xref: /linux/include/net/xdp.h (revision 1a9239bb4253f9076b5b4b2a1a4e8d7defd77a95)
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 };
80 
81 struct xdp_buff {
82 	void *data;
83 	void *data_end;
84 	void *data_meta;
85 	void *data_hard_start;
86 	struct xdp_rxq_info *rxq;
87 	struct xdp_txq_info *txq;
88 	u32 frame_sz; /* frame size to deduce data_hard_end/reserved tailroom*/
89 	u32 flags; /* supported values defined in xdp_buff_flags */
90 };
91 
xdp_buff_has_frags(const struct xdp_buff * xdp)92 static __always_inline bool xdp_buff_has_frags(const struct xdp_buff *xdp)
93 {
94 	return !!(xdp->flags & XDP_FLAGS_HAS_FRAGS);
95 }
96 
xdp_buff_set_frags_flag(struct xdp_buff * xdp)97 static __always_inline void xdp_buff_set_frags_flag(struct xdp_buff *xdp)
98 {
99 	xdp->flags |= XDP_FLAGS_HAS_FRAGS;
100 }
101 
xdp_buff_clear_frags_flag(struct xdp_buff * xdp)102 static __always_inline void xdp_buff_clear_frags_flag(struct xdp_buff *xdp)
103 {
104 	xdp->flags &= ~XDP_FLAGS_HAS_FRAGS;
105 }
106 
107 static __always_inline bool
xdp_buff_is_frag_pfmemalloc(const struct xdp_buff * xdp)108 xdp_buff_is_frag_pfmemalloc(const struct xdp_buff *xdp)
109 {
110 	return !!(xdp->flags & XDP_FLAGS_FRAGS_PF_MEMALLOC);
111 }
112 
xdp_buff_set_frag_pfmemalloc(struct xdp_buff * xdp)113 static __always_inline void xdp_buff_set_frag_pfmemalloc(struct xdp_buff *xdp)
114 {
115 	xdp->flags |= XDP_FLAGS_FRAGS_PF_MEMALLOC;
116 }
117 
118 static __always_inline void
xdp_init_buff(struct xdp_buff * xdp,u32 frame_sz,struct xdp_rxq_info * rxq)119 xdp_init_buff(struct xdp_buff *xdp, u32 frame_sz, struct xdp_rxq_info *rxq)
120 {
121 	xdp->frame_sz = frame_sz;
122 	xdp->rxq = rxq;
123 	xdp->flags = 0;
124 }
125 
126 static __always_inline void
xdp_prepare_buff(struct xdp_buff * xdp,unsigned char * hard_start,int headroom,int data_len,const bool meta_valid)127 xdp_prepare_buff(struct xdp_buff *xdp, unsigned char *hard_start,
128 		 int headroom, int data_len, const bool meta_valid)
129 {
130 	unsigned char *data = hard_start + headroom;
131 
132 	xdp->data_hard_start = hard_start;
133 	xdp->data = data;
134 	xdp->data_end = data + data_len;
135 	xdp->data_meta = meta_valid ? data : data + 1;
136 }
137 
138 /* Reserve memory area at end-of data area.
139  *
140  * This macro reserves tailroom in the XDP buffer by limiting the
141  * XDP/BPF data access to data_hard_end.  Notice same area (and size)
142  * is used for XDP_PASS, when constructing the SKB via build_skb().
143  */
144 #define xdp_data_hard_end(xdp)				\
145 	((xdp)->data_hard_start + (xdp)->frame_sz -	\
146 	 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
147 
148 static inline struct skb_shared_info *
xdp_get_shared_info_from_buff(const struct xdp_buff * xdp)149 xdp_get_shared_info_from_buff(const struct xdp_buff *xdp)
150 {
151 	return (struct skb_shared_info *)xdp_data_hard_end(xdp);
152 }
153 
154 static __always_inline unsigned int
xdp_get_buff_len(const struct xdp_buff * xdp)155 xdp_get_buff_len(const struct xdp_buff *xdp)
156 {
157 	unsigned int len = xdp->data_end - xdp->data;
158 	const struct skb_shared_info *sinfo;
159 
160 	if (likely(!xdp_buff_has_frags(xdp)))
161 		goto out;
162 
163 	sinfo = xdp_get_shared_info_from_buff(xdp);
164 	len += sinfo->xdp_frags_size;
165 out:
166 	return len;
167 }
168 
169 void xdp_return_frag(netmem_ref netmem, const struct xdp_buff *xdp);
170 
171 /**
172  * __xdp_buff_add_frag - attach frag to &xdp_buff
173  * @xdp: XDP buffer to attach the frag to
174  * @netmem: network memory containing the frag
175  * @offset: offset at which the frag starts
176  * @size: size of the frag
177  * @truesize: total memory size occupied by the frag
178  * @try_coalesce: whether to try coalescing the frags (not valid for XSk)
179  *
180  * Attach frag to the XDP buffer. If it currently has no frags attached,
181  * initialize the related fields, otherwise check that the frag number
182  * didn't reach the limit of ``MAX_SKB_FRAGS``. If possible, try coalescing
183  * the frag with the previous one.
184  * The function doesn't check/update the pfmemalloc bit. Please use the
185  * non-underscored wrapper in drivers.
186  *
187  * Return: true on success, false if there's no space for the frag in
188  * the shared info struct.
189  */
__xdp_buff_add_frag(struct xdp_buff * xdp,netmem_ref netmem,u32 offset,u32 size,u32 truesize,bool try_coalesce)190 static inline bool __xdp_buff_add_frag(struct xdp_buff *xdp, netmem_ref netmem,
191 				       u32 offset, u32 size, u32 truesize,
192 				       bool try_coalesce)
193 {
194 	struct skb_shared_info *sinfo = xdp_get_shared_info_from_buff(xdp);
195 	skb_frag_t *prev;
196 	u32 nr_frags;
197 
198 	if (!xdp_buff_has_frags(xdp)) {
199 		xdp_buff_set_frags_flag(xdp);
200 
201 		nr_frags = 0;
202 		sinfo->xdp_frags_size = 0;
203 		sinfo->xdp_frags_truesize = 0;
204 
205 		goto fill;
206 	}
207 
208 	nr_frags = sinfo->nr_frags;
209 	prev = &sinfo->frags[nr_frags - 1];
210 
211 	if (try_coalesce && netmem == skb_frag_netmem(prev) &&
212 	    offset == skb_frag_off(prev) + skb_frag_size(prev)) {
213 		skb_frag_size_add(prev, size);
214 		/* Guaranteed to only decrement the refcount */
215 		xdp_return_frag(netmem, xdp);
216 	} else if (unlikely(nr_frags == MAX_SKB_FRAGS)) {
217 		return false;
218 	} else {
219 fill:
220 		__skb_fill_netmem_desc_noacc(sinfo, nr_frags++, netmem,
221 					     offset, size);
222 	}
223 
224 	sinfo->nr_frags = nr_frags;
225 	sinfo->xdp_frags_size += size;
226 	sinfo->xdp_frags_truesize += truesize;
227 
228 	return true;
229 }
230 
231 /**
232  * xdp_buff_add_frag - attach frag to &xdp_buff
233  * @xdp: XDP buffer to attach the frag to
234  * @netmem: network memory containing the frag
235  * @offset: offset at which the frag starts
236  * @size: size of the frag
237  * @truesize: total memory size occupied by the frag
238  *
239  * Version of __xdp_buff_add_frag() which takes care of the pfmemalloc bit.
240  *
241  * Return: true on success, false if there's no space for the frag in
242  * the shared info struct.
243  */
xdp_buff_add_frag(struct xdp_buff * xdp,netmem_ref netmem,u32 offset,u32 size,u32 truesize)244 static inline bool xdp_buff_add_frag(struct xdp_buff *xdp, netmem_ref netmem,
245 				     u32 offset, u32 size, u32 truesize)
246 {
247 	if (!__xdp_buff_add_frag(xdp, netmem, offset, size, truesize, true))
248 		return false;
249 
250 	if (unlikely(netmem_is_pfmemalloc(netmem)))
251 		xdp_buff_set_frag_pfmemalloc(xdp);
252 
253 	return true;
254 }
255 
256 struct xdp_frame {
257 	void *data;
258 	u32 len;
259 	u32 headroom;
260 	u32 metasize; /* uses lower 8-bits */
261 	/* Lifetime of xdp_rxq_info is limited to NAPI/enqueue time,
262 	 * while mem_type is valid on remote CPU.
263 	 */
264 	enum xdp_mem_type mem_type:32;
265 	struct net_device *dev_rx; /* used by cpumap */
266 	u32 frame_sz;
267 	u32 flags; /* supported values defined in xdp_buff_flags */
268 };
269 
xdp_frame_has_frags(const struct xdp_frame * frame)270 static __always_inline bool xdp_frame_has_frags(const struct xdp_frame *frame)
271 {
272 	return !!(frame->flags & XDP_FLAGS_HAS_FRAGS);
273 }
274 
275 static __always_inline bool
xdp_frame_is_frag_pfmemalloc(const struct xdp_frame * frame)276 xdp_frame_is_frag_pfmemalloc(const struct xdp_frame *frame)
277 {
278 	return !!(frame->flags & XDP_FLAGS_FRAGS_PF_MEMALLOC);
279 }
280 
281 #define XDP_BULK_QUEUE_SIZE	16
282 struct xdp_frame_bulk {
283 	int count;
284 	netmem_ref q[XDP_BULK_QUEUE_SIZE];
285 };
286 
xdp_frame_bulk_init(struct xdp_frame_bulk * bq)287 static __always_inline void xdp_frame_bulk_init(struct xdp_frame_bulk *bq)
288 {
289 	bq->count = 0;
290 }
291 
292 static inline struct skb_shared_info *
xdp_get_shared_info_from_frame(const struct xdp_frame * frame)293 xdp_get_shared_info_from_frame(const struct xdp_frame *frame)
294 {
295 	void *data_hard_start = frame->data - frame->headroom - sizeof(*frame);
296 
297 	return (struct skb_shared_info *)(data_hard_start + frame->frame_sz -
298 				SKB_DATA_ALIGN(sizeof(struct skb_shared_info)));
299 }
300 
301 struct xdp_cpumap_stats {
302 	unsigned int redirect;
303 	unsigned int pass;
304 	unsigned int drop;
305 };
306 
307 /* Clear kernel pointers in xdp_frame */
xdp_scrub_frame(struct xdp_frame * frame)308 static inline void xdp_scrub_frame(struct xdp_frame *frame)
309 {
310 	frame->data = NULL;
311 	frame->dev_rx = NULL;
312 }
313 
314 static inline void
xdp_update_skb_shared_info(struct sk_buff * skb,u8 nr_frags,unsigned int size,unsigned int truesize,bool pfmemalloc)315 xdp_update_skb_shared_info(struct sk_buff *skb, u8 nr_frags,
316 			   unsigned int size, unsigned int truesize,
317 			   bool pfmemalloc)
318 {
319 	struct skb_shared_info *sinfo = skb_shinfo(skb);
320 
321 	sinfo->nr_frags = nr_frags;
322 	/*
323 	 * ``destructor_arg`` is unionized with ``xdp_frags_{,true}size``,
324 	 * reset it after that these fields aren't used anymore.
325 	 */
326 	sinfo->destructor_arg = NULL;
327 
328 	skb->len += size;
329 	skb->data_len += size;
330 	skb->truesize += truesize;
331 	skb->pfmemalloc |= pfmemalloc;
332 }
333 
334 /* Avoids inlining WARN macro in fast-path */
335 void xdp_warn(const char *msg, const char *func, const int line);
336 #define XDP_WARN(msg) xdp_warn(msg, __func__, __LINE__)
337 
338 struct sk_buff *xdp_build_skb_from_buff(const struct xdp_buff *xdp);
339 struct sk_buff *xdp_build_skb_from_zc(struct xdp_buff *xdp);
340 struct xdp_frame *xdp_convert_zc_to_xdp_frame(struct xdp_buff *xdp);
341 struct sk_buff *__xdp_build_skb_from_frame(struct xdp_frame *xdpf,
342 					   struct sk_buff *skb,
343 					   struct net_device *dev);
344 struct sk_buff *xdp_build_skb_from_frame(struct xdp_frame *xdpf,
345 					 struct net_device *dev);
346 struct xdp_frame *xdpf_clone(struct xdp_frame *xdpf);
347 
348 static inline
xdp_convert_frame_to_buff(const struct xdp_frame * frame,struct xdp_buff * xdp)349 void xdp_convert_frame_to_buff(const struct xdp_frame *frame,
350 			       struct xdp_buff *xdp)
351 {
352 	xdp->data_hard_start = frame->data - frame->headroom - sizeof(*frame);
353 	xdp->data = frame->data;
354 	xdp->data_end = frame->data + frame->len;
355 	xdp->data_meta = frame->data - frame->metasize;
356 	xdp->frame_sz = frame->frame_sz;
357 	xdp->flags = frame->flags;
358 }
359 
360 static inline
xdp_update_frame_from_buff(const struct xdp_buff * xdp,struct xdp_frame * xdp_frame)361 int xdp_update_frame_from_buff(const struct xdp_buff *xdp,
362 			       struct xdp_frame *xdp_frame)
363 {
364 	int metasize, headroom;
365 
366 	/* Assure headroom is available for storing info */
367 	headroom = xdp->data - xdp->data_hard_start;
368 	metasize = xdp->data - xdp->data_meta;
369 	metasize = metasize > 0 ? metasize : 0;
370 	if (unlikely((headroom - metasize) < sizeof(*xdp_frame)))
371 		return -ENOSPC;
372 
373 	/* Catch if driver didn't reserve tailroom for skb_shared_info */
374 	if (unlikely(xdp->data_end > xdp_data_hard_end(xdp))) {
375 		XDP_WARN("Driver BUG: missing reserved tailroom");
376 		return -ENOSPC;
377 	}
378 
379 	xdp_frame->data = xdp->data;
380 	xdp_frame->len  = xdp->data_end - xdp->data;
381 	xdp_frame->headroom = headroom - sizeof(*xdp_frame);
382 	xdp_frame->metasize = metasize;
383 	xdp_frame->frame_sz = xdp->frame_sz;
384 	xdp_frame->flags = xdp->flags;
385 
386 	return 0;
387 }
388 
389 /* Convert xdp_buff to xdp_frame */
390 static inline
xdp_convert_buff_to_frame(struct xdp_buff * xdp)391 struct xdp_frame *xdp_convert_buff_to_frame(struct xdp_buff *xdp)
392 {
393 	struct xdp_frame *xdp_frame;
394 
395 	if (xdp->rxq->mem.type == MEM_TYPE_XSK_BUFF_POOL)
396 		return xdp_convert_zc_to_xdp_frame(xdp);
397 
398 	/* Store info in top of packet */
399 	xdp_frame = xdp->data_hard_start;
400 	if (unlikely(xdp_update_frame_from_buff(xdp, xdp_frame) < 0))
401 		return NULL;
402 
403 	/* rxq only valid until napi_schedule ends, convert to xdp_mem_type */
404 	xdp_frame->mem_type = xdp->rxq->mem.type;
405 
406 	return xdp_frame;
407 }
408 
409 void __xdp_return(netmem_ref netmem, enum xdp_mem_type mem_type,
410 		  bool napi_direct, struct xdp_buff *xdp);
411 void xdp_return_frame(struct xdp_frame *xdpf);
412 void xdp_return_frame_rx_napi(struct xdp_frame *xdpf);
413 void xdp_return_buff(struct xdp_buff *xdp);
414 void xdp_return_frame_bulk(struct xdp_frame *xdpf,
415 			   struct xdp_frame_bulk *bq);
416 
xdp_flush_frame_bulk(struct xdp_frame_bulk * bq)417 static inline void xdp_flush_frame_bulk(struct xdp_frame_bulk *bq)
418 {
419 	if (unlikely(!bq->count))
420 		return;
421 
422 	page_pool_put_netmem_bulk(bq->q, bq->count);
423 	bq->count = 0;
424 }
425 
426 static __always_inline unsigned int
xdp_get_frame_len(const struct xdp_frame * xdpf)427 xdp_get_frame_len(const struct xdp_frame *xdpf)
428 {
429 	const struct skb_shared_info *sinfo;
430 	unsigned int len = xdpf->len;
431 
432 	if (likely(!xdp_frame_has_frags(xdpf)))
433 		goto out;
434 
435 	sinfo = xdp_get_shared_info_from_frame(xdpf);
436 	len += sinfo->xdp_frags_size;
437 out:
438 	return len;
439 }
440 
441 int __xdp_rxq_info_reg(struct xdp_rxq_info *xdp_rxq,
442 		       struct net_device *dev, u32 queue_index,
443 		       unsigned int napi_id, u32 frag_size);
444 static inline int
xdp_rxq_info_reg(struct xdp_rxq_info * xdp_rxq,struct net_device * dev,u32 queue_index,unsigned int napi_id)445 xdp_rxq_info_reg(struct xdp_rxq_info *xdp_rxq,
446 		 struct net_device *dev, u32 queue_index,
447 		 unsigned int napi_id)
448 {
449 	return __xdp_rxq_info_reg(xdp_rxq, dev, queue_index, napi_id, 0);
450 }
451 
452 void xdp_rxq_info_unreg(struct xdp_rxq_info *xdp_rxq);
453 void xdp_rxq_info_unused(struct xdp_rxq_info *xdp_rxq);
454 bool xdp_rxq_info_is_reg(struct xdp_rxq_info *xdp_rxq);
455 int xdp_rxq_info_reg_mem_model(struct xdp_rxq_info *xdp_rxq,
456 			       enum xdp_mem_type type, void *allocator);
457 void xdp_rxq_info_unreg_mem_model(struct xdp_rxq_info *xdp_rxq);
458 int xdp_reg_mem_model(struct xdp_mem_info *mem,
459 		      enum xdp_mem_type type, void *allocator);
460 void xdp_unreg_mem_model(struct xdp_mem_info *mem);
461 int xdp_reg_page_pool(struct page_pool *pool);
462 void xdp_unreg_page_pool(const struct page_pool *pool);
463 void xdp_rxq_info_attach_page_pool(struct xdp_rxq_info *xdp_rxq,
464 				   const struct page_pool *pool);
465 
466 /**
467  * xdp_rxq_info_attach_mem_model - attach registered mem info to RxQ info
468  * @xdp_rxq: XDP RxQ info to attach the memory info to
469  * @mem: already registered memory info
470  *
471  * If the driver registers its memory providers manually, it must use this
472  * function instead of xdp_rxq_info_reg_mem_model().
473  */
474 static inline void
xdp_rxq_info_attach_mem_model(struct xdp_rxq_info * xdp_rxq,const struct xdp_mem_info * mem)475 xdp_rxq_info_attach_mem_model(struct xdp_rxq_info *xdp_rxq,
476 			      const struct xdp_mem_info *mem)
477 {
478 	xdp_rxq->mem = *mem;
479 }
480 
481 /**
482  * xdp_rxq_info_detach_mem_model - detach registered mem info from RxQ info
483  * @xdp_rxq: XDP RxQ info to detach the memory info from
484  *
485  * If the driver registers its memory providers manually and then attaches it
486  * via xdp_rxq_info_attach_mem_model(), it must call this function before
487  * xdp_rxq_info_unreg().
488  */
xdp_rxq_info_detach_mem_model(struct xdp_rxq_info * xdp_rxq)489 static inline void xdp_rxq_info_detach_mem_model(struct xdp_rxq_info *xdp_rxq)
490 {
491 	xdp_rxq->mem = (struct xdp_mem_info){ };
492 }
493 
494 /* Drivers not supporting XDP metadata can use this helper, which
495  * rejects any room expansion for metadata as a result.
496  */
497 static __always_inline void
xdp_set_data_meta_invalid(struct xdp_buff * xdp)498 xdp_set_data_meta_invalid(struct xdp_buff *xdp)
499 {
500 	xdp->data_meta = xdp->data + 1;
501 }
502 
503 static __always_inline bool
xdp_data_meta_unsupported(const struct xdp_buff * xdp)504 xdp_data_meta_unsupported(const struct xdp_buff *xdp)
505 {
506 	return unlikely(xdp->data_meta > xdp->data);
507 }
508 
xdp_metalen_invalid(unsigned long metalen)509 static inline bool xdp_metalen_invalid(unsigned long metalen)
510 {
511 	unsigned long meta_max;
512 
513 	meta_max = type_max(typeof_member(struct skb_shared_info, meta_len));
514 	BUILD_BUG_ON(!__builtin_constant_p(meta_max));
515 
516 	return !IS_ALIGNED(metalen, sizeof(u32)) || metalen > meta_max;
517 }
518 
519 struct xdp_attachment_info {
520 	struct bpf_prog *prog;
521 	u32 flags;
522 };
523 
524 struct netdev_bpf;
525 void xdp_attachment_setup(struct xdp_attachment_info *info,
526 			  struct netdev_bpf *bpf);
527 
528 #define DEV_MAP_BULK_SIZE XDP_BULK_QUEUE_SIZE
529 
530 /* Define the relationship between xdp-rx-metadata kfunc and
531  * various other entities:
532  * - xdp_rx_metadata enum
533  * - netdev netlink enum (Documentation/netlink/specs/netdev.yaml)
534  * - kfunc name
535  * - xdp_metadata_ops field
536  */
537 #define XDP_METADATA_KFUNC_xxx	\
538 	XDP_METADATA_KFUNC(XDP_METADATA_KFUNC_RX_TIMESTAMP, \
539 			   NETDEV_XDP_RX_METADATA_TIMESTAMP, \
540 			   bpf_xdp_metadata_rx_timestamp, \
541 			   xmo_rx_timestamp) \
542 	XDP_METADATA_KFUNC(XDP_METADATA_KFUNC_RX_HASH, \
543 			   NETDEV_XDP_RX_METADATA_HASH, \
544 			   bpf_xdp_metadata_rx_hash, \
545 			   xmo_rx_hash) \
546 	XDP_METADATA_KFUNC(XDP_METADATA_KFUNC_RX_VLAN_TAG, \
547 			   NETDEV_XDP_RX_METADATA_VLAN_TAG, \
548 			   bpf_xdp_metadata_rx_vlan_tag, \
549 			   xmo_rx_vlan_tag) \
550 
551 enum xdp_rx_metadata {
552 #define XDP_METADATA_KFUNC(name, _, __, ___) name,
553 XDP_METADATA_KFUNC_xxx
554 #undef XDP_METADATA_KFUNC
555 MAX_XDP_METADATA_KFUNC,
556 };
557 
558 enum xdp_rss_hash_type {
559 	/* First part: Individual bits for L3/L4 types */
560 	XDP_RSS_L3_IPV4		= BIT(0),
561 	XDP_RSS_L3_IPV6		= BIT(1),
562 
563 	/* The fixed (L3) IPv4 and IPv6 headers can both be followed by
564 	 * variable/dynamic headers, IPv4 called Options and IPv6 called
565 	 * Extension Headers. HW RSS type can contain this info.
566 	 */
567 	XDP_RSS_L3_DYNHDR	= BIT(2),
568 
569 	/* When RSS hash covers L4 then drivers MUST set XDP_RSS_L4 bit in
570 	 * addition to the protocol specific bit.  This ease interaction with
571 	 * SKBs and avoids reserving a fixed mask for future L4 protocol bits.
572 	 */
573 	XDP_RSS_L4		= BIT(3), /* L4 based hash, proto can be unknown */
574 	XDP_RSS_L4_TCP		= BIT(4),
575 	XDP_RSS_L4_UDP		= BIT(5),
576 	XDP_RSS_L4_SCTP		= BIT(6),
577 	XDP_RSS_L4_IPSEC	= BIT(7), /* L4 based hash include IPSEC SPI */
578 	XDP_RSS_L4_ICMP		= BIT(8),
579 
580 	/* Second part: RSS hash type combinations used for driver HW mapping */
581 	XDP_RSS_TYPE_NONE            = 0,
582 	XDP_RSS_TYPE_L2              = XDP_RSS_TYPE_NONE,
583 
584 	XDP_RSS_TYPE_L3_IPV4         = XDP_RSS_L3_IPV4,
585 	XDP_RSS_TYPE_L3_IPV6         = XDP_RSS_L3_IPV6,
586 	XDP_RSS_TYPE_L3_IPV4_OPT     = XDP_RSS_L3_IPV4 | XDP_RSS_L3_DYNHDR,
587 	XDP_RSS_TYPE_L3_IPV6_EX      = XDP_RSS_L3_IPV6 | XDP_RSS_L3_DYNHDR,
588 
589 	XDP_RSS_TYPE_L4_ANY          = XDP_RSS_L4,
590 	XDP_RSS_TYPE_L4_IPV4_TCP     = XDP_RSS_L3_IPV4 | XDP_RSS_L4 | XDP_RSS_L4_TCP,
591 	XDP_RSS_TYPE_L4_IPV4_UDP     = XDP_RSS_L3_IPV4 | XDP_RSS_L4 | XDP_RSS_L4_UDP,
592 	XDP_RSS_TYPE_L4_IPV4_SCTP    = XDP_RSS_L3_IPV4 | XDP_RSS_L4 | XDP_RSS_L4_SCTP,
593 	XDP_RSS_TYPE_L4_IPV4_IPSEC   = XDP_RSS_L3_IPV4 | XDP_RSS_L4 | XDP_RSS_L4_IPSEC,
594 	XDP_RSS_TYPE_L4_IPV4_ICMP    = XDP_RSS_L3_IPV4 | XDP_RSS_L4 | XDP_RSS_L4_ICMP,
595 
596 	XDP_RSS_TYPE_L4_IPV6_TCP     = XDP_RSS_L3_IPV6 | XDP_RSS_L4 | XDP_RSS_L4_TCP,
597 	XDP_RSS_TYPE_L4_IPV6_UDP     = XDP_RSS_L3_IPV6 | XDP_RSS_L4 | XDP_RSS_L4_UDP,
598 	XDP_RSS_TYPE_L4_IPV6_SCTP    = XDP_RSS_L3_IPV6 | XDP_RSS_L4 | XDP_RSS_L4_SCTP,
599 	XDP_RSS_TYPE_L4_IPV6_IPSEC   = XDP_RSS_L3_IPV6 | XDP_RSS_L4 | XDP_RSS_L4_IPSEC,
600 	XDP_RSS_TYPE_L4_IPV6_ICMP    = XDP_RSS_L3_IPV6 | XDP_RSS_L4 | XDP_RSS_L4_ICMP,
601 
602 	XDP_RSS_TYPE_L4_IPV6_TCP_EX  = XDP_RSS_TYPE_L4_IPV6_TCP  | XDP_RSS_L3_DYNHDR,
603 	XDP_RSS_TYPE_L4_IPV6_UDP_EX  = XDP_RSS_TYPE_L4_IPV6_UDP  | XDP_RSS_L3_DYNHDR,
604 	XDP_RSS_TYPE_L4_IPV6_SCTP_EX = XDP_RSS_TYPE_L4_IPV6_SCTP | XDP_RSS_L3_DYNHDR,
605 };
606 
607 struct xdp_metadata_ops {
608 	int	(*xmo_rx_timestamp)(const struct xdp_md *ctx, u64 *timestamp);
609 	int	(*xmo_rx_hash)(const struct xdp_md *ctx, u32 *hash,
610 			       enum xdp_rss_hash_type *rss_type);
611 	int	(*xmo_rx_vlan_tag)(const struct xdp_md *ctx, __be16 *vlan_proto,
612 				   u16 *vlan_tci);
613 };
614 
615 #ifdef CONFIG_NET
616 u32 bpf_xdp_metadata_kfunc_id(int id);
617 bool bpf_dev_bound_kfunc_id(u32 btf_id);
618 void xdp_set_features_flag(struct net_device *dev, xdp_features_t val);
619 void xdp_features_set_redirect_target(struct net_device *dev, bool support_sg);
620 void xdp_features_clear_redirect_target(struct net_device *dev);
621 #else
bpf_xdp_metadata_kfunc_id(int id)622 static inline u32 bpf_xdp_metadata_kfunc_id(int id) { return 0; }
bpf_dev_bound_kfunc_id(u32 btf_id)623 static inline bool bpf_dev_bound_kfunc_id(u32 btf_id) { return false; }
624 
625 static inline void
xdp_set_features_flag(struct net_device * dev,xdp_features_t val)626 xdp_set_features_flag(struct net_device *dev, xdp_features_t val)
627 {
628 }
629 
630 static inline void
xdp_features_set_redirect_target(struct net_device * dev,bool support_sg)631 xdp_features_set_redirect_target(struct net_device *dev, bool support_sg)
632 {
633 }
634 
635 static inline void
xdp_features_clear_redirect_target(struct net_device * dev)636 xdp_features_clear_redirect_target(struct net_device *dev)
637 {
638 }
639 #endif
640 
xdp_clear_features_flag(struct net_device * dev)641 static inline void xdp_clear_features_flag(struct net_device *dev)
642 {
643 	xdp_set_features_flag(dev, 0);
644 }
645 
bpf_prog_run_xdp(const struct bpf_prog * prog,struct xdp_buff * xdp)646 static __always_inline u32 bpf_prog_run_xdp(const struct bpf_prog *prog,
647 					    struct xdp_buff *xdp)
648 {
649 	/* Driver XDP hooks are invoked within a single NAPI poll cycle and thus
650 	 * under local_bh_disable(), which provides the needed RCU protection
651 	 * for accessing map entries.
652 	 */
653 	u32 act = __bpf_prog_run(prog, xdp, BPF_DISPATCHER_FUNC(xdp));
654 
655 	if (static_branch_unlikely(&bpf_master_redirect_enabled_key)) {
656 		if (act == XDP_TX && netif_is_bond_slave(xdp->rxq->dev))
657 			act = xdp_master_redirect(xdp);
658 	}
659 
660 	return act;
661 }
662 #endif /* __LINUX_NET_XDP_H__ */
663