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_queue_config_validate(struct net_device *dev, int rxq_idx, 108 struct netdev_queue_config *qcfg, 109 struct netlink_ext_ack *extack); 110 111 bool netif_rxq_has_mp(struct net_device *dev, unsigned int rxq_idx); 112 bool netif_rxq_is_leased(struct net_device *dev, unsigned int rxq_idx); 113 bool netif_is_queue_leasee(const struct net_device *dev); 114 115 void __netif_mp_uninstall_rxq(struct netdev_rx_queue *rxq, 116 const struct pp_memory_provider_params *p); 117 118 void netif_rxq_cleanup_unlease(struct netdev_rx_queue *phys_rxq, 119 struct netdev_rx_queue *virt_rxq); 120 121 /* netdev management, shared between various uAPI entry points */ 122 struct netdev_name_node { 123 struct hlist_node hlist; 124 struct list_head list; 125 struct net_device *dev; 126 const char *name; 127 struct rcu_head rcu; 128 }; 129 130 int netdev_get_name(struct net *net, char *name, int ifindex); 131 int netif_change_name(struct net_device *dev, const char *newname); 132 int dev_change_name(struct net_device *dev, const char *newname); 133 134 #define netdev_for_each_altname(dev, namenode) \ 135 list_for_each_entry((namenode), &(dev)->name_node->list, list) 136 #define netdev_for_each_altname_safe(dev, namenode, next) \ 137 list_for_each_entry_safe((namenode), (next), &(dev)->name_node->list, \ 138 list) 139 140 int netdev_name_node_alt_create(struct net_device *dev, const char *name); 141 int netdev_name_node_alt_destroy(struct net_device *dev, const char *name); 142 143 int dev_validate_mtu(struct net_device *dev, int mtu, 144 struct netlink_ext_ack *extack); 145 int netif_set_mtu_ext(struct net_device *dev, int new_mtu, 146 struct netlink_ext_ack *extack); 147 148 int dev_get_phys_port_id(struct net_device *dev, 149 struct netdev_phys_item_id *ppid); 150 int dev_get_phys_port_name(struct net_device *dev, 151 char *name, size_t len); 152 153 int netif_change_proto_down(struct net_device *dev, bool proto_down); 154 int dev_change_proto_down(struct net_device *dev, bool proto_down); 155 void netdev_change_proto_down_reason_locked(struct net_device *dev, 156 unsigned long mask, u32 value); 157 158 typedef int (*bpf_op_t)(struct net_device *dev, struct netdev_bpf *bpf); 159 int dev_change_xdp_fd(struct net_device *dev, struct netlink_ext_ack *extack, 160 int fd, int expected_fd, u32 flags); 161 162 int netif_change_tx_queue_len(struct net_device *dev, unsigned long new_len); 163 int dev_change_tx_queue_len(struct net_device *dev, unsigned long new_len); 164 void netif_set_group(struct net_device *dev, int new_group); 165 void dev_set_group(struct net_device *dev, int new_group); 166 int netif_change_carrier(struct net_device *dev, bool new_carrier); 167 int dev_change_carrier(struct net_device *dev, bool new_carrier); 168 169 void __dev_set_rx_mode(struct net_device *dev); 170 int __dev_set_promiscuity(struct net_device *dev, int inc, bool notify); 171 void netif_rx_mode_init(struct net_device *dev); 172 void netif_rx_mode_run(struct net_device *dev); 173 void netif_rx_mode_sync(struct net_device *dev); 174 void netif_rx_mode_cancel_retry(struct net_device *dev); 175 176 /* Events for the async netdev work, tracked in netdev->work_core_pending. */ 177 enum netdev_work_core { 178 NETDEV_WORK_RX_MODE = BIT(0), /* run the rx_mode update */ 179 }; 180 181 void __netdev_work_core_sched(struct net_device *dev, unsigned long event); 182 unsigned long 183 __netdev_work_core_cancel(struct net_device *dev, unsigned long mask); 184 void netdev_work_cancel_all(struct net_device *dev); 185 186 void __dev_notify_flags(struct net_device *dev, unsigned int old_flags, 187 unsigned int gchanges, u32 portid, 188 const struct nlmsghdr *nlh); 189 190 void unregister_netdevice_many_notify(struct list_head *head, 191 u32 portid, const struct nlmsghdr *nlh); 192 193 static inline void netif_set_up(struct net_device *dev, bool value) 194 { 195 if (value) 196 dev->flags |= IFF_UP; 197 else 198 dev->flags &= ~IFF_UP; 199 200 if (!netdev_need_ops_lock(dev)) 201 netdev_lock(dev); 202 dev->up = value; 203 if (!netdev_need_ops_lock(dev)) 204 netdev_unlock(dev); 205 } 206 207 static inline void netif_set_gso_max_size(struct net_device *dev, 208 unsigned int size) 209 { 210 /* dev->gso_max_size is read locklessly from sk_setup_caps() */ 211 WRITE_ONCE(dev->gso_max_size, size); 212 if (size <= GSO_LEGACY_MAX_SIZE) 213 WRITE_ONCE(dev->gso_ipv4_max_size, size); 214 } 215 216 static inline void netif_set_gso_max_segs(struct net_device *dev, 217 unsigned int segs) 218 { 219 /* dev->gso_max_segs is read locklessly from sk_setup_caps() */ 220 WRITE_ONCE(dev->gso_max_segs, segs); 221 } 222 223 static inline void netif_set_gro_max_size(struct net_device *dev, 224 unsigned int size) 225 { 226 /* This pairs with the READ_ONCE() in skb_gro_receive() */ 227 WRITE_ONCE(dev->gro_max_size, size); 228 if (size <= GRO_LEGACY_MAX_SIZE) 229 WRITE_ONCE(dev->gro_ipv4_max_size, size); 230 } 231 232 static inline void netif_set_gso_ipv4_max_size(struct net_device *dev, 233 unsigned int size) 234 { 235 /* dev->gso_ipv4_max_size is read locklessly from sk_setup_caps() */ 236 WRITE_ONCE(dev->gso_ipv4_max_size, size); 237 } 238 239 static inline void netif_set_gro_ipv4_max_size(struct net_device *dev, 240 unsigned int size) 241 { 242 /* This pairs with the READ_ONCE() in skb_gro_receive() */ 243 WRITE_ONCE(dev->gro_ipv4_max_size, size); 244 } 245 246 /** 247 * napi_get_defer_hard_irqs - get the NAPI's defer_hard_irqs 248 * @n: napi struct to get the defer_hard_irqs field from 249 * 250 * Return: the per-NAPI value of the defar_hard_irqs field. 251 */ 252 static inline u32 napi_get_defer_hard_irqs(const struct napi_struct *n) 253 { 254 return READ_ONCE(n->defer_hard_irqs); 255 } 256 257 /** 258 * napi_set_defer_hard_irqs - set the defer_hard_irqs for a napi 259 * @n: napi_struct to set the defer_hard_irqs field 260 * @defer: the value the field should be set to 261 */ 262 static inline void napi_set_defer_hard_irqs(struct napi_struct *n, u32 defer) 263 { 264 WRITE_ONCE(n->defer_hard_irqs, defer); 265 } 266 267 /** 268 * netdev_set_defer_hard_irqs - set defer_hard_irqs for all NAPIs of a netdev 269 * @netdev: the net_device for which all NAPIs will have defer_hard_irqs set 270 * @defer: the defer_hard_irqs value to set 271 */ 272 static inline void netdev_set_defer_hard_irqs(struct net_device *netdev, 273 u32 defer) 274 { 275 unsigned int count = max(netdev->num_rx_queues, 276 netdev->num_tx_queues); 277 struct napi_struct *napi; 278 int i; 279 280 WRITE_ONCE(netdev->napi_defer_hard_irqs, defer); 281 list_for_each_entry(napi, &netdev->napi_list, dev_list) 282 napi_set_defer_hard_irqs(napi, defer); 283 284 for (i = 0; i < count; i++) 285 netdev->napi_config[i].defer_hard_irqs = defer; 286 } 287 288 /** 289 * napi_get_gro_flush_timeout - get the gro_flush_timeout 290 * @n: napi struct to get the gro_flush_timeout from 291 * 292 * Return: the per-NAPI value of the gro_flush_timeout field. 293 */ 294 static inline unsigned long 295 napi_get_gro_flush_timeout(const struct napi_struct *n) 296 { 297 return READ_ONCE(n->gro_flush_timeout); 298 } 299 300 /** 301 * napi_set_gro_flush_timeout - set the gro_flush_timeout for a napi 302 * @n: napi struct to set the gro_flush_timeout 303 * @timeout: timeout value to set 304 * 305 * napi_set_gro_flush_timeout sets the per-NAPI gro_flush_timeout 306 */ 307 static inline void napi_set_gro_flush_timeout(struct napi_struct *n, 308 unsigned long timeout) 309 { 310 WRITE_ONCE(n->gro_flush_timeout, timeout); 311 } 312 313 /** 314 * netdev_set_gro_flush_timeout - set gro_flush_timeout of a netdev's NAPIs 315 * @netdev: the net_device for which all NAPIs will have gro_flush_timeout set 316 * @timeout: the timeout value to set 317 */ 318 static inline void netdev_set_gro_flush_timeout(struct net_device *netdev, 319 unsigned long timeout) 320 { 321 unsigned int count = max(netdev->num_rx_queues, 322 netdev->num_tx_queues); 323 struct napi_struct *napi; 324 int i; 325 326 WRITE_ONCE(netdev->gro_flush_timeout, timeout); 327 list_for_each_entry(napi, &netdev->napi_list, dev_list) 328 napi_set_gro_flush_timeout(napi, timeout); 329 330 for (i = 0; i < count; i++) 331 netdev->napi_config[i].gro_flush_timeout = timeout; 332 } 333 334 /** 335 * napi_get_irq_suspend_timeout - get the irq_suspend_timeout 336 * @n: napi struct to get the irq_suspend_timeout from 337 * 338 * Return: the per-NAPI value of the irq_suspend_timeout field. 339 */ 340 static inline unsigned long 341 napi_get_irq_suspend_timeout(const struct napi_struct *n) 342 { 343 return READ_ONCE(n->irq_suspend_timeout); 344 } 345 346 /** 347 * napi_set_irq_suspend_timeout - set the irq_suspend_timeout for a napi 348 * @n: napi struct to set the irq_suspend_timeout 349 * @timeout: timeout value to set 350 * 351 * napi_set_irq_suspend_timeout sets the per-NAPI irq_suspend_timeout 352 */ 353 static inline void napi_set_irq_suspend_timeout(struct napi_struct *n, 354 unsigned long timeout) 355 { 356 WRITE_ONCE(n->irq_suspend_timeout, timeout); 357 } 358 359 static inline enum netdev_napi_threaded napi_get_threaded(struct napi_struct *n) 360 { 361 if (test_bit(NAPI_STATE_THREADED_BUSY_POLL, &n->state)) 362 return NETDEV_NAPI_THREADED_BUSY_POLL; 363 364 if (test_bit(NAPI_STATE_THREADED, &n->state)) 365 return NETDEV_NAPI_THREADED_ENABLED; 366 367 return NETDEV_NAPI_THREADED_DISABLED; 368 } 369 370 static inline enum netdev_napi_threaded 371 napi_get_threaded_config(struct net_device *dev, struct napi_struct *n) 372 { 373 if (n->config) 374 return n->config->threaded; 375 return dev->threaded; 376 } 377 378 int napi_set_threaded(struct napi_struct *n, 379 enum netdev_napi_threaded threaded); 380 381 int netif_set_threaded(struct net_device *dev, 382 enum netdev_napi_threaded threaded); 383 384 int rps_cpumask_housekeeping(struct cpumask *mask); 385 386 #if defined(CONFIG_DEBUG_NET) && defined(CONFIG_BPF_SYSCALL) 387 void xdp_do_check_flushed(struct napi_struct *napi); 388 #else 389 static inline void xdp_do_check_flushed(struct napi_struct *napi) { } 390 #endif 391 392 /* Best effort check that NAPI is not idle (can't be scheduled to run) */ 393 static inline void napi_assert_will_not_race(const struct napi_struct *napi) 394 { 395 /* uninitialized instance, can't race */ 396 if (!napi->poll_list.next) 397 return; 398 399 /* SCHED bit is set on disabled instances */ 400 WARN_ON(!test_bit(NAPI_STATE_SCHED, &napi->state)); 401 WARN_ON(READ_ONCE(napi->list_owner) != -1); 402 } 403 404 struct skb_defer_node; 405 void skb_defer_node_flush(struct skb_defer_node *sdn); 406 void kick_defer_list_purge(unsigned int cpu); 407 408 int dev_set_hwtstamp_phylib(struct net_device *dev, 409 struct kernel_hwtstamp_config *cfg, 410 struct netlink_ext_ack *extack); 411 int dev_get_hwtstamp_phylib(struct net_device *dev, 412 struct kernel_hwtstamp_config *cfg); 413 int net_hwtstamp_validate(const struct kernel_hwtstamp_config *cfg); 414 415 /* Caller holds RTNL, netdev->lock or RCU */ 416 static inline bool dev_isalive(const struct net_device *dev) 417 { 418 return READ_ONCE(dev->reg_state) <= NETREG_REGISTERED; 419 } 420 421 #endif 422