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