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#include <net/netdev_rx_queue.h>
27
28#include "dev.h"
29#include "net-sysfs.h"
30
31#ifdef CONFIG_SYSFS
32static const char fmt_hex[] = "%#x\n";
33static const char fmt_dec[] = "%d\n";
34static const char fmt_ulong[] = "%lu\n";
35static const char fmt_u64[] = "%llu\n";
36
37/* Caller holds RTNL or dev_base_lock */
38static inline int dev_isalive(const struct net_device *dev)
39{
40 return dev->reg_state <= NETREG_REGISTERED;
41}
42
43/* use same locking rules as GIF* ioctl's */
44static ssize_t netdev_show(const struct device *dev,
45 struct device_attribute *attr, char *buf,
46 ssize_t (*format)(const struct net_device *, char *))
47{
48 struct net_device *ndev = to_net_dev(dev);
49 ssize_t ret = -EINVAL;
50
51 read_lock(&dev_base_lock);
52 if (dev_isalive(ndev))
53 ret = (*format)(ndev, buf);
54 read_unlock(&dev_base_lock);
55
56 return ret;
57}
58
59/* generate a show function for simple field */
60#define NETDEVICE_SHOW(field, format_string) \
61static ssize_t format_##field(const struct net_device *dev, char *buf) \
62{ \
63 return sysfs_emit(buf, format_string, dev->field); \
64} \
65static ssize_t field##_show(struct device *dev, \
66 struct device_attribute *attr, char *buf) \
67{ \
68 return netdev_show(dev, attr, buf, format_##field); \
69} \
70
71#define NETDEVICE_SHOW_RO(field, format_string) \
72NETDEVICE_SHOW(field, format_string); \
73static DEVICE_ATTR_RO(field)
74
75#define NETDEVICE_SHOW_RW(field, format_string) \
76NETDEVICE_SHOW(field, format_string); \
77static DEVICE_ATTR_RW(field)
78
79/* use same locking and permission rules as SIF* ioctl's */
80static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
81 const char *buf, size_t len,
82 int (*set)(struct net_device *, unsigned long))
83{
84 struct net_device *netdev = to_net_dev(dev);
85 struct net *net = dev_net(netdev);
86 unsigned long new;
87 int ret;
88
89 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
90 return -EPERM;
91
92 ret = kstrtoul(buf, 0, &new);
93 if (ret)
94 goto err;
95
96 if (!rtnl_trylock())
97 return restart_syscall();
98
99 if (dev_isalive(netdev)) {
100 ret = (*set)(netdev, new);
101 if (ret == 0)
102 ret = len;
103 }
104 rtnl_unlock();
105 err:
106 return ret;
107}
108
109NETDEVICE_SHOW_RO(dev_id, fmt_hex);
110NETDEVICE_SHOW_RO(dev_port, fmt_dec);
111NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
112NETDEVICE_SHOW_RO(addr_len, fmt_dec);
113NETDEVICE_SHOW_RO(ifindex, fmt_dec);
114NETDEVICE_SHOW_RO(type, fmt_dec);
115NETDEVICE_SHOW_RO(link_mode, fmt_dec);
116
117static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
118 char *buf)
119{
120 struct net_device *ndev = to_net_dev(dev);
121
122 return sysfs_emit(buf, fmt_dec, dev_get_iflink(ndev));
123}
124static DEVICE_ATTR_RO(iflink);
125
126static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
127{
128 return sysfs_emit(buf, fmt_dec, dev->name_assign_type);
129}
130
131static ssize_t name_assign_type_show(struct device *dev,
132 struct device_attribute *attr,
133 char *buf)
134{
135 struct net_device *ndev = to_net_dev(dev);
136 ssize_t ret = -EINVAL;
137
138 if (ndev->name_assign_type != NET_NAME_UNKNOWN)
139 ret = netdev_show(dev, attr, buf, format_name_assign_type);
140
141 return ret;
142}
143static DEVICE_ATTR_RO(name_assign_type);
144
145/* use same locking rules as GIFHWADDR ioctl's */
146static ssize_t address_show(struct device *dev, struct device_attribute *attr,
147 char *buf)
148{
149 struct net_device *ndev = to_net_dev(dev);
150 ssize_t ret = -EINVAL;
151
152 read_lock(&dev_base_lock);
153 if (dev_isalive(ndev))
154 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
155 read_unlock(&dev_base_lock);
156 return ret;
157}
158static DEVICE_ATTR_RO(address);
159
160static ssize_t broadcast_show(struct device *dev,
161 struct device_attribute *attr, char *buf)
162{
163 struct net_device *ndev = to_net_dev(dev);
164
165 if (dev_isalive(ndev))
166 return sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
167 return -EINVAL;
168}
169static DEVICE_ATTR_RO(broadcast);
170
171static int change_carrier(struct net_device *dev, unsigned long new_carrier)
172{
173 if (!netif_running(dev))
174 return -EINVAL;
175 return dev_change_carrier(dev, (bool)new_carrier);
176}
177
178static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
179 const char *buf, size_t len)
180{
181 struct net_device *netdev = to_net_dev(dev);
182
183 /* The check is also done in change_carrier; this helps returning early
184 * without hitting the trylock/restart in netdev_store.
185 */
186 if (!netdev->netdev_ops->ndo_change_carrier)
187 return -EOPNOTSUPP;
188
189 return netdev_store(dev, attr, buf, len, change_carrier);
190}
191
192static ssize_t carrier_show(struct device *dev,
193 struct device_attribute *attr, char *buf)
194{
195 struct net_device *netdev = to_net_dev(dev);
196 int ret = -EINVAL;
197
198 if (!rtnl_trylock())
199 return restart_syscall();
200
201 if (netif_running(netdev)) {
202 /* Synchronize carrier state with link watch,
203 * see also rtnl_getlink().
204 */
205 linkwatch_sync_dev(netdev);
206
207 ret = sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev));
208 }
209 rtnl_unlock();
210
211 return ret;
212}
213static DEVICE_ATTR_RW(carrier);
214
215static ssize_t speed_show(struct device *dev,
216 struct device_attribute *attr, char *buf)
217{
218 struct net_device *netdev = to_net_dev(dev);
219 int ret = -EINVAL;
220
221 /* The check is also done in __ethtool_get_link_ksettings; this helps
222 * returning early without hitting the trylock/restart below.
223 */
224 if (!netdev->ethtool_ops->get_link_ksettings)
225 return ret;
226
227 if (!rtnl_trylock())
228 return restart_syscall();
229
230 if (netif_running(netdev) && netif_device_present(netdev)) {
231 struct ethtool_link_ksettings cmd;
232
233 if (!__ethtool_get_link_ksettings(netdev, &cmd))
234 ret = sysfs_emit(buf, fmt_dec, cmd.base.speed);
235 }
236 rtnl_unlock();
237 return ret;
238}
239static DEVICE_ATTR_RO(speed);
240
241static ssize_t duplex_show(struct device *dev,
242 struct device_attribute *attr, char *buf)
243{
244 struct net_device *netdev = to_net_dev(dev);
245 int ret = -EINVAL;
246
247 /* The check is also done in __ethtool_get_link_ksettings; this helps
248 * returning early without hitting the trylock/restart below.
249 */
250 if (!netdev->ethtool_ops->get_link_ksettings)
251 return ret;
252
253 if (!rtnl_trylock())
254 return restart_syscall();
255
256 if (netif_running(netdev)) {
257 struct ethtool_link_ksettings cmd;
258
259 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
260 const char *duplex;
261
262 switch (cmd.base.duplex) {
263 case DUPLEX_HALF:
264 duplex = "half";
265 break;
266 case DUPLEX_FULL:
267 duplex = "full";
268 break;
269 default:
270 duplex = "unknown";
271 break;
272 }
273 ret = sysfs_emit(buf, "%s\n", duplex);
274 }
275 }
276 rtnl_unlock();
277 return ret;
278}
279static DEVICE_ATTR_RO(duplex);
280
281static ssize_t testing_show(struct device *dev,
282 struct device_attribute *attr, char *buf)
283{
284 struct net_device *netdev = to_net_dev(dev);
285
286 if (netif_running(netdev))
287 return sysfs_emit(buf, fmt_dec, !!netif_testing(netdev));
288
289 return -EINVAL;
290}
291static DEVICE_ATTR_RO(testing);
292
293static ssize_t dormant_show(struct device *dev,
294 struct device_attribute *attr, char *buf)
295{
296 struct net_device *netdev = to_net_dev(dev);
297
298 if (netif_running(netdev))
299 return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev));
300
301 return -EINVAL;
302}
303static DEVICE_ATTR_RO(dormant);
304
305static const char *const operstates[] = {
306 "unknown",
307 "notpresent", /* currently unused */
308 "down",
309 "lowerlayerdown",
310 "testing",
311 "dormant",
312 "up"
313};
314
315static ssize_t operstate_show(struct device *dev,
316 struct device_attribute *attr, char *buf)
317{
318 const struct net_device *netdev = to_net_dev(dev);
319 unsigned char operstate;
320
321 read_lock(&dev_base_lock);
322 operstate = netdev->operstate;
323 if (!netif_running(netdev))
324 operstate = IF_OPER_DOWN;
325 read_unlock(&dev_base_lock);
326
327 if (operstate >= ARRAY_SIZE(operstates))
328 return -EINVAL; /* should not happen */
329
330 return sysfs_emit(buf, "%s\n", operstates[operstate]);
331}
332static DEVICE_ATTR_RO(operstate);
333
334static ssize_t carrier_changes_show(struct device *dev,
335 struct device_attribute *attr,
336 char *buf)
337{
338 struct net_device *netdev = to_net_dev(dev);
339
340 return sysfs_emit(buf, fmt_dec,
341 atomic_read(&netdev->carrier_up_count) +
342 atomic_read(&netdev->carrier_down_count));
343}
344static DEVICE_ATTR_RO(carrier_changes);
345
346static ssize_t carrier_up_count_show(struct device *dev,
347 struct device_attribute *attr,
348 char *buf)
349{
350 struct net_device *netdev = to_net_dev(dev);
351
352 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
353}
354static DEVICE_ATTR_RO(carrier_up_count);
355
356static ssize_t carrier_down_count_show(struct device *dev,
357 struct device_attribute *attr,
358 char *buf)
359{
360 struct net_device *netdev = to_net_dev(dev);
361
362 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
363}
364static DEVICE_ATTR_RO(carrier_down_count);
365
366/* read-write attributes */
367
368static int change_mtu(struct net_device *dev, unsigned long new_mtu)
369{
370 return dev_set_mtu(dev, (int)new_mtu);
371}
372
373static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
374 const char *buf, size_t len)
375{
376 return netdev_store(dev, attr, buf, len, change_mtu);
377}
378NETDEVICE_SHOW_RW(mtu, fmt_dec);
379
380static int change_flags(struct net_device *dev, unsigned long new_flags)
381{
382 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
383}
384
385static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
386 const char *buf, size_t len)
387{
388 return netdev_store(dev, attr, buf, len, change_flags);
389}
390NETDEVICE_SHOW_RW(flags, fmt_hex);
391
392static ssize_t tx_queue_len_store(struct device *dev,
393 struct device_attribute *attr,
394 const char *buf, size_t len)
395{
396 if (!capable(CAP_NET_ADMIN))
397 return -EPERM;
398
399 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
400}
401NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
402
403static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
404{
405 WRITE_ONCE(dev->gro_flush_timeout, val);
406 return 0;
407}
408
409static ssize_t gro_flush_timeout_store(struct device *dev,
410 struct device_attribute *attr,
411 const char *buf, size_t len)
412{
413 if (!capable(CAP_NET_ADMIN))
414 return -EPERM;
415
416 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
417}
418NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
419
420static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
421{
422 WRITE_ONCE(dev->napi_defer_hard_irqs, val);
423 return 0;
424}
425
426static ssize_t napi_defer_hard_irqs_store(struct device *dev,
427 struct device_attribute *attr,
428 const char *buf, size_t len)
429{
430 if (!capable(CAP_NET_ADMIN))
431 return -EPERM;
432
433 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
434}
435NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_dec);
436
437static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
438 const char *buf, size_t len)
439{
440 struct net_device *netdev = to_net_dev(dev);
441 struct net *net = dev_net(netdev);
442 size_t count = len;
443 ssize_t ret = 0;
444
445 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
446 return -EPERM;
447
448 /* ignore trailing newline */
449 if (len > 0 && buf[len - 1] == '\n')
450 --count;
451
452 if (!rtnl_trylock())
453 return restart_syscall();
454
455 if (dev_isalive(netdev)) {
456 ret = dev_set_alias(netdev, buf, count);
457 if (ret < 0)
458 goto err;
459 ret = len;
460 netdev_state_change(netdev);
461 }
462err:
463 rtnl_unlock();
464
465 return ret;
466}
467
468static ssize_t ifalias_show(struct device *dev,
469 struct device_attribute *attr, char *buf)
470{
471 const struct net_device *netdev = to_net_dev(dev);
472 char tmp[IFALIASZ];
473 ssize_t ret = 0;
474
475 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
476 if (ret > 0)
477 ret = sysfs_emit(buf, "%s\n", tmp);
478 return ret;
479}
480static DEVICE_ATTR_RW(ifalias);
481
482static int change_group(struct net_device *dev, unsigned long new_group)
483{
484 dev_set_group(dev, (int)new_group);
485 return 0;
486}
487
488static ssize_t group_store(struct device *dev, struct device_attribute *attr,
489 const char *buf, size_t len)
490{
491 return netdev_store(dev, attr, buf, len, change_group);
492}
493NETDEVICE_SHOW(group, fmt_dec);
494static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
495
496static int change_proto_down(struct net_device *dev, unsigned long proto_down)
497{
498 return dev_change_proto_down(dev, (bool)proto_down);
499}
500
501static ssize_t proto_down_store(struct device *dev,
502 struct device_attribute *attr,
503 const char *buf, size_t len)
504{
505 return netdev_store(dev, attr, buf, len, change_proto_down);
506}
507NETDEVICE_SHOW_RW(proto_down, fmt_dec);
508
509static ssize_t phys_port_id_show(struct device *dev,
510 struct device_attribute *attr, char *buf)
511{
512 struct net_device *netdev = to_net_dev(dev);
513 ssize_t ret = -EINVAL;
514
515 /* The check is also done in dev_get_phys_port_id; this helps returning
516 * early without hitting the trylock/restart below.
517 */
518 if (!netdev->netdev_ops->ndo_get_phys_port_id)
519 return -EOPNOTSUPP;
520
521 if (!rtnl_trylock())
522 return restart_syscall();
523
524 if (dev_isalive(netdev)) {
525 struct netdev_phys_item_id ppid;
526
527 ret = dev_get_phys_port_id(netdev, &ppid);
528 if (!ret)
529 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
530 }
531 rtnl_unlock();
532
533 return ret;
534}
535static DEVICE_ATTR_RO(phys_port_id);
536
537static ssize_t phys_port_name_show(struct device *dev,
538 struct device_attribute *attr, char *buf)
539{
540 struct net_device *netdev = to_net_dev(dev);
541 ssize_t ret = -EINVAL;
542
543 /* The checks are also done in dev_get_phys_port_name; this helps
544 * returning early without hitting the trylock/restart below.
545 */
546 if (!netdev->netdev_ops->ndo_get_phys_port_name &&
547 !netdev->devlink_port)
548 return -EOPNOTSUPP;
549
550 if (!rtnl_trylock())
551 return restart_syscall();
552
553 if (dev_isalive(netdev)) {
554 char name[IFNAMSIZ];
555
556 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
557 if (!ret)
558 ret = sysfs_emit(buf, "%s\n", name);
559 }
560 rtnl_unlock();
561
562 return ret;
563}
564static DEVICE_ATTR_RO(phys_port_name);
565
566static ssize_t phys_switch_id_show(struct device *dev,
567 struct device_attribute *attr, char *buf)
568{
569 struct net_device *netdev = to_net_dev(dev);
570 ssize_t ret = -EINVAL;
571
572 /* The checks are also done in dev_get_phys_port_name; this helps
573 * returning early without hitting the trylock/restart below. This works
574 * because recurse is false when calling dev_get_port_parent_id.
575 */
576 if (!netdev->netdev_ops->ndo_get_port_parent_id &&
577 !netdev->devlink_port)
578 return -EOPNOTSUPP;
579
580 if (!rtnl_trylock())
581 return restart_syscall();
582
583 if (dev_isalive(netdev)) {
584 struct netdev_phys_item_id ppid = { };
585
586 ret = dev_get_port_parent_id(netdev, &ppid, false);
587 if (!ret)
588 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
589 }
590 rtnl_unlock();
591
592 return ret;
593}
594static DEVICE_ATTR_RO(phys_switch_id);
595
596static ssize_t threaded_show(struct device *dev,
597 struct device_attribute *attr, char *buf)
598{
599 struct net_device *netdev = to_net_dev(dev);
600 ssize_t ret = -EINVAL;
601
602 if (!rtnl_trylock())
603 return restart_syscall();
604
605 if (dev_isalive(netdev))
606 ret = sysfs_emit(buf, fmt_dec, netdev->threaded);
607
608 rtnl_unlock();
609 return ret;
610}
611
612static int modify_napi_threaded(struct net_device *dev, unsigned long val)
613{
614 int ret;
615
616 if (list_empty(&dev->napi_list))
617 return -EOPNOTSUPP;
618
619 if (val != 0 && val != 1)
620 return -EOPNOTSUPP;
621
622 ret = dev_set_threaded(dev, val);
623
624 return ret;
625}
626
627static ssize_t threaded_store(struct device *dev,
628 struct device_attribute *attr,
629 const char *buf, size_t len)
630{
631 return netdev_store(dev, attr, buf, len, modify_napi_threaded);
632}
633static DEVICE_ATTR_RW(threaded);
634
635static struct attribute *net_class_attrs[] __ro_after_init = {
636 &dev_attr_netdev_group.attr,
637 &dev_attr_type.attr,
638 &dev_attr_dev_id.attr,
639 &dev_attr_dev_port.attr,
640 &dev_attr_iflink.attr,
641 &dev_attr_ifindex.attr,
642 &dev_attr_name_assign_type.attr,
643 &dev_attr_addr_assign_type.attr,
644 &dev_attr_addr_len.attr,
645 &dev_attr_link_mode.attr,
646 &dev_attr_address.attr,
647 &dev_attr_broadcast.attr,
648 &dev_attr_speed.attr,
649 &dev_attr_duplex.attr,
650 &dev_attr_dormant.attr,
651 &dev_attr_testing.attr,
652 &dev_attr_operstate.attr,
653 &dev_attr_carrier_changes.attr,
654 &dev_attr_ifalias.attr,
655 &dev_attr_carrier.attr,
656 &dev_attr_mtu.attr,
657 &dev_attr_flags.attr,
658 &dev_attr_tx_queue_len.attr,
659 &dev_attr_gro_flush_timeout.attr,
660 &dev_attr_napi_defer_hard_irqs.attr,
661 &dev_attr_phys_port_id.attr,
662 &dev_attr_phys_port_name.attr,
663 &dev_attr_phys_switch_id.attr,
664 &dev_attr_proto_down.attr,
665 &dev_attr_carrier_up_count.attr,
666 &dev_attr_carrier_down_count.attr,
667 &dev_attr_threaded.attr,
668 NULL,
669};
670ATTRIBUTE_GROUPS(net_class);
671
672/* Show a given an attribute in the statistics group */
673static ssize_t netstat_show(const struct device *d,
674 struct device_attribute *attr, char *buf,
675 unsigned long offset)
676{
677 struct net_device *dev = to_net_dev(d);
678 ssize_t ret = -EINVAL;
679
680 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
681 offset % sizeof(u64) != 0);
682
683 read_lock(&dev_base_lock);
684 if (dev_isalive(dev)) {
685 struct rtnl_link_stats64 temp;
686 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
687
688 ret = sysfs_emit(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
689 }
690 read_unlock(&dev_base_lock);
691 return ret;
692}
693
694/* generate a read-only statistics attribute */
695#define NETSTAT_ENTRY(name) \
696static ssize_t name##_show(struct device *d, \
697 struct device_attribute *attr, char *buf) \
698{ \
699 return netstat_show(d, attr, buf, \
700 offsetof(struct rtnl_link_stats64, name)); \
701} \
702static DEVICE_ATTR_RO(name)
703
704NETSTAT_ENTRY(rx_packets);
705NETSTAT_ENTRY(tx_packets);
706NETSTAT_ENTRY(rx_bytes);
707NETSTAT_ENTRY(tx_bytes);
708NETSTAT_ENTRY(rx_errors);
709NETSTAT_ENTRY(tx_errors);
710NETSTAT_ENTRY(rx_dropped);
711NETSTAT_ENTRY(tx_dropped);
712NETSTAT_ENTRY(multicast);
713NETSTAT_ENTRY(collisions);
714NETSTAT_ENTRY(rx_length_errors);
715NETSTAT_ENTRY(rx_over_errors);
716NETSTAT_ENTRY(rx_crc_errors);
717NETSTAT_ENTRY(rx_frame_errors);
718NETSTAT_ENTRY(rx_fifo_errors);
719NETSTAT_ENTRY(rx_missed_errors);
720NETSTAT_ENTRY(tx_aborted_errors);
721NETSTAT_ENTRY(tx_carrier_errors);
722NETSTAT_ENTRY(tx_fifo_errors);
723NETSTAT_ENTRY(tx_heartbeat_errors);
724NETSTAT_ENTRY(tx_window_errors);
725NETSTAT_ENTRY(rx_compressed);
726NETSTAT_ENTRY(tx_compressed);
727NETSTAT_ENTRY(rx_nohandler);
728
729static struct attribute *netstat_attrs[] __ro_after_init = {
730 &dev_attr_rx_packets.attr,
731 &dev_attr_tx_packets.attr,
732 &dev_attr_rx_bytes.attr,
733 &dev_attr_tx_bytes.attr,
734 &dev_attr_rx_errors.attr,
735 &dev_attr_tx_errors.attr,
736 &dev_attr_rx_dropped.attr,
737 &dev_attr_tx_dropped.attr,
738 &dev_attr_multicast.attr,
739 &dev_attr_collisions.attr,
740 &dev_attr_rx_length_errors.attr,
741 &dev_attr_rx_over_errors.attr,
742 &dev_attr_rx_crc_errors.attr,
743 &dev_attr_rx_frame_errors.attr,
744 &dev_attr_rx_fifo_errors.attr,
745 &dev_attr_rx_missed_errors.attr,
746 &dev_attr_tx_aborted_errors.attr,
747 &dev_attr_tx_carrier_errors.attr,
748 &dev_attr_tx_fifo_errors.attr,
749 &dev_attr_tx_heartbeat_errors.attr,
750 &dev_attr_tx_window_errors.attr,
751 &dev_attr_rx_compressed.attr,
752 &dev_attr_tx_compressed.attr,
753 &dev_attr_rx_nohandler.attr,
754 NULL
755};
756
757static const struct attribute_group netstat_group = {
758 .name = "statistics",
759 .attrs = netstat_attrs,
760};
761
762static struct attribute *wireless_attrs[] = {
763 NULL
764};
765
766static const struct attribute_group wireless_group = {
767 .name = "wireless",
768 .attrs = wireless_attrs,
769};
770
771static bool wireless_group_needed(struct net_device *ndev)
772{
773#if IS_ENABLED(CONFIG_CFG80211)
774 if (ndev->ieee80211_ptr)
775 return true;
776#endif
777#if IS_ENABLED(CONFIG_WIRELESS_EXT)
778 if (ndev->wireless_handlers)
779 return true;
780#endif
781 return false;
782}
783
784#else /* CONFIG_SYSFS */
785#define net_class_groups NULL
786#endif /* CONFIG_SYSFS */
787
788#ifdef CONFIG_SYSFS
789#define to_rx_queue_attr(_attr) \
790 container_of(_attr, struct rx_queue_attribute, attr)
791
792#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
793
794static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
795 char *buf)
796{
797 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
798 struct netdev_rx_queue *queue = to_rx_queue(kobj);
799
800 if (!attribute->show)
801 return -EIO;
802
803 return attribute->show(queue, buf);
804}
805
806static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
807 const char *buf, size_t count)
808{
809 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
810 struct netdev_rx_queue *queue = to_rx_queue(kobj);
811
812 if (!attribute->store)
813 return -EIO;
814
815 return attribute->store(queue, buf, count);
816}
817
818static const struct sysfs_ops rx_queue_sysfs_ops = {
819 .show = rx_queue_attr_show,
820 .store = rx_queue_attr_store,
821};
822
823#ifdef CONFIG_RPS
824static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
825{
826 struct rps_map *map;
827 cpumask_var_t mask;
828 int i, len;
829
830 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
831 return -ENOMEM;
832
833 rcu_read_lock();
834 map = rcu_dereference(queue->rps_map);
835 if (map)
836 for (i = 0; i < map->len; i++)
837 cpumask_set_cpu(map->cpus[i], mask);
838
839 len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask));
840 rcu_read_unlock();
841 free_cpumask_var(mask);
842
843 return len < PAGE_SIZE ? len : -EINVAL;
844}
845
846static int netdev_rx_queue_set_rps_mask(struct netdev_rx_queue *queue,
847 cpumask_var_t mask)
848{
849 static DEFINE_MUTEX(rps_map_mutex);
850 struct rps_map *old_map, *map;
851 int cpu, i;
852
853 map = kzalloc(max_t(unsigned int,
854 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
855 GFP_KERNEL);
856 if (!map)
857 return -ENOMEM;
858
859 i = 0;
860 for_each_cpu_and(cpu, mask, cpu_online_mask)
861 map->cpus[i++] = cpu;
862
863 if (i) {
864 map->len = i;
865 } else {
866 kfree(map);
867 map = NULL;
868 }
869
870 mutex_lock(&rps_map_mutex);
871 old_map = rcu_dereference_protected(queue->rps_map,
872 mutex_is_locked(&rps_map_mutex));
873 rcu_assign_pointer(queue->rps_map, map);
874
875 if (map)
876 static_branch_inc(&rps_needed);
877 if (old_map)
878 static_branch_dec(&rps_needed);
879
880 mutex_unlock(&rps_map_mutex);
881
882 if (old_map)
883 kfree_rcu(old_map, rcu);
884 return 0;
885}
886
887int rps_cpumask_housekeeping(struct cpumask *mask)
888{
889 if (!cpumask_empty(mask)) {
890 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
891 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
892 if (cpumask_empty(mask))
893 return -EINVAL;
894 }
895 return 0;
896}
897
898static ssize_t store_rps_map(struct netdev_rx_queue *queue,
899 const char *buf, size_t len)
900{
901 cpumask_var_t mask;
902 int err;
903
904 if (!capable(CAP_NET_ADMIN))
905 return -EPERM;
906
907 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
908 return -ENOMEM;
909
910 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
911 if (err)
912 goto out;
913
914 err = rps_cpumask_housekeeping(mask);
915 if (err)
916 goto out;
917
918 err = netdev_rx_queue_set_rps_mask(queue, mask);
919
920out:
921 free_cpumask_var(mask);
922 return err ? : len;
923}
924
925static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
926 char *buf)
927{
928 struct rps_dev_flow_table *flow_table;
929 unsigned long val = 0;
930
931 rcu_read_lock();
932 flow_table = rcu_dereference(queue->rps_flow_table);
933 if (flow_table)
934 val = (unsigned long)flow_table->mask + 1;
935 rcu_read_unlock();
936
937 return sysfs_emit(buf, "%lu\n", val);
938}
939
940static void rps_dev_flow_table_release(struct rcu_head *rcu)
941{
942 struct rps_dev_flow_table *table = container_of(rcu,
943 struct rps_dev_flow_table, rcu);
944 vfree(table);
945}
946
947static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
948 const char *buf, size_t len)
949{
950 unsigned long mask, count;
951 struct rps_dev_flow_table *table, *old_table;
952 static DEFINE_SPINLOCK(rps_dev_flow_lock);
953 int rc;
954
955 if (!capable(CAP_NET_ADMIN))
956 return -EPERM;
957
958 rc = kstrtoul(buf, 0, &count);
959 if (rc < 0)
960 return rc;
961
962 if (count) {
963 mask = count - 1;
964 /* mask = roundup_pow_of_two(count) - 1;
965 * without overflows...
966 */
967 while ((mask | (mask >> 1)) != mask)
968 mask |= (mask >> 1);
969 /* On 64 bit arches, must check mask fits in table->mask (u32),
970 * and on 32bit arches, must check
971 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
972 */
973#if BITS_PER_LONG > 32
974 if (mask > (unsigned long)(u32)mask)
975 return -EINVAL;
976#else
977 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
978 / sizeof(struct rps_dev_flow)) {
979 /* Enforce a limit to prevent overflow */
980 return -EINVAL;
981 }
982#endif
983 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
984 if (!table)
985 return -ENOMEM;
986
987 table->mask = mask;
988 for (count = 0; count <= mask; count++)
989 table->flows[count].cpu = RPS_NO_CPU;
990 } else {
991 table = NULL;
992 }
993
994 spin_lock(&rps_dev_flow_lock);
995 old_table = rcu_dereference_protected(queue->rps_flow_table,
996 lockdep_is_held(&rps_dev_flow_lock));
997 rcu_assign_pointer(queue->rps_flow_table, table);
998 spin_unlock(&rps_dev_flow_lock);
999
1000 if (old_table)
1001 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
1002
1003 return len;
1004}
1005
1006static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
1007 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
1008
1009static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
1010 = __ATTR(rps_flow_cnt, 0644,
1011 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
1012#endif /* CONFIG_RPS */
1013
1014static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
1015#ifdef CONFIG_RPS
1016 &rps_cpus_attribute.attr,
1017 &rps_dev_flow_table_cnt_attribute.attr,
1018#endif
1019 NULL
1020};
1021ATTRIBUTE_GROUPS(rx_queue_default);
1022
1023static void rx_queue_release(struct kobject *kobj)
1024{
1025 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1026#ifdef CONFIG_RPS
1027 struct rps_map *map;
1028 struct rps_dev_flow_table *flow_table;
1029
1030 map = rcu_dereference_protected(queue->rps_map, 1);
1031 if (map) {
1032 RCU_INIT_POINTER(queue->rps_map, NULL);
1033 kfree_rcu(map, rcu);
1034 }
1035
1036 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1037 if (flow_table) {
1038 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1039 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1040 }
1041#endif
1042
1043 memset(kobj, 0, sizeof(*kobj));
1044 netdev_put(queue->dev, &queue->dev_tracker);
1045}
1046
1047static const void *rx_queue_namespace(const struct kobject *kobj)
1048{
1049 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1050 struct device *dev = &queue->dev->dev;
1051 const void *ns = NULL;
1052
1053 if (dev->class && dev->class->ns_type)
1054 ns = dev->class->namespace(dev);
1055
1056 return ns;
1057}
1058
1059static void rx_queue_get_ownership(const struct kobject *kobj,
1060 kuid_t *uid, kgid_t *gid)
1061{
1062 const struct net *net = rx_queue_namespace(kobj);
1063
1064 net_ns_get_ownership(net, uid, gid);
1065}
1066
1067static const struct kobj_type rx_queue_ktype = {
1068 .sysfs_ops = &rx_queue_sysfs_ops,
1069 .release = rx_queue_release,
1070 .default_groups = rx_queue_default_groups,
1071 .namespace = rx_queue_namespace,
1072 .get_ownership = rx_queue_get_ownership,
1073};
1074
1075static int rx_queue_default_mask(struct net_device *dev,
1076 struct netdev_rx_queue *queue)
1077{
1078#if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL)
1079 struct cpumask *rps_default_mask = READ_ONCE(dev_net(dev)->core.rps_default_mask);
1080
1081 if (rps_default_mask && !cpumask_empty(rps_default_mask))
1082 return netdev_rx_queue_set_rps_mask(queue, rps_default_mask);
1083#endif
1084 return 0;
1085}
1086
1087static int rx_queue_add_kobject(struct net_device *dev, int index)
1088{
1089 struct netdev_rx_queue *queue = dev->_rx + index;
1090 struct kobject *kobj = &queue->kobj;
1091 int error = 0;
1092
1093 /* Kobject_put later will trigger rx_queue_release call which
1094 * decreases dev refcount: Take that reference here
1095 */
1096 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1097
1098 kobj->kset = dev->queues_kset;
1099 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1100 "rx-%u", index);
1101 if (error)
1102 goto err;
1103
1104 if (dev->sysfs_rx_queue_group) {
1105 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1106 if (error)
1107 goto err;
1108 }
1109
1110 error = rx_queue_default_mask(dev, queue);
1111 if (error)
1112 goto err;
1113
1114 kobject_uevent(kobj, KOBJ_ADD);
1115
1116 return error;
1117
1118err:
1119 kobject_put(kobj);
1120 return error;
1121}
1122
1123static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1124 kgid_t kgid)
1125{
1126 struct netdev_rx_queue *queue = dev->_rx + index;
1127 struct kobject *kobj = &queue->kobj;
1128 int error;
1129
1130 error = sysfs_change_owner(kobj, kuid, kgid);
1131 if (error)
1132 return error;
1133
1134 if (dev->sysfs_rx_queue_group)
1135 error = sysfs_group_change_owner(
1136 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1137
1138 return error;
1139}
1140#endif /* CONFIG_SYSFS */
1141
1142int
1143net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1144{
1145#ifdef CONFIG_SYSFS
1146 int i;
1147 int error = 0;
1148
1149#ifndef CONFIG_RPS
1150 if (!dev->sysfs_rx_queue_group)
1151 return 0;
1152#endif
1153 for (i = old_num; i < new_num; i++) {
1154 error = rx_queue_add_kobject(dev, i);
1155 if (error) {
1156 new_num = old_num;
1157 break;
1158 }
1159 }
1160
1161 while (--i >= new_num) {
1162 struct kobject *kobj = &dev->_rx[i].kobj;
1163
1164 if (!refcount_read(&dev_net(dev)->ns.count))
1165 kobj->uevent_suppress = 1;
1166 if (dev->sysfs_rx_queue_group)
1167 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1168 kobject_put(kobj);
1169 }
1170
1171 return error;
1172#else
1173 return 0;
1174#endif
1175}
1176
1177static int net_rx_queue_change_owner(struct net_device *dev, int num,
1178 kuid_t kuid, kgid_t kgid)
1179{
1180#ifdef CONFIG_SYSFS
1181 int error = 0;
1182 int i;
1183
1184#ifndef CONFIG_RPS
1185 if (!dev->sysfs_rx_queue_group)
1186 return 0;
1187#endif
1188 for (i = 0; i < num; i++) {
1189 error = rx_queue_change_owner(dev, i, kuid, kgid);
1190 if (error)
1191 break;
1192 }
1193
1194 return error;
1195#else
1196 return 0;
1197#endif
1198}
1199
1200#ifdef CONFIG_SYSFS
1201/*
1202 * netdev_queue sysfs structures and functions.
1203 */
1204struct netdev_queue_attribute {
1205 struct attribute attr;
1206 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1207 ssize_t (*store)(struct netdev_queue *queue,
1208 const char *buf, size_t len);
1209};
1210#define to_netdev_queue_attr(_attr) \
1211 container_of(_attr, struct netdev_queue_attribute, attr)
1212
1213#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1214
1215static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1216 struct attribute *attr, char *buf)
1217{
1218 const struct netdev_queue_attribute *attribute
1219 = to_netdev_queue_attr(attr);
1220 struct netdev_queue *queue = to_netdev_queue(kobj);
1221
1222 if (!attribute->show)
1223 return -EIO;
1224
1225 return attribute->show(queue, buf);
1226}
1227
1228static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1229 struct attribute *attr,
1230 const char *buf, size_t count)
1231{
1232 const struct netdev_queue_attribute *attribute
1233 = to_netdev_queue_attr(attr);
1234 struct netdev_queue *queue = to_netdev_queue(kobj);
1235
1236 if (!attribute->store)
1237 return -EIO;
1238
1239 return attribute->store(queue, buf, count);
1240}
1241
1242static const struct sysfs_ops netdev_queue_sysfs_ops = {
1243 .show = netdev_queue_attr_show,
1244 .store = netdev_queue_attr_store,
1245};
1246
1247static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1248{
1249 unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1250
1251 return sysfs_emit(buf, fmt_ulong, trans_timeout);
1252}
1253
1254static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1255{
1256 struct net_device *dev = queue->dev;
1257 unsigned int i;
1258
1259 i = queue - dev->_tx;
1260 BUG_ON(i >= dev->num_tx_queues);
1261
1262 return i;
1263}
1264
1265static ssize_t traffic_class_show(struct netdev_queue *queue,
1266 char *buf)
1267{
1268 struct net_device *dev = queue->dev;
1269 int num_tc, tc;
1270 int index;
1271
1272 if (!netif_is_multiqueue(dev))
1273 return -ENOENT;
1274
1275 if (!rtnl_trylock())
1276 return restart_syscall();
1277
1278 index = get_netdev_queue_index(queue);
1279
1280 /* If queue belongs to subordinate dev use its TC mapping */
1281 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1282
1283 num_tc = dev->num_tc;
1284 tc = netdev_txq_to_tc(dev, index);
1285
1286 rtnl_unlock();
1287
1288 if (tc < 0)
1289 return -EINVAL;
1290
1291 /* We can report the traffic class one of two ways:
1292 * Subordinate device traffic classes are reported with the traffic
1293 * class first, and then the subordinate class so for example TC0 on
1294 * subordinate device 2 will be reported as "0-2". If the queue
1295 * belongs to the root device it will be reported with just the
1296 * traffic class, so just "0" for TC 0 for example.
1297 */
1298 return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) :
1299 sysfs_emit(buf, "%d\n", tc);
1300}
1301
1302#ifdef CONFIG_XPS
1303static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1304 char *buf)
1305{
1306 return sysfs_emit(buf, "%lu\n", queue->tx_maxrate);
1307}
1308
1309static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1310 const char *buf, size_t len)
1311{
1312 struct net_device *dev = queue->dev;
1313 int err, index = get_netdev_queue_index(queue);
1314 u32 rate = 0;
1315
1316 if (!capable(CAP_NET_ADMIN))
1317 return -EPERM;
1318
1319 /* The check is also done later; this helps returning early without
1320 * hitting the trylock/restart below.
1321 */
1322 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1323 return -EOPNOTSUPP;
1324
1325 err = kstrtou32(buf, 10, &rate);
1326 if (err < 0)
1327 return err;
1328
1329 if (!rtnl_trylock())
1330 return restart_syscall();
1331
1332 err = -EOPNOTSUPP;
1333 if (dev->netdev_ops->ndo_set_tx_maxrate)
1334 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1335
1336 rtnl_unlock();
1337 if (!err) {
1338 queue->tx_maxrate = rate;
1339 return len;
1340 }
1341 return err;
1342}
1343
1344static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1345 = __ATTR_RW(tx_maxrate);
1346#endif
1347
1348static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1349 = __ATTR_RO(tx_timeout);
1350
1351static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1352 = __ATTR_RO(traffic_class);
1353
1354#ifdef CONFIG_BQL
1355/*
1356 * Byte queue limits sysfs structures and functions.
1357 */
1358static ssize_t bql_show(char *buf, unsigned int value)
1359{
1360 return sysfs_emit(buf, "%u\n", value);
1361}
1362
1363static ssize_t bql_set(const char *buf, const size_t count,
1364 unsigned int *pvalue)
1365{
1366 unsigned int value;
1367 int err;
1368
1369 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1370 value = DQL_MAX_LIMIT;
1371 } else {
1372 err = kstrtouint(buf, 10, &value);
1373 if (err < 0)
1374 return err;
1375 if (value > DQL_MAX_LIMIT)
1376 return -EINVAL;
1377 }
1378
1379 *pvalue = value;
1380
1381 return count;
1382}
1383
1384static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1385 char *buf)
1386{
1387 struct dql *dql = &queue->dql;
1388
1389 return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1390}
1391
1392static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1393 const char *buf, size_t len)
1394{
1395 struct dql *dql = &queue->dql;
1396 unsigned int value;
1397 int err;
1398
1399 err = kstrtouint(buf, 10, &value);
1400 if (err < 0)
1401 return err;
1402
1403 dql->slack_hold_time = msecs_to_jiffies(value);
1404
1405 return len;
1406}
1407
1408static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1409 = __ATTR(hold_time, 0644,
1410 bql_show_hold_time, bql_set_hold_time);
1411
1412static ssize_t bql_show_inflight(struct netdev_queue *queue,
1413 char *buf)
1414{
1415 struct dql *dql = &queue->dql;
1416
1417 return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed);
1418}
1419
1420static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1421 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1422
1423#define BQL_ATTR(NAME, FIELD) \
1424static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1425 char *buf) \
1426{ \
1427 return bql_show(buf, queue->dql.FIELD); \
1428} \
1429 \
1430static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1431 const char *buf, size_t len) \
1432{ \
1433 return bql_set(buf, len, &queue->dql.FIELD); \
1434} \
1435 \
1436static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1437 = __ATTR(NAME, 0644, \
1438 bql_show_ ## NAME, bql_set_ ## NAME)
1439
1440BQL_ATTR(limit, limit);
1441BQL_ATTR(limit_max, max_limit);
1442BQL_ATTR(limit_min, min_limit);
1443
1444static struct attribute *dql_attrs[] __ro_after_init = {
1445 &bql_limit_attribute.attr,
1446 &bql_limit_max_attribute.attr,
1447 &bql_limit_min_attribute.attr,
1448 &bql_hold_time_attribute.attr,
1449 &bql_inflight_attribute.attr,
1450 NULL
1451};
1452
1453static const struct attribute_group dql_group = {
1454 .name = "byte_queue_limits",
1455 .attrs = dql_attrs,
1456};
1457#endif /* CONFIG_BQL */
1458
1459#ifdef CONFIG_XPS
1460static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1461 int tc, char *buf, enum xps_map_type type)
1462{
1463 struct xps_dev_maps *dev_maps;
1464 unsigned long *mask;
1465 unsigned int nr_ids;
1466 int j, len;
1467
1468 rcu_read_lock();
1469 dev_maps = rcu_dereference(dev->xps_maps[type]);
1470
1471 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1472 * when dev_maps hasn't been allocated yet, to be backward compatible.
1473 */
1474 nr_ids = dev_maps ? dev_maps->nr_ids :
1475 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1476
1477 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1478 if (!mask) {
1479 rcu_read_unlock();
1480 return -ENOMEM;
1481 }
1482
1483 if (!dev_maps || tc >= dev_maps->num_tc)
1484 goto out_no_maps;
1485
1486 for (j = 0; j < nr_ids; j++) {
1487 int i, tci = j * dev_maps->num_tc + tc;
1488 struct xps_map *map;
1489
1490 map = rcu_dereference(dev_maps->attr_map[tci]);
1491 if (!map)
1492 continue;
1493
1494 for (i = map->len; i--;) {
1495 if (map->queues[i] == index) {
1496 __set_bit(j, mask);
1497 break;
1498 }
1499 }
1500 }
1501out_no_maps:
1502 rcu_read_unlock();
1503
1504 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1505 bitmap_free(mask);
1506
1507 return len < PAGE_SIZE ? len : -EINVAL;
1508}
1509
1510static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1511{
1512 struct net_device *dev = queue->dev;
1513 unsigned int index;
1514 int len, tc;
1515
1516 if (!netif_is_multiqueue(dev))
1517 return -ENOENT;
1518
1519 index = get_netdev_queue_index(queue);
1520
1521 if (!rtnl_trylock())
1522 return restart_syscall();
1523
1524 /* If queue belongs to subordinate dev use its map */
1525 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1526
1527 tc = netdev_txq_to_tc(dev, index);
1528 if (tc < 0) {
1529 rtnl_unlock();
1530 return -EINVAL;
1531 }
1532
1533 /* Make sure the subordinate device can't be freed */
1534 get_device(&dev->dev);
1535 rtnl_unlock();
1536
1537 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1538
1539 put_device(&dev->dev);
1540 return len;
1541}
1542
1543static ssize_t xps_cpus_store(struct netdev_queue *queue,
1544 const char *buf, size_t len)
1545{
1546 struct net_device *dev = queue->dev;
1547 unsigned int index;
1548 cpumask_var_t mask;
1549 int err;
1550
1551 if (!netif_is_multiqueue(dev))
1552 return -ENOENT;
1553
1554 if (!capable(CAP_NET_ADMIN))
1555 return -EPERM;
1556
1557 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1558 return -ENOMEM;
1559
1560 index = get_netdev_queue_index(queue);
1561
1562 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1563 if (err) {
1564 free_cpumask_var(mask);
1565 return err;
1566 }
1567
1568 if (!rtnl_trylock()) {
1569 free_cpumask_var(mask);
1570 return restart_syscall();
1571 }
1572
1573 err = netif_set_xps_queue(dev, mask, index);
1574 rtnl_unlock();
1575
1576 free_cpumask_var(mask);
1577
1578 return err ? : len;
1579}
1580
1581static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1582 = __ATTR_RW(xps_cpus);
1583
1584static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1585{
1586 struct net_device *dev = queue->dev;
1587 unsigned int index;
1588 int tc;
1589
1590 index = get_netdev_queue_index(queue);
1591
1592 if (!rtnl_trylock())
1593 return restart_syscall();
1594
1595 tc = netdev_txq_to_tc(dev, index);
1596 rtnl_unlock();
1597 if (tc < 0)
1598 return -EINVAL;
1599
1600 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1601}
1602
1603static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1604 size_t len)
1605{
1606 struct net_device *dev = queue->dev;
1607 struct net *net = dev_net(dev);
1608 unsigned long *mask;
1609 unsigned int index;
1610 int err;
1611
1612 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1613 return -EPERM;
1614
1615 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1616 if (!mask)
1617 return -ENOMEM;
1618
1619 index = get_netdev_queue_index(queue);
1620
1621 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1622 if (err) {
1623 bitmap_free(mask);
1624 return err;
1625 }
1626
1627 if (!rtnl_trylock()) {
1628 bitmap_free(mask);
1629 return restart_syscall();
1630 }
1631
1632 cpus_read_lock();
1633 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1634 cpus_read_unlock();
1635
1636 rtnl_unlock();
1637
1638 bitmap_free(mask);
1639 return err ? : len;
1640}
1641
1642static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1643 = __ATTR_RW(xps_rxqs);
1644#endif /* CONFIG_XPS */
1645
1646static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1647 &queue_trans_timeout.attr,
1648 &queue_traffic_class.attr,
1649#ifdef CONFIG_XPS
1650 &xps_cpus_attribute.attr,
1651 &xps_rxqs_attribute.attr,
1652 &queue_tx_maxrate.attr,
1653#endif
1654 NULL
1655};
1656ATTRIBUTE_GROUPS(netdev_queue_default);
1657
1658static void netdev_queue_release(struct kobject *kobj)
1659{
1660 struct netdev_queue *queue = to_netdev_queue(kobj);
1661
1662 memset(kobj, 0, sizeof(*kobj));
1663 netdev_put(queue->dev, &queue->dev_tracker);
1664}
1665
1666static const void *netdev_queue_namespace(const struct kobject *kobj)
1667{
1668 struct netdev_queue *queue = to_netdev_queue(kobj);
1669 struct device *dev = &queue->dev->dev;
1670 const void *ns = NULL;
1671
1672 if (dev->class && dev->class->ns_type)
1673 ns = dev->class->namespace(dev);
1674
1675 return ns;
1676}
1677
1678static void netdev_queue_get_ownership(const struct kobject *kobj,
1679 kuid_t *uid, kgid_t *gid)
1680{
1681 const struct net *net = netdev_queue_namespace(kobj);
1682
1683 net_ns_get_ownership(net, uid, gid);
1684}
1685
1686static const struct kobj_type netdev_queue_ktype = {
1687 .sysfs_ops = &netdev_queue_sysfs_ops,
1688 .release = netdev_queue_release,
1689 .default_groups = netdev_queue_default_groups,
1690 .namespace = netdev_queue_namespace,
1691 .get_ownership = netdev_queue_get_ownership,
1692};
1693
1694static int netdev_queue_add_kobject(struct net_device *dev, int index)
1695{
1696 struct netdev_queue *queue = dev->_tx + index;
1697 struct kobject *kobj = &queue->kobj;
1698 int error = 0;
1699
1700 /* Kobject_put later will trigger netdev_queue_release call
1701 * which decreases dev refcount: Take that reference here
1702 */
1703 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1704
1705 kobj->kset = dev->queues_kset;
1706 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1707 "tx-%u", index);
1708 if (error)
1709 goto err;
1710
1711#ifdef CONFIG_BQL
1712 error = sysfs_create_group(kobj, &dql_group);
1713 if (error)
1714 goto err;
1715#endif
1716
1717 kobject_uevent(kobj, KOBJ_ADD);
1718 return 0;
1719
1720err:
1721 kobject_put(kobj);
1722 return error;
1723}
1724
1725static int tx_queue_change_owner(struct net_device *ndev, int index,
1726 kuid_t kuid, kgid_t kgid)
1727{
1728 struct netdev_queue *queue = ndev->_tx + index;
1729 struct kobject *kobj = &queue->kobj;
1730 int error;
1731
1732 error = sysfs_change_owner(kobj, kuid, kgid);
1733 if (error)
1734 return error;
1735
1736#ifdef CONFIG_BQL
1737 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1738#endif
1739 return error;
1740}
1741#endif /* CONFIG_SYSFS */
1742
1743int
1744netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1745{
1746#ifdef CONFIG_SYSFS
1747 int i;
1748 int error = 0;
1749
1750 /* Tx queue kobjects are allowed to be updated when a device is being
1751 * unregistered, but solely to remove queues from qdiscs. Any path
1752 * adding queues should be fixed.
1753 */
1754 WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1755 "New queues can't be registered after device unregistration.");
1756
1757 for (i = old_num; i < new_num; i++) {
1758 error = netdev_queue_add_kobject(dev, i);
1759 if (error) {
1760 new_num = old_num;
1761 break;
1762 }
1763 }
1764
1765 while (--i >= new_num) {
1766 struct netdev_queue *queue = dev->_tx + i;
1767
1768 if (!refcount_read(&dev_net(dev)->ns.count))
1769 queue->kobj.uevent_suppress = 1;
1770#ifdef CONFIG_BQL
1771 sysfs_remove_group(&queue->kobj, &dql_group);
1772#endif
1773 kobject_put(&queue->kobj);
1774 }
1775
1776 return error;
1777#else
1778 return 0;
1779#endif /* CONFIG_SYSFS */
1780}
1781
1782static int net_tx_queue_change_owner(struct net_device *dev, int num,
1783 kuid_t kuid, kgid_t kgid)
1784{
1785#ifdef CONFIG_SYSFS
1786 int error = 0;
1787 int i;
1788
1789 for (i = 0; i < num; i++) {
1790 error = tx_queue_change_owner(dev, i, kuid, kgid);
1791 if (error)
1792 break;
1793 }
1794
1795 return error;
1796#else
1797 return 0;
1798#endif /* CONFIG_SYSFS */
1799}
1800
1801static int register_queue_kobjects(struct net_device *dev)
1802{
1803 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1804
1805#ifdef CONFIG_SYSFS
1806 dev->queues_kset = kset_create_and_add("queues",
1807 NULL, &dev->dev.kobj);
1808 if (!dev->queues_kset)
1809 return -ENOMEM;
1810 real_rx = dev->real_num_rx_queues;
1811#endif
1812 real_tx = dev->real_num_tx_queues;
1813
1814 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1815 if (error)
1816 goto error;
1817 rxq = real_rx;
1818
1819 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1820 if (error)
1821 goto error;
1822 txq = real_tx;
1823
1824 return 0;
1825
1826error:
1827 netdev_queue_update_kobjects(dev, txq, 0);
1828 net_rx_queue_update_kobjects(dev, rxq, 0);
1829#ifdef CONFIG_SYSFS
1830 kset_unregister(dev->queues_kset);
1831#endif
1832 return error;
1833}
1834
1835static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1836{
1837 int error = 0, real_rx = 0, real_tx = 0;
1838
1839#ifdef CONFIG_SYSFS
1840 if (ndev->queues_kset) {
1841 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1842 if (error)
1843 return error;
1844 }
1845 real_rx = ndev->real_num_rx_queues;
1846#endif
1847 real_tx = ndev->real_num_tx_queues;
1848
1849 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1850 if (error)
1851 return error;
1852
1853 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1854 if (error)
1855 return error;
1856
1857 return 0;
1858}
1859
1860static void remove_queue_kobjects(struct net_device *dev)
1861{
1862 int real_rx = 0, real_tx = 0;
1863
1864#ifdef CONFIG_SYSFS
1865 real_rx = dev->real_num_rx_queues;
1866#endif
1867 real_tx = dev->real_num_tx_queues;
1868
1869 net_rx_queue_update_kobjects(dev, real_rx, 0);
1870 netdev_queue_update_kobjects(dev, real_tx, 0);
1871
1872 dev->real_num_rx_queues = 0;
1873 dev->real_num_tx_queues = 0;
1874#ifdef CONFIG_SYSFS
1875 kset_unregister(dev->queues_kset);
1876#endif
1877}
1878
1879static bool net_current_may_mount(void)
1880{
1881 struct net *net = current->nsproxy->net_ns;
1882
1883 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1884}
1885
1886static void *net_grab_current_ns(void)
1887{
1888 struct net *ns = current->nsproxy->net_ns;
1889#ifdef CONFIG_NET_NS
1890 if (ns)
1891 refcount_inc(&ns->passive);
1892#endif
1893 return ns;
1894}
1895
1896static const void *net_initial_ns(void)
1897{
1898 return &init_net;
1899}
1900
1901static const void *net_netlink_ns(struct sock *sk)
1902{
1903 return sock_net(sk);
1904}
1905
1906const struct kobj_ns_type_operations net_ns_type_operations = {
1907 .type = KOBJ_NS_TYPE_NET,
1908 .current_may_mount = net_current_may_mount,
1909 .grab_current_ns = net_grab_current_ns,
1910 .netlink_ns = net_netlink_ns,
1911 .initial_ns = net_initial_ns,
1912 .drop_ns = net_drop_ns,
1913};
1914EXPORT_SYMBOL_GPL(net_ns_type_operations);
1915
1916static int netdev_uevent(const struct device *d, struct kobj_uevent_env *env)
1917{
1918 const struct net_device *dev = to_net_dev(d);
1919 int retval;
1920
1921 /* pass interface to uevent. */
1922 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
1923 if (retval)
1924 goto exit;
1925
1926 /* pass ifindex to uevent.
1927 * ifindex is useful as it won't change (interface name may change)
1928 * and is what RtNetlink uses natively.
1929 */
1930 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
1931
1932exit:
1933 return retval;
1934}
1935
1936/*
1937 * netdev_release -- destroy and free a dead device.
1938 * Called when last reference to device kobject is gone.
1939 */
1940static void netdev_release(struct device *d)
1941{
1942 struct net_device *dev = to_net_dev(d);
1943
1944 BUG_ON(dev->reg_state != NETREG_RELEASED);
1945
1946 /* no need to wait for rcu grace period:
1947 * device is dead and about to be freed.
1948 */
1949 kfree(rcu_access_pointer(dev->ifalias));
1950 netdev_freemem(dev);
1951}
1952
1953static const void *net_namespace(const struct device *d)
1954{
1955 const struct net_device *dev = to_net_dev(d);
1956
1957 return dev_net(dev);
1958}
1959
1960static void net_get_ownership(const struct device *d, kuid_t *uid, kgid_t *gid)
1961{
1962 const struct net_device *dev = to_net_dev(d);
1963 const struct net *net = dev_net(dev);
1964
1965 net_ns_get_ownership(net, uid, gid);
1966}
1967
1968static struct class net_class __ro_after_init = {
1969 .name = "net",
1970 .dev_release = netdev_release,
1971 .dev_groups = net_class_groups,
1972 .dev_uevent = netdev_uevent,
1973 .ns_type = &net_ns_type_operations,
1974 .namespace = net_namespace,
1975 .get_ownership = net_get_ownership,
1976};
1977
1978#ifdef CONFIG_OF
1979static int of_dev_node_match(struct device *dev, const void *data)
1980{
1981 for (; dev; dev = dev->parent) {
1982 if (dev->of_node == data)
1983 return 1;
1984 }
1985
1986 return 0;
1987}
1988
1989/*
1990 * of_find_net_device_by_node - lookup the net device for the device node
1991 * @np: OF device node
1992 *
1993 * Looks up the net_device structure corresponding with the device node.
1994 * If successful, returns a pointer to the net_device with the embedded
1995 * struct device refcount incremented by one, or NULL on failure. The
1996 * refcount must be dropped when done with the net_device.
1997 */
1998struct net_device *of_find_net_device_by_node(struct device_node *np)
1999{
2000 struct device *dev;
2001
2002 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
2003 if (!dev)
2004 return NULL;
2005
2006 return to_net_dev(dev);
2007}
2008EXPORT_SYMBOL(of_find_net_device_by_node);
2009#endif
2010
2011/* Delete sysfs entries but hold kobject reference until after all
2012 * netdev references are gone.
2013 */
2014void netdev_unregister_kobject(struct net_device *ndev)
2015{
2016 struct device *dev = &ndev->dev;
2017
2018 if (!refcount_read(&dev_net(ndev)->ns.count))
2019 dev_set_uevent_suppress(dev, 1);
2020
2021 kobject_get(&dev->kobj);
2022
2023 remove_queue_kobjects(ndev);
2024
2025 pm_runtime_set_memalloc_noio(dev, false);
2026
2027 device_del(dev);
2028}
2029
2030/* Create sysfs entries for network device. */
2031int netdev_register_kobject(struct net_device *ndev)
2032{
2033 struct device *dev = &ndev->dev;
2034 const struct attribute_group **groups = ndev->sysfs_groups;
2035 int error = 0;
2036
2037 device_initialize(dev);
2038 dev->class = &net_class;
2039 dev->platform_data = ndev;
2040 dev->groups = groups;
2041
2042 dev_set_name(dev, "%s", ndev->name);
2043
2044#ifdef CONFIG_SYSFS
2045 /* Allow for a device specific group */
2046 if (*groups)
2047 groups++;
2048
2049 *groups++ = &netstat_group;
2050
2051 if (wireless_group_needed(ndev))
2052 *groups++ = &wireless_group;
2053#endif /* CONFIG_SYSFS */
2054
2055 error = device_add(dev);
2056 if (error)
2057 return error;
2058
2059 error = register_queue_kobjects(ndev);
2060 if (error) {
2061 device_del(dev);
2062 return error;
2063 }
2064
2065 pm_runtime_set_memalloc_noio(dev, true);
2066
2067 return error;
2068}
2069
2070/* Change owner for sysfs entries when moving network devices across network
2071 * namespaces owned by different user namespaces.
2072 */
2073int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2074 const struct net *net_new)
2075{
2076 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2077 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2078 struct device *dev = &ndev->dev;
2079 int error;
2080
2081 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2082 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2083
2084 /* The network namespace was changed but the owning user namespace is
2085 * identical so there's no need to change the owner of sysfs entries.
2086 */
2087 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2088 return 0;
2089
2090 error = device_change_owner(dev, new_uid, new_gid);
2091 if (error)
2092 return error;
2093
2094 error = queue_change_owner(ndev, new_uid, new_gid);
2095 if (error)
2096 return error;
2097
2098 return 0;
2099}
2100
2101int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2102 const void *ns)
2103{
2104 return class_create_file_ns(&net_class, class_attr, ns);
2105}
2106EXPORT_SYMBOL(netdev_class_create_file_ns);
2107
2108void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2109 const void *ns)
2110{
2111 class_remove_file_ns(&net_class, class_attr, ns);
2112}
2113EXPORT_SYMBOL(netdev_class_remove_file_ns);
2114
2115int __init netdev_kobject_init(void)
2116{
2117 kobj_ns_type_register(&net_ns_type_operations);
2118 return class_register(&net_class);
2119}
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_uint[] = "%u\n";
36static const char fmt_ulong[] = "%lu\n";
37static const char fmt_u64[] = "%llu\n";
38
39/* Caller holds RTNL or RCU */
40static inline int dev_isalive(const struct net_device *dev)
41{
42 return READ_ONCE(dev->reg_state) <= NETREG_REGISTERED;
43}
44
45/* use same locking rules as GIF* ioctl's */
46static ssize_t netdev_show(const struct device *dev,
47 struct device_attribute *attr, char *buf,
48 ssize_t (*format)(const struct net_device *, char *))
49{
50 struct net_device *ndev = to_net_dev(dev);
51 ssize_t ret = -EINVAL;
52
53 rcu_read_lock();
54 if (dev_isalive(ndev))
55 ret = (*format)(ndev, buf);
56 rcu_read_unlock();
57
58 return ret;
59}
60
61/* generate a show function for simple field */
62#define NETDEVICE_SHOW(field, format_string) \
63static ssize_t format_##field(const struct net_device *dev, char *buf) \
64{ \
65 return sysfs_emit(buf, format_string, READ_ONCE(dev->field)); \
66} \
67static ssize_t field##_show(struct device *dev, \
68 struct device_attribute *attr, char *buf) \
69{ \
70 return netdev_show(dev, attr, buf, format_##field); \
71} \
72
73#define NETDEVICE_SHOW_RO(field, format_string) \
74NETDEVICE_SHOW(field, format_string); \
75static DEVICE_ATTR_RO(field)
76
77#define NETDEVICE_SHOW_RW(field, format_string) \
78NETDEVICE_SHOW(field, format_string); \
79static DEVICE_ATTR_RW(field)
80
81/* use same locking and permission rules as SIF* ioctl's */
82static ssize_t netdev_store(struct device *dev, struct device_attribute *attr,
83 const char *buf, size_t len,
84 int (*set)(struct net_device *, unsigned long))
85{
86 struct net_device *netdev = to_net_dev(dev);
87 struct net *net = dev_net(netdev);
88 unsigned long new;
89 int ret;
90
91 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
92 return -EPERM;
93
94 ret = kstrtoul(buf, 0, &new);
95 if (ret)
96 goto err;
97
98 if (!rtnl_trylock())
99 return restart_syscall();
100
101 if (dev_isalive(netdev)) {
102 ret = (*set)(netdev, new);
103 if (ret == 0)
104 ret = len;
105 }
106 rtnl_unlock();
107 err:
108 return ret;
109}
110
111NETDEVICE_SHOW_RO(dev_id, fmt_hex);
112NETDEVICE_SHOW_RO(dev_port, fmt_dec);
113NETDEVICE_SHOW_RO(addr_assign_type, fmt_dec);
114NETDEVICE_SHOW_RO(addr_len, fmt_dec);
115NETDEVICE_SHOW_RO(ifindex, fmt_dec);
116NETDEVICE_SHOW_RO(type, fmt_dec);
117NETDEVICE_SHOW_RO(link_mode, fmt_dec);
118
119static ssize_t iflink_show(struct device *dev, struct device_attribute *attr,
120 char *buf)
121{
122 struct net_device *ndev = to_net_dev(dev);
123
124 return sysfs_emit(buf, fmt_dec, dev_get_iflink(ndev));
125}
126static DEVICE_ATTR_RO(iflink);
127
128static ssize_t format_name_assign_type(const struct net_device *dev, char *buf)
129{
130 return sysfs_emit(buf, fmt_dec, READ_ONCE(dev->name_assign_type));
131}
132
133static ssize_t name_assign_type_show(struct device *dev,
134 struct device_attribute *attr,
135 char *buf)
136{
137 struct net_device *ndev = to_net_dev(dev);
138 ssize_t ret = -EINVAL;
139
140 if (READ_ONCE(ndev->name_assign_type) != NET_NAME_UNKNOWN)
141 ret = netdev_show(dev, attr, buf, format_name_assign_type);
142
143 return ret;
144}
145static DEVICE_ATTR_RO(name_assign_type);
146
147/* use same locking rules as GIFHWADDR ioctl's (dev_get_mac_address()) */
148static ssize_t address_show(struct device *dev, struct device_attribute *attr,
149 char *buf)
150{
151 struct net_device *ndev = to_net_dev(dev);
152 ssize_t ret = -EINVAL;
153
154 down_read(&dev_addr_sem);
155
156 rcu_read_lock();
157 if (dev_isalive(ndev))
158 ret = sysfs_format_mac(buf, ndev->dev_addr, ndev->addr_len);
159 rcu_read_unlock();
160
161 up_read(&dev_addr_sem);
162 return ret;
163}
164static DEVICE_ATTR_RO(address);
165
166static ssize_t broadcast_show(struct device *dev,
167 struct device_attribute *attr, char *buf)
168{
169 struct net_device *ndev = to_net_dev(dev);
170 int ret = -EINVAL;
171
172 rcu_read_lock();
173 if (dev_isalive(ndev))
174 ret = sysfs_format_mac(buf, ndev->broadcast, ndev->addr_len);
175 rcu_read_unlock();
176 return ret;
177}
178static DEVICE_ATTR_RO(broadcast);
179
180static int change_carrier(struct net_device *dev, unsigned long new_carrier)
181{
182 if (!netif_running(dev))
183 return -EINVAL;
184 return dev_change_carrier(dev, (bool)new_carrier);
185}
186
187static ssize_t carrier_store(struct device *dev, struct device_attribute *attr,
188 const char *buf, size_t len)
189{
190 struct net_device *netdev = to_net_dev(dev);
191
192 /* The check is also done in change_carrier; this helps returning early
193 * without hitting the trylock/restart in netdev_store.
194 */
195 if (!netdev->netdev_ops->ndo_change_carrier)
196 return -EOPNOTSUPP;
197
198 return netdev_store(dev, attr, buf, len, change_carrier);
199}
200
201static ssize_t carrier_show(struct device *dev,
202 struct device_attribute *attr, char *buf)
203{
204 struct net_device *netdev = to_net_dev(dev);
205 int ret = -EINVAL;
206
207 if (!rtnl_trylock())
208 return restart_syscall();
209
210 if (netif_running(netdev)) {
211 /* Synchronize carrier state with link watch,
212 * see also rtnl_getlink().
213 */
214 linkwatch_sync_dev(netdev);
215
216 ret = sysfs_emit(buf, fmt_dec, !!netif_carrier_ok(netdev));
217 }
218 rtnl_unlock();
219
220 return ret;
221}
222static DEVICE_ATTR_RW(carrier);
223
224static ssize_t speed_show(struct device *dev,
225 struct device_attribute *attr, char *buf)
226{
227 struct net_device *netdev = to_net_dev(dev);
228 int ret = -EINVAL;
229
230 /* The check is also done in __ethtool_get_link_ksettings; this helps
231 * returning early without hitting the trylock/restart below.
232 */
233 if (!netdev->ethtool_ops->get_link_ksettings)
234 return ret;
235
236 if (!rtnl_trylock())
237 return restart_syscall();
238
239 if (netif_running(netdev)) {
240 struct ethtool_link_ksettings cmd;
241
242 if (!__ethtool_get_link_ksettings(netdev, &cmd))
243 ret = sysfs_emit(buf, fmt_dec, cmd.base.speed);
244 }
245 rtnl_unlock();
246 return ret;
247}
248static DEVICE_ATTR_RO(speed);
249
250static ssize_t duplex_show(struct device *dev,
251 struct device_attribute *attr, char *buf)
252{
253 struct net_device *netdev = to_net_dev(dev);
254 int ret = -EINVAL;
255
256 /* The check is also done in __ethtool_get_link_ksettings; this helps
257 * returning early without hitting the trylock/restart below.
258 */
259 if (!netdev->ethtool_ops->get_link_ksettings)
260 return ret;
261
262 if (!rtnl_trylock())
263 return restart_syscall();
264
265 if (netif_running(netdev)) {
266 struct ethtool_link_ksettings cmd;
267
268 if (!__ethtool_get_link_ksettings(netdev, &cmd)) {
269 const char *duplex;
270
271 switch (cmd.base.duplex) {
272 case DUPLEX_HALF:
273 duplex = "half";
274 break;
275 case DUPLEX_FULL:
276 duplex = "full";
277 break;
278 default:
279 duplex = "unknown";
280 break;
281 }
282 ret = sysfs_emit(buf, "%s\n", duplex);
283 }
284 }
285 rtnl_unlock();
286 return ret;
287}
288static DEVICE_ATTR_RO(duplex);
289
290static ssize_t testing_show(struct device *dev,
291 struct device_attribute *attr, char *buf)
292{
293 struct net_device *netdev = to_net_dev(dev);
294
295 if (netif_running(netdev))
296 return sysfs_emit(buf, fmt_dec, !!netif_testing(netdev));
297
298 return -EINVAL;
299}
300static DEVICE_ATTR_RO(testing);
301
302static ssize_t dormant_show(struct device *dev,
303 struct device_attribute *attr, char *buf)
304{
305 struct net_device *netdev = to_net_dev(dev);
306
307 if (netif_running(netdev))
308 return sysfs_emit(buf, fmt_dec, !!netif_dormant(netdev));
309
310 return -EINVAL;
311}
312static DEVICE_ATTR_RO(dormant);
313
314static const char *const operstates[] = {
315 "unknown",
316 "notpresent", /* currently unused */
317 "down",
318 "lowerlayerdown",
319 "testing",
320 "dormant",
321 "up"
322};
323
324static ssize_t operstate_show(struct device *dev,
325 struct device_attribute *attr, char *buf)
326{
327 const struct net_device *netdev = to_net_dev(dev);
328 unsigned char operstate;
329
330 operstate = READ_ONCE(netdev->operstate);
331 if (!netif_running(netdev))
332 operstate = IF_OPER_DOWN;
333
334 if (operstate >= ARRAY_SIZE(operstates))
335 return -EINVAL; /* should not happen */
336
337 return sysfs_emit(buf, "%s\n", operstates[operstate]);
338}
339static DEVICE_ATTR_RO(operstate);
340
341static ssize_t carrier_changes_show(struct device *dev,
342 struct device_attribute *attr,
343 char *buf)
344{
345 struct net_device *netdev = to_net_dev(dev);
346
347 return sysfs_emit(buf, fmt_dec,
348 atomic_read(&netdev->carrier_up_count) +
349 atomic_read(&netdev->carrier_down_count));
350}
351static DEVICE_ATTR_RO(carrier_changes);
352
353static ssize_t carrier_up_count_show(struct device *dev,
354 struct device_attribute *attr,
355 char *buf)
356{
357 struct net_device *netdev = to_net_dev(dev);
358
359 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_up_count));
360}
361static DEVICE_ATTR_RO(carrier_up_count);
362
363static ssize_t carrier_down_count_show(struct device *dev,
364 struct device_attribute *attr,
365 char *buf)
366{
367 struct net_device *netdev = to_net_dev(dev);
368
369 return sysfs_emit(buf, fmt_dec, atomic_read(&netdev->carrier_down_count));
370}
371static DEVICE_ATTR_RO(carrier_down_count);
372
373/* read-write attributes */
374
375static int change_mtu(struct net_device *dev, unsigned long new_mtu)
376{
377 return dev_set_mtu(dev, (int)new_mtu);
378}
379
380static ssize_t mtu_store(struct device *dev, struct device_attribute *attr,
381 const char *buf, size_t len)
382{
383 return netdev_store(dev, attr, buf, len, change_mtu);
384}
385NETDEVICE_SHOW_RW(mtu, fmt_dec);
386
387static int change_flags(struct net_device *dev, unsigned long new_flags)
388{
389 return dev_change_flags(dev, (unsigned int)new_flags, NULL);
390}
391
392static ssize_t flags_store(struct device *dev, struct device_attribute *attr,
393 const char *buf, size_t len)
394{
395 return netdev_store(dev, attr, buf, len, change_flags);
396}
397NETDEVICE_SHOW_RW(flags, fmt_hex);
398
399static ssize_t tx_queue_len_store(struct device *dev,
400 struct device_attribute *attr,
401 const char *buf, size_t len)
402{
403 if (!capable(CAP_NET_ADMIN))
404 return -EPERM;
405
406 return netdev_store(dev, attr, buf, len, dev_change_tx_queue_len);
407}
408NETDEVICE_SHOW_RW(tx_queue_len, fmt_dec);
409
410static int change_gro_flush_timeout(struct net_device *dev, unsigned long val)
411{
412 netdev_set_gro_flush_timeout(dev, val);
413 return 0;
414}
415
416static ssize_t gro_flush_timeout_store(struct device *dev,
417 struct device_attribute *attr,
418 const char *buf, size_t len)
419{
420 if (!capable(CAP_NET_ADMIN))
421 return -EPERM;
422
423 return netdev_store(dev, attr, buf, len, change_gro_flush_timeout);
424}
425NETDEVICE_SHOW_RW(gro_flush_timeout, fmt_ulong);
426
427static int change_napi_defer_hard_irqs(struct net_device *dev, unsigned long val)
428{
429 if (val > S32_MAX)
430 return -ERANGE;
431
432 netdev_set_defer_hard_irqs(dev, (u32)val);
433 return 0;
434}
435
436static ssize_t napi_defer_hard_irqs_store(struct device *dev,
437 struct device_attribute *attr,
438 const char *buf, size_t len)
439{
440 if (!capable(CAP_NET_ADMIN))
441 return -EPERM;
442
443 return netdev_store(dev, attr, buf, len, change_napi_defer_hard_irqs);
444}
445NETDEVICE_SHOW_RW(napi_defer_hard_irqs, fmt_uint);
446
447static ssize_t ifalias_store(struct device *dev, struct device_attribute *attr,
448 const char *buf, size_t len)
449{
450 struct net_device *netdev = to_net_dev(dev);
451 struct net *net = dev_net(netdev);
452 size_t count = len;
453 ssize_t ret = 0;
454
455 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
456 return -EPERM;
457
458 /* ignore trailing newline */
459 if (len > 0 && buf[len - 1] == '\n')
460 --count;
461
462 if (!rtnl_trylock())
463 return restart_syscall();
464
465 if (dev_isalive(netdev)) {
466 ret = dev_set_alias(netdev, buf, count);
467 if (ret < 0)
468 goto err;
469 ret = len;
470 netdev_state_change(netdev);
471 }
472err:
473 rtnl_unlock();
474
475 return ret;
476}
477
478static ssize_t ifalias_show(struct device *dev,
479 struct device_attribute *attr, char *buf)
480{
481 const struct net_device *netdev = to_net_dev(dev);
482 char tmp[IFALIASZ];
483 ssize_t ret = 0;
484
485 ret = dev_get_alias(netdev, tmp, sizeof(tmp));
486 if (ret > 0)
487 ret = sysfs_emit(buf, "%s\n", tmp);
488 return ret;
489}
490static DEVICE_ATTR_RW(ifalias);
491
492static int change_group(struct net_device *dev, unsigned long new_group)
493{
494 dev_set_group(dev, (int)new_group);
495 return 0;
496}
497
498static ssize_t group_store(struct device *dev, struct device_attribute *attr,
499 const char *buf, size_t len)
500{
501 return netdev_store(dev, attr, buf, len, change_group);
502}
503NETDEVICE_SHOW(group, fmt_dec);
504static DEVICE_ATTR(netdev_group, 0644, group_show, group_store);
505
506static int change_proto_down(struct net_device *dev, unsigned long proto_down)
507{
508 return dev_change_proto_down(dev, (bool)proto_down);
509}
510
511static ssize_t proto_down_store(struct device *dev,
512 struct device_attribute *attr,
513 const char *buf, size_t len)
514{
515 return netdev_store(dev, attr, buf, len, change_proto_down);
516}
517NETDEVICE_SHOW_RW(proto_down, fmt_dec);
518
519static ssize_t phys_port_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 /* The check is also done in dev_get_phys_port_id; this helps returning
526 * early without hitting the trylock/restart below.
527 */
528 if (!netdev->netdev_ops->ndo_get_phys_port_id)
529 return -EOPNOTSUPP;
530
531 if (!rtnl_trylock())
532 return restart_syscall();
533
534 if (dev_isalive(netdev)) {
535 struct netdev_phys_item_id ppid;
536
537 ret = dev_get_phys_port_id(netdev, &ppid);
538 if (!ret)
539 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
540 }
541 rtnl_unlock();
542
543 return ret;
544}
545static DEVICE_ATTR_RO(phys_port_id);
546
547static ssize_t phys_port_name_show(struct device *dev,
548 struct device_attribute *attr, char *buf)
549{
550 struct net_device *netdev = to_net_dev(dev);
551 ssize_t ret = -EINVAL;
552
553 /* The checks are also done in dev_get_phys_port_name; this helps
554 * returning early without hitting the trylock/restart below.
555 */
556 if (!netdev->netdev_ops->ndo_get_phys_port_name &&
557 !netdev->devlink_port)
558 return -EOPNOTSUPP;
559
560 if (!rtnl_trylock())
561 return restart_syscall();
562
563 if (dev_isalive(netdev)) {
564 char name[IFNAMSIZ];
565
566 ret = dev_get_phys_port_name(netdev, name, sizeof(name));
567 if (!ret)
568 ret = sysfs_emit(buf, "%s\n", name);
569 }
570 rtnl_unlock();
571
572 return ret;
573}
574static DEVICE_ATTR_RO(phys_port_name);
575
576static ssize_t phys_switch_id_show(struct device *dev,
577 struct device_attribute *attr, char *buf)
578{
579 struct net_device *netdev = to_net_dev(dev);
580 ssize_t ret = -EINVAL;
581
582 /* The checks are also done in dev_get_phys_port_name; this helps
583 * returning early without hitting the trylock/restart below. This works
584 * because recurse is false when calling dev_get_port_parent_id.
585 */
586 if (!netdev->netdev_ops->ndo_get_port_parent_id &&
587 !netdev->devlink_port)
588 return -EOPNOTSUPP;
589
590 if (!rtnl_trylock())
591 return restart_syscall();
592
593 if (dev_isalive(netdev)) {
594 struct netdev_phys_item_id ppid = { };
595
596 ret = dev_get_port_parent_id(netdev, &ppid, false);
597 if (!ret)
598 ret = sysfs_emit(buf, "%*phN\n", ppid.id_len, ppid.id);
599 }
600 rtnl_unlock();
601
602 return ret;
603}
604static DEVICE_ATTR_RO(phys_switch_id);
605
606static ssize_t threaded_show(struct device *dev,
607 struct device_attribute *attr, char *buf)
608{
609 struct net_device *netdev = to_net_dev(dev);
610 ssize_t ret = -EINVAL;
611
612 rcu_read_lock();
613
614 if (dev_isalive(netdev))
615 ret = sysfs_emit(buf, fmt_dec, READ_ONCE(netdev->threaded));
616
617 rcu_read_unlock();
618
619 return ret;
620}
621
622static int modify_napi_threaded(struct net_device *dev, unsigned long val)
623{
624 int ret;
625
626 if (list_empty(&dev->napi_list))
627 return -EOPNOTSUPP;
628
629 if (val != 0 && val != 1)
630 return -EOPNOTSUPP;
631
632 ret = dev_set_threaded(dev, val);
633
634 return ret;
635}
636
637static ssize_t threaded_store(struct device *dev,
638 struct device_attribute *attr,
639 const char *buf, size_t len)
640{
641 return netdev_store(dev, attr, buf, len, modify_napi_threaded);
642}
643static DEVICE_ATTR_RW(threaded);
644
645static struct attribute *net_class_attrs[] __ro_after_init = {
646 &dev_attr_netdev_group.attr,
647 &dev_attr_type.attr,
648 &dev_attr_dev_id.attr,
649 &dev_attr_dev_port.attr,
650 &dev_attr_iflink.attr,
651 &dev_attr_ifindex.attr,
652 &dev_attr_name_assign_type.attr,
653 &dev_attr_addr_assign_type.attr,
654 &dev_attr_addr_len.attr,
655 &dev_attr_link_mode.attr,
656 &dev_attr_address.attr,
657 &dev_attr_broadcast.attr,
658 &dev_attr_speed.attr,
659 &dev_attr_duplex.attr,
660 &dev_attr_dormant.attr,
661 &dev_attr_testing.attr,
662 &dev_attr_operstate.attr,
663 &dev_attr_carrier_changes.attr,
664 &dev_attr_ifalias.attr,
665 &dev_attr_carrier.attr,
666 &dev_attr_mtu.attr,
667 &dev_attr_flags.attr,
668 &dev_attr_tx_queue_len.attr,
669 &dev_attr_gro_flush_timeout.attr,
670 &dev_attr_napi_defer_hard_irqs.attr,
671 &dev_attr_phys_port_id.attr,
672 &dev_attr_phys_port_name.attr,
673 &dev_attr_phys_switch_id.attr,
674 &dev_attr_proto_down.attr,
675 &dev_attr_carrier_up_count.attr,
676 &dev_attr_carrier_down_count.attr,
677 &dev_attr_threaded.attr,
678 NULL,
679};
680ATTRIBUTE_GROUPS(net_class);
681
682/* Show a given an attribute in the statistics group */
683static ssize_t netstat_show(const struct device *d,
684 struct device_attribute *attr, char *buf,
685 unsigned long offset)
686{
687 struct net_device *dev = to_net_dev(d);
688 ssize_t ret = -EINVAL;
689
690 WARN_ON(offset > sizeof(struct rtnl_link_stats64) ||
691 offset % sizeof(u64) != 0);
692
693 rcu_read_lock();
694 if (dev_isalive(dev)) {
695 struct rtnl_link_stats64 temp;
696 const struct rtnl_link_stats64 *stats = dev_get_stats(dev, &temp);
697
698 ret = sysfs_emit(buf, fmt_u64, *(u64 *)(((u8 *)stats) + offset));
699 }
700 rcu_read_unlock();
701 return ret;
702}
703
704/* generate a read-only statistics attribute */
705#define NETSTAT_ENTRY(name) \
706static ssize_t name##_show(struct device *d, \
707 struct device_attribute *attr, char *buf) \
708{ \
709 return netstat_show(d, attr, buf, \
710 offsetof(struct rtnl_link_stats64, name)); \
711} \
712static DEVICE_ATTR_RO(name)
713
714NETSTAT_ENTRY(rx_packets);
715NETSTAT_ENTRY(tx_packets);
716NETSTAT_ENTRY(rx_bytes);
717NETSTAT_ENTRY(tx_bytes);
718NETSTAT_ENTRY(rx_errors);
719NETSTAT_ENTRY(tx_errors);
720NETSTAT_ENTRY(rx_dropped);
721NETSTAT_ENTRY(tx_dropped);
722NETSTAT_ENTRY(multicast);
723NETSTAT_ENTRY(collisions);
724NETSTAT_ENTRY(rx_length_errors);
725NETSTAT_ENTRY(rx_over_errors);
726NETSTAT_ENTRY(rx_crc_errors);
727NETSTAT_ENTRY(rx_frame_errors);
728NETSTAT_ENTRY(rx_fifo_errors);
729NETSTAT_ENTRY(rx_missed_errors);
730NETSTAT_ENTRY(tx_aborted_errors);
731NETSTAT_ENTRY(tx_carrier_errors);
732NETSTAT_ENTRY(tx_fifo_errors);
733NETSTAT_ENTRY(tx_heartbeat_errors);
734NETSTAT_ENTRY(tx_window_errors);
735NETSTAT_ENTRY(rx_compressed);
736NETSTAT_ENTRY(tx_compressed);
737NETSTAT_ENTRY(rx_nohandler);
738
739static struct attribute *netstat_attrs[] __ro_after_init = {
740 &dev_attr_rx_packets.attr,
741 &dev_attr_tx_packets.attr,
742 &dev_attr_rx_bytes.attr,
743 &dev_attr_tx_bytes.attr,
744 &dev_attr_rx_errors.attr,
745 &dev_attr_tx_errors.attr,
746 &dev_attr_rx_dropped.attr,
747 &dev_attr_tx_dropped.attr,
748 &dev_attr_multicast.attr,
749 &dev_attr_collisions.attr,
750 &dev_attr_rx_length_errors.attr,
751 &dev_attr_rx_over_errors.attr,
752 &dev_attr_rx_crc_errors.attr,
753 &dev_attr_rx_frame_errors.attr,
754 &dev_attr_rx_fifo_errors.attr,
755 &dev_attr_rx_missed_errors.attr,
756 &dev_attr_tx_aborted_errors.attr,
757 &dev_attr_tx_carrier_errors.attr,
758 &dev_attr_tx_fifo_errors.attr,
759 &dev_attr_tx_heartbeat_errors.attr,
760 &dev_attr_tx_window_errors.attr,
761 &dev_attr_rx_compressed.attr,
762 &dev_attr_tx_compressed.attr,
763 &dev_attr_rx_nohandler.attr,
764 NULL
765};
766
767static const struct attribute_group netstat_group = {
768 .name = "statistics",
769 .attrs = netstat_attrs,
770};
771
772static struct attribute *wireless_attrs[] = {
773 NULL
774};
775
776static const struct attribute_group wireless_group = {
777 .name = "wireless",
778 .attrs = wireless_attrs,
779};
780
781static bool wireless_group_needed(struct net_device *ndev)
782{
783#if IS_ENABLED(CONFIG_CFG80211)
784 if (ndev->ieee80211_ptr)
785 return true;
786#endif
787#if IS_ENABLED(CONFIG_WIRELESS_EXT)
788 if (ndev->wireless_handlers)
789 return true;
790#endif
791 return false;
792}
793
794#else /* CONFIG_SYSFS */
795#define net_class_groups NULL
796#endif /* CONFIG_SYSFS */
797
798#ifdef CONFIG_SYSFS
799#define to_rx_queue_attr(_attr) \
800 container_of(_attr, struct rx_queue_attribute, attr)
801
802#define to_rx_queue(obj) container_of(obj, struct netdev_rx_queue, kobj)
803
804static ssize_t rx_queue_attr_show(struct kobject *kobj, struct attribute *attr,
805 char *buf)
806{
807 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
808 struct netdev_rx_queue *queue = to_rx_queue(kobj);
809
810 if (!attribute->show)
811 return -EIO;
812
813 return attribute->show(queue, buf);
814}
815
816static ssize_t rx_queue_attr_store(struct kobject *kobj, struct attribute *attr,
817 const char *buf, size_t count)
818{
819 const struct rx_queue_attribute *attribute = to_rx_queue_attr(attr);
820 struct netdev_rx_queue *queue = to_rx_queue(kobj);
821
822 if (!attribute->store)
823 return -EIO;
824
825 return attribute->store(queue, buf, count);
826}
827
828static const struct sysfs_ops rx_queue_sysfs_ops = {
829 .show = rx_queue_attr_show,
830 .store = rx_queue_attr_store,
831};
832
833#ifdef CONFIG_RPS
834static ssize_t show_rps_map(struct netdev_rx_queue *queue, char *buf)
835{
836 struct rps_map *map;
837 cpumask_var_t mask;
838 int i, len;
839
840 if (!zalloc_cpumask_var(&mask, GFP_KERNEL))
841 return -ENOMEM;
842
843 rcu_read_lock();
844 map = rcu_dereference(queue->rps_map);
845 if (map)
846 for (i = 0; i < map->len; i++)
847 cpumask_set_cpu(map->cpus[i], mask);
848
849 len = sysfs_emit(buf, "%*pb\n", cpumask_pr_args(mask));
850 rcu_read_unlock();
851 free_cpumask_var(mask);
852
853 return len < PAGE_SIZE ? len : -EINVAL;
854}
855
856static int netdev_rx_queue_set_rps_mask(struct netdev_rx_queue *queue,
857 cpumask_var_t mask)
858{
859 static DEFINE_MUTEX(rps_map_mutex);
860 struct rps_map *old_map, *map;
861 int cpu, i;
862
863 map = kzalloc(max_t(unsigned int,
864 RPS_MAP_SIZE(cpumask_weight(mask)), L1_CACHE_BYTES),
865 GFP_KERNEL);
866 if (!map)
867 return -ENOMEM;
868
869 i = 0;
870 for_each_cpu_and(cpu, mask, cpu_online_mask)
871 map->cpus[i++] = cpu;
872
873 if (i) {
874 map->len = i;
875 } else {
876 kfree(map);
877 map = NULL;
878 }
879
880 mutex_lock(&rps_map_mutex);
881 old_map = rcu_dereference_protected(queue->rps_map,
882 mutex_is_locked(&rps_map_mutex));
883 rcu_assign_pointer(queue->rps_map, map);
884
885 if (map)
886 static_branch_inc(&rps_needed);
887 if (old_map)
888 static_branch_dec(&rps_needed);
889
890 mutex_unlock(&rps_map_mutex);
891
892 if (old_map)
893 kfree_rcu(old_map, rcu);
894 return 0;
895}
896
897int rps_cpumask_housekeeping(struct cpumask *mask)
898{
899 if (!cpumask_empty(mask)) {
900 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_DOMAIN));
901 cpumask_and(mask, mask, housekeeping_cpumask(HK_TYPE_WQ));
902 if (cpumask_empty(mask))
903 return -EINVAL;
904 }
905 return 0;
906}
907
908static ssize_t store_rps_map(struct netdev_rx_queue *queue,
909 const char *buf, size_t len)
910{
911 cpumask_var_t mask;
912 int err;
913
914 if (!capable(CAP_NET_ADMIN))
915 return -EPERM;
916
917 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
918 return -ENOMEM;
919
920 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
921 if (err)
922 goto out;
923
924 err = rps_cpumask_housekeeping(mask);
925 if (err)
926 goto out;
927
928 err = netdev_rx_queue_set_rps_mask(queue, mask);
929
930out:
931 free_cpumask_var(mask);
932 return err ? : len;
933}
934
935static ssize_t show_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
936 char *buf)
937{
938 struct rps_dev_flow_table *flow_table;
939 unsigned long val = 0;
940
941 rcu_read_lock();
942 flow_table = rcu_dereference(queue->rps_flow_table);
943 if (flow_table)
944 val = (unsigned long)flow_table->mask + 1;
945 rcu_read_unlock();
946
947 return sysfs_emit(buf, "%lu\n", val);
948}
949
950static void rps_dev_flow_table_release(struct rcu_head *rcu)
951{
952 struct rps_dev_flow_table *table = container_of(rcu,
953 struct rps_dev_flow_table, rcu);
954 vfree(table);
955}
956
957static ssize_t store_rps_dev_flow_table_cnt(struct netdev_rx_queue *queue,
958 const char *buf, size_t len)
959{
960 unsigned long mask, count;
961 struct rps_dev_flow_table *table, *old_table;
962 static DEFINE_SPINLOCK(rps_dev_flow_lock);
963 int rc;
964
965 if (!capable(CAP_NET_ADMIN))
966 return -EPERM;
967
968 rc = kstrtoul(buf, 0, &count);
969 if (rc < 0)
970 return rc;
971
972 if (count) {
973 mask = count - 1;
974 /* mask = roundup_pow_of_two(count) - 1;
975 * without overflows...
976 */
977 while ((mask | (mask >> 1)) != mask)
978 mask |= (mask >> 1);
979 /* On 64 bit arches, must check mask fits in table->mask (u32),
980 * and on 32bit arches, must check
981 * RPS_DEV_FLOW_TABLE_SIZE(mask + 1) doesn't overflow.
982 */
983#if BITS_PER_LONG > 32
984 if (mask > (unsigned long)(u32)mask)
985 return -EINVAL;
986#else
987 if (mask > (ULONG_MAX - RPS_DEV_FLOW_TABLE_SIZE(1))
988 / sizeof(struct rps_dev_flow)) {
989 /* Enforce a limit to prevent overflow */
990 return -EINVAL;
991 }
992#endif
993 table = vmalloc(RPS_DEV_FLOW_TABLE_SIZE(mask + 1));
994 if (!table)
995 return -ENOMEM;
996
997 table->mask = mask;
998 for (count = 0; count <= mask; count++)
999 table->flows[count].cpu = RPS_NO_CPU;
1000 } else {
1001 table = NULL;
1002 }
1003
1004 spin_lock(&rps_dev_flow_lock);
1005 old_table = rcu_dereference_protected(queue->rps_flow_table,
1006 lockdep_is_held(&rps_dev_flow_lock));
1007 rcu_assign_pointer(queue->rps_flow_table, table);
1008 spin_unlock(&rps_dev_flow_lock);
1009
1010 if (old_table)
1011 call_rcu(&old_table->rcu, rps_dev_flow_table_release);
1012
1013 return len;
1014}
1015
1016static struct rx_queue_attribute rps_cpus_attribute __ro_after_init
1017 = __ATTR(rps_cpus, 0644, show_rps_map, store_rps_map);
1018
1019static struct rx_queue_attribute rps_dev_flow_table_cnt_attribute __ro_after_init
1020 = __ATTR(rps_flow_cnt, 0644,
1021 show_rps_dev_flow_table_cnt, store_rps_dev_flow_table_cnt);
1022#endif /* CONFIG_RPS */
1023
1024static struct attribute *rx_queue_default_attrs[] __ro_after_init = {
1025#ifdef CONFIG_RPS
1026 &rps_cpus_attribute.attr,
1027 &rps_dev_flow_table_cnt_attribute.attr,
1028#endif
1029 NULL
1030};
1031ATTRIBUTE_GROUPS(rx_queue_default);
1032
1033static void rx_queue_release(struct kobject *kobj)
1034{
1035 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1036#ifdef CONFIG_RPS
1037 struct rps_map *map;
1038 struct rps_dev_flow_table *flow_table;
1039
1040 map = rcu_dereference_protected(queue->rps_map, 1);
1041 if (map) {
1042 RCU_INIT_POINTER(queue->rps_map, NULL);
1043 kfree_rcu(map, rcu);
1044 }
1045
1046 flow_table = rcu_dereference_protected(queue->rps_flow_table, 1);
1047 if (flow_table) {
1048 RCU_INIT_POINTER(queue->rps_flow_table, NULL);
1049 call_rcu(&flow_table->rcu, rps_dev_flow_table_release);
1050 }
1051#endif
1052
1053 memset(kobj, 0, sizeof(*kobj));
1054 netdev_put(queue->dev, &queue->dev_tracker);
1055}
1056
1057static const void *rx_queue_namespace(const struct kobject *kobj)
1058{
1059 struct netdev_rx_queue *queue = to_rx_queue(kobj);
1060 struct device *dev = &queue->dev->dev;
1061 const void *ns = NULL;
1062
1063 if (dev->class && dev->class->namespace)
1064 ns = dev->class->namespace(dev);
1065
1066 return ns;
1067}
1068
1069static void rx_queue_get_ownership(const struct kobject *kobj,
1070 kuid_t *uid, kgid_t *gid)
1071{
1072 const struct net *net = rx_queue_namespace(kobj);
1073
1074 net_ns_get_ownership(net, uid, gid);
1075}
1076
1077static const struct kobj_type rx_queue_ktype = {
1078 .sysfs_ops = &rx_queue_sysfs_ops,
1079 .release = rx_queue_release,
1080 .default_groups = rx_queue_default_groups,
1081 .namespace = rx_queue_namespace,
1082 .get_ownership = rx_queue_get_ownership,
1083};
1084
1085static int rx_queue_default_mask(struct net_device *dev,
1086 struct netdev_rx_queue *queue)
1087{
1088#if IS_ENABLED(CONFIG_RPS) && IS_ENABLED(CONFIG_SYSCTL)
1089 struct cpumask *rps_default_mask = READ_ONCE(dev_net(dev)->core.rps_default_mask);
1090
1091 if (rps_default_mask && !cpumask_empty(rps_default_mask))
1092 return netdev_rx_queue_set_rps_mask(queue, rps_default_mask);
1093#endif
1094 return 0;
1095}
1096
1097static int rx_queue_add_kobject(struct net_device *dev, int index)
1098{
1099 struct netdev_rx_queue *queue = dev->_rx + index;
1100 struct kobject *kobj = &queue->kobj;
1101 int error = 0;
1102
1103 /* Kobject_put later will trigger rx_queue_release call which
1104 * decreases dev refcount: Take that reference here
1105 */
1106 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1107
1108 kobj->kset = dev->queues_kset;
1109 error = kobject_init_and_add(kobj, &rx_queue_ktype, NULL,
1110 "rx-%u", index);
1111 if (error)
1112 goto err;
1113
1114 if (dev->sysfs_rx_queue_group) {
1115 error = sysfs_create_group(kobj, dev->sysfs_rx_queue_group);
1116 if (error)
1117 goto err;
1118 }
1119
1120 error = rx_queue_default_mask(dev, queue);
1121 if (error)
1122 goto err;
1123
1124 kobject_uevent(kobj, KOBJ_ADD);
1125
1126 return error;
1127
1128err:
1129 kobject_put(kobj);
1130 return error;
1131}
1132
1133static int rx_queue_change_owner(struct net_device *dev, int index, kuid_t kuid,
1134 kgid_t kgid)
1135{
1136 struct netdev_rx_queue *queue = dev->_rx + index;
1137 struct kobject *kobj = &queue->kobj;
1138 int error;
1139
1140 error = sysfs_change_owner(kobj, kuid, kgid);
1141 if (error)
1142 return error;
1143
1144 if (dev->sysfs_rx_queue_group)
1145 error = sysfs_group_change_owner(
1146 kobj, dev->sysfs_rx_queue_group, kuid, kgid);
1147
1148 return error;
1149}
1150#endif /* CONFIG_SYSFS */
1151
1152int
1153net_rx_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1154{
1155#ifdef CONFIG_SYSFS
1156 int i;
1157 int error = 0;
1158
1159#ifndef CONFIG_RPS
1160 if (!dev->sysfs_rx_queue_group)
1161 return 0;
1162#endif
1163 for (i = old_num; i < new_num; i++) {
1164 error = rx_queue_add_kobject(dev, i);
1165 if (error) {
1166 new_num = old_num;
1167 break;
1168 }
1169 }
1170
1171 while (--i >= new_num) {
1172 struct kobject *kobj = &dev->_rx[i].kobj;
1173
1174 if (!refcount_read(&dev_net(dev)->ns.count))
1175 kobj->uevent_suppress = 1;
1176 if (dev->sysfs_rx_queue_group)
1177 sysfs_remove_group(kobj, dev->sysfs_rx_queue_group);
1178 kobject_put(kobj);
1179 }
1180
1181 return error;
1182#else
1183 return 0;
1184#endif
1185}
1186
1187static int net_rx_queue_change_owner(struct net_device *dev, int num,
1188 kuid_t kuid, kgid_t kgid)
1189{
1190#ifdef CONFIG_SYSFS
1191 int error = 0;
1192 int i;
1193
1194#ifndef CONFIG_RPS
1195 if (!dev->sysfs_rx_queue_group)
1196 return 0;
1197#endif
1198 for (i = 0; i < num; i++) {
1199 error = rx_queue_change_owner(dev, i, kuid, kgid);
1200 if (error)
1201 break;
1202 }
1203
1204 return error;
1205#else
1206 return 0;
1207#endif
1208}
1209
1210#ifdef CONFIG_SYSFS
1211/*
1212 * netdev_queue sysfs structures and functions.
1213 */
1214struct netdev_queue_attribute {
1215 struct attribute attr;
1216 ssize_t (*show)(struct netdev_queue *queue, char *buf);
1217 ssize_t (*store)(struct netdev_queue *queue,
1218 const char *buf, size_t len);
1219};
1220#define to_netdev_queue_attr(_attr) \
1221 container_of(_attr, struct netdev_queue_attribute, attr)
1222
1223#define to_netdev_queue(obj) container_of(obj, struct netdev_queue, kobj)
1224
1225static ssize_t netdev_queue_attr_show(struct kobject *kobj,
1226 struct attribute *attr, char *buf)
1227{
1228 const struct netdev_queue_attribute *attribute
1229 = to_netdev_queue_attr(attr);
1230 struct netdev_queue *queue = to_netdev_queue(kobj);
1231
1232 if (!attribute->show)
1233 return -EIO;
1234
1235 return attribute->show(queue, buf);
1236}
1237
1238static ssize_t netdev_queue_attr_store(struct kobject *kobj,
1239 struct attribute *attr,
1240 const char *buf, size_t count)
1241{
1242 const struct netdev_queue_attribute *attribute
1243 = to_netdev_queue_attr(attr);
1244 struct netdev_queue *queue = to_netdev_queue(kobj);
1245
1246 if (!attribute->store)
1247 return -EIO;
1248
1249 return attribute->store(queue, buf, count);
1250}
1251
1252static const struct sysfs_ops netdev_queue_sysfs_ops = {
1253 .show = netdev_queue_attr_show,
1254 .store = netdev_queue_attr_store,
1255};
1256
1257static ssize_t tx_timeout_show(struct netdev_queue *queue, char *buf)
1258{
1259 unsigned long trans_timeout = atomic_long_read(&queue->trans_timeout);
1260
1261 return sysfs_emit(buf, fmt_ulong, trans_timeout);
1262}
1263
1264static unsigned int get_netdev_queue_index(struct netdev_queue *queue)
1265{
1266 struct net_device *dev = queue->dev;
1267 unsigned int i;
1268
1269 i = queue - dev->_tx;
1270 BUG_ON(i >= dev->num_tx_queues);
1271
1272 return i;
1273}
1274
1275static ssize_t traffic_class_show(struct netdev_queue *queue,
1276 char *buf)
1277{
1278 struct net_device *dev = queue->dev;
1279 int num_tc, tc;
1280 int index;
1281
1282 if (!netif_is_multiqueue(dev))
1283 return -ENOENT;
1284
1285 if (!rtnl_trylock())
1286 return restart_syscall();
1287
1288 index = get_netdev_queue_index(queue);
1289
1290 /* If queue belongs to subordinate dev use its TC mapping */
1291 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1292
1293 num_tc = dev->num_tc;
1294 tc = netdev_txq_to_tc(dev, index);
1295
1296 rtnl_unlock();
1297
1298 if (tc < 0)
1299 return -EINVAL;
1300
1301 /* We can report the traffic class one of two ways:
1302 * Subordinate device traffic classes are reported with the traffic
1303 * class first, and then the subordinate class so for example TC0 on
1304 * subordinate device 2 will be reported as "0-2". If the queue
1305 * belongs to the root device it will be reported with just the
1306 * traffic class, so just "0" for TC 0 for example.
1307 */
1308 return num_tc < 0 ? sysfs_emit(buf, "%d%d\n", tc, num_tc) :
1309 sysfs_emit(buf, "%d\n", tc);
1310}
1311
1312#ifdef CONFIG_XPS
1313static ssize_t tx_maxrate_show(struct netdev_queue *queue,
1314 char *buf)
1315{
1316 return sysfs_emit(buf, "%lu\n", queue->tx_maxrate);
1317}
1318
1319static ssize_t tx_maxrate_store(struct netdev_queue *queue,
1320 const char *buf, size_t len)
1321{
1322 struct net_device *dev = queue->dev;
1323 int err, index = get_netdev_queue_index(queue);
1324 u32 rate = 0;
1325
1326 if (!capable(CAP_NET_ADMIN))
1327 return -EPERM;
1328
1329 /* The check is also done later; this helps returning early without
1330 * hitting the trylock/restart below.
1331 */
1332 if (!dev->netdev_ops->ndo_set_tx_maxrate)
1333 return -EOPNOTSUPP;
1334
1335 err = kstrtou32(buf, 10, &rate);
1336 if (err < 0)
1337 return err;
1338
1339 if (!rtnl_trylock())
1340 return restart_syscall();
1341
1342 err = -EOPNOTSUPP;
1343 if (dev->netdev_ops->ndo_set_tx_maxrate)
1344 err = dev->netdev_ops->ndo_set_tx_maxrate(dev, index, rate);
1345
1346 rtnl_unlock();
1347 if (!err) {
1348 queue->tx_maxrate = rate;
1349 return len;
1350 }
1351 return err;
1352}
1353
1354static struct netdev_queue_attribute queue_tx_maxrate __ro_after_init
1355 = __ATTR_RW(tx_maxrate);
1356#endif
1357
1358static struct netdev_queue_attribute queue_trans_timeout __ro_after_init
1359 = __ATTR_RO(tx_timeout);
1360
1361static struct netdev_queue_attribute queue_traffic_class __ro_after_init
1362 = __ATTR_RO(traffic_class);
1363
1364#ifdef CONFIG_BQL
1365/*
1366 * Byte queue limits sysfs structures and functions.
1367 */
1368static ssize_t bql_show(char *buf, unsigned int value)
1369{
1370 return sysfs_emit(buf, "%u\n", value);
1371}
1372
1373static ssize_t bql_set(const char *buf, const size_t count,
1374 unsigned int *pvalue)
1375{
1376 unsigned int value;
1377 int err;
1378
1379 if (!strcmp(buf, "max") || !strcmp(buf, "max\n")) {
1380 value = DQL_MAX_LIMIT;
1381 } else {
1382 err = kstrtouint(buf, 10, &value);
1383 if (err < 0)
1384 return err;
1385 if (value > DQL_MAX_LIMIT)
1386 return -EINVAL;
1387 }
1388
1389 *pvalue = value;
1390
1391 return count;
1392}
1393
1394static ssize_t bql_show_hold_time(struct netdev_queue *queue,
1395 char *buf)
1396{
1397 struct dql *dql = &queue->dql;
1398
1399 return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->slack_hold_time));
1400}
1401
1402static ssize_t bql_set_hold_time(struct netdev_queue *queue,
1403 const char *buf, size_t len)
1404{
1405 struct dql *dql = &queue->dql;
1406 unsigned int value;
1407 int err;
1408
1409 err = kstrtouint(buf, 10, &value);
1410 if (err < 0)
1411 return err;
1412
1413 dql->slack_hold_time = msecs_to_jiffies(value);
1414
1415 return len;
1416}
1417
1418static struct netdev_queue_attribute bql_hold_time_attribute __ro_after_init
1419 = __ATTR(hold_time, 0644,
1420 bql_show_hold_time, bql_set_hold_time);
1421
1422static ssize_t bql_show_stall_thrs(struct netdev_queue *queue, char *buf)
1423{
1424 struct dql *dql = &queue->dql;
1425
1426 return sysfs_emit(buf, "%u\n", jiffies_to_msecs(dql->stall_thrs));
1427}
1428
1429static ssize_t bql_set_stall_thrs(struct netdev_queue *queue,
1430 const char *buf, size_t len)
1431{
1432 struct dql *dql = &queue->dql;
1433 unsigned int value;
1434 int err;
1435
1436 err = kstrtouint(buf, 10, &value);
1437 if (err < 0)
1438 return err;
1439
1440 value = msecs_to_jiffies(value);
1441 if (value && (value < 4 || value > 4 / 2 * BITS_PER_LONG))
1442 return -ERANGE;
1443
1444 if (!dql->stall_thrs && value)
1445 dql->last_reap = jiffies;
1446 /* Force last_reap to be live */
1447 smp_wmb();
1448 dql->stall_thrs = value;
1449
1450 return len;
1451}
1452
1453static struct netdev_queue_attribute bql_stall_thrs_attribute __ro_after_init =
1454 __ATTR(stall_thrs, 0644, bql_show_stall_thrs, bql_set_stall_thrs);
1455
1456static ssize_t bql_show_stall_max(struct netdev_queue *queue, char *buf)
1457{
1458 return sysfs_emit(buf, "%u\n", READ_ONCE(queue->dql.stall_max));
1459}
1460
1461static ssize_t bql_set_stall_max(struct netdev_queue *queue,
1462 const char *buf, size_t len)
1463{
1464 WRITE_ONCE(queue->dql.stall_max, 0);
1465 return len;
1466}
1467
1468static struct netdev_queue_attribute bql_stall_max_attribute __ro_after_init =
1469 __ATTR(stall_max, 0644, bql_show_stall_max, bql_set_stall_max);
1470
1471static ssize_t bql_show_stall_cnt(struct netdev_queue *queue, char *buf)
1472{
1473 struct dql *dql = &queue->dql;
1474
1475 return sysfs_emit(buf, "%lu\n", dql->stall_cnt);
1476}
1477
1478static struct netdev_queue_attribute bql_stall_cnt_attribute __ro_after_init =
1479 __ATTR(stall_cnt, 0444, bql_show_stall_cnt, NULL);
1480
1481static ssize_t bql_show_inflight(struct netdev_queue *queue,
1482 char *buf)
1483{
1484 struct dql *dql = &queue->dql;
1485
1486 return sysfs_emit(buf, "%u\n", dql->num_queued - dql->num_completed);
1487}
1488
1489static struct netdev_queue_attribute bql_inflight_attribute __ro_after_init =
1490 __ATTR(inflight, 0444, bql_show_inflight, NULL);
1491
1492#define BQL_ATTR(NAME, FIELD) \
1493static ssize_t bql_show_ ## NAME(struct netdev_queue *queue, \
1494 char *buf) \
1495{ \
1496 return bql_show(buf, queue->dql.FIELD); \
1497} \
1498 \
1499static ssize_t bql_set_ ## NAME(struct netdev_queue *queue, \
1500 const char *buf, size_t len) \
1501{ \
1502 return bql_set(buf, len, &queue->dql.FIELD); \
1503} \
1504 \
1505static struct netdev_queue_attribute bql_ ## NAME ## _attribute __ro_after_init \
1506 = __ATTR(NAME, 0644, \
1507 bql_show_ ## NAME, bql_set_ ## NAME)
1508
1509BQL_ATTR(limit, limit);
1510BQL_ATTR(limit_max, max_limit);
1511BQL_ATTR(limit_min, min_limit);
1512
1513static struct attribute *dql_attrs[] __ro_after_init = {
1514 &bql_limit_attribute.attr,
1515 &bql_limit_max_attribute.attr,
1516 &bql_limit_min_attribute.attr,
1517 &bql_hold_time_attribute.attr,
1518 &bql_inflight_attribute.attr,
1519 &bql_stall_thrs_attribute.attr,
1520 &bql_stall_cnt_attribute.attr,
1521 &bql_stall_max_attribute.attr,
1522 NULL
1523};
1524
1525static const struct attribute_group dql_group = {
1526 .name = "byte_queue_limits",
1527 .attrs = dql_attrs,
1528};
1529#else
1530/* Fake declaration, all the code using it should be dead */
1531static const struct attribute_group dql_group = {};
1532#endif /* CONFIG_BQL */
1533
1534#ifdef CONFIG_XPS
1535static ssize_t xps_queue_show(struct net_device *dev, unsigned int index,
1536 int tc, char *buf, enum xps_map_type type)
1537{
1538 struct xps_dev_maps *dev_maps;
1539 unsigned long *mask;
1540 unsigned int nr_ids;
1541 int j, len;
1542
1543 rcu_read_lock();
1544 dev_maps = rcu_dereference(dev->xps_maps[type]);
1545
1546 /* Default to nr_cpu_ids/dev->num_rx_queues and do not just return 0
1547 * when dev_maps hasn't been allocated yet, to be backward compatible.
1548 */
1549 nr_ids = dev_maps ? dev_maps->nr_ids :
1550 (type == XPS_CPUS ? nr_cpu_ids : dev->num_rx_queues);
1551
1552 mask = bitmap_zalloc(nr_ids, GFP_NOWAIT);
1553 if (!mask) {
1554 rcu_read_unlock();
1555 return -ENOMEM;
1556 }
1557
1558 if (!dev_maps || tc >= dev_maps->num_tc)
1559 goto out_no_maps;
1560
1561 for (j = 0; j < nr_ids; j++) {
1562 int i, tci = j * dev_maps->num_tc + tc;
1563 struct xps_map *map;
1564
1565 map = rcu_dereference(dev_maps->attr_map[tci]);
1566 if (!map)
1567 continue;
1568
1569 for (i = map->len; i--;) {
1570 if (map->queues[i] == index) {
1571 __set_bit(j, mask);
1572 break;
1573 }
1574 }
1575 }
1576out_no_maps:
1577 rcu_read_unlock();
1578
1579 len = bitmap_print_to_pagebuf(false, buf, mask, nr_ids);
1580 bitmap_free(mask);
1581
1582 return len < PAGE_SIZE ? len : -EINVAL;
1583}
1584
1585static ssize_t xps_cpus_show(struct netdev_queue *queue, char *buf)
1586{
1587 struct net_device *dev = queue->dev;
1588 unsigned int index;
1589 int len, tc;
1590
1591 if (!netif_is_multiqueue(dev))
1592 return -ENOENT;
1593
1594 index = get_netdev_queue_index(queue);
1595
1596 if (!rtnl_trylock())
1597 return restart_syscall();
1598
1599 /* If queue belongs to subordinate dev use its map */
1600 dev = netdev_get_tx_queue(dev, index)->sb_dev ? : dev;
1601
1602 tc = netdev_txq_to_tc(dev, index);
1603 if (tc < 0) {
1604 rtnl_unlock();
1605 return -EINVAL;
1606 }
1607
1608 /* Make sure the subordinate device can't be freed */
1609 get_device(&dev->dev);
1610 rtnl_unlock();
1611
1612 len = xps_queue_show(dev, index, tc, buf, XPS_CPUS);
1613
1614 put_device(&dev->dev);
1615 return len;
1616}
1617
1618static ssize_t xps_cpus_store(struct netdev_queue *queue,
1619 const char *buf, size_t len)
1620{
1621 struct net_device *dev = queue->dev;
1622 unsigned int index;
1623 cpumask_var_t mask;
1624 int err;
1625
1626 if (!netif_is_multiqueue(dev))
1627 return -ENOENT;
1628
1629 if (!capable(CAP_NET_ADMIN))
1630 return -EPERM;
1631
1632 if (!alloc_cpumask_var(&mask, GFP_KERNEL))
1633 return -ENOMEM;
1634
1635 index = get_netdev_queue_index(queue);
1636
1637 err = bitmap_parse(buf, len, cpumask_bits(mask), nr_cpumask_bits);
1638 if (err) {
1639 free_cpumask_var(mask);
1640 return err;
1641 }
1642
1643 if (!rtnl_trylock()) {
1644 free_cpumask_var(mask);
1645 return restart_syscall();
1646 }
1647
1648 err = netif_set_xps_queue(dev, mask, index);
1649 rtnl_unlock();
1650
1651 free_cpumask_var(mask);
1652
1653 return err ? : len;
1654}
1655
1656static struct netdev_queue_attribute xps_cpus_attribute __ro_after_init
1657 = __ATTR_RW(xps_cpus);
1658
1659static ssize_t xps_rxqs_show(struct netdev_queue *queue, char *buf)
1660{
1661 struct net_device *dev = queue->dev;
1662 unsigned int index;
1663 int tc;
1664
1665 index = get_netdev_queue_index(queue);
1666
1667 if (!rtnl_trylock())
1668 return restart_syscall();
1669
1670 tc = netdev_txq_to_tc(dev, index);
1671 rtnl_unlock();
1672 if (tc < 0)
1673 return -EINVAL;
1674
1675 return xps_queue_show(dev, index, tc, buf, XPS_RXQS);
1676}
1677
1678static ssize_t xps_rxqs_store(struct netdev_queue *queue, const char *buf,
1679 size_t len)
1680{
1681 struct net_device *dev = queue->dev;
1682 struct net *net = dev_net(dev);
1683 unsigned long *mask;
1684 unsigned int index;
1685 int err;
1686
1687 if (!ns_capable(net->user_ns, CAP_NET_ADMIN))
1688 return -EPERM;
1689
1690 mask = bitmap_zalloc(dev->num_rx_queues, GFP_KERNEL);
1691 if (!mask)
1692 return -ENOMEM;
1693
1694 index = get_netdev_queue_index(queue);
1695
1696 err = bitmap_parse(buf, len, mask, dev->num_rx_queues);
1697 if (err) {
1698 bitmap_free(mask);
1699 return err;
1700 }
1701
1702 if (!rtnl_trylock()) {
1703 bitmap_free(mask);
1704 return restart_syscall();
1705 }
1706
1707 cpus_read_lock();
1708 err = __netif_set_xps_queue(dev, mask, index, XPS_RXQS);
1709 cpus_read_unlock();
1710
1711 rtnl_unlock();
1712
1713 bitmap_free(mask);
1714 return err ? : len;
1715}
1716
1717static struct netdev_queue_attribute xps_rxqs_attribute __ro_after_init
1718 = __ATTR_RW(xps_rxqs);
1719#endif /* CONFIG_XPS */
1720
1721static struct attribute *netdev_queue_default_attrs[] __ro_after_init = {
1722 &queue_trans_timeout.attr,
1723 &queue_traffic_class.attr,
1724#ifdef CONFIG_XPS
1725 &xps_cpus_attribute.attr,
1726 &xps_rxqs_attribute.attr,
1727 &queue_tx_maxrate.attr,
1728#endif
1729 NULL
1730};
1731ATTRIBUTE_GROUPS(netdev_queue_default);
1732
1733static void netdev_queue_release(struct kobject *kobj)
1734{
1735 struct netdev_queue *queue = to_netdev_queue(kobj);
1736
1737 memset(kobj, 0, sizeof(*kobj));
1738 netdev_put(queue->dev, &queue->dev_tracker);
1739}
1740
1741static const void *netdev_queue_namespace(const struct kobject *kobj)
1742{
1743 struct netdev_queue *queue = to_netdev_queue(kobj);
1744 struct device *dev = &queue->dev->dev;
1745 const void *ns = NULL;
1746
1747 if (dev->class && dev->class->namespace)
1748 ns = dev->class->namespace(dev);
1749
1750 return ns;
1751}
1752
1753static void netdev_queue_get_ownership(const struct kobject *kobj,
1754 kuid_t *uid, kgid_t *gid)
1755{
1756 const struct net *net = netdev_queue_namespace(kobj);
1757
1758 net_ns_get_ownership(net, uid, gid);
1759}
1760
1761static const struct kobj_type netdev_queue_ktype = {
1762 .sysfs_ops = &netdev_queue_sysfs_ops,
1763 .release = netdev_queue_release,
1764 .default_groups = netdev_queue_default_groups,
1765 .namespace = netdev_queue_namespace,
1766 .get_ownership = netdev_queue_get_ownership,
1767};
1768
1769static bool netdev_uses_bql(const struct net_device *dev)
1770{
1771 if (dev->lltx || (dev->priv_flags & IFF_NO_QUEUE))
1772 return false;
1773
1774 return IS_ENABLED(CONFIG_BQL);
1775}
1776
1777static int netdev_queue_add_kobject(struct net_device *dev, int index)
1778{
1779 struct netdev_queue *queue = dev->_tx + index;
1780 struct kobject *kobj = &queue->kobj;
1781 int error = 0;
1782
1783 /* Kobject_put later will trigger netdev_queue_release call
1784 * which decreases dev refcount: Take that reference here
1785 */
1786 netdev_hold(queue->dev, &queue->dev_tracker, GFP_KERNEL);
1787
1788 kobj->kset = dev->queues_kset;
1789 error = kobject_init_and_add(kobj, &netdev_queue_ktype, NULL,
1790 "tx-%u", index);
1791 if (error)
1792 goto err;
1793
1794 if (netdev_uses_bql(dev)) {
1795 error = sysfs_create_group(kobj, &dql_group);
1796 if (error)
1797 goto err;
1798 }
1799
1800 kobject_uevent(kobj, KOBJ_ADD);
1801 return 0;
1802
1803err:
1804 kobject_put(kobj);
1805 return error;
1806}
1807
1808static int tx_queue_change_owner(struct net_device *ndev, int index,
1809 kuid_t kuid, kgid_t kgid)
1810{
1811 struct netdev_queue *queue = ndev->_tx + index;
1812 struct kobject *kobj = &queue->kobj;
1813 int error;
1814
1815 error = sysfs_change_owner(kobj, kuid, kgid);
1816 if (error)
1817 return error;
1818
1819 if (netdev_uses_bql(ndev))
1820 error = sysfs_group_change_owner(kobj, &dql_group, kuid, kgid);
1821
1822 return error;
1823}
1824#endif /* CONFIG_SYSFS */
1825
1826int
1827netdev_queue_update_kobjects(struct net_device *dev, int old_num, int new_num)
1828{
1829#ifdef CONFIG_SYSFS
1830 int i;
1831 int error = 0;
1832
1833 /* Tx queue kobjects are allowed to be updated when a device is being
1834 * unregistered, but solely to remove queues from qdiscs. Any path
1835 * adding queues should be fixed.
1836 */
1837 WARN(dev->reg_state == NETREG_UNREGISTERING && new_num > old_num,
1838 "New queues can't be registered after device unregistration.");
1839
1840 for (i = old_num; i < new_num; i++) {
1841 error = netdev_queue_add_kobject(dev, i);
1842 if (error) {
1843 new_num = old_num;
1844 break;
1845 }
1846 }
1847
1848 while (--i >= new_num) {
1849 struct netdev_queue *queue = dev->_tx + i;
1850
1851 if (!refcount_read(&dev_net(dev)->ns.count))
1852 queue->kobj.uevent_suppress = 1;
1853
1854 if (netdev_uses_bql(dev))
1855 sysfs_remove_group(&queue->kobj, &dql_group);
1856
1857 kobject_put(&queue->kobj);
1858 }
1859
1860 return error;
1861#else
1862 return 0;
1863#endif /* CONFIG_SYSFS */
1864}
1865
1866static int net_tx_queue_change_owner(struct net_device *dev, int num,
1867 kuid_t kuid, kgid_t kgid)
1868{
1869#ifdef CONFIG_SYSFS
1870 int error = 0;
1871 int i;
1872
1873 for (i = 0; i < num; i++) {
1874 error = tx_queue_change_owner(dev, i, kuid, kgid);
1875 if (error)
1876 break;
1877 }
1878
1879 return error;
1880#else
1881 return 0;
1882#endif /* CONFIG_SYSFS */
1883}
1884
1885static int register_queue_kobjects(struct net_device *dev)
1886{
1887 int error = 0, txq = 0, rxq = 0, real_rx = 0, real_tx = 0;
1888
1889#ifdef CONFIG_SYSFS
1890 dev->queues_kset = kset_create_and_add("queues",
1891 NULL, &dev->dev.kobj);
1892 if (!dev->queues_kset)
1893 return -ENOMEM;
1894 real_rx = dev->real_num_rx_queues;
1895#endif
1896 real_tx = dev->real_num_tx_queues;
1897
1898 error = net_rx_queue_update_kobjects(dev, 0, real_rx);
1899 if (error)
1900 goto error;
1901 rxq = real_rx;
1902
1903 error = netdev_queue_update_kobjects(dev, 0, real_tx);
1904 if (error)
1905 goto error;
1906 txq = real_tx;
1907
1908 return 0;
1909
1910error:
1911 netdev_queue_update_kobjects(dev, txq, 0);
1912 net_rx_queue_update_kobjects(dev, rxq, 0);
1913#ifdef CONFIG_SYSFS
1914 kset_unregister(dev->queues_kset);
1915#endif
1916 return error;
1917}
1918
1919static int queue_change_owner(struct net_device *ndev, kuid_t kuid, kgid_t kgid)
1920{
1921 int error = 0, real_rx = 0, real_tx = 0;
1922
1923#ifdef CONFIG_SYSFS
1924 if (ndev->queues_kset) {
1925 error = sysfs_change_owner(&ndev->queues_kset->kobj, kuid, kgid);
1926 if (error)
1927 return error;
1928 }
1929 real_rx = ndev->real_num_rx_queues;
1930#endif
1931 real_tx = ndev->real_num_tx_queues;
1932
1933 error = net_rx_queue_change_owner(ndev, real_rx, kuid, kgid);
1934 if (error)
1935 return error;
1936
1937 error = net_tx_queue_change_owner(ndev, real_tx, kuid, kgid);
1938 if (error)
1939 return error;
1940
1941 return 0;
1942}
1943
1944static void remove_queue_kobjects(struct net_device *dev)
1945{
1946 int real_rx = 0, real_tx = 0;
1947
1948#ifdef CONFIG_SYSFS
1949 real_rx = dev->real_num_rx_queues;
1950#endif
1951 real_tx = dev->real_num_tx_queues;
1952
1953 net_rx_queue_update_kobjects(dev, real_rx, 0);
1954 netdev_queue_update_kobjects(dev, real_tx, 0);
1955
1956 dev->real_num_rx_queues = 0;
1957 dev->real_num_tx_queues = 0;
1958#ifdef CONFIG_SYSFS
1959 kset_unregister(dev->queues_kset);
1960#endif
1961}
1962
1963static bool net_current_may_mount(void)
1964{
1965 struct net *net = current->nsproxy->net_ns;
1966
1967 return ns_capable(net->user_ns, CAP_SYS_ADMIN);
1968}
1969
1970static void *net_grab_current_ns(void)
1971{
1972 struct net *ns = current->nsproxy->net_ns;
1973#ifdef CONFIG_NET_NS
1974 if (ns)
1975 refcount_inc(&ns->passive);
1976#endif
1977 return ns;
1978}
1979
1980static const void *net_initial_ns(void)
1981{
1982 return &init_net;
1983}
1984
1985static const void *net_netlink_ns(struct sock *sk)
1986{
1987 return sock_net(sk);
1988}
1989
1990const struct kobj_ns_type_operations net_ns_type_operations = {
1991 .type = KOBJ_NS_TYPE_NET,
1992 .current_may_mount = net_current_may_mount,
1993 .grab_current_ns = net_grab_current_ns,
1994 .netlink_ns = net_netlink_ns,
1995 .initial_ns = net_initial_ns,
1996 .drop_ns = net_drop_ns,
1997};
1998EXPORT_SYMBOL_GPL(net_ns_type_operations);
1999
2000static int netdev_uevent(const struct device *d, struct kobj_uevent_env *env)
2001{
2002 const struct net_device *dev = to_net_dev(d);
2003 int retval;
2004
2005 /* pass interface to uevent. */
2006 retval = add_uevent_var(env, "INTERFACE=%s", dev->name);
2007 if (retval)
2008 goto exit;
2009
2010 /* pass ifindex to uevent.
2011 * ifindex is useful as it won't change (interface name may change)
2012 * and is what RtNetlink uses natively.
2013 */
2014 retval = add_uevent_var(env, "IFINDEX=%d", dev->ifindex);
2015
2016exit:
2017 return retval;
2018}
2019
2020/*
2021 * netdev_release -- destroy and free a dead device.
2022 * Called when last reference to device kobject is gone.
2023 */
2024static void netdev_release(struct device *d)
2025{
2026 struct net_device *dev = to_net_dev(d);
2027
2028 BUG_ON(dev->reg_state != NETREG_RELEASED);
2029
2030 /* no need to wait for rcu grace period:
2031 * device is dead and about to be freed.
2032 */
2033 kfree(rcu_access_pointer(dev->ifalias));
2034 kvfree(dev);
2035}
2036
2037static const void *net_namespace(const struct device *d)
2038{
2039 const struct net_device *dev = to_net_dev(d);
2040
2041 return dev_net(dev);
2042}
2043
2044static void net_get_ownership(const struct device *d, kuid_t *uid, kgid_t *gid)
2045{
2046 const struct net_device *dev = to_net_dev(d);
2047 const struct net *net = dev_net(dev);
2048
2049 net_ns_get_ownership(net, uid, gid);
2050}
2051
2052static const struct class net_class = {
2053 .name = "net",
2054 .dev_release = netdev_release,
2055 .dev_groups = net_class_groups,
2056 .dev_uevent = netdev_uevent,
2057 .ns_type = &net_ns_type_operations,
2058 .namespace = net_namespace,
2059 .get_ownership = net_get_ownership,
2060};
2061
2062#ifdef CONFIG_OF
2063static int of_dev_node_match(struct device *dev, const void *data)
2064{
2065 for (; dev; dev = dev->parent) {
2066 if (dev->of_node == data)
2067 return 1;
2068 }
2069
2070 return 0;
2071}
2072
2073/*
2074 * of_find_net_device_by_node - lookup the net device for the device node
2075 * @np: OF device node
2076 *
2077 * Looks up the net_device structure corresponding with the device node.
2078 * If successful, returns a pointer to the net_device with the embedded
2079 * struct device refcount incremented by one, or NULL on failure. The
2080 * refcount must be dropped when done with the net_device.
2081 */
2082struct net_device *of_find_net_device_by_node(struct device_node *np)
2083{
2084 struct device *dev;
2085
2086 dev = class_find_device(&net_class, NULL, np, of_dev_node_match);
2087 if (!dev)
2088 return NULL;
2089
2090 return to_net_dev(dev);
2091}
2092EXPORT_SYMBOL(of_find_net_device_by_node);
2093#endif
2094
2095/* Delete sysfs entries but hold kobject reference until after all
2096 * netdev references are gone.
2097 */
2098void netdev_unregister_kobject(struct net_device *ndev)
2099{
2100 struct device *dev = &ndev->dev;
2101
2102 if (!refcount_read(&dev_net(ndev)->ns.count))
2103 dev_set_uevent_suppress(dev, 1);
2104
2105 kobject_get(&dev->kobj);
2106
2107 remove_queue_kobjects(ndev);
2108
2109 pm_runtime_set_memalloc_noio(dev, false);
2110
2111 device_del(dev);
2112}
2113
2114/* Create sysfs entries for network device. */
2115int netdev_register_kobject(struct net_device *ndev)
2116{
2117 struct device *dev = &ndev->dev;
2118 const struct attribute_group **groups = ndev->sysfs_groups;
2119 int error = 0;
2120
2121 device_initialize(dev);
2122 dev->class = &net_class;
2123 dev->platform_data = ndev;
2124 dev->groups = groups;
2125
2126 dev_set_name(dev, "%s", ndev->name);
2127
2128#ifdef CONFIG_SYSFS
2129 /* Allow for a device specific group */
2130 if (*groups)
2131 groups++;
2132
2133 *groups++ = &netstat_group;
2134
2135 if (wireless_group_needed(ndev))
2136 *groups++ = &wireless_group;
2137#endif /* CONFIG_SYSFS */
2138
2139 error = device_add(dev);
2140 if (error)
2141 return error;
2142
2143 error = register_queue_kobjects(ndev);
2144 if (error) {
2145 device_del(dev);
2146 return error;
2147 }
2148
2149 pm_runtime_set_memalloc_noio(dev, true);
2150
2151 return error;
2152}
2153
2154/* Change owner for sysfs entries when moving network devices across network
2155 * namespaces owned by different user namespaces.
2156 */
2157int netdev_change_owner(struct net_device *ndev, const struct net *net_old,
2158 const struct net *net_new)
2159{
2160 kuid_t old_uid = GLOBAL_ROOT_UID, new_uid = GLOBAL_ROOT_UID;
2161 kgid_t old_gid = GLOBAL_ROOT_GID, new_gid = GLOBAL_ROOT_GID;
2162 struct device *dev = &ndev->dev;
2163 int error;
2164
2165 net_ns_get_ownership(net_old, &old_uid, &old_gid);
2166 net_ns_get_ownership(net_new, &new_uid, &new_gid);
2167
2168 /* The network namespace was changed but the owning user namespace is
2169 * identical so there's no need to change the owner of sysfs entries.
2170 */
2171 if (uid_eq(old_uid, new_uid) && gid_eq(old_gid, new_gid))
2172 return 0;
2173
2174 error = device_change_owner(dev, new_uid, new_gid);
2175 if (error)
2176 return error;
2177
2178 error = queue_change_owner(ndev, new_uid, new_gid);
2179 if (error)
2180 return error;
2181
2182 return 0;
2183}
2184
2185int netdev_class_create_file_ns(const struct class_attribute *class_attr,
2186 const void *ns)
2187{
2188 return class_create_file_ns(&net_class, class_attr, ns);
2189}
2190EXPORT_SYMBOL(netdev_class_create_file_ns);
2191
2192void netdev_class_remove_file_ns(const struct class_attribute *class_attr,
2193 const void *ns)
2194{
2195 class_remove_file_ns(&net_class, class_attr, ns);
2196}
2197EXPORT_SYMBOL(netdev_class_remove_file_ns);
2198
2199int __init netdev_kobject_init(void)
2200{
2201 kobj_ns_type_register(&net_ns_type_operations);
2202 return class_register(&net_class);
2203}