Loading...
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
27#include "net-sysfs.h"
28
29#ifdef CONFIG_SYSFS
30static const char fmt_hex[] = "%#x\n";
31static const char fmt_dec[] = "%d\n";
32static const char fmt_ulong[] = "%lu\n";
33static const char fmt_u64[] = "%llu\n";
34
35static inline int dev_isalive(const struct net_device *dev)
36{
37 return dev->reg_state <= NETREG_REGISTERED;
38}
39
40/* use same locking rules as GIF* ioctl's */
41static ssize_t netdev_show(const struct device *dev,
42 struct device_attribute *attr, char *buf,
43 ssize_t (*format)(const struct net_device *, char *))
44{
45 struct net_device *ndev = to_net_dev(dev);
46 ssize_t ret = -EINVAL;
47
48 read_lock(&dev_base_lock);
49 if (dev_isalive(ndev))
50 ret = (*format)(ndev, buf);
51 read_unlock(&dev_base_lock);
52
53 return ret;
54}
55
56/* generate a show function for simple field */
57#define NETDEVICE_SHOW(field, format_string) \
58static ssize_t format_##field(const struct net_device *dev, char *buf) \
59{ \
60 return sprintf(buf, format_string, dev->field); \
61} \
62static ssize_t field##_show(struct device *dev, \
63 struct device_attribute *attr, char *buf) \
64{ \
65 return netdev_show(dev, attr, buf, format_##field); \
66} \
67
68#define NETDEVICE_SHOW_RO(field, format_string) \
69NETDEVICE_SHOW(field, format_string); \
70static DEVICE_ATTR_RO(field)
71
72#define NETDEVICE_SHOW_RW(field, format_string) \
73NETDEVICE_SHOW(field, format_string); \
74static DEVICE_ATTR_RW(field)
75
76/* use same locking and permission rules as SIF* ioctl's */
77static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
78 const char *buf, size_t len,
79 int (*set)(struct net_device *, unsigned long))
80{
81 struct net_device *netdev = to_net_dev(dev);
82 struct net *net = dev_net(netdev);
83 unsigned long new;
84 int ret;
85
86 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
87 return -EPERM;
88
89 ret = kstrtoul(buf, 0, &new);
90 if (ret)
91 goto err;
92
93 if (!rtnl_trylock())
94 return restart_syscall();
95
96 if (dev_isalive(netdev)) {
97 ret = (*set)(netdev, new);
98 if (ret == 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(ifindex, fmt_dec);
111NETDEVICE_SHOW_RO(type, fmt_dec);
112NETDEVICE_SHOW_RO(link_mode, fmt_dec);
113
114static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
115 char *buf)
116{
117 struct net_device *ndev = to_net_dev(dev);
118
119 return sprintf(buf, fmt_dec, dev_get_iflink(ndev));
120}
121static DEVICE_ATTR_RO(iflink);
122
123static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
124{
125 return sprintf(buf, fmt_dec, dev->name_assign_type);
126}
127
128static ssize_t name_assign_type_show(struct device *dev,
129 struct device_attribute *attr,
130 char *buf)
131{
132 struct net_device *ndev = to_net_dev(dev);
133 ssize_t ret = -EINVAL;
134
135 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
136 ret = netdev_show(dev, attr, buf, format_name_assign_type);
137
138 return ret;
139}
140static DEVICE_ATTR_RO(name_assign_type);
141
142/* use same locking rules as GIFHWADDR ioctl's */
143static ssize_t address_show(struct device *dev, struct device_attribute *attr,
144 char *buf)
145{
146 struct net_device *ndev = to_net_dev(dev);
147 ssize_t ret = -EINVAL;
148
149 read_lock(&dev_base_lock);
150 if (dev_isalive(ndev))
151 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
152 read_unlock(&dev_base_lock);
153 return ret;
154}
155static DEVICE_ATTR_RO(address);
156
157static ssize_t broadcast_show(struct device *dev,
158 struct device_attribute *attr, char *buf)
159{
160 struct net_device *ndev = to_net_dev(dev);
161
162 if (dev_isalive(ndev))
163 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
164 return -EINVAL;
165}
166static DEVICE_ATTR_RO(broadcast);
167
168static int change_carrier(struct net_device *dev, unsigned long new_carrier)
169{
170 if (!netif_running(dev))
171 return -EINVAL;
172 return dev_change_carrier(dev, (bool)new_carrier);
173}
174
175static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
176 const char *buf, size_t len)
177{
178 return netdev_store(dev, attr, buf, len, change_carrier);
179}
180
181static ssize_t carrier_show(struct device *dev,
182 struct device_attribute *attr, char *buf)
183{
184 struct net_device *netdev = to_net_dev(dev);
185
186 if (netif_running(netdev))
187 return sprintf(buf, fmt_dec, !!netif_carrier_ok(netdev));
188
189 return -EINVAL;
190}
191static DEVICE_ATTR_RW(carrier);
192
193static ssize_t speed_show(struct device *dev,
194 struct device_attribute *attr, char *buf)
195{
196 struct net_device *netdev = to_net_dev(dev);
197 int ret = -EINVAL;
198
199 if (!rtnl_trylock())
200 return restart_syscall();
201
202 if (netif_running(netdev)) {
203 struct ethtool_link_ksettings cmd;
204
205 if (!__ethtool_get_link_ksettings(netdev, &cmd))
206 ret = sprintf(buf, fmt_dec, cmd.base.speed);
207 }
208 rtnl_unlock();
209 return ret;
210}
211static DEVICE_ATTR_RO(speed);
212
213static ssize_t duplex_show(struct device *dev,
214 struct device_attribute *attr, char *buf)
215{
216 struct net_device *netdev = to_net_dev(dev);
217 int ret = -EINVAL;
218
219 if (!rtnl_trylock())
220 return restart_syscall();
221
222 if (netif_running(netdev)) {
223 struct ethtool_link_ksettings cmd;
224
225 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
226 const char *duplex;
227
228 switch (cmd.base.duplex) {
229 case DUPLEX_HALF:
230 duplex = "half";
231 break;
232 case DUPLEX_FULL:
233 duplex = "full";
234 break;
235 default:
236 duplex = "unknown";
237 break;
238 }
239 ret = sprintf(buf, "%s\n", duplex);
240 }
241 }
242 rtnl_unlock();
243 return ret;
244}
245static DEVICE_ATTR_RO(duplex);
246
247static ssize_t testing_show(struct device *dev,
248 struct device_attribute *attr, char *buf)
249{
250 struct net_device *netdev = to_net_dev(dev);
251
252 if (netif_running(netdev))
253 return sprintf(buf, fmt_dec, !!netif_testing(netdev));
254
255 return -EINVAL;
256}
257static DEVICE_ATTR_RO(testing);
258
259static ssize_t dormant_show(struct device *dev,
260 struct device_attribute *attr, char *buf)
261{
262 struct net_device *netdev = to_net_dev(dev);
263
264 if (netif_running(netdev))
265 return sprintf(buf, fmt_dec, !!netif_dormant(netdev));
266
267 return -EINVAL;
268}
269static DEVICE_ATTR_RO(dormant);
270
271static const char *const operstates[] = {
272 "unknown",
273 "notpresent", /* currently unused */
274 "down",
275 "lowerlayerdown",
276 "testing",
277 "dormant",
278 "up"
279};
280
281static ssize_t operstate_show(struct device *dev,
282 struct device_attribute *attr, char *buf)
283{
284 const struct net_device *netdev = to_net_dev(dev);
285 unsigned char operstate;
286
287 read_lock(&dev_base_lock);
288 operstate = netdev->operstate;
289 if (!netif_running(netdev))
290 operstate = IF_OPER_DOWN;
291 read_unlock(&dev_base_lock);
292
293 if (operstate >= ARRAY_SIZE(operstates))
294 return -EINVAL; /* should not happen */
295
296 return sprintf(buf, "%s\n", operstates[operstate]);
297}
298static DEVICE_ATTR_RO(operstate);
299
300static ssize_t carrier_changes_show(struct device *dev,
301 struct device_attribute *attr,
302 char *buf)
303{
304 struct net_device *netdev = to_net_dev(dev);
305
306 return sprintf(buf, fmt_dec,
307 atomic_read(&netdev->carrier_up_count) +
308 atomic_read(&netdev->carrier_down_count));
309}
310static DEVICE_ATTR_RO(carrier_changes);
311
312static ssize_t carrier_up_count_show(struct device *dev,
313 struct device_attribute *attr,
314 char *buf)
315{
316 struct net_device *netdev = to_net_dev(dev);
317
318 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
319}
320static DEVICE_ATTR_RO(carrier_up_count);
321
322static ssize_t carrier_down_count_show(struct device *dev,
323 struct device_attribute *attr,
324 char *buf)
325{
326 struct net_device *netdev = to_net_dev(dev);
327
328 return sprintf(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
329}
330static DEVICE_ATTR_RO(carrier_down_count);
331
332/* read-write attributes */
333
334static int change_mtu(struct net_device *dev, unsigned long new_mtu)
335{
336 return dev_set_mtu(dev, (int)new_mtu);
337}
338
339static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
340 const char *buf, size_t len)
341{
342 return netdev_store(dev, attr, buf, len, change_mtu);
343}
344NETDEVICE_SHOW_RW(mtu, fmt_dec);
345
346static int change_flags(struct net_device *dev, unsigned long new_flags)
347{
348 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
349}
350
351static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
352 const char *buf, size_t len)
353{
354 return netdev_store(dev, attr, buf, len, change_flags);
355}
356NETDEVICE_SHOW_RW(flags, fmt_hex);
357
358static ssize_t tx_queue_len_store(struct device *dev,
359 struct device_attribute *attr,
360 const char *buf, size_t len)
361{
362 if (!capable(CAP_NET_ADMIN))
363 return -EPERM;
364
365 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
366}
367NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
368
369static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
370{
371 WRITE_ONCE(dev->gro_flush_timeout, val);
372 return 0;
373}
374
375static ssize_t gro_flush_timeout_store(struct device *dev,
376 struct device_attribute *attr,
377 const char *buf, size_t len)
378{
379 if (!capable(CAP_NET_ADMIN))
380 return -EPERM;
381
382 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
383}
384NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
385
386static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
387{
388 WRITE_ONCE(dev->napi_defer_hard_irqs, val);
389 return 0;
390}
391
392static ssize_t napi_defer_hard_irqs_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, change_napi_defer_hard_irqs);
400}
401NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
402
403static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
404 const char *buf, size_t len)
405{
406 struct net_device *netdev = to_net_dev(dev);
407 struct net *net = dev_net(netdev);
408 size_t count = len;
409 ssize_t ret = 0;
410
411 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
412 return -EPERM;
413
414 /* ignore trailing newline */
415 if (len > 0 && buf[len - 1] == '\n')
416 --count;
417
418 if (!rtnl_trylock())
419 return restart_syscall();
420
421 if (dev_isalive(netdev)) {
422 ret = dev_set_alias(netdev, buf, count);
423 if (ret < 0)
424 goto err;
425 ret = len;
426 netdev_state_change(netdev);
427 }
428err:
429 rtnl_unlock();
430
431 return ret;
432}
433
434static ssize_t ifalias_show(struct device *dev,
435 struct device_attribute *attr, char *buf)
436{
437 const struct net_device *netdev = to_net_dev(dev);
438 char tmp[IFALIASZ];
439 ssize_t ret = 0;
440
441 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
442 if (ret > 0)
443 ret = sprintf(buf, "%s\n", tmp);
444 return ret;
445}
446static DEVICE_ATTR_RW(ifalias);
447
448static int change_group(struct net_device *dev, unsigned long new_group)
449{
450 dev_set_group(dev, (int)new_group);
451 return 0;
452}
453
454static ssize_t group_store(struct device *dev, struct device_attribute *attr,
455 const char *buf, size_t len)
456{
457 return netdev_store(dev, attr, buf, len, change_group);
458}
459NETDEVICE_SHOW(group, fmt_dec);
460static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
461
462static int change_proto_down(struct net_device *dev, unsigned long proto_down)
463{
464 return dev_change_proto_down(dev, (bool)proto_down);
465}
466
467static ssize_t proto_down_store(struct device *dev,
468 struct device_attribute *attr,
469 const char *buf, size_t len)
470{
471 return netdev_store(dev, attr, buf, len, change_proto_down);
472}
473NETDEVICE_SHOW_RW(proto_down, fmt_dec);
474
475static ssize_t phys_port_id_show(struct device *dev,
476 struct device_attribute *attr, char *buf)
477{
478 struct net_device *netdev = to_net_dev(dev);
479 ssize_t ret = -EINVAL;
480
481 if (!rtnl_trylock())
482 return restart_syscall();
483
484 if (dev_isalive(netdev)) {
485 struct netdev_phys_item_id ppid;
486
487 ret = dev_get_phys_port_id(netdev, &ppid);
488 if (!ret)
489 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
490 }
491 rtnl_unlock();
492
493 return ret;
494}
495static DEVICE_ATTR_RO(phys_port_id);
496
497static ssize_t phys_port_name_show(struct device *dev,
498 struct device_attribute *attr, char *buf)
499{
500 struct net_device *netdev = to_net_dev(dev);
501 ssize_t ret = -EINVAL;
502
503 if (!rtnl_trylock())
504 return restart_syscall();
505
506 if (dev_isalive(netdev)) {
507 char name[IFNAMSIZ];
508
509 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
510 if (!ret)
511 ret = sprintf(buf, "%s\n", name);
512 }
513 rtnl_unlock();
514
515 return ret;
516}
517static DEVICE_ATTR_RO(phys_port_name);
518
519static ssize_t phys_switch_id_show(struct device *dev,
520 struct device_attribute *attr, char *buf)
521{
522 struct net_device *netdev = to_net_dev(dev);
523 ssize_t ret = -EINVAL;
524
525 if (!rtnl_trylock())
526 return restart_syscall();
527
528 if (dev_isalive(netdev)) {
529 struct netdev_phys_item_id ppid = { };
530
531 ret = dev_get_port_parent_id(netdev, &ppid, false);
532 if (!ret)
533 ret = sprintf(buf, "%*phN\n", ppid.id_len, ppid.id);
534 }
535 rtnl_unlock();
536
537 return ret;
538}
539static DEVICE_ATTR_RO(phys_switch_id);
540
541static ssize_t threaded_show(struct device *dev,
542 struct device_attribute *attr, char *buf)
543{
544 struct net_device *netdev = to_net_dev(dev);
545 ssize_t ret = -EINVAL;
546
547 if (!rtnl_trylock())
548 return restart_syscall();
549
550 if (dev_isalive(netdev))
551 ret = sprintf(buf, fmt_dec, netdev->threaded);
552
553 rtnl_unlock();
554 return ret;
555}
556
557static int modify_napi_threaded(struct net_device *dev, unsigned long val)
558{
559 int ret;
560
561 if (list_empty(&dev->napi_list))
562 return -EOPNOTSUPP;
563
564 if (val != 0 && val != 1)
565 return -EOPNOTSUPP;
566
567 ret = dev_set_threaded(dev, val);
568
569 return ret;
570}
571
572static ssize_t threaded_store(struct device *dev,
573 struct device_attribute *attr,
574 const char *buf, size_t len)
575{
576 return netdev_store(dev, attr, buf, len, modify_napi_threaded);
577}
578static DEVICE_ATTR_RW(threaded);
579
580static struct attribute *net_class_attrs[] __ro_after_init = {
581 &dev_attr_netdev_group.attr,
582 &dev_attr_type.attr,
583 &dev_attr_dev_id.attr,
584 &dev_attr_dev_port.attr,
585 &dev_attr_iflink.attr,
586 &dev_attr_ifindex.attr,
587 &dev_attr_name_assign_type.attr,
588 &dev_attr_addr_assign_type.attr,
589 &dev_attr_addr_len.attr,
590 &dev_attr_link_mode.attr,
591 &dev_attr_address.attr,
592 &dev_attr_broadcast.attr,
593 &dev_attr_speed.attr,
594 &dev_attr_duplex.attr,
595 &dev_attr_dormant.attr,
596 &dev_attr_testing.attr,
597 &dev_attr_operstate.attr,
598 &dev_attr_carrier_changes.attr,
599 &dev_attr_ifalias.attr,
600 &dev_attr_carrier.attr,
601 &dev_attr_mtu.attr,
602 &dev_attr_flags.attr,
603 &dev_attr_tx_queue_len.attr,
604 &dev_attr_gro_flush_timeout.attr,
605 &dev_attr_napi_defer_hard_irqs.attr,
606 &dev_attr_phys_port_id.attr,
607 &dev_attr_phys_port_name.attr,
608 &dev_attr_phys_switch_id.attr,
609 &dev_attr_proto_down.attr,
610 &dev_attr_carrier_up_count.attr,
611 &dev_attr_carrier_down_count.attr,
612 &dev_attr_threaded.attr,
613 NULL,
614};
615ATTRIBUTE_GROUPS(net_class);
616
617/* Show a given an attribute in the statistics group */
618static ssize_t netstat_show(const struct device *d,
619 struct device_attribute *attr, char *buf,
620 unsigned long offset)
621{
622 struct net_device *dev = to_net_dev(d);
623 ssize_t ret = -EINVAL;
624
625 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
626 offset % sizeof(u64) != 0);
627
628 read_lock(&dev_base_lock);
629 if (dev_isalive(dev)) {
630 struct rtnl_link_stats64 temp;
631 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
632
633 ret = sprintf(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
634 }
635 read_unlock(&dev_base_lock);
636 return ret;
637}
638
639/* generate a read-only statistics attribute */
640#define NETSTAT_ENTRY(name) \
641static ssize_t name##_show(struct device *d, \
642 struct device_attribute *attr, char *buf) \
643{ \
644 return netstat_show(d, attr, buf, \
645 offsetof(struct rtnl_link_stats64, name)); \
646} \
647static DEVICE_ATTR_RO(name)
648
649NETSTAT_ENTRY(rx_packets);
650NETSTAT_ENTRY(tx_packets);
651NETSTAT_ENTRY(rx_bytes);
652NETSTAT_ENTRY(tx_bytes);
653NETSTAT_ENTRY(rx_errors);
654NETSTAT_ENTRY(tx_errors);
655NETSTAT_ENTRY(rx_dropped);
656NETSTAT_ENTRY(tx_dropped);
657NETSTAT_ENTRY(multicast);
658NETSTAT_ENTRY(collisions);
659NETSTAT_ENTRY(rx_length_errors);
660NETSTAT_ENTRY(rx_over_errors);
661NETSTAT_ENTRY(rx_crc_errors);
662NETSTAT_ENTRY(rx_frame_errors);
663NETSTAT_ENTRY(rx_fifo_errors);
664NETSTAT_ENTRY(rx_missed_errors);
665NETSTAT_ENTRY(tx_aborted_errors);
666NETSTAT_ENTRY(tx_carrier_errors);
667NETSTAT_ENTRY(tx_fifo_errors);
668NETSTAT_ENTRY(tx_heartbeat_errors);
669NETSTAT_ENTRY(tx_window_errors);
670NETSTAT_ENTRY(rx_compressed);
671NETSTAT_ENTRY(tx_compressed);
672NETSTAT_ENTRY(rx_nohandler);
673
674static struct attribute *netstat_attrs[] __ro_after_init = {
675 &dev_attr_rx_packets.attr,
676 &dev_attr_tx_packets.attr,
677 &dev_attr_rx_bytes.attr,
678 &dev_attr_tx_bytes.attr,
679 &dev_attr_rx_errors.attr,
680 &dev_attr_tx_errors.attr,
681 &dev_attr_rx_dropped.attr,
682 &dev_attr_tx_dropped.attr,
683 &dev_attr_multicast.attr,
684 &dev_attr_collisions.attr,
685 &dev_attr_rx_length_errors.attr,
686 &dev_attr_rx_over_errors.attr,
687 &dev_attr_rx_crc_errors.attr,
688 &dev_attr_rx_frame_errors.attr,
689 &dev_attr_rx_fifo_errors.attr,
690 &dev_attr_rx_missed_errors.attr,
691 &dev_attr_tx_aborted_errors.attr,
692 &dev_attr_tx_carrier_errors.attr,
693 &dev_attr_tx_fifo_errors.attr,
694 &dev_attr_tx_heartbeat_errors.attr,
695 &dev_attr_tx_window_errors.attr,
696 &dev_attr_rx_compressed.attr,
697 &dev_attr_tx_compressed.attr,
698 &dev_attr_rx_nohandler.attr,
699 NULL
700};
701
702static const struct attribute_group netstat_group = {
703 .name = "statistics",
704 .attrs = netstat_attrs,
705};
706
707#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
708static struct attribute *wireless_attrs[] = {
709 NULL
710};
711
712static const struct attribute_group wireless_group = {
713 .name = "wireless",
714 .attrs = wireless_attrs,
715};
716#endif
717
718#else /* CONFIG_SYSFS */
719#define net_class_groups NULL
720#endif /* CONFIG_SYSFS */
721
722#ifdef CONFIG_SYSFS
723#define to_rx_queue_attr(_attr) \
724 container_of(_attr, struct rx_queue_attribute, attr)
725
726#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
727
728static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
729 char *buf)
730{
731 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
732 struct netdev_rx_queue *queue = to_rx_queue(kobj);
733
734 if (!attribute->show)
735 return -EIO;
736
737 return attribute->show(queue, buf);
738}
739
740static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
741 const char *buf, size_t count)
742{
743 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
744 struct netdev_rx_queue *queue = to_rx_queue(kobj);
745
746 if (!attribute->store)
747 return -EIO;
748
749 return attribute->store(queue, buf, count);
750}
751
752static const struct sysfs_ops rx_queue_sysfs_ops = {
753 .show = rx_queue_attr_show,
754 .store = rx_queue_attr_store,
755};
756
757#ifdef CONFIG_RPS
758static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
759{
760 struct rps_map *map;
761 cpumask_var_t mask;
762 int i, len;
763
764 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
765 return -ENOMEM;
766
767 rcu_read_lock();
768 map = rcu_dereference(queue->rps_map);
769 if (map)
770 for (i = 0; i < map->len; i++)
771 cpumask_set_cpu(map->cpus[i], mask);
772
773 len = snprintf(buf, PAGE_SIZE, "%*pb\n", cpumask_pr_args(mask));
774 rcu_read_unlock();
775 free_cpumask_var(mask);
776
777 return len < PAGE_SIZE ? len : -EINVAL;
778}
779
780static ssize_t store_rps_map(struct netdev_rx_queue *queue,
781 const char *buf, size_t len)
782{
783 struct rps_map *old_map, *map;
784 cpumask_var_t mask;
785 int err, cpu, i, hk_flags;
786 static DEFINE_MUTEX(rps_map_mutex);
787
788 if (!capable(CAP_NET_ADMIN))
789 return -EPERM;
790
791 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
792 return -ENOMEM;
793
794 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
795 if (err) {
796 free_cpumask_var(mask);
797 return err;
798 }
799
800 if (!cpumask_empty(mask)) {
801 hk_flags = HK_FLAG_DOMAIN | HK_FLAG_WQ;
802 cpumask_and(mask, mask, housekeeping_cpumask(hk_flags));
803 if (cpumask_empty(mask)) {
804 free_cpumask_var(mask);
805 return -EINVAL;
806 }
807 }
808
809 map = kzalloc(max_t(unsigned int,
810 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
811 GFP_KERNEL);
812 if (!map) {
813 free_cpumask_var(mask);
814 return -ENOMEM;
815 }
816
817 i = 0;
818 for_each_cpu_and(cpu, mask, cpu_online_mask)
819 map->cpus[i++] = cpu;
820
821 if (i) {
822 map->len = i;
823 } else {
824 kfree(map);
825 map = NULL;
826 }
827
828 mutex_lock(&rps_map_mutex);
829 old_map = rcu_dereference_protected(queue->rps_map,
830 mutex_is_locked(&rps_map_mutex));
831 rcu_assign_pointer(queue->rps_map, map);
832
833 if (map)
834 static_branch_inc(&rps_needed);
835 if (old_map)
836 static_branch_dec(&rps_needed);
837
838 mutex_unlock(&rps_map_mutex);
839
840 if (old_map)
841 kfree_rcu(old_map, rcu);
842
843 free_cpumask_var(mask);
844 return len;
845}
846
847static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
848 char *buf)
849{
850 struct rps_dev_flow_table *flow_table;
851 unsigned long val = 0;
852
853 rcu_read_lock();
854 flow_table = rcu_dereference(queue->rps_flow_table);
855 if (flow_table)
856 val = (unsigned long)flow_table->mask + 1;
857 rcu_read_unlock();
858
859 return sprintf(buf, "%lu\n", val);
860}
861
862static void rps_dev_flow_table_release(struct rcu_head *rcu)
863{
864 struct rps_dev_flow_table *table = container_of(rcu,
865 struct rps_dev_flow_table, rcu);
866 vfree(table);
867}
868
869static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
870 const char *buf, size_t len)
871{
872 unsigned long mask, count;
873 struct rps_dev_flow_table *table, *old_table;
874 static DEFINE_SPINLOCK(rps_dev_flow_lock);
875 int rc;
876
877 if (!capable(CAP_NET_ADMIN))
878 return -EPERM;
879
880 rc = kstrtoul(buf, 0, &count);
881 if (rc < 0)
882 return rc;
883
884 if (count) {
885 mask = count - 1;
886 /* mask = roundup_pow_of_two(count) - 1;
887 * without overflows...
888 */
889 while ((mask | (mask >> 1)) != mask)
890 mask |= (mask >> 1);
891 /* On 64 bit arches, must check mask fits in table->mask (u32),
892 * and on 32bit arches, must check
893 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
894 */
895#if BITS_PER_LONG > 32
896 if (mask > (unsigned long)(u32)mask)
897 return -EINVAL;
898#else
899 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
900 / sizeof(struct rps_dev_flow)) {
901 /* Enforce a limit to prevent overflow */
902 return -EINVAL;
903 }
904#endif
905 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
906 if (!table)
907 return -ENOMEM;
908
909 table->mask = mask;
910 for (count = 0; count <= mask; count++)
911 table->flows[count].cpu = RPS_NO_CPU;
912 } else {
913 table = NULL;
914 }
915
916 spin_lock(&rps_dev_flow_lock);
917 old_table = rcu_dereference_protected(queue->rps_flow_table,
918 lockdep_is_held(&rps_dev_flow_lock));
919 rcu_assign_pointer(queue->rps_flow_table, table);
920 spin_unlock(&rps_dev_flow_lock);
921
922 if (old_table)
923 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
924
925 return len;
926}
927
928static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
929 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
930
931static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
932 = __ATTR(rps_flow_cnt, 0644,
933 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
934#endif /* CONFIG_RPS */
935
936static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
937#ifdef CONFIG_RPS
938 &rps_cpus_attribute.attr,
939 &rps_dev_flow_table_cnt_attribute.attr,
940#endif
941 NULL
942};
943ATTRIBUTE_GROUPS(rx_queue_default);
944
945static void rx_queue_release(struct kobject *kobj)
946{
947 struct netdev_rx_queue *queue = to_rx_queue(kobj);
948#ifdef CONFIG_RPS
949 struct rps_map *map;
950 struct rps_dev_flow_table *flow_table;
951
952 map = rcu_dereference_protected(queue->rps_map, 1);
953 if (map) {
954 RCU_INIT_POINTER(queue->rps_map, NULL);
955 kfree_rcu(map, rcu);
956 }
957
958 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
959 if (flow_table) {
960 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
961 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
962 }
963#endif
964
965 memset(kobj, 0, sizeof(*kobj));
966 dev_put(queue->dev);
967}
968
969static const void *rx_queue_namespace(struct kobject *kobj)
970{
971 struct netdev_rx_queue *queue = to_rx_queue(kobj);
972 struct device *dev = &queue->dev->dev;
973 const void *ns = NULL;
974
975 if (dev->class && dev->class->ns_type)
976 ns = dev->class->namespace(dev);
977
978 return ns;
979}
980
981static void rx_queue_get_ownership(struct kobject *kobj,
982 kuid_t *uid, kgid_t *gid)
983{
984 const struct net *net = rx_queue_namespace(kobj);
985
986 net_ns_get_ownership(net, uid, gid);
987}
988
989static struct kobj_type rx_queue_ktype __ro_after_init = {
990 .sysfs_ops = &rx_queue_sysfs_ops,
991 .release = rx_queue_release,
992 .default_groups = rx_queue_default_groups,
993 .namespace = rx_queue_namespace,
994 .get_ownership = rx_queue_get_ownership,
995};
996
997static int rx_queue_add_kobject(struct net_device *dev, int index)
998{
999 struct netdev_rx_queue *queue = dev->_rx + index;
1000 struct kobject *kobj = &queue->kobj;
1001 int error = 0;
1002
1003 /* Kobject_put later will trigger rx_queue_release call which
1004 * decreases dev refcount: Take that reference here
1005 */
1006 dev_hold(queue->dev);
1007
1008 kobj->kset = dev->queues_kset;
1009 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1010 "rx-%u", index);
1011 if (error)
1012 goto err;
1013
1014 if (dev->sysfs_rx_queue_group) {
1015 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1016 if (error)
1017 goto err;
1018 }
1019
1020 kobject_uevent(kobj, KOBJ_ADD);
1021
1022 return error;
1023
1024err:
1025 kobject_put(kobj);
1026 return error;
1027}
1028
1029static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1030 kgid_t kgid)
1031{
1032 struct netdev_rx_queue *queue = dev->_rx + index;
1033 struct kobject *kobj = &queue->kobj;
1034 int error;
1035
1036 error = sysfs_change_owner(kobj, kuid, kgid);
1037 if (error)
1038 return error;
1039
1040 if (dev->sysfs_rx_queue_group)
1041 error = sysfs_group_change_owner(
1042 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1043
1044 return error;
1045}
1046#endif /* CONFIG_SYSFS */
1047
1048int
1049net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1050{
1051#ifdef CONFIG_SYSFS
1052 int i;
1053 int error = 0;
1054
1055#ifndef CONFIG_RPS
1056 if (!dev->sysfs_rx_queue_group)
1057 return 0;
1058#endif
1059 for (i = old_num; i < new_num; i++) {
1060 error = rx_queue_add_kobject(dev, i);
1061 if (error) {
1062 new_num = old_num;
1063 break;
1064 }
1065 }
1066
1067 while (--i >= new_num) {
1068 struct kobject *kobj = &dev->_rx[i].kobj;
1069
1070 if (!refcount_read(&dev_net(dev)->ns.count))
1071 kobj->uevent_suppress = 1;
1072 if (dev->sysfs_rx_queue_group)
1073 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1074 kobject_put(kobj);
1075 }
1076
1077 return error;
1078#else
1079 return 0;
1080#endif
1081}
1082
1083static int net_rx_queue_change_owner(struct net_device *dev, int num,
1084 kuid_t kuid, kgid_t kgid)
1085{
1086#ifdef CONFIG_SYSFS
1087 int error = 0;
1088 int i;
1089
1090#ifndef CONFIG_RPS
1091 if (!dev->sysfs_rx_queue_group)
1092 return 0;
1093#endif
1094 for (i = 0; i < num; i++) {
1095 error = rx_queue_change_owner(dev, i, kuid, kgid);
1096 if (error)
1097 break;
1098 }
1099
1100 return error;
1101#else
1102 return 0;
1103#endif
1104}
1105
1106#ifdef CONFIG_SYSFS
1107/*
1108 * netdev_queue sysfs structures and functions.
1109 */
1110struct netdev_queue_attribute {
1111 struct attribute attr;
1112 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1113 ssize_t (*store)(struct netdev_queue *queue,
1114 const char *buf, size_t len);
1115};
1116#define to_netdev_queue_attr(_attr) \
1117 container_of(_attr, struct netdev_queue_attribute, attr)
1118
1119#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1120
1121static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1122 struct attribute *attr, char *buf)
1123{
1124 const struct netdev_queue_attribute *attribute
1125 = to_netdev_queue_attr(attr);
1126 struct netdev_queue *queue = to_netdev_queue(kobj);
1127
1128 if (!attribute->show)
1129 return -EIO;
1130
1131 return attribute->show(queue, buf);
1132}
1133
1134static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1135 struct attribute *attr,
1136 const char *buf, size_t count)
1137{
1138 const struct netdev_queue_attribute *attribute
1139 = to_netdev_queue_attr(attr);
1140 struct netdev_queue *queue = to_netdev_queue(kobj);
1141
1142 if (!attribute->store)
1143 return -EIO;
1144
1145 return attribute->store(queue, buf, count);
1146}
1147
1148static const struct sysfs_ops netdev_queue_sysfs_ops = {
1149 .show = netdev_queue_attr_show,
1150 .store = netdev_queue_attr_store,
1151};
1152
1153static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1154{
1155 unsigned long trans_timeout;
1156
1157 spin_lock_irq(&queue->_xmit_lock);
1158 trans_timeout = queue->trans_timeout;
1159 spin_unlock_irq(&queue->_xmit_lock);
1160
1161 return sprintf(buf, fmt_ulong, trans_timeout);
1162}
1163
1164static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1165{
1166 struct net_device *dev = queue->dev;
1167 unsigned int i;
1168
1169 i = queue - dev->_tx;
1170 BUG_ON(i >= dev->num_tx_queues);
1171
1172 return i;
1173}
1174
1175static ssize_t traffic_class_show(struct netdev_queue *queue,
1176 char *buf)
1177{
1178 struct net_device *dev = queue->dev;
1179 int num_tc, tc;
1180 int index;
1181
1182 if (!netif_is_multiqueue(dev))
1183 return -ENOENT;
1184
1185 if (!rtnl_trylock())
1186 return restart_syscall();
1187
1188 index = get_netdev_queue_index(queue);
1189
1190 /* If queue belongs to subordinate dev use its TC mapping */
1191 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1192
1193 num_tc = dev->num_tc;
1194 tc = netdev_txq_to_tc(dev, index);
1195
1196 rtnl_unlock();
1197
1198 if (tc < 0)
1199 return -EINVAL;
1200
1201 /* We can report the traffic class one of two ways:
1202 * Subordinate device traffic classes are reported with the traffic
1203 * class first, and then the subordinate class so for example TC0 on
1204 * subordinate device 2 will be reported as "0-2". If the queue
1205 * belongs to the root device it will be reported with just the
1206 * traffic class, so just "0" for TC 0 for example.
1207 */
1208 return num_tc < 0 ? sprintf(buf, "%d%d\n", tc, num_tc) :
1209 sprintf(buf, "%d\n", tc);
1210}
1211
1212#ifdef CONFIG_XPS
1213static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1214 char *buf)
1215{
1216 return sprintf(buf, "%lu\n", queue->tx_maxrate);
1217}
1218
1219static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1220 const char *buf, size_t len)
1221{
1222 struct net_device *dev = queue->dev;
1223 int err, index = get_netdev_queue_index(queue);
1224 u32 rate = 0;
1225
1226 if (!capable(CAP_NET_ADMIN))
1227 return -EPERM;
1228
1229 err = kstrtou32(buf, 10, &rate);
1230 if (err < 0)
1231 return err;
1232
1233 if (!rtnl_trylock())
1234 return restart_syscall();
1235
1236 err = -EOPNOTSUPP;
1237 if (dev->netdev_ops->ndo_set_tx_maxrate)
1238 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1239
1240 rtnl_unlock();
1241 if (!err) {
1242 queue->tx_maxrate = rate;
1243 return len;
1244 }
1245 return err;
1246}
1247
1248static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1249 = __ATTR_RW(tx_maxrate);
1250#endif
1251
1252static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1253 = __ATTR_RO(tx_timeout);
1254
1255static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1256 = __ATTR_RO(traffic_class);
1257
1258#ifdef CONFIG_BQL
1259/*
1260 * Byte queue limits sysfs structures and functions.
1261 */
1262static ssize_t bql_show(char *buf, unsigned int value)
1263{
1264 return sprintf(buf, "%u\n", value);
1265}
1266
1267static ssize_t bql_set(const char *buf, const size_t count,
1268 unsigned int *pvalue)
1269{
1270 unsigned int value;
1271 int err;
1272
1273 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1274 value = DQL_MAX_LIMIT;
1275 } else {
1276 err = kstrtouint(buf, 10, &value);
1277 if (err < 0)
1278 return err;
1279 if (value > DQL_MAX_LIMIT)
1280 return -EINVAL;
1281 }
1282
1283 *pvalue = value;
1284
1285 return count;
1286}
1287
1288static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1289 char *buf)
1290{
1291 struct dql *dql = &queue->dql;
1292
1293 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1294}
1295
1296static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1297 const char *buf, size_t len)
1298{
1299 struct dql *dql = &queue->dql;
1300 unsigned int value;
1301 int err;
1302
1303 err = kstrtouint(buf, 10, &value);
1304 if (err < 0)
1305 return err;
1306
1307 dql->slack_hold_time = msecs_to_jiffies(value);
1308
1309 return len;
1310}
1311
1312static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1313 = __ATTR(hold_time, 0644,
1314 bql_show_hold_time, bql_set_hold_time);
1315
1316static ssize_t bql_show_inflight(struct netdev_queue *queue,
1317 char *buf)
1318{
1319 struct dql *dql = &queue->dql;
1320
1321 return sprintf(buf, "%u\n", dql->num_queued - dql->num_completed);
1322}
1323
1324static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1325 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1326
1327#define BQL_ATTR(NAME, FIELD) \
1328static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1329 char *buf) \
1330{ \
1331 return bql_show(buf, queue->dql.FIELD); \
1332} \
1333 \
1334static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1335 const char *buf, size_t len) \
1336{ \
1337 return bql_set(buf, len, &queue->dql.FIELD); \
1338} \
1339 \
1340static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1341 = __ATTR(NAME, 0644, \
1342 bql_show_ ## NAME, bql_set_ ## NAME)
1343
1344BQL_ATTR(limit, limit);
1345BQL_ATTR(limit_max, max_limit);
1346BQL_ATTR(limit_min, min_limit);
1347
1348static struct attribute *dql_attrs[] __ro_after_init = {
1349 &bql_limit_attribute.attr,
1350 &bql_limit_max_attribute.attr,
1351 &bql_limit_min_attribute.attr,
1352 &bql_hold_time_attribute.attr,
1353 &bql_inflight_attribute.attr,
1354 NULL
1355};
1356
1357static const struct attribute_group dql_group = {
1358 .name = "byte_queue_limits",
1359 .attrs = dql_attrs,
1360};
1361#endif /* CONFIG_BQL */
1362
1363#ifdef CONFIG_XPS
1364static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1365 int tc, char *buf, enum xps_map_type type)
1366{
1367 struct xps_dev_maps *dev_maps;
1368 unsigned long *mask;
1369 unsigned int nr_ids;
1370 int j, len;
1371
1372 rcu_read_lock();
1373 dev_maps = rcu_dereference(dev->xps_maps[type]);
1374
1375 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1376 * when dev_maps hasn't been allocated yet, to be backward compatible.
1377 */
1378 nr_ids = dev_maps ? dev_maps->nr_ids :
1379 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1380
1381 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1382 if (!mask) {
1383 rcu_read_unlock();
1384 return -ENOMEM;
1385 }
1386
1387 if (!dev_maps || tc >= dev_maps->num_tc)
1388 goto out_no_maps;
1389
1390 for (j = 0; j < nr_ids; j++) {
1391 int i, tci = j * dev_maps->num_tc + tc;
1392 struct xps_map *map;
1393
1394 map = rcu_dereference(dev_maps->attr_map[tci]);
1395 if (!map)
1396 continue;
1397
1398 for (i = map->len; i--;) {
1399 if (map->queues[i] == index) {
1400 set_bit(j, mask);
1401 break;
1402 }
1403 }
1404 }
1405out_no_maps:
1406 rcu_read_unlock();
1407
1408 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1409 bitmap_free(mask);
1410
1411 return len < PAGE_SIZE ? len : -EINVAL;
1412}
1413
1414static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1415{
1416 struct net_device *dev = queue->dev;
1417 unsigned int index;
1418 int len, tc;
1419
1420 if (!netif_is_multiqueue(dev))
1421 return -ENOENT;
1422
1423 index = get_netdev_queue_index(queue);
1424
1425 if (!rtnl_trylock())
1426 return restart_syscall();
1427
1428 /* If queue belongs to subordinate dev use its map */
1429 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1430
1431 tc = netdev_txq_to_tc(dev, index);
1432 if (tc < 0) {
1433 rtnl_unlock();
1434 return -EINVAL;
1435 }
1436
1437 /* Make sure the subordinate device can't be freed */
1438 get_device(&dev->dev);
1439 rtnl_unlock();
1440
1441 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1442
1443 put_device(&dev->dev);
1444 return len;
1445}
1446
1447static ssize_t xps_cpus_store(struct netdev_queue *queue,
1448 const char *buf, size_t len)
1449{
1450 struct net_device *dev = queue->dev;
1451 unsigned int index;
1452 cpumask_var_t mask;
1453 int err;
1454
1455 if (!netif_is_multiqueue(dev))
1456 return -ENOENT;
1457
1458 if (!capable(CAP_NET_ADMIN))
1459 return -EPERM;
1460
1461 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1462 return -ENOMEM;
1463
1464 index = get_netdev_queue_index(queue);
1465
1466 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1467 if (err) {
1468 free_cpumask_var(mask);
1469 return err;
1470 }
1471
1472 if (!rtnl_trylock()) {
1473 free_cpumask_var(mask);
1474 return restart_syscall();
1475 }
1476
1477 err = netif_set_xps_queue(dev, mask, index);
1478 rtnl_unlock();
1479
1480 free_cpumask_var(mask);
1481
1482 return err ? : len;
1483}
1484
1485static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1486 = __ATTR_RW(xps_cpus);
1487
1488static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1489{
1490 struct net_device *dev = queue->dev;
1491 unsigned int index;
1492 int tc;
1493
1494 index = get_netdev_queue_index(queue);
1495
1496 if (!rtnl_trylock())
1497 return restart_syscall();
1498
1499 tc = netdev_txq_to_tc(dev, index);
1500 rtnl_unlock();
1501 if (tc < 0)
1502 return -EINVAL;
1503
1504 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1505}
1506
1507static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1508 size_t len)
1509{
1510 struct net_device *dev = queue->dev;
1511 struct net *net = dev_net(dev);
1512 unsigned long *mask;
1513 unsigned int index;
1514 int err;
1515
1516 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1517 return -EPERM;
1518
1519 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1520 if (!mask)
1521 return -ENOMEM;
1522
1523 index = get_netdev_queue_index(queue);
1524
1525 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1526 if (err) {
1527 bitmap_free(mask);
1528 return err;
1529 }
1530
1531 if (!rtnl_trylock()) {
1532 bitmap_free(mask);
1533 return restart_syscall();
1534 }
1535
1536 cpus_read_lock();
1537 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1538 cpus_read_unlock();
1539
1540 rtnl_unlock();
1541
1542 bitmap_free(mask);
1543 return err ? : len;
1544}
1545
1546static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1547 = __ATTR_RW(xps_rxqs);
1548#endif /* CONFIG_XPS */
1549
1550static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1551 &queue_trans_timeout.attr,
1552 &queue_traffic_class.attr,
1553#ifdef CONFIG_XPS
1554 &xps_cpus_attribute.attr,
1555 &xps_rxqs_attribute.attr,
1556 &queue_tx_maxrate.attr,
1557#endif
1558 NULL
1559};
1560ATTRIBUTE_GROUPS(netdev_queue_default);
1561
1562static void netdev_queue_release(struct kobject *kobj)
1563{
1564 struct netdev_queue *queue = to_netdev_queue(kobj);
1565
1566 memset(kobj, 0, sizeof(*kobj));
1567 dev_put(queue->dev);
1568}
1569
1570static const void *netdev_queue_namespace(struct kobject *kobj)
1571{
1572 struct netdev_queue *queue = to_netdev_queue(kobj);
1573 struct device *dev = &queue->dev->dev;
1574 const void *ns = NULL;
1575
1576 if (dev->class && dev->class->ns_type)
1577 ns = dev->class->namespace(dev);
1578
1579 return ns;
1580}
1581
1582static void netdev_queue_get_ownership(struct kobject *kobj,
1583 kuid_t *uid, kgid_t *gid)
1584{
1585 const struct net *net = netdev_queue_namespace(kobj);
1586
1587 net_ns_get_ownership(net, uid, gid);
1588}
1589
1590static struct kobj_type netdev_queue_ktype __ro_after_init = {
1591 .sysfs_ops = &netdev_queue_sysfs_ops,
1592 .release = netdev_queue_release,
1593 .default_groups = netdev_queue_default_groups,
1594 .namespace = netdev_queue_namespace,
1595 .get_ownership = netdev_queue_get_ownership,
1596};
1597
1598static int netdev_queue_add_kobject(struct net_device *dev, int index)
1599{
1600 struct netdev_queue *queue = dev->_tx + index;
1601 struct kobject *kobj = &queue->kobj;
1602 int error = 0;
1603
1604 /* Kobject_put later will trigger netdev_queue_release call
1605 * which decreases dev refcount: Take that reference here
1606 */
1607 dev_hold(queue->dev);
1608
1609 kobj->kset = dev->queues_kset;
1610 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1611 "tx-%u", index);
1612 if (error)
1613 goto err;
1614
1615#ifdef CONFIG_BQL
1616 error = sysfs_create_group(kobj, &dql_group);
1617 if (error)
1618 goto err;
1619#endif
1620
1621 kobject_uevent(kobj, KOBJ_ADD);
1622 return 0;
1623
1624err:
1625 kobject_put(kobj);
1626 return error;
1627}
1628
1629static int tx_queue_change_owner(struct net_device *ndev, int index,
1630 kuid_t kuid, kgid_t kgid)
1631{
1632 struct netdev_queue *queue = ndev->_tx + index;
1633 struct kobject *kobj = &queue->kobj;
1634 int error;
1635
1636 error = sysfs_change_owner(kobj, kuid, kgid);
1637 if (error)
1638 return error;
1639
1640#ifdef CONFIG_BQL
1641 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1642#endif
1643 return error;
1644}
1645#endif /* CONFIG_SYSFS */
1646
1647int
1648netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1649{
1650#ifdef CONFIG_SYSFS
1651 int i;
1652 int error = 0;
1653
1654 for (i = old_num; i < new_num; i++) {
1655 error = netdev_queue_add_kobject(dev, i);
1656 if (error) {
1657 new_num = old_num;
1658 break;
1659 }
1660 }
1661
1662 while (--i >= new_num) {
1663 struct netdev_queue *queue = dev->_tx + i;
1664
1665 if (!refcount_read(&dev_net(dev)->ns.count))
1666 queue->kobj.uevent_suppress = 1;
1667#ifdef CONFIG_BQL
1668 sysfs_remove_group(&queue->kobj, &dql_group);
1669#endif
1670 kobject_put(&queue->kobj);
1671 }
1672
1673 return error;
1674#else
1675 return 0;
1676#endif /* CONFIG_SYSFS */
1677}
1678
1679static int net_tx_queue_change_owner(struct net_device *dev, int num,
1680 kuid_t kuid, kgid_t kgid)
1681{
1682#ifdef CONFIG_SYSFS
1683 int error = 0;
1684 int i;
1685
1686 for (i = 0; i < num; i++) {
1687 error = tx_queue_change_owner(dev, i, kuid, kgid);
1688 if (error)
1689 break;
1690 }
1691
1692 return error;
1693#else
1694 return 0;
1695#endif /* CONFIG_SYSFS */
1696}
1697
1698static int register_queue_kobjects(struct net_device *dev)
1699{
1700 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1701
1702#ifdef CONFIG_SYSFS
1703 dev->queues_kset = kset_create_and_add("queues",
1704 NULL, &dev->dev.kobj);
1705 if (!dev->queues_kset)
1706 return -ENOMEM;
1707 real_rx = dev->real_num_rx_queues;
1708#endif
1709 real_tx = dev->real_num_tx_queues;
1710
1711 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1712 if (error)
1713 goto error;
1714 rxq = real_rx;
1715
1716 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1717 if (error)
1718 goto error;
1719 txq = real_tx;
1720
1721 return 0;
1722
1723error:
1724 netdev_queue_update_kobjects(dev, txq, 0);
1725 net_rx_queue_update_kobjects(dev, rxq, 0);
1726#ifdef CONFIG_SYSFS
1727 kset_unregister(dev->queues_kset);
1728#endif
1729 return error;
1730}
1731
1732static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1733{
1734 int error = 0, real_rx = 0, real_tx = 0;
1735
1736#ifdef CONFIG_SYSFS
1737 if (ndev->queues_kset) {
1738 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1739 if (error)
1740 return error;
1741 }
1742 real_rx = ndev->real_num_rx_queues;
1743#endif
1744 real_tx = ndev->real_num_tx_queues;
1745
1746 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1747 if (error)
1748 return error;
1749
1750 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1751 if (error)
1752 return error;
1753
1754 return 0;
1755}
1756
1757static void remove_queue_kobjects(struct net_device *dev)
1758{
1759 int real_rx = 0, real_tx = 0;
1760
1761#ifdef CONFIG_SYSFS
1762 real_rx = dev->real_num_rx_queues;
1763#endif
1764 real_tx = dev->real_num_tx_queues;
1765
1766 net_rx_queue_update_kobjects(dev, real_rx, 0);
1767 netdev_queue_update_kobjects(dev, real_tx, 0);
1768#ifdef CONFIG_SYSFS
1769 kset_unregister(dev->queues_kset);
1770#endif
1771}
1772
1773static bool net_current_may_mount(void)
1774{
1775 struct net *net = current->nsproxy->net_ns;
1776
1777 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1778}
1779
1780static void *net_grab_current_ns(void)
1781{
1782 struct net *ns = current->nsproxy->net_ns;
1783#ifdef CONFIG_NET_NS
1784 if (ns)
1785 refcount_inc(&ns->passive);
1786#endif
1787 return ns;
1788}
1789
1790static const void *net_initial_ns(void)
1791{
1792 return &init_net;
1793}
1794
1795static const void *net_netlink_ns(struct sock *sk)
1796{
1797 return sock_net(sk);
1798}
1799
1800const struct kobj_ns_type_operations net_ns_type_operations = {
1801 .type = KOBJ_NS_TYPE_NET,
1802 .current_may_mount = net_current_may_mount,
1803 .grab_current_ns = net_grab_current_ns,
1804 .netlink_ns = net_netlink_ns,
1805 .initial_ns = net_initial_ns,
1806 .drop_ns = net_drop_ns,
1807};
1808EXPORT_SYMBOL_GPL(net_ns_type_operations);
1809
1810static int netdev_uevent(struct device *d, struct kobj_uevent_env *env)
1811{
1812 struct net_device *dev = to_net_dev(d);
1813 int retval;
1814
1815 /* pass interface to uevent. */
1816 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1817 if (retval)
1818 goto exit;
1819
1820 /* pass ifindex to uevent.
1821 * ifindex is useful as it won't change (interface name may change)
1822 * and is what RtNetlink uses natively.
1823 */
1824 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1825
1826exit:
1827 return retval;
1828}
1829
1830/*
1831 * netdev_release -- destroy and free a dead device.
1832 * Called when last reference to device kobject is gone.
1833 */
1834static void netdev_release(struct device *d)
1835{
1836 struct net_device *dev = to_net_dev(d);
1837
1838 BUG_ON(dev->reg_state != NETREG_RELEASED);
1839
1840 /* no need to wait for rcu grace period:
1841 * device is dead and about to be freed.
1842 */
1843 kfree(rcu_access_pointer(dev->ifalias));
1844 netdev_freemem(dev);
1845}
1846
1847static const void *net_namespace(struct device *d)
1848{
1849 struct net_device *dev = to_net_dev(d);
1850
1851 return dev_net(dev);
1852}
1853
1854static void net_get_ownership(struct device *d, kuid_t *uid, kgid_t *gid)
1855{
1856 struct net_device *dev = to_net_dev(d);
1857 const struct net *net = dev_net(dev);
1858
1859 net_ns_get_ownership(net, uid, gid);
1860}
1861
1862static struct class net_class __ro_after_init = {
1863 .name = "net",
1864 .dev_release = netdev_release,
1865 .dev_groups = net_class_groups,
1866 .dev_uevent = netdev_uevent,
1867 .ns_type = &net_ns_type_operations,
1868 .namespace = net_namespace,
1869 .get_ownership = net_get_ownership,
1870};
1871
1872#ifdef CONFIG_OF_NET
1873static int of_dev_node_match(struct device *dev, const void *data)
1874{
1875 for (; dev; dev = dev->parent) {
1876 if (dev->of_node == data)
1877 return 1;
1878 }
1879
1880 return 0;
1881}
1882
1883/*
1884 * of_find_net_device_by_node - lookup the net device for the device node
1885 * @np: OF device node
1886 *
1887 * Looks up the net_device structure corresponding with the device node.
1888 * If successful, returns a pointer to the net_device with the embedded
1889 * struct device refcount incremented by one, or NULL on failure. The
1890 * refcount must be dropped when done with the net_device.
1891 */
1892struct net_device *of_find_net_device_by_node(struct device_node *np)
1893{
1894 struct device *dev;
1895
1896 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
1897 if (!dev)
1898 return NULL;
1899
1900 return to_net_dev(dev);
1901}
1902EXPORT_SYMBOL(of_find_net_device_by_node);
1903#endif
1904
1905/* Delete sysfs entries but hold kobject reference until after all
1906 * netdev references are gone.
1907 */
1908void netdev_unregister_kobject(struct net_device *ndev)
1909{
1910 struct device *dev = &ndev->dev;
1911
1912 if (!refcount_read(&dev_net(ndev)->ns.count))
1913 dev_set_uevent_suppress(dev, 1);
1914
1915 kobject_get(&dev->kobj);
1916
1917 remove_queue_kobjects(ndev);
1918
1919 pm_runtime_set_memalloc_noio(dev, false);
1920
1921 device_del(dev);
1922}
1923
1924/* Create sysfs entries for network device. */
1925int netdev_register_kobject(struct net_device *ndev)
1926{
1927 struct device *dev = &ndev->dev;
1928 const struct attribute_group **groups = ndev->sysfs_groups;
1929 int error = 0;
1930
1931 device_initialize(dev);
1932 dev->class = &net_class;
1933 dev->platform_data = ndev;
1934 dev->groups = groups;
1935
1936 dev_set_name(dev, "%s", ndev->name);
1937
1938#ifdef CONFIG_SYSFS
1939 /* Allow for a device specific group */
1940 if (*groups)
1941 groups++;
1942
1943 *groups++ = &netstat_group;
1944
1945#if IS_ENABLED(CONFIG_WIRELESS_EXT) || IS_ENABLED(CONFIG_CFG80211)
1946 if (ndev->ieee80211_ptr)
1947 *groups++ = &wireless_group;
1948#if IS_ENABLED(CONFIG_WIRELESS_EXT)
1949 else if (ndev->wireless_handlers)
1950 *groups++ = &wireless_group;
1951#endif
1952#endif
1953#endif /* CONFIG_SYSFS */
1954
1955 error = device_add(dev);
1956 if (error)
1957 return error;
1958
1959 error = register_queue_kobjects(ndev);
1960 if (error) {
1961 device_del(dev);
1962 return error;
1963 }
1964
1965 pm_runtime_set_memalloc_noio(dev, true);
1966
1967 return error;
1968}
1969
1970/* Change owner for sysfs entries when moving network devices across network
1971 * namespaces owned by different user namespaces.
1972 */
1973int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
1974 const struct net *net_new)
1975{
1976 struct device *dev = &ndev->dev;
1977 kuid_t old_uid, new_uid;
1978 kgid_t old_gid, new_gid;
1979 int error;
1980
1981 net_ns_get_ownership(net_old, &old_uid, &old_gid);
1982 net_ns_get_ownership(net_new, &new_uid, &new_gid);
1983
1984 /* The network namespace was changed but the owning user namespace is
1985 * identical so there's no need to change the owner of sysfs entries.
1986 */
1987 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
1988 return 0;
1989
1990 error = device_change_owner(dev, new_uid, new_gid);
1991 if (error)
1992 return error;
1993
1994 error = queue_change_owner(ndev, new_uid, new_gid);
1995 if (error)
1996 return error;
1997
1998 return 0;
1999}
2000
2001int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2002 const void *ns)
2003{
2004 return class_create_file_ns(&net_class, class_attr, ns);
2005}
2006EXPORT_SYMBOL(netdev_class_create_file_ns);
2007
2008void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2009 const void *ns)
2010{
2011 class_remove_file_ns(&net_class, class_attr, ns);
2012}
2013EXPORT_SYMBOL(netdev_class_remove_file_ns);
2014
2015int __init netdev_kobject_init(void)
2016{
2017 kobj_ns_type_register(&net_ns_type_operations);
2018 return class_register(&net_class);
2019}
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#include <net/rps.h>
28
29#include "dev.h"
30#include "net-sysfs.h"
31
32#ifdef CONFIG_SYSFS
33static const char fmt_hex[] = "%#x\n";
34static const char fmt_dec[] = "%d\n";
35static const char fmt_ulong[] = "%lu\n";
36static const char fmt_u64[] = "%llu\n";
37
38/* Caller holds RTNL or RCU */
39static inline int dev_isalive(const struct net_device *dev)
40{
41 return READ_ONCE(dev->reg_state) <= NETREG_REGISTERED;
42}
43
44/* use same locking rules as GIF* ioctl's */
45static ssize_t netdev_show(const struct device *dev,
46 struct device_attribute *attr, char *buf,
47 ssize_t (*format)(const struct net_device *, char *))
48{
49 struct net_device *ndev = to_net_dev(dev);
50 ssize_t ret = -EINVAL;
51
52 rcu_read_lock();
53 if (dev_isalive(ndev))
54 ret = (*format)(ndev, buf);
55 rcu_read_unlock();
56
57 return ret;
58}
59
60/* generate a show function for simple field */
61#define NETDEVICE_SHOW(field, format_string) \
62static ssize_t format_##field(const struct net_device *dev, char *buf) \
63{ \
64 return sysfs_emit(buf, format_string, READ_ONCE(dev->field)); \
65} \
66static ssize_t field##_show(struct device *dev, \
67 struct device_attribute *attr, char *buf) \
68{ \
69 return netdev_show(dev, attr, buf, format_##field); \
70} \
71
72#define NETDEVICE_SHOW_RO(field, format_string) \
73NETDEVICE_SHOW(field, format_string); \
74static DEVICE_ATTR_RO(field)
75
76#define NETDEVICE_SHOW_RW(field, format_string) \
77NETDEVICE_SHOW(field, format_string); \
78static DEVICE_ATTR_RW(field)
79
80/* use same locking and permission rules as SIF* ioctl's */
81static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
82 const char *buf, size_t len,
83 int (*set)(struct net_device *, unsigned long))
84{
85 struct net_device *netdev = to_net_dev(dev);
86 struct net *net = dev_net(netdev);
87 unsigned long new;
88 int ret;
89
90 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
91 return -EPERM;
92
93 ret = kstrtoul(buf, 0, &new);
94 if (ret)
95 goto err;
96
97 if (!rtnl_trylock())
98 return restart_syscall();
99
100 if (dev_isalive(netdev)) {
101 ret = (*set)(netdev, new);
102 if (ret == 0)
103 ret = len;
104 }
105 rtnl_unlock();
106 err:
107 return ret;
108}
109
110NETDEVICE_SHOW_RO(dev_id, fmt_hex);
111NETDEVICE_SHOW_RO(dev_port, fmt_dec);
112NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
113NETDEVICE_SHOW_RO(addr_len, fmt_dec);
114NETDEVICE_SHOW_RO(ifindex, fmt_dec);
115NETDEVICE_SHOW_RO(type, fmt_dec);
116NETDEVICE_SHOW_RO(link_mode, fmt_dec);
117
118static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
119 char *buf)
120{
121 struct net_device *ndev = to_net_dev(dev);
122
123 return sysfs_emit(buf, fmt_dec, dev_get_iflink(ndev));
124}
125static DEVICE_ATTR_RO(iflink);
126
127static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
128{
129 return sysfs_emit(buf, fmt_dec, READ_ONCE(dev->name_assign_type));
130}
131
132static ssize_t name_assign_type_show(struct device *dev,
133 struct device_attribute *attr,
134 char *buf)
135{
136 struct net_device *ndev = to_net_dev(dev);
137 ssize_t ret = -EINVAL;
138
139 if (READ_ONCE(ndev->name_assign_type) != NET_NAME_UNKNOWN)
140 ret = netdev_show(dev, attr, buf, format_name_assign_type);
141
142 return ret;
143}
144static DEVICE_ATTR_RO(name_assign_type);
145
146/* use same locking rules as GIFHWADDR ioctl's (dev_get_mac_address()) */
147static ssize_t address_show(struct device *dev, struct device_attribute *attr,
148 char *buf)
149{
150 struct net_device *ndev = to_net_dev(dev);
151 ssize_t ret = -EINVAL;
152
153 down_read(&dev_addr_sem);
154
155 rcu_read_lock();
156 if (dev_isalive(ndev))
157 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
158 rcu_read_unlock();
159
160 up_read(&dev_addr_sem);
161 return ret;
162}
163static DEVICE_ATTR_RO(address);
164
165static ssize_t broadcast_show(struct device *dev,
166 struct device_attribute *attr, char *buf)
167{
168 struct net_device *ndev = to_net_dev(dev);
169 int ret = -EINVAL;
170
171 rcu_read_lock();
172 if (dev_isalive(ndev))
173 ret = sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
174 rcu_read_unlock();
175 return ret;
176}
177static DEVICE_ATTR_RO(broadcast);
178
179static int change_carrier(struct net_device *dev, unsigned long new_carrier)
180{
181 if (!netif_running(dev))
182 return -EINVAL;
183 return dev_change_carrier(dev, (bool)new_carrier);
184}
185
186static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
187 const char *buf, size_t len)
188{
189 struct net_device *netdev = to_net_dev(dev);
190
191 /* The check is also done in change_carrier; this helps returning early
192 * without hitting the trylock/restart in netdev_store.
193 */
194 if (!netdev->netdev_ops->ndo_change_carrier)
195 return -EOPNOTSUPP;
196
197 return netdev_store(dev, attr, buf, len, change_carrier);
198}
199
200static ssize_t carrier_show(struct device *dev,
201 struct device_attribute *attr, char *buf)
202{
203 struct net_device *netdev = to_net_dev(dev);
204 int ret = -EINVAL;
205
206 if (!rtnl_trylock())
207 return restart_syscall();
208
209 if (netif_running(netdev)) {
210 /* Synchronize carrier state with link watch,
211 * see also rtnl_getlink().
212 */
213 linkwatch_sync_dev(netdev);
214
215 ret = sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev));
216 }
217 rtnl_unlock();
218
219 return ret;
220}
221static DEVICE_ATTR_RW(carrier);
222
223static ssize_t speed_show(struct device *dev,
224 struct device_attribute *attr, char *buf)
225{
226 struct net_device *netdev = to_net_dev(dev);
227 int ret = -EINVAL;
228
229 /* The check is also done in __ethtool_get_link_ksettings; this helps
230 * returning early without hitting the trylock/restart below.
231 */
232 if (!netdev->ethtool_ops->get_link_ksettings)
233 return ret;
234
235 if (!rtnl_trylock())
236 return restart_syscall();
237
238 if (netif_running(netdev) && netif_device_present(netdev)) {
239 struct ethtool_link_ksettings cmd;
240
241 if (!__ethtool_get_link_ksettings(netdev, &cmd))
242 ret = sysfs_emit(buf, fmt_dec, cmd.base.speed);
243 }
244 rtnl_unlock();
245 return ret;
246}
247static DEVICE_ATTR_RO(speed);
248
249static ssize_t duplex_show(struct device *dev,
250 struct device_attribute *attr, char *buf)
251{
252 struct net_device *netdev = to_net_dev(dev);
253 int ret = -EINVAL;
254
255 /* The check is also done in __ethtool_get_link_ksettings; this helps
256 * returning early without hitting the trylock/restart below.
257 */
258 if (!netdev->ethtool_ops->get_link_ksettings)
259 return ret;
260
261 if (!rtnl_trylock())
262 return restart_syscall();
263
264 if (netif_running(netdev)) {
265 struct ethtool_link_ksettings cmd;
266
267 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
268 const char *duplex;
269
270 switch (cmd.base.duplex) {
271 case DUPLEX_HALF:
272 duplex = "half";
273 break;
274 case DUPLEX_FULL:
275 duplex = "full";
276 break;
277 default:
278 duplex = "unknown";
279 break;
280 }
281 ret = sysfs_emit(buf, "%s\n", duplex);
282 }
283 }
284 rtnl_unlock();
285 return ret;
286}
287static DEVICE_ATTR_RO(duplex);
288
289static ssize_t testing_show(struct device *dev,
290 struct device_attribute *attr, char *buf)
291{
292 struct net_device *netdev = to_net_dev(dev);
293
294 if (netif_running(netdev))
295 return sysfs_emit(buf, fmt_dec, !!netif_testing(netdev));
296
297 return -EINVAL;
298}
299static DEVICE_ATTR_RO(testing);
300
301static ssize_t dormant_show(struct device *dev,
302 struct device_attribute *attr, char *buf)
303{
304 struct net_device *netdev = to_net_dev(dev);
305
306 if (netif_running(netdev))
307 return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev));
308
309 return -EINVAL;
310}
311static DEVICE_ATTR_RO(dormant);
312
313static const char *const operstates[] = {
314 "unknown",
315 "notpresent", /* currently unused */
316 "down",
317 "lowerlayerdown",
318 "testing",
319 "dormant",
320 "up"
321};
322
323static ssize_t operstate_show(struct device *dev,
324 struct device_attribute *attr, char *buf)
325{
326 const struct net_device *netdev = to_net_dev(dev);
327 unsigned char operstate;
328
329 operstate = READ_ONCE(netdev->operstate);
330 if (!netif_running(netdev))
331 operstate = IF_OPER_DOWN;
332
333 if (operstate >= ARRAY_SIZE(operstates))
334 return -EINVAL; /* should not happen */
335
336 return sysfs_emit(buf, "%s\n", operstates[operstate]);
337}
338static DEVICE_ATTR_RO(operstate);
339
340static ssize_t carrier_changes_show(struct device *dev,
341 struct device_attribute *attr,
342 char *buf)
343{
344 struct net_device *netdev = to_net_dev(dev);
345
346 return sysfs_emit(buf, fmt_dec,
347 atomic_read(&netdev->carrier_up_count) +
348 atomic_read(&netdev->carrier_down_count));
349}
350static DEVICE_ATTR_RO(carrier_changes);
351
352static ssize_t carrier_up_count_show(struct device *dev,
353 struct device_attribute *attr,
354 char *buf)
355{
356 struct net_device *netdev = to_net_dev(dev);
357
358 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
359}
360static DEVICE_ATTR_RO(carrier_up_count);
361
362static ssize_t carrier_down_count_show(struct device *dev,
363 struct device_attribute *attr,
364 char *buf)
365{
366 struct net_device *netdev = to_net_dev(dev);
367
368 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
369}
370static DEVICE_ATTR_RO(carrier_down_count);
371
372/* read-write attributes */
373
374static int change_mtu(struct net_device *dev, unsigned long new_mtu)
375{
376 return dev_set_mtu(dev, (int)new_mtu);
377}
378
379static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
380 const char *buf, size_t len)
381{
382 return netdev_store(dev, attr, buf, len, change_mtu);
383}
384NETDEVICE_SHOW_RW(mtu, fmt_dec);
385
386static int change_flags(struct net_device *dev, unsigned long new_flags)
387{
388 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
389}
390
391static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
392 const char *buf, size_t len)
393{
394 return netdev_store(dev, attr, buf, len, change_flags);
395}
396NETDEVICE_SHOW_RW(flags, fmt_hex);
397
398static ssize_t tx_queue_len_store(struct device *dev,
399 struct device_attribute *attr,
400 const char *buf, size_t len)
401{
402 if (!capable(CAP_NET_ADMIN))
403 return -EPERM;
404
405 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
406}
407NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
408
409static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
410{
411 WRITE_ONCE(dev->gro_flush_timeout, val);
412 return 0;
413}
414
415static ssize_t gro_flush_timeout_store(struct device *dev,
416 struct device_attribute *attr,
417 const char *buf, size_t len)
418{
419 if (!capable(CAP_NET_ADMIN))
420 return -EPERM;
421
422 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
423}
424NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
425
426static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
427{
428 WRITE_ONCE(dev->napi_defer_hard_irqs, val);
429 return 0;
430}
431
432static ssize_t napi_defer_hard_irqs_store(struct device *dev,
433 struct device_attribute *attr,
434 const char *buf, size_t len)
435{
436 if (!capable(CAP_NET_ADMIN))
437 return -EPERM;
438
439 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
440}
441NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
442
443static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
444 const char *buf, size_t len)
445{
446 struct net_device *netdev = to_net_dev(dev);
447 struct net *net = dev_net(netdev);
448 size_t count = len;
449 ssize_t ret = 0;
450
451 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
452 return -EPERM;
453
454 /* ignore trailing newline */
455 if (len > 0 && buf[len - 1] == '\n')
456 --count;
457
458 if (!rtnl_trylock())
459 return restart_syscall();
460
461 if (dev_isalive(netdev)) {
462 ret = dev_set_alias(netdev, buf, count);
463 if (ret < 0)
464 goto err;
465 ret = len;
466 netdev_state_change(netdev);
467 }
468err:
469 rtnl_unlock();
470
471 return ret;
472}
473
474static ssize_t ifalias_show(struct device *dev,
475 struct device_attribute *attr, char *buf)
476{
477 const struct net_device *netdev = to_net_dev(dev);
478 char tmp[IFALIASZ];
479 ssize_t ret = 0;
480
481 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
482 if (ret > 0)
483 ret = sysfs_emit(buf, "%s\n", tmp);
484 return ret;
485}
486static DEVICE_ATTR_RW(ifalias);
487
488static int change_group(struct net_device *dev, unsigned long new_group)
489{
490 dev_set_group(dev, (int)new_group);
491 return 0;
492}
493
494static ssize_t group_store(struct device *dev, struct device_attribute *attr,
495 const char *buf, size_t len)
496{
497 return netdev_store(dev, attr, buf, len, change_group);
498}
499NETDEVICE_SHOW(group, fmt_dec);
500static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
501
502static int change_proto_down(struct net_device *dev, unsigned long proto_down)
503{
504 return dev_change_proto_down(dev, (bool)proto_down);
505}
506
507static ssize_t proto_down_store(struct device *dev,
508 struct device_attribute *attr,
509 const char *buf, size_t len)
510{
511 return netdev_store(dev, attr, buf, len, change_proto_down);
512}
513NETDEVICE_SHOW_RW(proto_down, fmt_dec);
514
515static ssize_t phys_port_id_show(struct device *dev,
516 struct device_attribute *attr, char *buf)
517{
518 struct net_device *netdev = to_net_dev(dev);
519 ssize_t ret = -EINVAL;
520
521 /* The check is also done in dev_get_phys_port_id; this helps returning
522 * early without hitting the trylock/restart below.
523 */
524 if (!netdev->netdev_ops->ndo_get_phys_port_id)
525 return -EOPNOTSUPP;
526
527 if (!rtnl_trylock())
528 return restart_syscall();
529
530 if (dev_isalive(netdev)) {
531 struct netdev_phys_item_id ppid;
532
533 ret = dev_get_phys_port_id(netdev, &ppid);
534 if (!ret)
535 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
536 }
537 rtnl_unlock();
538
539 return ret;
540}
541static DEVICE_ATTR_RO(phys_port_id);
542
543static ssize_t phys_port_name_show(struct device *dev,
544 struct device_attribute *attr, char *buf)
545{
546 struct net_device *netdev = to_net_dev(dev);
547 ssize_t ret = -EINVAL;
548
549 /* The checks are also done in dev_get_phys_port_name; this helps
550 * returning early without hitting the trylock/restart below.
551 */
552 if (!netdev->netdev_ops->ndo_get_phys_port_name &&
553 !netdev->devlink_port)
554 return -EOPNOTSUPP;
555
556 if (!rtnl_trylock())
557 return restart_syscall();
558
559 if (dev_isalive(netdev)) {
560 char name[IFNAMSIZ];
561
562 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
563 if (!ret)
564 ret = sysfs_emit(buf, "%s\n", name);
565 }
566 rtnl_unlock();
567
568 return ret;
569}
570static DEVICE_ATTR_RO(phys_port_name);
571
572static ssize_t phys_switch_id_show(struct device *dev,
573 struct device_attribute *attr, char *buf)
574{
575 struct net_device *netdev = to_net_dev(dev);
576 ssize_t ret = -EINVAL;
577
578 /* The checks are also done in dev_get_phys_port_name; this helps
579 * returning early without hitting the trylock/restart below. This works
580 * because recurse is false when calling dev_get_port_parent_id.
581 */
582 if (!netdev->netdev_ops->ndo_get_port_parent_id &&
583 !netdev->devlink_port)
584 return -EOPNOTSUPP;
585
586 if (!rtnl_trylock())
587 return restart_syscall();
588
589 if (dev_isalive(netdev)) {
590 struct netdev_phys_item_id ppid = { };
591
592 ret = dev_get_port_parent_id(netdev, &ppid, false);
593 if (!ret)
594 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
595 }
596 rtnl_unlock();
597
598 return ret;
599}
600static DEVICE_ATTR_RO(phys_switch_id);
601
602static ssize_t threaded_show(struct device *dev,
603 struct device_attribute *attr, char *buf)
604{
605 struct net_device *netdev = to_net_dev(dev);
606 ssize_t ret = -EINVAL;
607
608 if (!rtnl_trylock())
609 return restart_syscall();
610
611 if (dev_isalive(netdev))
612 ret = sysfs_emit(buf, fmt_dec, netdev->threaded);
613
614 rtnl_unlock();
615 return ret;
616}
617
618static int modify_napi_threaded(struct net_device *dev, unsigned long val)
619{
620 int ret;
621
622 if (list_empty(&dev->napi_list))
623 return -EOPNOTSUPP;
624
625 if (val != 0 && val != 1)
626 return -EOPNOTSUPP;
627
628 ret = dev_set_threaded(dev, val);
629
630 return ret;
631}
632
633static ssize_t threaded_store(struct device *dev,
634 struct device_attribute *attr,
635 const char *buf, size_t len)
636{
637 return netdev_store(dev, attr, buf, len, modify_napi_threaded);
638}
639static DEVICE_ATTR_RW(threaded);
640
641static struct attribute *net_class_attrs[] __ro_after_init = {
642 &dev_attr_netdev_group.attr,
643 &dev_attr_type.attr,
644 &dev_attr_dev_id.attr,
645 &dev_attr_dev_port.attr,
646 &dev_attr_iflink.attr,
647 &dev_attr_ifindex.attr,
648 &dev_attr_name_assign_type.attr,
649 &dev_attr_addr_assign_type.attr,
650 &dev_attr_addr_len.attr,
651 &dev_attr_link_mode.attr,
652 &dev_attr_address.attr,
653 &dev_attr_broadcast.attr,
654 &dev_attr_speed.attr,
655 &dev_attr_duplex.attr,
656 &dev_attr_dormant.attr,
657 &dev_attr_testing.attr,
658 &dev_attr_operstate.attr,
659 &dev_attr_carrier_changes.attr,
660 &dev_attr_ifalias.attr,
661 &dev_attr_carrier.attr,
662 &dev_attr_mtu.attr,
663 &dev_attr_flags.attr,
664 &dev_attr_tx_queue_len.attr,
665 &dev_attr_gro_flush_timeout.attr,
666 &dev_attr_napi_defer_hard_irqs.attr,
667 &dev_attr_phys_port_id.attr,
668 &dev_attr_phys_port_name.attr,
669 &dev_attr_phys_switch_id.attr,
670 &dev_attr_proto_down.attr,
671 &dev_attr_carrier_up_count.attr,
672 &dev_attr_carrier_down_count.attr,
673 &dev_attr_threaded.attr,
674 NULL,
675};
676ATTRIBUTE_GROUPS(net_class);
677
678/* Show a given an attribute in the statistics group */
679static ssize_t netstat_show(const struct device *d,
680 struct device_attribute *attr, char *buf,
681 unsigned long offset)
682{
683 struct net_device *dev = to_net_dev(d);
684 ssize_t ret = -EINVAL;
685
686 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
687 offset % sizeof(u64) != 0);
688
689 rcu_read_lock();
690 if (dev_isalive(dev)) {
691 struct rtnl_link_stats64 temp;
692 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
693
694 ret = sysfs_emit(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
695 }
696 rcu_read_unlock();
697 return ret;
698}
699
700/* generate a read-only statistics attribute */
701#define NETSTAT_ENTRY(name) \
702static ssize_t name##_show(struct device *d, \
703 struct device_attribute *attr, char *buf) \
704{ \
705 return netstat_show(d, attr, buf, \
706 offsetof(struct rtnl_link_stats64, name)); \
707} \
708static DEVICE_ATTR_RO(name)
709
710NETSTAT_ENTRY(rx_packets);
711NETSTAT_ENTRY(tx_packets);
712NETSTAT_ENTRY(rx_bytes);
713NETSTAT_ENTRY(tx_bytes);
714NETSTAT_ENTRY(rx_errors);
715NETSTAT_ENTRY(tx_errors);
716NETSTAT_ENTRY(rx_dropped);
717NETSTAT_ENTRY(tx_dropped);
718NETSTAT_ENTRY(multicast);
719NETSTAT_ENTRY(collisions);
720NETSTAT_ENTRY(rx_length_errors);
721NETSTAT_ENTRY(rx_over_errors);
722NETSTAT_ENTRY(rx_crc_errors);
723NETSTAT_ENTRY(rx_frame_errors);
724NETSTAT_ENTRY(rx_fifo_errors);
725NETSTAT_ENTRY(rx_missed_errors);
726NETSTAT_ENTRY(tx_aborted_errors);
727NETSTAT_ENTRY(tx_carrier_errors);
728NETSTAT_ENTRY(tx_fifo_errors);
729NETSTAT_ENTRY(tx_heartbeat_errors);
730NETSTAT_ENTRY(tx_window_errors);
731NETSTAT_ENTRY(rx_compressed);
732NETSTAT_ENTRY(tx_compressed);
733NETSTAT_ENTRY(rx_nohandler);
734
735static struct attribute *netstat_attrs[] __ro_after_init = {
736 &dev_attr_rx_packets.attr,
737 &dev_attr_tx_packets.attr,
738 &dev_attr_rx_bytes.attr,
739 &dev_attr_tx_bytes.attr,
740 &dev_attr_rx_errors.attr,
741 &dev_attr_tx_errors.attr,
742 &dev_attr_rx_dropped.attr,
743 &dev_attr_tx_dropped.attr,
744 &dev_attr_multicast.attr,
745 &dev_attr_collisions.attr,
746 &dev_attr_rx_length_errors.attr,
747 &dev_attr_rx_over_errors.attr,
748 &dev_attr_rx_crc_errors.attr,
749 &dev_attr_rx_frame_errors.attr,
750 &dev_attr_rx_fifo_errors.attr,
751 &dev_attr_rx_missed_errors.attr,
752 &dev_attr_tx_aborted_errors.attr,
753 &dev_attr_tx_carrier_errors.attr,
754 &dev_attr_tx_fifo_errors.attr,
755 &dev_attr_tx_heartbeat_errors.attr,
756 &dev_attr_tx_window_errors.attr,
757 &dev_attr_rx_compressed.attr,
758 &dev_attr_tx_compressed.attr,
759 &dev_attr_rx_nohandler.attr,
760 NULL
761};
762
763static const struct attribute_group netstat_group = {
764 .name = "statistics",
765 .attrs = netstat_attrs,
766};
767
768static struct attribute *wireless_attrs[] = {
769 NULL
770};
771
772static const struct attribute_group wireless_group = {
773 .name = "wireless",
774 .attrs = wireless_attrs,
775};
776
777static bool wireless_group_needed(struct net_device *ndev)
778{
779#if IS_ENABLED(CONFIG_CFG80211)
780 if (ndev->ieee80211_ptr)
781 return true;
782#endif
783#if IS_ENABLED(CONFIG_WIRELESS_EXT)
784 if (ndev->wireless_handlers)
785 return true;
786#endif
787 return false;
788}
789
790#else /* CONFIG_SYSFS */
791#define net_class_groups NULL
792#endif /* CONFIG_SYSFS */
793
794#ifdef CONFIG_SYSFS
795#define to_rx_queue_attr(_attr) \
796 container_of(_attr, struct rx_queue_attribute, attr)
797
798#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
799
800static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
801 char *buf)
802{
803 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
804 struct netdev_rx_queue *queue = to_rx_queue(kobj);
805
806 if (!attribute->show)
807 return -EIO;
808
809 return attribute->show(queue, buf);
810}
811
812static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
813 const char *buf, size_t count)
814{
815 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
816 struct netdev_rx_queue *queue = to_rx_queue(kobj);
817
818 if (!attribute->store)
819 return -EIO;
820
821 return attribute->store(queue, buf, count);
822}
823
824static const struct sysfs_ops rx_queue_sysfs_ops = {
825 .show = rx_queue_attr_show,
826 .store = rx_queue_attr_store,
827};
828
829#ifdef CONFIG_RPS
830static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
831{
832 struct rps_map *map;
833 cpumask_var_t mask;
834 int i, len;
835
836 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
837 return -ENOMEM;
838
839 rcu_read_lock();
840 map = rcu_dereference(queue->rps_map);
841 if (map)
842 for (i = 0; i < map->len; i++)
843 cpumask_set_cpu(map->cpus[i], mask);
844
845 len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask));
846 rcu_read_unlock();
847 free_cpumask_var(mask);
848
849 return len < PAGE_SIZE ? len : -EINVAL;
850}
851
852static int netdev_rx_queue_set_rps_mask(struct netdev_rx_queue *queue,
853 cpumask_var_t mask)
854{
855 static DEFINE_MUTEX(rps_map_mutex);
856 struct rps_map *old_map, *map;
857 int cpu, i;
858
859 map = kzalloc(max_t(unsigned int,
860 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
861 GFP_KERNEL);
862 if (!map)
863 return -ENOMEM;
864
865 i = 0;
866 for_each_cpu_and(cpu, mask, cpu_online_mask)
867 map->cpus[i++] = cpu;
868
869 if (i) {
870 map->len = i;
871 } else {
872 kfree(map);
873 map = NULL;
874 }
875
876 mutex_lock(&rps_map_mutex);
877 old_map = rcu_dereference_protected(queue->rps_map,
878 mutex_is_locked(&rps_map_mutex));
879 rcu_assign_pointer(queue->rps_map, map);
880
881 if (map)
882 static_branch_inc(&rps_needed);
883 if (old_map)
884 static_branch_dec(&rps_needed);
885
886 mutex_unlock(&rps_map_mutex);
887
888 if (old_map)
889 kfree_rcu(old_map, rcu);
890 return 0;
891}
892
893int rps_cpumask_housekeeping(struct cpumask *mask)
894{
895 if (!cpumask_empty(mask)) {
896 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
897 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
898 if (cpumask_empty(mask))
899 return -EINVAL;
900 }
901 return 0;
902}
903
904static ssize_t store_rps_map(struct netdev_rx_queue *queue,
905 const char *buf, size_t len)
906{
907 cpumask_var_t mask;
908 int err;
909
910 if (!capable(CAP_NET_ADMIN))
911 return -EPERM;
912
913 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
914 return -ENOMEM;
915
916 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
917 if (err)
918 goto out;
919
920 err = rps_cpumask_housekeeping(mask);
921 if (err)
922 goto out;
923
924 err = netdev_rx_queue_set_rps_mask(queue, mask);
925
926out:
927 free_cpumask_var(mask);
928 return err ? : len;
929}
930
931static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
932 char *buf)
933{
934 struct rps_dev_flow_table *flow_table;
935 unsigned long val = 0;
936
937 rcu_read_lock();
938 flow_table = rcu_dereference(queue->rps_flow_table);
939 if (flow_table)
940 val = (unsigned long)flow_table->mask + 1;
941 rcu_read_unlock();
942
943 return sysfs_emit(buf, "%lu\n", val);
944}
945
946static void rps_dev_flow_table_release(struct rcu_head *rcu)
947{
948 struct rps_dev_flow_table *table = container_of(rcu,
949 struct rps_dev_flow_table, rcu);
950 vfree(table);
951}
952
953static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
954 const char *buf, size_t len)
955{
956 unsigned long mask, count;
957 struct rps_dev_flow_table *table, *old_table;
958 static DEFINE_SPINLOCK(rps_dev_flow_lock);
959 int rc;
960
961 if (!capable(CAP_NET_ADMIN))
962 return -EPERM;
963
964 rc = kstrtoul(buf, 0, &count);
965 if (rc < 0)
966 return rc;
967
968 if (count) {
969 mask = count - 1;
970 /* mask = roundup_pow_of_two(count) - 1;
971 * without overflows...
972 */
973 while ((mask | (mask >> 1)) != mask)
974 mask |= (mask >> 1);
975 /* On 64 bit arches, must check mask fits in table->mask (u32),
976 * and on 32bit arches, must check
977 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
978 */
979#if BITS_PER_LONG > 32
980 if (mask > (unsigned long)(u32)mask)
981 return -EINVAL;
982#else
983 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
984 / sizeof(struct rps_dev_flow)) {
985 /* Enforce a limit to prevent overflow */
986 return -EINVAL;
987 }
988#endif
989 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
990 if (!table)
991 return -ENOMEM;
992
993 table->mask = mask;
994 for (count = 0; count <= mask; count++)
995 table->flows[count].cpu = RPS_NO_CPU;
996 } else {
997 table = NULL;
998 }
999
1000 spin_lock(&rps_dev_flow_lock);
1001 old_table = rcu_dereference_protected(queue->rps_flow_table,
1002 lockdep_is_held(&rps_dev_flow_lock));
1003 rcu_assign_pointer(queue->rps_flow_table, table);
1004 spin_unlock(&rps_dev_flow_lock);
1005
1006 if (old_table)
1007 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
1008
1009 return len;
1010}
1011
1012static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
1013 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
1014
1015static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
1016 = __ATTR(rps_flow_cnt, 0644,
1017 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
1018#endif /* CONFIG_RPS */
1019
1020static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
1021#ifdef CONFIG_RPS
1022 &rps_cpus_attribute.attr,
1023 &rps_dev_flow_table_cnt_attribute.attr,
1024#endif
1025 NULL
1026};
1027ATTRIBUTE_GROUPS(rx_queue_default);
1028
1029static void rx_queue_release(struct kobject *kobj)
1030{
1031 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1032#ifdef CONFIG_RPS
1033 struct rps_map *map;
1034 struct rps_dev_flow_table *flow_table;
1035
1036 map = rcu_dereference_protected(queue->rps_map, 1);
1037 if (map) {
1038 RCU_INIT_POINTER(queue->rps_map, NULL);
1039 kfree_rcu(map, rcu);
1040 }
1041
1042 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1043 if (flow_table) {
1044 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1045 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1046 }
1047#endif
1048
1049 memset(kobj, 0, sizeof(*kobj));
1050 netdev_put(queue->dev, &queue->dev_tracker);
1051}
1052
1053static const void *rx_queue_namespace(const struct kobject *kobj)
1054{
1055 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1056 struct device *dev = &queue->dev->dev;
1057 const void *ns = NULL;
1058
1059 if (dev->class && dev->class->ns_type)
1060 ns = dev->class->namespace(dev);
1061
1062 return ns;
1063}
1064
1065static void rx_queue_get_ownership(const struct kobject *kobj,
1066 kuid_t *uid, kgid_t *gid)
1067{
1068 const struct net *net = rx_queue_namespace(kobj);
1069
1070 net_ns_get_ownership(net, uid, gid);
1071}
1072
1073static const struct kobj_type rx_queue_ktype = {
1074 .sysfs_ops = &rx_queue_sysfs_ops,
1075 .release = rx_queue_release,
1076 .default_groups = rx_queue_default_groups,
1077 .namespace = rx_queue_namespace,
1078 .get_ownership = rx_queue_get_ownership,
1079};
1080
1081static int rx_queue_default_mask(struct net_device *dev,
1082 struct netdev_rx_queue *queue)
1083{
1084#if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL)
1085 struct cpumask *rps_default_mask = READ_ONCE(dev_net(dev)->core.rps_default_mask);
1086
1087 if (rps_default_mask && !cpumask_empty(rps_default_mask))
1088 return netdev_rx_queue_set_rps_mask(queue, rps_default_mask);
1089#endif
1090 return 0;
1091}
1092
1093static int rx_queue_add_kobject(struct net_device *dev, int index)
1094{
1095 struct netdev_rx_queue *queue = dev->_rx + index;
1096 struct kobject *kobj = &queue->kobj;
1097 int error = 0;
1098
1099 /* Kobject_put later will trigger rx_queue_release call which
1100 * decreases dev refcount: Take that reference here
1101 */
1102 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1103
1104 kobj->kset = dev->queues_kset;
1105 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1106 "rx-%u", index);
1107 if (error)
1108 goto err;
1109
1110 if (dev->sysfs_rx_queue_group) {
1111 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1112 if (error)
1113 goto err;
1114 }
1115
1116 error = rx_queue_default_mask(dev, queue);
1117 if (error)
1118 goto err;
1119
1120 kobject_uevent(kobj, KOBJ_ADD);
1121
1122 return error;
1123
1124err:
1125 kobject_put(kobj);
1126 return error;
1127}
1128
1129static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1130 kgid_t kgid)
1131{
1132 struct netdev_rx_queue *queue = dev->_rx + index;
1133 struct kobject *kobj = &queue->kobj;
1134 int error;
1135
1136 error = sysfs_change_owner(kobj, kuid, kgid);
1137 if (error)
1138 return error;
1139
1140 if (dev->sysfs_rx_queue_group)
1141 error = sysfs_group_change_owner(
1142 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1143
1144 return error;
1145}
1146#endif /* CONFIG_SYSFS */
1147
1148int
1149net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1150{
1151#ifdef CONFIG_SYSFS
1152 int i;
1153 int error = 0;
1154
1155#ifndef CONFIG_RPS
1156 if (!dev->sysfs_rx_queue_group)
1157 return 0;
1158#endif
1159 for (i = old_num; i < new_num; i++) {
1160 error = rx_queue_add_kobject(dev, i);
1161 if (error) {
1162 new_num = old_num;
1163 break;
1164 }
1165 }
1166
1167 while (--i >= new_num) {
1168 struct kobject *kobj = &dev->_rx[i].kobj;
1169
1170 if (!refcount_read(&dev_net(dev)->ns.count))
1171 kobj->uevent_suppress = 1;
1172 if (dev->sysfs_rx_queue_group)
1173 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1174 kobject_put(kobj);
1175 }
1176
1177 return error;
1178#else
1179 return 0;
1180#endif
1181}
1182
1183static int net_rx_queue_change_owner(struct net_device *dev, int num,
1184 kuid_t kuid, kgid_t kgid)
1185{
1186#ifdef CONFIG_SYSFS
1187 int error = 0;
1188 int i;
1189
1190#ifndef CONFIG_RPS
1191 if (!dev->sysfs_rx_queue_group)
1192 return 0;
1193#endif
1194 for (i = 0; i < num; i++) {
1195 error = rx_queue_change_owner(dev, i, kuid, kgid);
1196 if (error)
1197 break;
1198 }
1199
1200 return error;
1201#else
1202 return 0;
1203#endif
1204}
1205
1206#ifdef CONFIG_SYSFS
1207/*
1208 * netdev_queue sysfs structures and functions.
1209 */
1210struct netdev_queue_attribute {
1211 struct attribute attr;
1212 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1213 ssize_t (*store)(struct netdev_queue *queue,
1214 const char *buf, size_t len);
1215};
1216#define to_netdev_queue_attr(_attr) \
1217 container_of(_attr, struct netdev_queue_attribute, attr)
1218
1219#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1220
1221static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1222 struct attribute *attr, char *buf)
1223{
1224 const struct netdev_queue_attribute *attribute
1225 = to_netdev_queue_attr(attr);
1226 struct netdev_queue *queue = to_netdev_queue(kobj);
1227
1228 if (!attribute->show)
1229 return -EIO;
1230
1231 return attribute->show(queue, buf);
1232}
1233
1234static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1235 struct attribute *attr,
1236 const char *buf, size_t count)
1237{
1238 const struct netdev_queue_attribute *attribute
1239 = to_netdev_queue_attr(attr);
1240 struct netdev_queue *queue = to_netdev_queue(kobj);
1241
1242 if (!attribute->store)
1243 return -EIO;
1244
1245 return attribute->store(queue, buf, count);
1246}
1247
1248static const struct sysfs_ops netdev_queue_sysfs_ops = {
1249 .show = netdev_queue_attr_show,
1250 .store = netdev_queue_attr_store,
1251};
1252
1253static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1254{
1255 unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1256
1257 return sysfs_emit(buf, fmt_ulong, trans_timeout);
1258}
1259
1260static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1261{
1262 struct net_device *dev = queue->dev;
1263 unsigned int i;
1264
1265 i = queue - dev->_tx;
1266 BUG_ON(i >= dev->num_tx_queues);
1267
1268 return i;
1269}
1270
1271static ssize_t traffic_class_show(struct netdev_queue *queue,
1272 char *buf)
1273{
1274 struct net_device *dev = queue->dev;
1275 int num_tc, tc;
1276 int index;
1277
1278 if (!netif_is_multiqueue(dev))
1279 return -ENOENT;
1280
1281 if (!rtnl_trylock())
1282 return restart_syscall();
1283
1284 index = get_netdev_queue_index(queue);
1285
1286 /* If queue belongs to subordinate dev use its TC mapping */
1287 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1288
1289 num_tc = dev->num_tc;
1290 tc = netdev_txq_to_tc(dev, index);
1291
1292 rtnl_unlock();
1293
1294 if (tc < 0)
1295 return -EINVAL;
1296
1297 /* We can report the traffic class one of two ways:
1298 * Subordinate device traffic classes are reported with the traffic
1299 * class first, and then the subordinate class so for example TC0 on
1300 * subordinate device 2 will be reported as "0-2". If the queue
1301 * belongs to the root device it will be reported with just the
1302 * traffic class, so just "0" for TC 0 for example.
1303 */
1304 return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) :
1305 sysfs_emit(buf, "%d\n", tc);
1306}
1307
1308#ifdef CONFIG_XPS
1309static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1310 char *buf)
1311{
1312 return sysfs_emit(buf, "%lu\n", queue->tx_maxrate);
1313}
1314
1315static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1316 const char *buf, size_t len)
1317{
1318 struct net_device *dev = queue->dev;
1319 int err, index = get_netdev_queue_index(queue);
1320 u32 rate = 0;
1321
1322 if (!capable(CAP_NET_ADMIN))
1323 return -EPERM;
1324
1325 /* The check is also done later; this helps returning early without
1326 * hitting the trylock/restart below.
1327 */
1328 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1329 return -EOPNOTSUPP;
1330
1331 err = kstrtou32(buf, 10, &rate);
1332 if (err < 0)
1333 return err;
1334
1335 if (!rtnl_trylock())
1336 return restart_syscall();
1337
1338 err = -EOPNOTSUPP;
1339 if (dev->netdev_ops->ndo_set_tx_maxrate)
1340 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1341
1342 rtnl_unlock();
1343 if (!err) {
1344 queue->tx_maxrate = rate;
1345 return len;
1346 }
1347 return err;
1348}
1349
1350static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1351 = __ATTR_RW(tx_maxrate);
1352#endif
1353
1354static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1355 = __ATTR_RO(tx_timeout);
1356
1357static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1358 = __ATTR_RO(traffic_class);
1359
1360#ifdef CONFIG_BQL
1361/*
1362 * Byte queue limits sysfs structures and functions.
1363 */
1364static ssize_t bql_show(char *buf, unsigned int value)
1365{
1366 return sysfs_emit(buf, "%u\n", value);
1367}
1368
1369static ssize_t bql_set(const char *buf, const size_t count,
1370 unsigned int *pvalue)
1371{
1372 unsigned int value;
1373 int err;
1374
1375 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1376 value = DQL_MAX_LIMIT;
1377 } else {
1378 err = kstrtouint(buf, 10, &value);
1379 if (err < 0)
1380 return err;
1381 if (value > DQL_MAX_LIMIT)
1382 return -EINVAL;
1383 }
1384
1385 *pvalue = value;
1386
1387 return count;
1388}
1389
1390static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1391 char *buf)
1392{
1393 struct dql *dql = &queue->dql;
1394
1395 return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1396}
1397
1398static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1399 const char *buf, size_t len)
1400{
1401 struct dql *dql = &queue->dql;
1402 unsigned int value;
1403 int err;
1404
1405 err = kstrtouint(buf, 10, &value);
1406 if (err < 0)
1407 return err;
1408
1409 dql->slack_hold_time = msecs_to_jiffies(value);
1410
1411 return len;
1412}
1413
1414static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1415 = __ATTR(hold_time, 0644,
1416 bql_show_hold_time, bql_set_hold_time);
1417
1418static ssize_t bql_show_stall_thrs(struct netdev_queue *queue, char *buf)
1419{
1420 struct dql *dql = &queue->dql;
1421
1422 return sprintf(buf, "%u\n", jiffies_to_msecs(dql->stall_thrs));
1423}
1424
1425static ssize_t bql_set_stall_thrs(struct netdev_queue *queue,
1426 const char *buf, size_t len)
1427{
1428 struct dql *dql = &queue->dql;
1429 unsigned int value;
1430 int err;
1431
1432 err = kstrtouint(buf, 10, &value);
1433 if (err < 0)
1434 return err;
1435
1436 value = msecs_to_jiffies(value);
1437 if (value && (value < 4 || value > 4 / 2 * BITS_PER_LONG))
1438 return -ERANGE;
1439
1440 if (!dql->stall_thrs && value)
1441 dql->last_reap = jiffies;
1442 /* Force last_reap to be live */
1443 smp_wmb();
1444 dql->stall_thrs = value;
1445
1446 return len;
1447}
1448
1449static struct netdev_queue_attribute bql_stall_thrs_attribute __ro_after_init =
1450 __ATTR(stall_thrs, 0644, bql_show_stall_thrs, bql_set_stall_thrs);
1451
1452static ssize_t bql_show_stall_max(struct netdev_queue *queue, char *buf)
1453{
1454 return sprintf(buf, "%u\n", READ_ONCE(queue->dql.stall_max));
1455}
1456
1457static ssize_t bql_set_stall_max(struct netdev_queue *queue,
1458 const char *buf, size_t len)
1459{
1460 WRITE_ONCE(queue->dql.stall_max, 0);
1461 return len;
1462}
1463
1464static struct netdev_queue_attribute bql_stall_max_attribute __ro_after_init =
1465 __ATTR(stall_max, 0644, bql_show_stall_max, bql_set_stall_max);
1466
1467static ssize_t bql_show_stall_cnt(struct netdev_queue *queue, char *buf)
1468{
1469 struct dql *dql = &queue->dql;
1470
1471 return sprintf(buf, "%lu\n", dql->stall_cnt);
1472}
1473
1474static struct netdev_queue_attribute bql_stall_cnt_attribute __ro_after_init =
1475 __ATTR(stall_cnt, 0444, bql_show_stall_cnt, NULL);
1476
1477static ssize_t bql_show_inflight(struct netdev_queue *queue,
1478 char *buf)
1479{
1480 struct dql *dql = &queue->dql;
1481
1482 return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed);
1483}
1484
1485static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1486 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1487
1488#define BQL_ATTR(NAME, FIELD) \
1489static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1490 char *buf) \
1491{ \
1492 return bql_show(buf, queue->dql.FIELD); \
1493} \
1494 \
1495static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1496 const char *buf, size_t len) \
1497{ \
1498 return bql_set(buf, len, &queue->dql.FIELD); \
1499} \
1500 \
1501static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1502 = __ATTR(NAME, 0644, \
1503 bql_show_ ## NAME, bql_set_ ## NAME)
1504
1505BQL_ATTR(limit, limit);
1506BQL_ATTR(limit_max, max_limit);
1507BQL_ATTR(limit_min, min_limit);
1508
1509static struct attribute *dql_attrs[] __ro_after_init = {
1510 &bql_limit_attribute.attr,
1511 &bql_limit_max_attribute.attr,
1512 &bql_limit_min_attribute.attr,
1513 &bql_hold_time_attribute.attr,
1514 &bql_inflight_attribute.attr,
1515 &bql_stall_thrs_attribute.attr,
1516 &bql_stall_cnt_attribute.attr,
1517 &bql_stall_max_attribute.attr,
1518 NULL
1519};
1520
1521static const struct attribute_group dql_group = {
1522 .name = "byte_queue_limits",
1523 .attrs = dql_attrs,
1524};
1525#else
1526/* Fake declaration, all the code using it should be dead */
1527extern const struct attribute_group dql_group;
1528#endif /* CONFIG_BQL */
1529
1530#ifdef CONFIG_XPS
1531static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1532 int tc, char *buf, enum xps_map_type type)
1533{
1534 struct xps_dev_maps *dev_maps;
1535 unsigned long *mask;
1536 unsigned int nr_ids;
1537 int j, len;
1538
1539 rcu_read_lock();
1540 dev_maps = rcu_dereference(dev->xps_maps[type]);
1541
1542 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1543 * when dev_maps hasn't been allocated yet, to be backward compatible.
1544 */
1545 nr_ids = dev_maps ? dev_maps->nr_ids :
1546 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1547
1548 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1549 if (!mask) {
1550 rcu_read_unlock();
1551 return -ENOMEM;
1552 }
1553
1554 if (!dev_maps || tc >= dev_maps->num_tc)
1555 goto out_no_maps;
1556
1557 for (j = 0; j < nr_ids; j++) {
1558 int i, tci = j * dev_maps->num_tc + tc;
1559 struct xps_map *map;
1560
1561 map = rcu_dereference(dev_maps->attr_map[tci]);
1562 if (!map)
1563 continue;
1564
1565 for (i = map->len; i--;) {
1566 if (map->queues[i] == index) {
1567 __set_bit(j, mask);
1568 break;
1569 }
1570 }
1571 }
1572out_no_maps:
1573 rcu_read_unlock();
1574
1575 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1576 bitmap_free(mask);
1577
1578 return len < PAGE_SIZE ? len : -EINVAL;
1579}
1580
1581static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1582{
1583 struct net_device *dev = queue->dev;
1584 unsigned int index;
1585 int len, tc;
1586
1587 if (!netif_is_multiqueue(dev))
1588 return -ENOENT;
1589
1590 index = get_netdev_queue_index(queue);
1591
1592 if (!rtnl_trylock())
1593 return restart_syscall();
1594
1595 /* If queue belongs to subordinate dev use its map */
1596 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1597
1598 tc = netdev_txq_to_tc(dev, index);
1599 if (tc < 0) {
1600 rtnl_unlock();
1601 return -EINVAL;
1602 }
1603
1604 /* Make sure the subordinate device can't be freed */
1605 get_device(&dev->dev);
1606 rtnl_unlock();
1607
1608 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1609
1610 put_device(&dev->dev);
1611 return len;
1612}
1613
1614static ssize_t xps_cpus_store(struct netdev_queue *queue,
1615 const char *buf, size_t len)
1616{
1617 struct net_device *dev = queue->dev;
1618 unsigned int index;
1619 cpumask_var_t mask;
1620 int err;
1621
1622 if (!netif_is_multiqueue(dev))
1623 return -ENOENT;
1624
1625 if (!capable(CAP_NET_ADMIN))
1626 return -EPERM;
1627
1628 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1629 return -ENOMEM;
1630
1631 index = get_netdev_queue_index(queue);
1632
1633 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1634 if (err) {
1635 free_cpumask_var(mask);
1636 return err;
1637 }
1638
1639 if (!rtnl_trylock()) {
1640 free_cpumask_var(mask);
1641 return restart_syscall();
1642 }
1643
1644 err = netif_set_xps_queue(dev, mask, index);
1645 rtnl_unlock();
1646
1647 free_cpumask_var(mask);
1648
1649 return err ? : len;
1650}
1651
1652static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1653 = __ATTR_RW(xps_cpus);
1654
1655static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1656{
1657 struct net_device *dev = queue->dev;
1658 unsigned int index;
1659 int tc;
1660
1661 index = get_netdev_queue_index(queue);
1662
1663 if (!rtnl_trylock())
1664 return restart_syscall();
1665
1666 tc = netdev_txq_to_tc(dev, index);
1667 rtnl_unlock();
1668 if (tc < 0)
1669 return -EINVAL;
1670
1671 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1672}
1673
1674static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1675 size_t len)
1676{
1677 struct net_device *dev = queue->dev;
1678 struct net *net = dev_net(dev);
1679 unsigned long *mask;
1680 unsigned int index;
1681 int err;
1682
1683 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1684 return -EPERM;
1685
1686 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1687 if (!mask)
1688 return -ENOMEM;
1689
1690 index = get_netdev_queue_index(queue);
1691
1692 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1693 if (err) {
1694 bitmap_free(mask);
1695 return err;
1696 }
1697
1698 if (!rtnl_trylock()) {
1699 bitmap_free(mask);
1700 return restart_syscall();
1701 }
1702
1703 cpus_read_lock();
1704 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1705 cpus_read_unlock();
1706
1707 rtnl_unlock();
1708
1709 bitmap_free(mask);
1710 return err ? : len;
1711}
1712
1713static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1714 = __ATTR_RW(xps_rxqs);
1715#endif /* CONFIG_XPS */
1716
1717static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1718 &queue_trans_timeout.attr,
1719 &queue_traffic_class.attr,
1720#ifdef CONFIG_XPS
1721 &xps_cpus_attribute.attr,
1722 &xps_rxqs_attribute.attr,
1723 &queue_tx_maxrate.attr,
1724#endif
1725 NULL
1726};
1727ATTRIBUTE_GROUPS(netdev_queue_default);
1728
1729static void netdev_queue_release(struct kobject *kobj)
1730{
1731 struct netdev_queue *queue = to_netdev_queue(kobj);
1732
1733 memset(kobj, 0, sizeof(*kobj));
1734 netdev_put(queue->dev, &queue->dev_tracker);
1735}
1736
1737static const void *netdev_queue_namespace(const struct kobject *kobj)
1738{
1739 struct netdev_queue *queue = to_netdev_queue(kobj);
1740 struct device *dev = &queue->dev->dev;
1741 const void *ns = NULL;
1742
1743 if (dev->class && dev->class->ns_type)
1744 ns = dev->class->namespace(dev);
1745
1746 return ns;
1747}
1748
1749static void netdev_queue_get_ownership(const struct kobject *kobj,
1750 kuid_t *uid, kgid_t *gid)
1751{
1752 const struct net *net = netdev_queue_namespace(kobj);
1753
1754 net_ns_get_ownership(net, uid, gid);
1755}
1756
1757static const struct kobj_type netdev_queue_ktype = {
1758 .sysfs_ops = &netdev_queue_sysfs_ops,
1759 .release = netdev_queue_release,
1760 .default_groups = netdev_queue_default_groups,
1761 .namespace = netdev_queue_namespace,
1762 .get_ownership = netdev_queue_get_ownership,
1763};
1764
1765static bool netdev_uses_bql(const struct net_device *dev)
1766{
1767 if (dev->features & NETIF_F_LLTX ||
1768 dev->priv_flags & IFF_NO_QUEUE)
1769 return false;
1770
1771 return IS_ENABLED(CONFIG_BQL);
1772}
1773
1774static int netdev_queue_add_kobject(struct net_device *dev, int index)
1775{
1776 struct netdev_queue *queue = dev->_tx + index;
1777 struct kobject *kobj = &queue->kobj;
1778 int error = 0;
1779
1780 /* Kobject_put later will trigger netdev_queue_release call
1781 * which decreases dev refcount: Take that reference here
1782 */
1783 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1784
1785 kobj->kset = dev->queues_kset;
1786 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1787 "tx-%u", index);
1788 if (error)
1789 goto err;
1790
1791 if (netdev_uses_bql(dev)) {
1792 error = sysfs_create_group(kobj, &dql_group);
1793 if (error)
1794 goto err;
1795 }
1796
1797 kobject_uevent(kobj, KOBJ_ADD);
1798 return 0;
1799
1800err:
1801 kobject_put(kobj);
1802 return error;
1803}
1804
1805static int tx_queue_change_owner(struct net_device *ndev, int index,
1806 kuid_t kuid, kgid_t kgid)
1807{
1808 struct netdev_queue *queue = ndev->_tx + index;
1809 struct kobject *kobj = &queue->kobj;
1810 int error;
1811
1812 error = sysfs_change_owner(kobj, kuid, kgid);
1813 if (error)
1814 return error;
1815
1816 if (netdev_uses_bql(ndev))
1817 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1818
1819 return error;
1820}
1821#endif /* CONFIG_SYSFS */
1822
1823int
1824netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1825{
1826#ifdef CONFIG_SYSFS
1827 int i;
1828 int error = 0;
1829
1830 /* Tx queue kobjects are allowed to be updated when a device is being
1831 * unregistered, but solely to remove queues from qdiscs. Any path
1832 * adding queues should be fixed.
1833 */
1834 WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1835 "New queues can't be registered after device unregistration.");
1836
1837 for (i = old_num; i < new_num; i++) {
1838 error = netdev_queue_add_kobject(dev, i);
1839 if (error) {
1840 new_num = old_num;
1841 break;
1842 }
1843 }
1844
1845 while (--i >= new_num) {
1846 struct netdev_queue *queue = dev->_tx + i;
1847
1848 if (!refcount_read(&dev_net(dev)->ns.count))
1849 queue->kobj.uevent_suppress = 1;
1850
1851 if (netdev_uses_bql(dev))
1852 sysfs_remove_group(&queue->kobj, &dql_group);
1853
1854 kobject_put(&queue->kobj);
1855 }
1856
1857 return error;
1858#else
1859 return 0;
1860#endif /* CONFIG_SYSFS */
1861}
1862
1863static int net_tx_queue_change_owner(struct net_device *dev, int num,
1864 kuid_t kuid, kgid_t kgid)
1865{
1866#ifdef CONFIG_SYSFS
1867 int error = 0;
1868 int i;
1869
1870 for (i = 0; i < num; i++) {
1871 error = tx_queue_change_owner(dev, i, kuid, kgid);
1872 if (error)
1873 break;
1874 }
1875
1876 return error;
1877#else
1878 return 0;
1879#endif /* CONFIG_SYSFS */
1880}
1881
1882static int register_queue_kobjects(struct net_device *dev)
1883{
1884 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1885
1886#ifdef CONFIG_SYSFS
1887 dev->queues_kset = kset_create_and_add("queues",
1888 NULL, &dev->dev.kobj);
1889 if (!dev->queues_kset)
1890 return -ENOMEM;
1891 real_rx = dev->real_num_rx_queues;
1892#endif
1893 real_tx = dev->real_num_tx_queues;
1894
1895 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1896 if (error)
1897 goto error;
1898 rxq = real_rx;
1899
1900 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1901 if (error)
1902 goto error;
1903 txq = real_tx;
1904
1905 return 0;
1906
1907error:
1908 netdev_queue_update_kobjects(dev, txq, 0);
1909 net_rx_queue_update_kobjects(dev, rxq, 0);
1910#ifdef CONFIG_SYSFS
1911 kset_unregister(dev->queues_kset);
1912#endif
1913 return error;
1914}
1915
1916static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1917{
1918 int error = 0, real_rx = 0, real_tx = 0;
1919
1920#ifdef CONFIG_SYSFS
1921 if (ndev->queues_kset) {
1922 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1923 if (error)
1924 return error;
1925 }
1926 real_rx = ndev->real_num_rx_queues;
1927#endif
1928 real_tx = ndev->real_num_tx_queues;
1929
1930 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1931 if (error)
1932 return error;
1933
1934 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1935 if (error)
1936 return error;
1937
1938 return 0;
1939}
1940
1941static void remove_queue_kobjects(struct net_device *dev)
1942{
1943 int real_rx = 0, real_tx = 0;
1944
1945#ifdef CONFIG_SYSFS
1946 real_rx = dev->real_num_rx_queues;
1947#endif
1948 real_tx = dev->real_num_tx_queues;
1949
1950 net_rx_queue_update_kobjects(dev, real_rx, 0);
1951 netdev_queue_update_kobjects(dev, real_tx, 0);
1952
1953 dev->real_num_rx_queues = 0;
1954 dev->real_num_tx_queues = 0;
1955#ifdef CONFIG_SYSFS
1956 kset_unregister(dev->queues_kset);
1957#endif
1958}
1959
1960static bool net_current_may_mount(void)
1961{
1962 struct net *net = current->nsproxy->net_ns;
1963
1964 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1965}
1966
1967static void *net_grab_current_ns(void)
1968{
1969 struct net *ns = current->nsproxy->net_ns;
1970#ifdef CONFIG_NET_NS
1971 if (ns)
1972 refcount_inc(&ns->passive);
1973#endif
1974 return ns;
1975}
1976
1977static const void *net_initial_ns(void)
1978{
1979 return &init_net;
1980}
1981
1982static const void *net_netlink_ns(struct sock *sk)
1983{
1984 return sock_net(sk);
1985}
1986
1987const struct kobj_ns_type_operations net_ns_type_operations = {
1988 .type = KOBJ_NS_TYPE_NET,
1989 .current_may_mount = net_current_may_mount,
1990 .grab_current_ns = net_grab_current_ns,
1991 .netlink_ns = net_netlink_ns,
1992 .initial_ns = net_initial_ns,
1993 .drop_ns = net_drop_ns,
1994};
1995EXPORT_SYMBOL_GPL(net_ns_type_operations);
1996
1997static int netdev_uevent(const struct device *d, struct kobj_uevent_env *env)
1998{
1999 const struct net_device *dev = to_net_dev(d);
2000 int retval;
2001
2002 /* pass interface to uevent. */
2003 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
2004 if (retval)
2005 goto exit;
2006
2007 /* pass ifindex to uevent.
2008 * ifindex is useful as it won't change (interface name may change)
2009 * and is what RtNetlink uses natively.
2010 */
2011 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
2012
2013exit:
2014 return retval;
2015}
2016
2017/*
2018 * netdev_release -- destroy and free a dead device.
2019 * Called when last reference to device kobject is gone.
2020 */
2021static void netdev_release(struct device *d)
2022{
2023 struct net_device *dev = to_net_dev(d);
2024
2025 BUG_ON(dev->reg_state != NETREG_RELEASED);
2026
2027 /* no need to wait for rcu grace period:
2028 * device is dead and about to be freed.
2029 */
2030 kfree(rcu_access_pointer(dev->ifalias));
2031 netdev_freemem(dev);
2032}
2033
2034static const void *net_namespace(const struct device *d)
2035{
2036 const struct net_device *dev = to_net_dev(d);
2037
2038 return dev_net(dev);
2039}
2040
2041static void net_get_ownership(const struct device *d, kuid_t *uid, kgid_t *gid)
2042{
2043 const struct net_device *dev = to_net_dev(d);
2044 const struct net *net = dev_net(dev);
2045
2046 net_ns_get_ownership(net, uid, gid);
2047}
2048
2049static struct class net_class __ro_after_init = {
2050 .name = "net",
2051 .dev_release = netdev_release,
2052 .dev_groups = net_class_groups,
2053 .dev_uevent = netdev_uevent,
2054 .ns_type = &net_ns_type_operations,
2055 .namespace = net_namespace,
2056 .get_ownership = net_get_ownership,
2057};
2058
2059#ifdef CONFIG_OF
2060static int of_dev_node_match(struct device *dev, const void *data)
2061{
2062 for (; dev; dev = dev->parent) {
2063 if (dev->of_node == data)
2064 return 1;
2065 }
2066
2067 return 0;
2068}
2069
2070/*
2071 * of_find_net_device_by_node - lookup the net device for the device node
2072 * @np: OF device node
2073 *
2074 * Looks up the net_device structure corresponding with the device node.
2075 * If successful, returns a pointer to the net_device with the embedded
2076 * struct device refcount incremented by one, or NULL on failure. The
2077 * refcount must be dropped when done with the net_device.
2078 */
2079struct net_device *of_find_net_device_by_node(struct device_node *np)
2080{
2081 struct device *dev;
2082
2083 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
2084 if (!dev)
2085 return NULL;
2086
2087 return to_net_dev(dev);
2088}
2089EXPORT_SYMBOL(of_find_net_device_by_node);
2090#endif
2091
2092/* Delete sysfs entries but hold kobject reference until after all
2093 * netdev references are gone.
2094 */
2095void netdev_unregister_kobject(struct net_device *ndev)
2096{
2097 struct device *dev = &ndev->dev;
2098
2099 if (!refcount_read(&dev_net(ndev)->ns.count))
2100 dev_set_uevent_suppress(dev, 1);
2101
2102 kobject_get(&dev->kobj);
2103
2104 remove_queue_kobjects(ndev);
2105
2106 pm_runtime_set_memalloc_noio(dev, false);
2107
2108 device_del(dev);
2109}
2110
2111/* Create sysfs entries for network device. */
2112int netdev_register_kobject(struct net_device *ndev)
2113{
2114 struct device *dev = &ndev->dev;
2115 const struct attribute_group **groups = ndev->sysfs_groups;
2116 int error = 0;
2117
2118 device_initialize(dev);
2119 dev->class = &net_class;
2120 dev->platform_data = ndev;
2121 dev->groups = groups;
2122
2123 dev_set_name(dev, "%s", ndev->name);
2124
2125#ifdef CONFIG_SYSFS
2126 /* Allow for a device specific group */
2127 if (*groups)
2128 groups++;
2129
2130 *groups++ = &netstat_group;
2131
2132 if (wireless_group_needed(ndev))
2133 *groups++ = &wireless_group;
2134#endif /* CONFIG_SYSFS */
2135
2136 error = device_add(dev);
2137 if (error)
2138 return error;
2139
2140 error = register_queue_kobjects(ndev);
2141 if (error) {
2142 device_del(dev);
2143 return error;
2144 }
2145
2146 pm_runtime_set_memalloc_noio(dev, true);
2147
2148 return error;
2149}
2150
2151/* Change owner for sysfs entries when moving network devices across network
2152 * namespaces owned by different user namespaces.
2153 */
2154int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2155 const struct net *net_new)
2156{
2157 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2158 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2159 struct device *dev = &ndev->dev;
2160 int error;
2161
2162 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2163 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2164
2165 /* The network namespace was changed but the owning user namespace is
2166 * identical so there's no need to change the owner of sysfs entries.
2167 */
2168 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2169 return 0;
2170
2171 error = device_change_owner(dev, new_uid, new_gid);
2172 if (error)
2173 return error;
2174
2175 error = queue_change_owner(ndev, new_uid, new_gid);
2176 if (error)
2177 return error;
2178
2179 return 0;
2180}
2181
2182int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2183 const void *ns)
2184{
2185 return class_create_file_ns(&net_class, class_attr, ns);
2186}
2187EXPORT_SYMBOL(netdev_class_create_file_ns);
2188
2189void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2190 const void *ns)
2191{
2192 class_remove_file_ns(&net_class, class_attr, ns);
2193}
2194EXPORT_SYMBOL(netdev_class_remove_file_ns);
2195
2196int __init netdev_kobject_init(void)
2197{
2198 kobj_ns_type_register(&net_ns_type_operations);
2199 return class_register(&net_class);
2200}