1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
36
37#include <linux/module.h>
38#include <linux/types.h>
39#include <linux/kernel.h>
40#include <linux/mm.h>
41#include <linux/interrupt.h>
42#include <linux/in.h>
43#include <linux/inet.h>
44#include <linux/slab.h>
45#include <linux/tcp.h>
46#include <linux/udp.h>
47#include <linux/sctp.h>
48#include <linux/netdevice.h>
49#ifdef CONFIG_NET_CLS_ACT
50#include <net/pkt_sched.h>
51#endif
52#include <linux/string.h>
53#include <linux/skbuff.h>
54#include <linux/splice.h>
55#include <linux/cache.h>
56#include <linux/rtnetlink.h>
57#include <linux/init.h>
58#include <linux/scatterlist.h>
59#include <linux/errqueue.h>
60#include <linux/prefetch.h>
61#include <linux/if_vlan.h>
62#include <linux/mpls.h>
63
64#include <net/protocol.h>
65#include <net/dst.h>
66#include <net/sock.h>
67#include <net/checksum.h>
68#include <net/ip6_checksum.h>
69#include <net/xfrm.h>
70#include <net/mpls.h>
71#include <net/mptcp.h>
72
73#include <linux/uaccess.h>
74#include <trace/events/skb.h>
75#include <linux/highmem.h>
76#include <linux/capability.h>
77#include <linux/user_namespace.h>
78#include <linux/indirect_call_wrapper.h>
79
80#include "datagram.h"
81
82struct kmem_cache *skbuff_head_cache __ro_after_init;
83static struct kmem_cache *skbuff_fclone_cache __ro_after_init;
84#ifdef CONFIG_SKB_EXTENSIONS
85static struct kmem_cache *skbuff_ext_cache __ro_after_init;
86#endif
87int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
88EXPORT_SYMBOL(sysctl_max_skb_frags);
89
90
91
92
93
94
95
96
97
98
99
100
101
102static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
103 const char msg[])
104{
105 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
106 msg, addr, skb->len, sz, skb->head, skb->data,
107 (unsigned long)skb->tail, (unsigned long)skb->end,
108 skb->dev ? skb->dev->name : "<NULL>");
109 BUG();
110}
111
112static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
113{
114 skb_panic(skb, sz, addr, __func__);
115}
116
117static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
118{
119 skb_panic(skb, sz, addr, __func__);
120}
121
122
123
124
125
126
127
128
129#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
130 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
131
132static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
133 unsigned long ip, bool *pfmemalloc)
134{
135 void *obj;
136 bool ret_pfmemalloc = false;
137
138
139
140
141
142 obj = kmalloc_node_track_caller(size,
143 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
144 node);
145 if (obj || !(gfp_pfmemalloc_allowed(flags)))
146 goto out;
147
148
149 ret_pfmemalloc = true;
150 obj = kmalloc_node_track_caller(size, flags, node);
151
152out:
153 if (pfmemalloc)
154 *pfmemalloc = ret_pfmemalloc;
155
156 return obj;
157}
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
183 int flags, int node)
184{
185 struct kmem_cache *cache;
186 struct skb_shared_info *shinfo;
187 struct sk_buff *skb;
188 u8 *data;
189 bool pfmemalloc;
190
191 cache = (flags & SKB_ALLOC_FCLONE)
192 ? skbuff_fclone_cache : skbuff_head_cache;
193
194 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
195 gfp_mask |= __GFP_MEMALLOC;
196
197
198 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
199 if (!skb)
200 goto out;
201 prefetchw(skb);
202
203
204
205
206
207
208 size = SKB_DATA_ALIGN(size);
209 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
210 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
211 if (!data)
212 goto nodata;
213
214
215
216
217 size = SKB_WITH_OVERHEAD(ksize(data));
218 prefetchw(data + size);
219
220
221
222
223
224
225 memset(skb, 0, offsetof(struct sk_buff, tail));
226
227 skb->truesize = SKB_TRUESIZE(size);
228 skb->pfmemalloc = pfmemalloc;
229 refcount_set(&skb->users, 1);
230 skb->head = data;
231 skb->data = data;
232 skb_reset_tail_pointer(skb);
233 skb->end = skb->tail + size;
234 skb->mac_header = (typeof(skb->mac_header))~0U;
235 skb->transport_header = (typeof(skb->transport_header))~0U;
236
237
238 shinfo = skb_shinfo(skb);
239 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
240 atomic_set(&shinfo->dataref, 1);
241
242 if (flags & SKB_ALLOC_FCLONE) {
243 struct sk_buff_fclones *fclones;
244
245 fclones = container_of(skb, struct sk_buff_fclones, skb1);
246
247 skb->fclone = SKB_FCLONE_ORIG;
248 refcount_set(&fclones->fclone_ref, 1);
249
250 fclones->skb2.fclone = SKB_FCLONE_CLONE;
251 }
252out:
253 return skb;
254nodata:
255 kmem_cache_free(cache, skb);
256 skb = NULL;
257 goto out;
258}
259EXPORT_SYMBOL(__alloc_skb);
260
261
262static struct sk_buff *__build_skb_around(struct sk_buff *skb,
263 void *data, unsigned int frag_size)
264{
265 struct skb_shared_info *shinfo;
266 unsigned int size = frag_size ? : ksize(data);
267
268 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
269
270
271 skb->truesize = SKB_TRUESIZE(size);
272 refcount_set(&skb->users, 1);
273 skb->head = data;
274 skb->data = data;
275 skb_reset_tail_pointer(skb);
276 skb->end = skb->tail + size;
277 skb->mac_header = (typeof(skb->mac_header))~0U;
278 skb->transport_header = (typeof(skb->transport_header))~0U;
279
280
281 shinfo = skb_shinfo(skb);
282 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
283 atomic_set(&shinfo->dataref, 1);
284
285 return skb;
286}
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307struct sk_buff *__build_skb(void *data, unsigned int frag_size)
308{
309 struct sk_buff *skb;
310
311 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
312 if (unlikely(!skb))
313 return NULL;
314
315 memset(skb, 0, offsetof(struct sk_buff, tail));
316
317 return __build_skb_around(skb, data, frag_size);
318}
319
320
321
322
323
324
325struct sk_buff *build_skb(void *data, unsigned int frag_size)
326{
327 struct sk_buff *skb = __build_skb(data, frag_size);
328
329 if (skb && frag_size) {
330 skb->head_frag = 1;
331 if (page_is_pfmemalloc(virt_to_head_page(data)))
332 skb->pfmemalloc = 1;
333 }
334 return skb;
335}
336EXPORT_SYMBOL(build_skb);
337
338
339
340
341
342
343
344struct sk_buff *build_skb_around(struct sk_buff *skb,
345 void *data, unsigned int frag_size)
346{
347 if (unlikely(!skb))
348 return NULL;
349
350 skb = __build_skb_around(skb, data, frag_size);
351
352 if (skb && frag_size) {
353 skb->head_frag = 1;
354 if (page_is_pfmemalloc(virt_to_head_page(data)))
355 skb->pfmemalloc = 1;
356 }
357 return skb;
358}
359EXPORT_SYMBOL(build_skb_around);
360
361#define NAPI_SKB_CACHE_SIZE 64
362
363struct napi_alloc_cache {
364 struct page_frag_cache page;
365 unsigned int skb_count;
366 void *skb_cache[NAPI_SKB_CACHE_SIZE];
367};
368
369static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
370static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
371
372static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
373{
374 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
375
376 return page_frag_alloc(&nc->page, fragsz, gfp_mask);
377}
378
379void *napi_alloc_frag(unsigned int fragsz)
380{
381 fragsz = SKB_DATA_ALIGN(fragsz);
382
383 return __napi_alloc_frag(fragsz, GFP_ATOMIC);
384}
385EXPORT_SYMBOL(napi_alloc_frag);
386
387
388
389
390
391
392
393
394void *netdev_alloc_frag(unsigned int fragsz)
395{
396 struct page_frag_cache *nc;
397 void *data;
398
399 fragsz = SKB_DATA_ALIGN(fragsz);
400 if (in_irq() || irqs_disabled()) {
401 nc = this_cpu_ptr(&netdev_alloc_cache);
402 data = page_frag_alloc(nc, fragsz, GFP_ATOMIC);
403 } else {
404 local_bh_disable();
405 data = __napi_alloc_frag(fragsz, GFP_ATOMIC);
406 local_bh_enable();
407 }
408 return data;
409}
410EXPORT_SYMBOL(netdev_alloc_frag);
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
426 gfp_t gfp_mask)
427{
428 struct page_frag_cache *nc;
429 struct sk_buff *skb;
430 bool pfmemalloc;
431 void *data;
432
433 len += NET_SKB_PAD;
434
435 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
436 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
437 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
438 if (!skb)
439 goto skb_fail;
440 goto skb_success;
441 }
442
443 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
444 len = SKB_DATA_ALIGN(len);
445
446 if (sk_memalloc_socks())
447 gfp_mask |= __GFP_MEMALLOC;
448
449 if (in_irq() || irqs_disabled()) {
450 nc = this_cpu_ptr(&netdev_alloc_cache);
451 data = page_frag_alloc(nc, len, gfp_mask);
452 pfmemalloc = nc->pfmemalloc;
453 } else {
454 local_bh_disable();
455 nc = this_cpu_ptr(&napi_alloc_cache.page);
456 data = page_frag_alloc(nc, len, gfp_mask);
457 pfmemalloc = nc->pfmemalloc;
458 local_bh_enable();
459 }
460
461 if (unlikely(!data))
462 return NULL;
463
464 skb = __build_skb(data, len);
465 if (unlikely(!skb)) {
466 skb_free_frag(data);
467 return NULL;
468 }
469
470 if (pfmemalloc)
471 skb->pfmemalloc = 1;
472 skb->head_frag = 1;
473
474skb_success:
475 skb_reserve(skb, NET_SKB_PAD);
476 skb->dev = dev;
477
478skb_fail:
479 return skb;
480}
481EXPORT_SYMBOL(__netdev_alloc_skb);
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
497 gfp_t gfp_mask)
498{
499 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
500 struct sk_buff *skb;
501 void *data;
502
503 len += NET_SKB_PAD + NET_IP_ALIGN;
504
505 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
506 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
507 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
508 if (!skb)
509 goto skb_fail;
510 goto skb_success;
511 }
512
513 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
514 len = SKB_DATA_ALIGN(len);
515
516 if (sk_memalloc_socks())
517 gfp_mask |= __GFP_MEMALLOC;
518
519 data = page_frag_alloc(&nc->page, len, gfp_mask);
520 if (unlikely(!data))
521 return NULL;
522
523 skb = __build_skb(data, len);
524 if (unlikely(!skb)) {
525 skb_free_frag(data);
526 return NULL;
527 }
528
529 if (nc->page.pfmemalloc)
530 skb->pfmemalloc = 1;
531 skb->head_frag = 1;
532
533skb_success:
534 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
535 skb->dev = napi->dev;
536
537skb_fail:
538 return skb;
539}
540EXPORT_SYMBOL(__napi_alloc_skb);
541
542void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
543 int size, unsigned int truesize)
544{
545 skb_fill_page_desc(skb, i, page, off, size);
546 skb->len += size;
547 skb->data_len += size;
548 skb->truesize += truesize;
549}
550EXPORT_SYMBOL(skb_add_rx_frag);
551
552void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
553 unsigned int truesize)
554{
555 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
556
557 skb_frag_size_add(frag, size);
558 skb->len += size;
559 skb->data_len += size;
560 skb->truesize += truesize;
561}
562EXPORT_SYMBOL(skb_coalesce_rx_frag);
563
564static void skb_drop_list(struct sk_buff **listp)
565{
566 kfree_skb_list(*listp);
567 *listp = NULL;
568}
569
570static inline void skb_drop_fraglist(struct sk_buff *skb)
571{
572 skb_drop_list(&skb_shinfo(skb)->frag_list);
573}
574
575static void skb_clone_fraglist(struct sk_buff *skb)
576{
577 struct sk_buff *list;
578
579 skb_walk_frags(skb, list)
580 skb_get(list);
581}
582
583static void skb_free_head(struct sk_buff *skb)
584{
585 unsigned char *head = skb->head;
586
587 if (skb->head_frag)
588 skb_free_frag(head);
589 else
590 kfree(head);
591}
592
593static void skb_release_data(struct sk_buff *skb)
594{
595 struct skb_shared_info *shinfo = skb_shinfo(skb);
596 int i;
597
598 if (skb->cloned &&
599 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
600 &shinfo->dataref))
601 return;
602
603 for (i = 0; i < shinfo->nr_frags; i++)
604 __skb_frag_unref(&shinfo->frags[i]);
605
606 if (shinfo->frag_list)
607 kfree_skb_list(shinfo->frag_list);
608
609 skb_zcopy_clear(skb, true);
610 skb_free_head(skb);
611}
612
613
614
615
616static void kfree_skbmem(struct sk_buff *skb)
617{
618 struct sk_buff_fclones *fclones;
619
620 switch (skb->fclone) {
621 case SKB_FCLONE_UNAVAILABLE:
622 kmem_cache_free(skbuff_head_cache, skb);
623 return;
624
625 case SKB_FCLONE_ORIG:
626 fclones = container_of(skb, struct sk_buff_fclones, skb1);
627
628
629
630
631
632 if (refcount_read(&fclones->fclone_ref) == 1)
633 goto fastpath;
634 break;
635
636 default:
637 fclones = container_of(skb, struct sk_buff_fclones, skb2);
638 break;
639 }
640 if (!refcount_dec_and_test(&fclones->fclone_ref))
641 return;
642fastpath:
643 kmem_cache_free(skbuff_fclone_cache, fclones);
644}
645
646void skb_release_head_state(struct sk_buff *skb)
647{
648 skb_dst_drop(skb);
649 if (skb->destructor) {
650 WARN_ON(in_irq());
651 skb->destructor(skb);
652 }
653#if IS_ENABLED(CONFIG_NF_CONNTRACK)
654 nf_conntrack_put(skb_nfct(skb));
655#endif
656 skb_ext_put(skb);
657}
658
659
660static void skb_release_all(struct sk_buff *skb)
661{
662 skb_release_head_state(skb);
663 if (likely(skb->head))
664 skb_release_data(skb);
665}
666
667
668
669
670
671
672
673
674
675
676void __kfree_skb(struct sk_buff *skb)
677{
678 skb_release_all(skb);
679 kfree_skbmem(skb);
680}
681EXPORT_SYMBOL(__kfree_skb);
682
683
684
685
686
687
688
689
690void kfree_skb(struct sk_buff *skb)
691{
692 if (!skb_unref(skb))
693 return;
694
695 trace_kfree_skb(skb, __builtin_return_address(0));
696 __kfree_skb(skb);
697}
698EXPORT_SYMBOL(kfree_skb);
699
700void kfree_skb_list(struct sk_buff *segs)
701{
702 while (segs) {
703 struct sk_buff *next = segs->next;
704
705 kfree_skb(segs);
706 segs = next;
707 }
708}
709EXPORT_SYMBOL(kfree_skb_list);
710
711
712
713
714
715
716
717void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt)
718{
719 static atomic_t can_dump_full = ATOMIC_INIT(5);
720 struct skb_shared_info *sh = skb_shinfo(skb);
721 struct net_device *dev = skb->dev;
722 struct sock *sk = skb->sk;
723 struct sk_buff *list_skb;
724 bool has_mac, has_trans;
725 int headroom, tailroom;
726 int i, len, seg_len;
727
728 if (full_pkt)
729 full_pkt = atomic_dec_if_positive(&can_dump_full) >= 0;
730
731 if (full_pkt)
732 len = skb->len;
733 else
734 len = min_t(int, skb->len, MAX_HEADER + 128);
735
736 headroom = skb_headroom(skb);
737 tailroom = skb_tailroom(skb);
738
739 has_mac = skb_mac_header_was_set(skb);
740 has_trans = skb_transport_header_was_set(skb);
741
742 printk("%sskb len=%u headroom=%u headlen=%u tailroom=%u\n"
743 "mac=(%d,%d) net=(%d,%d) trans=%d\n"
744 "shinfo(txflags=%u nr_frags=%u gso(size=%hu type=%u segs=%hu))\n"
745 "csum(0x%x ip_summed=%u complete_sw=%u valid=%u level=%u)\n"
746 "hash(0x%x sw=%u l4=%u) proto=0x%04x pkttype=%u iif=%d\n",
747 level, skb->len, headroom, skb_headlen(skb), tailroom,
748 has_mac ? skb->mac_header : -1,
749 has_mac ? skb_mac_header_len(skb) : -1,
750 skb->network_header,
751 has_trans ? skb_network_header_len(skb) : -1,
752 has_trans ? skb->transport_header : -1,
753 sh->tx_flags, sh->nr_frags,
754 sh->gso_size, sh->gso_type, sh->gso_segs,
755 skb->csum, skb->ip_summed, skb->csum_complete_sw,
756 skb->csum_valid, skb->csum_level,
757 skb->hash, skb->sw_hash, skb->l4_hash,
758 ntohs(skb->protocol), skb->pkt_type, skb->skb_iif);
759
760 if (dev)
761 printk("%sdev name=%s feat=0x%pNF\n",
762 level, dev->name, &dev->features);
763 if (sk)
764 printk("%ssk family=%hu type=%u proto=%u\n",
765 level, sk->sk_family, sk->sk_type, sk->sk_protocol);
766
767 if (full_pkt && headroom)
768 print_hex_dump(level, "skb headroom: ", DUMP_PREFIX_OFFSET,
769 16, 1, skb->head, headroom, false);
770
771 seg_len = min_t(int, skb_headlen(skb), len);
772 if (seg_len)
773 print_hex_dump(level, "skb linear: ", DUMP_PREFIX_OFFSET,
774 16, 1, skb->data, seg_len, false);
775 len -= seg_len;
776
777 if (full_pkt && tailroom)
778 print_hex_dump(level, "skb tailroom: ", DUMP_PREFIX_OFFSET,
779 16, 1, skb_tail_pointer(skb), tailroom, false);
780
781 for (i = 0; len && i < skb_shinfo(skb)->nr_frags; i++) {
782 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
783 u32 p_off, p_len, copied;
784 struct page *p;
785 u8 *vaddr;
786
787 skb_frag_foreach_page(frag, skb_frag_off(frag),
788 skb_frag_size(frag), p, p_off, p_len,
789 copied) {
790 seg_len = min_t(int, p_len, len);
791 vaddr = kmap_atomic(p);
792 print_hex_dump(level, "skb frag: ",
793 DUMP_PREFIX_OFFSET,
794 16, 1, vaddr + p_off, seg_len, false);
795 kunmap_atomic(vaddr);
796 len -= seg_len;
797 if (!len)
798 break;
799 }
800 }
801
802 if (full_pkt && skb_has_frag_list(skb)) {
803 printk("skb fraglist:\n");
804 skb_walk_frags(skb, list_skb)
805 skb_dump(level, list_skb, true);
806 }
807}
808EXPORT_SYMBOL(skb_dump);
809
810
811
812
813
814
815
816
817void skb_tx_error(struct sk_buff *skb)
818{
819 skb_zcopy_clear(skb, true);
820}
821EXPORT_SYMBOL(skb_tx_error);
822
823
824
825
826
827
828
829
830
831void consume_skb(struct sk_buff *skb)
832{
833 if (!skb_unref(skb))
834 return;
835
836 trace_consume_skb(skb);
837 __kfree_skb(skb);
838}
839EXPORT_SYMBOL(consume_skb);
840
841
842
843
844
845
846
847
848void __consume_stateless_skb(struct sk_buff *skb)
849{
850 trace_consume_skb(skb);
851 skb_release_data(skb);
852 kfree_skbmem(skb);
853}
854
855void __kfree_skb_flush(void)
856{
857 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
858
859
860 if (nc->skb_count) {
861 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
862 nc->skb_cache);
863 nc->skb_count = 0;
864 }
865}
866
867static inline void _kfree_skb_defer(struct sk_buff *skb)
868{
869 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
870
871
872 skb_release_all(skb);
873
874
875 nc->skb_cache[nc->skb_count++] = skb;
876
877#ifdef CONFIG_SLUB
878
879 prefetchw(skb);
880#endif
881
882
883 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
884 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
885 nc->skb_cache);
886 nc->skb_count = 0;
887 }
888}
889void __kfree_skb_defer(struct sk_buff *skb)
890{
891 _kfree_skb_defer(skb);
892}
893
894void napi_consume_skb(struct sk_buff *skb, int budget)
895{
896 if (unlikely(!skb))
897 return;
898
899
900 if (unlikely(!budget)) {
901 dev_consume_skb_any(skb);
902 return;
903 }
904
905 if (!skb_unref(skb))
906 return;
907
908
909 trace_consume_skb(skb);
910
911
912 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
913 __kfree_skb(skb);
914 return;
915 }
916
917 _kfree_skb_defer(skb);
918}
919EXPORT_SYMBOL(napi_consume_skb);
920
921
922#define CHECK_SKB_FIELD(field) \
923 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
924 offsetof(struct sk_buff, headers_start)); \
925 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
926 offsetof(struct sk_buff, headers_end)); \
927
928static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
929{
930 new->tstamp = old->tstamp;
931
932 new->dev = old->dev;
933 memcpy(new->cb, old->cb, sizeof(old->cb));
934 skb_dst_copy(new, old);
935 __skb_ext_copy(new, old);
936 __nf_copy(new, old, false);
937
938
939
940
941 new->queue_mapping = old->queue_mapping;
942
943 memcpy(&new->headers_start, &old->headers_start,
944 offsetof(struct sk_buff, headers_end) -
945 offsetof(struct sk_buff, headers_start));
946 CHECK_SKB_FIELD(protocol);
947 CHECK_SKB_FIELD(csum);
948 CHECK_SKB_FIELD(hash);
949 CHECK_SKB_FIELD(priority);
950 CHECK_SKB_FIELD(skb_iif);
951 CHECK_SKB_FIELD(vlan_proto);
952 CHECK_SKB_FIELD(vlan_tci);
953 CHECK_SKB_FIELD(transport_header);
954 CHECK_SKB_FIELD(network_header);
955 CHECK_SKB_FIELD(mac_header);
956 CHECK_SKB_FIELD(inner_protocol);
957 CHECK_SKB_FIELD(inner_transport_header);
958 CHECK_SKB_FIELD(inner_network_header);
959 CHECK_SKB_FIELD(inner_mac_header);
960 CHECK_SKB_FIELD(mark);
961#ifdef CONFIG_NETWORK_SECMARK
962 CHECK_SKB_FIELD(secmark);
963#endif
964#ifdef CONFIG_NET_RX_BUSY_POLL
965 CHECK_SKB_FIELD(napi_id);
966#endif
967#ifdef CONFIG_XPS
968 CHECK_SKB_FIELD(sender_cpu);
969#endif
970#ifdef CONFIG_NET_SCHED
971 CHECK_SKB_FIELD(tc_index);
972#endif
973
974}
975
976
977
978
979
980static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
981{
982#define C(x) n->x = skb->x
983
984 n->next = n->prev = NULL;
985 n->sk = NULL;
986 __copy_skb_header(n, skb);
987
988 C(len);
989 C(data_len);
990 C(mac_len);
991 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
992 n->cloned = 1;
993 n->nohdr = 0;
994 n->peeked = 0;
995 C(pfmemalloc);
996 n->destructor = NULL;
997 C(tail);
998 C(end);
999 C(head);
1000 C(head_frag);
1001 C(data);
1002 C(truesize);
1003 refcount_set(&n->users, 1);
1004
1005 atomic_inc(&(skb_shinfo(skb)->dataref));
1006 skb->cloned = 1;
1007
1008 return n;
1009#undef C
1010}
1011
1012
1013
1014
1015
1016struct sk_buff *alloc_skb_for_msg(struct sk_buff *first)
1017{
1018 struct sk_buff *n;
1019
1020 n = alloc_skb(0, GFP_ATOMIC);
1021 if (!n)
1022 return NULL;
1023
1024 n->len = first->len;
1025 n->data_len = first->len;
1026 n->truesize = first->truesize;
1027
1028 skb_shinfo(n)->frag_list = first;
1029
1030 __copy_skb_header(n, first);
1031 n->destructor = NULL;
1032
1033 return n;
1034}
1035EXPORT_SYMBOL_GPL(alloc_skb_for_msg);
1036
1037
1038
1039
1040
1041
1042
1043
1044
1045
1046
1047struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
1048{
1049 skb_release_all(dst);
1050 return __skb_clone(dst, src);
1051}
1052EXPORT_SYMBOL_GPL(skb_morph);
1053
1054int mm_account_pinned_pages(struct mmpin *mmp, size_t size)
1055{
1056 unsigned long max_pg, num_pg, new_pg, old_pg;
1057 struct user_struct *user;
1058
1059 if (capable(CAP_IPC_LOCK) || !size)
1060 return 0;
1061
1062 num_pg = (size >> PAGE_SHIFT) + 2;
1063 max_pg = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
1064 user = mmp->user ? : current_user();
1065
1066 do {
1067 old_pg = atomic_long_read(&user->locked_vm);
1068 new_pg = old_pg + num_pg;
1069 if (new_pg > max_pg)
1070 return -ENOBUFS;
1071 } while (atomic_long_cmpxchg(&user->locked_vm, old_pg, new_pg) !=
1072 old_pg);
1073
1074 if (!mmp->user) {
1075 mmp->user = get_uid(user);
1076 mmp->num_pg = num_pg;
1077 } else {
1078 mmp->num_pg += num_pg;
1079 }
1080
1081 return 0;
1082}
1083EXPORT_SYMBOL_GPL(mm_account_pinned_pages);
1084
1085void mm_unaccount_pinned_pages(struct mmpin *mmp)
1086{
1087 if (mmp->user) {
1088 atomic_long_sub(mmp->num_pg, &mmp->user->locked_vm);
1089 free_uid(mmp->user);
1090 }
1091}
1092EXPORT_SYMBOL_GPL(mm_unaccount_pinned_pages);
1093
1094struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size)
1095{
1096 struct ubuf_info *uarg;
1097 struct sk_buff *skb;
1098
1099 WARN_ON_ONCE(!in_task());
1100
1101 skb = sock_omalloc(sk, 0, GFP_KERNEL);
1102 if (!skb)
1103 return NULL;
1104
1105 BUILD_BUG_ON(sizeof(*uarg) > sizeof(skb->cb));
1106 uarg = (void *)skb->cb;
1107 uarg->mmp.user = NULL;
1108
1109 if (mm_account_pinned_pages(&uarg->mmp, size)) {
1110 kfree_skb(skb);
1111 return NULL;
1112 }
1113
1114 uarg->callback = sock_zerocopy_callback;
1115 uarg->id = ((u32)atomic_inc_return(&sk->sk_zckey)) - 1;
1116 uarg->len = 1;
1117 uarg->bytelen = size;
1118 uarg->zerocopy = 1;
1119 refcount_set(&uarg->refcnt, 1);
1120 sock_hold(sk);
1121
1122 return uarg;
1123}
1124EXPORT_SYMBOL_GPL(sock_zerocopy_alloc);
1125
1126static inline struct sk_buff *skb_from_uarg(struct ubuf_info *uarg)
1127{
1128 return container_of((void *)uarg, struct sk_buff, cb);
1129}
1130
1131struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
1132 struct ubuf_info *uarg)
1133{
1134 if (uarg) {
1135 const u32 byte_limit = 1 << 19;
1136 u32 bytelen, next;
1137
1138
1139
1140
1141 if (!sock_owned_by_user(sk)) {
1142 WARN_ON_ONCE(1);
1143 return NULL;
1144 }
1145
1146 bytelen = uarg->bytelen + size;
1147 if (uarg->len == USHRT_MAX - 1 || bytelen > byte_limit) {
1148
1149 if (sk->sk_type == SOCK_STREAM)
1150 goto new_alloc;
1151 return NULL;
1152 }
1153
1154 next = (u32)atomic_read(&sk->sk_zckey);
1155 if ((u32)(uarg->id + uarg->len) == next) {
1156 if (mm_account_pinned_pages(&uarg->mmp, size))
1157 return NULL;
1158 uarg->len++;
1159 uarg->bytelen = bytelen;
1160 atomic_set(&sk->sk_zckey, ++next);
1161
1162
1163 if (sk->sk_type == SOCK_STREAM)
1164 sock_zerocopy_get(uarg);
1165
1166 return uarg;
1167 }
1168 }
1169
1170new_alloc:
1171 return sock_zerocopy_alloc(sk, size);
1172}
1173EXPORT_SYMBOL_GPL(sock_zerocopy_realloc);
1174
1175static bool skb_zerocopy_notify_extend(struct sk_buff *skb, u32 lo, u16 len)
1176{
1177 struct sock_exterr_skb *serr = SKB_EXT_ERR(skb);
1178 u32 old_lo, old_hi;
1179 u64 sum_len;
1180
1181 old_lo = serr->ee.ee_info;
1182 old_hi = serr->ee.ee_data;
1183 sum_len = old_hi - old_lo + 1ULL + len;
1184
1185 if (sum_len >= (1ULL << 32))
1186 return false;
1187
1188 if (lo != old_hi + 1)
1189 return false;
1190
1191 serr->ee.ee_data += len;
1192 return true;
1193}
1194
1195void sock_zerocopy_callback(struct ubuf_info *uarg, bool success)
1196{
1197 struct sk_buff *tail, *skb = skb_from_uarg(uarg);
1198 struct sock_exterr_skb *serr;
1199 struct sock *sk = skb->sk;
1200 struct sk_buff_head *q;
1201 unsigned long flags;
1202 u32 lo, hi;
1203 u16 len;
1204
1205 mm_unaccount_pinned_pages(&uarg->mmp);
1206
1207
1208
1209
1210 if (!uarg->len || sock_flag(sk, SOCK_DEAD))
1211 goto release;
1212
1213 len = uarg->len;
1214 lo = uarg->id;
1215 hi = uarg->id + len - 1;
1216
1217 serr = SKB_EXT_ERR(skb);
1218 memset(serr, 0, sizeof(*serr));
1219 serr->ee.ee_errno = 0;
1220 serr->ee.ee_origin = SO_EE_ORIGIN_ZEROCOPY;
1221 serr->ee.ee_data = hi;
1222 serr->ee.ee_info = lo;
1223 if (!success)
1224 serr->ee.ee_code |= SO_EE_CODE_ZEROCOPY_COPIED;
1225
1226 q = &sk->sk_error_queue;
1227 spin_lock_irqsave(&q->lock, flags);
1228 tail = skb_peek_tail(q);
1229 if (!tail || SKB_EXT_ERR(tail)->ee.ee_origin != SO_EE_ORIGIN_ZEROCOPY ||
1230 !skb_zerocopy_notify_extend(tail, lo, len)) {
1231 __skb_queue_tail(q, skb);
1232 skb = NULL;
1233 }
1234 spin_unlock_irqrestore(&q->lock, flags);
1235
1236 sk->sk_error_report(sk);
1237
1238release:
1239 consume_skb(skb);
1240 sock_put(sk);
1241}
1242EXPORT_SYMBOL_GPL(sock_zerocopy_callback);
1243
1244void sock_zerocopy_put(struct ubuf_info *uarg)
1245{
1246 if (uarg && refcount_dec_and_test(&uarg->refcnt)) {
1247 if (uarg->callback)
1248 uarg->callback(uarg, uarg->zerocopy);
1249 else
1250 consume_skb(skb_from_uarg(uarg));
1251 }
1252}
1253EXPORT_SYMBOL_GPL(sock_zerocopy_put);
1254
1255void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref)
1256{
1257 if (uarg) {
1258 struct sock *sk = skb_from_uarg(uarg)->sk;
1259
1260 atomic_dec(&sk->sk_zckey);
1261 uarg->len--;
1262
1263 if (have_uref)
1264 sock_zerocopy_put(uarg);
1265 }
1266}
1267EXPORT_SYMBOL_GPL(sock_zerocopy_put_abort);
1268
1269int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len)
1270{
1271 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len);
1272}
1273EXPORT_SYMBOL_GPL(skb_zerocopy_iter_dgram);
1274
1275int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
1276 struct msghdr *msg, int len,
1277 struct ubuf_info *uarg)
1278{
1279 struct ubuf_info *orig_uarg = skb_zcopy(skb);
1280 struct iov_iter orig_iter = msg->msg_iter;
1281 int err, orig_len = skb->len;
1282
1283
1284
1285
1286 if (orig_uarg && uarg != orig_uarg)
1287 return -EEXIST;
1288
1289 err = __zerocopy_sg_from_iter(sk, skb, &msg->msg_iter, len);
1290 if (err == -EFAULT || (err == -EMSGSIZE && skb->len == orig_len)) {
1291 struct sock *save_sk = skb->sk;
1292
1293
1294 msg->msg_iter = orig_iter;
1295 skb->sk = sk;
1296 ___pskb_trim(skb, orig_len);
1297 skb->sk = save_sk;
1298 return err;
1299 }
1300
1301 skb_zcopy_set(skb, uarg, NULL);
1302 return skb->len - orig_len;
1303}
1304EXPORT_SYMBOL_GPL(skb_zerocopy_iter_stream);
1305
1306static int skb_zerocopy_clone(struct sk_buff *nskb, struct sk_buff *orig,
1307 gfp_t gfp_mask)
1308{
1309 if (skb_zcopy(orig)) {
1310 if (skb_zcopy(nskb)) {
1311
1312 if (!gfp_mask) {
1313 WARN_ON_ONCE(1);
1314 return -ENOMEM;
1315 }
1316 if (skb_uarg(nskb) == skb_uarg(orig))
1317 return 0;
1318 if (skb_copy_ubufs(nskb, GFP_ATOMIC))
1319 return -EIO;
1320 }
1321 skb_zcopy_set(nskb, skb_uarg(orig), NULL);
1322 }
1323 return 0;
1324}
1325
1326
1327
1328
1329
1330
1331
1332
1333
1334
1335
1336
1337
1338
1339
1340
1341int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
1342{
1343 int num_frags = skb_shinfo(skb)->nr_frags;
1344 struct page *page, *head = NULL;
1345 int i, new_frags;
1346 u32 d_off;
1347
1348 if (skb_shared(skb) || skb_unclone(skb, gfp_mask))
1349 return -EINVAL;
1350
1351 if (!num_frags)
1352 goto release;
1353
1354 new_frags = (__skb_pagelen(skb) + PAGE_SIZE - 1) >> PAGE_SHIFT;
1355 for (i = 0; i < new_frags; i++) {
1356 page = alloc_page(gfp_mask);
1357 if (!page) {
1358 while (head) {
1359 struct page *next = (struct page *)page_private(head);
1360 put_page(head);
1361 head = next;
1362 }
1363 return -ENOMEM;
1364 }
1365 set_page_private(page, (unsigned long)head);
1366 head = page;
1367 }
1368
1369 page = head;
1370 d_off = 0;
1371 for (i = 0; i < num_frags; i++) {
1372 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1373 u32 p_off, p_len, copied;
1374 struct page *p;
1375 u8 *vaddr;
1376
1377 skb_frag_foreach_page(f, skb_frag_off(f), skb_frag_size(f),
1378 p, p_off, p_len, copied) {
1379 u32 copy, done = 0;
1380 vaddr = kmap_atomic(p);
1381
1382 while (done < p_len) {
1383 if (d_off == PAGE_SIZE) {
1384 d_off = 0;
1385 page = (struct page *)page_private(page);
1386 }
1387 copy = min_t(u32, PAGE_SIZE - d_off, p_len - done);
1388 memcpy(page_address(page) + d_off,
1389 vaddr + p_off + done, copy);
1390 done += copy;
1391 d_off += copy;
1392 }
1393 kunmap_atomic(vaddr);
1394 }
1395 }
1396
1397
1398 for (i = 0; i < num_frags; i++)
1399 skb_frag_unref(skb, i);
1400
1401
1402 for (i = 0; i < new_frags - 1; i++) {
1403 __skb_fill_page_desc(skb, i, head, 0, PAGE_SIZE);
1404 head = (struct page *)page_private(head);
1405 }
1406 __skb_fill_page_desc(skb, new_frags - 1, head, 0, d_off);
1407 skb_shinfo(skb)->nr_frags = new_frags;
1408
1409release:
1410 skb_zcopy_clear(skb, false);
1411 return 0;
1412}
1413EXPORT_SYMBOL_GPL(skb_copy_ubufs);
1414
1415
1416
1417
1418
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1430{
1431 struct sk_buff_fclones *fclones = container_of(skb,
1432 struct sk_buff_fclones,
1433 skb1);
1434 struct sk_buff *n;
1435
1436 if (skb_orphan_frags(skb, gfp_mask))
1437 return NULL;
1438
1439 if (skb->fclone == SKB_FCLONE_ORIG &&
1440 refcount_read(&fclones->fclone_ref) == 1) {
1441 n = &fclones->skb2;
1442 refcount_set(&fclones->fclone_ref, 2);
1443 } else {
1444 if (skb_pfmemalloc(skb))
1445 gfp_mask |= __GFP_MEMALLOC;
1446
1447 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1448 if (!n)
1449 return NULL;
1450
1451 n->fclone = SKB_FCLONE_UNAVAILABLE;
1452 }
1453
1454 return __skb_clone(n, skb);
1455}
1456EXPORT_SYMBOL(skb_clone);
1457
1458void skb_headers_offset_update(struct sk_buff *skb, int off)
1459{
1460
1461 if (skb->ip_summed == CHECKSUM_PARTIAL)
1462 skb->csum_start += off;
1463
1464 skb->transport_header += off;
1465 skb->network_header += off;
1466 if (skb_mac_header_was_set(skb))
1467 skb->mac_header += off;
1468 skb->inner_transport_header += off;
1469 skb->inner_network_header += off;
1470 skb->inner_mac_header += off;
1471}
1472EXPORT_SYMBOL(skb_headers_offset_update);
1473
1474void skb_copy_header(struct sk_buff *new, const struct sk_buff *old)
1475{
1476 __copy_skb_header(new, old);
1477
1478 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1479 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1480 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1481}
1482EXPORT_SYMBOL(skb_copy_header);
1483
1484static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1485{
1486 if (skb_pfmemalloc(skb))
1487 return SKB_ALLOC_RX;
1488 return 0;
1489}
1490
1491
1492
1493
1494
1495
1496
1497
1498
1499
1500
1501
1502
1503
1504
1505
1506
1507
1508struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1509{
1510 int headerlen = skb_headroom(skb);
1511 unsigned int size = skb_end_offset(skb) + skb->data_len;
1512 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1513 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1514
1515 if (!n)
1516 return NULL;
1517
1518
1519 skb_reserve(n, headerlen);
1520
1521 skb_put(n, skb->len);
1522
1523 BUG_ON(skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len));
1524
1525 skb_copy_header(n, skb);
1526 return n;
1527}
1528EXPORT_SYMBOL(skb_copy);
1529
1530
1531
1532
1533
1534
1535
1536
1537
1538
1539
1540
1541
1542
1543
1544
1545
1546
1547struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1548 gfp_t gfp_mask, bool fclone)
1549{
1550 unsigned int size = skb_headlen(skb) + headroom;
1551 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1552 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1553
1554 if (!n)
1555 goto out;
1556
1557
1558 skb_reserve(n, headroom);
1559
1560 skb_put(n, skb_headlen(skb));
1561
1562 skb_copy_from_linear_data(skb, n->data, n->len);
1563
1564 n->truesize += skb->data_len;
1565 n->data_len = skb->data_len;
1566 n->len = skb->len;
1567
1568 if (skb_shinfo(skb)->nr_frags) {
1569 int i;
1570
1571 if (skb_orphan_frags(skb, gfp_mask) ||
1572 skb_zerocopy_clone(n, skb, gfp_mask)) {
1573 kfree_skb(n);
1574 n = NULL;
1575 goto out;
1576 }
1577 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1578 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1579 skb_frag_ref(skb, i);
1580 }
1581 skb_shinfo(n)->nr_frags = i;
1582 }
1583
1584 if (skb_has_frag_list(skb)) {
1585 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1586 skb_clone_fraglist(n);
1587 }
1588
1589 skb_copy_header(n, skb);
1590out:
1591 return n;
1592}
1593EXPORT_SYMBOL(__pskb_copy_fclone);
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603
1604
1605
1606
1607
1608
1609
1610
1611int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1612 gfp_t gfp_mask)
1613{
1614 int i, osize = skb_end_offset(skb);
1615 int size = osize + nhead + ntail;
1616 long off;
1617 u8 *data;
1618
1619 BUG_ON(nhead < 0);
1620
1621 BUG_ON(skb_shared(skb));
1622
1623 size = SKB_DATA_ALIGN(size);
1624
1625 if (skb_pfmemalloc(skb))
1626 gfp_mask |= __GFP_MEMALLOC;
1627 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1628 gfp_mask, NUMA_NO_NODE, NULL);
1629 if (!data)
1630 goto nodata;
1631 size = SKB_WITH_OVERHEAD(ksize(data));
1632
1633
1634
1635
1636 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1637
1638 memcpy((struct skb_shared_info *)(data + size),
1639 skb_shinfo(skb),
1640 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1641
1642
1643
1644
1645
1646
1647 if (skb_cloned(skb)) {
1648 if (skb_orphan_frags(skb, gfp_mask))
1649 goto nofrags;
1650 if (skb_zcopy(skb))
1651 refcount_inc(&skb_uarg(skb)->refcnt);
1652 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1653 skb_frag_ref(skb, i);
1654
1655 if (skb_has_frag_list(skb))
1656 skb_clone_fraglist(skb);
1657
1658 skb_release_data(skb);
1659 } else {
1660 skb_free_head(skb);
1661 }
1662 off = (data + nhead) - skb->head;
1663
1664 skb->head = data;
1665 skb->head_frag = 0;
1666 skb->data += off;
1667#ifdef NET_SKBUFF_DATA_USES_OFFSET
1668 skb->end = size;
1669 off = nhead;
1670#else
1671 skb->end = skb->head + size;
1672#endif
1673 skb->tail += off;
1674 skb_headers_offset_update(skb, nhead);
1675 skb->cloned = 0;
1676 skb->hdr_len = 0;
1677 skb->nohdr = 0;
1678 atomic_set(&skb_shinfo(skb)->dataref, 1);
1679
1680 skb_metadata_clear(skb);
1681
1682
1683
1684
1685
1686 if (!skb->sk || skb->destructor == sock_edemux)
1687 skb->truesize += size - osize;
1688
1689 return 0;
1690
1691nofrags:
1692 kfree(data);
1693nodata:
1694 return -ENOMEM;
1695}
1696EXPORT_SYMBOL(pskb_expand_head);
1697
1698
1699
1700struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1701{
1702 struct sk_buff *skb2;
1703 int delta = headroom - skb_headroom(skb);
1704
1705 if (delta <= 0)
1706 skb2 = pskb_copy(skb, GFP_ATOMIC);
1707 else {
1708 skb2 = skb_clone(skb, GFP_ATOMIC);
1709 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1710 GFP_ATOMIC)) {
1711 kfree_skb(skb2);
1712 skb2 = NULL;
1713 }
1714 }
1715 return skb2;
1716}
1717EXPORT_SYMBOL(skb_realloc_headroom);
1718
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735
1736
1737struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1738 int newheadroom, int newtailroom,
1739 gfp_t gfp_mask)
1740{
1741
1742
1743
1744 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1745 gfp_mask, skb_alloc_rx_flag(skb),
1746 NUMA_NO_NODE);
1747 int oldheadroom = skb_headroom(skb);
1748 int head_copy_len, head_copy_off;
1749
1750 if (!n)
1751 return NULL;
1752
1753 skb_reserve(n, newheadroom);
1754
1755
1756 skb_put(n, skb->len);
1757
1758 head_copy_len = oldheadroom;
1759 head_copy_off = 0;
1760 if (newheadroom <= head_copy_len)
1761 head_copy_len = newheadroom;
1762 else
1763 head_copy_off = newheadroom - head_copy_len;
1764
1765
1766 BUG_ON(skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1767 skb->len + head_copy_len));
1768
1769 skb_copy_header(n, skb);
1770
1771 skb_headers_offset_update(n, newheadroom - oldheadroom);
1772
1773 return n;
1774}
1775EXPORT_SYMBOL(skb_copy_expand);
1776
1777
1778
1779
1780
1781
1782
1783
1784
1785
1786
1787
1788
1789
1790
1791int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error)
1792{
1793 int err;
1794 int ntail;
1795
1796
1797 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1798 memset(skb->data+skb->len, 0, pad);
1799 return 0;
1800 }
1801
1802 ntail = skb->data_len + pad - (skb->end - skb->tail);
1803 if (likely(skb_cloned(skb) || ntail > 0)) {
1804 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1805 if (unlikely(err))
1806 goto free_skb;
1807 }
1808
1809
1810
1811
1812 err = skb_linearize(skb);
1813 if (unlikely(err))
1814 goto free_skb;
1815
1816 memset(skb->data + skb->len, 0, pad);
1817 return 0;
1818
1819free_skb:
1820 if (free_on_error)
1821 kfree_skb(skb);
1822 return err;
1823}
1824EXPORT_SYMBOL(__skb_pad);
1825
1826
1827
1828
1829
1830
1831
1832
1833
1834
1835
1836
1837
1838
1839void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1840{
1841 if (tail != skb) {
1842 skb->data_len += len;
1843 skb->len += len;
1844 }
1845 return skb_put(tail, len);
1846}
1847EXPORT_SYMBOL_GPL(pskb_put);
1848
1849
1850
1851
1852
1853
1854
1855
1856
1857
1858void *skb_put(struct sk_buff *skb, unsigned int len)
1859{
1860 void *tmp = skb_tail_pointer(skb);
1861 SKB_LINEAR_ASSERT(skb);
1862 skb->tail += len;
1863 skb->len += len;
1864 if (unlikely(skb->tail > skb->end))
1865 skb_over_panic(skb, len, __builtin_return_address(0));
1866 return tmp;
1867}
1868EXPORT_SYMBOL(skb_put);
1869
1870
1871
1872
1873
1874
1875
1876
1877
1878
1879void *skb_push(struct sk_buff *skb, unsigned int len)
1880{
1881 skb->data -= len;
1882 skb->len += len;
1883 if (unlikely(skb->data < skb->head))
1884 skb_under_panic(skb, len, __builtin_return_address(0));
1885 return skb->data;
1886}
1887EXPORT_SYMBOL(skb_push);
1888
1889
1890
1891
1892
1893
1894
1895
1896
1897
1898
1899void *skb_pull(struct sk_buff *skb, unsigned int len)
1900{
1901 return skb_pull_inline(skb, len);
1902}
1903EXPORT_SYMBOL(skb_pull);
1904
1905
1906
1907
1908
1909
1910
1911
1912
1913
1914void skb_trim(struct sk_buff *skb, unsigned int len)
1915{
1916 if (skb->len > len)
1917 __skb_trim(skb, len);
1918}
1919EXPORT_SYMBOL(skb_trim);
1920
1921
1922
1923
1924int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1925{
1926 struct sk_buff **fragp;
1927 struct sk_buff *frag;
1928 int offset = skb_headlen(skb);
1929 int nfrags = skb_shinfo(skb)->nr_frags;
1930 int i;
1931 int err;
1932
1933 if (skb_cloned(skb) &&
1934 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1935 return err;
1936
1937 i = 0;
1938 if (offset >= len)
1939 goto drop_pages;
1940
1941 for (; i < nfrags; i++) {
1942 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1943
1944 if (end < len) {
1945 offset = end;
1946 continue;
1947 }
1948
1949 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1950
1951drop_pages:
1952 skb_shinfo(skb)->nr_frags = i;
1953
1954 for (; i < nfrags; i++)
1955 skb_frag_unref(skb, i);
1956
1957 if (skb_has_frag_list(skb))
1958 skb_drop_fraglist(skb);
1959 goto done;
1960 }
1961
1962 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1963 fragp = &frag->next) {
1964 int end = offset + frag->len;
1965
1966 if (skb_shared(frag)) {
1967 struct sk_buff *nfrag;
1968
1969 nfrag = skb_clone(frag, GFP_ATOMIC);
1970 if (unlikely(!nfrag))
1971 return -ENOMEM;
1972
1973 nfrag->next = frag->next;
1974 consume_skb(frag);
1975 frag = nfrag;
1976 *fragp = frag;
1977 }
1978
1979 if (end < len) {
1980 offset = end;
1981 continue;
1982 }
1983
1984 if (end > len &&
1985 unlikely((err = pskb_trim(frag, len - offset))))
1986 return err;
1987
1988 if (frag->next)
1989 skb_drop_list(&frag->next);
1990 break;
1991 }
1992
1993done:
1994 if (len > skb_headlen(skb)) {
1995 skb->data_len -= skb->len - len;
1996 skb->len = len;
1997 } else {
1998 skb->len = len;
1999 skb->data_len = 0;
2000 skb_set_tail_pointer(skb, len);
2001 }
2002
2003 if (!skb->sk || skb->destructor == sock_edemux)
2004 skb_condense(skb);
2005 return 0;
2006}
2007EXPORT_SYMBOL(___pskb_trim);
2008
2009
2010
2011int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len)
2012{
2013 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2014 int delta = skb->len - len;
2015
2016 skb->csum = csum_block_sub(skb->csum,
2017 skb_checksum(skb, len, delta, 0),
2018 len);
2019 }
2020 return __pskb_trim(skb, len);
2021}
2022EXPORT_SYMBOL(pskb_trim_rcsum_slow);
2023
2024
2025
2026
2027
2028
2029
2030
2031
2032
2033
2034
2035
2036
2037
2038
2039
2040
2041
2042
2043
2044
2045
2046
2047
2048
2049void *__pskb_pull_tail(struct sk_buff *skb, int delta)
2050{
2051
2052
2053
2054
2055 int i, k, eat = (skb->tail + delta) - skb->end;
2056
2057 if (eat > 0 || skb_cloned(skb)) {
2058 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
2059 GFP_ATOMIC))
2060 return NULL;
2061 }
2062
2063 BUG_ON(skb_copy_bits(skb, skb_headlen(skb),
2064 skb_tail_pointer(skb), delta));
2065
2066
2067
2068
2069 if (!skb_has_frag_list(skb))
2070 goto pull_pages;
2071
2072
2073 eat = delta;
2074 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2075 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2076
2077 if (size >= eat)
2078 goto pull_pages;
2079 eat -= size;
2080 }
2081
2082
2083
2084
2085
2086
2087
2088
2089 if (eat) {
2090 struct sk_buff *list = skb_shinfo(skb)->frag_list;
2091 struct sk_buff *clone = NULL;
2092 struct sk_buff *insp = NULL;
2093
2094 do {
2095 if (list->len <= eat) {
2096
2097 eat -= list->len;
2098 list = list->next;
2099 insp = list;
2100 } else {
2101
2102
2103 if (skb_shared(list)) {
2104
2105 clone = skb_clone(list, GFP_ATOMIC);
2106 if (!clone)
2107 return NULL;
2108 insp = list->next;
2109 list = clone;
2110 } else {
2111
2112
2113 insp = list;
2114 }
2115 if (!pskb_pull(list, eat)) {
2116 kfree_skb(clone);
2117 return NULL;
2118 }
2119 break;
2120 }
2121 } while (eat);
2122
2123
2124 while ((list = skb_shinfo(skb)->frag_list) != insp) {
2125 skb_shinfo(skb)->frag_list = list->next;
2126 kfree_skb(list);
2127 }
2128
2129 if (clone) {
2130 clone->next = list;
2131 skb_shinfo(skb)->frag_list = clone;
2132 }
2133 }
2134
2135
2136pull_pages:
2137 eat = delta;
2138 k = 0;
2139 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2140 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2141
2142 if (size <= eat) {
2143 skb_frag_unref(skb, i);
2144 eat -= size;
2145 } else {
2146 skb_frag_t *frag = &skb_shinfo(skb)->frags[k];
2147
2148 *frag = skb_shinfo(skb)->frags[i];
2149 if (eat) {
2150 skb_frag_off_add(frag, eat);
2151 skb_frag_size_sub(frag, eat);
2152 if (!i)
2153 goto end;
2154 eat = 0;
2155 }
2156 k++;
2157 }
2158 }
2159 skb_shinfo(skb)->nr_frags = k;
2160
2161end:
2162 skb->tail += delta;
2163 skb->data_len -= delta;
2164
2165 if (!skb->data_len)
2166 skb_zcopy_clear(skb, false);
2167
2168 return skb_tail_pointer(skb);
2169}
2170EXPORT_SYMBOL(__pskb_pull_tail);
2171
2172
2173
2174
2175
2176
2177
2178
2179
2180
2181
2182
2183
2184
2185
2186
2187int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
2188{
2189 int start = skb_headlen(skb);
2190 struct sk_buff *frag_iter;
2191 int i, copy;
2192
2193 if (offset > (int)skb->len - len)
2194 goto fault;
2195
2196
2197 if ((copy = start - offset) > 0) {
2198 if (copy > len)
2199 copy = len;
2200 skb_copy_from_linear_data_offset(skb, offset, to, copy);
2201 if ((len -= copy) == 0)
2202 return 0;
2203 offset += copy;
2204 to += copy;
2205 }
2206
2207 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2208 int end;
2209 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
2210
2211 WARN_ON(start > offset + len);
2212
2213 end = start + skb_frag_size(f);
2214 if ((copy = end - offset) > 0) {
2215 u32 p_off, p_len, copied;
2216 struct page *p;
2217 u8 *vaddr;
2218
2219 if (copy > len)
2220 copy = len;
2221
2222 skb_frag_foreach_page(f,
2223 skb_frag_off(f) + offset - start,
2224 copy, p, p_off, p_len, copied) {
2225 vaddr = kmap_atomic(p);
2226 memcpy(to + copied, vaddr + p_off, p_len);
2227 kunmap_atomic(vaddr);
2228 }
2229
2230 if ((len -= copy) == 0)
2231 return 0;
2232 offset += copy;
2233 to += copy;
2234 }
2235 start = end;
2236 }
2237
2238 skb_walk_frags(skb, frag_iter) {
2239 int end;
2240
2241 WARN_ON(start > offset + len);
2242
2243 end = start + frag_iter->len;
2244 if ((copy = end - offset) > 0) {
2245 if (copy > len)
2246 copy = len;
2247 if (skb_copy_bits(frag_iter, offset - start, to, copy))
2248 goto fault;
2249 if ((len -= copy) == 0)
2250 return 0;
2251 offset += copy;
2252 to += copy;
2253 }
2254 start = end;
2255 }
2256
2257 if (!len)
2258 return 0;
2259
2260fault:
2261 return -EFAULT;
2262}
2263EXPORT_SYMBOL(skb_copy_bits);
2264
2265
2266
2267
2268
2269static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
2270{
2271 put_page(spd->pages[i]);
2272}
2273
2274static struct page *linear_to_page(struct page *page, unsigned int *len,
2275 unsigned int *offset,
2276 struct sock *sk)
2277{
2278 struct page_frag *pfrag = sk_page_frag(sk);
2279
2280 if (!sk_page_frag_refill(sk, pfrag))
2281 return NULL;
2282
2283 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
2284
2285 memcpy(page_address(pfrag->page) + pfrag->offset,
2286 page_address(page) + *offset, *len);
2287 *offset = pfrag->offset;
2288 pfrag->offset += *len;
2289
2290 return pfrag->page;
2291}
2292
2293static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
2294 struct page *page,
2295 unsigned int offset)
2296{
2297 return spd->nr_pages &&
2298 spd->pages[spd->nr_pages - 1] == page &&
2299 (spd->partial[spd->nr_pages - 1].offset +
2300 spd->partial[spd->nr_pages - 1].len == offset);
2301}
2302
2303
2304
2305
2306static bool spd_fill_page(struct splice_pipe_desc *spd,
2307 struct pipe_inode_info *pipe, struct page *page,
2308 unsigned int *len, unsigned int offset,
2309 bool linear,
2310 struct sock *sk)
2311{
2312 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
2313 return true;
2314
2315 if (linear) {
2316 page = linear_to_page(page, len, &offset, sk);
2317 if (!page)
2318 return true;
2319 }
2320 if (spd_can_coalesce(spd, page, offset)) {
2321 spd->partial[spd->nr_pages - 1].len += *len;
2322 return false;
2323 }
2324 get_page(page);
2325 spd->pages[spd->nr_pages] = page;
2326 spd->partial[spd->nr_pages].len = *len;
2327 spd->partial[spd->nr_pages].offset = offset;
2328 spd->nr_pages++;
2329
2330 return false;
2331}
2332
2333static bool __splice_segment(struct page *page, unsigned int poff,
2334 unsigned int plen, unsigned int *off,
2335 unsigned int *len,
2336 struct splice_pipe_desc *spd, bool linear,
2337 struct sock *sk,
2338 struct pipe_inode_info *pipe)
2339{
2340 if (!*len)
2341 return true;
2342
2343
2344 if (*off >= plen) {
2345 *off -= plen;
2346 return false;
2347 }
2348
2349
2350 poff += *off;
2351 plen -= *off;
2352 *off = 0;
2353
2354 do {
2355 unsigned int flen = min(*len, plen);
2356
2357 if (spd_fill_page(spd, pipe, page, &flen, poff,
2358 linear, sk))
2359 return true;
2360 poff += flen;
2361 plen -= flen;
2362 *len -= flen;
2363 } while (*len && plen);
2364
2365 return false;
2366}
2367
2368
2369
2370
2371
2372static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
2373 unsigned int *offset, unsigned int *len,
2374 struct splice_pipe_desc *spd, struct sock *sk)
2375{
2376 int seg;
2377 struct sk_buff *iter;
2378
2379
2380
2381
2382
2383
2384 if (__splice_segment(virt_to_page(skb->data),
2385 (unsigned long) skb->data & (PAGE_SIZE - 1),
2386 skb_headlen(skb),
2387 offset, len, spd,
2388 skb_head_is_locked(skb),
2389 sk, pipe))
2390 return true;
2391
2392
2393
2394
2395 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
2396 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
2397
2398 if (__splice_segment(skb_frag_page(f),
2399 skb_frag_off(f), skb_frag_size(f),
2400 offset, len, spd, false, sk, pipe))
2401 return true;
2402 }
2403
2404 skb_walk_frags(skb, iter) {
2405 if (*offset >= iter->len) {
2406 *offset -= iter->len;
2407 continue;
2408 }
2409
2410
2411
2412
2413 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
2414 return true;
2415 }
2416
2417 return false;
2418}
2419
2420
2421
2422
2423
2424int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
2425 struct pipe_inode_info *pipe, unsigned int tlen,
2426 unsigned int flags)
2427{
2428 struct partial_page partial[MAX_SKB_FRAGS];
2429 struct page *pages[MAX_SKB_FRAGS];
2430 struct splice_pipe_desc spd = {
2431 .pages = pages,
2432 .partial = partial,
2433 .nr_pages_max = MAX_SKB_FRAGS,
2434 .ops = &nosteal_pipe_buf_ops,
2435 .spd_release = sock_spd_release,
2436 };
2437 int ret = 0;
2438
2439 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2440
2441 if (spd.nr_pages)
2442 ret = splice_to_pipe(pipe, &spd);
2443
2444 return ret;
2445}
2446EXPORT_SYMBOL_GPL(skb_splice_bits);
2447
2448
2449int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
2450 int len)
2451{
2452 unsigned int orig_len = len;
2453 struct sk_buff *head = skb;
2454 unsigned short fragidx;
2455 int slen, ret;
2456
2457do_frag_list:
2458
2459
2460 while (offset < skb_headlen(skb) && len) {
2461 struct kvec kv;
2462 struct msghdr msg;
2463
2464 slen = min_t(int, len, skb_headlen(skb) - offset);
2465 kv.iov_base = skb->data + offset;
2466 kv.iov_len = slen;
2467 memset(&msg, 0, sizeof(msg));
2468 msg.msg_flags = MSG_DONTWAIT;
2469
2470 ret = kernel_sendmsg_locked(sk, &msg, &kv, 1, slen);
2471 if (ret <= 0)
2472 goto error;
2473
2474 offset += ret;
2475 len -= ret;
2476 }
2477
2478
2479 if (!len)
2480 goto out;
2481
2482
2483 offset -= skb_headlen(skb);
2484
2485
2486 for (fragidx = 0; fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2487 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2488
2489 if (offset < skb_frag_size(frag))
2490 break;
2491
2492 offset -= skb_frag_size(frag);
2493 }
2494
2495 for (; len && fragidx < skb_shinfo(skb)->nr_frags; fragidx++) {
2496 skb_frag_t *frag = &skb_shinfo(skb)->frags[fragidx];
2497
2498 slen = min_t(size_t, len, skb_frag_size(frag) - offset);
2499
2500 while (slen) {
2501 ret = kernel_sendpage_locked(sk, skb_frag_page(frag),
2502 skb_frag_off(frag) + offset,
2503 slen, MSG_DONTWAIT);
2504 if (ret <= 0)
2505 goto error;
2506
2507 len -= ret;
2508 offset += ret;
2509 slen -= ret;
2510 }
2511
2512 offset = 0;
2513 }
2514
2515 if (len) {
2516
2517
2518 if (skb == head) {
2519 if (skb_has_frag_list(skb)) {
2520 skb = skb_shinfo(skb)->frag_list;
2521 goto do_frag_list;
2522 }
2523 } else if (skb->next) {
2524 skb = skb->next;
2525 goto do_frag_list;
2526 }
2527 }
2528
2529out:
2530 return orig_len - len;
2531
2532error:
2533 return orig_len == len ? ret : orig_len - len;
2534}
2535EXPORT_SYMBOL_GPL(skb_send_sock_locked);
2536
2537
2538
2539
2540
2541
2542
2543
2544
2545
2546
2547
2548
2549int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2550{
2551 int start = skb_headlen(skb);
2552 struct sk_buff *frag_iter;
2553 int i, copy;
2554
2555 if (offset > (int)skb->len - len)
2556 goto fault;
2557
2558 if ((copy = start - offset) > 0) {
2559 if (copy > len)
2560 copy = len;
2561 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2562 if ((len -= copy) == 0)
2563 return 0;
2564 offset += copy;
2565 from += copy;
2566 }
2567
2568 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2569 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2570 int end;
2571
2572 WARN_ON(start > offset + len);
2573
2574 end = start + skb_frag_size(frag);
2575 if ((copy = end - offset) > 0) {
2576 u32 p_off, p_len, copied;
2577 struct page *p;
2578 u8 *vaddr;
2579
2580 if (copy > len)
2581 copy = len;
2582
2583 skb_frag_foreach_page(frag,
2584 skb_frag_off(frag) + offset - start,
2585 copy, p, p_off, p_len, copied) {
2586 vaddr = kmap_atomic(p);
2587 memcpy(vaddr + p_off, from + copied, p_len);
2588 kunmap_atomic(vaddr);
2589 }
2590
2591 if ((len -= copy) == 0)
2592 return 0;
2593 offset += copy;
2594 from += copy;
2595 }
2596 start = end;
2597 }
2598
2599 skb_walk_frags(skb, frag_iter) {
2600 int end;
2601
2602 WARN_ON(start > offset + len);
2603
2604 end = start + frag_iter->len;
2605 if ((copy = end - offset) > 0) {
2606 if (copy > len)
2607 copy = len;
2608 if (skb_store_bits(frag_iter, offset - start,
2609 from, copy))
2610 goto fault;
2611 if ((len -= copy) == 0)
2612 return 0;
2613 offset += copy;
2614 from += copy;
2615 }
2616 start = end;
2617 }
2618 if (!len)
2619 return 0;
2620
2621fault:
2622 return -EFAULT;
2623}
2624EXPORT_SYMBOL(skb_store_bits);
2625
2626
2627__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2628 __wsum csum, const struct skb_checksum_ops *ops)
2629{
2630 int start = skb_headlen(skb);
2631 int i, copy = start - offset;
2632 struct sk_buff *frag_iter;
2633 int pos = 0;
2634
2635
2636 if (copy > 0) {
2637 if (copy > len)
2638 copy = len;
2639 csum = INDIRECT_CALL_1(ops->update, csum_partial_ext,
2640 skb->data + offset, copy, csum);
2641 if ((len -= copy) == 0)
2642 return csum;
2643 offset += copy;
2644 pos = copy;
2645 }
2646
2647 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2648 int end;
2649 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2650
2651 WARN_ON(start > offset + len);
2652
2653 end = start + skb_frag_size(frag);
2654 if ((copy = end - offset) > 0) {
2655 u32 p_off, p_len, copied;
2656 struct page *p;
2657 __wsum csum2;
2658 u8 *vaddr;
2659
2660 if (copy > len)
2661 copy = len;
2662
2663 skb_frag_foreach_page(frag,
2664 skb_frag_off(frag) + offset - start,
2665 copy, p, p_off, p_len, copied) {
2666 vaddr = kmap_atomic(p);
2667 csum2 = INDIRECT_CALL_1(ops->update,
2668 csum_partial_ext,
2669 vaddr + p_off, p_len, 0);
2670 kunmap_atomic(vaddr);
2671 csum = INDIRECT_CALL_1(ops->combine,
2672 csum_block_add_ext, csum,
2673 csum2, pos, p_len);
2674 pos += p_len;
2675 }
2676
2677 if (!(len -= copy))
2678 return csum;
2679 offset += copy;
2680 }
2681 start = end;
2682 }
2683
2684 skb_walk_frags(skb, frag_iter) {
2685 int end;
2686
2687 WARN_ON(start > offset + len);
2688
2689 end = start + frag_iter->len;
2690 if ((copy = end - offset) > 0) {
2691 __wsum csum2;
2692 if (copy > len)
2693 copy = len;
2694 csum2 = __skb_checksum(frag_iter, offset - start,
2695 copy, 0, ops);
2696 csum = INDIRECT_CALL_1(ops->combine, csum_block_add_ext,
2697 csum, csum2, pos, copy);
2698 if ((len -= copy) == 0)
2699 return csum;
2700 offset += copy;
2701 pos += copy;
2702 }
2703 start = end;
2704 }
2705 BUG_ON(len);
2706
2707 return csum;
2708}
2709EXPORT_SYMBOL(__skb_checksum);
2710
2711__wsum skb_checksum(const struct sk_buff *skb, int offset,
2712 int len, __wsum csum)
2713{
2714 const struct skb_checksum_ops ops = {
2715 .update = csum_partial_ext,
2716 .combine = csum_block_add_ext,
2717 };
2718
2719 return __skb_checksum(skb, offset, len, csum, &ops);
2720}
2721EXPORT_SYMBOL(skb_checksum);
2722
2723
2724
2725__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2726 u8 *to, int len, __wsum csum)
2727{
2728 int start = skb_headlen(skb);
2729 int i, copy = start - offset;
2730 struct sk_buff *frag_iter;
2731 int pos = 0;
2732
2733
2734 if (copy > 0) {
2735 if (copy > len)
2736 copy = len;
2737 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2738 copy, csum);
2739 if ((len -= copy) == 0)
2740 return csum;
2741 offset += copy;
2742 to += copy;
2743 pos = copy;
2744 }
2745
2746 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2747 int end;
2748
2749 WARN_ON(start > offset + len);
2750
2751 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2752 if ((copy = end - offset) > 0) {
2753 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2754 u32 p_off, p_len, copied;
2755 struct page *p;
2756 __wsum csum2;
2757 u8 *vaddr;
2758
2759 if (copy > len)
2760 copy = len;
2761
2762 skb_frag_foreach_page(frag,
2763 skb_frag_off(frag) + offset - start,
2764 copy, p, p_off, p_len, copied) {
2765 vaddr = kmap_atomic(p);
2766 csum2 = csum_partial_copy_nocheck(vaddr + p_off,
2767 to + copied,
2768 p_len, 0);
2769 kunmap_atomic(vaddr);
2770 csum = csum_block_add(csum, csum2, pos);
2771 pos += p_len;
2772 }
2773
2774 if (!(len -= copy))
2775 return csum;
2776 offset += copy;
2777 to += copy;
2778 }
2779 start = end;
2780 }
2781
2782 skb_walk_frags(skb, frag_iter) {
2783 __wsum csum2;
2784 int end;
2785
2786 WARN_ON(start > offset + len);
2787
2788 end = start + frag_iter->len;
2789 if ((copy = end - offset) > 0) {
2790 if (copy > len)
2791 copy = len;
2792 csum2 = skb_copy_and_csum_bits(frag_iter,
2793 offset - start,
2794 to, copy, 0);
2795 csum = csum_block_add(csum, csum2, pos);
2796 if ((len -= copy) == 0)
2797 return csum;
2798 offset += copy;
2799 to += copy;
2800 pos += copy;
2801 }
2802 start = end;
2803 }
2804 BUG_ON(len);
2805 return csum;
2806}
2807EXPORT_SYMBOL(skb_copy_and_csum_bits);
2808
2809__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len)
2810{
2811 __sum16 sum;
2812
2813 sum = csum_fold(skb_checksum(skb, 0, len, skb->csum));
2814
2815 if (likely(!sum)) {
2816 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2817 !skb->csum_complete_sw)
2818 netdev_rx_csum_fault(skb->dev, skb);
2819 }
2820 if (!skb_shared(skb))
2821 skb->csum_valid = !sum;
2822 return sum;
2823}
2824EXPORT_SYMBOL(__skb_checksum_complete_head);
2825
2826
2827
2828
2829
2830
2831
2832
2833
2834
2835__sum16 __skb_checksum_complete(struct sk_buff *skb)
2836{
2837 __wsum csum;
2838 __sum16 sum;
2839
2840 csum = skb_checksum(skb, 0, skb->len, 0);
2841
2842 sum = csum_fold(csum_add(skb->csum, csum));
2843
2844
2845
2846
2847
2848
2849
2850 if (likely(!sum)) {
2851 if (unlikely(skb->ip_summed == CHECKSUM_COMPLETE) &&
2852 !skb->csum_complete_sw)
2853 netdev_rx_csum_fault(skb->dev, skb);
2854 }
2855
2856 if (!skb_shared(skb)) {
2857
2858 skb->csum = csum;
2859 skb->ip_summed = CHECKSUM_COMPLETE;
2860 skb->csum_complete_sw = 1;
2861 skb->csum_valid = !sum;
2862 }
2863
2864 return sum;
2865}
2866EXPORT_SYMBOL(__skb_checksum_complete);
2867
2868static __wsum warn_crc32c_csum_update(const void *buff, int len, __wsum sum)
2869{
2870 net_warn_ratelimited(
2871 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2872 __func__);
2873 return 0;
2874}
2875
2876static __wsum warn_crc32c_csum_combine(__wsum csum, __wsum csum2,
2877 int offset, int len)
2878{
2879 net_warn_ratelimited(
2880 "%s: attempt to compute crc32c without libcrc32c.ko\n",
2881 __func__);
2882 return 0;
2883}
2884
2885static const struct skb_checksum_ops default_crc32c_ops = {
2886 .update = warn_crc32c_csum_update,
2887 .combine = warn_crc32c_csum_combine,
2888};
2889
2890const struct skb_checksum_ops *crc32c_csum_stub __read_mostly =
2891 &default_crc32c_ops;
2892EXPORT_SYMBOL(crc32c_csum_stub);
2893
2894
2895
2896
2897
2898
2899
2900
2901unsigned int
2902skb_zerocopy_headlen(const struct sk_buff *from)
2903{
2904 unsigned int hlen = 0;
2905
2906 if (!from->head_frag ||
2907 skb_headlen(from) < L1_CACHE_BYTES ||
2908 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2909 hlen = skb_headlen(from);
2910
2911 if (skb_has_frag_list(from))
2912 hlen = from->len;
2913
2914 return hlen;
2915}
2916EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926
2927
2928
2929
2930
2931
2932
2933
2934
2935
2936int
2937skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2938{
2939 int i, j = 0;
2940 int plen = 0;
2941 int ret;
2942 struct page *page;
2943 unsigned int offset;
2944
2945 BUG_ON(!from->head_frag && !hlen);
2946
2947
2948 if (len <= skb_tailroom(to))
2949 return skb_copy_bits(from, 0, skb_put(to, len), len);
2950
2951 if (hlen) {
2952 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2953 if (unlikely(ret))
2954 return ret;
2955 len -= hlen;
2956 } else {
2957 plen = min_t(int, skb_headlen(from), len);
2958 if (plen) {
2959 page = virt_to_head_page(from->head);
2960 offset = from->data - (unsigned char *)page_address(page);
2961 __skb_fill_page_desc(to, 0, page, offset, plen);
2962 get_page(page);
2963 j = 1;
2964 len -= plen;
2965 }
2966 }
2967
2968 to->truesize += len + plen;
2969 to->len += len + plen;
2970 to->data_len += len + plen;
2971
2972 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2973 skb_tx_error(from);
2974 return -ENOMEM;
2975 }
2976 skb_zerocopy_clone(to, from, GFP_ATOMIC);
2977
2978 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2979 int size;
2980
2981 if (!len)
2982 break;
2983 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2984 size = min_t(int, skb_frag_size(&skb_shinfo(to)->frags[j]),
2985 len);
2986 skb_frag_size_set(&skb_shinfo(to)->frags[j], size);
2987 len -= size;
2988 skb_frag_ref(to, j);
2989 j++;
2990 }
2991 skb_shinfo(to)->nr_frags = j;
2992
2993 return 0;
2994}
2995EXPORT_SYMBOL_GPL(skb_zerocopy);
2996
2997void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2998{
2999 __wsum csum;
3000 long csstart;
3001
3002 if (skb->ip_summed == CHECKSUM_PARTIAL)
3003 csstart = skb_checksum_start_offset(skb);
3004 else
3005 csstart = skb_headlen(skb);
3006
3007 BUG_ON(csstart > skb_headlen(skb));
3008
3009 skb_copy_from_linear_data(skb, to, csstart);
3010
3011 csum = 0;
3012 if (csstart != skb->len)
3013 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
3014 skb->len - csstart, 0);
3015
3016 if (skb->ip_summed == CHECKSUM_PARTIAL) {
3017 long csstuff = csstart + skb->csum_offset;
3018
3019 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
3020 }
3021}
3022EXPORT_SYMBOL(skb_copy_and_csum_dev);
3023
3024
3025
3026
3027
3028
3029
3030
3031
3032
3033struct sk_buff *skb_dequeue(struct sk_buff_head *list)
3034{
3035 unsigned long flags;
3036 struct sk_buff *result;
3037
3038 spin_lock_irqsave(&list->lock, flags);
3039 result = __skb_dequeue(list);
3040 spin_unlock_irqrestore(&list->lock, flags);
3041 return result;
3042}
3043EXPORT_SYMBOL(skb_dequeue);
3044
3045
3046
3047
3048
3049
3050
3051
3052
3053struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
3054{
3055 unsigned long flags;
3056 struct sk_buff *result;
3057
3058 spin_lock_irqsave(&list->lock, flags);
3059 result = __skb_dequeue_tail(list);
3060 spin_unlock_irqrestore(&list->lock, flags);
3061 return result;
3062}
3063EXPORT_SYMBOL(skb_dequeue_tail);
3064
3065
3066
3067
3068
3069
3070
3071
3072
3073void skb_queue_purge(struct sk_buff_head *list)
3074{
3075 struct sk_buff *skb;
3076 while ((skb = skb_dequeue(list)) != NULL)
3077 kfree_skb(skb);
3078}
3079EXPORT_SYMBOL(skb_queue_purge);
3080
3081
3082
3083
3084
3085
3086
3087
3088
3089
3090
3091unsigned int skb_rbtree_purge(struct rb_root *root)
3092{
3093 struct rb_node *p = rb_first(root);
3094 unsigned int sum = 0;
3095
3096 while (p) {
3097 struct sk_buff *skb = rb_entry(p, struct sk_buff, rbnode);
3098
3099 p = rb_next(p);
3100 rb_erase(&skb->rbnode, root);
3101 sum += skb->truesize;
3102 kfree_skb(skb);
3103 }
3104 return sum;
3105}
3106
3107
3108
3109
3110
3111
3112
3113
3114
3115
3116
3117
3118void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
3119{
3120 unsigned long flags;
3121
3122 spin_lock_irqsave(&list->lock, flags);
3123 __skb_queue_head(list, newsk);
3124 spin_unlock_irqrestore(&list->lock, flags);
3125}
3126EXPORT_SYMBOL(skb_queue_head);
3127
3128
3129
3130
3131
3132
3133
3134
3135
3136
3137
3138
3139void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
3140{
3141 unsigned long flags;
3142
3143 spin_lock_irqsave(&list->lock, flags);
3144 __skb_queue_tail(list, newsk);
3145 spin_unlock_irqrestore(&list->lock, flags);
3146}
3147EXPORT_SYMBOL(skb_queue_tail);
3148
3149
3150
3151
3152
3153
3154
3155
3156
3157
3158
3159void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
3160{
3161 unsigned long flags;
3162
3163 spin_lock_irqsave(&list->lock, flags);
3164 __skb_unlink(skb, list);
3165 spin_unlock_irqrestore(&list->lock, flags);
3166}
3167EXPORT_SYMBOL(skb_unlink);
3168
3169
3170
3171
3172
3173
3174
3175
3176
3177
3178
3179void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
3180{
3181 unsigned long flags;
3182
3183 spin_lock_irqsave(&list->lock, flags);
3184 __skb_queue_after(list, old, newsk);
3185 spin_unlock_irqrestore(&list->lock, flags);
3186}
3187EXPORT_SYMBOL(skb_append);
3188
3189static inline void skb_split_inside_header(struct sk_buff *skb,
3190 struct sk_buff* skb1,
3191 const u32 len, const int pos)
3192{
3193 int i;
3194
3195 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
3196 pos - len);
3197
3198 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
3199 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
3200
3201 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
3202 skb_shinfo(skb)->nr_frags = 0;
3203 skb1->data_len = skb->data_len;
3204 skb1->len += skb1->data_len;
3205 skb->data_len = 0;
3206 skb->len = len;
3207 skb_set_tail_pointer(skb, len);
3208}
3209
3210static inline void skb_split_no_header(struct sk_buff *skb,
3211 struct sk_buff* skb1,
3212 const u32 len, int pos)
3213{
3214 int i, k = 0;
3215 const int nfrags = skb_shinfo(skb)->nr_frags;
3216
3217 skb_shinfo(skb)->nr_frags = 0;
3218 skb1->len = skb1->data_len = skb->len - len;
3219 skb->len = len;
3220 skb->data_len = len - pos;
3221
3222 for (i = 0; i < nfrags; i++) {
3223 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
3224
3225 if (pos + size > len) {
3226 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
3227
3228 if (pos < len) {
3229
3230
3231
3232
3233
3234
3235
3236
3237 skb_frag_ref(skb, i);
3238 skb_frag_off_add(&skb_shinfo(skb1)->frags[0], len - pos);
3239 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
3240 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
3241 skb_shinfo(skb)->nr_frags++;
3242 }
3243 k++;
3244 } else
3245 skb_shinfo(skb)->nr_frags++;
3246 pos += size;
3247 }
3248 skb_shinfo(skb1)->nr_frags = k;
3249}
3250
3251
3252
3253
3254
3255
3256
3257void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
3258{
3259 int pos = skb_headlen(skb);
3260
3261 skb_shinfo(skb1)->tx_flags |= skb_shinfo(skb)->tx_flags &
3262 SKBTX_SHARED_FRAG;
3263 skb_zerocopy_clone(skb1, skb, 0);
3264 if (len < pos)
3265 skb_split_inside_header(skb, skb1, len, pos);
3266 else
3267 skb_split_no_header(skb, skb1, len, pos);
3268}
3269EXPORT_SYMBOL(skb_split);
3270
3271
3272
3273
3274
3275static int skb_prepare_for_shift(struct sk_buff *skb)
3276{
3277 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3278}
3279
3280
3281
3282
3283
3284
3285
3286
3287
3288
3289
3290
3291
3292
3293
3294
3295
3296
3297
3298int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
3299{
3300 int from, to, merge, todo;
3301 skb_frag_t *fragfrom, *fragto;
3302
3303 BUG_ON(shiftlen > skb->len);
3304
3305 if (skb_headlen(skb))
3306 return 0;
3307 if (skb_zcopy(tgt) || skb_zcopy(skb))
3308 return 0;
3309
3310 todo = shiftlen;
3311 from = 0;
3312 to = skb_shinfo(tgt)->nr_frags;
3313 fragfrom = &skb_shinfo(skb)->frags[from];
3314
3315
3316
3317
3318 if (!to ||
3319 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
3320 skb_frag_off(fragfrom))) {
3321 merge = -1;
3322 } else {
3323 merge = to - 1;
3324
3325 todo -= skb_frag_size(fragfrom);
3326 if (todo < 0) {
3327 if (skb_prepare_for_shift(skb) ||
3328 skb_prepare_for_shift(tgt))
3329 return 0;
3330
3331
3332 fragfrom = &skb_shinfo(skb)->frags[from];
3333 fragto = &skb_shinfo(tgt)->frags[merge];
3334
3335 skb_frag_size_add(fragto, shiftlen);
3336 skb_frag_size_sub(fragfrom, shiftlen);
3337 skb_frag_off_add(fragfrom, shiftlen);
3338
3339 goto onlymerged;
3340 }
3341
3342 from++;
3343 }
3344
3345
3346 if ((shiftlen == skb->len) &&
3347 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
3348 return 0;
3349
3350 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
3351 return 0;
3352
3353 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
3354 if (to == MAX_SKB_FRAGS)
3355 return 0;
3356
3357 fragfrom = &skb_shinfo(skb)->frags[from];
3358 fragto = &skb_shinfo(tgt)->frags[to];
3359
3360 if (todo >= skb_frag_size(fragfrom)) {
3361 *fragto = *fragfrom;
3362 todo -= skb_frag_size(fragfrom);
3363 from++;
3364 to++;
3365
3366 } else {
3367 __skb_frag_ref(fragfrom);
3368 skb_frag_page_copy(fragto, fragfrom);
3369 skb_frag_off_copy(fragto, fragfrom);
3370 skb_frag_size_set(fragto, todo);
3371
3372 skb_frag_off_add(fragfrom, todo);
3373 skb_frag_size_sub(fragfrom, todo);
3374 todo = 0;
3375
3376 to++;
3377 break;
3378 }
3379 }
3380
3381
3382 skb_shinfo(tgt)->nr_frags = to;
3383
3384 if (merge >= 0) {
3385 fragfrom = &skb_shinfo(skb)->frags[0];
3386 fragto = &skb_shinfo(tgt)->frags[merge];
3387
3388 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
3389 __skb_frag_unref(fragfrom);
3390 }
3391
3392
3393 to = 0;
3394 while (from < skb_shinfo(skb)->nr_frags)
3395 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
3396 skb_shinfo(skb)->nr_frags = to;
3397
3398 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
3399
3400onlymerged:
3401
3402
3403
3404 tgt->ip_summed = CHECKSUM_PARTIAL;
3405 skb->ip_summed = CHECKSUM_PARTIAL;
3406
3407
3408 skb->len -= shiftlen;
3409 skb->data_len -= shiftlen;
3410 skb->truesize -= shiftlen;
3411 tgt->len += shiftlen;
3412 tgt->data_len += shiftlen;
3413 tgt->truesize += shiftlen;
3414
3415 return shiftlen;
3416}
3417
3418
3419
3420
3421
3422
3423
3424
3425
3426
3427
3428void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
3429 unsigned int to, struct skb_seq_state *st)
3430{
3431 st->lower_offset = from;
3432 st->upper_offset = to;
3433 st->root_skb = st->cur_skb = skb;
3434 st->frag_idx = st->stepped_offset = 0;
3435 st->frag_data = NULL;
3436}
3437EXPORT_SYMBOL(skb_prepare_seq_read);
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448
3449
3450
3451
3452
3453
3454
3455
3456
3457
3458
3459
3460
3461
3462
3463
3464unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
3465 struct skb_seq_state *st)
3466{
3467 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
3468 skb_frag_t *frag;
3469
3470 if (unlikely(abs_offset >= st->upper_offset)) {
3471 if (st->frag_data) {
3472 kunmap_atomic(st->frag_data);
3473 st->frag_data = NULL;
3474 }
3475 return 0;
3476 }
3477
3478next_skb:
3479 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
3480
3481 if (abs_offset < block_limit && !st->frag_data) {
3482 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
3483 return block_limit - abs_offset;
3484 }
3485
3486 if (st->frag_idx == 0 && !st->frag_data)
3487 st->stepped_offset += skb_headlen(st->cur_skb);
3488
3489 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
3490 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
3491 block_limit = skb_frag_size(frag) + st->stepped_offset;
3492
3493 if (abs_offset < block_limit) {
3494 if (!st->frag_data)
3495 st->frag_data = kmap_atomic(skb_frag_page(frag));
3496
3497 *data = (u8 *) st->frag_data + skb_frag_off(frag) +
3498 (abs_offset - st->stepped_offset);
3499
3500 return block_limit - abs_offset;
3501 }
3502
3503 if (st->frag_data) {
3504 kunmap_atomic(st->frag_data);
3505 st->frag_data = NULL;
3506 }
3507
3508 st->frag_idx++;
3509 st->stepped_offset += skb_frag_size(frag);
3510 }
3511
3512 if (st->frag_data) {
3513 kunmap_atomic(st->frag_data);
3514 st->frag_data = NULL;
3515 }
3516
3517 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
3518 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
3519 st->frag_idx = 0;
3520 goto next_skb;
3521 } else if (st->cur_skb->next) {
3522 st->cur_skb = st->cur_skb->next;
3523 st->frag_idx = 0;
3524 goto next_skb;
3525 }
3526
3527 return 0;
3528}
3529EXPORT_SYMBOL(skb_seq_read);
3530
3531
3532
3533
3534
3535
3536
3537
3538void skb_abort_seq_read(struct skb_seq_state *st)
3539{
3540 if (st->frag_data)
3541 kunmap_atomic(st->frag_data);
3542}
3543EXPORT_SYMBOL(skb_abort_seq_read);
3544
3545#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
3546
3547static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
3548 struct ts_config *conf,
3549 struct ts_state *state)
3550{
3551 return skb_seq_read(offset, text, TS_SKB_CB(state));
3552}
3553
3554static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
3555{
3556 skb_abort_seq_read(TS_SKB_CB(state));
3557}
3558
3559
3560
3561
3562
3563
3564
3565
3566
3567
3568
3569
3570
3571unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
3572 unsigned int to, struct ts_config *config)
3573{
3574 struct ts_state state;
3575 unsigned int ret;
3576
3577 config->get_next_block = skb_ts_get_next_block;
3578 config->finish = skb_ts_finish;
3579
3580 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
3581
3582 ret = textsearch_find(config, &state);
3583 return (ret <= to - from ? ret : UINT_MAX);
3584}
3585EXPORT_SYMBOL(skb_find_text);
3586
3587int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3588 int offset, size_t size)
3589{
3590 int i = skb_shinfo(skb)->nr_frags;
3591
3592 if (skb_can_coalesce(skb, i, page, offset)) {
3593 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3594 } else if (i < MAX_SKB_FRAGS) {
3595 get_page(page);
3596 skb_fill_page_desc(skb, i, page, offset, size);
3597 } else {
3598 return -EMSGSIZE;
3599 }
3600
3601 return 0;
3602}
3603EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3604
3605
3606
3607
3608
3609
3610
3611
3612
3613
3614
3615
3616void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3617{
3618 unsigned char *data = skb->data;
3619
3620 BUG_ON(len > skb->len);
3621 __skb_pull(skb, len);
3622 skb_postpull_rcsum(skb, data, len);
3623 return skb->data;
3624}
3625EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3626
3627static inline skb_frag_t skb_head_frag_to_page_desc(struct sk_buff *frag_skb)
3628{
3629 skb_frag_t head_frag;
3630 struct page *page;
3631
3632 page = virt_to_head_page(frag_skb->head);
3633 __skb_frag_set_page(&head_frag, page);
3634 skb_frag_off_set(&head_frag, frag_skb->data -
3635 (unsigned char *)page_address(page));
3636 skb_frag_size_set(&head_frag, skb_headlen(frag_skb));
3637 return head_frag;
3638}
3639
3640struct sk_buff *skb_segment_list(struct sk_buff *skb,
3641 netdev_features_t features,
3642 unsigned int offset)
3643{
3644 struct sk_buff *list_skb = skb_shinfo(skb)->frag_list;
3645 unsigned int tnl_hlen = skb_tnl_header_len(skb);
3646 unsigned int delta_truesize = 0;
3647 unsigned int delta_len = 0;
3648 struct sk_buff *tail = NULL;
3649 struct sk_buff *nskb;
3650
3651 skb_push(skb, -skb_network_offset(skb) + offset);
3652
3653 skb_shinfo(skb)->frag_list = NULL;
3654
3655 do {
3656 nskb = list_skb;
3657 list_skb = list_skb->next;
3658
3659 if (!tail)
3660 skb->next = nskb;
3661 else
3662 tail->next = nskb;
3663
3664 tail = nskb;
3665
3666 delta_len += nskb->len;
3667 delta_truesize += nskb->truesize;
3668
3669 skb_push(nskb, -skb_network_offset(nskb) + offset);
3670
3671 skb_release_head_state(nskb);
3672 __copy_skb_header(nskb, skb);
3673
3674 skb_headers_offset_update(nskb, skb_headroom(nskb) - skb_headroom(skb));
3675 skb_copy_from_linear_data_offset(skb, -tnl_hlen,
3676 nskb->data - tnl_hlen,
3677 offset + tnl_hlen);
3678
3679 if (skb_needs_linearize(nskb, features) &&
3680 __skb_linearize(nskb))
3681 goto err_linearize;
3682
3683 } while (list_skb);
3684
3685 skb->truesize = skb->truesize - delta_truesize;
3686 skb->data_len = skb->data_len - delta_len;
3687 skb->len = skb->len - delta_len;
3688
3689 skb_gso_reset(skb);
3690
3691 skb->prev = tail;
3692
3693 if (skb_needs_linearize(skb, features) &&
3694 __skb_linearize(skb))
3695 goto err_linearize;
3696
3697 skb_get(skb);
3698
3699 return skb;
3700
3701err_linearize:
3702 kfree_skb_list(skb->next);
3703 skb->next = NULL;
3704 return ERR_PTR(-ENOMEM);
3705}
3706EXPORT_SYMBOL_GPL(skb_segment_list);
3707
3708int skb_gro_receive_list(struct sk_buff *p, struct sk_buff *skb)
3709{
3710 if (unlikely(p->len + skb->len >= 65536))
3711 return -E2BIG;
3712
3713 if (NAPI_GRO_CB(p)->last == p)
3714 skb_shinfo(p)->frag_list = skb;
3715 else
3716 NAPI_GRO_CB(p)->last->next = skb;
3717
3718 skb_pull(skb, skb_gro_offset(skb));
3719
3720 NAPI_GRO_CB(p)->last = skb;
3721 NAPI_GRO_CB(p)->count++;
3722 p->data_len += skb->len;
3723 p->truesize += skb->truesize;
3724 p->len += skb->len;
3725
3726 NAPI_GRO_CB(skb)->same_flow = 1;
3727
3728 return 0;
3729}
3730EXPORT_SYMBOL_GPL(skb_gro_receive_list);
3731
3732
3733
3734
3735
3736
3737
3738
3739
3740
3741struct sk_buff *skb_segment(struct sk_buff *head_skb,
3742 netdev_features_t features)
3743{
3744 struct sk_buff *segs = NULL;
3745 struct sk_buff *tail = NULL;
3746 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3747 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3748 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3749 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3750 struct sk_buff *frag_skb = head_skb;
3751 unsigned int offset = doffset;
3752 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3753 unsigned int partial_segs = 0;
3754 unsigned int headroom;
3755 unsigned int len = head_skb->len;
3756 __be16 proto;
3757 bool csum, sg;
3758 int nfrags = skb_shinfo(head_skb)->nr_frags;
3759 int err = -ENOMEM;
3760 int i = 0;
3761 int pos;
3762 int dummy;
3763
3764 if (list_skb && !list_skb->head_frag && skb_headlen(list_skb) &&
3765 (skb_shinfo(head_skb)->gso_type & SKB_GSO_DODGY)) {
3766
3767
3768
3769
3770
3771
3772
3773
3774
3775
3776
3777
3778
3779 if (mss != GSO_BY_FRAGS && mss != skb_headlen(head_skb))
3780 features &= ~NETIF_F_SG;
3781 }
3782
3783 __skb_push(head_skb, doffset);
3784 proto = skb_network_protocol(head_skb, &dummy);
3785 if (unlikely(!proto))
3786 return ERR_PTR(-EINVAL);
3787
3788 sg = !!(features & NETIF_F_SG);
3789 csum = !!can_checksum_protocol(features, proto);
3790
3791 if (sg && csum && (mss != GSO_BY_FRAGS)) {
3792 if (!(features & NETIF_F_GSO_PARTIAL)) {
3793 struct sk_buff *iter;
3794 unsigned int frag_len;
3795
3796 if (!list_skb ||
3797 !net_gso_ok(features, skb_shinfo(head_skb)->gso_type))
3798 goto normal;
3799
3800
3801
3802
3803
3804
3805
3806
3807
3808 frag_len = list_skb->len;
3809 skb_walk_frags(head_skb, iter) {
3810 if (frag_len != iter->len && iter->next)
3811 goto normal;
3812 if (skb_headlen(iter) && !iter->head_frag)
3813 goto normal;
3814
3815 len -= iter->len;
3816 }
3817
3818 if (len != frag_len)
3819 goto normal;
3820 }
3821
3822
3823
3824
3825
3826 partial_segs = len / mss;
3827 if (partial_segs > 1)
3828 mss *= partial_segs;
3829 else
3830 partial_segs = 0;
3831 }
3832
3833normal:
3834 headroom = skb_headroom(head_skb);
3835 pos = skb_headlen(head_skb);
3836
3837 do {
3838 struct sk_buff *nskb;
3839 skb_frag_t *nskb_frag;
3840 int hsize;
3841 int size;
3842
3843 if (unlikely(mss == GSO_BY_FRAGS)) {
3844 len = list_skb->len;
3845 } else {
3846 len = head_skb->len - offset;
3847 if (len > mss)
3848 len = mss;
3849 }
3850
3851 hsize = skb_headlen(head_skb) - offset;
3852 if (hsize < 0)
3853 hsize = 0;
3854 if (hsize > len || !sg)
3855 hsize = len;
3856
3857 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3858 (skb_headlen(list_skb) == len || sg)) {
3859 BUG_ON(skb_headlen(list_skb) > len);
3860
3861 i = 0;
3862 nfrags = skb_shinfo(list_skb)->nr_frags;
3863 frag = skb_shinfo(list_skb)->frags;
3864 frag_skb = list_skb;
3865 pos += skb_headlen(list_skb);
3866
3867 while (pos < offset + len) {
3868 BUG_ON(i >= nfrags);
3869
3870 size = skb_frag_size(frag);
3871 if (pos + size > offset + len)
3872 break;
3873
3874 i++;
3875 pos += size;
3876 frag++;
3877 }
3878
3879 nskb = skb_clone(list_skb, GFP_ATOMIC);
3880 list_skb = list_skb->next;
3881
3882 if (unlikely(!nskb))
3883 goto err;
3884
3885 if (unlikely(pskb_trim(nskb, len))) {
3886 kfree_skb(nskb);
3887 goto err;
3888 }
3889
3890 hsize = skb_end_offset(nskb);
3891 if (skb_cow_head(nskb, doffset + headroom)) {
3892 kfree_skb(nskb);
3893 goto err;
3894 }
3895
3896 nskb->truesize += skb_end_offset(nskb) - hsize;
3897 skb_release_head_state(nskb);
3898 __skb_push(nskb, doffset);
3899 } else {
3900 nskb = __alloc_skb(hsize + doffset + headroom,
3901 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3902 NUMA_NO_NODE);
3903
3904 if (unlikely(!nskb))
3905 goto err;
3906
3907 skb_reserve(nskb, headroom);
3908 __skb_put(nskb, doffset);
3909 }
3910
3911 if (segs)
3912 tail->next = nskb;
3913 else
3914 segs = nskb;
3915 tail = nskb;
3916
3917 __copy_skb_header(nskb, head_skb);
3918
3919 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3920 skb_reset_mac_len(nskb);
3921
3922 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3923 nskb->data - tnl_hlen,
3924 doffset + tnl_hlen);
3925
3926 if (nskb->len == len + doffset)
3927 goto perform_csum_check;
3928
3929 if (!sg) {
3930 if (!csum) {
3931 if (!nskb->remcsum_offload)
3932 nskb->ip_summed = CHECKSUM_NONE;
3933 SKB_GSO_CB(nskb)->csum =
3934 skb_copy_and_csum_bits(head_skb, offset,
3935 skb_put(nskb,
3936 len),
3937 len, 0);
3938 SKB_GSO_CB(nskb)->csum_start =
3939 skb_headroom(nskb) + doffset;
3940 } else {
3941 skb_copy_bits(head_skb, offset,
3942 skb_put(nskb, len),
3943 len);
3944 }
3945 continue;
3946 }
3947
3948 nskb_frag = skb_shinfo(nskb)->frags;
3949
3950 skb_copy_from_linear_data_offset(head_skb, offset,
3951 skb_put(nskb, hsize), hsize);
3952
3953 skb_shinfo(nskb)->tx_flags |= skb_shinfo(head_skb)->tx_flags &
3954 SKBTX_SHARED_FRAG;
3955
3956 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3957 skb_zerocopy_clone(nskb, frag_skb, GFP_ATOMIC))
3958 goto err;
3959
3960 while (pos < offset + len) {
3961 if (i >= nfrags) {
3962 i = 0;
3963 nfrags = skb_shinfo(list_skb)->nr_frags;
3964 frag = skb_shinfo(list_skb)->frags;
3965 frag_skb = list_skb;
3966 if (!skb_headlen(list_skb)) {
3967 BUG_ON(!nfrags);
3968 } else {
3969 BUG_ON(!list_skb->head_frag);
3970
3971
3972 i--;
3973 frag--;
3974 }
3975 if (skb_orphan_frags(frag_skb, GFP_ATOMIC) ||
3976 skb_zerocopy_clone(nskb, frag_skb,
3977 GFP_ATOMIC))
3978 goto err;
3979
3980 list_skb = list_skb->next;
3981 }
3982
3983 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3984 MAX_SKB_FRAGS)) {
3985 net_warn_ratelimited(
3986 "skb_segment: too many frags: %u %u\n",
3987 pos, mss);
3988 err = -EINVAL;
3989 goto err;
3990 }
3991
3992 *nskb_frag = (i < 0) ? skb_head_frag_to_page_desc(frag_skb) : *frag;
3993 __skb_frag_ref(nskb_frag);
3994 size = skb_frag_size(nskb_frag);
3995
3996 if (pos < offset) {
3997 skb_frag_off_add(nskb_frag, offset - pos);
3998 skb_frag_size_sub(nskb_frag, offset - pos);
3999 }
4000
4001 skb_shinfo(nskb)->nr_frags++;
4002
4003 if (pos + size <= offset + len) {
4004 i++;
4005 frag++;
4006 pos += size;
4007 } else {
4008 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
4009 goto skip_fraglist;
4010 }
4011
4012 nskb_frag++;
4013 }
4014
4015skip_fraglist:
4016 nskb->data_len = len - hsize;
4017 nskb->len += nskb->data_len;
4018 nskb->truesize += nskb->data_len;
4019
4020perform_csum_check:
4021 if (!csum) {
4022 if (skb_has_shared_frag(nskb) &&
4023 __skb_linearize(nskb))
4024 goto err;
4025
4026 if (!nskb->remcsum_offload)
4027 nskb->ip_summed = CHECKSUM_NONE;
4028 SKB_GSO_CB(nskb)->csum =
4029 skb_checksum(nskb, doffset,
4030 nskb->len - doffset, 0);
4031 SKB_GSO_CB(nskb)->csum_start =
4032 skb_headroom(nskb) + doffset;
4033 }
4034 } while ((offset += len) < head_skb->len);
4035
4036
4037
4038
4039
4040 segs->prev = tail;
4041
4042 if (partial_segs) {
4043 struct sk_buff *iter;
4044 int type = skb_shinfo(head_skb)->gso_type;
4045 unsigned short gso_size = skb_shinfo(head_skb)->gso_size;
4046
4047
4048 type |= (features & NETIF_F_GSO_PARTIAL) / NETIF_F_GSO_PARTIAL * SKB_GSO_PARTIAL;
4049 type &= ~SKB_GSO_DODGY;
4050
4051
4052
4053
4054 for (iter = segs; iter; iter = iter->next) {
4055 skb_shinfo(iter)->gso_size = gso_size;
4056 skb_shinfo(iter)->gso_segs = partial_segs;
4057 skb_shinfo(iter)->gso_type = type;
4058 SKB_GSO_CB(iter)->data_offset = skb_headroom(iter) + doffset;
4059 }
4060
4061 if (tail->len - doffset <= gso_size)
4062 skb_shinfo(tail)->gso_size = 0;
4063 else if (tail != segs)
4064 skb_shinfo(tail)->gso_segs = DIV_ROUND_UP(tail->len - doffset, gso_size);
4065 }
4066
4067
4068
4069
4070
4071 if (head_skb->destructor == sock_wfree) {
4072 swap(tail->truesize, head_skb->truesize);
4073 swap(tail->destructor, head_skb->destructor);
4074 swap(tail->sk, head_skb->sk);
4075 }
4076 return segs;
4077
4078err:
4079 kfree_skb_list(segs);
4080 return ERR_PTR(err);
4081}
4082EXPORT_SYMBOL_GPL(skb_segment);
4083
4084int skb_gro_receive(struct sk_buff *p, struct sk_buff *skb)
4085{
4086 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
4087 unsigned int offset = skb_gro_offset(skb);
4088 unsigned int headlen = skb_headlen(skb);
4089 unsigned int len = skb_gro_len(skb);
4090 unsigned int delta_truesize;
4091 struct sk_buff *lp;
4092
4093 if (unlikely(p->len + len >= 65536 || NAPI_GRO_CB(skb)->flush))
4094 return -E2BIG;
4095
4096 lp = NAPI_GRO_CB(p)->last;
4097 pinfo = skb_shinfo(lp);
4098
4099 if (headlen <= offset) {
4100 skb_frag_t *frag;
4101 skb_frag_t *frag2;
4102 int i = skbinfo->nr_frags;
4103 int nr_frags = pinfo->nr_frags + i;
4104
4105 if (nr_frags > MAX_SKB_FRAGS)
4106 goto merge;
4107
4108 offset -= headlen;
4109 pinfo->nr_frags = nr_frags;
4110 skbinfo->nr_frags = 0;
4111
4112 frag = pinfo->frags + nr_frags;
4113 frag2 = skbinfo->frags + i;
4114 do {
4115 *--frag = *--frag2;
4116 } while (--i);
4117
4118 skb_frag_off_add(frag, offset);
4119 skb_frag_size_sub(frag, offset);
4120
4121
4122 delta_truesize = skb->truesize -
4123 SKB_TRUESIZE(skb_end_offset(skb));
4124
4125 skb->truesize -= skb->data_len;
4126 skb->len -= skb->data_len;
4127 skb->data_len = 0;
4128
4129 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
4130 goto done;
4131 } else if (skb->head_frag) {
4132 int nr_frags = pinfo->nr_frags;
4133 skb_frag_t *frag = pinfo->frags + nr_frags;
4134 struct page *page = virt_to_head_page(skb->head);
4135 unsigned int first_size = headlen - offset;
4136 unsigned int first_offset;
4137
4138 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
4139 goto merge;
4140
4141 first_offset = skb->data -
4142 (unsigned char *)page_address(page) +
4143 offset;
4144
4145 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
4146
4147 __skb_frag_set_page(frag, page);
4148 skb_frag_off_set(frag, first_offset);
4149 skb_frag_size_set(frag, first_size);
4150
4151 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
4152
4153
4154 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4155 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
4156 goto done;
4157 }
4158
4159merge:
4160 delta_truesize = skb->truesize;
4161 if (offset > headlen) {
4162 unsigned int eat = offset - headlen;
4163
4164 skb_frag_off_add(&skbinfo->frags[0], eat);
4165 skb_frag_size_sub(&skbinfo->frags[0], eat);
4166 skb->data_len -= eat;
4167 skb->len -= eat;
4168 offset = headlen;
4169 }
4170
4171 __skb_pull(skb, offset);
4172
4173 if (NAPI_GRO_CB(p)->last == p)
4174 skb_shinfo(p)->frag_list = skb;
4175 else
4176 NAPI_GRO_CB(p)->last->next = skb;
4177 NAPI_GRO_CB(p)->last = skb;
4178 __skb_header_release(skb);
4179 lp = p;
4180
4181done:
4182 NAPI_GRO_CB(p)->count++;
4183 p->data_len += len;
4184 p->truesize += delta_truesize;
4185 p->len += len;
4186 if (lp != p) {
4187 lp->data_len += len;
4188 lp->truesize += delta_truesize;
4189 lp->len += len;
4190 }
4191 NAPI_GRO_CB(skb)->same_flow = 1;
4192 return 0;
4193}
4194EXPORT_SYMBOL_GPL(skb_gro_receive);
4195
4196#ifdef CONFIG_SKB_EXTENSIONS
4197#define SKB_EXT_ALIGN_VALUE 8
4198#define SKB_EXT_CHUNKSIZEOF(x) (ALIGN((sizeof(x)), SKB_EXT_ALIGN_VALUE) / SKB_EXT_ALIGN_VALUE)
4199
4200static const u8 skb_ext_type_len[] = {
4201#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4202 [SKB_EXT_BRIDGE_NF] = SKB_EXT_CHUNKSIZEOF(struct nf_bridge_info),
4203#endif
4204#ifdef CONFIG_XFRM
4205 [SKB_EXT_SEC_PATH] = SKB_EXT_CHUNKSIZEOF(struct sec_path),
4206#endif
4207#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4208 [TC_SKB_EXT] = SKB_EXT_CHUNKSIZEOF(struct tc_skb_ext),
4209#endif
4210#if IS_ENABLED(CONFIG_MPTCP)
4211 [SKB_EXT_MPTCP] = SKB_EXT_CHUNKSIZEOF(struct mptcp_ext),
4212#endif
4213};
4214
4215static __always_inline unsigned int skb_ext_total_length(void)
4216{
4217 return SKB_EXT_CHUNKSIZEOF(struct skb_ext) +
4218#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4219 skb_ext_type_len[SKB_EXT_BRIDGE_NF] +
4220#endif
4221#ifdef CONFIG_XFRM
4222 skb_ext_type_len[SKB_EXT_SEC_PATH] +
4223#endif
4224#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT)
4225 skb_ext_type_len[TC_SKB_EXT] +
4226#endif
4227#if IS_ENABLED(CONFIG_MPTCP)
4228 skb_ext_type_len[SKB_EXT_MPTCP] +
4229#endif
4230 0;
4231}
4232
4233static void skb_extensions_init(void)
4234{
4235 BUILD_BUG_ON(SKB_EXT_NUM >= 8);
4236 BUILD_BUG_ON(skb_ext_total_length() > 255);
4237
4238 skbuff_ext_cache = kmem_cache_create("skbuff_ext_cache",
4239 SKB_EXT_ALIGN_VALUE * skb_ext_total_length(),
4240 0,
4241 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4242 NULL);
4243}
4244#else
4245static void skb_extensions_init(void) {}
4246#endif
4247
4248void __init skb_init(void)
4249{
4250 skbuff_head_cache = kmem_cache_create_usercopy("skbuff_head_cache",
4251 sizeof(struct sk_buff),
4252 0,
4253 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4254 offsetof(struct sk_buff, cb),
4255 sizeof_field(struct sk_buff, cb),
4256 NULL);
4257 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
4258 sizeof(struct sk_buff_fclones),
4259 0,
4260 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
4261 NULL);
4262 skb_extensions_init();
4263}
4264
4265static int
4266__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len,
4267 unsigned int recursion_level)
4268{
4269 int start = skb_headlen(skb);
4270 int i, copy = start - offset;
4271 struct sk_buff *frag_iter;
4272 int elt = 0;
4273
4274 if (unlikely(recursion_level >= 24))
4275 return -EMSGSIZE;
4276
4277 if (copy > 0) {
4278 if (copy > len)
4279 copy = len;
4280 sg_set_buf(sg, skb->data + offset, copy);
4281 elt++;
4282 if ((len -= copy) == 0)
4283 return elt;
4284 offset += copy;
4285 }
4286
4287 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
4288 int end;
4289
4290 WARN_ON(start > offset + len);
4291
4292 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
4293 if ((copy = end - offset) > 0) {
4294 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
4295 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4296 return -EMSGSIZE;
4297
4298 if (copy > len)
4299 copy = len;
4300 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
4301 skb_frag_off(frag) + offset - start);
4302 elt++;
4303 if (!(len -= copy))
4304 return elt;
4305 offset += copy;
4306 }
4307 start = end;
4308 }
4309
4310 skb_walk_frags(skb, frag_iter) {
4311 int end, ret;
4312
4313 WARN_ON(start > offset + len);
4314
4315 end = start + frag_iter->len;
4316 if ((copy = end - offset) > 0) {
4317 if (unlikely(elt && sg_is_last(&sg[elt - 1])))
4318 return -EMSGSIZE;
4319
4320 if (copy > len)
4321 copy = len;
4322 ret = __skb_to_sgvec(frag_iter, sg+elt, offset - start,
4323 copy, recursion_level + 1);
4324 if (unlikely(ret < 0))
4325 return ret;
4326 elt += ret;
4327 if ((len -= copy) == 0)
4328 return elt;
4329 offset += copy;
4330 }
4331 start = end;
4332 }
4333 BUG_ON(len);
4334 return elt;
4335}
4336
4337
4338
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
4350{
4351 int nsg = __skb_to_sgvec(skb, sg, offset, len, 0);
4352
4353 if (nsg <= 0)
4354 return nsg;
4355
4356 sg_mark_end(&sg[nsg - 1]);
4357
4358 return nsg;
4359}
4360EXPORT_SYMBOL_GPL(skb_to_sgvec);
4361
4362
4363
4364
4365
4366
4367
4368
4369
4370
4371
4372
4373
4374
4375
4376
4377
4378
4379
4380
4381int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
4382 int offset, int len)
4383{
4384 return __skb_to_sgvec(skb, sg, offset, len, 0);
4385}
4386EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
4387
4388
4389
4390
4391
4392
4393
4394
4395
4396
4397
4398
4399
4400
4401
4402
4403
4404
4405
4406
4407int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
4408{
4409 int copyflag;
4410 int elt;
4411 struct sk_buff *skb1, **skb_p;
4412
4413
4414
4415
4416
4417 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
4418 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
4419 return -ENOMEM;
4420
4421
4422 if (!skb_has_frag_list(skb)) {
4423
4424
4425
4426
4427
4428 if (skb_tailroom(skb) < tailbits &&
4429 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
4430 return -ENOMEM;
4431
4432
4433 *trailer = skb;
4434 return 1;
4435 }
4436
4437
4438
4439 elt = 1;
4440 skb_p = &skb_shinfo(skb)->frag_list;
4441 copyflag = 0;
4442
4443 while ((skb1 = *skb_p) != NULL) {
4444 int ntail = 0;
4445
4446
4447
4448
4449
4450 if (skb_shared(skb1))
4451 copyflag = 1;
4452
4453
4454
4455 if (skb1->next == NULL && tailbits) {
4456 if (skb_shinfo(skb1)->nr_frags ||
4457 skb_has_frag_list(skb1) ||
4458 skb_tailroom(skb1) < tailbits)
4459 ntail = tailbits + 128;
4460 }
4461
4462 if (copyflag ||
4463 skb_cloned(skb1) ||
4464 ntail ||
4465 skb_shinfo(skb1)->nr_frags ||
4466 skb_has_frag_list(skb1)) {
4467 struct sk_buff *skb2;
4468
4469
4470 if (ntail == 0)
4471 skb2 = skb_copy(skb1, GFP_ATOMIC);
4472 else
4473 skb2 = skb_copy_expand(skb1,
4474 skb_headroom(skb1),
4475 ntail,
4476 GFP_ATOMIC);
4477 if (unlikely(skb2 == NULL))
4478 return -ENOMEM;
4479
4480 if (skb1->sk)
4481 skb_set_owner_w(skb2, skb1->sk);
4482
4483
4484
4485
4486 skb2->next = skb1->next;
4487 *skb_p = skb2;
4488 kfree_skb(skb1);
4489 skb1 = skb2;
4490 }
4491 elt++;
4492 *trailer = skb1;
4493 skb_p = &skb1->next;
4494 }
4495
4496 return elt;
4497}
4498EXPORT_SYMBOL_GPL(skb_cow_data);
4499
4500static void sock_rmem_free(struct sk_buff *skb)
4501{
4502 struct sock *sk = skb->sk;
4503
4504 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
4505}
4506
4507static void skb_set_err_queue(struct sk_buff *skb)
4508{
4509
4510
4511
4512 skb->pkt_type = PACKET_OUTGOING;
4513 BUILD_BUG_ON(PACKET_OUTGOING == 0);
4514}
4515
4516
4517
4518
4519int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
4520{
4521 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
4522 (unsigned int)READ_ONCE(sk->sk_rcvbuf))
4523 return -ENOMEM;
4524
4525 skb_orphan(skb);
4526 skb->sk = sk;
4527 skb->destructor = sock_rmem_free;
4528 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
4529 skb_set_err_queue(skb);
4530
4531
4532 skb_dst_force(skb);
4533
4534 skb_queue_tail(&sk->sk_error_queue, skb);
4535 if (!sock_flag(sk, SOCK_DEAD))
4536 sk->sk_error_report(sk);
4537 return 0;
4538}
4539EXPORT_SYMBOL(sock_queue_err_skb);
4540
4541static bool is_icmp_err_skb(const struct sk_buff *skb)
4542{
4543 return skb && (SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP ||
4544 SKB_EXT_ERR(skb)->ee.ee_origin == SO_EE_ORIGIN_ICMP6);
4545}
4546
4547struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
4548{
4549 struct sk_buff_head *q = &sk->sk_error_queue;
4550 struct sk_buff *skb, *skb_next = NULL;
4551 bool icmp_next = false;
4552 unsigned long flags;
4553
4554 spin_lock_irqsave(&q->lock, flags);
4555 skb = __skb_dequeue(q);
4556 if (skb && (skb_next = skb_peek(q))) {
4557 icmp_next = is_icmp_err_skb(skb_next);
4558 if (icmp_next)
4559 sk->sk_err = SKB_EXT_ERR(skb_next)->ee.ee_origin;
4560 }
4561 spin_unlock_irqrestore(&q->lock, flags);
4562
4563 if (is_icmp_err_skb(skb) && !icmp_next)
4564 sk->sk_err = 0;
4565
4566 if (skb_next)
4567 sk->sk_error_report(sk);
4568
4569 return skb;
4570}
4571EXPORT_SYMBOL(sock_dequeue_err_skb);
4572
4573
4574
4575
4576
4577
4578
4579
4580
4581
4582
4583
4584
4585
4586struct sk_buff *skb_clone_sk(struct sk_buff *skb)
4587{
4588 struct sock *sk = skb->sk;
4589 struct sk_buff *clone;
4590
4591 if (!sk || !refcount_inc_not_zero(&sk->sk_refcnt))
4592 return NULL;
4593
4594 clone = skb_clone(skb, GFP_ATOMIC);
4595 if (!clone) {
4596 sock_put(sk);
4597 return NULL;
4598 }
4599
4600 clone->sk = sk;
4601 clone->destructor = sock_efree;
4602
4603 return clone;
4604}
4605EXPORT_SYMBOL(skb_clone_sk);
4606
4607static void __skb_complete_tx_timestamp(struct sk_buff *skb,
4608 struct sock *sk,
4609 int tstype,
4610 bool opt_stats)
4611{
4612 struct sock_exterr_skb *serr;
4613 int err;
4614
4615 BUILD_BUG_ON(sizeof(struct sock_exterr_skb) > sizeof(skb->cb));
4616
4617 serr = SKB_EXT_ERR(skb);
4618 memset(serr, 0, sizeof(*serr));
4619 serr->ee.ee_errno = ENOMSG;
4620 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
4621 serr->ee.ee_info = tstype;
4622 serr->opt_stats = opt_stats;
4623 serr->header.h4.iif = skb->dev ? skb->dev->ifindex : 0;
4624 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
4625 serr->ee.ee_data = skb_shinfo(skb)->tskey;
4626 if (sk->sk_protocol == IPPROTO_TCP &&
4627 sk->sk_type == SOCK_STREAM)
4628 serr->ee.ee_data -= sk->sk_tskey;
4629 }
4630
4631 err = sock_queue_err_skb(sk, skb);
4632
4633 if (err)
4634 kfree_skb(skb);
4635}
4636
4637static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
4638{
4639 bool ret;
4640
4641 if (likely(sysctl_tstamp_allow_data || tsonly))
4642 return true;
4643
4644 read_lock_bh(&sk->sk_callback_lock);
4645 ret = sk->sk_socket && sk->sk_socket->file &&
4646 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
4647 read_unlock_bh(&sk->sk_callback_lock);
4648 return ret;
4649}
4650
4651void skb_complete_tx_timestamp(struct sk_buff *skb,
4652 struct skb_shared_hwtstamps *hwtstamps)
4653{
4654 struct sock *sk = skb->sk;
4655
4656 if (!skb_may_tx_timestamp(sk, false))
4657 goto err;
4658
4659
4660
4661
4662 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4663 *skb_hwtstamps(skb) = *hwtstamps;
4664 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND, false);
4665 sock_put(sk);
4666 return;
4667 }
4668
4669err:
4670 kfree_skb(skb);
4671}
4672EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
4673
4674void __skb_tstamp_tx(struct sk_buff *orig_skb,
4675 struct skb_shared_hwtstamps *hwtstamps,
4676 struct sock *sk, int tstype)
4677{
4678 struct sk_buff *skb;
4679 bool tsonly, opt_stats = false;
4680
4681 if (!sk)
4682 return;
4683
4684 if (!hwtstamps && !(sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TX_SWHW) &&
4685 skb_shinfo(orig_skb)->tx_flags & SKBTX_IN_PROGRESS)
4686 return;
4687
4688 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
4689 if (!skb_may_tx_timestamp(sk, tsonly))
4690 return;
4691
4692 if (tsonly) {
4693#ifdef CONFIG_INET
4694 if ((sk->sk_tsflags & SOF_TIMESTAMPING_OPT_STATS) &&
4695 sk->sk_protocol == IPPROTO_TCP &&
4696 sk->sk_type == SOCK_STREAM) {
4697 skb = tcp_get_timestamping_opt_stats(sk);
4698 opt_stats = true;
4699 } else
4700#endif
4701 skb = alloc_skb(0, GFP_ATOMIC);
4702 } else {
4703 skb = skb_clone(orig_skb, GFP_ATOMIC);
4704 }
4705 if (!skb)
4706 return;
4707
4708 if (tsonly) {
4709 skb_shinfo(skb)->tx_flags |= skb_shinfo(orig_skb)->tx_flags &
4710 SKBTX_ANY_TSTAMP;
4711 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
4712 }
4713
4714 if (hwtstamps)
4715 *skb_hwtstamps(skb) = *hwtstamps;
4716 else
4717 skb->tstamp = ktime_get_real();
4718
4719 __skb_complete_tx_timestamp(skb, sk, tstype, opt_stats);
4720}
4721EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
4722
4723void skb_tstamp_tx(struct sk_buff *orig_skb,
4724 struct skb_shared_hwtstamps *hwtstamps)
4725{
4726 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
4727 SCM_TSTAMP_SND);
4728}
4729EXPORT_SYMBOL_GPL(skb_tstamp_tx);
4730
4731void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
4732{
4733 struct sock *sk = skb->sk;
4734 struct sock_exterr_skb *serr;
4735 int err = 1;
4736
4737 skb->wifi_acked_valid = 1;
4738 skb->wifi_acked = acked;
4739
4740 serr = SKB_EXT_ERR(skb);
4741 memset(serr, 0, sizeof(*serr));
4742 serr->ee.ee_errno = ENOMSG;
4743 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
4744
4745
4746
4747
4748 if (likely(refcount_inc_not_zero(&sk->sk_refcnt))) {
4749 err = sock_queue_err_skb(sk, skb);
4750 sock_put(sk);
4751 }
4752 if (err)
4753 kfree_skb(skb);
4754}
4755EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
4756
4757
4758
4759
4760
4761
4762
4763
4764
4765
4766
4767
4768
4769bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
4770{
4771 u32 csum_end = (u32)start + (u32)off + sizeof(__sum16);
4772 u32 csum_start = skb_headroom(skb) + (u32)start;
4773
4774 if (unlikely(csum_start > U16_MAX || csum_end > skb_headlen(skb))) {
4775 net_warn_ratelimited("bad partial csum: csum=%u/%u headroom=%u headlen=%u\n",
4776 start, off, skb_headroom(skb), skb_headlen(skb));
4777 return false;
4778 }
4779 skb->ip_summed = CHECKSUM_PARTIAL;
4780 skb->csum_start = csum_start;
4781 skb->csum_offset = off;
4782 skb_set_transport_header(skb, start);
4783 return true;
4784}
4785EXPORT_SYMBOL_GPL(skb_partial_csum_set);
4786
4787static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
4788 unsigned int max)
4789{
4790 if (skb_headlen(skb) >= len)
4791 return 0;
4792
4793
4794
4795
4796 if (max > skb->len)
4797 max = skb->len;
4798
4799 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
4800 return -ENOMEM;
4801
4802 if (skb_headlen(skb) < len)
4803 return -EPROTO;
4804
4805 return 0;
4806}
4807
4808#define MAX_TCP_HDR_LEN (15 * 4)
4809
4810static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
4811 typeof(IPPROTO_IP) proto,
4812 unsigned int off)
4813{
4814 int err;
4815
4816 switch (proto) {
4817 case IPPROTO_TCP:
4818 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
4819 off + MAX_TCP_HDR_LEN);
4820 if (!err && !skb_partial_csum_set(skb, off,
4821 offsetof(struct tcphdr,
4822 check)))
4823 err = -EPROTO;
4824 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
4825
4826 case IPPROTO_UDP:
4827 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
4828 off + sizeof(struct udphdr));
4829 if (!err && !skb_partial_csum_set(skb, off,
4830 offsetof(struct udphdr,
4831 check)))
4832 err = -EPROTO;
4833 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
4834 }
4835
4836 return ERR_PTR(-EPROTO);
4837}
4838
4839
4840
4841
4842#define MAX_IP_HDR_LEN 128
4843
4844static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
4845{
4846 unsigned int off;
4847 bool fragment;
4848 __sum16 *csum;
4849 int err;
4850
4851 fragment = false;
4852
4853 err = skb_maybe_pull_tail(skb,
4854 sizeof(struct iphdr),
4855 MAX_IP_HDR_LEN);
4856 if (err < 0)
4857 goto out;
4858
4859 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
4860 fragment = true;
4861
4862 off = ip_hdrlen(skb);
4863
4864 err = -EPROTO;
4865
4866 if (fragment)
4867 goto out;
4868
4869 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
4870 if (IS_ERR(csum))
4871 return PTR_ERR(csum);
4872
4873 if (recalculate)
4874 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
4875 ip_hdr(skb)->daddr,
4876 skb->len - off,
4877 ip_hdr(skb)->protocol, 0);
4878 err = 0;
4879
4880out:
4881 return err;
4882}
4883
4884
4885
4886
4887#define MAX_IPV6_HDR_LEN 256
4888
4889#define OPT_HDR(type, skb, off) \
4890 (type *)(skb_network_header(skb) + (off))
4891
4892static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
4893{
4894 int err;
4895 u8 nexthdr;
4896 unsigned int off;
4897 unsigned int len;
4898 bool fragment;
4899 bool done;
4900 __sum16 *csum;
4901
4902 fragment = false;
4903 done = false;
4904
4905 off = sizeof(struct ipv6hdr);
4906
4907 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4908 if (err < 0)
4909 goto out;
4910
4911 nexthdr = ipv6_hdr(skb)->nexthdr;
4912
4913 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4914 while (off <= len && !done) {
4915 switch (nexthdr) {
4916 case IPPROTO_DSTOPTS:
4917 case IPPROTO_HOPOPTS:
4918 case IPPROTO_ROUTING: {
4919 struct ipv6_opt_hdr *hp;
4920
4921 err = skb_maybe_pull_tail(skb,
4922 off +
4923 sizeof(struct ipv6_opt_hdr),
4924 MAX_IPV6_HDR_LEN);
4925 if (err < 0)
4926 goto out;
4927
4928 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4929 nexthdr = hp->nexthdr;
4930 off += ipv6_optlen(hp);
4931 break;
4932 }
4933 case IPPROTO_AH: {
4934 struct ip_auth_hdr *hp;
4935
4936 err = skb_maybe_pull_tail(skb,
4937 off +
4938 sizeof(struct ip_auth_hdr),
4939 MAX_IPV6_HDR_LEN);
4940 if (err < 0)
4941 goto out;
4942
4943 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4944 nexthdr = hp->nexthdr;
4945 off += ipv6_authlen(hp);
4946 break;
4947 }
4948 case IPPROTO_FRAGMENT: {
4949 struct frag_hdr *hp;
4950
4951 err = skb_maybe_pull_tail(skb,
4952 off +
4953 sizeof(struct frag_hdr),
4954 MAX_IPV6_HDR_LEN);
4955 if (err < 0)
4956 goto out;
4957
4958 hp = OPT_HDR(struct frag_hdr, skb, off);
4959
4960 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4961 fragment = true;
4962
4963 nexthdr = hp->nexthdr;
4964 off += sizeof(struct frag_hdr);
4965 break;
4966 }
4967 default:
4968 done = true;
4969 break;
4970 }
4971 }
4972
4973 err = -EPROTO;
4974
4975 if (!done || fragment)
4976 goto out;
4977
4978 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4979 if (IS_ERR(csum))
4980 return PTR_ERR(csum);
4981
4982 if (recalculate)
4983 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4984 &ipv6_hdr(skb)->daddr,
4985 skb->len - off, nexthdr, 0);
4986 err = 0;
4987
4988out:
4989 return err;
4990}
4991
4992
4993
4994
4995
4996
4997int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4998{
4999 int err;
5000
5001 switch (skb->protocol) {
5002 case htons(ETH_P_IP):
5003 err = skb_checksum_setup_ipv4(skb, recalculate);
5004 break;
5005
5006 case htons(ETH_P_IPV6):
5007 err = skb_checksum_setup_ipv6(skb, recalculate);
5008 break;
5009
5010 default:
5011 err = -EPROTO;
5012 break;
5013 }
5014
5015 return err;
5016}
5017EXPORT_SYMBOL(skb_checksum_setup);
5018
5019
5020
5021
5022
5023
5024
5025
5026
5027
5028
5029
5030
5031
5032static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
5033 unsigned int transport_len)
5034{
5035 struct sk_buff *skb_chk;
5036 unsigned int len = skb_transport_offset(skb) + transport_len;
5037 int ret;
5038
5039 if (skb->len < len)
5040 return NULL;
5041 else if (skb->len == len)
5042 return skb;
5043
5044 skb_chk = skb_clone(skb, GFP_ATOMIC);
5045 if (!skb_chk)
5046 return NULL;
5047
5048 ret = pskb_trim_rcsum(skb_chk, len);
5049 if (ret) {
5050 kfree_skb(skb_chk);
5051 return NULL;
5052 }
5053
5054 return skb_chk;
5055}
5056
5057
5058
5059
5060
5061
5062
5063
5064
5065
5066
5067
5068
5069
5070
5071
5072struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
5073 unsigned int transport_len,
5074 __sum16(*skb_chkf)(struct sk_buff *skb))
5075{
5076 struct sk_buff *skb_chk;
5077 unsigned int offset = skb_transport_offset(skb);
5078 __sum16 ret;
5079
5080 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
5081 if (!skb_chk)
5082 goto err;
5083
5084 if (!pskb_may_pull(skb_chk, offset))
5085 goto err;
5086
5087 skb_pull_rcsum(skb_chk, offset);
5088 ret = skb_chkf(skb_chk);
5089 skb_push_rcsum(skb_chk, offset);
5090
5091 if (ret)
5092 goto err;
5093
5094 return skb_chk;
5095
5096err:
5097 if (skb_chk && skb_chk != skb)
5098 kfree_skb(skb_chk);
5099
5100 return NULL;
5101
5102}
5103EXPORT_SYMBOL(skb_checksum_trimmed);
5104
5105void __skb_warn_lro_forwarding(const struct sk_buff *skb)
5106{
5107 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
5108 skb->dev->name);
5109}
5110EXPORT_SYMBOL(__skb_warn_lro_forwarding);
5111
5112void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
5113{
5114 if (head_stolen) {
5115 skb_release_head_state(skb);
5116 kmem_cache_free(skbuff_head_cache, skb);
5117 } else {
5118 __kfree_skb(skb);
5119 }
5120}
5121EXPORT_SYMBOL(kfree_skb_partial);
5122
5123
5124
5125
5126
5127
5128
5129
5130bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
5131 bool *fragstolen, int *delta_truesize)
5132{
5133 struct skb_shared_info *to_shinfo, *from_shinfo;
5134 int i, delta, len = from->len;
5135
5136 *fragstolen = false;
5137
5138 if (skb_cloned(to))
5139 return false;
5140
5141 if (len <= skb_tailroom(to)) {
5142 if (len)
5143 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
5144 *delta_truesize = 0;
5145 return true;
5146 }
5147
5148 to_shinfo = skb_shinfo(to);
5149 from_shinfo = skb_shinfo(from);
5150 if (to_shinfo->frag_list || from_shinfo->frag_list)
5151 return false;
5152 if (skb_zcopy(to) || skb_zcopy(from))
5153 return false;
5154
5155 if (skb_headlen(from) != 0) {
5156 struct page *page;
5157 unsigned int offset;
5158
5159 if (to_shinfo->nr_frags +
5160 from_shinfo->nr_frags >= MAX_SKB_FRAGS)
5161 return false;
5162
5163 if (skb_head_is_locked(from))
5164 return false;
5165
5166 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
5167
5168 page = virt_to_head_page(from->head);
5169 offset = from->data - (unsigned char *)page_address(page);
5170
5171 skb_fill_page_desc(to, to_shinfo->nr_frags,
5172 page, offset, skb_headlen(from));
5173 *fragstolen = true;
5174 } else {
5175 if (to_shinfo->nr_frags +
5176 from_shinfo->nr_frags > MAX_SKB_FRAGS)
5177 return false;
5178
5179 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
5180 }
5181
5182 WARN_ON_ONCE(delta < len);
5183
5184 memcpy(to_shinfo->frags + to_shinfo->nr_frags,
5185 from_shinfo->frags,
5186 from_shinfo->nr_frags * sizeof(skb_frag_t));
5187 to_shinfo->nr_frags += from_shinfo->nr_frags;
5188
5189 if (!skb_cloned(from))
5190 from_shinfo->nr_frags = 0;
5191
5192
5193
5194
5195 for (i = 0; i < from_shinfo->nr_frags; i++)
5196 __skb_frag_ref(&from_shinfo->frags[i]);
5197
5198 to->truesize += delta;
5199 to->len += len;
5200 to->data_len += len;
5201
5202 *delta_truesize = delta;
5203 return true;
5204}
5205EXPORT_SYMBOL(skb_try_coalesce);
5206
5207
5208
5209
5210
5211
5212
5213
5214
5215
5216
5217
5218
5219
5220void skb_scrub_packet(struct sk_buff *skb, bool xnet)
5221{
5222 skb->pkt_type = PACKET_HOST;
5223 skb->skb_iif = 0;
5224 skb->ignore_df = 0;
5225 skb_dst_drop(skb);
5226 skb_ext_reset(skb);
5227 nf_reset_ct(skb);
5228 nf_reset_trace(skb);
5229
5230#ifdef CONFIG_NET_SWITCHDEV
5231 skb->offload_fwd_mark = 0;
5232 skb->offload_l3_fwd_mark = 0;
5233#endif
5234
5235 if (!xnet)
5236 return;
5237
5238 ipvs_reset(skb);
5239 skb->mark = 0;
5240 skb->tstamp = 0;
5241}
5242EXPORT_SYMBOL_GPL(skb_scrub_packet);
5243
5244
5245
5246
5247
5248
5249
5250
5251
5252
5253
5254static unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
5255{
5256 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5257 unsigned int thlen = 0;
5258
5259 if (skb->encapsulation) {
5260 thlen = skb_inner_transport_header(skb) -
5261 skb_transport_header(skb);
5262
5263 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
5264 thlen += inner_tcp_hdrlen(skb);
5265 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
5266 thlen = tcp_hdrlen(skb);
5267 } else if (unlikely(skb_is_gso_sctp(skb))) {
5268 thlen = sizeof(struct sctphdr);
5269 } else if (shinfo->gso_type & SKB_GSO_UDP_L4) {
5270 thlen = sizeof(struct udphdr);
5271 }
5272
5273
5274
5275
5276 return thlen + shinfo->gso_size;
5277}
5278
5279
5280
5281
5282
5283
5284
5285
5286
5287
5288
5289static unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
5290{
5291 unsigned int hdr_len = skb_transport_header(skb) -
5292 skb_network_header(skb);
5293
5294 return hdr_len + skb_gso_transport_seglen(skb);
5295}
5296
5297
5298
5299
5300
5301
5302
5303
5304
5305
5306static unsigned int skb_gso_mac_seglen(const struct sk_buff *skb)
5307{
5308 unsigned int hdr_len = skb_transport_header(skb) - skb_mac_header(skb);
5309
5310 return hdr_len + skb_gso_transport_seglen(skb);
5311}
5312
5313
5314
5315
5316
5317
5318
5319
5320
5321
5322
5323
5324
5325
5326
5327
5328
5329
5330
5331
5332
5333
5334static inline bool skb_gso_size_check(const struct sk_buff *skb,
5335 unsigned int seg_len,
5336 unsigned int max_len) {
5337 const struct skb_shared_info *shinfo = skb_shinfo(skb);
5338 const struct sk_buff *iter;
5339
5340 if (shinfo->gso_size != GSO_BY_FRAGS)
5341 return seg_len <= max_len;
5342
5343
5344 seg_len -= GSO_BY_FRAGS;
5345
5346 skb_walk_frags(skb, iter) {
5347 if (seg_len + skb_headlen(iter) > max_len)
5348 return false;
5349 }
5350
5351 return true;
5352}
5353
5354
5355
5356
5357
5358
5359
5360
5361
5362
5363
5364bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu)
5365{
5366 return skb_gso_size_check(skb, skb_gso_network_seglen(skb), mtu);
5367}
5368EXPORT_SYMBOL_GPL(skb_gso_validate_network_len);
5369
5370
5371
5372
5373
5374
5375
5376
5377
5378
5379bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len)
5380{
5381 return skb_gso_size_check(skb, skb_gso_mac_seglen(skb), len);
5382}
5383EXPORT_SYMBOL_GPL(skb_gso_validate_mac_len);
5384
5385static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
5386{
5387 int mac_len, meta_len;
5388 void *meta;
5389
5390 if (skb_cow(skb, skb_headroom(skb)) < 0) {
5391 kfree_skb(skb);
5392 return NULL;
5393 }
5394
5395 mac_len = skb->data - skb_mac_header(skb);
5396 if (likely(mac_len > VLAN_HLEN + ETH_TLEN)) {
5397 memmove(skb_mac_header(skb) + VLAN_HLEN, skb_mac_header(skb),
5398 mac_len - VLAN_HLEN - ETH_TLEN);
5399 }
5400
5401 meta_len = skb_metadata_len(skb);
5402 if (meta_len) {
5403 meta = skb_metadata_end(skb) - meta_len;
5404 memmove(meta + VLAN_HLEN, meta, meta_len);
5405 }
5406
5407 skb->mac_header += VLAN_HLEN;
5408 return skb;
5409}
5410
5411struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
5412{
5413 struct vlan_hdr *vhdr;
5414 u16 vlan_tci;
5415
5416 if (unlikely(skb_vlan_tag_present(skb))) {
5417
5418 return skb;
5419 }
5420
5421 skb = skb_share_check(skb, GFP_ATOMIC);
5422 if (unlikely(!skb))
5423 goto err_free;
5424
5425 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
5426 goto err_free;
5427
5428 vhdr = (struct vlan_hdr *)skb->data;
5429 vlan_tci = ntohs(vhdr->h_vlan_TCI);
5430 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
5431
5432 skb_pull_rcsum(skb, VLAN_HLEN);
5433 vlan_set_encap_proto(skb, vhdr);
5434
5435 skb = skb_reorder_vlan_header(skb);
5436 if (unlikely(!skb))
5437 goto err_free;
5438
5439 skb_reset_network_header(skb);
5440 skb_reset_transport_header(skb);
5441 skb_reset_mac_len(skb);
5442
5443 return skb;
5444
5445err_free:
5446 kfree_skb(skb);
5447 return NULL;
5448}
5449EXPORT_SYMBOL(skb_vlan_untag);
5450
5451int skb_ensure_writable(struct sk_buff *skb, int write_len)
5452{
5453 if (!pskb_may_pull(skb, write_len))
5454 return -ENOMEM;
5455
5456 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
5457 return 0;
5458
5459 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
5460}
5461EXPORT_SYMBOL(skb_ensure_writable);
5462
5463
5464
5465
5466int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
5467{
5468 struct vlan_hdr *vhdr;
5469 int offset = skb->data - skb_mac_header(skb);
5470 int err;
5471
5472 if (WARN_ONCE(offset,
5473 "__skb_vlan_pop got skb with skb->data not at mac header (offset %d)\n",
5474 offset)) {
5475 return -EINVAL;
5476 }
5477
5478 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
5479 if (unlikely(err))
5480 return err;
5481
5482 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5483
5484 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
5485 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
5486
5487 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
5488 __skb_pull(skb, VLAN_HLEN);
5489
5490 vlan_set_encap_proto(skb, vhdr);
5491 skb->mac_header += VLAN_HLEN;
5492
5493 if (skb_network_offset(skb) < ETH_HLEN)
5494 skb_set_network_header(skb, ETH_HLEN);
5495
5496 skb_reset_mac_len(skb);
5497
5498 return err;
5499}
5500EXPORT_SYMBOL(__skb_vlan_pop);
5501
5502
5503
5504
5505int skb_vlan_pop(struct sk_buff *skb)
5506{
5507 u16 vlan_tci;
5508 __be16 vlan_proto;
5509 int err;
5510
5511 if (likely(skb_vlan_tag_present(skb))) {
5512 __vlan_hwaccel_clear_tag(skb);
5513 } else {
5514 if (unlikely(!eth_type_vlan(skb->protocol)))
5515 return 0;
5516
5517 err = __skb_vlan_pop(skb, &vlan_tci);
5518 if (err)
5519 return err;
5520 }
5521
5522 if (likely(!eth_type_vlan(skb->protocol)))
5523 return 0;
5524
5525 vlan_proto = skb->protocol;
5526 err = __skb_vlan_pop(skb, &vlan_tci);
5527 if (unlikely(err))
5528 return err;
5529
5530 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5531 return 0;
5532}
5533EXPORT_SYMBOL(skb_vlan_pop);
5534
5535
5536
5537
5538int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
5539{
5540 if (skb_vlan_tag_present(skb)) {
5541 int offset = skb->data - skb_mac_header(skb);
5542 int err;
5543
5544 if (WARN_ONCE(offset,
5545 "skb_vlan_push got skb with skb->data not at mac header (offset %d)\n",
5546 offset)) {
5547 return -EINVAL;
5548 }
5549
5550 err = __vlan_insert_tag(skb, skb->vlan_proto,
5551 skb_vlan_tag_get(skb));
5552 if (err)
5553 return err;
5554
5555 skb->protocol = skb->vlan_proto;
5556 skb->mac_len += VLAN_HLEN;
5557
5558 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
5559 }
5560 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
5561 return 0;
5562}
5563EXPORT_SYMBOL(skb_vlan_push);
5564
5565
5566static void skb_mod_eth_type(struct sk_buff *skb, struct ethhdr *hdr,
5567 __be16 ethertype)
5568{
5569 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5570 __be16 diff[] = { ~hdr->h_proto, ethertype };
5571
5572 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5573 }
5574
5575 hdr->h_proto = ethertype;
5576}
5577
5578
5579
5580
5581
5582
5583
5584
5585
5586
5587
5588
5589
5590
5591
5592
5593int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto,
5594 int mac_len, bool ethernet)
5595{
5596 struct mpls_shim_hdr *lse;
5597 int err;
5598
5599 if (unlikely(!eth_p_mpls(mpls_proto)))
5600 return -EINVAL;
5601
5602
5603 if (skb->encapsulation)
5604 return -EINVAL;
5605
5606 err = skb_cow_head(skb, MPLS_HLEN);
5607 if (unlikely(err))
5608 return err;
5609
5610 if (!skb->inner_protocol) {
5611 skb_set_inner_network_header(skb, skb_network_offset(skb));
5612 skb_set_inner_protocol(skb, skb->protocol);
5613 }
5614
5615 skb_push(skb, MPLS_HLEN);
5616 memmove(skb_mac_header(skb) - MPLS_HLEN, skb_mac_header(skb),
5617 mac_len);
5618 skb_reset_mac_header(skb);
5619 skb_set_network_header(skb, mac_len);
5620 skb_reset_mac_len(skb);
5621
5622 lse = mpls_hdr(skb);
5623 lse->label_stack_entry = mpls_lse;
5624 skb_postpush_rcsum(skb, lse, MPLS_HLEN);
5625
5626 if (ethernet)
5627 skb_mod_eth_type(skb, eth_hdr(skb), mpls_proto);
5628 skb->protocol = mpls_proto;
5629
5630 return 0;
5631}
5632EXPORT_SYMBOL_GPL(skb_mpls_push);
5633
5634
5635
5636
5637
5638
5639
5640
5641
5642
5643
5644
5645
5646int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len,
5647 bool ethernet)
5648{
5649 int err;
5650
5651 if (unlikely(!eth_p_mpls(skb->protocol)))
5652 return 0;
5653
5654 err = skb_ensure_writable(skb, mac_len + MPLS_HLEN);
5655 if (unlikely(err))
5656 return err;
5657
5658 skb_postpull_rcsum(skb, mpls_hdr(skb), MPLS_HLEN);
5659 memmove(skb_mac_header(skb) + MPLS_HLEN, skb_mac_header(skb),
5660 mac_len);
5661
5662 __skb_pull(skb, MPLS_HLEN);
5663 skb_reset_mac_header(skb);
5664 skb_set_network_header(skb, mac_len);
5665
5666 if (ethernet) {
5667 struct ethhdr *hdr;
5668
5669
5670 hdr = (struct ethhdr *)((void *)mpls_hdr(skb) - ETH_HLEN);
5671 skb_mod_eth_type(skb, hdr, next_proto);
5672 }
5673 skb->protocol = next_proto;
5674
5675 return 0;
5676}
5677EXPORT_SYMBOL_GPL(skb_mpls_pop);
5678
5679
5680
5681
5682
5683
5684
5685
5686
5687
5688
5689int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse)
5690{
5691 int err;
5692
5693 if (unlikely(!eth_p_mpls(skb->protocol)))
5694 return -EINVAL;
5695
5696 err = skb_ensure_writable(skb, skb->mac_len + MPLS_HLEN);
5697 if (unlikely(err))
5698 return err;
5699
5700 if (skb->ip_summed == CHECKSUM_COMPLETE) {
5701 __be32 diff[] = { ~mpls_hdr(skb)->label_stack_entry, mpls_lse };
5702
5703 skb->csum = csum_partial((char *)diff, sizeof(diff), skb->csum);
5704 }
5705
5706 mpls_hdr(skb)->label_stack_entry = mpls_lse;
5707
5708 return 0;
5709}
5710EXPORT_SYMBOL_GPL(skb_mpls_update_lse);
5711
5712
5713
5714
5715
5716
5717
5718
5719
5720
5721int skb_mpls_dec_ttl(struct sk_buff *skb)
5722{
5723 u32 lse;
5724 u8 ttl;
5725
5726 if (unlikely(!eth_p_mpls(skb->protocol)))
5727 return -EINVAL;
5728
5729 lse = be32_to_cpu(mpls_hdr(skb)->label_stack_entry);
5730 ttl = (lse & MPLS_LS_TTL_MASK) >> MPLS_LS_TTL_SHIFT;
5731 if (!--ttl)
5732 return -EINVAL;
5733
5734 lse &= ~MPLS_LS_TTL_MASK;
5735 lse |= ttl << MPLS_LS_TTL_SHIFT;
5736
5737 return skb_mpls_update_lse(skb, cpu_to_be32(lse));
5738}
5739EXPORT_SYMBOL_GPL(skb_mpls_dec_ttl);
5740
5741
5742
5743
5744
5745
5746
5747
5748
5749
5750
5751
5752struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
5753 unsigned long data_len,
5754 int max_page_order,
5755 int *errcode,
5756 gfp_t gfp_mask)
5757{
5758 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
5759 unsigned long chunk;
5760 struct sk_buff *skb;
5761 struct page *page;
5762 int i;
5763
5764 *errcode = -EMSGSIZE;
5765
5766
5767
5768 if (npages > MAX_SKB_FRAGS)
5769 return NULL;
5770
5771 *errcode = -ENOBUFS;
5772 skb = alloc_skb(header_len, gfp_mask);
5773 if (!skb)
5774 return NULL;
5775
5776 skb->truesize += npages << PAGE_SHIFT;
5777
5778 for (i = 0; npages > 0; i++) {
5779 int order = max_page_order;
5780
5781 while (order) {
5782 if (npages >= 1 << order) {
5783 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
5784 __GFP_COMP |
5785 __GFP_NOWARN,
5786 order);
5787 if (page)
5788 goto fill_page;
5789
5790 order = 1;
5791 max_page_order = 0;
5792 }
5793 order--;
5794 }
5795 page = alloc_page(gfp_mask);
5796 if (!page)
5797 goto failure;
5798fill_page:
5799 chunk = min_t(unsigned long, data_len,
5800 PAGE_SIZE << order);
5801 skb_fill_page_desc(skb, i, page, 0, chunk);
5802 data_len -= chunk;
5803 npages -= 1 << order;
5804 }
5805 return skb;
5806
5807failure:
5808 kfree_skb(skb);
5809 return NULL;
5810}
5811EXPORT_SYMBOL(alloc_skb_with_frags);
5812
5813
5814static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
5815 const int headlen, gfp_t gfp_mask)
5816{
5817 int i;
5818 int size = skb_end_offset(skb);
5819 int new_hlen = headlen - off;
5820 u8 *data;
5821
5822 size = SKB_DATA_ALIGN(size);
5823
5824 if (skb_pfmemalloc(skb))
5825 gfp_mask |= __GFP_MEMALLOC;
5826 data = kmalloc_reserve(size +
5827 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5828 gfp_mask, NUMA_NO_NODE, NULL);
5829 if (!data)
5830 return -ENOMEM;
5831
5832 size = SKB_WITH_OVERHEAD(ksize(data));
5833
5834
5835 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
5836 skb->len -= off;
5837
5838 memcpy((struct skb_shared_info *)(data + size),
5839 skb_shinfo(skb),
5840 offsetof(struct skb_shared_info,
5841 frags[skb_shinfo(skb)->nr_frags]));
5842 if (skb_cloned(skb)) {
5843
5844 if (skb_orphan_frags(skb, gfp_mask)) {
5845 kfree(data);
5846 return -ENOMEM;
5847 }
5848 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
5849 skb_frag_ref(skb, i);
5850 if (skb_has_frag_list(skb))
5851 skb_clone_fraglist(skb);
5852 skb_release_data(skb);
5853 } else {
5854
5855
5856
5857 skb_free_head(skb);
5858 }
5859
5860 skb->head = data;
5861 skb->data = data;
5862 skb->head_frag = 0;
5863#ifdef NET_SKBUFF_DATA_USES_OFFSET
5864 skb->end = size;
5865#else
5866 skb->end = skb->head + size;
5867#endif
5868 skb_set_tail_pointer(skb, skb_headlen(skb));
5869 skb_headers_offset_update(skb, 0);
5870 skb->cloned = 0;
5871 skb->hdr_len = 0;
5872 skb->nohdr = 0;
5873 atomic_set(&skb_shinfo(skb)->dataref, 1);
5874
5875 return 0;
5876}
5877
5878static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
5879
5880
5881
5882
5883static int pskb_carve_frag_list(struct sk_buff *skb,
5884 struct skb_shared_info *shinfo, int eat,
5885 gfp_t gfp_mask)
5886{
5887 struct sk_buff *list = shinfo->frag_list;
5888 struct sk_buff *clone = NULL;
5889 struct sk_buff *insp = NULL;
5890
5891 do {
5892 if (!list) {
5893 pr_err("Not enough bytes to eat. Want %d\n", eat);
5894 return -EFAULT;
5895 }
5896 if (list->len <= eat) {
5897
5898 eat -= list->len;
5899 list = list->next;
5900 insp = list;
5901 } else {
5902
5903 if (skb_shared(list)) {
5904 clone = skb_clone(list, gfp_mask);
5905 if (!clone)
5906 return -ENOMEM;
5907 insp = list->next;
5908 list = clone;
5909 } else {
5910
5911 insp = list;
5912 }
5913 if (pskb_carve(list, eat, gfp_mask) < 0) {
5914 kfree_skb(clone);
5915 return -ENOMEM;
5916 }
5917 break;
5918 }
5919 } while (eat);
5920
5921
5922 while ((list = shinfo->frag_list) != insp) {
5923 shinfo->frag_list = list->next;
5924 kfree_skb(list);
5925 }
5926
5927 if (clone) {
5928 clone->next = list;
5929 shinfo->frag_list = clone;
5930 }
5931 return 0;
5932}
5933
5934
5935
5936
5937static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
5938 int pos, gfp_t gfp_mask)
5939{
5940 int i, k = 0;
5941 int size = skb_end_offset(skb);
5942 u8 *data;
5943 const int nfrags = skb_shinfo(skb)->nr_frags;
5944 struct skb_shared_info *shinfo;
5945
5946 size = SKB_DATA_ALIGN(size);
5947
5948 if (skb_pfmemalloc(skb))
5949 gfp_mask |= __GFP_MEMALLOC;
5950 data = kmalloc_reserve(size +
5951 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
5952 gfp_mask, NUMA_NO_NODE, NULL);
5953 if (!data)
5954 return -ENOMEM;
5955
5956 size = SKB_WITH_OVERHEAD(ksize(data));
5957
5958 memcpy((struct skb_shared_info *)(data + size),
5959 skb_shinfo(skb), offsetof(struct skb_shared_info,
5960 frags[skb_shinfo(skb)->nr_frags]));
5961 if (skb_orphan_frags(skb, gfp_mask)) {
5962 kfree(data);
5963 return -ENOMEM;
5964 }
5965 shinfo = (struct skb_shared_info *)(data + size);
5966 for (i = 0; i < nfrags; i++) {
5967 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
5968
5969 if (pos + fsize > off) {
5970 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
5971
5972 if (pos < off) {
5973
5974
5975
5976
5977
5978
5979
5980
5981 skb_frag_off_add(&shinfo->frags[0], off - pos);
5982 skb_frag_size_sub(&shinfo->frags[0], off - pos);
5983 }
5984 skb_frag_ref(skb, i);
5985 k++;
5986 }
5987 pos += fsize;
5988 }
5989 shinfo->nr_frags = k;
5990 if (skb_has_frag_list(skb))
5991 skb_clone_fraglist(skb);
5992
5993 if (k == 0) {
5994
5995 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
5996 }
5997 skb_release_data(skb);
5998
5999 skb->head = data;
6000 skb->head_frag = 0;
6001 skb->data = data;
6002#ifdef NET_SKBUFF_DATA_USES_OFFSET
6003 skb->end = size;
6004#else
6005 skb->end = skb->head + size;
6006#endif
6007 skb_reset_tail_pointer(skb);
6008 skb_headers_offset_update(skb, 0);
6009 skb->cloned = 0;
6010 skb->hdr_len = 0;
6011 skb->nohdr = 0;
6012 skb->len -= off;
6013 skb->data_len = skb->len;
6014 atomic_set(&skb_shinfo(skb)->dataref, 1);
6015 return 0;
6016}
6017
6018
6019static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
6020{
6021 int headlen = skb_headlen(skb);
6022
6023 if (len < headlen)
6024 return pskb_carve_inside_header(skb, len, headlen, gfp);
6025 else
6026 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
6027}
6028
6029
6030
6031
6032struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
6033 int to_copy, gfp_t gfp)
6034{
6035 struct sk_buff *clone = skb_clone(skb, gfp);
6036
6037 if (!clone)
6038 return NULL;
6039
6040 if (pskb_carve(clone, off, gfp) < 0 ||
6041 pskb_trim(clone, to_copy)) {
6042 kfree_skb(clone);
6043 return NULL;
6044 }
6045 return clone;
6046}
6047EXPORT_SYMBOL(pskb_extract);
6048
6049
6050
6051
6052
6053
6054
6055
6056
6057
6058
6059
6060
6061void skb_condense(struct sk_buff *skb)
6062{
6063 if (skb->data_len) {
6064 if (skb->data_len > skb->end - skb->tail ||
6065 skb_cloned(skb))
6066 return;
6067
6068
6069 __pskb_pull_tail(skb, skb->data_len);
6070 }
6071
6072
6073
6074
6075
6076
6077
6078 skb->truesize = SKB_TRUESIZE(skb_end_offset(skb));
6079}
6080
6081#ifdef CONFIG_SKB_EXTENSIONS
6082static void *skb_ext_get_ptr(struct skb_ext *ext, enum skb_ext_id id)
6083{
6084 return (void *)ext + (ext->offset[id] * SKB_EXT_ALIGN_VALUE);
6085}
6086
6087
6088
6089
6090
6091
6092
6093
6094struct skb_ext *__skb_ext_alloc(void)
6095{
6096 struct skb_ext *new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6097
6098 if (new) {
6099 memset(new->offset, 0, sizeof(new->offset));
6100 refcount_set(&new->refcnt, 1);
6101 }
6102
6103 return new;
6104}
6105
6106static struct skb_ext *skb_ext_maybe_cow(struct skb_ext *old,
6107 unsigned int old_active)
6108{
6109 struct skb_ext *new;
6110
6111 if (refcount_read(&old->refcnt) == 1)
6112 return old;
6113
6114 new = kmem_cache_alloc(skbuff_ext_cache, GFP_ATOMIC);
6115 if (!new)
6116 return NULL;
6117
6118 memcpy(new, old, old->chunks * SKB_EXT_ALIGN_VALUE);
6119 refcount_set(&new->refcnt, 1);
6120
6121#ifdef CONFIG_XFRM
6122 if (old_active & (1 << SKB_EXT_SEC_PATH)) {
6123 struct sec_path *sp = skb_ext_get_ptr(old, SKB_EXT_SEC_PATH);
6124 unsigned int i;
6125
6126 for (i = 0; i < sp->len; i++)
6127 xfrm_state_hold(sp->xvec[i]);
6128 }
6129#endif
6130 __skb_ext_put(old);
6131 return new;
6132}
6133
6134
6135
6136
6137
6138
6139
6140
6141
6142
6143
6144void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id,
6145 struct skb_ext *ext)
6146{
6147 unsigned int newlen, newoff = SKB_EXT_CHUNKSIZEOF(*ext);
6148
6149 skb_ext_put(skb);
6150 newlen = newoff + skb_ext_type_len[id];
6151 ext->chunks = newlen;
6152 ext->offset[id] = newoff;
6153 skb->extensions = ext;
6154 skb->active_extensions = 1 << id;
6155 return skb_ext_get_ptr(ext, id);
6156}
6157
6158
6159
6160
6161
6162
6163
6164
6165
6166
6167
6168
6169
6170
6171
6172void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id)
6173{
6174 struct skb_ext *new, *old = NULL;
6175 unsigned int newlen, newoff;
6176
6177 if (skb->active_extensions) {
6178 old = skb->extensions;
6179
6180 new = skb_ext_maybe_cow(old, skb->active_extensions);
6181 if (!new)
6182 return NULL;
6183
6184 if (__skb_ext_exist(new, id))
6185 goto set_active;
6186
6187 newoff = new->chunks;
6188 } else {
6189 newoff = SKB_EXT_CHUNKSIZEOF(*new);
6190
6191 new = __skb_ext_alloc();
6192 if (!new)
6193 return NULL;
6194 }
6195
6196 newlen = newoff + skb_ext_type_len[id];
6197 new->chunks = newlen;
6198 new->offset[id] = newoff;
6199set_active:
6200 skb->extensions = new;
6201 skb->active_extensions |= 1 << id;
6202 return skb_ext_get_ptr(new, id);
6203}
6204EXPORT_SYMBOL(skb_ext_add);
6205
6206#ifdef CONFIG_XFRM
6207static void skb_ext_put_sp(struct sec_path *sp)
6208{
6209 unsigned int i;
6210
6211 for (i = 0; i < sp->len; i++)
6212 xfrm_state_put(sp->xvec[i]);
6213}
6214#endif
6215
6216void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
6217{
6218 struct skb_ext *ext = skb->extensions;
6219
6220 skb->active_extensions &= ~(1 << id);
6221 if (skb->active_extensions == 0) {
6222 skb->extensions = NULL;
6223 __skb_ext_put(ext);
6224#ifdef CONFIG_XFRM
6225 } else if (id == SKB_EXT_SEC_PATH &&
6226 refcount_read(&ext->refcnt) == 1) {
6227 struct sec_path *sp = skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH);
6228
6229 skb_ext_put_sp(sp);
6230 sp->len = 0;
6231#endif
6232 }
6233}
6234EXPORT_SYMBOL(__skb_ext_del);
6235
6236void __skb_ext_put(struct skb_ext *ext)
6237{
6238
6239
6240
6241 if (refcount_read(&ext->refcnt) == 1)
6242 goto free_now;
6243
6244 if (!refcount_dec_and_test(&ext->refcnt))
6245 return;
6246free_now:
6247#ifdef CONFIG_XFRM
6248 if (__skb_ext_exist(ext, SKB_EXT_SEC_PATH))
6249 skb_ext_put_sp(skb_ext_get_ptr(ext, SKB_EXT_SEC_PATH));
6250#endif
6251
6252 kmem_cache_free(skbuff_ext_cache, ext);
6253}
6254EXPORT_SYMBOL(__skb_ext_put);
6255#endif
6256