1 // SPDX-License-Identifier: GPL-2.0-only
2 /* net/core/xdp.c
3 *
4 * Copyright (c) 2017 Jesper Dangaard Brouer, Red Hat Inc.
5 */
6 #include <linux/bpf.h>
7 #include <linux/btf.h>
8 #include <linux/btf_ids.h>
9 #include <linux/filter.h>
10 #include <linux/types.h>
11 #include <linux/mm.h>
12 #include <linux/netdevice.h>
13 #include <linux/slab.h>
14 #include <linux/idr.h>
15 #include <linux/rhashtable.h>
16 #include <linux/bug.h>
17 #include <net/page_pool/helpers.h>
18
19 #include <net/hotdata.h>
20 #include <net/xdp.h>
21 #include <net/xdp_priv.h> /* struct xdp_mem_allocator */
22 #include <trace/events/xdp.h>
23 #include <net/xdp_sock_drv.h>
24
25 #define REG_STATE_NEW 0x0
26 #define REG_STATE_REGISTERED 0x1
27 #define REG_STATE_UNREGISTERED 0x2
28 #define REG_STATE_UNUSED 0x3
29
30 static DEFINE_IDA(mem_id_pool);
31 static DEFINE_MUTEX(mem_id_lock);
32 #define MEM_ID_MAX 0xFFFE
33 #define MEM_ID_MIN 1
34 static int mem_id_next = MEM_ID_MIN;
35
36 static bool mem_id_init; /* false */
37 static struct rhashtable *mem_id_ht;
38
39 static u32 xdp_mem_id_hashfn(const void *data, u32 len, u32 seed)
40 {
41 const u32 *k = data;
42 const u32 key = *k;
43
44 BUILD_BUG_ON(sizeof_field(struct xdp_mem_allocator, mem.id)
45 != sizeof(u32));
46
47 /* Use cyclic increasing ID as direct hash key */
48 return key;
49 }
50
51 static int xdp_mem_id_cmp(struct rhashtable_compare_arg *arg,
52 const void *ptr)
53 {
54 const struct xdp_mem_allocator *xa = ptr;
55 u32 mem_id = *(u32 *)arg->key;
56
57 return xa->mem.id != mem_id;
58 }
59
60 static const struct rhashtable_params mem_id_rht_params = {
61 .nelem_hint = 64,
62 .head_offset = offsetof(struct xdp_mem_allocator, node),
63 .key_offset = offsetof(struct xdp_mem_allocator, mem.id),
64 .key_len = sizeof_field(struct xdp_mem_allocator, mem.id),
65 .max_size = MEM_ID_MAX,
66 .min_size = 8,
67 .automatic_shrinking = true,
68 .hashfn = xdp_mem_id_hashfn,
69 .obj_cmpfn = xdp_mem_id_cmp,
70 };
71
72 static void __xdp_mem_allocator_rcu_free(struct rcu_head *rcu)
73 {
74 struct xdp_mem_allocator *xa;
75
76 xa = container_of(rcu, struct xdp_mem_allocator, rcu);
77
78 /* Allow this ID to be reused */
79 ida_free(&mem_id_pool, xa->mem.id);
80
81 kfree(xa);
82 }
83
84 static void mem_xa_remove(struct xdp_mem_allocator *xa)
85 {
86 trace_mem_disconnect(xa);
87
88 if (!rhashtable_remove_fast(mem_id_ht, &xa->node, mem_id_rht_params))
89 call_rcu(&xa->rcu, __xdp_mem_allocator_rcu_free);
90 }
91
92 static void mem_allocator_disconnect(void *allocator)
93 {
94 struct xdp_mem_allocator *xa;
95 struct rhashtable_iter iter;
96
97 mutex_lock(&mem_id_lock);
98
99 rhashtable_walk_enter(mem_id_ht, &iter);
100 do {
101 rhashtable_walk_start(&iter);
102
103 while ((xa = rhashtable_walk_next(&iter)) && !IS_ERR(xa)) {
104 if (xa->allocator == allocator)
105 mem_xa_remove(xa);
106 }
107
108 rhashtable_walk_stop(&iter);
109
110 } while (xa == ERR_PTR(-EAGAIN));
111 rhashtable_walk_exit(&iter);
112
113 mutex_unlock(&mem_id_lock);
114 }
115
116 void xdp_unreg_mem_model(struct xdp_mem_info *mem)
117 {
118 struct xdp_mem_allocator *xa;
119 int type = mem->type;
120 int id = mem->id;
121
122 /* Reset mem info to defaults */
123 mem->id = 0;
124 mem->type = 0;
125
126 if (id == 0)
127 return;
128
129 if (type == MEM_TYPE_PAGE_POOL) {
130 xa = rhashtable_lookup_fast(mem_id_ht, &id, mem_id_rht_params);
131 page_pool_destroy(xa->page_pool);
132 }
133 }
134 EXPORT_SYMBOL_GPL(xdp_unreg_mem_model);
135
136 void xdp_rxq_info_unreg_mem_model(struct xdp_rxq_info *xdp_rxq)
137 {
138 if (xdp_rxq->reg_state != REG_STATE_REGISTERED) {
139 WARN(1, "Missing register, driver bug");
140 return;
141 }
142
143 xdp_unreg_mem_model(&xdp_rxq->mem);
144 }
145 EXPORT_SYMBOL_GPL(xdp_rxq_info_unreg_mem_model);
146
147 void xdp_rxq_info_unreg(struct xdp_rxq_info *xdp_rxq)
148 {
149 /* Simplify driver cleanup code paths, allow unreg "unused" */
150 if (xdp_rxq->reg_state == REG_STATE_UNUSED)
151 return;
152
153 xdp_rxq_info_unreg_mem_model(xdp_rxq);
154
155 xdp_rxq->reg_state = REG_STATE_UNREGISTERED;
156 xdp_rxq->dev = NULL;
157 }
158 EXPORT_SYMBOL_GPL(xdp_rxq_info_unreg);
159
160 static void xdp_rxq_info_init(struct xdp_rxq_info *xdp_rxq)
161 {
162 memset(xdp_rxq, 0, sizeof(*xdp_rxq));
163 }
164
165 /* Returns 0 on success, negative on failure */
166 int __xdp_rxq_info_reg(struct xdp_rxq_info *xdp_rxq,
167 struct net_device *dev, u32 queue_index,
168 unsigned int napi_id, u32 frag_size)
169 {
170 if (!dev) {
171 WARN(1, "Missing net_device from driver");
172 return -ENODEV;
173 }
174
175 if (xdp_rxq->reg_state == REG_STATE_UNUSED) {
176 WARN(1, "Driver promised not to register this");
177 return -EINVAL;
178 }
179
180 if (xdp_rxq->reg_state == REG_STATE_REGISTERED) {
181 WARN(1, "Missing unregister, handled but fix driver");
182 xdp_rxq_info_unreg(xdp_rxq);
183 }
184
185 /* State either UNREGISTERED or NEW */
186 xdp_rxq_info_init(xdp_rxq);
187 xdp_rxq->dev = dev;
188 xdp_rxq->queue_index = queue_index;
189 xdp_rxq->napi_id = napi_id;
190 xdp_rxq->frag_size = frag_size;
191
192 xdp_rxq->reg_state = REG_STATE_REGISTERED;
193 return 0;
194 }
195 EXPORT_SYMBOL_GPL(__xdp_rxq_info_reg);
196
197 void xdp_rxq_info_unused(struct xdp_rxq_info *xdp_rxq)
198 {
199 xdp_rxq->reg_state = REG_STATE_UNUSED;
200 }
201 EXPORT_SYMBOL_GPL(xdp_rxq_info_unused);
202
203 bool xdp_rxq_info_is_reg(struct xdp_rxq_info *xdp_rxq)
204 {
205 return (xdp_rxq->reg_state == REG_STATE_REGISTERED);
206 }
207 EXPORT_SYMBOL_GPL(xdp_rxq_info_is_reg);
208
209 static int __mem_id_init_hash_table(void)
210 {
211 struct rhashtable *rht;
212 int ret;
213
214 if (unlikely(mem_id_init))
215 return 0;
216
217 rht = kzalloc(sizeof(*rht), GFP_KERNEL);
218 if (!rht)
219 return -ENOMEM;
220
221 ret = rhashtable_init(rht, &mem_id_rht_params);
222 if (ret < 0) {
223 kfree(rht);
224 return ret;
225 }
226 mem_id_ht = rht;
227 smp_mb(); /* mutex lock should provide enough pairing */
228 mem_id_init = true;
229
230 return 0;
231 }
232
233 /* Allocate a cyclic ID that maps to allocator pointer.
234 * See: https://www.kernel.org/doc/html/latest/core-api/idr.html
235 *
236 * Caller must lock mem_id_lock.
237 */
238 static int __mem_id_cyclic_get(gfp_t gfp)
239 {
240 int retries = 1;
241 int id;
242
243 again:
244 id = ida_alloc_range(&mem_id_pool, mem_id_next, MEM_ID_MAX - 1, gfp);
245 if (id < 0) {
246 if (id == -ENOSPC) {
247 /* Cyclic allocator, reset next id */
248 if (retries--) {
249 mem_id_next = MEM_ID_MIN;
250 goto again;
251 }
252 }
253 return id; /* errno */
254 }
255 mem_id_next = id + 1;
256
257 return id;
258 }
259
260 static bool __is_supported_mem_type(enum xdp_mem_type type)
261 {
262 if (type == MEM_TYPE_PAGE_POOL)
263 return is_page_pool_compiled_in();
264
265 if (type >= MEM_TYPE_MAX)
266 return false;
267
268 return true;
269 }
270
271 static struct xdp_mem_allocator *__xdp_reg_mem_model(struct xdp_mem_info *mem,
272 enum xdp_mem_type type,
273 void *allocator)
274 {
275 struct xdp_mem_allocator *xdp_alloc;
276 gfp_t gfp = GFP_KERNEL;
277 int id, errno, ret;
278 void *ptr;
279
280 if (!__is_supported_mem_type(type))
281 return ERR_PTR(-EOPNOTSUPP);
282
283 mem->type = type;
284
285 if (!allocator) {
286 if (type == MEM_TYPE_PAGE_POOL)
287 return ERR_PTR(-EINVAL); /* Setup time check page_pool req */
288 return NULL;
289 }
290
291 /* Delay init of rhashtable to save memory if feature isn't used */
292 if (!mem_id_init) {
293 mutex_lock(&mem_id_lock);
294 ret = __mem_id_init_hash_table();
295 mutex_unlock(&mem_id_lock);
296 if (ret < 0)
297 return ERR_PTR(ret);
298 }
299
300 xdp_alloc = kzalloc(sizeof(*xdp_alloc), gfp);
301 if (!xdp_alloc)
302 return ERR_PTR(-ENOMEM);
303
304 mutex_lock(&mem_id_lock);
305 id = __mem_id_cyclic_get(gfp);
306 if (id < 0) {
307 errno = id;
308 goto err;
309 }
310 mem->id = id;
311 xdp_alloc->mem = *mem;
312 xdp_alloc->allocator = allocator;
313
314 /* Insert allocator into ID lookup table */
315 ptr = rhashtable_insert_slow(mem_id_ht, &id, &xdp_alloc->node);
316 if (IS_ERR(ptr)) {
317 ida_free(&mem_id_pool, mem->id);
318 mem->id = 0;
319 errno = PTR_ERR(ptr);
320 goto err;
321 }
322
323 if (type == MEM_TYPE_PAGE_POOL)
324 page_pool_use_xdp_mem(allocator, mem_allocator_disconnect, mem);
325
326 mutex_unlock(&mem_id_lock);
327
328 return xdp_alloc;
329 err:
330 mutex_unlock(&mem_id_lock);
331 kfree(xdp_alloc);
332 return ERR_PTR(errno);
333 }
334
335 int xdp_reg_mem_model(struct xdp_mem_info *mem,
336 enum xdp_mem_type type, void *allocator)
337 {
338 struct xdp_mem_allocator *xdp_alloc;
339
340 xdp_alloc = __xdp_reg_mem_model(mem, type, allocator);
341 if (IS_ERR(xdp_alloc))
342 return PTR_ERR(xdp_alloc);
343 return 0;
344 }
345 EXPORT_SYMBOL_GPL(xdp_reg_mem_model);
346
347 int xdp_rxq_info_reg_mem_model(struct xdp_rxq_info *xdp_rxq,
348 enum xdp_mem_type type, void *allocator)
349 {
350 struct xdp_mem_allocator *xdp_alloc;
351
352 if (xdp_rxq->reg_state != REG_STATE_REGISTERED) {
353 WARN(1, "Missing register, driver bug");
354 return -EFAULT;
355 }
356
357 xdp_alloc = __xdp_reg_mem_model(&xdp_rxq->mem, type, allocator);
358 if (IS_ERR(xdp_alloc))
359 return PTR_ERR(xdp_alloc);
360
361 if (trace_mem_connect_enabled() && xdp_alloc)
362 trace_mem_connect(xdp_alloc, xdp_rxq);
363 return 0;
364 }
365
366 EXPORT_SYMBOL_GPL(xdp_rxq_info_reg_mem_model);
367
368 /* XDP RX runs under NAPI protection, and in different delivery error
369 * scenarios (e.g. queue full), it is possible to return the xdp_frame
370 * while still leveraging this protection. The @napi_direct boolean
371 * is used for those calls sites. Thus, allowing for faster recycling
372 * of xdp_frames/pages in those cases.
373 */
374 void __xdp_return(void *data, struct xdp_mem_info *mem, bool napi_direct,
375 struct xdp_buff *xdp)
376 {
377 struct page *page;
378
379 switch (mem->type) {
380 case MEM_TYPE_PAGE_POOL:
381 page = virt_to_head_page(data);
382 if (napi_direct && xdp_return_frame_no_direct())
383 napi_direct = false;
384 /* No need to check ((page->pp_magic & ~0x3UL) == PP_SIGNATURE)
385 * as mem->type knows this a page_pool page
386 */
387 page_pool_put_full_page(page->pp, page, napi_direct);
388 break;
389 case MEM_TYPE_PAGE_SHARED:
390 page_frag_free(data);
391 break;
392 case MEM_TYPE_PAGE_ORDER0:
393 page = virt_to_page(data); /* Assumes order0 page*/
394 put_page(page);
395 break;
396 case MEM_TYPE_XSK_BUFF_POOL:
397 /* NB! Only valid from an xdp_buff! */
398 xsk_buff_free(xdp);
399 break;
400 default:
401 /* Not possible, checked in xdp_rxq_info_reg_mem_model() */
402 WARN(1, "Incorrect XDP memory type (%d) usage", mem->type);
403 break;
404 }
405 }
406
407 void xdp_return_frame(struct xdp_frame *xdpf)
408 {
409 struct skb_shared_info *sinfo;
410 int i;
411
412 if (likely(!xdp_frame_has_frags(xdpf)))
413 goto out;
414
415 sinfo = xdp_get_shared_info_from_frame(xdpf);
416 for (i = 0; i < sinfo->nr_frags; i++) {
417 struct page *page = skb_frag_page(&sinfo->frags[i]);
418
419 __xdp_return(page_address(page), &xdpf->mem, false, NULL);
420 }
421 out:
422 __xdp_return(xdpf->data, &xdpf->mem, false, NULL);
423 }
424 EXPORT_SYMBOL_GPL(xdp_return_frame);
425
426 void xdp_return_frame_rx_napi(struct xdp_frame *xdpf)
427 {
428 struct skb_shared_info *sinfo;
429 int i;
430
431 if (likely(!xdp_frame_has_frags(xdpf)))
432 goto out;
433
434 sinfo = xdp_get_shared_info_from_frame(xdpf);
435 for (i = 0; i < sinfo->nr_frags; i++) {
436 struct page *page = skb_frag_page(&sinfo->frags[i]);
437
438 __xdp_return(page_address(page), &xdpf->mem, true, NULL);
439 }
440 out:
441 __xdp_return(xdpf->data, &xdpf->mem, true, NULL);
442 }
443 EXPORT_SYMBOL_GPL(xdp_return_frame_rx_napi);
444
445 /* XDP bulk APIs introduce a defer/flush mechanism to return
446 * pages belonging to the same xdp_mem_allocator object
447 * (identified via the mem.id field) in bulk to optimize
448 * I-cache and D-cache.
449 * The bulk queue size is set to 16 to be aligned to how
450 * XDP_REDIRECT bulking works. The bulk is flushed when
451 * it is full or when mem.id changes.
452 * xdp_frame_bulk is usually stored/allocated on the function
453 * call-stack to avoid locking penalties.
454 */
455 void xdp_flush_frame_bulk(struct xdp_frame_bulk *bq)
456 {
457 struct xdp_mem_allocator *xa = bq->xa;
458
459 if (unlikely(!xa || !bq->count))
460 return;
461
462 page_pool_put_page_bulk(xa->page_pool, bq->q, bq->count);
463 /* bq->xa is not cleared to save lookup, if mem.id same in next bulk */
464 bq->count = 0;
465 }
466 EXPORT_SYMBOL_GPL(xdp_flush_frame_bulk);
467
468 /* Must be called with rcu_read_lock held */
469 void xdp_return_frame_bulk(struct xdp_frame *xdpf,
470 struct xdp_frame_bulk *bq)
471 {
472 struct xdp_mem_info *mem = &xdpf->mem;
473 struct xdp_mem_allocator *xa;
474
475 if (mem->type != MEM_TYPE_PAGE_POOL) {
476 xdp_return_frame(xdpf);
477 return;
478 }
479
480 xa = bq->xa;
481 if (unlikely(!xa)) {
482 xa = rhashtable_lookup(mem_id_ht, &mem->id, mem_id_rht_params);
483 bq->count = 0;
484 bq->xa = xa;
485 }
486
487 if (bq->count == XDP_BULK_QUEUE_SIZE)
488 xdp_flush_frame_bulk(bq);
489
490 if (unlikely(mem->id != xa->mem.id)) {
491 xdp_flush_frame_bulk(bq);
492 bq->xa = rhashtable_lookup(mem_id_ht, &mem->id, mem_id_rht_params);
493 }
494
495 if (unlikely(xdp_frame_has_frags(xdpf))) {
496 struct skb_shared_info *sinfo;
497 int i;
498
499 sinfo = xdp_get_shared_info_from_frame(xdpf);
500 for (i = 0; i < sinfo->nr_frags; i++) {
501 skb_frag_t *frag = &sinfo->frags[i];
502
503 bq->q[bq->count++] = skb_frag_address(frag);
504 if (bq->count == XDP_BULK_QUEUE_SIZE)
505 xdp_flush_frame_bulk(bq);
506 }
507 }
508 bq->q[bq->count++] = xdpf->data;
509 }
510 EXPORT_SYMBOL_GPL(xdp_return_frame_bulk);
511
512 void xdp_return_buff(struct xdp_buff *xdp)
513 {
514 struct skb_shared_info *sinfo;
515 int i;
516
517 if (likely(!xdp_buff_has_frags(xdp)))
518 goto out;
519
520 sinfo = xdp_get_shared_info_from_buff(xdp);
521 for (i = 0; i < sinfo->nr_frags; i++) {
522 struct page *page = skb_frag_page(&sinfo->frags[i]);
523
524 __xdp_return(page_address(page), &xdp->rxq->mem, true, xdp);
525 }
526 out:
527 __xdp_return(xdp->data, &xdp->rxq->mem, true, xdp);
528 }
529 EXPORT_SYMBOL_GPL(xdp_return_buff);
530
531 void xdp_attachment_setup(struct xdp_attachment_info *info,
532 struct netdev_bpf *bpf)
533 {
534 if (info->prog)
535 bpf_prog_put(info->prog);
536 info->prog = bpf->prog;
537 info->flags = bpf->flags;
538 }
539 EXPORT_SYMBOL_GPL(xdp_attachment_setup);
540
541 struct xdp_frame *xdp_convert_zc_to_xdp_frame(struct xdp_buff *xdp)
542 {
543 unsigned int metasize, totsize;
544 void *addr, *data_to_copy;
545 struct xdp_frame *xdpf;
546 struct page *page;
547
548 /* Clone into a MEM_TYPE_PAGE_ORDER0 xdp_frame. */
549 metasize = xdp_data_meta_unsupported(xdp) ? 0 :
550 xdp->data - xdp->data_meta;
551 totsize = xdp->data_end - xdp->data + metasize;
552
553 if (sizeof(*xdpf) + totsize > PAGE_SIZE)
554 return NULL;
555
556 page = dev_alloc_page();
557 if (!page)
558 return NULL;
559
560 addr = page_to_virt(page);
561 xdpf = addr;
562 memset(xdpf, 0, sizeof(*xdpf));
563
564 addr += sizeof(*xdpf);
565 data_to_copy = metasize ? xdp->data_meta : xdp->data;
566 memcpy(addr, data_to_copy, totsize);
567
568 xdpf->data = addr + metasize;
569 xdpf->len = totsize - metasize;
570 xdpf->headroom = 0;
571 xdpf->metasize = metasize;
572 xdpf->frame_sz = PAGE_SIZE;
573 xdpf->mem.type = MEM_TYPE_PAGE_ORDER0;
574
575 xsk_buff_free(xdp);
576 return xdpf;
577 }
578 EXPORT_SYMBOL_GPL(xdp_convert_zc_to_xdp_frame);
579
580 /* Used by XDP_WARN macro, to avoid inlining WARN() in fast-path */
581 void xdp_warn(const char *msg, const char *func, const int line)
582 {
583 WARN(1, "XDP_WARN: %s(line:%d): %s\n", func, line, msg);
584 };
585 EXPORT_SYMBOL_GPL(xdp_warn);
586
587 int xdp_alloc_skb_bulk(void **skbs, int n_skb, gfp_t gfp)
588 {
589 n_skb = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache, gfp, n_skb, skbs);
590 if (unlikely(!n_skb))
591 return -ENOMEM;
592
593 return 0;
594 }
595 EXPORT_SYMBOL_GPL(xdp_alloc_skb_bulk);
596
597 struct sk_buff *__xdp_build_skb_from_frame(struct xdp_frame *xdpf,
598 struct sk_buff *skb,
599 struct net_device *dev)
600 {
601 struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf);
602 unsigned int headroom, frame_size;
603 void *hard_start;
604 u8 nr_frags;
605
606 /* xdp frags frame */
607 if (unlikely(xdp_frame_has_frags(xdpf)))
608 nr_frags = sinfo->nr_frags;
609
610 /* Part of headroom was reserved to xdpf */
611 headroom = sizeof(*xdpf) + xdpf->headroom;
612
613 /* Memory size backing xdp_frame data already have reserved
614 * room for build_skb to place skb_shared_info in tailroom.
615 */
616 frame_size = xdpf->frame_sz;
617
618 hard_start = xdpf->data - headroom;
619 skb = build_skb_around(skb, hard_start, frame_size);
620 if (unlikely(!skb))
621 return NULL;
622
623 skb_reserve(skb, headroom);
624 __skb_put(skb, xdpf->len);
625 if (xdpf->metasize)
626 skb_metadata_set(skb, xdpf->metasize);
627
628 if (unlikely(xdp_frame_has_frags(xdpf)))
629 xdp_update_skb_shared_info(skb, nr_frags,
630 sinfo->xdp_frags_size,
631 nr_frags * xdpf->frame_sz,
632 xdp_frame_is_frag_pfmemalloc(xdpf));
633
634 /* Essential SKB info: protocol and skb->dev */
635 skb->protocol = eth_type_trans(skb, dev);
636
637 /* Optional SKB info, currently missing:
638 * - HW checksum info (skb->ip_summed)
639 * - HW RX hash (skb_set_hash)
640 * - RX ring dev queue index (skb_record_rx_queue)
641 */
642
643 if (xdpf->mem.type == MEM_TYPE_PAGE_POOL)
644 skb_mark_for_recycle(skb);
645
646 /* Allow SKB to reuse area used by xdp_frame */
647 xdp_scrub_frame(xdpf);
648
649 return skb;
650 }
651 EXPORT_SYMBOL_GPL(__xdp_build_skb_from_frame);
652
653 struct sk_buff *xdp_build_skb_from_frame(struct xdp_frame *xdpf,
654 struct net_device *dev)
655 {
656 struct sk_buff *skb;
657
658 skb = kmem_cache_alloc(net_hotdata.skbuff_cache, GFP_ATOMIC);
659 if (unlikely(!skb))
660 return NULL;
661
662 memset(skb, 0, offsetof(struct sk_buff, tail));
663
664 return __xdp_build_skb_from_frame(xdpf, skb, dev);
665 }
666 EXPORT_SYMBOL_GPL(xdp_build_skb_from_frame);
667
668 struct xdp_frame *xdpf_clone(struct xdp_frame *xdpf)
669 {
670 unsigned int headroom, totalsize;
671 struct xdp_frame *nxdpf;
672 struct page *page;
673 void *addr;
674
675 headroom = xdpf->headroom + sizeof(*xdpf);
676 totalsize = headroom + xdpf->len;
677
678 if (unlikely(totalsize > PAGE_SIZE))
679 return NULL;
680 page = dev_alloc_page();
681 if (!page)
682 return NULL;
683 addr = page_to_virt(page);
684
685 memcpy(addr, xdpf, totalsize);
686
687 nxdpf = addr;
688 nxdpf->data = addr + headroom;
689 nxdpf->frame_sz = PAGE_SIZE;
690 nxdpf->mem.type = MEM_TYPE_PAGE_ORDER0;
691 nxdpf->mem.id = 0;
692
693 return nxdpf;
694 }
695
696 __bpf_kfunc_start_defs();
697
698 /**
699 * bpf_xdp_metadata_rx_timestamp - Read XDP frame RX timestamp.
700 * @ctx: XDP context pointer.
701 * @timestamp: Return value pointer.
702 *
703 * Return:
704 * * Returns 0 on success or ``-errno`` on error.
705 * * ``-EOPNOTSUPP`` : means device driver does not implement kfunc
706 * * ``-ENODATA`` : means no RX-timestamp available for this frame
707 */
708 __bpf_kfunc int bpf_xdp_metadata_rx_timestamp(const struct xdp_md *ctx, u64 *timestamp)
709 {
710 return -EOPNOTSUPP;
711 }
712
713 /**
714 * bpf_xdp_metadata_rx_hash - Read XDP frame RX hash.
715 * @ctx: XDP context pointer.
716 * @hash: Return value pointer.
717 * @rss_type: Return value pointer for RSS type.
718 *
719 * The RSS hash type (@rss_type) specifies what portion of packet headers NIC
720 * hardware used when calculating RSS hash value. The RSS type can be decoded
721 * via &enum xdp_rss_hash_type either matching on individual L3/L4 bits
722 * ``XDP_RSS_L*`` or by combined traditional *RSS Hashing Types*
723 * ``XDP_RSS_TYPE_L*``.
724 *
725 * Return:
726 * * Returns 0 on success or ``-errno`` on error.
727 * * ``-EOPNOTSUPP`` : means device driver doesn't implement kfunc
728 * * ``-ENODATA`` : means no RX-hash available for this frame
729 */
730 __bpf_kfunc int bpf_xdp_metadata_rx_hash(const struct xdp_md *ctx, u32 *hash,
731 enum xdp_rss_hash_type *rss_type)
732 {
733 return -EOPNOTSUPP;
734 }
735
736 /**
737 * bpf_xdp_metadata_rx_vlan_tag - Get XDP packet outermost VLAN tag
738 * @ctx: XDP context pointer.
739 * @vlan_proto: Destination pointer for VLAN Tag protocol identifier (TPID).
740 * @vlan_tci: Destination pointer for VLAN TCI (VID + DEI + PCP)
741 *
742 * In case of success, ``vlan_proto`` contains *Tag protocol identifier (TPID)*,
743 * usually ``ETH_P_8021Q`` or ``ETH_P_8021AD``, but some networks can use
744 * custom TPIDs. ``vlan_proto`` is stored in **network byte order (BE)**
745 * and should be used as follows:
746 * ``if (vlan_proto == bpf_htons(ETH_P_8021Q)) do_something();``
747 *
748 * ``vlan_tci`` contains the remaining 16 bits of a VLAN tag.
749 * Driver is expected to provide those in **host byte order (usually LE)**,
750 * so the bpf program should not perform byte conversion.
751 * According to 802.1Q standard, *VLAN TCI (Tag control information)*
752 * is a bit field that contains:
753 * *VLAN identifier (VID)* that can be read with ``vlan_tci & 0xfff``,
754 * *Drop eligible indicator (DEI)* - 1 bit,
755 * *Priority code point (PCP)* - 3 bits.
756 * For detailed meaning of DEI and PCP, please refer to other sources.
757 *
758 * Return:
759 * * Returns 0 on success or ``-errno`` on error.
760 * * ``-EOPNOTSUPP`` : device driver doesn't implement kfunc
761 * * ``-ENODATA`` : VLAN tag was not stripped or is not available
762 */
763 __bpf_kfunc int bpf_xdp_metadata_rx_vlan_tag(const struct xdp_md *ctx,
764 __be16 *vlan_proto, u16 *vlan_tci)
765 {
766 return -EOPNOTSUPP;
767 }
768
769 __bpf_kfunc_end_defs();
770
771 BTF_KFUNCS_START(xdp_metadata_kfunc_ids)
772 #define XDP_METADATA_KFUNC(_, __, name, ___) BTF_ID_FLAGS(func, name, KF_TRUSTED_ARGS)
773 XDP_METADATA_KFUNC_xxx
774 #undef XDP_METADATA_KFUNC
775 BTF_KFUNCS_END(xdp_metadata_kfunc_ids)
776
777 static const struct btf_kfunc_id_set xdp_metadata_kfunc_set = {
778 .owner = THIS_MODULE,
779 .set = &xdp_metadata_kfunc_ids,
780 };
781
782 BTF_ID_LIST(xdp_metadata_kfunc_ids_unsorted)
783 #define XDP_METADATA_KFUNC(name, _, str, __) BTF_ID(func, str)
784 XDP_METADATA_KFUNC_xxx
785 #undef XDP_METADATA_KFUNC
786
787 u32 bpf_xdp_metadata_kfunc_id(int id)
788 {
789 /* xdp_metadata_kfunc_ids is sorted and can't be used */
790 return xdp_metadata_kfunc_ids_unsorted[id];
791 }
792
793 bool bpf_dev_bound_kfunc_id(u32 btf_id)
794 {
795 return btf_id_set8_contains(&xdp_metadata_kfunc_ids, btf_id);
796 }
797
798 static int __init xdp_metadata_init(void)
799 {
800 return register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &xdp_metadata_kfunc_set);
801 }
802 late_initcall(xdp_metadata_init);
803
804 void xdp_set_features_flag(struct net_device *dev, xdp_features_t val)
805 {
806 val &= NETDEV_XDP_ACT_MASK;
807 if (dev->xdp_features == val)
808 return;
809
810 dev->xdp_features = val;
811
812 if (dev->reg_state == NETREG_REGISTERED)
813 call_netdevice_notifiers(NETDEV_XDP_FEAT_CHANGE, dev);
814 }
815 EXPORT_SYMBOL_GPL(xdp_set_features_flag);
816
817 void xdp_features_set_redirect_target(struct net_device *dev, bool support_sg)
818 {
819 xdp_features_t val = (dev->xdp_features | NETDEV_XDP_ACT_NDO_XMIT);
820
821 if (support_sg)
822 val |= NETDEV_XDP_ACT_NDO_XMIT_SG;
823 xdp_set_features_flag(dev, val);
824 }
825 EXPORT_SYMBOL_GPL(xdp_features_set_redirect_target);
826
827 void xdp_features_clear_redirect_target(struct net_device *dev)
828 {
829 xdp_features_t val = dev->xdp_features;
830
831 val &= ~(NETDEV_XDP_ACT_NDO_XMIT | NETDEV_XDP_ACT_NDO_XMIT_SG);
832 xdp_set_features_flag(dev, val);
833 }
834 EXPORT_SYMBOL_GPL(xdp_features_clear_redirect_target);
835
Linux® is a registered trademark of Linus Torvalds in the United States and other countries.
TOMOYO® is a registered trademark of NTT DATA CORPORATION.