1 // SPDX-License-Identifier: GPL-2.0-or-later
2 #include <net/gro.h>
3 #include <net/dst_metadata.h>
4 #include <net/busy_poll.h>
5 #include <trace/events/net.h>
6 #include <linux/skbuff_ref.h>
7
8 #define MAX_GRO_SKBS 8
9
10 static DEFINE_SPINLOCK(offload_lock);
11
12 /**
13 * dev_add_offload - register offload handlers
14 * @po: protocol offload declaration
15 *
16 * Add protocol offload handlers to the networking stack. The passed
17 * &proto_offload is linked into kernel lists and may not be freed until
18 * it has been removed from the kernel lists.
19 *
20 * This call does not sleep therefore it can not
21 * guarantee all CPU's that are in middle of receiving packets
22 * will see the new offload handlers (until the next received packet).
23 */
dev_add_offload(struct packet_offload * po)24 void dev_add_offload(struct packet_offload *po)
25 {
26 struct packet_offload *elem;
27
28 spin_lock(&offload_lock);
29 list_for_each_entry(elem, &net_hotdata.offload_base, list) {
30 if (po->priority < elem->priority)
31 break;
32 }
33 list_add_rcu(&po->list, elem->list.prev);
34 spin_unlock(&offload_lock);
35 }
36 EXPORT_SYMBOL(dev_add_offload);
37
38 /**
39 * __dev_remove_offload - remove offload handler
40 * @po: packet offload declaration
41 *
42 * Remove a protocol offload handler that was previously added to the
43 * kernel offload handlers by dev_add_offload(). The passed &offload_type
44 * is removed from the kernel lists and can be freed or reused once this
45 * function returns.
46 *
47 * The packet type might still be in use by receivers
48 * and must not be freed until after all the CPU's have gone
49 * through a quiescent state.
50 */
__dev_remove_offload(struct packet_offload * po)51 static void __dev_remove_offload(struct packet_offload *po)
52 {
53 struct list_head *head = &net_hotdata.offload_base;
54 struct packet_offload *po1;
55
56 spin_lock(&offload_lock);
57
58 list_for_each_entry(po1, head, list) {
59 if (po == po1) {
60 list_del_rcu(&po->list);
61 goto out;
62 }
63 }
64
65 pr_warn("dev_remove_offload: %p not found\n", po);
66 out:
67 spin_unlock(&offload_lock);
68 }
69
70 /**
71 * dev_remove_offload - remove packet offload handler
72 * @po: packet offload declaration
73 *
74 * Remove a packet offload handler that was previously added to the kernel
75 * offload handlers by dev_add_offload(). The passed &offload_type is
76 * removed from the kernel lists and can be freed or reused once this
77 * function returns.
78 *
79 * This call sleeps to guarantee that no CPU is looking at the packet
80 * type after return.
81 */
dev_remove_offload(struct packet_offload * po)82 void dev_remove_offload(struct packet_offload *po)
83 {
84 __dev_remove_offload(po);
85
86 synchronize_net();
87 }
88 EXPORT_SYMBOL(dev_remove_offload);
89
90
skb_gro_receive(struct sk_buff * p,struct sk_buff * skb)91 int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
92 {
93 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
94 unsigned int offset = skb_gro_offset(skb);
95 unsigned int headlen = skb_headlen(skb);
96 unsigned int len = skb_gro_len(skb);
97 unsigned int delta_truesize;
98 unsigned int new_truesize;
99 struct sk_buff *lp;
100 int segs;
101
102 /* Do not splice page pool based packets w/ non-page pool
103 * packets. This can result in reference count issues as page
104 * pool pages will not decrement the reference count and will
105 * instead be immediately returned to the pool or have frag
106 * count decremented.
107 */
108 if (p->pp_recycle != skb->pp_recycle)
109 return -ETOOMANYREFS;
110
111 if (unlikely(p->len + len >= netif_get_gro_max_size(p->dev, p) ||
112 NAPI_GRO_CB(skb)->flush))
113 return -E2BIG;
114
115 if (unlikely(p->len + len >= GRO_LEGACY_MAX_SIZE)) {
116 if (NAPI_GRO_CB(skb)->proto != IPPROTO_TCP ||
117 (p->protocol == htons(ETH_P_IPV6) &&
118 skb_headroom(p) < sizeof(struct hop_jumbo_hdr)) ||
119 p->encapsulation)
120 return -E2BIG;
121 }
122
123 segs = NAPI_GRO_CB(skb)->count;
124 lp = NAPI_GRO_CB(p)->last;
125 pinfo = skb_shinfo(lp);
126
127 if (headlen <= offset) {
128 skb_frag_t *frag;
129 skb_frag_t *frag2;
130 int i = skbinfo->nr_frags;
131 int nr_frags = pinfo->nr_frags + i;
132
133 if (nr_frags > MAX_SKB_FRAGS)
134 goto merge;
135
136 offset -= headlen;
137 pinfo->nr_frags = nr_frags;
138 skbinfo->nr_frags = 0;
139
140 frag = pinfo->frags + nr_frags;
141 frag2 = skbinfo->frags + i;
142 do {
143 *--frag = *--frag2;
144 } while (--i);
145
146 skb_frag_off_add(frag, offset);
147 skb_frag_size_sub(frag, offset);
148
149 /* all fragments truesize : remove (head size + sk_buff) */
150 new_truesize = SKB_TRUESIZE(skb_end_offset(skb));
151 delta_truesize = skb->truesize - new_truesize;
152
153 skb->truesize = new_truesize;
154 skb->len -= skb->data_len;
155 skb->data_len = 0;
156
157 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
158 goto done;
159 } else if (skb->head_frag) {
160 int nr_frags = pinfo->nr_frags;
161 skb_frag_t *frag = pinfo->frags + nr_frags;
162 struct page *page = virt_to_head_page(skb->head);
163 unsigned int first_size = headlen - offset;
164 unsigned int first_offset;
165
166 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
167 goto merge;
168
169 first_offset = skb->data -
170 (unsigned char *)page_address(page) +
171 offset;
172
173 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
174
175 skb_frag_fill_page_desc(frag, page, first_offset, first_size);
176
177 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
178 /* We dont need to clear skbinfo->nr_frags here */
179
180 new_truesize = SKB_DATA_ALIGN(sizeof(struct sk_buff));
181 delta_truesize = skb->truesize - new_truesize;
182 skb->truesize = new_truesize;
183 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
184 goto done;
185 }
186
187 merge:
188 /* sk ownership - if any - completely transferred to the aggregated packet */
189 skb->destructor = NULL;
190 skb->sk = NULL;
191 delta_truesize = skb->truesize;
192 if (offset > headlen) {
193 unsigned int eat = offset - headlen;
194
195 skb_frag_off_add(&skbinfo->frags[0], eat);
196 skb_frag_size_sub(&skbinfo->frags[0], eat);
197 skb->data_len -= eat;
198 skb->len -= eat;
199 offset = headlen;
200 }
201
202 __skb_pull(skb, offset);
203
204 if (NAPI_GRO_CB(p)->last == p)
205 skb_shinfo(p)->frag_list = skb;
206 else
207 NAPI_GRO_CB(p)->last->next = skb;
208 NAPI_GRO_CB(p)->last = skb;
209 __skb_header_release(skb);
210 lp = p;
211
212 done:
213 NAPI_GRO_CB(p)->count += segs;
214 p->data_len += len;
215 p->truesize += delta_truesize;
216 p->len += len;
217 if (lp != p) {
218 lp->data_len += len;
219 lp->truesize += delta_truesize;
220 lp->len += len;
221 }
222 NAPI_GRO_CB(skb)->same_flow = 1;
223 return 0;
224 }
225
skb_gro_receive_list(struct sk_buff * p,struct sk_buff * skb)226 int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
227 {
228 if (unlikely(p->len + skb->len >= 65536))
229 return -E2BIG;
230
231 if (NAPI_GRO_CB(p)->last == p)
232 skb_shinfo(p)->frag_list = skb;
233 else
234 NAPI_GRO_CB(p)->last->next = skb;
235
236 skb_pull(skb, skb_gro_offset(skb));
237
238 NAPI_GRO_CB(p)->last = skb;
239 NAPI_GRO_CB(p)->count++;
240 p->data_len += skb->len;
241
242 /* sk ownership - if any - completely transferred to the aggregated packet */
243 skb->destructor = NULL;
244 skb->sk = NULL;
245 p->truesize += skb->truesize;
246 p->len += skb->len;
247
248 NAPI_GRO_CB(skb)->same_flow = 1;
249
250 return 0;
251 }
252
gro_complete(struct gro_node * gro,struct sk_buff * skb)253 static void gro_complete(struct gro_node *gro, struct sk_buff *skb)
254 {
255 struct list_head *head = &net_hotdata.offload_base;
256 struct packet_offload *ptype;
257 __be16 type = skb->protocol;
258 int err = -ENOENT;
259
260 BUILD_BUG_ON(sizeof(struct napi_gro_cb) > sizeof(skb->cb));
261
262 if (NAPI_GRO_CB(skb)->count == 1) {
263 skb_shinfo(skb)->gso_size = 0;
264 goto out;
265 }
266
267 rcu_read_lock();
268 list_for_each_entry_rcu(ptype, head, list) {
269 if (ptype->type != type || !ptype->callbacks.gro_complete)
270 continue;
271
272 err = INDIRECT_CALL_INET(ptype->callbacks.gro_complete,
273 ipv6_gro_complete, inet_gro_complete,
274 skb, 0);
275 break;
276 }
277 rcu_read_unlock();
278
279 if (err) {
280 WARN_ON(&ptype->list == head);
281 kfree_skb(skb);
282 return;
283 }
284
285 out:
286 gro_normal_one(gro, skb, NAPI_GRO_CB(skb)->count);
287 }
288
__gro_flush_chain(struct gro_node * gro,u32 index,bool flush_old)289 static void __gro_flush_chain(struct gro_node *gro, u32 index, bool flush_old)
290 {
291 struct list_head *head = &gro->hash[index].list;
292 struct sk_buff *skb, *p;
293
294 list_for_each_entry_safe_reverse(skb, p, head, list) {
295 if (flush_old && NAPI_GRO_CB(skb)->age == jiffies)
296 return;
297 skb_list_del_init(skb);
298 gro_complete(gro, skb);
299 gro->hash[index].count--;
300 }
301
302 if (!gro->hash[index].count)
303 __clear_bit(index, &gro->bitmask);
304 }
305
306 /*
307 * gro->hash[].list contains packets ordered by age.
308 * youngest packets at the head of it.
309 * Complete skbs in reverse order to reduce latencies.
310 */
__gro_flush(struct gro_node * gro,bool flush_old)311 void __gro_flush(struct gro_node *gro, bool flush_old)
312 {
313 unsigned long bitmask = gro->bitmask;
314 unsigned int i, base = ~0U;
315
316 while ((i = ffs(bitmask)) != 0) {
317 bitmask >>= i;
318 base += i;
319 __gro_flush_chain(gro, base, flush_old);
320 }
321 }
322 EXPORT_SYMBOL(__gro_flush);
323
gro_list_prepare_tc_ext(const struct sk_buff * skb,const struct sk_buff * p,unsigned long diffs)324 static unsigned long gro_list_prepare_tc_ext(const struct sk_buff *skb,
325 const struct sk_buff *p,
326 unsigned long diffs)
327 {
328 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
329 struct tc_skb_ext *skb_ext;
330 struct tc_skb_ext *p_ext;
331
332 skb_ext = skb_ext_find(skb, TC_SKB_EXT);
333 p_ext = skb_ext_find(p, TC_SKB_EXT);
334
335 diffs |= (!!p_ext) ^ (!!skb_ext);
336 if (!diffs && unlikely(skb_ext))
337 diffs |= p_ext->chain ^ skb_ext->chain;
338 #endif
339 return diffs;
340 }
341
gro_list_prepare(const struct list_head * head,const struct sk_buff * skb)342 static void gro_list_prepare(const struct list_head *head,
343 const struct sk_buff *skb)
344 {
345 unsigned int maclen = skb->dev->hard_header_len;
346 u32 hash = skb_get_hash_raw(skb);
347 struct sk_buff *p;
348
349 list_for_each_entry(p, head, list) {
350 unsigned long diffs;
351
352 if (hash != skb_get_hash_raw(p)) {
353 NAPI_GRO_CB(p)->same_flow = 0;
354 continue;
355 }
356
357 diffs = (unsigned long)p->dev ^ (unsigned long)skb->dev;
358 diffs |= p->vlan_all ^ skb->vlan_all;
359 diffs |= skb_metadata_differs(p, skb);
360 if (maclen == ETH_HLEN)
361 diffs |= compare_ether_header(skb_mac_header(p),
362 skb_mac_header(skb));
363 else if (!diffs)
364 diffs = memcmp(skb_mac_header(p),
365 skb_mac_header(skb),
366 maclen);
367
368 /* in most common scenarios 'slow_gro' is 0
369 * otherwise we are already on some slower paths
370 * either skip all the infrequent tests altogether or
371 * avoid trying too hard to skip each of them individually
372 */
373 if (!diffs && unlikely(skb->slow_gro | p->slow_gro)) {
374 diffs |= p->sk != skb->sk;
375 diffs |= skb_metadata_dst_cmp(p, skb);
376 diffs |= skb_get_nfct(p) ^ skb_get_nfct(skb);
377
378 diffs |= gro_list_prepare_tc_ext(skb, p, diffs);
379 }
380
381 NAPI_GRO_CB(p)->same_flow = !diffs;
382 }
383 }
384
skb_gro_reset_offset(struct sk_buff * skb,u32 nhoff)385 static inline void skb_gro_reset_offset(struct sk_buff *skb, u32 nhoff)
386 {
387 const struct skb_shared_info *pinfo;
388 const skb_frag_t *frag0;
389 unsigned int headlen;
390
391 NAPI_GRO_CB(skb)->network_offset = 0;
392 NAPI_GRO_CB(skb)->data_offset = 0;
393 headlen = skb_headlen(skb);
394 NAPI_GRO_CB(skb)->frag0 = skb->data;
395 NAPI_GRO_CB(skb)->frag0_len = headlen;
396 if (headlen)
397 return;
398
399 pinfo = skb_shinfo(skb);
400 frag0 = &pinfo->frags[0];
401
402 if (pinfo->nr_frags && skb_frag_page(frag0) &&
403 !PageHighMem(skb_frag_page(frag0)) &&
404 (!NET_IP_ALIGN || !((skb_frag_off(frag0) + nhoff) & 3))) {
405 NAPI_GRO_CB(skb)->frag0 = skb_frag_address(frag0);
406 NAPI_GRO_CB(skb)->frag0_len = min_t(unsigned int,
407 skb_frag_size(frag0),
408 skb->end - skb->tail);
409 }
410 }
411
gro_pull_from_frag0(struct sk_buff * skb,int grow)412 static void gro_pull_from_frag0(struct sk_buff *skb, int grow)
413 {
414 struct skb_shared_info *pinfo = skb_shinfo(skb);
415
416 BUG_ON(skb->end - skb->tail < grow);
417
418 memcpy(skb_tail_pointer(skb), NAPI_GRO_CB(skb)->frag0, grow);
419
420 skb->data_len -= grow;
421 skb->tail += grow;
422
423 skb_frag_off_add(&pinfo->frags[0], grow);
424 skb_frag_size_sub(&pinfo->frags[0], grow);
425
426 if (unlikely(!skb_frag_size(&pinfo->frags[0]))) {
427 skb_frag_unref(skb, 0);
428 memmove(pinfo->frags, pinfo->frags + 1,
429 --pinfo->nr_frags * sizeof(pinfo->frags[0]));
430 }
431 }
432
gro_try_pull_from_frag0(struct sk_buff * skb)433 static void gro_try_pull_from_frag0(struct sk_buff *skb)
434 {
435 int grow = skb_gro_offset(skb) - skb_headlen(skb);
436
437 if (grow > 0)
438 gro_pull_from_frag0(skb, grow);
439 }
440
gro_flush_oldest(struct gro_node * gro,struct list_head * head)441 static void gro_flush_oldest(struct gro_node *gro, struct list_head *head)
442 {
443 struct sk_buff *oldest;
444
445 oldest = list_last_entry(head, struct sk_buff, list);
446
447 /* We are called with head length >= MAX_GRO_SKBS, so this is
448 * impossible.
449 */
450 if (WARN_ON_ONCE(!oldest))
451 return;
452
453 /* Do not adjust napi->gro_hash[].count, caller is adding a new
454 * SKB to the chain.
455 */
456 skb_list_del_init(oldest);
457 gro_complete(gro, oldest);
458 }
459
dev_gro_receive(struct gro_node * gro,struct sk_buff * skb)460 static enum gro_result dev_gro_receive(struct gro_node *gro,
461 struct sk_buff *skb)
462 {
463 u32 bucket = skb_get_hash_raw(skb) & (GRO_HASH_BUCKETS - 1);
464 struct list_head *head = &net_hotdata.offload_base;
465 struct gro_list *gro_list = &gro->hash[bucket];
466 struct packet_offload *ptype;
467 __be16 type = skb->protocol;
468 struct sk_buff *pp = NULL;
469 enum gro_result ret;
470 int same_flow;
471
472 if (netif_elide_gro(skb->dev))
473 goto normal;
474
475 gro_list_prepare(&gro_list->list, skb);
476
477 rcu_read_lock();
478 list_for_each_entry_rcu(ptype, head, list) {
479 if (ptype->type == type && ptype->callbacks.gro_receive)
480 goto found_ptype;
481 }
482 rcu_read_unlock();
483 goto normal;
484
485 found_ptype:
486 skb_set_network_header(skb, skb_gro_offset(skb));
487 skb_reset_mac_len(skb);
488 BUILD_BUG_ON(sizeof_field(struct napi_gro_cb, zeroed) != sizeof(u32));
489 BUILD_BUG_ON(!IS_ALIGNED(offsetof(struct napi_gro_cb, zeroed),
490 sizeof(u32))); /* Avoid slow unaligned acc */
491 *(u32 *)&NAPI_GRO_CB(skb)->zeroed = 0;
492 NAPI_GRO_CB(skb)->flush = skb_has_frag_list(skb);
493 NAPI_GRO_CB(skb)->count = 1;
494 if (unlikely(skb_is_gso(skb))) {
495 NAPI_GRO_CB(skb)->count = skb_shinfo(skb)->gso_segs;
496 /* Only support TCP and non DODGY users. */
497 if (!skb_is_gso_tcp(skb) ||
498 (skb_shinfo(skb)->gso_type & SKB_GSO_DODGY))
499 NAPI_GRO_CB(skb)->flush = 1;
500 }
501
502 /* Setup for GRO checksum validation */
503 switch (skb->ip_summed) {
504 case CHECKSUM_COMPLETE:
505 NAPI_GRO_CB(skb)->csum = skb->csum;
506 NAPI_GRO_CB(skb)->csum_valid = 1;
507 break;
508 case CHECKSUM_UNNECESSARY:
509 NAPI_GRO_CB(skb)->csum_cnt = skb->csum_level + 1;
510 break;
511 }
512
513 pp = INDIRECT_CALL_INET(ptype->callbacks.gro_receive,
514 ipv6_gro_receive, inet_gro_receive,
515 &gro_list->list, skb);
516
517 rcu_read_unlock();
518
519 if (PTR_ERR(pp) == -EINPROGRESS) {
520 ret = GRO_CONSUMED;
521 goto ok;
522 }
523
524 same_flow = NAPI_GRO_CB(skb)->same_flow;
525 ret = NAPI_GRO_CB(skb)->free ? GRO_MERGED_FREE : GRO_MERGED;
526
527 if (pp) {
528 skb_list_del_init(pp);
529 gro_complete(gro, pp);
530 gro_list->count--;
531 }
532
533 if (same_flow)
534 goto ok;
535
536 if (NAPI_GRO_CB(skb)->flush)
537 goto normal;
538
539 if (unlikely(gro_list->count >= MAX_GRO_SKBS))
540 gro_flush_oldest(gro, &gro_list->list);
541 else
542 gro_list->count++;
543
544 /* Must be called before setting NAPI_GRO_CB(skb)->{age|last} */
545 gro_try_pull_from_frag0(skb);
546 NAPI_GRO_CB(skb)->age = jiffies;
547 NAPI_GRO_CB(skb)->last = skb;
548 if (!skb_is_gso(skb))
549 skb_shinfo(skb)->gso_size = skb_gro_len(skb);
550 list_add(&skb->list, &gro_list->list);
551 ret = GRO_HELD;
552 ok:
553 if (gro_list->count) {
554 if (!test_bit(bucket, &gro->bitmask))
555 __set_bit(bucket, &gro->bitmask);
556 } else if (test_bit(bucket, &gro->bitmask)) {
557 __clear_bit(bucket, &gro->bitmask);
558 }
559
560 return ret;
561
562 normal:
563 ret = GRO_NORMAL;
564 gro_try_pull_from_frag0(skb);
565 goto ok;
566 }
567
gro_find_receive_by_type(__be16 type)568 struct packet_offload *gro_find_receive_by_type(__be16 type)
569 {
570 struct list_head *offload_head = &net_hotdata.offload_base;
571 struct packet_offload *ptype;
572
573 list_for_each_entry_rcu(ptype, offload_head, list) {
574 if (ptype->type != type || !ptype->callbacks.gro_receive)
575 continue;
576 return ptype;
577 }
578 return NULL;
579 }
580 EXPORT_SYMBOL(gro_find_receive_by_type);
581
gro_find_complete_by_type(__be16 type)582 struct packet_offload *gro_find_complete_by_type(__be16 type)
583 {
584 struct list_head *offload_head = &net_hotdata.offload_base;
585 struct packet_offload *ptype;
586
587 list_for_each_entry_rcu(ptype, offload_head, list) {
588 if (ptype->type != type || !ptype->callbacks.gro_complete)
589 continue;
590 return ptype;
591 }
592 return NULL;
593 }
594 EXPORT_SYMBOL(gro_find_complete_by_type);
595
gro_skb_finish(struct gro_node * gro,struct sk_buff * skb,gro_result_t ret)596 static gro_result_t gro_skb_finish(struct gro_node *gro, struct sk_buff *skb,
597 gro_result_t ret)
598 {
599 switch (ret) {
600 case GRO_NORMAL:
601 gro_normal_one(gro, skb, 1);
602 break;
603
604 case GRO_MERGED_FREE:
605 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
606 napi_skb_free_stolen_head(skb);
607 else if (skb->fclone != SKB_FCLONE_UNAVAILABLE)
608 __kfree_skb(skb);
609 else
610 __napi_kfree_skb(skb, SKB_CONSUMED);
611 break;
612
613 case GRO_HELD:
614 case GRO_MERGED:
615 case GRO_CONSUMED:
616 break;
617 }
618
619 return ret;
620 }
621
gro_receive_skb(struct gro_node * gro,struct sk_buff * skb)622 gro_result_t gro_receive_skb(struct gro_node *gro, struct sk_buff *skb)
623 {
624 gro_result_t ret;
625
626 __skb_mark_napi_id(skb, gro);
627 trace_napi_gro_receive_entry(skb);
628
629 skb_gro_reset_offset(skb, 0);
630
631 ret = gro_skb_finish(gro, skb, dev_gro_receive(gro, skb));
632 trace_napi_gro_receive_exit(ret);
633
634 return ret;
635 }
636 EXPORT_SYMBOL(gro_receive_skb);
637
napi_reuse_skb(struct napi_struct * napi,struct sk_buff * skb)638 static void napi_reuse_skb(struct napi_struct *napi, struct sk_buff *skb)
639 {
640 if (unlikely(skb->pfmemalloc)) {
641 consume_skb(skb);
642 return;
643 }
644 __skb_pull(skb, skb_headlen(skb));
645 /* restore the reserve we had after netdev_alloc_skb_ip_align() */
646 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN - skb_headroom(skb));
647 __vlan_hwaccel_clear_tag(skb);
648 skb->dev = napi->dev;
649 skb->skb_iif = 0;
650
651 /* eth_type_trans() assumes pkt_type is PACKET_HOST */
652 skb->pkt_type = PACKET_HOST;
653
654 skb->encapsulation = 0;
655 skb->ip_summed = CHECKSUM_NONE;
656 skb_shinfo(skb)->gso_type = 0;
657 skb_shinfo(skb)->gso_size = 0;
658 if (unlikely(skb->slow_gro)) {
659 skb_orphan(skb);
660 skb_ext_reset(skb);
661 nf_reset_ct(skb);
662 skb->slow_gro = 0;
663 }
664
665 napi->skb = skb;
666 }
667
napi_get_frags(struct napi_struct * napi)668 struct sk_buff *napi_get_frags(struct napi_struct *napi)
669 {
670 struct sk_buff *skb = napi->skb;
671
672 if (!skb) {
673 skb = napi_alloc_skb(napi, GRO_MAX_HEAD);
674 if (skb) {
675 napi->skb = skb;
676 skb_mark_napi_id(skb, napi);
677 }
678 }
679 return skb;
680 }
681 EXPORT_SYMBOL(napi_get_frags);
682
napi_frags_finish(struct napi_struct * napi,struct sk_buff * skb,gro_result_t ret)683 static gro_result_t napi_frags_finish(struct napi_struct *napi,
684 struct sk_buff *skb,
685 gro_result_t ret)
686 {
687 switch (ret) {
688 case GRO_NORMAL:
689 case GRO_HELD:
690 __skb_push(skb, ETH_HLEN);
691 skb->protocol = eth_type_trans(skb, skb->dev);
692 if (ret == GRO_NORMAL)
693 gro_normal_one(&napi->gro, skb, 1);
694 break;
695
696 case GRO_MERGED_FREE:
697 if (NAPI_GRO_CB(skb)->free == NAPI_GRO_FREE_STOLEN_HEAD)
698 napi_skb_free_stolen_head(skb);
699 else
700 napi_reuse_skb(napi, skb);
701 break;
702
703 case GRO_MERGED:
704 case GRO_CONSUMED:
705 break;
706 }
707
708 return ret;
709 }
710
711 /* Upper GRO stack assumes network header starts at gro_offset=0
712 * Drivers could call both napi_gro_frags() and napi_gro_receive()
713 * We copy ethernet header into skb->data to have a common layout.
714 */
napi_frags_skb(struct napi_struct * napi)715 static struct sk_buff *napi_frags_skb(struct napi_struct *napi)
716 {
717 struct sk_buff *skb = napi->skb;
718 const struct ethhdr *eth;
719 unsigned int hlen = sizeof(*eth);
720
721 napi->skb = NULL;
722
723 skb_reset_mac_header(skb);
724 skb_gro_reset_offset(skb, hlen);
725
726 if (unlikely(!skb_gro_may_pull(skb, hlen))) {
727 eth = skb_gro_header_slow(skb, hlen, 0);
728 if (unlikely(!eth)) {
729 net_warn_ratelimited("%s: dropping impossible skb from %s\n",
730 __func__, napi->dev->name);
731 napi_reuse_skb(napi, skb);
732 return NULL;
733 }
734 } else {
735 eth = (const struct ethhdr *)skb->data;
736
737 if (NAPI_GRO_CB(skb)->frag0 != skb->data)
738 gro_pull_from_frag0(skb, hlen);
739
740 NAPI_GRO_CB(skb)->frag0 += hlen;
741 NAPI_GRO_CB(skb)->frag0_len -= hlen;
742 }
743 __skb_pull(skb, hlen);
744
745 /*
746 * This works because the only protocols we care about don't require
747 * special handling.
748 * We'll fix it up properly in napi_frags_finish()
749 */
750 skb->protocol = eth->h_proto;
751
752 return skb;
753 }
754
napi_gro_frags(struct napi_struct * napi)755 gro_result_t napi_gro_frags(struct napi_struct *napi)
756 {
757 gro_result_t ret;
758 struct sk_buff *skb = napi_frags_skb(napi);
759
760 trace_napi_gro_frags_entry(skb);
761
762 ret = napi_frags_finish(napi, skb, dev_gro_receive(&napi->gro, skb));
763 trace_napi_gro_frags_exit(ret);
764
765 return ret;
766 }
767 EXPORT_SYMBOL(napi_gro_frags);
768
769 /* Compute the checksum from gro_offset and return the folded value
770 * after adding in any pseudo checksum.
771 */
__skb_gro_checksum_complete(struct sk_buff * skb)772 __sum16 __skb_gro_checksum_complete(struct sk_buff *skb)
773 {
774 __wsum wsum;
775 __sum16 sum;
776
777 wsum = skb_checksum(skb, skb_gro_offset(skb), skb_gro_len(skb), 0);
778
779 /* NAPI_GRO_CB(skb)->csum holds pseudo checksum */
780 sum = csum_fold(csum_add(NAPI_GRO_CB(skb)->csum, wsum));
781 /* See comments in __skb_checksum_complete(). */
782 if (likely(!sum)) {
783 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
784 !skb->csum_complete_sw)
785 netdev_rx_csum_fault(skb->dev, skb);
786 }
787
788 NAPI_GRO_CB(skb)->csum = wsum;
789 NAPI_GRO_CB(skb)->csum_valid = 1;
790
791 return sum;
792 }
793 EXPORT_SYMBOL(__skb_gro_checksum_complete);
794
gro_init(struct gro_node * gro)795 void gro_init(struct gro_node *gro)
796 {
797 for (u32 i = 0; i < GRO_HASH_BUCKETS; i++) {
798 INIT_LIST_HEAD(&gro->hash[i].list);
799 gro->hash[i].count = 0;
800 }
801
802 gro->bitmask = 0;
803 gro->cached_napi_id = 0;
804
805 INIT_LIST_HEAD(&gro->rx_list);
806 gro->rx_count = 0;
807 }
808
gro_cleanup(struct gro_node * gro)809 void gro_cleanup(struct gro_node *gro)
810 {
811 struct sk_buff *skb, *n;
812
813 for (u32 i = 0; i < GRO_HASH_BUCKETS; i++) {
814 list_for_each_entry_safe(skb, n, &gro->hash[i].list, list)
815 kfree_skb(skb);
816
817 gro->hash[i].count = 0;
818 }
819
820 gro->bitmask = 0;
821 gro->cached_napi_id = 0;
822
823 list_for_each_entry_safe(skb, n, &gro->rx_list, list)
824 kfree_skb(skb);
825
826 gro->rx_count = 0;
827 }
828