xref: /linux/net/core/dev.h (revision 0e59523b440c90ed1899476f32ad4e949282f43e)
1 /* SPDX-License-Identifier: GPL-2.0-or-later */
2 #ifndef _NET_CORE_DEV_H
3 #define _NET_CORE_DEV_H
4 
5 #include <linux/cleanup.h>
6 #include <linux/types.h>
7 #include <linux/rwsem.h>
8 #include <linux/netdevice.h>
9 #include <net/netdev_lock.h>
10 
11 struct net;
12 struct netlink_ext_ack;
13 struct netdev_queue_config;
14 struct cpumask;
15 struct pp_memory_provider_params;
16 
17 /* Random bits of netdevice that don't need to be exposed */
18 #define FLOW_LIMIT_HISTORY	(1 << 7)  /* must be ^2 and !overflow buckets */
19 struct sd_flow_limit {
20 	struct rcu_head		rcu;
21 	unsigned int		count;
22 	u8			log_buckets;
23 	unsigned int		history_head;
24 	u16			history[FLOW_LIMIT_HISTORY];
25 	u8			buckets[];
26 };
27 
28 extern int netdev_flow_limit_table_len;
29 
30 struct napi_struct *
31 netdev_napi_by_id_lock(struct net *net, unsigned int napi_id);
32 struct net_device *dev_get_by_napi_id(unsigned int napi_id);
33 
34 struct net_device *netdev_put_lock(struct net_device *dev, struct net *net,
35 				   netdevice_tracker *tracker);
36 
37 static inline struct net_device *
38 __netdev_put_lock(struct net_device *dev, struct net *net)
39 {
40 	return netdev_put_lock(dev, net, NULL);
41 }
42 
43 struct net_device *
44 netdev_xa_find_lock(struct net *net, struct net_device *dev,
45 		    unsigned long *index);
46 
47 DEFINE_FREE(netdev_unlock, struct net_device *, if (_T) netdev_unlock(_T));
48 
49 #define for_each_netdev_lock_scoped(net, var_name, ifindex)		\
50 	for (struct net_device *var_name __free(netdev_unlock) = NULL;	\
51 	     (var_name = netdev_xa_find_lock(net, var_name, &ifindex)); \
52 	     ifindex++)
53 
54 struct net_device *
55 netdev_get_by_index_lock_ops_compat(struct net *net, int ifindex);
56 struct net_device *
57 netdev_xa_find_lock_ops_compat(struct net *net, struct net_device *dev,
58 			       unsigned long *index);
59 
60 DEFINE_FREE(netdev_unlock_ops_compat, struct net_device *,
61 	    if (_T) netdev_unlock_ops_compat(_T));
62 
63 #define for_each_netdev_lock_ops_compat_scoped(net, var_name, ifindex)	\
64 	for (struct net_device *var_name __free(netdev_unlock_ops_compat) = NULL; \
65 	     (var_name = netdev_xa_find_lock_ops_compat(net, var_name,	\
66 							&ifindex));	\
67 	     ifindex++)
68 
69 #ifdef CONFIG_PROC_FS
70 int __init dev_proc_init(void);
71 #else
72 #define dev_proc_init() 0
73 #endif
74 
75 void linkwatch_init_dev(struct net_device *dev);
76 void linkwatch_run_queue(void);
77 
78 void dev_addr_flush(struct net_device *dev);
79 int dev_addr_init(struct net_device *dev);
80 void dev_addr_check(struct net_device *dev);
81 void __hw_addr_flush(struct netdev_hw_addr_list *list);
82 
83 #if IS_ENABLED(CONFIG_NET_SHAPER)
84 void net_shaper_flush_netdev(struct net_device *dev);
85 void net_shaper_set_real_num_tx_queues(struct net_device *dev,
86 				       unsigned int txq);
87 #else
88 static inline void net_shaper_flush_netdev(struct net_device *dev) {}
89 static inline void net_shaper_set_real_num_tx_queues(struct net_device *dev,
90 						     unsigned int txq) {}
91 #endif
92 
93 /* sysctls not referred to from outside net/core/ */
94 extern int		netdev_unregister_timeout_secs;
95 extern int		weight_p;
96 extern int		dev_weight_rx_bias;
97 extern int		dev_weight_tx_bias;
98 extern u8		netdev_rss_key[NETDEV_RSS_KEY_LEN] __read_mostly;
99 extern bool		netdev_rss_key_initialized;
100 
101 extern struct rw_semaphore dev_addr_sem;
102 
103 /* rtnl helpers */
104 extern struct list_head net_todo_list;
105 void netdev_run_todo(void);
106 
107 int netdev_alloc_config(struct net_device *dev);
108 void __netdev_free_config(struct netdev_config *cfg);
109 void netdev_free_config(struct net_device *dev);
110 int netdev_reconfig_start(struct net_device *dev);
111 
112 int netdev_queue_config_validate(struct net_device *dev, int rxq_idx,
113 				 struct netdev_queue_config *qcfg,
114 				 struct netlink_ext_ack *extack);
115 int netdev_queue_config_revalidate(struct net_device *dev,
116 				   struct netlink_ext_ack *extack);
117 
118 bool netif_rxq_has_mp(struct net_device *dev, unsigned int rxq_idx);
119 bool netif_rxq_is_leased(struct net_device *dev, unsigned int rxq_idx);
120 bool netif_is_queue_leasee(const struct net_device *dev);
121 
122 void __netif_mp_uninstall_rxq(struct netdev_rx_queue *rxq,
123 			      const struct pp_memory_provider_params *p);
124 
125 void netif_rxq_cleanup_unlease(struct netdev_rx_queue *phys_rxq,
126 			       struct netdev_rx_queue *virt_rxq);
127 
128 /* netdev management, shared between various uAPI entry points */
129 struct netdev_name_node {
130 	struct hlist_node hlist;
131 	struct list_head list;
132 	struct net_device *dev;
133 	const char *name;
134 	struct rcu_head rcu;
135 };
136 
137 int netdev_get_name(struct net *net, char *name, int ifindex);
138 int netif_change_name(struct net_device *dev, const char *newname);
139 int dev_change_name(struct net_device *dev, const char *newname);
140 
141 #define netdev_for_each_altname(dev, namenode)				\
142 	list_for_each_entry((namenode), &(dev)->name_node->list, list)
143 #define netdev_for_each_altname_safe(dev, namenode, next)		\
144 	list_for_each_entry_safe((namenode), (next), &(dev)->name_node->list, \
145 				 list)
146 
147 int netdev_name_node_alt_create(struct net_device *dev, const char *name);
148 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name);
149 
150 int dev_validate_mtu(struct net_device *dev, int mtu,
151 		     struct netlink_ext_ack *extack);
152 int netif_set_mtu_ext(struct net_device *dev, int new_mtu,
153 		      struct netlink_ext_ack *extack);
154 
155 int dev_get_phys_port_id(struct net_device *dev,
156 			 struct netdev_phys_item_id *ppid);
157 int dev_get_phys_port_name(struct net_device *dev,
158 			   char *name, size_t len);
159 
160 int netif_change_proto_down(struct net_device *dev, bool proto_down);
161 int dev_change_proto_down(struct net_device *dev, bool proto_down);
162 void netdev_change_proto_down_reason_locked(struct net_device *dev,
163 					    unsigned long mask, u32 value);
164 
165 typedef int (*bpf_op_t)(struct net_device *dev, struct netdev_bpf *bpf);
166 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack,
167 		      int fd, int expected_fd, u32 flags);
168 
169 int netif_change_tx_queue_len(struct net_device *dev, unsigned long new_len);
170 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len);
171 void netif_set_group(struct net_device *dev, int new_group);
172 void dev_set_group(struct net_device *dev, int new_group);
173 int netif_change_carrier(struct net_device *dev, bool new_carrier);
174 int dev_change_carrier(struct net_device *dev, bool new_carrier);
175 
176 void __dev_set_rx_mode(struct net_device *dev);
177 int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify);
178 void netif_rx_mode_init(struct net_device *dev);
179 void netif_rx_mode_run(struct net_device *dev);
180 void netif_rx_mode_sync(struct net_device *dev);
181 void netif_rx_mode_cancel_retry(struct net_device *dev);
182 
183 /* Events for the async netdev work, tracked in netdev->work_core_pending. */
184 enum netdev_work_core {
185 	NETDEV_WORK_RX_MODE	= BIT(0),	/* run the rx_mode update */
186 };
187 
188 void __netdev_work_core_sched(struct net_device *dev, unsigned long event);
189 unsigned long
190 __netdev_work_core_cancel(struct net_device *dev, unsigned long mask);
191 void netdev_work_cancel_all(struct net_device *dev);
192 
193 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags,
194 			unsigned int gchanges, u32 portid,
195 			const struct nlmsghdr *nlh);
196 
197 void unregister_netdevice_many_notify(struct list_head *head,
198 				      u32 portid, const struct nlmsghdr *nlh);
199 
200 static inline void netif_set_up(struct net_device *dev, bool value)
201 {
202 	if (value)
203 		dev->flags |= IFF_UP;
204 	else
205 		dev->flags &= ~IFF_UP;
206 
207 	if (!netdev_need_ops_lock(dev))
208 		netdev_lock(dev);
209 	dev->up = value;
210 	if (!netdev_need_ops_lock(dev))
211 		netdev_unlock(dev);
212 }
213 
214 static inline void netif_set_gso_max_size(struct net_device *dev,
215 					  unsigned int size)
216 {
217 	/* dev->gso_max_size is read locklessly from sk_setup_caps() */
218 	WRITE_ONCE(dev->gso_max_size, size);
219 	if (size <= GSO_LEGACY_MAX_SIZE)
220 		WRITE_ONCE(dev->gso_ipv4_max_size, size);
221 }
222 
223 static inline void netif_set_gso_max_segs(struct net_device *dev,
224 					  unsigned int segs)
225 {
226 	/* dev->gso_max_segs is read locklessly from sk_setup_caps() */
227 	WRITE_ONCE(dev->gso_max_segs, segs);
228 }
229 
230 static inline void netif_set_gro_max_size(struct net_device *dev,
231 					  unsigned int size)
232 {
233 	/* This pairs with the READ_ONCE() in skb_gro_receive() */
234 	WRITE_ONCE(dev->gro_max_size, size);
235 	if (size <= GRO_LEGACY_MAX_SIZE)
236 		WRITE_ONCE(dev->gro_ipv4_max_size, size);
237 }
238 
239 static inline void netif_set_gso_ipv4_max_size(struct net_device *dev,
240 					       unsigned int size)
241 {
242 	/* dev->gso_ipv4_max_size is read locklessly from sk_setup_caps() */
243 	WRITE_ONCE(dev->gso_ipv4_max_size, size);
244 }
245 
246 static inline void netif_set_gro_ipv4_max_size(struct net_device *dev,
247 					       unsigned int size)
248 {
249 	/* This pairs with the READ_ONCE() in skb_gro_receive() */
250 	WRITE_ONCE(dev->gro_ipv4_max_size, size);
251 }
252 
253 /**
254  * napi_get_defer_hard_irqs - get the NAPI's defer_hard_irqs
255  * @n: napi struct to get the defer_hard_irqs field from
256  *
257  * Return: the per-NAPI value of the defar_hard_irqs field.
258  */
259 static inline u32 napi_get_defer_hard_irqs(const struct napi_struct *n)
260 {
261 	return READ_ONCE(n->defer_hard_irqs);
262 }
263 
264 /**
265  * napi_set_defer_hard_irqs - set the defer_hard_irqs for a napi
266  * @n: napi_struct to set the defer_hard_irqs field
267  * @defer: the value the field should be set to
268  */
269 static inline void napi_set_defer_hard_irqs(struct napi_struct *n, u32 defer)
270 {
271 	WRITE_ONCE(n->defer_hard_irqs, defer);
272 }
273 
274 /**
275  * netdev_set_defer_hard_irqs - set defer_hard_irqs for all NAPIs of a netdev
276  * @netdev: the net_device for which all NAPIs will have defer_hard_irqs set
277  * @defer: the defer_hard_irqs value to set
278  */
279 static inline void netdev_set_defer_hard_irqs(struct net_device *netdev,
280 					      u32 defer)
281 {
282 	unsigned int count = max(netdev->num_rx_queues,
283 				 netdev->num_tx_queues);
284 	struct napi_struct *napi;
285 	int i;
286 
287 	WRITE_ONCE(netdev->napi_defer_hard_irqs, defer);
288 	list_for_each_entry(napi, &netdev->napi_list, dev_list)
289 		napi_set_defer_hard_irqs(napi, defer);
290 
291 	for (i = 0; i < count; i++)
292 		netdev->napi_config[i].defer_hard_irqs = defer;
293 }
294 
295 /**
296  * napi_get_gro_flush_timeout - get the gro_flush_timeout
297  * @n: napi struct to get the gro_flush_timeout from
298  *
299  * Return: the per-NAPI value of the gro_flush_timeout field.
300  */
301 static inline unsigned long
302 napi_get_gro_flush_timeout(const struct napi_struct *n)
303 {
304 	return READ_ONCE(n->gro_flush_timeout);
305 }
306 
307 /**
308  * napi_set_gro_flush_timeout - set the gro_flush_timeout for a napi
309  * @n: napi struct to set the gro_flush_timeout
310  * @timeout: timeout value to set
311  *
312  * napi_set_gro_flush_timeout sets the per-NAPI gro_flush_timeout
313  */
314 static inline void napi_set_gro_flush_timeout(struct napi_struct *n,
315 					      unsigned long timeout)
316 {
317 	WRITE_ONCE(n->gro_flush_timeout, timeout);
318 }
319 
320 /**
321  * netdev_set_gro_flush_timeout - set gro_flush_timeout of a netdev's NAPIs
322  * @netdev: the net_device for which all NAPIs will have gro_flush_timeout set
323  * @timeout: the timeout value to set
324  */
325 static inline void netdev_set_gro_flush_timeout(struct net_device *netdev,
326 						unsigned long timeout)
327 {
328 	unsigned int count = max(netdev->num_rx_queues,
329 				 netdev->num_tx_queues);
330 	struct napi_struct *napi;
331 	int i;
332 
333 	WRITE_ONCE(netdev->gro_flush_timeout, timeout);
334 	list_for_each_entry(napi, &netdev->napi_list, dev_list)
335 		napi_set_gro_flush_timeout(napi, timeout);
336 
337 	for (i = 0; i < count; i++)
338 		netdev->napi_config[i].gro_flush_timeout = timeout;
339 }
340 
341 /**
342  * napi_get_irq_suspend_timeout - get the irq_suspend_timeout
343  * @n: napi struct to get the irq_suspend_timeout from
344  *
345  * Return: the per-NAPI value of the irq_suspend_timeout field.
346  */
347 static inline unsigned long
348 napi_get_irq_suspend_timeout(const struct napi_struct *n)
349 {
350 	return READ_ONCE(n->irq_suspend_timeout);
351 }
352 
353 /**
354  * napi_set_irq_suspend_timeout - set the irq_suspend_timeout for a napi
355  * @n: napi struct to set the irq_suspend_timeout
356  * @timeout: timeout value to set
357  *
358  * napi_set_irq_suspend_timeout sets the per-NAPI irq_suspend_timeout
359  */
360 static inline void napi_set_irq_suspend_timeout(struct napi_struct *n,
361 						unsigned long timeout)
362 {
363 	WRITE_ONCE(n->irq_suspend_timeout, timeout);
364 }
365 
366 static inline enum netdev_napi_threaded napi_get_threaded(struct napi_struct *n)
367 {
368 	if (test_bit(NAPI_STATE_THREADED_BUSY_POLL, &n->state))
369 		return NETDEV_NAPI_THREADED_BUSY_POLL;
370 
371 	if (test_bit(NAPI_STATE_THREADED, &n->state))
372 		return NETDEV_NAPI_THREADED_ENABLED;
373 
374 	return NETDEV_NAPI_THREADED_DISABLED;
375 }
376 
377 static inline enum netdev_napi_threaded
378 napi_get_threaded_config(struct net_device *dev, struct napi_struct *n)
379 {
380 	if (n->config)
381 		return n->config->threaded;
382 	return dev->threaded;
383 }
384 
385 int napi_set_threaded(struct napi_struct *n,
386 		      enum netdev_napi_threaded threaded);
387 
388 int netif_set_threaded(struct net_device *dev,
389 		       enum netdev_napi_threaded threaded);
390 
391 int rps_cpumask_housekeeping(struct cpumask *mask);
392 
393 #if defined(CONFIG_DEBUG_NET) && defined(CONFIG_BPF_SYSCALL)
394 void xdp_do_check_flushed(struct napi_struct *napi);
395 #else
396 static inline void xdp_do_check_flushed(struct napi_struct *napi) { }
397 #endif
398 
399 /* Best effort check that NAPI is not idle (can't be scheduled to run) */
400 static inline void napi_assert_will_not_race(const struct napi_struct *napi)
401 {
402 	/* uninitialized instance, can't race */
403 	if (!napi->poll_list.next)
404 		return;
405 
406 	/* SCHED bit is set on disabled instances */
407 	WARN_ON(!test_bit(NAPI_STATE_SCHED, &napi->state));
408 	WARN_ON(READ_ONCE(napi->list_owner) != -1);
409 }
410 
411 struct skb_defer_node;
412 void skb_defer_node_flush(struct skb_defer_node *sdn);
413 void kick_defer_list_purge(unsigned int cpu);
414 
415 int dev_set_hwtstamp_phylib(struct net_device *dev,
416 			    struct kernel_hwtstamp_config *cfg,
417 			    struct netlink_ext_ack *extack);
418 int dev_get_hwtstamp_phylib(struct net_device *dev,
419 			    struct kernel_hwtstamp_config *cfg);
420 int net_hwtstamp_validate(const struct kernel_hwtstamp_config *cfg);
421 
422 #endif
423