Loading...
1/*
2 * net-sysfs.c - network device class and attributes
3 *
4 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
5 *
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.
10 */
11
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/vmalloc.h>
22#include <linux/export.h>
23#include <linux/jiffies.h>
24#include <linux/pm_runtime.h>
25
26#include "net-sysfs.h"
27
28#ifdef CONFIG_SYSFS
29static const char fmt_hex[] = "%#x\n";
30static const char fmt_long_hex[] = "%#lx\n";
31static const char fmt_dec[] = "%d\n";
32static const char fmt_udec[] = "%u\n";
33static const char fmt_ulong[] = "%lu\n";
34static const char fmt_u64[] = "%llu\n";
35
36static inline int dev_isalive(const struct net_device *dev)
37{
38 return dev->reg_state <= NETREG_REGISTERED;
39}
40
41/* use same locking rules as GIF* ioctl's */
42static ssize_t netdev_show(const struct device *dev,
43 struct device_attribute *attr, char *buf,
44 ssize_t (*format)(const struct net_device *, char *))
45{
46 struct net_device *net = to_net_dev(dev);
47 ssize_t ret = -EINVAL;
48
49 read_lock(&dev_base_lock);
50 if (dev_isalive(net))
51 ret = (*format)(net, buf);
52 read_unlock(&dev_base_lock);
53
54 return ret;
55}
56
57/* generate a show function for simple field */
58#define NETDEVICE_SHOW(field, format_string) \
59static ssize_t format_##field(const struct net_device *net, char *buf) \
60{ \
61 return sprintf(buf, format_string, net->field); \
62} \
63static ssize_t field##_show(struct device *dev, \
64 struct device_attribute *attr, char *buf) \
65{ \
66 return netdev_show(dev, attr, buf, format_##field); \
67} \
68
69#define NETDEVICE_SHOW_RO(field, format_string) \
70NETDEVICE_SHOW(field, format_string); \
71static DEVICE_ATTR_RO(field)
72
73#define NETDEVICE_SHOW_RW(field, format_string) \
74NETDEVICE_SHOW(field, format_string); \
75static DEVICE_ATTR_RW(field)
76
77/* use same locking and permission rules as SIF* ioctl's */
78static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
79 const char *buf, size_t len,
80 int (*set)(struct net_device *, unsigned long))
81{
82 struct net_device *netdev = to_net_dev(dev);
83 struct net *net = dev_net(netdev);
84 unsigned long new;
85 int ret = -EINVAL;
86
87 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
88 return -EPERM;
89
90 ret = kstrtoul(buf, 0, &new);
91 if (ret)
92 goto err;
93
94 if (!rtnl_trylock())
95 return restart_syscall();
96
97 if (dev_isalive(netdev)) {
98 if ((ret = (*set)(netdev, new)) == 0)
99 ret = len;
100 }
101 rtnl_unlock();
102 err:
103 return ret;
104}
105
106NETDEVICE_SHOW_RO(dev_id, fmt_hex);
107NETDEVICE_SHOW_RO(dev_port, fmt_dec);
108NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
109NETDEVICE_SHOW_RO(addr_len, fmt_dec);
110NETDEVICE_SHOW_RO(iflink, fmt_dec);
111NETDEVICE_SHOW_RO(ifindex, fmt_dec);
112NETDEVICE_SHOW_RO(type, fmt_dec);
113NETDEVICE_SHOW_RO(link_mode, fmt_dec);
114
115/* use same locking rules as GIFHWADDR ioctl's */
116static ssize_t address_show(struct device *dev, struct device_attribute *attr,
117 char *buf)
118{
119 struct net_device *net = to_net_dev(dev);
120 ssize_t ret = -EINVAL;
121
122 read_lock(&dev_base_lock);
123 if (dev_isalive(net))
124 ret = sysfs_format_mac(buf, net->dev_addr, net->addr_len);
125 read_unlock(&dev_base_lock);
126 return ret;
127}
128static DEVICE_ATTR_RO(address);
129
130static ssize_t broadcast_show(struct device *dev,
131 struct device_attribute *attr, char *buf)
132{
133 struct net_device *net = to_net_dev(dev);
134 if (dev_isalive(net))
135 return sysfs_format_mac(buf, net->broadcast, net->addr_len);
136 return -EINVAL;
137}
138static DEVICE_ATTR_RO(broadcast);
139
140static int change_carrier(struct net_device *net, unsigned long new_carrier)
141{
142 if (!netif_running(net))
143 return -EINVAL;
144 return dev_change_carrier(net, (bool) new_carrier);
145}
146
147static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
148 const char *buf, size_t len)
149{
150 return netdev_store(dev, attr, buf, len, change_carrier);
151}
152
153static ssize_t carrier_show(struct device *dev,
154 struct device_attribute *attr, char *buf)
155{
156 struct net_device *netdev = to_net_dev(dev);
157 if (netif_running(netdev)) {
158 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
159 }
160 return -EINVAL;
161}
162static DEVICE_ATTR_RW(carrier);
163
164static ssize_t speed_show(struct device *dev,
165 struct device_attribute *attr, char *buf)
166{
167 struct net_device *netdev = to_net_dev(dev);
168 int ret = -EINVAL;
169
170 if (!rtnl_trylock())
171 return restart_syscall();
172
173 if (netif_running(netdev)) {
174 struct ethtool_cmd cmd;
175 if (!__ethtool_get_settings(netdev, &cmd))
176 ret = sprintf(buf, fmt_udec, ethtool_cmd_speed(&cmd));
177 }
178 rtnl_unlock();
179 return ret;
180}
181static DEVICE_ATTR_RO(speed);
182
183static ssize_t duplex_show(struct device *dev,
184 struct device_attribute *attr, char *buf)
185{
186 struct net_device *netdev = to_net_dev(dev);
187 int ret = -EINVAL;
188
189 if (!rtnl_trylock())
190 return restart_syscall();
191
192 if (netif_running(netdev)) {
193 struct ethtool_cmd cmd;
194 if (!__ethtool_get_settings(netdev, &cmd)) {
195 const char *duplex;
196 switch (cmd.duplex) {
197 case DUPLEX_HALF:
198 duplex = "half";
199 break;
200 case DUPLEX_FULL:
201 duplex = "full";
202 break;
203 default:
204 duplex = "unknown";
205 break;
206 }
207 ret = sprintf(buf, "%s\n", duplex);
208 }
209 }
210 rtnl_unlock();
211 return ret;
212}
213static DEVICE_ATTR_RO(duplex);
214
215static ssize_t dormant_show(struct device *dev,
216 struct device_attribute *attr, char *buf)
217{
218 struct net_device *netdev = to_net_dev(dev);
219
220 if (netif_running(netdev))
221 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
222
223 return -EINVAL;
224}
225static DEVICE_ATTR_RO(dormant);
226
227static const char *const operstates[] = {
228 "unknown",
229 "notpresent", /* currently unused */
230 "down",
231 "lowerlayerdown",
232 "testing", /* currently unused */
233 "dormant",
234 "up"
235};
236
237static ssize_t operstate_show(struct device *dev,
238 struct device_attribute *attr, char *buf)
239{
240 const struct net_device *netdev = to_net_dev(dev);
241 unsigned char operstate;
242
243 read_lock(&dev_base_lock);
244 operstate = netdev->operstate;
245 if (!netif_running(netdev))
246 operstate = IF_OPER_DOWN;
247 read_unlock(&dev_base_lock);
248
249 if (operstate >= ARRAY_SIZE(operstates))
250 return -EINVAL; /* should not happen */
251
252 return sprintf(buf, "%s\n", operstates[operstate]);
253}
254static DEVICE_ATTR_RO(operstate);
255
256static ssize_t carrier_changes_show(struct device *dev,
257 struct device_attribute *attr,
258 char *buf)
259{
260 struct net_device *netdev = to_net_dev(dev);
261 return sprintf(buf, fmt_dec,
262 atomic_read(&netdev->carrier_changes));
263}
264static DEVICE_ATTR_RO(carrier_changes);
265
266/* read-write attributes */
267
268static int change_mtu(struct net_device *net, unsigned long new_mtu)
269{
270 return dev_set_mtu(net, (int) new_mtu);
271}
272
273static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
274 const char *buf, size_t len)
275{
276 return netdev_store(dev, attr, buf, len, change_mtu);
277}
278NETDEVICE_SHOW_RW(mtu, fmt_dec);
279
280static int change_flags(struct net_device *net, unsigned long new_flags)
281{
282 return dev_change_flags(net, (unsigned int) new_flags);
283}
284
285static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
286 const char *buf, size_t len)
287{
288 return netdev_store(dev, attr, buf, len, change_flags);
289}
290NETDEVICE_SHOW_RW(flags, fmt_hex);
291
292static int change_tx_queue_len(struct net_device *net, unsigned long new_len)
293{
294 net->tx_queue_len = new_len;
295 return 0;
296}
297
298static ssize_t tx_queue_len_store(struct device *dev,
299 struct device_attribute *attr,
300 const char *buf, size_t len)
301{
302 if (!capable(CAP_NET_ADMIN))
303 return -EPERM;
304
305 return netdev_store(dev, attr, buf, len, change_tx_queue_len);
306}
307NETDEVICE_SHOW_RW(tx_queue_len, fmt_ulong);
308
309static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
310 const char *buf, size_t len)
311{
312 struct net_device *netdev = to_net_dev(dev);
313 struct net *net = dev_net(netdev);
314 size_t count = len;
315 ssize_t ret;
316
317 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
318 return -EPERM;
319
320 /* ignore trailing newline */
321 if (len > 0 && buf[len - 1] == '\n')
322 --count;
323
324 if (!rtnl_trylock())
325 return restart_syscall();
326 ret = dev_set_alias(netdev, buf, count);
327 rtnl_unlock();
328
329 return ret < 0 ? ret : len;
330}
331
332static ssize_t ifalias_show(struct device *dev,
333 struct device_attribute *attr, char *buf)
334{
335 const struct net_device *netdev = to_net_dev(dev);
336 ssize_t ret = 0;
337
338 if (!rtnl_trylock())
339 return restart_syscall();
340 if (netdev->ifalias)
341 ret = sprintf(buf, "%s\n", netdev->ifalias);
342 rtnl_unlock();
343 return ret;
344}
345static DEVICE_ATTR_RW(ifalias);
346
347static int change_group(struct net_device *net, unsigned long new_group)
348{
349 dev_set_group(net, (int) new_group);
350 return 0;
351}
352
353static ssize_t group_store(struct device *dev, struct device_attribute *attr,
354 const char *buf, size_t len)
355{
356 return netdev_store(dev, attr, buf, len, change_group);
357}
358NETDEVICE_SHOW(group, fmt_dec);
359static DEVICE_ATTR(netdev_group, S_IRUGO | S_IWUSR, group_show, group_store);
360
361static ssize_t phys_port_id_show(struct device *dev,
362 struct device_attribute *attr, char *buf)
363{
364 struct net_device *netdev = to_net_dev(dev);
365 ssize_t ret = -EINVAL;
366
367 if (!rtnl_trylock())
368 return restart_syscall();
369
370 if (dev_isalive(netdev)) {
371 struct netdev_phys_port_id ppid;
372
373 ret = dev_get_phys_port_id(netdev, &ppid);
374 if (!ret)
375 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
376 }
377 rtnl_unlock();
378
379 return ret;
380}
381static DEVICE_ATTR_RO(phys_port_id);
382
383static struct attribute *net_class_attrs[] = {
384 &dev_attr_netdev_group.attr,
385 &dev_attr_type.attr,
386 &dev_attr_dev_id.attr,
387 &dev_attr_dev_port.attr,
388 &dev_attr_iflink.attr,
389 &dev_attr_ifindex.attr,
390 &dev_attr_addr_assign_type.attr,
391 &dev_attr_addr_len.attr,
392 &dev_attr_link_mode.attr,
393 &dev_attr_address.attr,
394 &dev_attr_broadcast.attr,
395 &dev_attr_speed.attr,
396 &dev_attr_duplex.attr,
397 &dev_attr_dormant.attr,
398 &dev_attr_operstate.attr,
399 &dev_attr_carrier_changes.attr,
400 &dev_attr_ifalias.attr,
401 &dev_attr_carrier.attr,
402 &dev_attr_mtu.attr,
403 &dev_attr_flags.attr,
404 &dev_attr_tx_queue_len.attr,
405 &dev_attr_phys_port_id.attr,
406 NULL,
407};
408ATTRIBUTE_GROUPS(net_class);
409
410/* Show a given an attribute in the statistics group */
411static ssize_t netstat_show(const struct device *d,
412 struct device_attribute *attr, char *buf,
413 unsigned long offset)
414{
415 struct net_device *dev = to_net_dev(d);
416 ssize_t ret = -EINVAL;
417
418 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
419 offset % sizeof(u64) != 0);
420
421 read_lock(&dev_base_lock);
422 if (dev_isalive(dev)) {
423 struct rtnl_link_stats64 temp;
424 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
425
426 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *) stats) + offset));
427 }
428 read_unlock(&dev_base_lock);
429 return ret;
430}
431
432/* generate a read-only statistics attribute */
433#define NETSTAT_ENTRY(name) \
434static ssize_t name##_show(struct device *d, \
435 struct device_attribute *attr, char *buf) \
436{ \
437 return netstat_show(d, attr, buf, \
438 offsetof(struct rtnl_link_stats64, name)); \
439} \
440static DEVICE_ATTR_RO(name)
441
442NETSTAT_ENTRY(rx_packets);
443NETSTAT_ENTRY(tx_packets);
444NETSTAT_ENTRY(rx_bytes);
445NETSTAT_ENTRY(tx_bytes);
446NETSTAT_ENTRY(rx_errors);
447NETSTAT_ENTRY(tx_errors);
448NETSTAT_ENTRY(rx_dropped);
449NETSTAT_ENTRY(tx_dropped);
450NETSTAT_ENTRY(multicast);
451NETSTAT_ENTRY(collisions);
452NETSTAT_ENTRY(rx_length_errors);
453NETSTAT_ENTRY(rx_over_errors);
454NETSTAT_ENTRY(rx_crc_errors);
455NETSTAT_ENTRY(rx_frame_errors);
456NETSTAT_ENTRY(rx_fifo_errors);
457NETSTAT_ENTRY(rx_missed_errors);
458NETSTAT_ENTRY(tx_aborted_errors);
459NETSTAT_ENTRY(tx_carrier_errors);
460NETSTAT_ENTRY(tx_fifo_errors);
461NETSTAT_ENTRY(tx_heartbeat_errors);
462NETSTAT_ENTRY(tx_window_errors);
463NETSTAT_ENTRY(rx_compressed);
464NETSTAT_ENTRY(tx_compressed);
465
466static struct attribute *netstat_attrs[] = {
467 &dev_attr_rx_packets.attr,
468 &dev_attr_tx_packets.attr,
469 &dev_attr_rx_bytes.attr,
470 &dev_attr_tx_bytes.attr,
471 &dev_attr_rx_errors.attr,
472 &dev_attr_tx_errors.attr,
473 &dev_attr_rx_dropped.attr,
474 &dev_attr_tx_dropped.attr,
475 &dev_attr_multicast.attr,
476 &dev_attr_collisions.attr,
477 &dev_attr_rx_length_errors.attr,
478 &dev_attr_rx_over_errors.attr,
479 &dev_attr_rx_crc_errors.attr,
480 &dev_attr_rx_frame_errors.attr,
481 &dev_attr_rx_fifo_errors.attr,
482 &dev_attr_rx_missed_errors.attr,
483 &dev_attr_tx_aborted_errors.attr,
484 &dev_attr_tx_carrier_errors.attr,
485 &dev_attr_tx_fifo_errors.attr,
486 &dev_attr_tx_heartbeat_errors.attr,
487 &dev_attr_tx_window_errors.attr,
488 &dev_attr_rx_compressed.attr,
489 &dev_attr_tx_compressed.attr,
490 NULL
491};
492
493
494static struct attribute_group netstat_group = {
495 .name = "statistics",
496 .attrs = netstat_attrs,
497};
498
499#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
500static struct attribute *wireless_attrs[] = {
501 NULL
502};
503
504static struct attribute_group wireless_group = {
505 .name = "wireless",
506 .attrs = wireless_attrs,
507};
508#endif
509
510#else /* CONFIG_SYSFS */
511#define net_class_groups NULL
512#endif /* CONFIG_SYSFS */
513
514#ifdef CONFIG_SYSFS
515#define to_rx_queue_attr(_attr) container_of(_attr, \
516 struct rx_queue_attribute, attr)
517
518#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
519
520static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
521 char *buf)
522{
523 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
524 struct netdev_rx_queue *queue = to_rx_queue(kobj);
525
526 if (!attribute->show)
527 return -EIO;
528
529 return attribute->show(queue, attribute, buf);
530}
531
532static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
533 const char *buf, size_t count)
534{
535 struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
536 struct netdev_rx_queue *queue = to_rx_queue(kobj);
537
538 if (!attribute->store)
539 return -EIO;
540
541 return attribute->store(queue, attribute, buf, count);
542}
543
544static const struct sysfs_ops rx_queue_sysfs_ops = {
545 .show = rx_queue_attr_show,
546 .store = rx_queue_attr_store,
547};
548
549#ifdef CONFIG_RPS
550static ssize_t show_rps_map(struct netdev_rx_queue *queue,
551 struct rx_queue_attribute *attribute, char *buf)
552{
553 struct rps_map *map;
554 cpumask_var_t mask;
555 size_t len = 0;
556 int i;
557
558 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
559 return -ENOMEM;
560
561 rcu_read_lock();
562 map = rcu_dereference(queue->rps_map);
563 if (map)
564 for (i = 0; i < map->len; i++)
565 cpumask_set_cpu(map->cpus[i], mask);
566
567 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
568 if (PAGE_SIZE - len < 3) {
569 rcu_read_unlock();
570 free_cpumask_var(mask);
571 return -EINVAL;
572 }
573 rcu_read_unlock();
574
575 free_cpumask_var(mask);
576 len += sprintf(buf + len, "\n");
577 return len;
578}
579
580static ssize_t store_rps_map(struct netdev_rx_queue *queue,
581 struct rx_queue_attribute *attribute,
582 const char *buf, size_t len)
583{
584 struct rps_map *old_map, *map;
585 cpumask_var_t mask;
586 int err, cpu, i;
587 static DEFINE_SPINLOCK(rps_map_lock);
588
589 if (!capable(CAP_NET_ADMIN))
590 return -EPERM;
591
592 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
593 return -ENOMEM;
594
595 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
596 if (err) {
597 free_cpumask_var(mask);
598 return err;
599 }
600
601 map = kzalloc(max_t(unsigned int,
602 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
603 GFP_KERNEL);
604 if (!map) {
605 free_cpumask_var(mask);
606 return -ENOMEM;
607 }
608
609 i = 0;
610 for_each_cpu_and(cpu, mask, cpu_online_mask)
611 map->cpus[i++] = cpu;
612
613 if (i)
614 map->len = i;
615 else {
616 kfree(map);
617 map = NULL;
618 }
619
620 spin_lock(&rps_map_lock);
621 old_map = rcu_dereference_protected(queue->rps_map,
622 lockdep_is_held(&rps_map_lock));
623 rcu_assign_pointer(queue->rps_map, map);
624 spin_unlock(&rps_map_lock);
625
626 if (map)
627 static_key_slow_inc(&rps_needed);
628 if (old_map) {
629 kfree_rcu(old_map, rcu);
630 static_key_slow_dec(&rps_needed);
631 }
632 free_cpumask_var(mask);
633 return len;
634}
635
636static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
637 struct rx_queue_attribute *attr,
638 char *buf)
639{
640 struct rps_dev_flow_table *flow_table;
641 unsigned long val = 0;
642
643 rcu_read_lock();
644 flow_table = rcu_dereference(queue->rps_flow_table);
645 if (flow_table)
646 val = (unsigned long)flow_table->mask + 1;
647 rcu_read_unlock();
648
649 return sprintf(buf, "%lu\n", val);
650}
651
652static void rps_dev_flow_table_release(struct rcu_head *rcu)
653{
654 struct rps_dev_flow_table *table = container_of(rcu,
655 struct rps_dev_flow_table, rcu);
656 vfree(table);
657}
658
659static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
660 struct rx_queue_attribute *attr,
661 const char *buf, size_t len)
662{
663 unsigned long mask, count;
664 struct rps_dev_flow_table *table, *old_table;
665 static DEFINE_SPINLOCK(rps_dev_flow_lock);
666 int rc;
667
668 if (!capable(CAP_NET_ADMIN))
669 return -EPERM;
670
671 rc = kstrtoul(buf, 0, &count);
672 if (rc < 0)
673 return rc;
674
675 if (count) {
676 mask = count - 1;
677 /* mask = roundup_pow_of_two(count) - 1;
678 * without overflows...
679 */
680 while ((mask | (mask >> 1)) != mask)
681 mask |= (mask >> 1);
682 /* On 64 bit arches, must check mask fits in table->mask (u32),
683 * and on 32bit arches, must check
684 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
685 */
686#if BITS_PER_LONG > 32
687 if (mask > (unsigned long)(u32)mask)
688 return -EINVAL;
689#else
690 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
691 / sizeof(struct rps_dev_flow)) {
692 /* Enforce a limit to prevent overflow */
693 return -EINVAL;
694 }
695#endif
696 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
697 if (!table)
698 return -ENOMEM;
699
700 table->mask = mask;
701 for (count = 0; count <= mask; count++)
702 table->flows[count].cpu = RPS_NO_CPU;
703 } else
704 table = NULL;
705
706 spin_lock(&rps_dev_flow_lock);
707 old_table = rcu_dereference_protected(queue->rps_flow_table,
708 lockdep_is_held(&rps_dev_flow_lock));
709 rcu_assign_pointer(queue->rps_flow_table, table);
710 spin_unlock(&rps_dev_flow_lock);
711
712 if (old_table)
713 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
714
715 return len;
716}
717
718static struct rx_queue_attribute rps_cpus_attribute =
719 __ATTR(rps_cpus, S_IRUGO | S_IWUSR, show_rps_map, store_rps_map);
720
721
722static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute =
723 __ATTR(rps_flow_cnt, S_IRUGO | S_IWUSR,
724 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
725#endif /* CONFIG_RPS */
726
727static struct attribute *rx_queue_default_attrs[] = {
728#ifdef CONFIG_RPS
729 &rps_cpus_attribute.attr,
730 &rps_dev_flow_table_cnt_attribute.attr,
731#endif
732 NULL
733};
734
735static void rx_queue_release(struct kobject *kobj)
736{
737 struct netdev_rx_queue *queue = to_rx_queue(kobj);
738#ifdef CONFIG_RPS
739 struct rps_map *map;
740 struct rps_dev_flow_table *flow_table;
741
742
743 map = rcu_dereference_protected(queue->rps_map, 1);
744 if (map) {
745 RCU_INIT_POINTER(queue->rps_map, NULL);
746 kfree_rcu(map, rcu);
747 }
748
749 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
750 if (flow_table) {
751 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
752 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
753 }
754#endif
755
756 memset(kobj, 0, sizeof(*kobj));
757 dev_put(queue->dev);
758}
759
760static const void *rx_queue_namespace(struct kobject *kobj)
761{
762 struct netdev_rx_queue *queue = to_rx_queue(kobj);
763 struct device *dev = &queue->dev->dev;
764 const void *ns = NULL;
765
766 if (dev->class && dev->class->ns_type)
767 ns = dev->class->namespace(dev);
768
769 return ns;
770}
771
772static struct kobj_type rx_queue_ktype = {
773 .sysfs_ops = &rx_queue_sysfs_ops,
774 .release = rx_queue_release,
775 .default_attrs = rx_queue_default_attrs,
776 .namespace = rx_queue_namespace
777};
778
779static int rx_queue_add_kobject(struct net_device *net, int index)
780{
781 struct netdev_rx_queue *queue = net->_rx + index;
782 struct kobject *kobj = &queue->kobj;
783 int error = 0;
784
785 kobj->kset = net->queues_kset;
786 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
787 "rx-%u", index);
788 if (error)
789 goto exit;
790
791 if (net->sysfs_rx_queue_group) {
792 error = sysfs_create_group(kobj, net->sysfs_rx_queue_group);
793 if (error)
794 goto exit;
795 }
796
797 kobject_uevent(kobj, KOBJ_ADD);
798 dev_hold(queue->dev);
799
800 return error;
801exit:
802 kobject_put(kobj);
803 return error;
804}
805#endif /* CONFIG_SYSFS */
806
807int
808net_rx_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
809{
810#ifdef CONFIG_SYSFS
811 int i;
812 int error = 0;
813
814#ifndef CONFIG_RPS
815 if (!net->sysfs_rx_queue_group)
816 return 0;
817#endif
818 for (i = old_num; i < new_num; i++) {
819 error = rx_queue_add_kobject(net, i);
820 if (error) {
821 new_num = old_num;
822 break;
823 }
824 }
825
826 while (--i >= new_num) {
827 if (net->sysfs_rx_queue_group)
828 sysfs_remove_group(&net->_rx[i].kobj,
829 net->sysfs_rx_queue_group);
830 kobject_put(&net->_rx[i].kobj);
831 }
832
833 return error;
834#else
835 return 0;
836#endif
837}
838
839#ifdef CONFIG_SYSFS
840/*
841 * netdev_queue sysfs structures and functions.
842 */
843struct netdev_queue_attribute {
844 struct attribute attr;
845 ssize_t (*show)(struct netdev_queue *queue,
846 struct netdev_queue_attribute *attr, char *buf);
847 ssize_t (*store)(struct netdev_queue *queue,
848 struct netdev_queue_attribute *attr, const char *buf, size_t len);
849};
850#define to_netdev_queue_attr(_attr) container_of(_attr, \
851 struct netdev_queue_attribute, attr)
852
853#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
854
855static ssize_t netdev_queue_attr_show(struct kobject *kobj,
856 struct attribute *attr, char *buf)
857{
858 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
859 struct netdev_queue *queue = to_netdev_queue(kobj);
860
861 if (!attribute->show)
862 return -EIO;
863
864 return attribute->show(queue, attribute, buf);
865}
866
867static ssize_t netdev_queue_attr_store(struct kobject *kobj,
868 struct attribute *attr,
869 const char *buf, size_t count)
870{
871 struct netdev_queue_attribute *attribute = to_netdev_queue_attr(attr);
872 struct netdev_queue *queue = to_netdev_queue(kobj);
873
874 if (!attribute->store)
875 return -EIO;
876
877 return attribute->store(queue, attribute, buf, count);
878}
879
880static const struct sysfs_ops netdev_queue_sysfs_ops = {
881 .show = netdev_queue_attr_show,
882 .store = netdev_queue_attr_store,
883};
884
885static ssize_t show_trans_timeout(struct netdev_queue *queue,
886 struct netdev_queue_attribute *attribute,
887 char *buf)
888{
889 unsigned long trans_timeout;
890
891 spin_lock_irq(&queue->_xmit_lock);
892 trans_timeout = queue->trans_timeout;
893 spin_unlock_irq(&queue->_xmit_lock);
894
895 return sprintf(buf, "%lu", trans_timeout);
896}
897
898static struct netdev_queue_attribute queue_trans_timeout =
899 __ATTR(tx_timeout, S_IRUGO, show_trans_timeout, NULL);
900
901#ifdef CONFIG_BQL
902/*
903 * Byte queue limits sysfs structures and functions.
904 */
905static ssize_t bql_show(char *buf, unsigned int value)
906{
907 return sprintf(buf, "%u\n", value);
908}
909
910static ssize_t bql_set(const char *buf, const size_t count,
911 unsigned int *pvalue)
912{
913 unsigned int value;
914 int err;
915
916 if (!strcmp(buf, "max") || !strcmp(buf, "max\n"))
917 value = DQL_MAX_LIMIT;
918 else {
919 err = kstrtouint(buf, 10, &value);
920 if (err < 0)
921 return err;
922 if (value > DQL_MAX_LIMIT)
923 return -EINVAL;
924 }
925
926 *pvalue = value;
927
928 return count;
929}
930
931static ssize_t bql_show_hold_time(struct netdev_queue *queue,
932 struct netdev_queue_attribute *attr,
933 char *buf)
934{
935 struct dql *dql = &queue->dql;
936
937 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
938}
939
940static ssize_t bql_set_hold_time(struct netdev_queue *queue,
941 struct netdev_queue_attribute *attribute,
942 const char *buf, size_t len)
943{
944 struct dql *dql = &queue->dql;
945 unsigned int value;
946 int err;
947
948 err = kstrtouint(buf, 10, &value);
949 if (err < 0)
950 return err;
951
952 dql->slack_hold_time = msecs_to_jiffies(value);
953
954 return len;
955}
956
957static struct netdev_queue_attribute bql_hold_time_attribute =
958 __ATTR(hold_time, S_IRUGO | S_IWUSR, bql_show_hold_time,
959 bql_set_hold_time);
960
961static ssize_t bql_show_inflight(struct netdev_queue *queue,
962 struct netdev_queue_attribute *attr,
963 char *buf)
964{
965 struct dql *dql = &queue->dql;
966
967 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
968}
969
970static struct netdev_queue_attribute bql_inflight_attribute =
971 __ATTR(inflight, S_IRUGO, bql_show_inflight, NULL);
972
973#define BQL_ATTR(NAME, FIELD) \
974static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
975 struct netdev_queue_attribute *attr, \
976 char *buf) \
977{ \
978 return bql_show(buf, queue->dql.FIELD); \
979} \
980 \
981static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
982 struct netdev_queue_attribute *attr, \
983 const char *buf, size_t len) \
984{ \
985 return bql_set(buf, len, &queue->dql.FIELD); \
986} \
987 \
988static struct netdev_queue_attribute bql_ ## NAME ## _attribute = \
989 __ATTR(NAME, S_IRUGO | S_IWUSR, bql_show_ ## NAME, \
990 bql_set_ ## NAME);
991
992BQL_ATTR(limit, limit)
993BQL_ATTR(limit_max, max_limit)
994BQL_ATTR(limit_min, min_limit)
995
996static struct attribute *dql_attrs[] = {
997 &bql_limit_attribute.attr,
998 &bql_limit_max_attribute.attr,
999 &bql_limit_min_attribute.attr,
1000 &bql_hold_time_attribute.attr,
1001 &bql_inflight_attribute.attr,
1002 NULL
1003};
1004
1005static struct attribute_group dql_group = {
1006 .name = "byte_queue_limits",
1007 .attrs = dql_attrs,
1008};
1009#endif /* CONFIG_BQL */
1010
1011#ifdef CONFIG_XPS
1012static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1013{
1014 struct net_device *dev = queue->dev;
1015 unsigned int i;
1016
1017 i = queue - dev->_tx;
1018 BUG_ON(i >= dev->num_tx_queues);
1019
1020 return i;
1021}
1022
1023
1024static ssize_t show_xps_map(struct netdev_queue *queue,
1025 struct netdev_queue_attribute *attribute, char *buf)
1026{
1027 struct net_device *dev = queue->dev;
1028 struct xps_dev_maps *dev_maps;
1029 cpumask_var_t mask;
1030 unsigned long index;
1031 size_t len = 0;
1032 int i;
1033
1034 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
1035 return -ENOMEM;
1036
1037 index = get_netdev_queue_index(queue);
1038
1039 rcu_read_lock();
1040 dev_maps = rcu_dereference(dev->xps_maps);
1041 if (dev_maps) {
1042 for_each_possible_cpu(i) {
1043 struct xps_map *map =
1044 rcu_dereference(dev_maps->cpu_map[i]);
1045 if (map) {
1046 int j;
1047 for (j = 0; j < map->len; j++) {
1048 if (map->queues[j] == index) {
1049 cpumask_set_cpu(i, mask);
1050 break;
1051 }
1052 }
1053 }
1054 }
1055 }
1056 rcu_read_unlock();
1057
1058 len += cpumask_scnprintf(buf + len, PAGE_SIZE, mask);
1059 if (PAGE_SIZE - len < 3) {
1060 free_cpumask_var(mask);
1061 return -EINVAL;
1062 }
1063
1064 free_cpumask_var(mask);
1065 len += sprintf(buf + len, "\n");
1066 return len;
1067}
1068
1069static ssize_t store_xps_map(struct netdev_queue *queue,
1070 struct netdev_queue_attribute *attribute,
1071 const char *buf, size_t len)
1072{
1073 struct net_device *dev = queue->dev;
1074 unsigned long index;
1075 cpumask_var_t mask;
1076 int err;
1077
1078 if (!capable(CAP_NET_ADMIN))
1079 return -EPERM;
1080
1081 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1082 return -ENOMEM;
1083
1084 index = get_netdev_queue_index(queue);
1085
1086 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1087 if (err) {
1088 free_cpumask_var(mask);
1089 return err;
1090 }
1091
1092 err = netif_set_xps_queue(dev, mask, index);
1093
1094 free_cpumask_var(mask);
1095
1096 return err ? : len;
1097}
1098
1099static struct netdev_queue_attribute xps_cpus_attribute =
1100 __ATTR(xps_cpus, S_IRUGO | S_IWUSR, show_xps_map, store_xps_map);
1101#endif /* CONFIG_XPS */
1102
1103static struct attribute *netdev_queue_default_attrs[] = {
1104 &queue_trans_timeout.attr,
1105#ifdef CONFIG_XPS
1106 &xps_cpus_attribute.attr,
1107#endif
1108 NULL
1109};
1110
1111static void netdev_queue_release(struct kobject *kobj)
1112{
1113 struct netdev_queue *queue = to_netdev_queue(kobj);
1114
1115 memset(kobj, 0, sizeof(*kobj));
1116 dev_put(queue->dev);
1117}
1118
1119static const void *netdev_queue_namespace(struct kobject *kobj)
1120{
1121 struct netdev_queue *queue = to_netdev_queue(kobj);
1122 struct device *dev = &queue->dev->dev;
1123 const void *ns = NULL;
1124
1125 if (dev->class && dev->class->ns_type)
1126 ns = dev->class->namespace(dev);
1127
1128 return ns;
1129}
1130
1131static struct kobj_type netdev_queue_ktype = {
1132 .sysfs_ops = &netdev_queue_sysfs_ops,
1133 .release = netdev_queue_release,
1134 .default_attrs = netdev_queue_default_attrs,
1135 .namespace = netdev_queue_namespace,
1136};
1137
1138static int netdev_queue_add_kobject(struct net_device *net, int index)
1139{
1140 struct netdev_queue *queue = net->_tx + index;
1141 struct kobject *kobj = &queue->kobj;
1142 int error = 0;
1143
1144 kobj->kset = net->queues_kset;
1145 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1146 "tx-%u", index);
1147 if (error)
1148 goto exit;
1149
1150#ifdef CONFIG_BQL
1151 error = sysfs_create_group(kobj, &dql_group);
1152 if (error)
1153 goto exit;
1154#endif
1155
1156 kobject_uevent(kobj, KOBJ_ADD);
1157 dev_hold(queue->dev);
1158
1159 return 0;
1160exit:
1161 kobject_put(kobj);
1162 return error;
1163}
1164#endif /* CONFIG_SYSFS */
1165
1166int
1167netdev_queue_update_kobjects(struct net_device *net, int old_num, int new_num)
1168{
1169#ifdef CONFIG_SYSFS
1170 int i;
1171 int error = 0;
1172
1173 for (i = old_num; i < new_num; i++) {
1174 error = netdev_queue_add_kobject(net, i);
1175 if (error) {
1176 new_num = old_num;
1177 break;
1178 }
1179 }
1180
1181 while (--i >= new_num) {
1182 struct netdev_queue *queue = net->_tx + i;
1183
1184#ifdef CONFIG_BQL
1185 sysfs_remove_group(&queue->kobj, &dql_group);
1186#endif
1187 kobject_put(&queue->kobj);
1188 }
1189
1190 return error;
1191#else
1192 return 0;
1193#endif /* CONFIG_SYSFS */
1194}
1195
1196static int register_queue_kobjects(struct net_device *net)
1197{
1198 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1199
1200#ifdef CONFIG_SYSFS
1201 net->queues_kset = kset_create_and_add("queues",
1202 NULL, &net->dev.kobj);
1203 if (!net->queues_kset)
1204 return -ENOMEM;
1205 real_rx = net->real_num_rx_queues;
1206#endif
1207 real_tx = net->real_num_tx_queues;
1208
1209 error = net_rx_queue_update_kobjects(net, 0, real_rx);
1210 if (error)
1211 goto error;
1212 rxq = real_rx;
1213
1214 error = netdev_queue_update_kobjects(net, 0, real_tx);
1215 if (error)
1216 goto error;
1217 txq = real_tx;
1218
1219 return 0;
1220
1221error:
1222 netdev_queue_update_kobjects(net, txq, 0);
1223 net_rx_queue_update_kobjects(net, rxq, 0);
1224 return error;
1225}
1226
1227static void remove_queue_kobjects(struct net_device *net)
1228{
1229 int real_rx = 0, real_tx = 0;
1230
1231#ifdef CONFIG_SYSFS
1232 real_rx = net->real_num_rx_queues;
1233#endif
1234 real_tx = net->real_num_tx_queues;
1235
1236 net_rx_queue_update_kobjects(net, real_rx, 0);
1237 netdev_queue_update_kobjects(net, real_tx, 0);
1238#ifdef CONFIG_SYSFS
1239 kset_unregister(net->queues_kset);
1240#endif
1241}
1242
1243static bool net_current_may_mount(void)
1244{
1245 struct net *net = current->nsproxy->net_ns;
1246
1247 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1248}
1249
1250static void *net_grab_current_ns(void)
1251{
1252 struct net *ns = current->nsproxy->net_ns;
1253#ifdef CONFIG_NET_NS
1254 if (ns)
1255 atomic_inc(&ns->passive);
1256#endif
1257 return ns;
1258}
1259
1260static const void *net_initial_ns(void)
1261{
1262 return &init_net;
1263}
1264
1265static const void *net_netlink_ns(struct sock *sk)
1266{
1267 return sock_net(sk);
1268}
1269
1270struct kobj_ns_type_operations net_ns_type_operations = {
1271 .type = KOBJ_NS_TYPE_NET,
1272 .current_may_mount = net_current_may_mount,
1273 .grab_current_ns = net_grab_current_ns,
1274 .netlink_ns = net_netlink_ns,
1275 .initial_ns = net_initial_ns,
1276 .drop_ns = net_drop_ns,
1277};
1278EXPORT_SYMBOL_GPL(net_ns_type_operations);
1279
1280static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1281{
1282 struct net_device *dev = to_net_dev(d);
1283 int retval;
1284
1285 /* pass interface to uevent. */
1286 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1287 if (retval)
1288 goto exit;
1289
1290 /* pass ifindex to uevent.
1291 * ifindex is useful as it won't change (interface name may change)
1292 * and is what RtNetlink uses natively. */
1293 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1294
1295exit:
1296 return retval;
1297}
1298
1299/*
1300 * netdev_release -- destroy and free a dead device.
1301 * Called when last reference to device kobject is gone.
1302 */
1303static void netdev_release(struct device *d)
1304{
1305 struct net_device *dev = to_net_dev(d);
1306
1307 BUG_ON(dev->reg_state != NETREG_RELEASED);
1308
1309 kfree(dev->ifalias);
1310 netdev_freemem(dev);
1311}
1312
1313static const void *net_namespace(struct device *d)
1314{
1315 struct net_device *dev;
1316 dev = container_of(d, struct net_device, dev);
1317 return dev_net(dev);
1318}
1319
1320static struct class net_class = {
1321 .name = "net",
1322 .dev_release = netdev_release,
1323 .dev_groups = net_class_groups,
1324 .dev_uevent = netdev_uevent,
1325 .ns_type = &net_ns_type_operations,
1326 .namespace = net_namespace,
1327};
1328
1329/* Delete sysfs entries but hold kobject reference until after all
1330 * netdev references are gone.
1331 */
1332void netdev_unregister_kobject(struct net_device * net)
1333{
1334 struct device *dev = &(net->dev);
1335
1336 kobject_get(&dev->kobj);
1337
1338 remove_queue_kobjects(net);
1339
1340 pm_runtime_set_memalloc_noio(dev, false);
1341
1342 device_del(dev);
1343}
1344
1345/* Create sysfs entries for network device. */
1346int netdev_register_kobject(struct net_device *net)
1347{
1348 struct device *dev = &(net->dev);
1349 const struct attribute_group **groups = net->sysfs_groups;
1350 int error = 0;
1351
1352 device_initialize(dev);
1353 dev->class = &net_class;
1354 dev->platform_data = net;
1355 dev->groups = groups;
1356
1357 dev_set_name(dev, "%s", net->name);
1358
1359#ifdef CONFIG_SYSFS
1360 /* Allow for a device specific group */
1361 if (*groups)
1362 groups++;
1363
1364 *groups++ = &netstat_group;
1365
1366#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1367 if (net->ieee80211_ptr)
1368 *groups++ = &wireless_group;
1369#if IS_ENABLED(CONFIG_WIRELESS_EXT)
1370 else if (net->wireless_handlers)
1371 *groups++ = &wireless_group;
1372#endif
1373#endif
1374#endif /* CONFIG_SYSFS */
1375
1376 error = device_add(dev);
1377 if (error)
1378 return error;
1379
1380 error = register_queue_kobjects(net);
1381 if (error) {
1382 device_del(dev);
1383 return error;
1384 }
1385
1386 pm_runtime_set_memalloc_noio(dev, true);
1387
1388 return error;
1389}
1390
1391int netdev_class_create_file_ns(struct class_attribute *class_attr,
1392 const void *ns)
1393{
1394 return class_create_file_ns(&net_class, class_attr, ns);
1395}
1396EXPORT_SYMBOL(netdev_class_create_file_ns);
1397
1398void netdev_class_remove_file_ns(struct class_attribute *class_attr,
1399 const void *ns)
1400{
1401 class_remove_file_ns(&net_class, class_attr, ns);
1402}
1403EXPORT_SYMBOL(netdev_class_remove_file_ns);
1404
1405int __init netdev_kobject_init(void)
1406{
1407 kobj_ns_type_register(&net_ns_type_operations);
1408 return class_register(&net_class);
1409}
1// SPDX-License-Identifier: GPL-2.0-or-later
2/*
3 * net-sysfs.c - network device class and attributes
4 *
5 * Copyright (c) 2003 Stephen Hemminger <shemminger@osdl.org>
6 */
7
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/sched/isolation.h>
15#include <linux/nsproxy.h>
16#include <net/sock.h>
17#include <net/net_namespace.h>
18#include <linux/rtnetlink.h>
19#include <linux/vmalloc.h>
20#include <linux/export.h>
21#include <linux/jiffies.h>
22#include <linux/pm_runtime.h>
23#include <linux/of.h>
24#include <linux/of_net.h>
25#include <linux/cpu.h>
26#include <net/netdev_rx_queue.h>
27
28#include "dev.h"
29#include "net-sysfs.h"
30
31#ifdef CONFIG_SYSFS
32static const char fmt_hex[] = "%#x\n";
33static const char fmt_dec[] = "%d\n";
34static const char fmt_ulong[] = "%lu\n";
35static const char fmt_u64[] = "%llu\n";
36
37/* Caller holds RTNL or dev_base_lock */
38static inline int dev_isalive(const struct net_device *dev)
39{
40 return dev->reg_state <= NETREG_REGISTERED;
41}
42
43/* use same locking rules as GIF* ioctl's */
44static ssize_t netdev_show(const struct device *dev,
45 struct device_attribute *attr, char *buf,
46 ssize_t (*format)(const struct net_device *, char *))
47{
48 struct net_device *ndev = to_net_dev(dev);
49 ssize_t ret = -EINVAL;
50
51 read_lock(&dev_base_lock);
52 if (dev_isalive(ndev))
53 ret = (*format)(ndev, buf);
54 read_unlock(&dev_base_lock);
55
56 return ret;
57}
58
59/* generate a show function for simple field */
60#define NETDEVICE_SHOW(field, format_string) \
61static ssize_t format_##field(const struct net_device *dev, char *buf) \
62{ \
63 return sysfs_emit(buf, format_string, dev->field); \
64} \
65static ssize_t field##_show(struct device *dev, \
66 struct device_attribute *attr, char *buf) \
67{ \
68 return netdev_show(dev, attr, buf, format_##field); \
69} \
70
71#define NETDEVICE_SHOW_RO(field, format_string) \
72NETDEVICE_SHOW(field, format_string); \
73static DEVICE_ATTR_RO(field)
74
75#define NETDEVICE_SHOW_RW(field, format_string) \
76NETDEVICE_SHOW(field, format_string); \
77static DEVICE_ATTR_RW(field)
78
79/* use same locking and permission rules as SIF* ioctl's */
80static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
81 const char *buf, size_t len,
82 int (*set)(struct net_device *, unsigned long))
83{
84 struct net_device *netdev = to_net_dev(dev);
85 struct net *net = dev_net(netdev);
86 unsigned long new;
87 int ret;
88
89 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
90 return -EPERM;
91
92 ret = kstrtoul(buf, 0, &new);
93 if (ret)
94 goto err;
95
96 if (!rtnl_trylock())
97 return restart_syscall();
98
99 if (dev_isalive(netdev)) {
100 ret = (*set)(netdev, new);
101 if (ret == 0)
102 ret = len;
103 }
104 rtnl_unlock();
105 err:
106 return ret;
107}
108
109NETDEVICE_SHOW_RO(dev_id, fmt_hex);
110NETDEVICE_SHOW_RO(dev_port, fmt_dec);
111NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
112NETDEVICE_SHOW_RO(addr_len, fmt_dec);
113NETDEVICE_SHOW_RO(ifindex, fmt_dec);
114NETDEVICE_SHOW_RO(type, fmt_dec);
115NETDEVICE_SHOW_RO(link_mode, fmt_dec);
116
117static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
118 char *buf)
119{
120 struct net_device *ndev = to_net_dev(dev);
121
122 return sysfs_emit(buf, fmt_dec, dev_get_iflink(ndev));
123}
124static DEVICE_ATTR_RO(iflink);
125
126static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
127{
128 return sysfs_emit(buf, fmt_dec, dev->name_assign_type);
129}
130
131static ssize_t name_assign_type_show(struct device *dev,
132 struct device_attribute *attr,
133 char *buf)
134{
135 struct net_device *ndev = to_net_dev(dev);
136 ssize_t ret = -EINVAL;
137
138 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
139 ret = netdev_show(dev, attr, buf, format_name_assign_type);
140
141 return ret;
142}
143static DEVICE_ATTR_RO(name_assign_type);
144
145/* use same locking rules as GIFHWADDR ioctl's */
146static ssize_t address_show(struct device *dev, struct device_attribute *attr,
147 char *buf)
148{
149 struct net_device *ndev = to_net_dev(dev);
150 ssize_t ret = -EINVAL;
151
152 read_lock(&dev_base_lock);
153 if (dev_isalive(ndev))
154 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
155 read_unlock(&dev_base_lock);
156 return ret;
157}
158static DEVICE_ATTR_RO(address);
159
160static ssize_t broadcast_show(struct device *dev,
161 struct device_attribute *attr, char *buf)
162{
163 struct net_device *ndev = to_net_dev(dev);
164
165 if (dev_isalive(ndev))
166 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
167 return -EINVAL;
168}
169static DEVICE_ATTR_RO(broadcast);
170
171static int change_carrier(struct net_device *dev, unsigned long new_carrier)
172{
173 if (!netif_running(dev))
174 return -EINVAL;
175 return dev_change_carrier(dev, (bool)new_carrier);
176}
177
178static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
179 const char *buf, size_t len)
180{
181 struct net_device *netdev = to_net_dev(dev);
182
183 /* The check is also done in change_carrier; this helps returning early
184 * without hitting the trylock/restart in netdev_store.
185 */
186 if (!netdev->netdev_ops->ndo_change_carrier)
187 return -EOPNOTSUPP;
188
189 return netdev_store(dev, attr, buf, len, change_carrier);
190}
191
192static ssize_t carrier_show(struct device *dev,
193 struct device_attribute *attr, char *buf)
194{
195 struct net_device *netdev = to_net_dev(dev);
196 int ret = -EINVAL;
197
198 if (!rtnl_trylock())
199 return restart_syscall();
200
201 if (netif_running(netdev)) {
202 /* Synchronize carrier state with link watch,
203 * see also rtnl_getlink().
204 */
205 linkwatch_sync_dev(netdev);
206
207 ret = sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev));
208 }
209 rtnl_unlock();
210
211 return ret;
212}
213static DEVICE_ATTR_RW(carrier);
214
215static ssize_t speed_show(struct device *dev,
216 struct device_attribute *attr, char *buf)
217{
218 struct net_device *netdev = to_net_dev(dev);
219 int ret = -EINVAL;
220
221 /* The check is also done in __ethtool_get_link_ksettings; this helps
222 * returning early without hitting the trylock/restart below.
223 */
224 if (!netdev->ethtool_ops->get_link_ksettings)
225 return ret;
226
227 if (!rtnl_trylock())
228 return restart_syscall();
229
230 if (netif_running(netdev) && netif_device_present(netdev)) {
231 struct ethtool_link_ksettings cmd;
232
233 if (!__ethtool_get_link_ksettings(netdev, &cmd))
234 ret = sysfs_emit(buf, fmt_dec, cmd.base.speed);
235 }
236 rtnl_unlock();
237 return ret;
238}
239static DEVICE_ATTR_RO(speed);
240
241static ssize_t duplex_show(struct device *dev,
242 struct device_attribute *attr, char *buf)
243{
244 struct net_device *netdev = to_net_dev(dev);
245 int ret = -EINVAL;
246
247 /* The check is also done in __ethtool_get_link_ksettings; this helps
248 * returning early without hitting the trylock/restart below.
249 */
250 if (!netdev->ethtool_ops->get_link_ksettings)
251 return ret;
252
253 if (!rtnl_trylock())
254 return restart_syscall();
255
256 if (netif_running(netdev)) {
257 struct ethtool_link_ksettings cmd;
258
259 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
260 const char *duplex;
261
262 switch (cmd.base.duplex) {
263 case DUPLEX_HALF:
264 duplex = "half";
265 break;
266 case DUPLEX_FULL:
267 duplex = "full";
268 break;
269 default:
270 duplex = "unknown";
271 break;
272 }
273 ret = sysfs_emit(buf, "%s\n", duplex);
274 }
275 }
276 rtnl_unlock();
277 return ret;
278}
279static DEVICE_ATTR_RO(duplex);
280
281static ssize_t testing_show(struct device *dev,
282 struct device_attribute *attr, char *buf)
283{
284 struct net_device *netdev = to_net_dev(dev);
285
286 if (netif_running(netdev))
287 return sysfs_emit(buf, fmt_dec, !!netif_testing(netdev));
288
289 return -EINVAL;
290}
291static DEVICE_ATTR_RO(testing);
292
293static ssize_t dormant_show(struct device *dev,
294 struct device_attribute *attr, char *buf)
295{
296 struct net_device *netdev = to_net_dev(dev);
297
298 if (netif_running(netdev))
299 return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev));
300
301 return -EINVAL;
302}
303static DEVICE_ATTR_RO(dormant);
304
305static const char *const operstates[] = {
306 "unknown",
307 "notpresent", /* currently unused */
308 "down",
309 "lowerlayerdown",
310 "testing",
311 "dormant",
312 "up"
313};
314
315static ssize_t operstate_show(struct device *dev,
316 struct device_attribute *attr, char *buf)
317{
318 const struct net_device *netdev = to_net_dev(dev);
319 unsigned char operstate;
320
321 read_lock(&dev_base_lock);
322 operstate = netdev->operstate;
323 if (!netif_running(netdev))
324 operstate = IF_OPER_DOWN;
325 read_unlock(&dev_base_lock);
326
327 if (operstate >= ARRAY_SIZE(operstates))
328 return -EINVAL; /* should not happen */
329
330 return sysfs_emit(buf, "%s\n", operstates[operstate]);
331}
332static DEVICE_ATTR_RO(operstate);
333
334static ssize_t carrier_changes_show(struct device *dev,
335 struct device_attribute *attr,
336 char *buf)
337{
338 struct net_device *netdev = to_net_dev(dev);
339
340 return sysfs_emit(buf, fmt_dec,
341 atomic_read(&netdev->carrier_up_count) +
342 atomic_read(&netdev->carrier_down_count));
343}
344static DEVICE_ATTR_RO(carrier_changes);
345
346static ssize_t carrier_up_count_show(struct device *dev,
347 struct device_attribute *attr,
348 char *buf)
349{
350 struct net_device *netdev = to_net_dev(dev);
351
352 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
353}
354static DEVICE_ATTR_RO(carrier_up_count);
355
356static ssize_t carrier_down_count_show(struct device *dev,
357 struct device_attribute *attr,
358 char *buf)
359{
360 struct net_device *netdev = to_net_dev(dev);
361
362 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
363}
364static DEVICE_ATTR_RO(carrier_down_count);
365
366/* read-write attributes */
367
368static int change_mtu(struct net_device *dev, unsigned long new_mtu)
369{
370 return dev_set_mtu(dev, (int)new_mtu);
371}
372
373static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
374 const char *buf, size_t len)
375{
376 return netdev_store(dev, attr, buf, len, change_mtu);
377}
378NETDEVICE_SHOW_RW(mtu, fmt_dec);
379
380static int change_flags(struct net_device *dev, unsigned long new_flags)
381{
382 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
383}
384
385static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
386 const char *buf, size_t len)
387{
388 return netdev_store(dev, attr, buf, len, change_flags);
389}
390NETDEVICE_SHOW_RW(flags, fmt_hex);
391
392static ssize_t tx_queue_len_store(struct device *dev,
393 struct device_attribute *attr,
394 const char *buf, size_t len)
395{
396 if (!capable(CAP_NET_ADMIN))
397 return -EPERM;
398
399 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
400}
401NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
402
403static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
404{
405 WRITE_ONCE(dev->gro_flush_timeout, val);
406 return 0;
407}
408
409static ssize_t gro_flush_timeout_store(struct device *dev,
410 struct device_attribute *attr,
411 const char *buf, size_t len)
412{
413 if (!capable(CAP_NET_ADMIN))
414 return -EPERM;
415
416 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
417}
418NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
419
420static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
421{
422 WRITE_ONCE(dev->napi_defer_hard_irqs, val);
423 return 0;
424}
425
426static ssize_t napi_defer_hard_irqs_store(struct device *dev,
427 struct device_attribute *attr,
428 const char *buf, size_t len)
429{
430 if (!capable(CAP_NET_ADMIN))
431 return -EPERM;
432
433 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
434}
435NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
436
437static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
438 const char *buf, size_t len)
439{
440 struct net_device *netdev = to_net_dev(dev);
441 struct net *net = dev_net(netdev);
442 size_t count = len;
443 ssize_t ret = 0;
444
445 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
446 return -EPERM;
447
448 /* ignore trailing newline */
449 if (len > 0 && buf[len - 1] == '\n')
450 --count;
451
452 if (!rtnl_trylock())
453 return restart_syscall();
454
455 if (dev_isalive(netdev)) {
456 ret = dev_set_alias(netdev, buf, count);
457 if (ret < 0)
458 goto err;
459 ret = len;
460 netdev_state_change(netdev);
461 }
462err:
463 rtnl_unlock();
464
465 return ret;
466}
467
468static ssize_t ifalias_show(struct device *dev,
469 struct device_attribute *attr, char *buf)
470{
471 const struct net_device *netdev = to_net_dev(dev);
472 char tmp[IFALIASZ];
473 ssize_t ret = 0;
474
475 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
476 if (ret > 0)
477 ret = sysfs_emit(buf, "%s\n", tmp);
478 return ret;
479}
480static DEVICE_ATTR_RW(ifalias);
481
482static int change_group(struct net_device *dev, unsigned long new_group)
483{
484 dev_set_group(dev, (int)new_group);
485 return 0;
486}
487
488static ssize_t group_store(struct device *dev, struct device_attribute *attr,
489 const char *buf, size_t len)
490{
491 return netdev_store(dev, attr, buf, len, change_group);
492}
493NETDEVICE_SHOW(group, fmt_dec);
494static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
495
496static int change_proto_down(struct net_device *dev, unsigned long proto_down)
497{
498 return dev_change_proto_down(dev, (bool)proto_down);
499}
500
501static ssize_t proto_down_store(struct device *dev,
502 struct device_attribute *attr,
503 const char *buf, size_t len)
504{
505 return netdev_store(dev, attr, buf, len, change_proto_down);
506}
507NETDEVICE_SHOW_RW(proto_down, fmt_dec);
508
509static ssize_t phys_port_id_show(struct device *dev,
510 struct device_attribute *attr, char *buf)
511{
512 struct net_device *netdev = to_net_dev(dev);
513 ssize_t ret = -EINVAL;
514
515 /* The check is also done in dev_get_phys_port_id; this helps returning
516 * early without hitting the trylock/restart below.
517 */
518 if (!netdev->netdev_ops->ndo_get_phys_port_id)
519 return -EOPNOTSUPP;
520
521 if (!rtnl_trylock())
522 return restart_syscall();
523
524 if (dev_isalive(netdev)) {
525 struct netdev_phys_item_id ppid;
526
527 ret = dev_get_phys_port_id(netdev, &ppid);
528 if (!ret)
529 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
530 }
531 rtnl_unlock();
532
533 return ret;
534}
535static DEVICE_ATTR_RO(phys_port_id);
536
537static ssize_t phys_port_name_show(struct device *dev,
538 struct device_attribute *attr, char *buf)
539{
540 struct net_device *netdev = to_net_dev(dev);
541 ssize_t ret = -EINVAL;
542
543 /* The checks are also done in dev_get_phys_port_name; this helps
544 * returning early without hitting the trylock/restart below.
545 */
546 if (!netdev->netdev_ops->ndo_get_phys_port_name &&
547 !netdev->devlink_port)
548 return -EOPNOTSUPP;
549
550 if (!rtnl_trylock())
551 return restart_syscall();
552
553 if (dev_isalive(netdev)) {
554 char name[IFNAMSIZ];
555
556 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
557 if (!ret)
558 ret = sysfs_emit(buf, "%s\n", name);
559 }
560 rtnl_unlock();
561
562 return ret;
563}
564static DEVICE_ATTR_RO(phys_port_name);
565
566static ssize_t phys_switch_id_show(struct device *dev,
567 struct device_attribute *attr, char *buf)
568{
569 struct net_device *netdev = to_net_dev(dev);
570 ssize_t ret = -EINVAL;
571
572 /* The checks are also done in dev_get_phys_port_name; this helps
573 * returning early without hitting the trylock/restart below. This works
574 * because recurse is false when calling dev_get_port_parent_id.
575 */
576 if (!netdev->netdev_ops->ndo_get_port_parent_id &&
577 !netdev->devlink_port)
578 return -EOPNOTSUPP;
579
580 if (!rtnl_trylock())
581 return restart_syscall();
582
583 if (dev_isalive(netdev)) {
584 struct netdev_phys_item_id ppid = { };
585
586 ret = dev_get_port_parent_id(netdev, &ppid, false);
587 if (!ret)
588 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
589 }
590 rtnl_unlock();
591
592 return ret;
593}
594static DEVICE_ATTR_RO(phys_switch_id);
595
596static ssize_t threaded_show(struct device *dev,
597 struct device_attribute *attr, char *buf)
598{
599 struct net_device *netdev = to_net_dev(dev);
600 ssize_t ret = -EINVAL;
601
602 if (!rtnl_trylock())
603 return restart_syscall();
604
605 if (dev_isalive(netdev))
606 ret = sysfs_emit(buf, fmt_dec, netdev->threaded);
607
608 rtnl_unlock();
609 return ret;
610}
611
612static int modify_napi_threaded(struct net_device *dev, unsigned long val)
613{
614 int ret;
615
616 if (list_empty(&dev->napi_list))
617 return -EOPNOTSUPP;
618
619 if (val != 0 && val != 1)
620 return -EOPNOTSUPP;
621
622 ret = dev_set_threaded(dev, val);
623
624 return ret;
625}
626
627static ssize_t threaded_store(struct device *dev,
628 struct device_attribute *attr,
629 const char *buf, size_t len)
630{
631 return netdev_store(dev, attr, buf, len, modify_napi_threaded);
632}
633static DEVICE_ATTR_RW(threaded);
634
635static struct attribute *net_class_attrs[] __ro_after_init = {
636 &dev_attr_netdev_group.attr,
637 &dev_attr_type.attr,
638 &dev_attr_dev_id.attr,
639 &dev_attr_dev_port.attr,
640 &dev_attr_iflink.attr,
641 &dev_attr_ifindex.attr,
642 &dev_attr_name_assign_type.attr,
643 &dev_attr_addr_assign_type.attr,
644 &dev_attr_addr_len.attr,
645 &dev_attr_link_mode.attr,
646 &dev_attr_address.attr,
647 &dev_attr_broadcast.attr,
648 &dev_attr_speed.attr,
649 &dev_attr_duplex.attr,
650 &dev_attr_dormant.attr,
651 &dev_attr_testing.attr,
652 &dev_attr_operstate.attr,
653 &dev_attr_carrier_changes.attr,
654 &dev_attr_ifalias.attr,
655 &dev_attr_carrier.attr,
656 &dev_attr_mtu.attr,
657 &dev_attr_flags.attr,
658 &dev_attr_tx_queue_len.attr,
659 &dev_attr_gro_flush_timeout.attr,
660 &dev_attr_napi_defer_hard_irqs.attr,
661 &dev_attr_phys_port_id.attr,
662 &dev_attr_phys_port_name.attr,
663 &dev_attr_phys_switch_id.attr,
664 &dev_attr_proto_down.attr,
665 &dev_attr_carrier_up_count.attr,
666 &dev_attr_carrier_down_count.attr,
667 &dev_attr_threaded.attr,
668 NULL,
669};
670ATTRIBUTE_GROUPS(net_class);
671
672/* Show a given an attribute in the statistics group */
673static ssize_t netstat_show(const struct device *d,
674 struct device_attribute *attr, char *buf,
675 unsigned long offset)
676{
677 struct net_device *dev = to_net_dev(d);
678 ssize_t ret = -EINVAL;
679
680 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
681 offset % sizeof(u64) != 0);
682
683 read_lock(&dev_base_lock);
684 if (dev_isalive(dev)) {
685 struct rtnl_link_stats64 temp;
686 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
687
688 ret = sysfs_emit(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
689 }
690 read_unlock(&dev_base_lock);
691 return ret;
692}
693
694/* generate a read-only statistics attribute */
695#define NETSTAT_ENTRY(name) \
696static ssize_t name##_show(struct device *d, \
697 struct device_attribute *attr, char *buf) \
698{ \
699 return netstat_show(d, attr, buf, \
700 offsetof(struct rtnl_link_stats64, name)); \
701} \
702static DEVICE_ATTR_RO(name)
703
704NETSTAT_ENTRY(rx_packets);
705NETSTAT_ENTRY(tx_packets);
706NETSTAT_ENTRY(rx_bytes);
707NETSTAT_ENTRY(tx_bytes);
708NETSTAT_ENTRY(rx_errors);
709NETSTAT_ENTRY(tx_errors);
710NETSTAT_ENTRY(rx_dropped);
711NETSTAT_ENTRY(tx_dropped);
712NETSTAT_ENTRY(multicast);
713NETSTAT_ENTRY(collisions);
714NETSTAT_ENTRY(rx_length_errors);
715NETSTAT_ENTRY(rx_over_errors);
716NETSTAT_ENTRY(rx_crc_errors);
717NETSTAT_ENTRY(rx_frame_errors);
718NETSTAT_ENTRY(rx_fifo_errors);
719NETSTAT_ENTRY(rx_missed_errors);
720NETSTAT_ENTRY(tx_aborted_errors);
721NETSTAT_ENTRY(tx_carrier_errors);
722NETSTAT_ENTRY(tx_fifo_errors);
723NETSTAT_ENTRY(tx_heartbeat_errors);
724NETSTAT_ENTRY(tx_window_errors);
725NETSTAT_ENTRY(rx_compressed);
726NETSTAT_ENTRY(tx_compressed);
727NETSTAT_ENTRY(rx_nohandler);
728
729static struct attribute *netstat_attrs[] __ro_after_init = {
730 &dev_attr_rx_packets.attr,
731 &dev_attr_tx_packets.attr,
732 &dev_attr_rx_bytes.attr,
733 &dev_attr_tx_bytes.attr,
734 &dev_attr_rx_errors.attr,
735 &dev_attr_tx_errors.attr,
736 &dev_attr_rx_dropped.attr,
737 &dev_attr_tx_dropped.attr,
738 &dev_attr_multicast.attr,
739 &dev_attr_collisions.attr,
740 &dev_attr_rx_length_errors.attr,
741 &dev_attr_rx_over_errors.attr,
742 &dev_attr_rx_crc_errors.attr,
743 &dev_attr_rx_frame_errors.attr,
744 &dev_attr_rx_fifo_errors.attr,
745 &dev_attr_rx_missed_errors.attr,
746 &dev_attr_tx_aborted_errors.attr,
747 &dev_attr_tx_carrier_errors.attr,
748 &dev_attr_tx_fifo_errors.attr,
749 &dev_attr_tx_heartbeat_errors.attr,
750 &dev_attr_tx_window_errors.attr,
751 &dev_attr_rx_compressed.attr,
752 &dev_attr_tx_compressed.attr,
753 &dev_attr_rx_nohandler.attr,
754 NULL
755};
756
757static const struct attribute_group netstat_group = {
758 .name = "statistics",
759 .attrs = netstat_attrs,
760};
761
762static struct attribute *wireless_attrs[] = {
763 NULL
764};
765
766static const struct attribute_group wireless_group = {
767 .name = "wireless",
768 .attrs = wireless_attrs,
769};
770
771static bool wireless_group_needed(struct net_device *ndev)
772{
773#if IS_ENABLED(CONFIG_CFG80211)
774 if (ndev->ieee80211_ptr)
775 return true;
776#endif
777#if IS_ENABLED(CONFIG_WIRELESS_EXT)
778 if (ndev->wireless_handlers)
779 return true;
780#endif
781 return false;
782}
783
784#else /* CONFIG_SYSFS */
785#define net_class_groups NULL
786#endif /* CONFIG_SYSFS */
787
788#ifdef CONFIG_SYSFS
789#define to_rx_queue_attr(_attr) \
790 container_of(_attr, struct rx_queue_attribute, attr)
791
792#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
793
794static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
795 char *buf)
796{
797 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
798 struct netdev_rx_queue *queue = to_rx_queue(kobj);
799
800 if (!attribute->show)
801 return -EIO;
802
803 return attribute->show(queue, buf);
804}
805
806static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
807 const char *buf, size_t count)
808{
809 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
810 struct netdev_rx_queue *queue = to_rx_queue(kobj);
811
812 if (!attribute->store)
813 return -EIO;
814
815 return attribute->store(queue, buf, count);
816}
817
818static const struct sysfs_ops rx_queue_sysfs_ops = {
819 .show = rx_queue_attr_show,
820 .store = rx_queue_attr_store,
821};
822
823#ifdef CONFIG_RPS
824static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
825{
826 struct rps_map *map;
827 cpumask_var_t mask;
828 int i, len;
829
830 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
831 return -ENOMEM;
832
833 rcu_read_lock();
834 map = rcu_dereference(queue->rps_map);
835 if (map)
836 for (i = 0; i < map->len; i++)
837 cpumask_set_cpu(map->cpus[i], mask);
838
839 len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask));
840 rcu_read_unlock();
841 free_cpumask_var(mask);
842
843 return len < PAGE_SIZE ? len : -EINVAL;
844}
845
846static int netdev_rx_queue_set_rps_mask(struct netdev_rx_queue *queue,
847 cpumask_var_t mask)
848{
849 static DEFINE_MUTEX(rps_map_mutex);
850 struct rps_map *old_map, *map;
851 int cpu, i;
852
853 map = kzalloc(max_t(unsigned int,
854 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
855 GFP_KERNEL);
856 if (!map)
857 return -ENOMEM;
858
859 i = 0;
860 for_each_cpu_and(cpu, mask, cpu_online_mask)
861 map->cpus[i++] = cpu;
862
863 if (i) {
864 map->len = i;
865 } else {
866 kfree(map);
867 map = NULL;
868 }
869
870 mutex_lock(&rps_map_mutex);
871 old_map = rcu_dereference_protected(queue->rps_map,
872 mutex_is_locked(&rps_map_mutex));
873 rcu_assign_pointer(queue->rps_map, map);
874
875 if (map)
876 static_branch_inc(&rps_needed);
877 if (old_map)
878 static_branch_dec(&rps_needed);
879
880 mutex_unlock(&rps_map_mutex);
881
882 if (old_map)
883 kfree_rcu(old_map, rcu);
884 return 0;
885}
886
887int rps_cpumask_housekeeping(struct cpumask *mask)
888{
889 if (!cpumask_empty(mask)) {
890 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
891 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
892 if (cpumask_empty(mask))
893 return -EINVAL;
894 }
895 return 0;
896}
897
898static ssize_t store_rps_map(struct netdev_rx_queue *queue,
899 const char *buf, size_t len)
900{
901 cpumask_var_t mask;
902 int err;
903
904 if (!capable(CAP_NET_ADMIN))
905 return -EPERM;
906
907 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
908 return -ENOMEM;
909
910 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
911 if (err)
912 goto out;
913
914 err = rps_cpumask_housekeeping(mask);
915 if (err)
916 goto out;
917
918 err = netdev_rx_queue_set_rps_mask(queue, mask);
919
920out:
921 free_cpumask_var(mask);
922 return err ? : len;
923}
924
925static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
926 char *buf)
927{
928 struct rps_dev_flow_table *flow_table;
929 unsigned long val = 0;
930
931 rcu_read_lock();
932 flow_table = rcu_dereference(queue->rps_flow_table);
933 if (flow_table)
934 val = (unsigned long)flow_table->mask + 1;
935 rcu_read_unlock();
936
937 return sysfs_emit(buf, "%lu\n", val);
938}
939
940static void rps_dev_flow_table_release(struct rcu_head *rcu)
941{
942 struct rps_dev_flow_table *table = container_of(rcu,
943 struct rps_dev_flow_table, rcu);
944 vfree(table);
945}
946
947static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
948 const char *buf, size_t len)
949{
950 unsigned long mask, count;
951 struct rps_dev_flow_table *table, *old_table;
952 static DEFINE_SPINLOCK(rps_dev_flow_lock);
953 int rc;
954
955 if (!capable(CAP_NET_ADMIN))
956 return -EPERM;
957
958 rc = kstrtoul(buf, 0, &count);
959 if (rc < 0)
960 return rc;
961
962 if (count) {
963 mask = count - 1;
964 /* mask = roundup_pow_of_two(count) - 1;
965 * without overflows...
966 */
967 while ((mask | (mask >> 1)) != mask)
968 mask |= (mask >> 1);
969 /* On 64 bit arches, must check mask fits in table->mask (u32),
970 * and on 32bit arches, must check
971 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
972 */
973#if BITS_PER_LONG > 32
974 if (mask > (unsigned long)(u32)mask)
975 return -EINVAL;
976#else
977 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
978 / sizeof(struct rps_dev_flow)) {
979 /* Enforce a limit to prevent overflow */
980 return -EINVAL;
981 }
982#endif
983 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
984 if (!table)
985 return -ENOMEM;
986
987 table->mask = mask;
988 for (count = 0; count <= mask; count++)
989 table->flows[count].cpu = RPS_NO_CPU;
990 } else {
991 table = NULL;
992 }
993
994 spin_lock(&rps_dev_flow_lock);
995 old_table = rcu_dereference_protected(queue->rps_flow_table,
996 lockdep_is_held(&rps_dev_flow_lock));
997 rcu_assign_pointer(queue->rps_flow_table, table);
998 spin_unlock(&rps_dev_flow_lock);
999
1000 if (old_table)
1001 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
1002
1003 return len;
1004}
1005
1006static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
1007 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
1008
1009static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
1010 = __ATTR(rps_flow_cnt, 0644,
1011 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
1012#endif /* CONFIG_RPS */
1013
1014static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
1015#ifdef CONFIG_RPS
1016 &rps_cpus_attribute.attr,
1017 &rps_dev_flow_table_cnt_attribute.attr,
1018#endif
1019 NULL
1020};
1021ATTRIBUTE_GROUPS(rx_queue_default);
1022
1023static void rx_queue_release(struct kobject *kobj)
1024{
1025 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1026#ifdef CONFIG_RPS
1027 struct rps_map *map;
1028 struct rps_dev_flow_table *flow_table;
1029
1030 map = rcu_dereference_protected(queue->rps_map, 1);
1031 if (map) {
1032 RCU_INIT_POINTER(queue->rps_map, NULL);
1033 kfree_rcu(map, rcu);
1034 }
1035
1036 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1037 if (flow_table) {
1038 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1039 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1040 }
1041#endif
1042
1043 memset(kobj, 0, sizeof(*kobj));
1044 netdev_put(queue->dev, &queue->dev_tracker);
1045}
1046
1047static const void *rx_queue_namespace(const struct kobject *kobj)
1048{
1049 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1050 struct device *dev = &queue->dev->dev;
1051 const void *ns = NULL;
1052
1053 if (dev->class && dev->class->ns_type)
1054 ns = dev->class->namespace(dev);
1055
1056 return ns;
1057}
1058
1059static void rx_queue_get_ownership(const struct kobject *kobj,
1060 kuid_t *uid, kgid_t *gid)
1061{
1062 const struct net *net = rx_queue_namespace(kobj);
1063
1064 net_ns_get_ownership(net, uid, gid);
1065}
1066
1067static const struct kobj_type rx_queue_ktype = {
1068 .sysfs_ops = &rx_queue_sysfs_ops,
1069 .release = rx_queue_release,
1070 .default_groups = rx_queue_default_groups,
1071 .namespace = rx_queue_namespace,
1072 .get_ownership = rx_queue_get_ownership,
1073};
1074
1075static int rx_queue_default_mask(struct net_device *dev,
1076 struct netdev_rx_queue *queue)
1077{
1078#if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL)
1079 struct cpumask *rps_default_mask = READ_ONCE(dev_net(dev)->core.rps_default_mask);
1080
1081 if (rps_default_mask && !cpumask_empty(rps_default_mask))
1082 return netdev_rx_queue_set_rps_mask(queue, rps_default_mask);
1083#endif
1084 return 0;
1085}
1086
1087static int rx_queue_add_kobject(struct net_device *dev, int index)
1088{
1089 struct netdev_rx_queue *queue = dev->_rx + index;
1090 struct kobject *kobj = &queue->kobj;
1091 int error = 0;
1092
1093 /* Kobject_put later will trigger rx_queue_release call which
1094 * decreases dev refcount: Take that reference here
1095 */
1096 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1097
1098 kobj->kset = dev->queues_kset;
1099 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1100 "rx-%u", index);
1101 if (error)
1102 goto err;
1103
1104 if (dev->sysfs_rx_queue_group) {
1105 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1106 if (error)
1107 goto err;
1108 }
1109
1110 error = rx_queue_default_mask(dev, queue);
1111 if (error)
1112 goto err;
1113
1114 kobject_uevent(kobj, KOBJ_ADD);
1115
1116 return error;
1117
1118err:
1119 kobject_put(kobj);
1120 return error;
1121}
1122
1123static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1124 kgid_t kgid)
1125{
1126 struct netdev_rx_queue *queue = dev->_rx + index;
1127 struct kobject *kobj = &queue->kobj;
1128 int error;
1129
1130 error = sysfs_change_owner(kobj, kuid, kgid);
1131 if (error)
1132 return error;
1133
1134 if (dev->sysfs_rx_queue_group)
1135 error = sysfs_group_change_owner(
1136 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1137
1138 return error;
1139}
1140#endif /* CONFIG_SYSFS */
1141
1142int
1143net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1144{
1145#ifdef CONFIG_SYSFS
1146 int i;
1147 int error = 0;
1148
1149#ifndef CONFIG_RPS
1150 if (!dev->sysfs_rx_queue_group)
1151 return 0;
1152#endif
1153 for (i = old_num; i < new_num; i++) {
1154 error = rx_queue_add_kobject(dev, i);
1155 if (error) {
1156 new_num = old_num;
1157 break;
1158 }
1159 }
1160
1161 while (--i >= new_num) {
1162 struct kobject *kobj = &dev->_rx[i].kobj;
1163
1164 if (!refcount_read(&dev_net(dev)->ns.count))
1165 kobj->uevent_suppress = 1;
1166 if (dev->sysfs_rx_queue_group)
1167 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1168 kobject_put(kobj);
1169 }
1170
1171 return error;
1172#else
1173 return 0;
1174#endif
1175}
1176
1177static int net_rx_queue_change_owner(struct net_device *dev, int num,
1178 kuid_t kuid, kgid_t kgid)
1179{
1180#ifdef CONFIG_SYSFS
1181 int error = 0;
1182 int i;
1183
1184#ifndef CONFIG_RPS
1185 if (!dev->sysfs_rx_queue_group)
1186 return 0;
1187#endif
1188 for (i = 0; i < num; i++) {
1189 error = rx_queue_change_owner(dev, i, kuid, kgid);
1190 if (error)
1191 break;
1192 }
1193
1194 return error;
1195#else
1196 return 0;
1197#endif
1198}
1199
1200#ifdef CONFIG_SYSFS
1201/*
1202 * netdev_queue sysfs structures and functions.
1203 */
1204struct netdev_queue_attribute {
1205 struct attribute attr;
1206 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1207 ssize_t (*store)(struct netdev_queue *queue,
1208 const char *buf, size_t len);
1209};
1210#define to_netdev_queue_attr(_attr) \
1211 container_of(_attr, struct netdev_queue_attribute, attr)
1212
1213#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1214
1215static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1216 struct attribute *attr, char *buf)
1217{
1218 const struct netdev_queue_attribute *attribute
1219 = to_netdev_queue_attr(attr);
1220 struct netdev_queue *queue = to_netdev_queue(kobj);
1221
1222 if (!attribute->show)
1223 return -EIO;
1224
1225 return attribute->show(queue, buf);
1226}
1227
1228static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1229 struct attribute *attr,
1230 const char *buf, size_t count)
1231{
1232 const struct netdev_queue_attribute *attribute
1233 = to_netdev_queue_attr(attr);
1234 struct netdev_queue *queue = to_netdev_queue(kobj);
1235
1236 if (!attribute->store)
1237 return -EIO;
1238
1239 return attribute->store(queue, buf, count);
1240}
1241
1242static const struct sysfs_ops netdev_queue_sysfs_ops = {
1243 .show = netdev_queue_attr_show,
1244 .store = netdev_queue_attr_store,
1245};
1246
1247static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1248{
1249 unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1250
1251 return sysfs_emit(buf, fmt_ulong, trans_timeout);
1252}
1253
1254static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1255{
1256 struct net_device *dev = queue->dev;
1257 unsigned int i;
1258
1259 i = queue - dev->_tx;
1260 BUG_ON(i >= dev->num_tx_queues);
1261
1262 return i;
1263}
1264
1265static ssize_t traffic_class_show(struct netdev_queue *queue,
1266 char *buf)
1267{
1268 struct net_device *dev = queue->dev;
1269 int num_tc, tc;
1270 int index;
1271
1272 if (!netif_is_multiqueue(dev))
1273 return -ENOENT;
1274
1275 if (!rtnl_trylock())
1276 return restart_syscall();
1277
1278 index = get_netdev_queue_index(queue);
1279
1280 /* If queue belongs to subordinate dev use its TC mapping */
1281 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1282
1283 num_tc = dev->num_tc;
1284 tc = netdev_txq_to_tc(dev, index);
1285
1286 rtnl_unlock();
1287
1288 if (tc < 0)
1289 return -EINVAL;
1290
1291 /* We can report the traffic class one of two ways:
1292 * Subordinate device traffic classes are reported with the traffic
1293 * class first, and then the subordinate class so for example TC0 on
1294 * subordinate device 2 will be reported as "0-2". If the queue
1295 * belongs to the root device it will be reported with just the
1296 * traffic class, so just "0" for TC 0 for example.
1297 */
1298 return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) :
1299 sysfs_emit(buf, "%d\n", tc);
1300}
1301
1302#ifdef CONFIG_XPS
1303static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1304 char *buf)
1305{
1306 return sysfs_emit(buf, "%lu\n", queue->tx_maxrate);
1307}
1308
1309static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1310 const char *buf, size_t len)
1311{
1312 struct net_device *dev = queue->dev;
1313 int err, index = get_netdev_queue_index(queue);
1314 u32 rate = 0;
1315
1316 if (!capable(CAP_NET_ADMIN))
1317 return -EPERM;
1318
1319 /* The check is also done later; this helps returning early without
1320 * hitting the trylock/restart below.
1321 */
1322 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1323 return -EOPNOTSUPP;
1324
1325 err = kstrtou32(buf, 10, &rate);
1326 if (err < 0)
1327 return err;
1328
1329 if (!rtnl_trylock())
1330 return restart_syscall();
1331
1332 err = -EOPNOTSUPP;
1333 if (dev->netdev_ops->ndo_set_tx_maxrate)
1334 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1335
1336 rtnl_unlock();
1337 if (!err) {
1338 queue->tx_maxrate = rate;
1339 return len;
1340 }
1341 return err;
1342}
1343
1344static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1345 = __ATTR_RW(tx_maxrate);
1346#endif
1347
1348static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1349 = __ATTR_RO(tx_timeout);
1350
1351static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1352 = __ATTR_RO(traffic_class);
1353
1354#ifdef CONFIG_BQL
1355/*
1356 * Byte queue limits sysfs structures and functions.
1357 */
1358static ssize_t bql_show(char *buf, unsigned int value)
1359{
1360 return sysfs_emit(buf, "%u\n", value);
1361}
1362
1363static ssize_t bql_set(const char *buf, const size_t count,
1364 unsigned int *pvalue)
1365{
1366 unsigned int value;
1367 int err;
1368
1369 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1370 value = DQL_MAX_LIMIT;
1371 } else {
1372 err = kstrtouint(buf, 10, &value);
1373 if (err < 0)
1374 return err;
1375 if (value > DQL_MAX_LIMIT)
1376 return -EINVAL;
1377 }
1378
1379 *pvalue = value;
1380
1381 return count;
1382}
1383
1384static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1385 char *buf)
1386{
1387 struct dql *dql = &queue->dql;
1388
1389 return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1390}
1391
1392static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1393 const char *buf, size_t len)
1394{
1395 struct dql *dql = &queue->dql;
1396 unsigned int value;
1397 int err;
1398
1399 err = kstrtouint(buf, 10, &value);
1400 if (err < 0)
1401 return err;
1402
1403 dql->slack_hold_time = msecs_to_jiffies(value);
1404
1405 return len;
1406}
1407
1408static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1409 = __ATTR(hold_time, 0644,
1410 bql_show_hold_time, bql_set_hold_time);
1411
1412static ssize_t bql_show_inflight(struct netdev_queue *queue,
1413 char *buf)
1414{
1415 struct dql *dql = &queue->dql;
1416
1417 return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed);
1418}
1419
1420static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1421 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1422
1423#define BQL_ATTR(NAME, FIELD) \
1424static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1425 char *buf) \
1426{ \
1427 return bql_show(buf, queue->dql.FIELD); \
1428} \
1429 \
1430static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1431 const char *buf, size_t len) \
1432{ \
1433 return bql_set(buf, len, &queue->dql.FIELD); \
1434} \
1435 \
1436static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1437 = __ATTR(NAME, 0644, \
1438 bql_show_ ## NAME, bql_set_ ## NAME)
1439
1440BQL_ATTR(limit, limit);
1441BQL_ATTR(limit_max, max_limit);
1442BQL_ATTR(limit_min, min_limit);
1443
1444static struct attribute *dql_attrs[] __ro_after_init = {
1445 &bql_limit_attribute.attr,
1446 &bql_limit_max_attribute.attr,
1447 &bql_limit_min_attribute.attr,
1448 &bql_hold_time_attribute.attr,
1449 &bql_inflight_attribute.attr,
1450 NULL
1451};
1452
1453static const struct attribute_group dql_group = {
1454 .name = "byte_queue_limits",
1455 .attrs = dql_attrs,
1456};
1457#endif /* CONFIG_BQL */
1458
1459#ifdef CONFIG_XPS
1460static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1461 int tc, char *buf, enum xps_map_type type)
1462{
1463 struct xps_dev_maps *dev_maps;
1464 unsigned long *mask;
1465 unsigned int nr_ids;
1466 int j, len;
1467
1468 rcu_read_lock();
1469 dev_maps = rcu_dereference(dev->xps_maps[type]);
1470
1471 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1472 * when dev_maps hasn't been allocated yet, to be backward compatible.
1473 */
1474 nr_ids = dev_maps ? dev_maps->nr_ids :
1475 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1476
1477 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1478 if (!mask) {
1479 rcu_read_unlock();
1480 return -ENOMEM;
1481 }
1482
1483 if (!dev_maps || tc >= dev_maps->num_tc)
1484 goto out_no_maps;
1485
1486 for (j = 0; j < nr_ids; j++) {
1487 int i, tci = j * dev_maps->num_tc + tc;
1488 struct xps_map *map;
1489
1490 map = rcu_dereference(dev_maps->attr_map[tci]);
1491 if (!map)
1492 continue;
1493
1494 for (i = map->len; i--;) {
1495 if (map->queues[i] == index) {
1496 __set_bit(j, mask);
1497 break;
1498 }
1499 }
1500 }
1501out_no_maps:
1502 rcu_read_unlock();
1503
1504 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1505 bitmap_free(mask);
1506
1507 return len < PAGE_SIZE ? len : -EINVAL;
1508}
1509
1510static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1511{
1512 struct net_device *dev = queue->dev;
1513 unsigned int index;
1514 int len, tc;
1515
1516 if (!netif_is_multiqueue(dev))
1517 return -ENOENT;
1518
1519 index = get_netdev_queue_index(queue);
1520
1521 if (!rtnl_trylock())
1522 return restart_syscall();
1523
1524 /* If queue belongs to subordinate dev use its map */
1525 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1526
1527 tc = netdev_txq_to_tc(dev, index);
1528 if (tc < 0) {
1529 rtnl_unlock();
1530 return -EINVAL;
1531 }
1532
1533 /* Make sure the subordinate device can't be freed */
1534 get_device(&dev->dev);
1535 rtnl_unlock();
1536
1537 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1538
1539 put_device(&dev->dev);
1540 return len;
1541}
1542
1543static ssize_t xps_cpus_store(struct netdev_queue *queue,
1544 const char *buf, size_t len)
1545{
1546 struct net_device *dev = queue->dev;
1547 unsigned int index;
1548 cpumask_var_t mask;
1549 int err;
1550
1551 if (!netif_is_multiqueue(dev))
1552 return -ENOENT;
1553
1554 if (!capable(CAP_NET_ADMIN))
1555 return -EPERM;
1556
1557 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1558 return -ENOMEM;
1559
1560 index = get_netdev_queue_index(queue);
1561
1562 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1563 if (err) {
1564 free_cpumask_var(mask);
1565 return err;
1566 }
1567
1568 if (!rtnl_trylock()) {
1569 free_cpumask_var(mask);
1570 return restart_syscall();
1571 }
1572
1573 err = netif_set_xps_queue(dev, mask, index);
1574 rtnl_unlock();
1575
1576 free_cpumask_var(mask);
1577
1578 return err ? : len;
1579}
1580
1581static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1582 = __ATTR_RW(xps_cpus);
1583
1584static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1585{
1586 struct net_device *dev = queue->dev;
1587 unsigned int index;
1588 int tc;
1589
1590 index = get_netdev_queue_index(queue);
1591
1592 if (!rtnl_trylock())
1593 return restart_syscall();
1594
1595 tc = netdev_txq_to_tc(dev, index);
1596 rtnl_unlock();
1597 if (tc < 0)
1598 return -EINVAL;
1599
1600 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1601}
1602
1603static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1604 size_t len)
1605{
1606 struct net_device *dev = queue->dev;
1607 struct net *net = dev_net(dev);
1608 unsigned long *mask;
1609 unsigned int index;
1610 int err;
1611
1612 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1613 return -EPERM;
1614
1615 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1616 if (!mask)
1617 return -ENOMEM;
1618
1619 index = get_netdev_queue_index(queue);
1620
1621 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1622 if (err) {
1623 bitmap_free(mask);
1624 return err;
1625 }
1626
1627 if (!rtnl_trylock()) {
1628 bitmap_free(mask);
1629 return restart_syscall();
1630 }
1631
1632 cpus_read_lock();
1633 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1634 cpus_read_unlock();
1635
1636 rtnl_unlock();
1637
1638 bitmap_free(mask);
1639 return err ? : len;
1640}
1641
1642static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1643 = __ATTR_RW(xps_rxqs);
1644#endif /* CONFIG_XPS */
1645
1646static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1647 &queue_trans_timeout.attr,
1648 &queue_traffic_class.attr,
1649#ifdef CONFIG_XPS
1650 &xps_cpus_attribute.attr,
1651 &xps_rxqs_attribute.attr,
1652 &queue_tx_maxrate.attr,
1653#endif
1654 NULL
1655};
1656ATTRIBUTE_GROUPS(netdev_queue_default);
1657
1658static void netdev_queue_release(struct kobject *kobj)
1659{
1660 struct netdev_queue *queue = to_netdev_queue(kobj);
1661
1662 memset(kobj, 0, sizeof(*kobj));
1663 netdev_put(queue->dev, &queue->dev_tracker);
1664}
1665
1666static const void *netdev_queue_namespace(const struct kobject *kobj)
1667{
1668 struct netdev_queue *queue = to_netdev_queue(kobj);
1669 struct device *dev = &queue->dev->dev;
1670 const void *ns = NULL;
1671
1672 if (dev->class && dev->class->ns_type)
1673 ns = dev->class->namespace(dev);
1674
1675 return ns;
1676}
1677
1678static void netdev_queue_get_ownership(const struct kobject *kobj,
1679 kuid_t *uid, kgid_t *gid)
1680{
1681 const struct net *net = netdev_queue_namespace(kobj);
1682
1683 net_ns_get_ownership(net, uid, gid);
1684}
1685
1686static const struct kobj_type netdev_queue_ktype = {
1687 .sysfs_ops = &netdev_queue_sysfs_ops,
1688 .release = netdev_queue_release,
1689 .default_groups = netdev_queue_default_groups,
1690 .namespace = netdev_queue_namespace,
1691 .get_ownership = netdev_queue_get_ownership,
1692};
1693
1694static int netdev_queue_add_kobject(struct net_device *dev, int index)
1695{
1696 struct netdev_queue *queue = dev->_tx + index;
1697 struct kobject *kobj = &queue->kobj;
1698 int error = 0;
1699
1700 /* Kobject_put later will trigger netdev_queue_release call
1701 * which decreases dev refcount: Take that reference here
1702 */
1703 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1704
1705 kobj->kset = dev->queues_kset;
1706 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1707 "tx-%u", index);
1708 if (error)
1709 goto err;
1710
1711#ifdef CONFIG_BQL
1712 error = sysfs_create_group(kobj, &dql_group);
1713 if (error)
1714 goto err;
1715#endif
1716
1717 kobject_uevent(kobj, KOBJ_ADD);
1718 return 0;
1719
1720err:
1721 kobject_put(kobj);
1722 return error;
1723}
1724
1725static int tx_queue_change_owner(struct net_device *ndev, int index,
1726 kuid_t kuid, kgid_t kgid)
1727{
1728 struct netdev_queue *queue = ndev->_tx + index;
1729 struct kobject *kobj = &queue->kobj;
1730 int error;
1731
1732 error = sysfs_change_owner(kobj, kuid, kgid);
1733 if (error)
1734 return error;
1735
1736#ifdef CONFIG_BQL
1737 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1738#endif
1739 return error;
1740}
1741#endif /* CONFIG_SYSFS */
1742
1743int
1744netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1745{
1746#ifdef CONFIG_SYSFS
1747 int i;
1748 int error = 0;
1749
1750 /* Tx queue kobjects are allowed to be updated when a device is being
1751 * unregistered, but solely to remove queues from qdiscs. Any path
1752 * adding queues should be fixed.
1753 */
1754 WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1755 "New queues can't be registered after device unregistration.");
1756
1757 for (i = old_num; i < new_num; i++) {
1758 error = netdev_queue_add_kobject(dev, i);
1759 if (error) {
1760 new_num = old_num;
1761 break;
1762 }
1763 }
1764
1765 while (--i >= new_num) {
1766 struct netdev_queue *queue = dev->_tx + i;
1767
1768 if (!refcount_read(&dev_net(dev)->ns.count))
1769 queue->kobj.uevent_suppress = 1;
1770#ifdef CONFIG_BQL
1771 sysfs_remove_group(&queue->kobj, &dql_group);
1772#endif
1773 kobject_put(&queue->kobj);
1774 }
1775
1776 return error;
1777#else
1778 return 0;
1779#endif /* CONFIG_SYSFS */
1780}
1781
1782static int net_tx_queue_change_owner(struct net_device *dev, int num,
1783 kuid_t kuid, kgid_t kgid)
1784{
1785#ifdef CONFIG_SYSFS
1786 int error = 0;
1787 int i;
1788
1789 for (i = 0; i < num; i++) {
1790 error = tx_queue_change_owner(dev, i, kuid, kgid);
1791 if (error)
1792 break;
1793 }
1794
1795 return error;
1796#else
1797 return 0;
1798#endif /* CONFIG_SYSFS */
1799}
1800
1801static int register_queue_kobjects(struct net_device *dev)
1802{
1803 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1804
1805#ifdef CONFIG_SYSFS
1806 dev->queues_kset = kset_create_and_add("queues",
1807 NULL, &dev->dev.kobj);
1808 if (!dev->queues_kset)
1809 return -ENOMEM;
1810 real_rx = dev->real_num_rx_queues;
1811#endif
1812 real_tx = dev->real_num_tx_queues;
1813
1814 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1815 if (error)
1816 goto error;
1817 rxq = real_rx;
1818
1819 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1820 if (error)
1821 goto error;
1822 txq = real_tx;
1823
1824 return 0;
1825
1826error:
1827 netdev_queue_update_kobjects(dev, txq, 0);
1828 net_rx_queue_update_kobjects(dev, rxq, 0);
1829#ifdef CONFIG_SYSFS
1830 kset_unregister(dev->queues_kset);
1831#endif
1832 return error;
1833}
1834
1835static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1836{
1837 int error = 0, real_rx = 0, real_tx = 0;
1838
1839#ifdef CONFIG_SYSFS
1840 if (ndev->queues_kset) {
1841 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1842 if (error)
1843 return error;
1844 }
1845 real_rx = ndev->real_num_rx_queues;
1846#endif
1847 real_tx = ndev->real_num_tx_queues;
1848
1849 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1850 if (error)
1851 return error;
1852
1853 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1854 if (error)
1855 return error;
1856
1857 return 0;
1858}
1859
1860static void remove_queue_kobjects(struct net_device *dev)
1861{
1862 int real_rx = 0, real_tx = 0;
1863
1864#ifdef CONFIG_SYSFS
1865 real_rx = dev->real_num_rx_queues;
1866#endif
1867 real_tx = dev->real_num_tx_queues;
1868
1869 net_rx_queue_update_kobjects(dev, real_rx, 0);
1870 netdev_queue_update_kobjects(dev, real_tx, 0);
1871
1872 dev->real_num_rx_queues = 0;
1873 dev->real_num_tx_queues = 0;
1874#ifdef CONFIG_SYSFS
1875 kset_unregister(dev->queues_kset);
1876#endif
1877}
1878
1879static bool net_current_may_mount(void)
1880{
1881 struct net *net = current->nsproxy->net_ns;
1882
1883 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1884}
1885
1886static void *net_grab_current_ns(void)
1887{
1888 struct net *ns = current->nsproxy->net_ns;
1889#ifdef CONFIG_NET_NS
1890 if (ns)
1891 refcount_inc(&ns->passive);
1892#endif
1893 return ns;
1894}
1895
1896static const void *net_initial_ns(void)
1897{
1898 return &init_net;
1899}
1900
1901static const void *net_netlink_ns(struct sock *sk)
1902{
1903 return sock_net(sk);
1904}
1905
1906const struct kobj_ns_type_operations net_ns_type_operations = {
1907 .type = KOBJ_NS_TYPE_NET,
1908 .current_may_mount = net_current_may_mount,
1909 .grab_current_ns = net_grab_current_ns,
1910 .netlink_ns = net_netlink_ns,
1911 .initial_ns = net_initial_ns,
1912 .drop_ns = net_drop_ns,
1913};
1914EXPORT_SYMBOL_GPL(net_ns_type_operations);
1915
1916static int netdev_uevent(const struct device *d, struct kobj_uevent_env *env)
1917{
1918 const struct net_device *dev = to_net_dev(d);
1919 int retval;
1920
1921 /* pass interface to uevent. */
1922 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1923 if (retval)
1924 goto exit;
1925
1926 /* pass ifindex to uevent.
1927 * ifindex is useful as it won't change (interface name may change)
1928 * and is what RtNetlink uses natively.
1929 */
1930 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1931
1932exit:
1933 return retval;
1934}
1935
1936/*
1937 * netdev_release -- destroy and free a dead device.
1938 * Called when last reference to device kobject is gone.
1939 */
1940static void netdev_release(struct device *d)
1941{
1942 struct net_device *dev = to_net_dev(d);
1943
1944 BUG_ON(dev->reg_state != NETREG_RELEASED);
1945
1946 /* no need to wait for rcu grace period:
1947 * device is dead and about to be freed.
1948 */
1949 kfree(rcu_access_pointer(dev->ifalias));
1950 netdev_freemem(dev);
1951}
1952
1953static const void *net_namespace(const struct device *d)
1954{
1955 const struct net_device *dev = to_net_dev(d);
1956
1957 return dev_net(dev);
1958}
1959
1960static void net_get_ownership(const struct device *d, kuid_t *uid, kgid_t *gid)
1961{
1962 const struct net_device *dev = to_net_dev(d);
1963 const struct net *net = dev_net(dev);
1964
1965 net_ns_get_ownership(net, uid, gid);
1966}
1967
1968static struct class net_class __ro_after_init = {
1969 .name = "net",
1970 .dev_release = netdev_release,
1971 .dev_groups = net_class_groups,
1972 .dev_uevent = netdev_uevent,
1973 .ns_type = &net_ns_type_operations,
1974 .namespace = net_namespace,
1975 .get_ownership = net_get_ownership,
1976};
1977
1978#ifdef CONFIG_OF
1979static int of_dev_node_match(struct device *dev, const void *data)
1980{
1981 for (; dev; dev = dev->parent) {
1982 if (dev->of_node == data)
1983 return 1;
1984 }
1985
1986 return 0;
1987}
1988
1989/*
1990 * of_find_net_device_by_node - lookup the net device for the device node
1991 * @np: OF device node
1992 *
1993 * Looks up the net_device structure corresponding with the device node.
1994 * If successful, returns a pointer to the net_device with the embedded
1995 * struct device refcount incremented by one, or NULL on failure. The
1996 * refcount must be dropped when done with the net_device.
1997 */
1998struct net_device *of_find_net_device_by_node(struct device_node *np)
1999{
2000 struct device *dev;
2001
2002 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
2003 if (!dev)
2004 return NULL;
2005
2006 return to_net_dev(dev);
2007}
2008EXPORT_SYMBOL(of_find_net_device_by_node);
2009#endif
2010
2011/* Delete sysfs entries but hold kobject reference until after all
2012 * netdev references are gone.
2013 */
2014void netdev_unregister_kobject(struct net_device *ndev)
2015{
2016 struct device *dev = &ndev->dev;
2017
2018 if (!refcount_read(&dev_net(ndev)->ns.count))
2019 dev_set_uevent_suppress(dev, 1);
2020
2021 kobject_get(&dev->kobj);
2022
2023 remove_queue_kobjects(ndev);
2024
2025 pm_runtime_set_memalloc_noio(dev, false);
2026
2027 device_del(dev);
2028}
2029
2030/* Create sysfs entries for network device. */
2031int netdev_register_kobject(struct net_device *ndev)
2032{
2033 struct device *dev = &ndev->dev;
2034 const struct attribute_group **groups = ndev->sysfs_groups;
2035 int error = 0;
2036
2037 device_initialize(dev);
2038 dev->class = &net_class;
2039 dev->platform_data = ndev;
2040 dev->groups = groups;
2041
2042 dev_set_name(dev, "%s", ndev->name);
2043
2044#ifdef CONFIG_SYSFS
2045 /* Allow for a device specific group */
2046 if (*groups)
2047 groups++;
2048
2049 *groups++ = &netstat_group;
2050
2051 if (wireless_group_needed(ndev))
2052 *groups++ = &wireless_group;
2053#endif /* CONFIG_SYSFS */
2054
2055 error = device_add(dev);
2056 if (error)
2057 return error;
2058
2059 error = register_queue_kobjects(ndev);
2060 if (error) {
2061 device_del(dev);
2062 return error;
2063 }
2064
2065 pm_runtime_set_memalloc_noio(dev, true);
2066
2067 return error;
2068}
2069
2070/* Change owner for sysfs entries when moving network devices across network
2071 * namespaces owned by different user namespaces.
2072 */
2073int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2074 const struct net *net_new)
2075{
2076 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2077 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2078 struct device *dev = &ndev->dev;
2079 int error;
2080
2081 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2082 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2083
2084 /* The network namespace was changed but the owning user namespace is
2085 * identical so there's no need to change the owner of sysfs entries.
2086 */
2087 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2088 return 0;
2089
2090 error = device_change_owner(dev, new_uid, new_gid);
2091 if (error)
2092 return error;
2093
2094 error = queue_change_owner(ndev, new_uid, new_gid);
2095 if (error)
2096 return error;
2097
2098 return 0;
2099}
2100
2101int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2102 const void *ns)
2103{
2104 return class_create_file_ns(&net_class, class_attr, ns);
2105}
2106EXPORT_SYMBOL(netdev_class_create_file_ns);
2107
2108void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2109 const void *ns)
2110{
2111 class_remove_file_ns(&net_class, class_attr, ns);
2112}
2113EXPORT_SYMBOL(netdev_class_remove_file_ns);
2114
2115int __init netdev_kobject_init(void)
2116{
2117 kobj_ns_type_register(&net_ns_type_operations);
2118 return class_register(&net_class);
2119}