Commit | Line | Data |
---|---|---|
ddc64d0a | 1 | // SPDX-License-Identifier: GPL-2.0-only |
aecd67b6 JDB |
2 | /* net/core/xdp.c |
3 | * | |
4 | * Copyright (c) 2017 Jesper Dangaard Brouer, Red Hat Inc. | |
aecd67b6 | 5 | */ |
05296620 | 6 | #include <linux/bpf.h> |
400031e0 | 7 | #include <linux/btf.h> |
3d76a4d3 | 8 | #include <linux/btf_ids.h> |
05296620 | 9 | #include <linux/filter.h> |
aecd67b6 JDB |
10 | #include <linux/types.h> |
11 | #include <linux/mm.h> | |
05296620 | 12 | #include <linux/netdevice.h> |
8d5d8852 JDB |
13 | #include <linux/slab.h> |
14 | #include <linux/idr.h> | |
15 | #include <linux/rhashtable.h> | |
34cc0b33 | 16 | #include <linux/bug.h> |
a9ca9f9c | 17 | #include <net/page_pool/helpers.h> |
aecd67b6 | 18 | |
aa70d2d1 | 19 | #include <net/hotdata.h> |
aecd67b6 | 20 | #include <net/xdp.h> |
f033b688 JDB |
21 | #include <net/xdp_priv.h> /* struct xdp_mem_allocator */ |
22 | #include <trace/events/xdp.h> | |
2b43470a | 23 | #include <net/xdp_sock_drv.h> |
aecd67b6 JDB |
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 | ||
8d5d8852 JDB |
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 | ||
8d5d8852 JDB |
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 | ||
c593642c | 44 | BUILD_BUG_ON(sizeof_field(struct xdp_mem_allocator, mem.id) |
8d5d8852 JDB |
45 | != sizeof(u32)); |
46 | ||
9f9a7077 N |
47 | /* Use cyclic increasing ID as direct hash key */ |
48 | return key; | |
8d5d8852 JDB |
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), | |
c593642c | 64 | .key_len = sizeof_field(struct xdp_mem_allocator, mem.id), |
8d5d8852 JDB |
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 */ | |
6a571895 | 79 | ida_free(&mem_id_pool, xa->mem.id); |
8d5d8852 | 80 | |
8d5d8852 JDB |
81 | kfree(xa); |
82 | } | |
83 | ||
c3f812ce | 84 | static void mem_xa_remove(struct xdp_mem_allocator *xa) |
99c07c43 | 85 | { |
c3f812ce | 86 | trace_mem_disconnect(xa); |
99c07c43 | 87 | |
c3f812ce | 88 | if (!rhashtable_remove_fast(mem_id_ht, &xa->node, mem_id_rht_params)) |
99c07c43 | 89 | call_rcu(&xa->rcu, __xdp_mem_allocator_rcu_free); |
99c07c43 JDB |
90 | } |
91 | ||
c3f812ce JL |
92 | static void mem_allocator_disconnect(void *allocator) |
93 | { | |
94 | struct xdp_mem_allocator *xa; | |
95 | struct rhashtable_iter iter; | |
96 | ||
86c76c09 JL |
97 | mutex_lock(&mem_id_lock); |
98 | ||
c3f812ce JL |
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); | |
99c07c43 | 109 | |
c3f812ce JL |
110 | } while (xa == ERR_PTR(-EAGAIN)); |
111 | rhashtable_walk_exit(&iter); | |
86c76c09 JL |
112 | |
113 | mutex_unlock(&mem_id_lock); | |
c3f812ce JL |
114 | } |
115 | ||
4a48ef70 | 116 | void xdp_unreg_mem_model(struct xdp_mem_info *mem) |
8d5d8852 JDB |
117 | { |
118 | struct xdp_mem_allocator *xa; | |
4a48ef70 THJ |
119 | int type = mem->type; |
120 | int id = mem->id; | |
8d5d8852 | 121 | |
a78cae24 | 122 | /* Reset mem info to defaults */ |
4a48ef70 THJ |
123 | mem->id = 0; |
124 | mem->type = 0; | |
dce5bd61 | 125 | |
8d5d8852 JDB |
126 | if (id == 0) |
127 | return; | |
128 | ||
a78cae24 | 129 | if (type == MEM_TYPE_PAGE_POOL) { |
c3f812ce JL |
130 | rcu_read_lock(); |
131 | xa = rhashtable_lookup(mem_id_ht, &id, mem_id_rht_params); | |
132 | page_pool_destroy(xa->page_pool); | |
133 | rcu_read_unlock(); | |
99c07c43 | 134 | } |
8d5d8852 | 135 | } |
4a48ef70 THJ |
136 | EXPORT_SYMBOL_GPL(xdp_unreg_mem_model); |
137 | ||
138 | void xdp_rxq_info_unreg_mem_model(struct xdp_rxq_info *xdp_rxq) | |
139 | { | |
140 | if (xdp_rxq->reg_state != REG_STATE_REGISTERED) { | |
141 | WARN(1, "Missing register, driver bug"); | |
142 | return; | |
143 | } | |
144 | ||
145 | xdp_unreg_mem_model(&xdp_rxq->mem); | |
146 | } | |
dce5bd61 | 147 | EXPORT_SYMBOL_GPL(xdp_rxq_info_unreg_mem_model); |
8d5d8852 | 148 | |
aecd67b6 JDB |
149 | void xdp_rxq_info_unreg(struct xdp_rxq_info *xdp_rxq) |
150 | { | |
151 | /* Simplify driver cleanup code paths, allow unreg "unused" */ | |
152 | if (xdp_rxq->reg_state == REG_STATE_UNUSED) | |
153 | return; | |
154 | ||
dce5bd61 | 155 | xdp_rxq_info_unreg_mem_model(xdp_rxq); |
8d5d8852 | 156 | |
aecd67b6 JDB |
157 | xdp_rxq->reg_state = REG_STATE_UNREGISTERED; |
158 | xdp_rxq->dev = NULL; | |
159 | } | |
160 | EXPORT_SYMBOL_GPL(xdp_rxq_info_unreg); | |
161 | ||
162 | static void xdp_rxq_info_init(struct xdp_rxq_info *xdp_rxq) | |
163 | { | |
164 | memset(xdp_rxq, 0, sizeof(*xdp_rxq)); | |
165 | } | |
166 | ||
167 | /* Returns 0 on success, negative on failure */ | |
bf25146a EC |
168 | int __xdp_rxq_info_reg(struct xdp_rxq_info *xdp_rxq, |
169 | struct net_device *dev, u32 queue_index, | |
170 | unsigned int napi_id, u32 frag_size) | |
aecd67b6 | 171 | { |
f85b244e YD |
172 | if (!dev) { |
173 | WARN(1, "Missing net_device from driver"); | |
174 | return -ENODEV; | |
175 | } | |
176 | ||
aecd67b6 JDB |
177 | if (xdp_rxq->reg_state == REG_STATE_UNUSED) { |
178 | WARN(1, "Driver promised not to register this"); | |
179 | return -EINVAL; | |
180 | } | |
181 | ||
182 | if (xdp_rxq->reg_state == REG_STATE_REGISTERED) { | |
183 | WARN(1, "Missing unregister, handled but fix driver"); | |
184 | xdp_rxq_info_unreg(xdp_rxq); | |
185 | } | |
186 | ||
aecd67b6 JDB |
187 | /* State either UNREGISTERED or NEW */ |
188 | xdp_rxq_info_init(xdp_rxq); | |
189 | xdp_rxq->dev = dev; | |
190 | xdp_rxq->queue_index = queue_index; | |
b02e5a0e | 191 | xdp_rxq->napi_id = napi_id; |
bf25146a | 192 | xdp_rxq->frag_size = frag_size; |
aecd67b6 JDB |
193 | |
194 | xdp_rxq->reg_state = REG_STATE_REGISTERED; | |
195 | return 0; | |
196 | } | |
bf25146a | 197 | EXPORT_SYMBOL_GPL(__xdp_rxq_info_reg); |
aecd67b6 JDB |
198 | |
199 | void xdp_rxq_info_unused(struct xdp_rxq_info *xdp_rxq) | |
200 | { | |
201 | xdp_rxq->reg_state = REG_STATE_UNUSED; | |
202 | } | |
203 | EXPORT_SYMBOL_GPL(xdp_rxq_info_unused); | |
c0124f32 JDB |
204 | |
205 | bool xdp_rxq_info_is_reg(struct xdp_rxq_info *xdp_rxq) | |
206 | { | |
207 | return (xdp_rxq->reg_state == REG_STATE_REGISTERED); | |
208 | } | |
209 | EXPORT_SYMBOL_GPL(xdp_rxq_info_is_reg); | |
5ab073ff | 210 | |
8d5d8852 JDB |
211 | static int __mem_id_init_hash_table(void) |
212 | { | |
213 | struct rhashtable *rht; | |
214 | int ret; | |
215 | ||
216 | if (unlikely(mem_id_init)) | |
217 | return 0; | |
218 | ||
219 | rht = kzalloc(sizeof(*rht), GFP_KERNEL); | |
220 | if (!rht) | |
221 | return -ENOMEM; | |
222 | ||
223 | ret = rhashtable_init(rht, &mem_id_rht_params); | |
224 | if (ret < 0) { | |
225 | kfree(rht); | |
226 | return ret; | |
227 | } | |
228 | mem_id_ht = rht; | |
229 | smp_mb(); /* mutex lock should provide enough pairing */ | |
230 | mem_id_init = true; | |
231 | ||
232 | return 0; | |
233 | } | |
234 | ||
235 | /* Allocate a cyclic ID that maps to allocator pointer. | |
236 | * See: https://www.kernel.org/doc/html/latest/core-api/idr.html | |
237 | * | |
238 | * Caller must lock mem_id_lock. | |
239 | */ | |
240 | static int __mem_id_cyclic_get(gfp_t gfp) | |
241 | { | |
242 | int retries = 1; | |
243 | int id; | |
244 | ||
245 | again: | |
6a571895 | 246 | id = ida_alloc_range(&mem_id_pool, mem_id_next, MEM_ID_MAX - 1, gfp); |
8d5d8852 JDB |
247 | if (id < 0) { |
248 | if (id == -ENOSPC) { | |
249 | /* Cyclic allocator, reset next id */ | |
250 | if (retries--) { | |
251 | mem_id_next = MEM_ID_MIN; | |
252 | goto again; | |
253 | } | |
254 | } | |
255 | return id; /* errno */ | |
256 | } | |
257 | mem_id_next = id + 1; | |
258 | ||
259 | return id; | |
260 | } | |
261 | ||
57d0a1c1 JDB |
262 | static bool __is_supported_mem_type(enum xdp_mem_type type) |
263 | { | |
264 | if (type == MEM_TYPE_PAGE_POOL) | |
265 | return is_page_pool_compiled_in(); | |
266 | ||
267 | if (type >= MEM_TYPE_MAX) | |
268 | return false; | |
269 | ||
270 | return true; | |
271 | } | |
272 | ||
4a48ef70 THJ |
273 | static struct xdp_mem_allocator *__xdp_reg_mem_model(struct xdp_mem_info *mem, |
274 | enum xdp_mem_type type, | |
275 | void *allocator) | |
5ab073ff | 276 | { |
8d5d8852 JDB |
277 | struct xdp_mem_allocator *xdp_alloc; |
278 | gfp_t gfp = GFP_KERNEL; | |
279 | int id, errno, ret; | |
280 | void *ptr; | |
281 | ||
57d0a1c1 | 282 | if (!__is_supported_mem_type(type)) |
4a48ef70 | 283 | return ERR_PTR(-EOPNOTSUPP); |
5ab073ff | 284 | |
4a48ef70 | 285 | mem->type = type; |
5ab073ff | 286 | |
57d0a1c1 | 287 | if (!allocator) { |
0807892e | 288 | if (type == MEM_TYPE_PAGE_POOL) |
4a48ef70 THJ |
289 | return ERR_PTR(-EINVAL); /* Setup time check page_pool req */ |
290 | return NULL; | |
57d0a1c1 | 291 | } |
8d5d8852 JDB |
292 | |
293 | /* Delay init of rhashtable to save memory if feature isn't used */ | |
294 | if (!mem_id_init) { | |
295 | mutex_lock(&mem_id_lock); | |
296 | ret = __mem_id_init_hash_table(); | |
297 | mutex_unlock(&mem_id_lock); | |
7e9f7942 | 298 | if (ret < 0) |
4a48ef70 | 299 | return ERR_PTR(ret); |
8d5d8852 JDB |
300 | } |
301 | ||
302 | xdp_alloc = kzalloc(sizeof(*xdp_alloc), gfp); | |
303 | if (!xdp_alloc) | |
4a48ef70 | 304 | return ERR_PTR(-ENOMEM); |
8d5d8852 JDB |
305 | |
306 | mutex_lock(&mem_id_lock); | |
307 | id = __mem_id_cyclic_get(gfp); | |
308 | if (id < 0) { | |
309 | errno = id; | |
310 | goto err; | |
311 | } | |
4a48ef70 THJ |
312 | mem->id = id; |
313 | xdp_alloc->mem = *mem; | |
8d5d8852 JDB |
314 | xdp_alloc->allocator = allocator; |
315 | ||
316 | /* Insert allocator into ID lookup table */ | |
317 | ptr = rhashtable_insert_slow(mem_id_ht, &id, &xdp_alloc->node); | |
318 | if (IS_ERR(ptr)) { | |
6a571895 | 319 | ida_free(&mem_id_pool, mem->id); |
4a48ef70 | 320 | mem->id = 0; |
8d5d8852 JDB |
321 | errno = PTR_ERR(ptr); |
322 | goto err; | |
323 | } | |
324 | ||
1da4bbef | 325 | if (type == MEM_TYPE_PAGE_POOL) |
64693ec7 | 326 | page_pool_use_xdp_mem(allocator, mem_allocator_disconnect, mem); |
1da4bbef | 327 | |
8d5d8852 | 328 | mutex_unlock(&mem_id_lock); |
5ab073ff | 329 | |
4a48ef70 | 330 | return xdp_alloc; |
8d5d8852 JDB |
331 | err: |
332 | mutex_unlock(&mem_id_lock); | |
333 | kfree(xdp_alloc); | |
4a48ef70 THJ |
334 | return ERR_PTR(errno); |
335 | } | |
336 | ||
337 | int xdp_reg_mem_model(struct xdp_mem_info *mem, | |
338 | enum xdp_mem_type type, void *allocator) | |
339 | { | |
340 | struct xdp_mem_allocator *xdp_alloc; | |
341 | ||
342 | xdp_alloc = __xdp_reg_mem_model(mem, type, allocator); | |
343 | if (IS_ERR(xdp_alloc)) | |
344 | return PTR_ERR(xdp_alloc); | |
345 | return 0; | |
346 | } | |
347 | EXPORT_SYMBOL_GPL(xdp_reg_mem_model); | |
348 | ||
349 | int xdp_rxq_info_reg_mem_model(struct xdp_rxq_info *xdp_rxq, | |
350 | enum xdp_mem_type type, void *allocator) | |
351 | { | |
352 | struct xdp_mem_allocator *xdp_alloc; | |
353 | ||
354 | if (xdp_rxq->reg_state != REG_STATE_REGISTERED) { | |
355 | WARN(1, "Missing register, driver bug"); | |
356 | return -EFAULT; | |
357 | } | |
358 | ||
359 | xdp_alloc = __xdp_reg_mem_model(&xdp_rxq->mem, type, allocator); | |
360 | if (IS_ERR(xdp_alloc)) | |
361 | return PTR_ERR(xdp_alloc); | |
362 | ||
e0ae7130 SAS |
363 | if (trace_mem_connect_enabled() && xdp_alloc) |
364 | trace_mem_connect(xdp_alloc, xdp_rxq); | |
4a48ef70 | 365 | return 0; |
5ab073ff | 366 | } |
4a48ef70 | 367 | |
5ab073ff | 368 | EXPORT_SYMBOL_GPL(xdp_rxq_info_reg_mem_model); |
8d5d8852 | 369 | |
389ab7f0 JDB |
370 | /* XDP RX runs under NAPI protection, and in different delivery error |
371 | * scenarios (e.g. queue full), it is possible to return the xdp_frame | |
baead859 | 372 | * while still leveraging this protection. The @napi_direct boolean |
389ab7f0 | 373 | * is used for those calls sites. Thus, allowing for faster recycling |
ed1182dc | 374 | * of xdp_frames/pages in those cases. |
389ab7f0 | 375 | */ |
bf25146a EC |
376 | void __xdp_return(void *data, struct xdp_mem_info *mem, bool napi_direct, |
377 | struct xdp_buff *xdp) | |
8d5d8852 | 378 | { |
57d0a1c1 JDB |
379 | struct page *page; |
380 | ||
381 | switch (mem->type) { | |
382 | case MEM_TYPE_PAGE_POOL: | |
57d0a1c1 | 383 | page = virt_to_head_page(data); |
622d1369 OBL |
384 | if (napi_direct && xdp_return_frame_no_direct()) |
385 | napi_direct = false; | |
fb33ec01 JDB |
386 | /* No need to check ((page->pp_magic & ~0x3UL) == PP_SIGNATURE) |
387 | * as mem->type knows this a page_pool page | |
388 | */ | |
389 | page_pool_put_full_page(page->pp, page, napi_direct); | |
57d0a1c1 JDB |
390 | break; |
391 | case MEM_TYPE_PAGE_SHARED: | |
8d5d8852 | 392 | page_frag_free(data); |
57d0a1c1 JDB |
393 | break; |
394 | case MEM_TYPE_PAGE_ORDER0: | |
395 | page = virt_to_page(data); /* Assumes order0 page*/ | |
8d5d8852 | 396 | put_page(page); |
57d0a1c1 | 397 | break; |
ed1182dc BT |
398 | case MEM_TYPE_XSK_BUFF_POOL: |
399 | /* NB! Only valid from an xdp_buff! */ | |
400 | xsk_buff_free(xdp); | |
401 | break; | |
57d0a1c1 JDB |
402 | default: |
403 | /* Not possible, checked in xdp_rxq_info_reg_mem_model() */ | |
82c41671 | 404 | WARN(1, "Incorrect XDP memory type (%d) usage", mem->type); |
57d0a1c1 | 405 | break; |
8d5d8852 JDB |
406 | } |
407 | } | |
c497176c BT |
408 | |
409 | void xdp_return_frame(struct xdp_frame *xdpf) | |
410 | { | |
7c48cb01 LB |
411 | struct skb_shared_info *sinfo; |
412 | int i; | |
413 | ||
414 | if (likely(!xdp_frame_has_frags(xdpf))) | |
415 | goto out; | |
416 | ||
417 | sinfo = xdp_get_shared_info_from_frame(xdpf); | |
418 | for (i = 0; i < sinfo->nr_frags; i++) { | |
419 | struct page *page = skb_frag_page(&sinfo->frags[i]); | |
420 | ||
421 | __xdp_return(page_address(page), &xdpf->mem, false, NULL); | |
422 | } | |
423 | out: | |
ed1182dc | 424 | __xdp_return(xdpf->data, &xdpf->mem, false, NULL); |
c497176c | 425 | } |
8d5d8852 | 426 | EXPORT_SYMBOL_GPL(xdp_return_frame); |
c497176c | 427 | |
389ab7f0 JDB |
428 | void xdp_return_frame_rx_napi(struct xdp_frame *xdpf) |
429 | { | |
7c48cb01 LB |
430 | struct skb_shared_info *sinfo; |
431 | int i; | |
432 | ||
433 | if (likely(!xdp_frame_has_frags(xdpf))) | |
434 | goto out; | |
435 | ||
436 | sinfo = xdp_get_shared_info_from_frame(xdpf); | |
437 | for (i = 0; i < sinfo->nr_frags; i++) { | |
438 | struct page *page = skb_frag_page(&sinfo->frags[i]); | |
439 | ||
440 | __xdp_return(page_address(page), &xdpf->mem, true, NULL); | |
441 | } | |
442 | out: | |
ed1182dc | 443 | __xdp_return(xdpf->data, &xdpf->mem, true, NULL); |
389ab7f0 JDB |
444 | } |
445 | EXPORT_SYMBOL_GPL(xdp_return_frame_rx_napi); | |
446 | ||
89653987 LB |
447 | /* XDP bulk APIs introduce a defer/flush mechanism to return |
448 | * pages belonging to the same xdp_mem_allocator object | |
449 | * (identified via the mem.id field) in bulk to optimize | |
450 | * I-cache and D-cache. | |
451 | * The bulk queue size is set to 16 to be aligned to how | |
452 | * XDP_REDIRECT bulking works. The bulk is flushed when | |
453 | * it is full or when mem.id changes. | |
454 | * xdp_frame_bulk is usually stored/allocated on the function | |
455 | * call-stack to avoid locking penalties. | |
456 | */ | |
457 | void xdp_flush_frame_bulk(struct xdp_frame_bulk *bq) | |
458 | { | |
459 | struct xdp_mem_allocator *xa = bq->xa; | |
89653987 | 460 | |
78862447 | 461 | if (unlikely(!xa || !bq->count)) |
89653987 LB |
462 | return; |
463 | ||
78862447 | 464 | page_pool_put_page_bulk(xa->page_pool, bq->q, bq->count); |
89653987 LB |
465 | /* bq->xa is not cleared to save lookup, if mem.id same in next bulk */ |
466 | bq->count = 0; | |
467 | } | |
468 | EXPORT_SYMBOL_GPL(xdp_flush_frame_bulk); | |
469 | ||
470 | /* Must be called with rcu_read_lock held */ | |
471 | void xdp_return_frame_bulk(struct xdp_frame *xdpf, | |
472 | struct xdp_frame_bulk *bq) | |
473 | { | |
474 | struct xdp_mem_info *mem = &xdpf->mem; | |
475 | struct xdp_mem_allocator *xa; | |
476 | ||
477 | if (mem->type != MEM_TYPE_PAGE_POOL) { | |
7c48cb01 | 478 | xdp_return_frame(xdpf); |
89653987 LB |
479 | return; |
480 | } | |
481 | ||
482 | xa = bq->xa; | |
483 | if (unlikely(!xa)) { | |
484 | xa = rhashtable_lookup(mem_id_ht, &mem->id, mem_id_rht_params); | |
485 | bq->count = 0; | |
486 | bq->xa = xa; | |
487 | } | |
488 | ||
489 | if (bq->count == XDP_BULK_QUEUE_SIZE) | |
490 | xdp_flush_frame_bulk(bq); | |
491 | ||
492 | if (unlikely(mem->id != xa->mem.id)) { | |
493 | xdp_flush_frame_bulk(bq); | |
494 | bq->xa = rhashtable_lookup(mem_id_ht, &mem->id, mem_id_rht_params); | |
495 | } | |
496 | ||
7c48cb01 LB |
497 | if (unlikely(xdp_frame_has_frags(xdpf))) { |
498 | struct skb_shared_info *sinfo; | |
499 | int i; | |
500 | ||
501 | sinfo = xdp_get_shared_info_from_frame(xdpf); | |
502 | for (i = 0; i < sinfo->nr_frags; i++) { | |
503 | skb_frag_t *frag = &sinfo->frags[i]; | |
504 | ||
505 | bq->q[bq->count++] = skb_frag_address(frag); | |
506 | if (bq->count == XDP_BULK_QUEUE_SIZE) | |
507 | xdp_flush_frame_bulk(bq); | |
508 | } | |
509 | } | |
89653987 LB |
510 | bq->q[bq->count++] = xdpf->data; |
511 | } | |
512 | EXPORT_SYMBOL_GPL(xdp_return_frame_bulk); | |
513 | ||
c497176c BT |
514 | void xdp_return_buff(struct xdp_buff *xdp) |
515 | { | |
7c48cb01 LB |
516 | struct skb_shared_info *sinfo; |
517 | int i; | |
518 | ||
519 | if (likely(!xdp_buff_has_frags(xdp))) | |
520 | goto out; | |
521 | ||
522 | sinfo = xdp_get_shared_info_from_buff(xdp); | |
523 | for (i = 0; i < sinfo->nr_frags; i++) { | |
524 | struct page *page = skb_frag_page(&sinfo->frags[i]); | |
525 | ||
526 | __xdp_return(page_address(page), &xdp->rxq->mem, true, xdp); | |
527 | } | |
528 | out: | |
ed1182dc | 529 | __xdp_return(xdp->data, &xdp->rxq->mem, true, xdp); |
c497176c | 530 | } |
718a18a0 | 531 | EXPORT_SYMBOL_GPL(xdp_return_buff); |
05296620 | 532 | |
05296620 JK |
533 | void xdp_attachment_setup(struct xdp_attachment_info *info, |
534 | struct netdev_bpf *bpf) | |
535 | { | |
536 | if (info->prog) | |
537 | bpf_prog_put(info->prog); | |
538 | info->prog = bpf->prog; | |
539 | info->flags = bpf->flags; | |
540 | } | |
541 | EXPORT_SYMBOL_GPL(xdp_attachment_setup); | |
b0d1beef BT |
542 | |
543 | struct xdp_frame *xdp_convert_zc_to_xdp_frame(struct xdp_buff *xdp) | |
544 | { | |
72962167 | 545 | unsigned int metasize, totsize; |
b0d1beef BT |
546 | void *addr, *data_to_copy; |
547 | struct xdp_frame *xdpf; | |
548 | struct page *page; | |
549 | ||
550 | /* Clone into a MEM_TYPE_PAGE_ORDER0 xdp_frame. */ | |
551 | metasize = xdp_data_meta_unsupported(xdp) ? 0 : | |
552 | xdp->data - xdp->data_meta; | |
b0d1beef BT |
553 | totsize = xdp->data_end - xdp->data + metasize; |
554 | ||
555 | if (sizeof(*xdpf) + totsize > PAGE_SIZE) | |
556 | return NULL; | |
557 | ||
558 | page = dev_alloc_page(); | |
559 | if (!page) | |
560 | return NULL; | |
561 | ||
562 | addr = page_to_virt(page); | |
563 | xdpf = addr; | |
564 | memset(xdpf, 0, sizeof(*xdpf)); | |
565 | ||
566 | addr += sizeof(*xdpf); | |
567 | data_to_copy = metasize ? xdp->data_meta : xdp->data; | |
568 | memcpy(addr, data_to_copy, totsize); | |
569 | ||
570 | xdpf->data = addr + metasize; | |
571 | xdpf->len = totsize - metasize; | |
572 | xdpf->headroom = 0; | |
573 | xdpf->metasize = metasize; | |
3ff23516 | 574 | xdpf->frame_sz = PAGE_SIZE; |
b0d1beef BT |
575 | xdpf->mem.type = MEM_TYPE_PAGE_ORDER0; |
576 | ||
82c41671 | 577 | xsk_buff_free(xdp); |
b0d1beef BT |
578 | return xdpf; |
579 | } | |
580 | EXPORT_SYMBOL_GPL(xdp_convert_zc_to_xdp_frame); | |
34cc0b33 JDB |
581 | |
582 | /* Used by XDP_WARN macro, to avoid inlining WARN() in fast-path */ | |
583 | void xdp_warn(const char *msg, const char *func, const int line) | |
584 | { | |
585 | WARN(1, "XDP_WARN: %s(line:%d): %s\n", func, line, msg); | |
586 | }; | |
587 | EXPORT_SYMBOL_GPL(xdp_warn); | |
97a0e1ea | 588 | |
65e6dcf7 LB |
589 | int xdp_alloc_skb_bulk(void **skbs, int n_skb, gfp_t gfp) |
590 | { | |
aa70d2d1 | 591 | n_skb = kmem_cache_alloc_bulk(net_hotdata.skbuff_cache, gfp, n_skb, skbs); |
65e6dcf7 LB |
592 | if (unlikely(!n_skb)) |
593 | return -ENOMEM; | |
594 | ||
595 | return 0; | |
596 | } | |
597 | EXPORT_SYMBOL_GPL(xdp_alloc_skb_bulk); | |
598 | ||
97a0e1ea LB |
599 | struct sk_buff *__xdp_build_skb_from_frame(struct xdp_frame *xdpf, |
600 | struct sk_buff *skb, | |
601 | struct net_device *dev) | |
602 | { | |
d65a1906 | 603 | struct skb_shared_info *sinfo = xdp_get_shared_info_from_frame(xdpf); |
97a0e1ea LB |
604 | unsigned int headroom, frame_size; |
605 | void *hard_start; | |
d65a1906 LB |
606 | u8 nr_frags; |
607 | ||
608 | /* xdp frags frame */ | |
609 | if (unlikely(xdp_frame_has_frags(xdpf))) | |
610 | nr_frags = sinfo->nr_frags; | |
97a0e1ea LB |
611 | |
612 | /* Part of headroom was reserved to xdpf */ | |
613 | headroom = sizeof(*xdpf) + xdpf->headroom; | |
614 | ||
615 | /* Memory size backing xdp_frame data already have reserved | |
616 | * room for build_skb to place skb_shared_info in tailroom. | |
617 | */ | |
618 | frame_size = xdpf->frame_sz; | |
619 | ||
620 | hard_start = xdpf->data - headroom; | |
621 | skb = build_skb_around(skb, hard_start, frame_size); | |
622 | if (unlikely(!skb)) | |
623 | return NULL; | |
624 | ||
625 | skb_reserve(skb, headroom); | |
626 | __skb_put(skb, xdpf->len); | |
627 | if (xdpf->metasize) | |
628 | skb_metadata_set(skb, xdpf->metasize); | |
629 | ||
d65a1906 LB |
630 | if (unlikely(xdp_frame_has_frags(xdpf))) |
631 | xdp_update_skb_shared_info(skb, nr_frags, | |
632 | sinfo->xdp_frags_size, | |
633 | nr_frags * xdpf->frame_sz, | |
634 | xdp_frame_is_frag_pfmemalloc(xdpf)); | |
635 | ||
97a0e1ea LB |
636 | /* Essential SKB info: protocol and skb->dev */ |
637 | skb->protocol = eth_type_trans(skb, dev); | |
638 | ||
639 | /* Optional SKB info, currently missing: | |
640 | * - HW checksum info (skb->ip_summed) | |
641 | * - HW RX hash (skb_set_hash) | |
642 | * - RX ring dev queue index (skb_record_rx_queue) | |
643 | */ | |
644 | ||
9c94bbf9 AL |
645 | if (xdpf->mem.type == MEM_TYPE_PAGE_POOL) |
646 | skb_mark_for_recycle(skb); | |
97a0e1ea LB |
647 | |
648 | /* Allow SKB to reuse area used by xdp_frame */ | |
649 | xdp_scrub_frame(xdpf); | |
650 | ||
651 | return skb; | |
652 | } | |
653 | EXPORT_SYMBOL_GPL(__xdp_build_skb_from_frame); | |
89f479f0 LB |
654 | |
655 | struct sk_buff *xdp_build_skb_from_frame(struct xdp_frame *xdpf, | |
656 | struct net_device *dev) | |
657 | { | |
658 | struct sk_buff *skb; | |
659 | ||
aa70d2d1 | 660 | skb = kmem_cache_alloc(net_hotdata.skbuff_cache, GFP_ATOMIC); |
89f479f0 LB |
661 | if (unlikely(!skb)) |
662 | return NULL; | |
663 | ||
664 | memset(skb, 0, offsetof(struct sk_buff, tail)); | |
665 | ||
666 | return __xdp_build_skb_from_frame(xdpf, skb, dev); | |
667 | } | |
668 | EXPORT_SYMBOL_GPL(xdp_build_skb_from_frame); | |
e624d4ed HL |
669 | |
670 | struct xdp_frame *xdpf_clone(struct xdp_frame *xdpf) | |
671 | { | |
672 | unsigned int headroom, totalsize; | |
673 | struct xdp_frame *nxdpf; | |
674 | struct page *page; | |
675 | void *addr; | |
676 | ||
677 | headroom = xdpf->headroom + sizeof(*xdpf); | |
678 | totalsize = headroom + xdpf->len; | |
679 | ||
680 | if (unlikely(totalsize > PAGE_SIZE)) | |
681 | return NULL; | |
682 | page = dev_alloc_page(); | |
683 | if (!page) | |
684 | return NULL; | |
685 | addr = page_to_virt(page); | |
686 | ||
687 | memcpy(addr, xdpf, totalsize); | |
688 | ||
689 | nxdpf = addr; | |
690 | nxdpf->data = addr + headroom; | |
691 | nxdpf->frame_sz = PAGE_SIZE; | |
692 | nxdpf->mem.type = MEM_TYPE_PAGE_ORDER0; | |
693 | nxdpf->mem.id = 0; | |
694 | ||
695 | return nxdpf; | |
696 | } | |
3d76a4d3 | 697 | |
391145ba | 698 | __bpf_kfunc_start_defs(); |
3d76a4d3 SF |
699 | |
700 | /** | |
701 | * bpf_xdp_metadata_rx_timestamp - Read XDP frame RX timestamp. | |
702 | * @ctx: XDP context pointer. | |
703 | * @timestamp: Return value pointer. | |
704 | * | |
915efd8a JDB |
705 | * Return: |
706 | * * Returns 0 on success or ``-errno`` on error. | |
707 | * * ``-EOPNOTSUPP`` : means device driver does not implement kfunc | |
708 | * * ``-ENODATA`` : means no RX-timestamp available for this frame | |
3d76a4d3 | 709 | */ |
400031e0 | 710 | __bpf_kfunc int bpf_xdp_metadata_rx_timestamp(const struct xdp_md *ctx, u64 *timestamp) |
3d76a4d3 SF |
711 | { |
712 | return -EOPNOTSUPP; | |
713 | } | |
714 | ||
715 | /** | |
716 | * bpf_xdp_metadata_rx_hash - Read XDP frame RX hash. | |
717 | * @ctx: XDP context pointer. | |
718 | * @hash: Return value pointer. | |
0cd917a4 JDB |
719 | * @rss_type: Return value pointer for RSS type. |
720 | * | |
721 | * The RSS hash type (@rss_type) specifies what portion of packet headers NIC | |
722 | * hardware used when calculating RSS hash value. The RSS type can be decoded | |
723 | * via &enum xdp_rss_hash_type either matching on individual L3/L4 bits | |
724 | * ``XDP_RSS_L*`` or by combined traditional *RSS Hashing Types* | |
725 | * ``XDP_RSS_TYPE_L*``. | |
3d76a4d3 | 726 | * |
915efd8a JDB |
727 | * Return: |
728 | * * Returns 0 on success or ``-errno`` on error. | |
729 | * * ``-EOPNOTSUPP`` : means device driver doesn't implement kfunc | |
730 | * * ``-ENODATA`` : means no RX-hash available for this frame | |
3d76a4d3 | 731 | */ |
0cd917a4 JDB |
732 | __bpf_kfunc int bpf_xdp_metadata_rx_hash(const struct xdp_md *ctx, u32 *hash, |
733 | enum xdp_rss_hash_type *rss_type) | |
3d76a4d3 SF |
734 | { |
735 | return -EOPNOTSUPP; | |
736 | } | |
737 | ||
e6795330 LZ |
738 | /** |
739 | * bpf_xdp_metadata_rx_vlan_tag - Get XDP packet outermost VLAN tag | |
740 | * @ctx: XDP context pointer. | |
741 | * @vlan_proto: Destination pointer for VLAN Tag protocol identifier (TPID). | |
742 | * @vlan_tci: Destination pointer for VLAN TCI (VID + DEI + PCP) | |
743 | * | |
744 | * In case of success, ``vlan_proto`` contains *Tag protocol identifier (TPID)*, | |
745 | * usually ``ETH_P_8021Q`` or ``ETH_P_8021AD``, but some networks can use | |
746 | * custom TPIDs. ``vlan_proto`` is stored in **network byte order (BE)** | |
747 | * and should be used as follows: | |
748 | * ``if (vlan_proto == bpf_htons(ETH_P_8021Q)) do_something();`` | |
749 | * | |
750 | * ``vlan_tci`` contains the remaining 16 bits of a VLAN tag. | |
751 | * Driver is expected to provide those in **host byte order (usually LE)**, | |
752 | * so the bpf program should not perform byte conversion. | |
753 | * According to 802.1Q standard, *VLAN TCI (Tag control information)* | |
754 | * is a bit field that contains: | |
755 | * *VLAN identifier (VID)* that can be read with ``vlan_tci & 0xfff``, | |
756 | * *Drop eligible indicator (DEI)* - 1 bit, | |
757 | * *Priority code point (PCP)* - 3 bits. | |
758 | * For detailed meaning of DEI and PCP, please refer to other sources. | |
759 | * | |
760 | * Return: | |
761 | * * Returns 0 on success or ``-errno`` on error. | |
762 | * * ``-EOPNOTSUPP`` : device driver doesn't implement kfunc | |
763 | * * ``-ENODATA`` : VLAN tag was not stripped or is not available | |
764 | */ | |
765 | __bpf_kfunc int bpf_xdp_metadata_rx_vlan_tag(const struct xdp_md *ctx, | |
766 | __be16 *vlan_proto, u16 *vlan_tci) | |
767 | { | |
768 | return -EOPNOTSUPP; | |
769 | } | |
770 | ||
391145ba | 771 | __bpf_kfunc_end_defs(); |
3d76a4d3 | 772 | |
6f3189f3 | 773 | BTF_KFUNCS_START(xdp_metadata_kfunc_ids) |
a9c2a608 | 774 | #define XDP_METADATA_KFUNC(_, __, name, ___) BTF_ID_FLAGS(func, name, KF_TRUSTED_ARGS) |
3d76a4d3 SF |
775 | XDP_METADATA_KFUNC_xxx |
776 | #undef XDP_METADATA_KFUNC | |
6f3189f3 | 777 | BTF_KFUNCS_END(xdp_metadata_kfunc_ids) |
3d76a4d3 SF |
778 | |
779 | static const struct btf_kfunc_id_set xdp_metadata_kfunc_set = { | |
780 | .owner = THIS_MODULE, | |
781 | .set = &xdp_metadata_kfunc_ids, | |
782 | }; | |
783 | ||
784 | BTF_ID_LIST(xdp_metadata_kfunc_ids_unsorted) | |
a9c2a608 | 785 | #define XDP_METADATA_KFUNC(name, _, str, __) BTF_ID(func, str) |
3d76a4d3 SF |
786 | XDP_METADATA_KFUNC_xxx |
787 | #undef XDP_METADATA_KFUNC | |
788 | ||
789 | u32 bpf_xdp_metadata_kfunc_id(int id) | |
790 | { | |
791 | /* xdp_metadata_kfunc_ids is sorted and can't be used */ | |
792 | return xdp_metadata_kfunc_ids_unsorted[id]; | |
793 | } | |
794 | ||
795 | bool bpf_dev_bound_kfunc_id(u32 btf_id) | |
796 | { | |
797 | return btf_id_set8_contains(&xdp_metadata_kfunc_ids, btf_id); | |
798 | } | |
799 | ||
800 | static int __init xdp_metadata_init(void) | |
801 | { | |
802 | return register_btf_kfunc_id_set(BPF_PROG_TYPE_XDP, &xdp_metadata_kfunc_set); | |
803 | } | |
804 | late_initcall(xdp_metadata_init); | |
66c0e13a | 805 | |
f85949f9 | 806 | void xdp_set_features_flag(struct net_device *dev, xdp_features_t val) |
66c0e13a | 807 | { |
f85949f9 LB |
808 | val &= NETDEV_XDP_ACT_MASK; |
809 | if (dev->xdp_features == val) | |
810 | return; | |
66c0e13a | 811 | |
f85949f9 | 812 | dev->xdp_features = val; |
769639c1 JK |
813 | |
814 | if (dev->reg_state == NETREG_REGISTERED) | |
815 | call_netdevice_notifiers(NETDEV_XDP_FEAT_CHANGE, dev); | |
66c0e13a | 816 | } |
f85949f9 LB |
817 | EXPORT_SYMBOL_GPL(xdp_set_features_flag); |
818 | ||
819 | void xdp_features_set_redirect_target(struct net_device *dev, bool support_sg) | |
820 | { | |
821 | xdp_features_t val = (dev->xdp_features | NETDEV_XDP_ACT_NDO_XMIT); | |
822 | ||
823 | if (support_sg) | |
824 | val |= NETDEV_XDP_ACT_NDO_XMIT_SG; | |
825 | xdp_set_features_flag(dev, val); | |
826 | } | |
66c0e13a MM |
827 | EXPORT_SYMBOL_GPL(xdp_features_set_redirect_target); |
828 | ||
829 | void xdp_features_clear_redirect_target(struct net_device *dev) | |
830 | { | |
f85949f9 LB |
831 | xdp_features_t val = dev->xdp_features; |
832 | ||
833 | val &= ~(NETDEV_XDP_ACT_NDO_XMIT | NETDEV_XDP_ACT_NDO_XMIT_SG); | |
834 | xdp_set_features_flag(dev, val); | |
66c0e13a MM |
835 | } |
836 | EXPORT_SYMBOL_GPL(xdp_features_clear_redirect_target); |