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