (!capable(CAP_NET_ADMIN))
81 return -EPERM;
83 ret = kstrtoul(buf, 0, &new);
84 if (ret)
85 goto err;
87 if (!rtnl_trylock())
88 return restart_syscall();
90 if (dev_isalive(net)) {
91 if ((ret = (*set)(net, new)) == 0)
92 ret = len;
94 rtnl_unlock();
95 err:
96 return ret;
99 NETDEVICE_SHOW(dev_id, fmt_hex);
100 NETDEVICE_SHOW(addr_assign_type, fmt_dec);
101 NETDEVICE_SHOW(addr_len, fmt_dec);
102 NETDEVICE_SHOW(iflink, fmt_dec);
103 NETDEVICE_SHOW(ifindex, fmt_dec);
104 NETDEVICE_SHOW(type, fmt_dec);
105 NETDEVICE_SHOW(link_mode, fmt_dec);
107 /* use same locking rules as GIFHWADDR ioctl's */
108 static ssize_t show_address(struct device *dev, struct device_attribute *attr,
109 char *buf)
111 struct net_device *net = to_net_dev(dev);
112 ssize_t ret = -EINVAL;
114 read_lock(&dev_base_lock);
115 if (dev_isalive(net))
116 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
117 read_unlock(&dev_base_lock);
118 return ret;
121 static ssize_t show_broadcast(struct device *dev,
122 struct device_attribute *attr, char *buf)
124 struct net_device *net = to_net_dev(dev);
125 if (dev_isalive(net))
126 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
127 return -EINVAL;
130 static ssize_t show_carrier(struct device *dev,
131 struct device_attribute *attr, char *buf)
133 struct net_device *netdev = to_net_dev(dev);
134 if (netif_running(netdev)) {
135 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
137 return -EINVAL;
140 static ssize_t show_speed(struct device *dev,
141 struct device_attribute *attr, char *buf)
143 struct net_device *netdev = to_net_dev(dev);
144 int ret = -EINVAL;
146 if (!rtnl_trylock())
147 return restart_syscall();
149 if (netif_running(netdev)) {
150 struct ethtool_cmd cmd;
151 if (!__ethtool_get_settings(netdev, &cmd))
152 ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
154 rtnl_unlock();
155 return ret;
158 static ssize_t show_duplex(struct device *dev,
159 struct device_attribute *attr, char *buf)
161 struct net_device *netdev = to_net_dev(dev);
162 int ret = -EINVAL;
164 if (!rtnl_trylock())
165 return restart_syscall();
167 if (netif_running(netdev)) {
168 struct ethtool_cmd cmd;
169 if (!__ethtool_get_settings(netdev, &cmd))
170 ret = sprintf(buf, "%s\n",
171 cmd.duplex ? "full" : "half");
173 rtnl_unlock();
174 return ret;
177 static ssize_t show_dormant(struct device *dev,
178 struct device_attribute *attr, char *buf)
180 struct net_device *netdev = to_net_dev(dev);
182 if (netif_running(netdev))
183 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
185 return -EINVAL;
188 static const char *const operstates[] = {
189 "unknown",
190 "notpresent", /* currently unused */
191 "down",
192 "lowerlayerdown",
193 "testing", /* currently unused */
194 "dormant",
195 "up"
198 static ssize_t show_operstate(struct device *dev,
199 struct device_attribute *attr, char *buf)
201 const struct net_device *netdev = to_net_dev(dev);
202 unsigned char operstate;
204 read_lock(&dev_base_lock);
205 operstate = netdev->operstate;
206 if (!netif_running(netdev))
207 operstate = IF_OPER_DOWN;
208 read_unlock(&dev_base_lock);
210 if (operstate >= ARRAY_SIZE(operstates))
211 return -EINVAL; /* should not happen */
213 return sprintf(buf, "%s\n", operstates[operstate]);
216 /* read-write attributes */
217 NETDEVICE_SHOW(mtu, fmt_dec);
219 static int change_mtu(struct net_device *net, unsigned long new_mtu)
221 return dev_set_mtu(net, (int) new_mtu);
224 static ssize_t store_mtu(struct device *dev, struct device_attribute *attr,
225 const char *buf, size_t len)
227 return netdev_store(dev, attr, buf, len, change_mtu);
230 NETDEVICE_SHOW(flags, fmt_hex);
232 static int change_flags(struct net_device *net, unsigned long new_flags)
234 return dev_change_flags(net, (unsigned int) new_flags);
237 static ssize_t store_flags(struct device *dev, struct device_attribute *attr,
238 const char *buf, size_t len)
240 return netdev_store(dev, attr, buf, len, change_flags);
243 NETDEVICE_SHOW(tx_queue_len, fmt_ulong);
245 static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
247 net->tx_queue_len = new_len;
248 return 0;
251 static ssize_t store_tx_queue_len(struct device *dev,
252 struct device_attribute *attr,
253 const char *buf, size_t len)
255 return netdev_store(dev, attr, buf, len, change_tx_queue_len);
258 static ssize_t store_ifalias(struct device *dev, struct device_attribute *attr,
259 const char *buf, size_t len)
261 struct net_device *netdev = to_net_dev(dev);
262 size_t count = len;
263 ssize_t ret;
265 if (!capable(CAP_NET_ADMIN))
266 return -EPERM;
268 /* ignore trailing newline */
269 if (len > 0 && buf[len - 1] == '\n')
270 --count;
272 if (!rtnl_trylock())
273 return restart_syscall();
274 ret = dev_set_alias(netdev, buf, count);
275 rtnl_unlock();
277 return ret < 0 ? ret : len;
280 static ssize_t show_ifalias(struct device *dev,
281 struct device_attribute *attr, char *buf)
283 const struct net_device *netdev = to_net_dev(dev);
284 ssize_t ret = 0;
286 if (!rtnl_trylock())
287 return restart_syscall();
288 if (netdev->ifalias)
289 ret = sprintf(buf, "%s\n", netdev->ifalias);
290 rtnl_unlock();
291 return ret;
294 NETDEVICE_SHOW(group, fmt_dec);
296 static int change_group(struct net_device *net, unsigned long new_group)
298 dev_set_group(net, (int) new_group);
299 return 0;
302 static ssize_t store_group(struct device *dev, struct device_attribute *attr,
303 const char *buf, size_t len)
305 return netdev_store(dev, attr, buf, len, change_group);
308 static struct device_attribute net_class_attributes[] = {
309 __ATTR(addr_assign_type, S_IRUGO, show_addr_assign_type, NULL),
310 __ATTR(addr_len, S_IRUGO, show_addr_len, NULL),
311 __ATTR(dev_id, S_IRUGO, show_dev_id, NULL),
312 __ATTR(ifalias, S_IRUGO | S_IWUSR, show_ifalias, store_ifalias),
313 __ATTR(iflink, S_IRUGO, show_iflink, NULL),
314 __ATTR(ifindex, S_IRUGO, show_ifindex, NULL),
315 __ATTR(type, S_IRUGO, show_type, NULL),
316 __ATTR(link_mode, S_IRUGO, show_link_mode, NULL),
317 __ATTR(address, S_IRUGO, show_address, NULL),
318 __ATTR(broadcast, S_IRUGO, show_broadcast, NULL),
319 __ATTR(carrier, S_IRUGO, show_carrier, NULL),
320 __ATTR(speed, S_IRUGO, show_speed, NULL),
321 __ATTR(duplex, S_IRUGO, show_duplex, NULL),
322 __ATTR(dormant, S_IRUGO, show_dormant, NULL),
323 __ATTR(operstate, S_IRUGO, show_operstate, NULL),
324 __ATTR(mtu, S_IRUGO | S_IWUSR, show_mtu, store_mtu),
325 __ATTR(flags, S_IRUGO | S_IWUSR, show_flags, store_flags),
326 __ATTR(tx_queue_len, S_IRUGO | S_IWUSR, show_tx_queue_len,
327 store_tx_queue_len),
328 __ATTR(netdev_group, S_IRUGO | S_IWUSR, show_group, store_group),
332 /* Show a given an attribute in the statistics group */
333 static ssize_t netstat_show(const struct device *d,
334 struct device_attribute *attr, char *buf,
335 unsigned long offset)
337 struct net_device *dev = to_net_dev(d);
338 ssize_t ret = -EINVAL;
340 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
341 offset % sizeof(u64) != 0);
343 read_lock(&dev_base_lock);
344 if (dev_isalive(dev)) {
345 struct rtnl_link_stats64 temp;
346 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
348 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
350 read_unlock(&dev_base_lock);
351 return ret;
354 /* generate a read-only statistics attribute */
355 #define NETSTAT_ENTRY(name) \
356 static ssize_t show_##name(struct device *d, \
357 struct device_attribute *attr, char *buf) \
359 return netstat_show(d, attr, buf, \
360 offsetof(struct rtnl_link_stats64, name)); \
362 static DEVICE_ATTR(name, S_IRUGO, show_##name, NULL)
364 NETSTAT_ENTRY(rx_packets);
365 NETSTAT_ENTRY(tx_packets);
366 NETSTAT_ENTRY(rx_bytes);
367 NETSTAT_ENTRY(tx_bytes);
368 NETSTAT_ENTRY(rx_errors);
369 NETSTAT_ENTRY(tx_errors);
370 NETSTAT_ENTRY(rx_dropped);
371 NETSTAT_ENTRY(tx_dropped);
372 NETSTAT_ENTRY(multicast);
373 NETSTAT_ENTRY(collisions);
374 NETSTAT_ENTRY(rx_length_errors);
375 NETSTAT_ENTRY(rx_over_errors);
376 NETSTAT_ENTRY(rx_crc_errors);
377 NETSTAT_ENTRY(rx_frame_errors);
378 NETSTAT_ENTRY(rx_fifo_errors);
379 NETSTAT_ENTRY(rx_missed_errors);
380 NETSTAT_ENTRY(tx_aborted_errors);
381 NETSTAT_ENTRY(tx_carrier_errors);
382 NETSTAT_ENTRY(tx_fifo_errors);
383 NETSTAT_ENTRY(tx_heartbeat_errors);
384 NETSTAT_ENTRY(tx_window_errors);
385 NETSTAT_ENTRY(rx_compressed);
386 NETSTAT_ENTRY(tx_compressed);
388 static struct attribute *netstat_attrs[] = {
389 &dev_attr_rx_packets.attr,
390 &dev_attr_tx_packets.attr,
391 &dev_attr_rx_bytes.attr,
392 &dev_attr_tx_bytes.attr,
393 &dev_attr_rx_errors.attr,
394 &dev_attr_tx_errors.attr,
395 &dev_attr_rx_dropped.attr,
396 &dev_attr_tx_dropped.attr,
397 &dev_attr_multicast.attr,
398 &dev_attr_collisions.attr,
399 &dev_attr_rx_length_errors.attr,
400 &dev_attr_rx_over_errors.attr,
401 &dev_attr_rx_crc_errors.attr,
402 &dev_attr_rx_frame_errors.attr,
403 &dev_attr_rx_fifo_errors.attr,
404 &dev_attr_rx_missed_errors.attr,
405 &dev_attr_tx_aborted_errors.attr,
406 &dev_attr_tx_carrier_errors.attr,
407 &dev_attr_tx_fifo_errors.attr,
408 &dev_attr_tx_heartbeat_errors.attr,
409 &dev_attr_tx_window_errors.attr,
410 &dev_attr_rx_compressed.attr,
411 &dev_attr_tx_compressed.attr,
412 NULL
416 static struct attribute_group netstat_group = {
417 .name = "statistics",
418 .attrs = netstat_attrs,
421 #ifdef CONFIG_WIRELESS_EXT_SYSFS
422 /* helper function that does all the locking etc for wireless stats */
423 static ssize_t wireless_show(struct device *d, char *buf,
424 ssize_t (*format)(const struct iw_statistics *,
425 char *))
427 struct net_device *dev = to_net_dev(d);
428 const struct iw_statistics *iw;
429 ssize_t ret = -EINVAL;
431 if (!rtnl_trylock())
432 return restart_syscall();
433 if (dev_isalive(dev)) {
434 iw = get_wireless_stats(dev);
435 if (iw)
436 ret = (*format)(iw, buf);
438 rtnl_unlock();
440 return ret;
443 /* show function template for wireless fields */
444 #define WIRELESS_SHOW(name, field, format_string) \
445 static ssize_t format_iw_##name(const struct iw_statistics *iw, char *buf) \
447 return sprintf(buf, format_string, iw->field); \
449 static ssize_t show_iw_##name(struct device *d, \
450 struct device_attribute *attr, char *buf) \
452 return wireless_show(d, buf, format_iw_##name); \
454 static DEVICE_ATTR(name, S_IRUGO, show_iw_##name, NULL)
456 WIRELESS_SHOW(status, status, fmt_hex);
457 WIRELESS_SHOW(link, qual.qual, fmt_dec);
458 WIRELESS_SHOW(level, qual.level, fmt_dec);
459 WIRELESS_SHOW(noise, qual.noise, fmt_dec);
460 WIRELESS_SHOW(nwid, discard.nwid, fmt_dec);
461 WIRELESS_SHOW(crypt, discard.code, fmt_dec);
462 WIRELESS_SHOW(fragment, discard.fragment, fmt_dec);
463 WIRELESS_SHOW(misc, discard.misc, fmt_dec);
464 WIRELESS_SHOW(retries, discard.retries, fmt_dec);
465 WIRELESS_SHOW(beacon, miss.beacon, fmt_dec);
467 static struct attribute *wireless_attrs[] = {
468 &dev_attr_status.attr,
469 &dev_attr_link.attr,
470 &dev_attr_level.attr,
471 &dev_attr_noise.attr,
472 &dev_attr_nwid.attr,
473 &dev_attr_crypt.attr,
474 &dev_attr_fragment.attr,
475 &dev_attr_retries.attr,
476 &dev_attr_misc.attr,
477 &dev_attr_beacon.attr,
478 NULL
481 static struct attribute_group wireless_group = {
482 .name = "wireless",
483 .attrs = wireless_attrs,
485 #endif
486 #endif /* CONFIG_SYSFS */
488 #ifdef CONFIG_RPS
490 * RX queue sysfs structures and functions.
492 struct rx_queue_attribute {
493 struct attribute attr;
494 ssize_t (*show)(struct netdev_rx_queue *queue,
495 struct rx_queue_attribute *attr, char *buf);
496 ssize_t (*store)(struct netdev_rx_queue *queue,
497 struct rx_queue_attribute *attr, const char *buf, size_t len);
499 #define to_rx_queue_attr(_attr) container_of(_attr, \
500 struct rx_queue_attribute, attr)
502 #define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
504 static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
505 char *buf)
507 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
508 struct netdev_rx_queue *queue = to_rx_queue(kobj);
510 if (!attribute->show)
511 return -EIO;
513 return attribute->show(queue, attribute, buf);
516 static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
517 const char *buf, size_t count)
519 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
520 struct netdev_rx_queue *queue = to_rx_queue(kobj);
522 if (!attribute->store)
523 return -EIO;
525 return attribute->store(queue, attribute, buf, count);
528 static const struct sysfs_ops rx_queue_sysfs_ops = {
529 .show = rx_queue_attr_show,
530 .store = rx_queue_attr_store,
533 static ssize_t show_rps_map(struct netdev_rx_queue *queue,
534 struct rx_queue_attribute *attribute, char *buf)
536 struct rps_map *map;
537 cpumask_var_t mask;
538 size_t len = 0;
539 int i;
541 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
542 return -ENOMEM;
544 rcu_read_lock();
545 map = rcu_dereference(queue->rps_map);
546 if (map)
547 for (i = 0; i < map->len; i++)
548 cpumask_set_cpu(map->cpus[i], mask);
550 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
551 if (PAGE_SIZE - len < 3) {
552 rcu_read_unlock();
553 free_cpumask_var(mask);
554 return -EINVAL;
556 rcu_read_unlock();
558 free_cpumask_var(mask);
559 len += sprintf(buf + len, "\n");
560 return len;
563 static ssize_t store_rps_map(struct netdev_rx_queue *queue,
564 struct rx_queue_attribute *attribute,
565 const char *buf, size_t len)
567 struct rps_map *old_map, *map;
568 cpumask_var_t mask;
569 int err, cpu, i;
570 static DEFINE_SPINLOCK(rps_map_lock);
572 if (!capable(CAP_NET_ADMIN))
573 return -EPERM;
575 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
576 return -ENOMEM;
578 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
579 if (err) {
580 free_cpumask_var(mask);
581 return err;
584 map = kzalloc(max_t(unsigned int,
585 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
586 GFP_KERNEL);
587 if (!map) {
588 free_cpumask_var(mask);
589 return -ENOMEM;
592 i = 0;
593 for_each_cpu_and(cpu, mask, cpu_online_mask)
594 map->cpus[i++] = cpu;
596 if (i)
597 map->len = i;
598 else {
599 kfree(map);
600 map = NULL;
603 spin_lock(&rps_map_lock);
604 old_map = rcu_dereference_protected(queue->rps_map,
605 lockdep_is_held(&rps_map_lock));
606 rcu_assign_pointer(queue->rps_map, map);
607 spin_unlock(&rps_map_lock);
609 if (map)
610 static_key_slow_inc(&rps_needed);
611 if (old_map) {
612 kfree_rcu(old_map, rcu);
613 static_key_slow_dec(&rps_needed);
615 free_cpumask_var(mask);
616 return len;
619 static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
620 struct rx_queue_attribute *attr,
621 char *buf)
623 struct rps_dev_flow_table *flow_table;
624 unsigned long val = 0;
626 rcu_read_lock();
627 flow_table = rcu_dereference(queue->rps_flow_table);
628 if (flow_table)
629 val = (unsigned long)flow_table->mask + 1;
630 rcu_read_unlock();
632 return sprintf(buf, "%lu\n", val);
635 static void rps_dev_flow_table_release_work(struct work_struct *work)
637 struct rps_dev_flow_table *table = container_of(work,
638 struct rps_dev_flow_table, free_work);
640 vfree(table);
643 static void rps_dev_flow_table_release(struct rcu_head *rcu)
645 struct rps_dev_flow_table *table = container_of(rcu,
646 struct rps_dev_flow_table, rcu);
648 INIT_WORK(&table->free_work, rps_dev_flow_table_release_work);
649 schedule_work(&table->free_work);
652 static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
653 struct rx_queue_attribute *attr,
654 const char *buf, size_t len)
656 unsigned long mask, count;
657 struct rps_dev_flow_table *table, *old_table;
658 static DEFINE_SPINLOCK(rps_dev_flow_lock);
659 int rc;
661 if (!capable(CAP_NET_ADMIN))
662 return -EPERM;
664 rc = kstrtoul(buf, 0, &count);
665 if (rc < 0)
666 return rc;
668 if (count) {
669 mask = count - 1;
670 /* mask = roundup_pow_of_two(count) - 1;
671 * without overflows...
673 while ((mask | (mask >> 1)) != mask)
674 mask |= (mask >> 1);
675 /* On 64 bit arches, must check mask fits in table->mask (u32),
676 * and on 32bit arches, must check RPS_DEV_FLOW_TABLE_SIZE(mask + 1)
677 * doesnt overflow.
679 #if BITS_PER_LONG > 32
680 if (mask > (unsigned long)(u32)mask)
681 return -EINVAL;
682 #else
683 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
684 / sizeof(struct rps_dev_flow)) {
685 /* Enforce a limit to prevent overflow */
686 return -EINVAL;
688 #endif
689 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
690 if (!table)
691 return -ENOMEM;
693 table->mask = mask;
694 for (count = 0; count <= mask; count++)
695 table->flows[count].cpu = RPS_NO_CPU;
696 } else
697 table = NULL;
699 spin_lock(&rps_dev_flow_lock);
700 old_table = rcu_dereference_protected(queue->rps_flow_table,
701 lockdep_is_held(&rps_dev_flow_lock));
702 rcu_assign_pointer(queue->rps_flow_table, table);
703 spin_unlock(&rps_dev_flow_lock);
705 if (old_table)
706 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
708 return len;
711 static struct rx_queue_attribute rps_cpus_attribute =
712 __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
715 static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
716 __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
717 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
719 static struct attribute *rx_queue_default_attrs[] = {
720 &rps_cpus_attribute.attr,
721 &rps_dev_flow_table_cnt_attribute.attr,
722 NULL
725 static void rx_queue_release(struct kobject *kobj)
727 struct netdev_rx_queue *queue = to_rx_queue(kobj);
728 struct rps_map *map;
729 struct rps_dev_flow_table *flow_table;
732 map = rcu_dereference_protected(queue->rps_map, 1);
733 if (map) {
734 RCU_INIT_POINTER(queue->rps_map, NULL);
735 kfree_rcu(map, rcu);
738 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
739 if (flow_table) {
740 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
741 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
744 memset(kobj, 0, sizeof(*kobj));
745 dev_put(queue->dev);
748 static struct kobj_type rx_queue_ktype = {
749 .sysfs_ops = &rx_queue_sysfs_ops,
750 .release = rx_queue_release,
751 .default_attrs = rx_queue_default_attrs,
754 static int rx_queue_add_kobject(struct net_device *net, int index)
756 struct netdev_rx_queue *queue = net->_rx + index;
757 struct kobject *kobj = &queue->kobj;
758 int error = 0;
760 kobj->kset = net->queues_kset;
761 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
762 "rx-%u", index);
763 if (error) {
764 kobject_put(kobj);
765 return error;
768 kobject_uevent(kobj, KOBJ_ADD);
769 dev_hold(queue->dev);
771 return error;
773 #endif /* CONFIG_RPS */
776 net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
778 #ifdef CONFIG_RPS
779 int i;
780 int error = 0;
782 for (i = old_num; i < new_num; i++) {
783 error = rx_queue_add_kobject(net, i);
784 if (error) {
785 new_num = old_num;
786 break;
790 while (--i >= new_num)
791 kobject_put(&net->_rx[i].kobj);
793 return error;
794 #else
795 return 0;
796 #endif
799 #ifdef CONFIG_SYSFS
801 * netdev_queue sysfs structures and functions.
803 struct netdev_queue_attribute {
804 struct attribute attr;
805 ssize_t (*show)(struct netdev_queue *queue,
806 struct netdev_queue_attribute *attr, char *buf);
807 ssize_t (*store)(struct netdev_queue *queue,
808 struct netdev_queue_attribute *attr, const char *buf, size_t len);
810 #define to_netdev_queue_attr(_attr) container_of(_attr, \
811 struct netdev_queue_attribute, attr)
813 #define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
815 static ssize_t netdev_queue_attr_show(struct kobject *kobj,
816 struct attribute *attr, char *buf)
818 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
819 struct netdev_queue *queue = to_netdev_queue(kobj);
821 if (!attribute->show)
822 return -EIO;
824 return attribute->show(queue, attribute, buf);
827 static ssize_t netdev_queue_attr_store(struct kobject *kobj,
828 struct attribute *attr,
829 const char *buf, size_t count)
831 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
832 struct netdev_queue *queue = to_netdev_queue(kobj);
834 if (!attribute->store)
835 return -EIO;
837 return attribute->store(queue, attribute, buf, count);
840 static const struct sysfs_ops netdev_queue_sysfs_ops = {
841 .show = netdev_queue_attr_show,
842 .store = netdev_queue_attr_store,
845 static ssize_t show_trans_timeout(struct netdev_queue *queue,
846 struct netdev_queue_attribute *attribute,
847 char *buf)
849 unsigned long trans_timeout;
851 spin_lock_irq(&queue->_xmit_lock);
852 trans_timeout = queue->trans_timeout;
853 spin_unlock_irq(&queue->_xmit_lock);
855 return sprintf(buf, "%lu", trans_timeout);
858 static struct netdev_queue_attribute queue_trans_timeout =
859 __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
861 #ifdef CONFIG_BQL
863 * Byte queue limits sysfs structures and functions.
865 static ssize_t bql_show(char *buf, unsigned int value)
867 return sprintf(buf, "%u\n", value);
870 static ssize_t bql_set(const char *buf, const size_t count,
871 unsigned int *pvalue)
873 unsigned int value;
874 int err;
876 if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
877 value = DQL_MAX_LIMIT;
878 else {
879 err = kstrtouint(buf, 10, &value);
880 if (err < 0)
881 return err;
882 if (value > DQL_MAX_LIMIT)
883 return -EINVAL;
886 *pvalue = value;
888 return count;
891 static ssize_t bql_show_hold_time(struct netdev_queue *queue,
892 struct netdev_queue_attribute *attr,
893 char *buf)
895 struct dql *dql = &queue->dql;
897 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
900 static ssize_t bql_set_hold_time(struct netdev_queue *queue,
901 struct netdev_queue_attribute *attribute,
902 const char *buf, size_t len)
904 struct dql *dql = &queue->dql;
905 unsigned int value;
906 int err;
908 err = kstrtouint(buf, 10, &value);
909 if (err < 0)
910 return err;
912 dql->slack_hold_time = msecs_to_jiffies(value);
914 return len;
917 static struct netdev_queue_attribute bql_hold_time_attribute =
918 __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
919 bql_set_hold_time);
921 static ssize_t bql_show_inflight(struct netdev_queue *queue,
922 struct netdev_queue_attribute *attr,
923 char *buf)
925 struct dql *dql = &queue->dql;
927 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
930 static struct netdev_queue_attribute bql_inflight_attribute =
931 __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
933 #define BQL_ATTR(NAME, FIELD) \
934 static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
935 struct netdev_queue_attribute *attr, \
936 char *buf) \
938 return bql_show(buf, queue->dql.FIELD); \
941 static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
942 struct netdev_queue_attribute *attr, \
943 const char *buf, size_t len) \
945 return bql_set(buf, len, &queue->dql.FIELD); \
948 static struct netdev_queue_attribute bql_ ## NAME ## _attribute = \
949 __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME, \
950 bql_set_ ## NAME);
952 BQL_ATTR(limit, limit)
953 BQL_ATTR(limit_max, max_limit)
954 BQL_ATTR(limit_min, min_limit)
956 static struct attribute *dql_attrs[] = {
957 &bql_limit_attribute.attr,
958 &bql_limit_max_attribute.attr,
959 &bql_limit_min_attribute.attr,
960 &bql_hold_time_attribute.attr,
961 &bql_inflight_attribute.attr,
962 NULL
965 static struct attribute_group dql_group = {
966 .name = "byte_queue_limits",
967 .attrs = dql_attrs,
969 #endif /* CONFIG_BQL */
971 #ifdef CONFIG_XPS
972 static inline unsigned int get_netdev_queue_index(struct netdev_queue *queue)
974 struct net_device *dev = queue->dev;
975 int i;
977 for (i = 0; i < dev->num_tx_queues; i++)
978 if (queue == &dev->_tx[i])
979 break;
981 BUG_ON(i >= dev->num_tx_queues);
983 return i;
987 static ssize_t show_xps_map(struct netdev_queue *queue,
988 struct netdev_queue_attribute *attribute, char *buf)
990 struct net_device *dev = queue->dev;
991 struct xps_dev_maps *dev_maps;
992 cpumask_var_t mask;
993 unsigned long index;
994 size_t len = 0;
995 int i;
997 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
998 return -ENOMEM;
1000 index = get_netdev_queue_index(queue);
1002 rcu_read_lock();
1003 dev_maps = rcu_dereference(dev->xps_maps);
1004 if (dev_maps) {
1005 for_each_possible_cpu(i) {
1006 struct xps_map *map =
1007 rcu_dereference(dev_maps->cpu_map[i]);
1008 if (map) {
1009 int j;
1010 for (j = 0; j < map->len; j++) {
1011 if (map->queues[j] == index) {
1012 cpumask_set_cpu(i, mask);
1013 break;
1019 rcu_read_unlock();
1021 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
1022 if (PAGE_SIZE - len < 3) {
1023 free_cpumask_var(mask);
1024 return -EINVAL;
1027 free_cpumask_var(mask);
1028 len += sprintf(buf + len, "\n");
1029 return len;
1032 static DEFINE_MUTEX(xps_map_mutex);
1033 #define xmap_dereference(P) \
1034 rcu_dereference_protected((P), lockdep_is_held(&xps_map_mutex))
1036 static void xps_queue_release(struct netdev_queue *queue)
1038 struct net_device *dev = queue->dev;
1039 struct xps_dev_maps *dev_maps;
1040 struct xps_map *map;
1041 unsigned long index;
1042 int i, pos, nonempty = 0;
1044 index = get_netdev_queue_index(queue);
1046 mutex_lock(&xps_map_mutex);
1047 dev_maps = xmap_dereference(dev->xps_maps);
1049 if (dev_maps) {
1050 for_each_possible_cpu(i) {
1051 map = xmap_dereference(dev_maps->cpu_map[i]);
1052 if (!map)
1053 continue;
1055 for (pos = 0; pos < map->len; pos++)
1056 if (map->queues[pos] == index)
1057 break;
1059 if (pos < map->len) {
1060 if (map->len > 1)
1061 map->queues[pos] =
1062 map->queues[--map->len];
1063 else {
1064 RCU_INIT_POINTER(dev_maps->cpu_map[i],
1065 NULL);
1066 kfree_rcu(map, rcu);
1067 map = NULL;
1070 if (map)
1071 nonempty = 1;
1074 if (!nonempty) {
1075 RCU_INIT_POINTER(dev->xps_maps, NULL);
1076 kfree_rcu(dev_maps, rcu);
1079 mutex_unlock(&xps_map_mutex);
1082 static ssize_t store_xps_map(struct netdev_queue *queue,
1083 struct netdev_queue_attribute *attribute,
1084 const char *buf, size_t len)
1086 struct net_device *dev = queue->dev;
1087 cpumask_var_t mask;
1088 int err, i, cpu, pos, map_len, alloc_len, need_set;
1089 unsigned long index;
1090 struct xps_map *map, *new_map;
1091 struct xps_dev_maps *dev_maps, *new_dev_maps;
1092 int nonempty = 0;
1093 int numa_node_id = -2;
1095 if (!capable(CAP_NET_ADMIN))
1096 return -EPERM;
1098 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1099 return -ENOMEM;
1101 index = get_netdev_queue_index(queue);
1103 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1104 if (err) {
1105 free_cpumask_var(mask);
1106 return err;
1109 new_dev_maps = kzalloc(max_t(unsigned int,
1110 XPS_DEV_MAPS_SIZE, L1_CACHE_BYTES), GFP_KERNEL);
1111 if (!new_dev_maps) {
1112 free_cpumask_var(mask);
1113 return -ENOMEM;
1116 mutex_lock(&xps_map_mutex);
1118 dev_maps = xmap_dereference(dev->xps_maps);
1120 for_each_possible_cpu(cpu) {
1121 map = dev_maps ?
1122 xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1123 new_map = map;
1124 if (map) {
1125 for (pos = 0; pos < map->len; pos++)
1126 if (map->queues[pos] == index)
1127 break;
1128 map_len = map->len;
1129 alloc_len = map->alloc_len;
1130 } else
1131 pos = map_len = alloc_len = 0;
1133 need_set = cpumask_test_cpu(cpu, mask) && cpu_online(cpu);
1134 #ifdef CONFIG_NUMA
1135 if (need_set) {
1136 if (numa_node_id == -2)
1137 numa_node_id = cpu_to_node(cpu);
1138 else if (numa_node_id != cpu_to_node(cpu))
1139 numa_node_id = -1;
1141 #endif
1142 if (need_set && pos >= map_len) {
1143 /* Need to add queue to this CPU's map */
1144 if (map_len >= alloc_len) {
1145 alloc_len = alloc_len ?
1146 2 * alloc_len : XPS_MIN_MAP_ALLOC;
1147 new_map = kzalloc_node(XPS_MAP_SIZE(alloc_len),
1148 GFP_KERNEL,
1149 cpu_to_node(cpu));
1150 if (!new_map)
1151 goto error;
1152 new_map->alloc_len = alloc_len;
1153 for (i = 0; i < map_len; i++)
1154 new_map->queues[i] = map->queues[i];
1155 new_map->len = map_len;
1157 new_map->queues[new_map->len++] = index;
1158 } else if (!need_set && pos < map_len) {
1159 /* Need to remove queue from this CPU's map */
1160 if (map_len > 1)
1161 new_map->queues[pos] =
1162 new_map->queues[--new_map->len];
1163 else
1164 new_map = NULL;
1166 RCU_INIT_POINTER(new_dev_maps->cpu_map[cpu], new_map);
1169 /* Cleanup old maps */
1170 for_each_possible_cpu(cpu) {
1171 map = dev_maps ?
1172 xmap_dereference(dev_maps->cpu_map[cpu]) : NULL;
1173 if (map && xmap_dereference(new_dev_maps->cpu_map[cpu]) != map)
1174 kfree_rcu(map, rcu);
1175 if (new_dev_maps->cpu_map[cpu])
1176 nonempty = 1;
1179 if (nonempty) {
1180 rcu_assign_pointer(dev->xps_maps, new_dev_maps);
1181 } else {
1182 kfree(new_dev_maps);
1183 RCU_INIT_POINTER(dev->xps_maps, NULL);
1186 if (dev_maps)
1187 kfree_rcu(dev_maps, rcu);
1189 netdev_queue_numa_node_write(queue, (numa_node_id >= 0) ? numa_node_id :
1190 NUMA_NO_NODE);
1192 mutex_unlock(&xps_map_mutex);
1194 free_cpumask_var(mask);
1195 return len;
1197 error:
1198 mutex_unlock(&xps_map_mutex);
1200 if (new_dev_maps)
1201 for_each_possible_cpu(i)
1202 kfree(rcu_dereference_protected(
1203 new_dev_maps->cpu_map[i],
1204 1));
1205 kfree(new_dev_maps);
1206 free_cpumask_var(mask);
1207 return -ENOMEM;
1210 static struct netdev_queue_attribute xps_cpus_attribute =
1211 __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1212 #endif /* CONFIG_XPS */
1214 static struct attribute *netdev_queue_default_attrs[] = {
1215 &queue_trans_timeout.attr,
1216 #ifdef CONFIG_XPS
1217 &xps_cpus_attribute.attr,
1218 #endif
1219 NULL
1222 static void netdev_queue_release(struct kobject *kobj)
1224 struct netdev_queue *queue = to_netdev_queue(kobj);
1226 #ifdef CONFIG_XPS
1227 xps_queue_release(queue);
1228 #endif
1230 memset(kobj, 0, sizeof(*kobj));
1231 dev_put(queue->dev);
1234 static struct kobj_type netdev_queue_ktype = {
1235 .sysfs_ops = &netdev_queue_sysfs_ops,
1236 .release = netdev_queue_release,
1237 .default_attrs = netdev_queue_default_attrs,
1240 static int netdev_queue_add_kobject(struct net_device *net, int index)
1242 struct netdev_queue *queue = net->_tx + index;
1243 struct kobject *kobj = &queue->kobj;
1244 int error = 0;
1246 kobj->kset = net->queues_kset;
1247 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1248 "tx-%u", index);
1249 if (error)
1250 goto exit;
1252 #ifdef CONFIG_BQL
1253 error = sysfs_create_group(kobj, &dql_group);
1254 if (error)
1255 goto exit;
1256 #endif
1258 kobject_uevent(kobj, KOBJ_ADD);
1259 dev_hold(queue->dev);
1261 return 0;
1262 exit:
1263 kobject_put(kobj);
1264 return error;
1266 #endif /* CONFIG_SYSFS */
1269 netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1271 #ifdef CONFIG_SYSFS
1272 int i;
1273 int error = 0;
1275 for (i = old_num; i < new_num; i++) {
1276 error = netdev_queue_add_kobject(net, i);
1277 if (error) {
1278 new_num = old_num;
1279 break;
1283 while (--i >= new_num) {
1284 struct netdev_queue *queue = net->_tx + i;
1286 #ifdef CONFIG_BQL
1287 sysfs_remove_group(&queue->kobj, &dql_group);
1288 #endif
1289 kobject_put(&queue->kobj);
1292 return error;
1293 #else
1294 return 0;
1295 #endif /* CONFIG_SYSFS */
1298 static int register_queue_kobjects(struct net_device *net)
1300 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1302 #ifdef CONFIG_SYSFS
1303 net->queues_kset = kset_create_and_add("queues",
1304 NULL, &net->dev.kobj);
1305 if (!net->queues_kset)
1306 return -ENOMEM;
1307 #endif
1309 #ifdef CONFIG_RPS
1310 real_rx = net->real_num_rx_queues;
1311 #endif
1312 real_tx = net->real_num_tx_queues;
1314 error = net_rx_queue_update_kobjects(net, 0, real_rx);
1315 if (error)
1316 goto error;
1317 rxq = real_rx;
1319 error = netdev_queue_update_kobjects(net, 0, real_tx);
1320 if (error)
1321 goto error;
1322 txq = real_tx;
1324 return 0;
1326 error:
1327 netdev_queue_update_kobjects(net, txq, 0);
1328 net_rx_queue_update_kobjects(net, rxq, 0);
1329 return error;
1332 static void remove_queue_kobjects(struct net_device *net)
1334 int real_rx = 0, real_tx = 0;
1336 #ifdef CONFIG_RPS
1337 real_rx = net->real_num_rx_queues;
1338 #endif
1339 real_tx = net->real_num_tx_queues;
1341 net_rx_queue_update_kobjects(net, real_rx, 0);
1342 netdev_queue_update_kobjects(net, real_tx, 0);
1343 #ifdef CONFIG_SYSFS
1344 kset_unregister(net->queues_kset);
1345 #endif
1348 static void *net_grab_current_ns(void)
1350 struct net *ns = current->nsproxy->net_ns;
1351 #ifdef CONFIG_NET_NS
1352 if (ns)
1353 atomic_inc(&ns->passive);
1354 #endif
1355 return ns;
1358 static const void *net_initial_ns(void)
1360 return &init_net;
1363 static const void *net_netlink_ns(struct sock *sk)
1365 return sock_net(sk);
1368 struct kobj_ns_type_operations net_ns_type_operations = {
1369 .type = KOBJ_NS_TYPE_NET,
1370 .grab_current_ns = net_grab_current_ns,
1371 .netlink_ns = net_netlink_ns,
1372 .initial_ns = net_initial_ns,
1373 .drop_ns = net_drop_ns,
1375 EXPORT_SYMBOL_GPL(net_ns_type_operations);
1377 #ifdef CONFIG_HOTPLUG
1378 static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1380 struct net_device *dev = to_net_dev(d);
1381 int retval;
1383 /* pass interface to uevent. */
1384 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1385 if (retval)
1386 goto exit;
1388 /* pass ifindex to uevent.
1389 * ifindex is useful as it won't change (interface name may change)
1390 * and is what RtNetlink uses natively. */
1391 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1393 exit:
1394 return retval;
1396 #endif
1399 * netdev_release -- destroy and free a dead device.
1400 * Called when last reference to device kobject is gone.
1402 static void netdev_release(struct device *d)
1404 struct net_device *dev = to_net_dev(d);
1406 BUG_ON(dev->reg_state != NETREG_RELEASED);
1408 kfree(dev->ifalias);
1409 kfree((char *)dev - dev->padded);
1412 static const void *net_namespace(struct device *d)
1414 struct net_device *dev;
1415 dev = container_of(d, struct net_device, dev);
1416 return dev_net(dev);
1419 static struct class net_class = {
1420 .name = "net",
1421 .dev_release = netdev_release,
1422 #ifdef CONFIG_SYSFS
1423 .dev_attrs = net_class_attributes,
1424 #endif /* CONFIG_SYSFS */
1425 #ifdef CONFIG_HOTPLUG
1426 .dev_uevent = netdev_uevent,
1427 #endif
1428 .ns_type = &net_ns_type_operations,
1429 .namespace = net_namespace,
1432 /* Delete sysfs entries but hold kobject reference until after all
1433 * netdev references are gone.
1435 void netdev_unregister_kobject(struct net_device * net)
1437 struct device *dev = &(net->dev);
1439 kobject_get(&dev->kobj);
1441 remove_queue_kobjects(net);
1443 device_del(dev);
1446 /* Create sysfs entries for network device. */
1447 int netdev_register_kobject(struct net_device *net)
1449 struct device *dev = &(net->dev);
1450 const struct attribute_group **groups = net->sysfs_groups;
1451 int error = 0;
1453 device_initialize(dev);
1454 dev->class = &net_class;
1455 dev->platform_data = net;
1456 dev->groups = groups;
1458 dev_set_name(dev, "%s", net->name);
1460 #ifdef CONFIG_SYSFS
1461 /* Allow for a device specific group */
1462 if (*groups)
1463 groups++;
1465 *groups++ = &netstat_group;
1466 #ifdef CONFIG_WIRELESS_EXT_SYSFS
1467 if (net->ieee80211_ptr)
1468 *groups++ = &wireless_group;
1469 #ifdef CONFIG_WIRELESS_EXT
1470 else if (net->wireless_handlers)
1471 *groups++ = &wireless_group;
1472 #endif
1473 #endif
1474 #endif /* CONFIG_SYSFS */
1476 error = device_add(dev);
1477 if (error)
1478 return error;
1480 error = register_queue_kobjects(net);
1481 if (error) {
1482 device_del(dev);
1483 return error;
1486 return error;
1489 int netdev_class_create_file(struct class_attribute *class_attr)
1491 return class_create_file(&net_class, class_attr);
1493 EXPORT_SYMBOL(netdev_class_create_file);
1495 void netdev_class_remove_file(struct class_attribute *class_attr)
1497 class_remove_file(&net_class, class_attr);
1499 EXPORT_SYMBOL(netdev_class_remove_file);
1501 int netdev_kobject_init(void)
1503 kobj_ns_type_register(&net_ns_type_operations);
1504 return class_register(&net_class);