1 /* SPDX-License-Identifier: GPL-2.0-only */ 2 /* include/net/xdp.h 3 * 4 * Copyright (c) 2017 Jesper Dangaard Brouer, Red Hat Inc. 5 */ 6 #ifndef __LINUX_NET_XDP_H__ 7 #define __LINUX_NET_XDP_H__ 8 9 #include <linux/bitfield.h> 10 #include <linux/filter.h> 11 #include <linux/netdevice.h> 12 #include <linux/skbuff.h> /* skb_shared_info */ 13 14 #include <net/page_pool/types.h> 15 16 /** 17 * DOC: XDP RX-queue information 18 * 19 * The XDP RX-queue info (xdp_rxq_info) is associated with the driver 20 * level RX-ring queues. It is information that is specific to how 21 * the driver has configured a given RX-ring queue. 22 * 23 * Each xdp_buff frame received in the driver carries a (pointer) 24 * reference to this xdp_rxq_info structure. This provides the XDP 25 * data-path read-access to RX-info for both kernel and bpf-side 26 * (limited subset). 27 * 28 * For now, direct access is only safe while running in NAPI/softirq 29 * context. Contents are read-mostly and must not be updated during 30 * driver NAPI/softirq poll. 31 * 32 * The driver usage API is a register and unregister API. 33 * 34 * The struct is not directly tied to the XDP prog. A new XDP prog 35 * can be attached as long as it doesn't change the underlying 36 * RX-ring. If the RX-ring does change significantly, the NIC driver 37 * naturally needs to stop the RX-ring before purging and reallocating 38 * memory. In that process the driver MUST call unregister (which 39 * also applies for driver shutdown and unload). The register API is 40 * also mandatory during RX-ring setup. 41 */ 42 43 enum xdp_mem_type { 44 MEM_TYPE_PAGE_SHARED = 0, /* Split-page refcnt based model */ 45 MEM_TYPE_PAGE_ORDER0, /* Orig XDP full page model */ 46 MEM_TYPE_PAGE_POOL, 47 MEM_TYPE_XSK_BUFF_POOL, 48 MEM_TYPE_MAX, 49 }; 50 51 /* XDP flags for ndo_xdp_xmit */ 52 #define XDP_XMIT_FLUSH (1U << 0) /* doorbell signal consumer */ 53 #define XDP_XMIT_FLAGS_MASK XDP_XMIT_FLUSH 54 55 struct xdp_mem_info { 56 u32 type; /* enum xdp_mem_type, but known size type */ 57 u32 id; 58 }; 59 60 struct page_pool; 61 62 struct xdp_rxq_info { 63 struct net_device *dev; 64 u32 queue_index; 65 u32 reg_state; 66 struct xdp_mem_info mem; 67 u32 frag_size; 68 } ____cacheline_aligned; /* perf critical, avoid false-sharing */ 69 70 struct xdp_txq_info { 71 struct net_device *dev; 72 }; 73 74 enum xdp_buff_flags { 75 XDP_FLAGS_HAS_FRAGS = BIT(0), /* non-linear xdp buff */ 76 XDP_FLAGS_FRAGS_PF_MEMALLOC = BIT(1), /* xdp paged memory is under 77 * pressure 78 */ 79 /* frags have unreadable mem, this can't be true for real XDP packets, 80 * but drivers may use XDP helpers to construct Rx pkt state even when 81 * XDP program is not attached. 82 */ 83 XDP_FLAGS_FRAGS_UNREADABLE = BIT(2), 84 }; 85 86 struct xdp_buff { 87 void *data; 88 void *data_end; 89 void *data_meta; 90 void *data_hard_start; 91 struct xdp_rxq_info *rxq; 92 struct xdp_txq_info *txq; 93 94 union { 95 struct { 96 /* frame size to deduce data_hard_end/tailroom */ 97 u32 frame_sz; 98 /* supported values defined in xdp_buff_flags */ 99 u32 flags; 100 }; 101 102 #ifdef __LITTLE_ENDIAN 103 /* Used to micro-optimize xdp_init_buff(), don't use directly */ 104 u64 frame_sz_flags_init; 105 #endif 106 }; 107 }; 108 109 static __always_inline void xdp_reinit_buff(struct xdp_buff *xdp) 110 { 111 xdp->flags = 0; 112 } 113 114 static __always_inline bool xdp_buff_has_frags(const struct xdp_buff *xdp) 115 { 116 return !!(xdp->flags & XDP_FLAGS_HAS_FRAGS); 117 } 118 119 static __always_inline void xdp_buff_set_frags_flag(struct xdp_buff *xdp) 120 { 121 xdp->flags |= XDP_FLAGS_HAS_FRAGS; 122 } 123 124 static __always_inline void xdp_buff_clear_frags_flag(struct xdp_buff *xdp) 125 { 126 xdp->flags &= ~XDP_FLAGS_HAS_FRAGS; 127 } 128 129 static __always_inline void xdp_buff_set_frag_pfmemalloc(struct xdp_buff *xdp) 130 { 131 xdp->flags |= XDP_FLAGS_FRAGS_PF_MEMALLOC; 132 } 133 134 static __always_inline void xdp_buff_set_frag_unreadable(struct xdp_buff *xdp) 135 { 136 xdp->flags |= XDP_FLAGS_FRAGS_UNREADABLE; 137 } 138 139 static __always_inline u32 xdp_buff_get_skb_flags(const struct xdp_buff *xdp) 140 { 141 return xdp->flags; 142 } 143 144 static __always_inline void xdp_buff_clear_frag_pfmemalloc(struct xdp_buff *xdp) 145 { 146 xdp->flags &= ~XDP_FLAGS_FRAGS_PF_MEMALLOC; 147 } 148 149 static __always_inline void 150 xdp_init_buff(struct xdp_buff *xdp, u32 frame_sz, struct xdp_rxq_info *rxq) 151 { 152 xdp->rxq = rxq; 153 154 #ifdef __LITTLE_ENDIAN 155 /* 156 * Force the compilers to initialize ::flags and assign ::frame_sz with 157 * one write on 64-bit LE architectures as they're often unable to do 158 * it themselves. 159 */ 160 xdp->frame_sz_flags_init = frame_sz; 161 #else 162 xdp->frame_sz = frame_sz; 163 xdp->flags = 0; 164 #endif 165 } 166 167 static __always_inline void 168 xdp_prepare_buff(struct xdp_buff *xdp, unsigned char *hard_start, 169 int headroom, int data_len, const bool meta_valid) 170 { 171 unsigned char *data = hard_start + headroom; 172 173 xdp->data_hard_start = hard_start; 174 xdp->data = data; 175 xdp->data_end = data + data_len; 176 xdp->data_meta = meta_valid ? data : data + 1; 177 } 178 179 /* Reserve memory area at end-of data area. 180 * 181 * This macro reserves tailroom in the XDP buffer by limiting the 182 * XDP/BPF data access to data_hard_end. Notice same area (and size) 183 * is used for XDP_PASS, when constructing the SKB via build_skb(). 184 */ 185 #define xdp_data_hard_end(xdp) \ 186 ((xdp)->data_hard_start + (xdp)->frame_sz - \ 187 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 188 189 static inline struct skb_shared_info * 190 xdp_get_shared_info_from_buff(const struct xdp_buff *xdp) 191 { 192 return (struct skb_shared_info *)xdp_data_hard_end(xdp); 193 } 194 195 static __always_inline unsigned int 196 xdp_get_buff_len(const struct xdp_buff *xdp) 197 { 198 unsigned int len = xdp->data_end - xdp->data; 199 const struct skb_shared_info *sinfo; 200 201 if (likely(!xdp_buff_has_frags(xdp))) 202 goto out; 203 204 sinfo = xdp_get_shared_info_from_buff(xdp); 205 len += sinfo->xdp_frags_size; 206 out: 207 return len; 208 } 209 210 void xdp_return_frag(netmem_ref netmem, const struct xdp_buff *xdp); 211 212 /** 213 * __xdp_buff_add_frag - attach frag to &xdp_buff 214 * @xdp: XDP buffer to attach the frag to 215 * @netmem: network memory containing the frag 216 * @offset: offset at which the frag starts 217 * @size: size of the frag 218 * @truesize: total memory size occupied by the frag 219 * @try_coalesce: whether to try coalescing the frags (not valid for XSk) 220 * 221 * Attach frag to the XDP buffer. If it currently has no frags attached, 222 * initialize the related fields, otherwise check that the frag number 223 * didn't reach the limit of ``MAX_SKB_FRAGS``. If possible, try coalescing 224 * the frag with the previous one. 225 * The function doesn't check/update the pfmemalloc bit. Please use the 226 * non-underscored wrapper in drivers. 227 * 228 * Return: true on success, false if there's no space for the frag in 229 * the shared info struct. 230 */ 231 static inline bool __xdp_buff_add_frag(struct xdp_buff *xdp, netmem_ref netmem, 232 u32 offset, u32 size, u32 truesize, 233 bool try_coalesce) 234 { 235 struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp); 236 skb_frag_t *prev; 237 u32 nr_frags; 238 239 if (!xdp_buff_has_frags(xdp)) { 240 xdp_buff_set_frags_flag(xdp); 241 242 nr_frags = 0; 243 sinfo->xdp_frags_size = 0; 244 sinfo->xdp_frags_truesize = 0; 245 246 goto fill; 247 } 248 249 nr_frags = sinfo->nr_frags; 250 prev = &sinfo->frags[nr_frags - 1]; 251 252 if (try_coalesce && netmem == skb_frag_netmem(prev) && 253 offset == skb_frag_off(prev) + skb_frag_size(prev)) { 254 skb_frag_size_add(prev, size); 255 /* Guaranteed to only decrement the refcount */ 256 xdp_return_frag(netmem, xdp); 257 } else if (unlikely(nr_frags == MAX_SKB_FRAGS)) { 258 return false; 259 } else { 260 fill: 261 __skb_fill_netmem_desc_noacc(sinfo, nr_frags++, netmem, 262 offset, size); 263 } 264 265 sinfo->nr_frags = nr_frags; 266 sinfo->xdp_frags_size += size; 267 sinfo->xdp_frags_truesize += truesize; 268 269 return true; 270 } 271 272 /** 273 * xdp_buff_add_frag - attach frag to &xdp_buff 274 * @xdp: XDP buffer to attach the frag to 275 * @netmem: network memory containing the frag 276 * @offset: offset at which the frag starts 277 * @size: size of the frag 278 * @truesize: total memory size occupied by the frag 279 * 280 * Version of __xdp_buff_add_frag() which takes care of the pfmemalloc bit. 281 * 282 * Return: true on success, false if there's no space for the frag in 283 * the shared info struct. 284 */ 285 static inline bool xdp_buff_add_frag(struct xdp_buff *xdp, netmem_ref netmem, 286 u32 offset, u32 size, u32 truesize) 287 { 288 if (!__xdp_buff_add_frag(xdp, netmem, offset, size, truesize, true)) 289 return false; 290 291 if (unlikely(netmem_is_pfmemalloc(netmem))) 292 xdp_buff_set_frag_pfmemalloc(xdp); 293 if (unlikely(netmem_is_net_iov(netmem))) 294 xdp_buff_set_frag_unreadable(xdp); 295 296 return true; 297 } 298 299 struct xdp_frame { 300 void *data; 301 u32 len; 302 u32 headroom; 303 u32 metasize; /* uses lower 8-bits */ 304 /* Lifetime of xdp_rxq_info is limited to NAPI/enqueue time, 305 * while mem_type is valid on remote CPU. 306 */ 307 enum xdp_mem_type mem_type:32; 308 struct net_device *dev_rx; /* used by cpumap */ 309 u32 frame_sz; 310 u32 flags; /* supported values defined in xdp_buff_flags */ 311 }; 312 313 static __always_inline bool xdp_frame_has_frags(const struct xdp_frame *frame) 314 { 315 return !!(frame->flags & XDP_FLAGS_HAS_FRAGS); 316 } 317 318 static __always_inline u32 319 xdp_frame_get_skb_flags(const struct xdp_frame *frame) 320 { 321 return frame->flags; 322 } 323 324 #define XDP_BULK_QUEUE_SIZE 16 325 struct xdp_frame_bulk { 326 int count; 327 netmem_ref q[XDP_BULK_QUEUE_SIZE]; 328 }; 329 330 static __always_inline void xdp_frame_bulk_init(struct xdp_frame_bulk *bq) 331 { 332 bq->count = 0; 333 } 334 335 static inline struct skb_shared_info * 336 xdp_get_shared_info_from_frame(const struct xdp_frame *frame) 337 { 338 void *data_hard_start = frame->data - frame->headroom - sizeof(*frame); 339 340 return (struct skb_shared_info *)(data_hard_start + frame->frame_sz - 341 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))); 342 } 343 344 struct xdp_cpumap_stats { 345 unsigned int redirect; 346 unsigned int pass; 347 unsigned int drop; 348 }; 349 350 /* Clear kernel pointers in xdp_frame */ 351 static inline void xdp_scrub_frame(struct xdp_frame *frame) 352 { 353 frame->data = NULL; 354 frame->dev_rx = NULL; 355 } 356 357 static inline void 358 xdp_update_skb_frags_info(struct sk_buff *skb, u8 nr_frags, 359 unsigned int size, unsigned int truesize, 360 u32 xdp_flags) 361 { 362 struct skb_shared_info *sinfo = skb_shinfo(skb); 363 364 sinfo->nr_frags = nr_frags; 365 /* 366 * ``destructor_arg`` is unionized with ``xdp_frags_{,true}size``, 367 * reset it after that these fields aren't used anymore. 368 */ 369 sinfo->destructor_arg = NULL; 370 371 skb->len += size; 372 skb->data_len += size; 373 skb->truesize += truesize; 374 skb->pfmemalloc |= !!(xdp_flags & XDP_FLAGS_FRAGS_PF_MEMALLOC); 375 skb->unreadable |= !!(xdp_flags & XDP_FLAGS_FRAGS_UNREADABLE); 376 } 377 378 /* Avoids inlining WARN macro in fast-path */ 379 void xdp_warn(const char *msg, const char *func, const int line); 380 #define XDP_WARN(msg) xdp_warn(msg, __func__, __LINE__) 381 382 struct sk_buff *xdp_build_skb_from_buff(const struct xdp_buff *xdp); 383 struct sk_buff *xdp_build_skb_from_zc(struct xdp_buff *xdp); 384 struct xdp_frame *xdp_convert_zc_to_xdp_frame(struct xdp_buff *xdp); 385 struct sk_buff *__xdp_build_skb_from_frame(struct xdp_frame *xdpf, 386 struct sk_buff *skb, 387 struct net_device *dev); 388 struct sk_buff *xdp_build_skb_from_frame(struct xdp_frame *xdpf, 389 struct net_device *dev); 390 struct xdp_frame *xdpf_clone(struct xdp_frame *xdpf); 391 392 static inline 393 void xdp_convert_frame_to_buff(const struct xdp_frame *frame, 394 struct xdp_buff *xdp) 395 { 396 xdp->data_hard_start = frame->data - frame->headroom - sizeof(*frame); 397 xdp->data = frame->data; 398 xdp->data_end = frame->data + frame->len; 399 xdp->data_meta = frame->data - frame->metasize; 400 xdp->frame_sz = frame->frame_sz; 401 xdp->flags = frame->flags; 402 } 403 404 static inline 405 int xdp_update_frame_from_buff(const struct xdp_buff *xdp, 406 struct xdp_frame *xdp_frame) 407 { 408 int metasize, headroom; 409 410 /* Assure headroom is available for storing info */ 411 headroom = xdp->data - xdp->data_hard_start; 412 metasize = xdp->data - xdp->data_meta; 413 metasize = metasize > 0 ? metasize : 0; 414 if (unlikely((headroom - metasize) < sizeof(*xdp_frame))) 415 return -ENOSPC; 416 417 /* Catch if driver didn't reserve tailroom for skb_shared_info */ 418 if (unlikely(xdp->data_end > xdp_data_hard_end(xdp))) { 419 XDP_WARN("Driver BUG: missing reserved tailroom"); 420 return -ENOSPC; 421 } 422 423 xdp_frame->data = xdp->data; 424 xdp_frame->len = xdp->data_end - xdp->data; 425 xdp_frame->headroom = headroom - sizeof(*xdp_frame); 426 xdp_frame->metasize = metasize; 427 xdp_frame->frame_sz = xdp->frame_sz; 428 xdp_frame->flags = xdp->flags; 429 430 return 0; 431 } 432 433 /* Convert xdp_buff to xdp_frame */ 434 static inline 435 struct xdp_frame *xdp_convert_buff_to_frame(struct xdp_buff *xdp) 436 { 437 struct xdp_frame *xdp_frame; 438 439 if (xdp->rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL) 440 return xdp_convert_zc_to_xdp_frame(xdp); 441 442 /* Store info in top of packet */ 443 xdp_frame = xdp->data_hard_start; 444 if (unlikely(xdp_update_frame_from_buff(xdp, xdp_frame) < 0)) 445 return NULL; 446 447 /* rxq only valid until napi_schedule ends, convert to xdp_mem_type */ 448 xdp_frame->mem_type = xdp->rxq->mem.type; 449 450 return xdp_frame; 451 } 452 453 void __xdp_return(netmem_ref netmem, enum xdp_mem_type mem_type, 454 bool napi_direct, struct xdp_buff *xdp); 455 void xdp_return_frame(struct xdp_frame *xdpf); 456 void xdp_return_frame_rx_napi(struct xdp_frame *xdpf); 457 void xdp_return_buff(struct xdp_buff *xdp); 458 void xdp_return_frame_bulk(struct xdp_frame *xdpf, 459 struct xdp_frame_bulk *bq); 460 461 static inline void xdp_flush_frame_bulk(struct xdp_frame_bulk *bq) 462 { 463 if (unlikely(!bq->count)) 464 return; 465 466 page_pool_put_netmem_bulk(bq->q, bq->count); 467 bq->count = 0; 468 } 469 470 static __always_inline unsigned int 471 xdp_get_frame_len(const struct xdp_frame *xdpf) 472 { 473 const struct skb_shared_info *sinfo; 474 unsigned int len = xdpf->len; 475 476 if (likely(!xdp_frame_has_frags(xdpf))) 477 goto out; 478 479 sinfo = xdp_get_shared_info_from_frame(xdpf); 480 len += sinfo->xdp_frags_size; 481 out: 482 return len; 483 } 484 485 int __xdp_rxq_info_reg(struct xdp_rxq_info *xdp_rxq, 486 struct net_device *dev, u32 queue_index, 487 unsigned int napi_id, u32 frag_size); 488 static inline int 489 xdp_rxq_info_reg(struct xdp_rxq_info *xdp_rxq, 490 struct net_device *dev, u32 queue_index, 491 unsigned int napi_id) 492 { 493 return __xdp_rxq_info_reg(xdp_rxq, dev, queue_index, napi_id, 0); 494 } 495 496 void xdp_rxq_info_unreg(struct xdp_rxq_info *xdp_rxq); 497 void xdp_rxq_info_unused(struct xdp_rxq_info *xdp_rxq); 498 bool xdp_rxq_info_is_reg(struct xdp_rxq_info *xdp_rxq); 499 int xdp_rxq_info_reg_mem_model(struct xdp_rxq_info *xdp_rxq, 500 enum xdp_mem_type type, void *allocator); 501 void xdp_rxq_info_unreg_mem_model(struct xdp_rxq_info *xdp_rxq); 502 int xdp_reg_mem_model(struct xdp_mem_info *mem, 503 enum xdp_mem_type type, void *allocator); 504 void xdp_unreg_mem_model(struct xdp_mem_info *mem); 505 int xdp_reg_page_pool(struct page_pool *pool); 506 void xdp_unreg_page_pool(const struct page_pool *pool); 507 void xdp_rxq_info_attach_page_pool(struct xdp_rxq_info *xdp_rxq, 508 const struct page_pool *pool); 509 510 /** 511 * xdp_rxq_info_attach_mem_model - attach registered mem info to RxQ info 512 * @xdp_rxq: XDP RxQ info to attach the memory info to 513 * @mem: already registered memory info 514 * 515 * If the driver registers its memory providers manually, it must use this 516 * function instead of xdp_rxq_info_reg_mem_model(). 517 */ 518 static inline void 519 xdp_rxq_info_attach_mem_model(struct xdp_rxq_info *xdp_rxq, 520 const struct xdp_mem_info *mem) 521 { 522 xdp_rxq->mem = *mem; 523 } 524 525 /** 526 * xdp_rxq_info_detach_mem_model - detach registered mem info from RxQ info 527 * @xdp_rxq: XDP RxQ info to detach the memory info from 528 * 529 * If the driver registers its memory providers manually and then attaches it 530 * via xdp_rxq_info_attach_mem_model(), it must call this function before 531 * xdp_rxq_info_unreg(). 532 */ 533 static inline void xdp_rxq_info_detach_mem_model(struct xdp_rxq_info *xdp_rxq) 534 { 535 xdp_rxq->mem = (struct xdp_mem_info){ }; 536 } 537 538 /* Drivers not supporting XDP metadata can use this helper, which 539 * rejects any room expansion for metadata as a result. 540 */ 541 static __always_inline void 542 xdp_set_data_meta_invalid(struct xdp_buff *xdp) 543 { 544 xdp->data_meta = xdp->data + 1; 545 } 546 547 static __always_inline bool 548 xdp_data_meta_unsupported(const struct xdp_buff *xdp) 549 { 550 return unlikely(xdp->data_meta > xdp->data); 551 } 552 553 static inline bool xdp_metalen_invalid(unsigned long metalen) 554 { 555 unsigned long meta_max; 556 557 meta_max = type_max(typeof_member(struct skb_shared_info, meta_len)); 558 BUILD_BUG_ON(!__builtin_constant_p(meta_max)); 559 560 return !IS_ALIGNED(metalen, sizeof(u32)) || metalen > meta_max; 561 } 562 563 struct xdp_attachment_info { 564 struct bpf_prog *prog; 565 u32 flags; 566 }; 567 568 struct netdev_bpf; 569 void xdp_attachment_setup(struct xdp_attachment_info *info, 570 struct netdev_bpf *bpf); 571 572 #define DEV_MAP_BULK_SIZE XDP_BULK_QUEUE_SIZE 573 574 /* Define the relationship between xdp-rx-metadata kfunc and 575 * various other entities: 576 * - xdp_rx_metadata enum 577 * - netdev netlink enum (Documentation/netlink/specs/netdev.yaml) 578 * - kfunc name 579 * - xdp_metadata_ops field 580 */ 581 #define XDP_METADATA_KFUNC_xxx \ 582 XDP_METADATA_KFUNC(XDP_METADATA_KFUNC_RX_TIMESTAMP, \ 583 NETDEV_XDP_RX_METADATA_TIMESTAMP, \ 584 bpf_xdp_metadata_rx_timestamp, \ 585 xmo_rx_timestamp) \ 586 XDP_METADATA_KFUNC(XDP_METADATA_KFUNC_RX_HASH, \ 587 NETDEV_XDP_RX_METADATA_HASH, \ 588 bpf_xdp_metadata_rx_hash, \ 589 xmo_rx_hash) \ 590 XDP_METADATA_KFUNC(XDP_METADATA_KFUNC_RX_VLAN_TAG, \ 591 NETDEV_XDP_RX_METADATA_VLAN_TAG, \ 592 bpf_xdp_metadata_rx_vlan_tag, \ 593 xmo_rx_vlan_tag) \ 594 595 enum xdp_rx_metadata { 596 #define XDP_METADATA_KFUNC(name, _, __, ___) name, 597 XDP_METADATA_KFUNC_xxx 598 #undef XDP_METADATA_KFUNC 599 MAX_XDP_METADATA_KFUNC, 600 }; 601 602 enum xdp_rss_hash_type { 603 /* First part: Individual bits for L3/L4 types */ 604 XDP_RSS_L3_IPV4 = BIT(0), 605 XDP_RSS_L3_IPV6 = BIT(1), 606 607 /* The fixed (L3) IPv4 and IPv6 headers can both be followed by 608 * variable/dynamic headers, IPv4 called Options and IPv6 called 609 * Extension Headers. HW RSS type can contain this info. 610 */ 611 XDP_RSS_L3_DYNHDR = BIT(2), 612 613 /* When RSS hash covers L4 then drivers MUST set XDP_RSS_L4 bit in 614 * addition to the protocol specific bit. This ease interaction with 615 * SKBs and avoids reserving a fixed mask for future L4 protocol bits. 616 */ 617 XDP_RSS_L4 = BIT(3), /* L4 based hash, proto can be unknown */ 618 XDP_RSS_L4_TCP = BIT(4), 619 XDP_RSS_L4_UDP = BIT(5), 620 XDP_RSS_L4_SCTP = BIT(6), 621 XDP_RSS_L4_IPSEC = BIT(7), /* L4 based hash include IPSEC SPI */ 622 XDP_RSS_L4_ICMP = BIT(8), 623 624 /* Second part: RSS hash type combinations used for driver HW mapping */ 625 XDP_RSS_TYPE_NONE = 0, 626 XDP_RSS_TYPE_L2 = XDP_RSS_TYPE_NONE, 627 628 XDP_RSS_TYPE_L3_IPV4 = XDP_RSS_L3_IPV4, 629 XDP_RSS_TYPE_L3_IPV6 = XDP_RSS_L3_IPV6, 630 XDP_RSS_TYPE_L3_IPV4_OPT = XDP_RSS_L3_IPV4 | XDP_RSS_L3_DYNHDR, 631 XDP_RSS_TYPE_L3_IPV6_EX = XDP_RSS_L3_IPV6 | XDP_RSS_L3_DYNHDR, 632 633 XDP_RSS_TYPE_L4_ANY = XDP_RSS_L4, 634 XDP_RSS_TYPE_L4_IPV4_TCP = XDP_RSS_L3_IPV4 | XDP_RSS_L4 | XDP_RSS_L4_TCP, 635 XDP_RSS_TYPE_L4_IPV4_UDP = XDP_RSS_L3_IPV4 | XDP_RSS_L4 | XDP_RSS_L4_UDP, 636 XDP_RSS_TYPE_L4_IPV4_SCTP = XDP_RSS_L3_IPV4 | XDP_RSS_L4 | XDP_RSS_L4_SCTP, 637 XDP_RSS_TYPE_L4_IPV4_IPSEC = XDP_RSS_L3_IPV4 | XDP_RSS_L4 | XDP_RSS_L4_IPSEC, 638 XDP_RSS_TYPE_L4_IPV4_ICMP = XDP_RSS_L3_IPV4 | XDP_RSS_L4 | XDP_RSS_L4_ICMP, 639 640 XDP_RSS_TYPE_L4_IPV6_TCP = XDP_RSS_L3_IPV6 | XDP_RSS_L4 | XDP_RSS_L4_TCP, 641 XDP_RSS_TYPE_L4_IPV6_UDP = XDP_RSS_L3_IPV6 | XDP_RSS_L4 | XDP_RSS_L4_UDP, 642 XDP_RSS_TYPE_L4_IPV6_SCTP = XDP_RSS_L3_IPV6 | XDP_RSS_L4 | XDP_RSS_L4_SCTP, 643 XDP_RSS_TYPE_L4_IPV6_IPSEC = XDP_RSS_L3_IPV6 | XDP_RSS_L4 | XDP_RSS_L4_IPSEC, 644 XDP_RSS_TYPE_L4_IPV6_ICMP = XDP_RSS_L3_IPV6 | XDP_RSS_L4 | XDP_RSS_L4_ICMP, 645 646 XDP_RSS_TYPE_L4_IPV6_TCP_EX = XDP_RSS_TYPE_L4_IPV6_TCP | XDP_RSS_L3_DYNHDR, 647 XDP_RSS_TYPE_L4_IPV6_UDP_EX = XDP_RSS_TYPE_L4_IPV6_UDP | XDP_RSS_L3_DYNHDR, 648 XDP_RSS_TYPE_L4_IPV6_SCTP_EX = XDP_RSS_TYPE_L4_IPV6_SCTP | XDP_RSS_L3_DYNHDR, 649 }; 650 651 struct xdp_metadata_ops { 652 int (*xmo_rx_timestamp)(const struct xdp_md *ctx, u64 *timestamp); 653 int (*xmo_rx_hash)(const struct xdp_md *ctx, u32 *hash, 654 enum xdp_rss_hash_type *rss_type); 655 int (*xmo_rx_vlan_tag)(const struct xdp_md *ctx, __be16 *vlan_proto, 656 u16 *vlan_tci); 657 }; 658 659 #ifdef CONFIG_NET 660 u32 bpf_xdp_metadata_kfunc_id(int id); 661 bool bpf_dev_bound_kfunc_id(u32 btf_id); 662 void xdp_set_features_flag(struct net_device *dev, xdp_features_t val); 663 void xdp_set_features_flag_locked(struct net_device *dev, xdp_features_t val); 664 void xdp_features_set_redirect_target(struct net_device *dev, bool support_sg); 665 void xdp_features_set_redirect_target_locked(struct net_device *dev, 666 bool support_sg); 667 void xdp_features_clear_redirect_target(struct net_device *dev); 668 void xdp_features_clear_redirect_target_locked(struct net_device *dev); 669 #else 670 static inline u32 bpf_xdp_metadata_kfunc_id(int id) { return 0; } 671 static inline bool bpf_dev_bound_kfunc_id(u32 btf_id) { return false; } 672 673 static inline void 674 xdp_set_features_flag(struct net_device *dev, xdp_features_t val) 675 { 676 } 677 678 static inline void 679 xdp_features_set_redirect_target(struct net_device *dev, bool support_sg) 680 { 681 } 682 683 static inline void 684 xdp_features_clear_redirect_target(struct net_device *dev) 685 { 686 } 687 #endif 688 689 static inline void xdp_clear_features_flag(struct net_device *dev) 690 { 691 xdp_set_features_flag(dev, 0); 692 } 693 694 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog, 695 struct xdp_buff *xdp) 696 { 697 /* Driver XDP hooks are invoked within a single NAPI poll cycle and thus 698 * under local_bh_disable(), which provides the needed RCU protection 699 * for accessing map entries. 700 */ 701 u32 act = __bpf_prog_run(prog, xdp, BPF_DISPATCHER_FUNC(xdp)); 702 703 if (static_branch_unlikely(&bpf_master_redirect_enabled_key)) { 704 if (act == XDP_TX && netif_is_bond_slave(xdp->rxq->dev)) 705 act = xdp_master_redirect(xdp); 706 } 707 708 return act; 709 } 710 #endif /* __LINUX_NET_XDP_H__ */ 711