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
36
37
38
39#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
40
41#include <linux/module.h>
42#include <linux/types.h>
43#include <linux/kernel.h>
44#include <linux/kmemcheck.h>
45#include <linux/mm.h>
46#include <linux/interrupt.h>
47#include <linux/in.h>
48#include <linux/inet.h>
49#include <linux/slab.h>
50#include <linux/tcp.h>
51#include <linux/udp.h>
52#include <linux/netdevice.h>
53#ifdef CONFIG_NET_CLS_ACT
54#include <net/pkt_sched.h>
55#endif
56#include <linux/string.h>
57#include <linux/skbuff.h>
58#include <linux/splice.h>
59#include <linux/cache.h>
60#include <linux/rtnetlink.h>
61#include <linux/init.h>
62#include <linux/scatterlist.h>
63#include <linux/errqueue.h>
64#include <linux/prefetch.h>
65#include <linux/if_vlan.h>
66
67#include <net/protocol.h>
68#include <net/dst.h>
69#include <net/sock.h>
70#include <net/checksum.h>
71#include <net/ip6_checksum.h>
72#include <net/xfrm.h>
73
74#include <asm/uaccess.h>
75#include <trace/events/skb.h>
76#include <linux/highmem.h>
77#include <linux/capability.h>
78#include <linux/user_namespace.h>
79
80struct kmem_cache *skbuff_head_cache __read_mostly;
81static struct kmem_cache *skbuff_fclone_cache __read_mostly;
82int sysctl_max_skb_frags __read_mostly = MAX_SKB_FRAGS;
83EXPORT_SYMBOL(sysctl_max_skb_frags);
84
85
86
87
88
89
90
91
92
93
94
95
96
97static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
98 const char msg[])
99{
100 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
101 msg, addr, skb->len, sz, skb->head, skb->data,
102 (unsigned long)skb->tail, (unsigned long)skb->end,
103 skb->dev ? skb->dev->name : "<NULL>");
104 BUG();
105}
106
107static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
108{
109 skb_panic(skb, sz, addr, __func__);
110}
111
112static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
113{
114 skb_panic(skb, sz, addr, __func__);
115}
116
117
118
119
120
121
122
123
124#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
125 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
126
127static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
128 unsigned long ip, bool *pfmemalloc)
129{
130 void *obj;
131 bool ret_pfmemalloc = false;
132
133
134
135
136
137 obj = kmalloc_node_track_caller(size,
138 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
139 node);
140 if (obj || !(gfp_pfmemalloc_allowed(flags)))
141 goto out;
142
143
144 ret_pfmemalloc = true;
145 obj = kmalloc_node_track_caller(size, flags, node);
146
147out:
148 if (pfmemalloc)
149 *pfmemalloc = ret_pfmemalloc;
150
151 return obj;
152}
153
154
155
156
157
158
159
160struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
161{
162 struct sk_buff *skb;
163
164
165 skb = kmem_cache_alloc_node(skbuff_head_cache,
166 gfp_mask & ~__GFP_DMA, node);
167 if (!skb)
168 goto out;
169
170
171
172
173
174
175 memset(skb, 0, offsetof(struct sk_buff, tail));
176 skb->head = NULL;
177 skb->truesize = sizeof(struct sk_buff);
178 atomic_set(&skb->users, 1);
179
180 skb->mac_header = (typeof(skb->mac_header))~0U;
181out:
182 return skb;
183}
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
203 int flags, int node)
204{
205 struct kmem_cache *cache;
206 struct skb_shared_info *shinfo;
207 struct sk_buff *skb;
208 u8 *data;
209 bool pfmemalloc;
210
211 cache = (flags & SKB_ALLOC_FCLONE)
212 ? skbuff_fclone_cache : skbuff_head_cache;
213
214 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
215 gfp_mask |= __GFP_MEMALLOC;
216
217
218 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
219 if (!skb)
220 goto out;
221 prefetchw(skb);
222
223
224
225
226
227
228 size = SKB_DATA_ALIGN(size);
229 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
230 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
231 if (!data)
232 goto nodata;
233
234
235
236
237 size = SKB_WITH_OVERHEAD(ksize(data));
238 prefetchw(data + size);
239
240
241
242
243
244
245 memset(skb, 0, offsetof(struct sk_buff, tail));
246
247 skb->truesize = SKB_TRUESIZE(size);
248 skb->pfmemalloc = pfmemalloc;
249 atomic_set(&skb->users, 1);
250 skb->head = data;
251 skb->data = data;
252 skb_reset_tail_pointer(skb);
253 skb->end = skb->tail + size;
254 skb->mac_header = (typeof(skb->mac_header))~0U;
255 skb->transport_header = (typeof(skb->transport_header))~0U;
256
257
258 shinfo = skb_shinfo(skb);
259 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
260 atomic_set(&shinfo->dataref, 1);
261 kmemcheck_annotate_variable(shinfo->destructor_arg);
262
263 if (flags & SKB_ALLOC_FCLONE) {
264 struct sk_buff_fclones *fclones;
265
266 fclones = container_of(skb, struct sk_buff_fclones, skb1);
267
268 kmemcheck_annotate_bitfield(&fclones->skb2, flags1);
269 skb->fclone = SKB_FCLONE_ORIG;
270 atomic_set(&fclones->fclone_ref, 1);
271
272 fclones->skb2.fclone = SKB_FCLONE_CLONE;
273 fclones->skb2.pfmemalloc = pfmemalloc;
274 }
275out:
276 return skb;
277nodata:
278 kmem_cache_free(cache, skb);
279 skb = NULL;
280 goto out;
281}
282EXPORT_SYMBOL(__alloc_skb);
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303struct sk_buff *__build_skb(void *data, unsigned int frag_size)
304{
305 struct skb_shared_info *shinfo;
306 struct sk_buff *skb;
307 unsigned int size = frag_size ? : ksize(data);
308
309 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
310 if (!skb)
311 return NULL;
312
313 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
314
315 memset(skb, 0, offsetof(struct sk_buff, tail));
316 skb->truesize = SKB_TRUESIZE(size);
317 atomic_set(&skb->users, 1);
318 skb->head = data;
319 skb->data = data;
320 skb_reset_tail_pointer(skb);
321 skb->end = skb->tail + size;
322 skb->mac_header = (typeof(skb->mac_header))~0U;
323 skb->transport_header = (typeof(skb->transport_header))~0U;
324
325
326 shinfo = skb_shinfo(skb);
327 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
328 atomic_set(&shinfo->dataref, 1);
329 kmemcheck_annotate_variable(shinfo->destructor_arg);
330
331 return skb;
332}
333
334
335
336
337
338
339struct sk_buff *build_skb(void *data, unsigned int frag_size)
340{
341 struct sk_buff *skb = __build_skb(data, frag_size);
342
343 if (skb && frag_size) {
344 skb->head_frag = 1;
345 if (page_is_pfmemalloc(virt_to_head_page(data)))
346 skb->pfmemalloc = 1;
347 }
348 return skb;
349}
350EXPORT_SYMBOL(build_skb);
351
352#define NAPI_SKB_CACHE_SIZE 64
353
354struct napi_alloc_cache {
355 struct page_frag_cache page;
356 size_t skb_count;
357 void *skb_cache[NAPI_SKB_CACHE_SIZE];
358};
359
360static DEFINE_PER_CPU(struct page_frag_cache, netdev_alloc_cache);
361static DEFINE_PER_CPU(struct napi_alloc_cache, napi_alloc_cache);
362
363static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
364{
365 struct page_frag_cache *nc;
366 unsigned long flags;
367 void *data;
368
369 local_irq_save(flags);
370 nc = this_cpu_ptr(&netdev_alloc_cache);
371 data = __alloc_page_frag(nc, fragsz, gfp_mask);
372 local_irq_restore(flags);
373 return data;
374}
375
376
377
378
379
380
381
382
383void *netdev_alloc_frag(unsigned int fragsz)
384{
385 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
386}
387EXPORT_SYMBOL(netdev_alloc_frag);
388
389static void *__napi_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
390{
391 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
392
393 return __alloc_page_frag(&nc->page, fragsz, gfp_mask);
394}
395
396void *napi_alloc_frag(unsigned int fragsz)
397{
398 return __napi_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
399}
400EXPORT_SYMBOL(napi_alloc_frag);
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int len,
416 gfp_t gfp_mask)
417{
418 struct page_frag_cache *nc;
419 unsigned long flags;
420 struct sk_buff *skb;
421 bool pfmemalloc;
422 void *data;
423
424 len += NET_SKB_PAD;
425
426 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
427 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
428 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
429 if (!skb)
430 goto skb_fail;
431 goto skb_success;
432 }
433
434 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
435 len = SKB_DATA_ALIGN(len);
436
437 if (sk_memalloc_socks())
438 gfp_mask |= __GFP_MEMALLOC;
439
440 local_irq_save(flags);
441
442 nc = this_cpu_ptr(&netdev_alloc_cache);
443 data = __alloc_page_frag(nc, len, gfp_mask);
444 pfmemalloc = nc->pfmemalloc;
445
446 local_irq_restore(flags);
447
448 if (unlikely(!data))
449 return NULL;
450
451 skb = __build_skb(data, len);
452 if (unlikely(!skb)) {
453 skb_free_frag(data);
454 return NULL;
455 }
456
457
458 if (pfmemalloc)
459 skb->pfmemalloc = 1;
460 skb->head_frag = 1;
461
462skb_success:
463 skb_reserve(skb, NET_SKB_PAD);
464 skb->dev = dev;
465
466skb_fail:
467 return skb;
468}
469EXPORT_SYMBOL(__netdev_alloc_skb);
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, unsigned int len,
485 gfp_t gfp_mask)
486{
487 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
488 struct sk_buff *skb;
489 void *data;
490
491 len += NET_SKB_PAD + NET_IP_ALIGN;
492
493 if ((len > SKB_WITH_OVERHEAD(PAGE_SIZE)) ||
494 (gfp_mask & (__GFP_DIRECT_RECLAIM | GFP_DMA))) {
495 skb = __alloc_skb(len, gfp_mask, SKB_ALLOC_RX, NUMA_NO_NODE);
496 if (!skb)
497 goto skb_fail;
498 goto skb_success;
499 }
500
501 len += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
502 len = SKB_DATA_ALIGN(len);
503
504 if (sk_memalloc_socks())
505 gfp_mask |= __GFP_MEMALLOC;
506
507 data = __alloc_page_frag(&nc->page, len, gfp_mask);
508 if (unlikely(!data))
509 return NULL;
510
511 skb = __build_skb(data, len);
512 if (unlikely(!skb)) {
513 skb_free_frag(data);
514 return NULL;
515 }
516
517
518 if (nc->page.pfmemalloc)
519 skb->pfmemalloc = 1;
520 skb->head_frag = 1;
521
522skb_success:
523 skb_reserve(skb, NET_SKB_PAD + NET_IP_ALIGN);
524 skb->dev = napi->dev;
525
526skb_fail:
527 return skb;
528}
529EXPORT_SYMBOL(__napi_alloc_skb);
530
531void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
532 int size, unsigned int truesize)
533{
534 skb_fill_page_desc(skb, i, page, off, size);
535 skb->len += size;
536 skb->data_len += size;
537 skb->truesize += truesize;
538}
539EXPORT_SYMBOL(skb_add_rx_frag);
540
541void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
542 unsigned int truesize)
543{
544 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
545
546 skb_frag_size_add(frag, size);
547 skb->len += size;
548 skb->data_len += size;
549 skb->truesize += truesize;
550}
551EXPORT_SYMBOL(skb_coalesce_rx_frag);
552
553static void skb_drop_list(struct sk_buff **listp)
554{
555 kfree_skb_list(*listp);
556 *listp = NULL;
557}
558
559static inline void skb_drop_fraglist(struct sk_buff *skb)
560{
561 skb_drop_list(&skb_shinfo(skb)->frag_list);
562}
563
564static void skb_clone_fraglist(struct sk_buff *skb)
565{
566 struct sk_buff *list;
567
568 skb_walk_frags(skb, list)
569 skb_get(list);
570}
571
572static void skb_free_head(struct sk_buff *skb)
573{
574 unsigned char *head = skb->head;
575
576 if (skb->head_frag)
577 skb_free_frag(head);
578 else
579 kfree(head);
580}
581
582static void skb_release_data(struct sk_buff *skb)
583{
584 struct skb_shared_info *shinfo = skb_shinfo(skb);
585 int i;
586
587 if (skb->cloned &&
588 atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
589 &shinfo->dataref))
590 return;
591
592 for (i = 0; i < shinfo->nr_frags; i++)
593 __skb_frag_unref(&shinfo->frags[i]);
594
595
596
597
598
599 if (shinfo->tx_flags & SKBTX_DEV_ZEROCOPY) {
600 struct ubuf_info *uarg;
601
602 uarg = shinfo->destructor_arg;
603 if (uarg->callback)
604 uarg->callback(uarg, true);
605 }
606
607 if (shinfo->frag_list)
608 kfree_skb_list(shinfo->frag_list);
609
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 (atomic_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 (!atomic_dec_and_test(&fclones->fclone_ref))
641 return;
642fastpath:
643 kmem_cache_free(skbuff_fclone_cache, fclones);
644}
645
646static void skb_release_head_state(struct sk_buff *skb)
647{
648 skb_dst_drop(skb);
649#ifdef CONFIG_XFRM
650 secpath_put(skb->sp);
651#endif
652 if (skb->destructor) {
653 WARN_ON(in_irq());
654 skb->destructor(skb);
655 }
656#if IS_ENABLED(CONFIG_NF_CONNTRACK)
657 nf_conntrack_put(skb->nfct);
658#endif
659#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
660 nf_bridge_put(skb->nf_bridge);
661#endif
662}
663
664
665static void skb_release_all(struct sk_buff *skb)
666{
667 skb_release_head_state(skb);
668 if (likely(skb->head))
669 skb_release_data(skb);
670}
671
672
673
674
675
676
677
678
679
680
681void __kfree_skb(struct sk_buff *skb)
682{
683 skb_release_all(skb);
684 kfree_skbmem(skb);
685}
686EXPORT_SYMBOL(__kfree_skb);
687
688
689
690
691
692
693
694
695void kfree_skb(struct sk_buff *skb)
696{
697 if (unlikely(!skb))
698 return;
699 if (likely(atomic_read(&skb->users) == 1))
700 smp_rmb();
701 else if (likely(!atomic_dec_and_test(&skb->users)))
702 return;
703 trace_kfree_skb(skb, __builtin_return_address(0));
704 __kfree_skb(skb);
705}
706EXPORT_SYMBOL(kfree_skb);
707
708void kfree_skb_list(struct sk_buff *segs)
709{
710 while (segs) {
711 struct sk_buff *next = segs->next;
712
713 kfree_skb(segs);
714 segs = next;
715 }
716}
717EXPORT_SYMBOL(kfree_skb_list);
718
719
720
721
722
723
724
725
726void skb_tx_error(struct sk_buff *skb)
727{
728 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
729 struct ubuf_info *uarg;
730
731 uarg = skb_shinfo(skb)->destructor_arg;
732 if (uarg->callback)
733 uarg->callback(uarg, false);
734 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
735 }
736}
737EXPORT_SYMBOL(skb_tx_error);
738
739
740
741
742
743
744
745
746
747void consume_skb(struct sk_buff *skb)
748{
749 if (unlikely(!skb))
750 return;
751 if (likely(atomic_read(&skb->users) == 1))
752 smp_rmb();
753 else if (likely(!atomic_dec_and_test(&skb->users)))
754 return;
755 trace_consume_skb(skb);
756 __kfree_skb(skb);
757}
758EXPORT_SYMBOL(consume_skb);
759
760void __kfree_skb_flush(void)
761{
762 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
763
764
765 if (nc->skb_count) {
766 kmem_cache_free_bulk(skbuff_head_cache, nc->skb_count,
767 nc->skb_cache);
768 nc->skb_count = 0;
769 }
770}
771
772static inline void _kfree_skb_defer(struct sk_buff *skb)
773{
774 struct napi_alloc_cache *nc = this_cpu_ptr(&napi_alloc_cache);
775
776
777 skb_release_all(skb);
778
779
780 nc->skb_cache[nc->skb_count++] = skb;
781
782#ifdef CONFIG_SLUB
783
784 prefetchw(skb);
785#endif
786
787
788 if (unlikely(nc->skb_count == NAPI_SKB_CACHE_SIZE)) {
789 kmem_cache_free_bulk(skbuff_head_cache, NAPI_SKB_CACHE_SIZE,
790 nc->skb_cache);
791 nc->skb_count = 0;
792 }
793}
794void __kfree_skb_defer(struct sk_buff *skb)
795{
796 _kfree_skb_defer(skb);
797}
798
799void napi_consume_skb(struct sk_buff *skb, int budget)
800{
801 if (unlikely(!skb))
802 return;
803
804
805 if (unlikely(!budget)) {
806 dev_consume_skb_any(skb);
807 return;
808 }
809
810 if (likely(atomic_read(&skb->users) == 1))
811 smp_rmb();
812 else if (likely(!atomic_dec_and_test(&skb->users)))
813 return;
814
815 trace_consume_skb(skb);
816
817
818 if (skb->fclone != SKB_FCLONE_UNAVAILABLE) {
819 __kfree_skb(skb);
820 return;
821 }
822
823 _kfree_skb_defer(skb);
824}
825EXPORT_SYMBOL(napi_consume_skb);
826
827
828#define CHECK_SKB_FIELD(field) \
829 BUILD_BUG_ON(offsetof(struct sk_buff, field) < \
830 offsetof(struct sk_buff, headers_start)); \
831 BUILD_BUG_ON(offsetof(struct sk_buff, field) > \
832 offsetof(struct sk_buff, headers_end)); \
833
834static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
835{
836 new->tstamp = old->tstamp;
837
838 new->dev = old->dev;
839 memcpy(new->cb, old->cb, sizeof(old->cb));
840 skb_dst_copy(new, old);
841#ifdef CONFIG_XFRM
842 new->sp = secpath_get(old->sp);
843#endif
844 __nf_copy(new, old, false);
845
846
847
848
849 new->queue_mapping = old->queue_mapping;
850
851 memcpy(&new->headers_start, &old->headers_start,
852 offsetof(struct sk_buff, headers_end) -
853 offsetof(struct sk_buff, headers_start));
854 CHECK_SKB_FIELD(protocol);
855 CHECK_SKB_FIELD(csum);
856 CHECK_SKB_FIELD(hash);
857 CHECK_SKB_FIELD(priority);
858 CHECK_SKB_FIELD(skb_iif);
859 CHECK_SKB_FIELD(vlan_proto);
860 CHECK_SKB_FIELD(vlan_tci);
861 CHECK_SKB_FIELD(transport_header);
862 CHECK_SKB_FIELD(network_header);
863 CHECK_SKB_FIELD(mac_header);
864 CHECK_SKB_FIELD(inner_protocol);
865 CHECK_SKB_FIELD(inner_transport_header);
866 CHECK_SKB_FIELD(inner_network_header);
867 CHECK_SKB_FIELD(inner_mac_header);
868 CHECK_SKB_FIELD(mark);
869#ifdef CONFIG_NETWORK_SECMARK
870 CHECK_SKB_FIELD(secmark);
871#endif
872#ifdef CONFIG_NET_RX_BUSY_POLL
873 CHECK_SKB_FIELD(napi_id);
874#endif
875#ifdef CONFIG_XPS
876 CHECK_SKB_FIELD(sender_cpu);
877#endif
878#ifdef CONFIG_NET_SCHED
879 CHECK_SKB_FIELD(tc_index);
880#ifdef CONFIG_NET_CLS_ACT
881 CHECK_SKB_FIELD(tc_verd);
882#endif
883#endif
884
885}
886
887
888
889
890
891static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
892{
893#define C(x) n->x = skb->x
894
895 n->next = n->prev = NULL;
896 n->sk = NULL;
897 __copy_skb_header(n, skb);
898
899 C(len);
900 C(data_len);
901 C(mac_len);
902 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
903 n->cloned = 1;
904 n->nohdr = 0;
905 n->destructor = NULL;
906 C(tail);
907 C(end);
908 C(head);
909 C(head_frag);
910 C(data);
911 C(truesize);
912 atomic_set(&n->users, 1);
913
914 atomic_inc(&(skb_shinfo(skb)->dataref));
915 skb->cloned = 1;
916
917 return n;
918#undef C
919}
920
921
922
923
924
925
926
927
928
929
930
931struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
932{
933 skb_release_all(dst);
934 return __skb_clone(dst, src);
935}
936EXPORT_SYMBOL_GPL(skb_morph);
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
954{
955 int i;
956 int num_frags = skb_shinfo(skb)->nr_frags;
957 struct page *page, *head = NULL;
958 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
959
960 for (i = 0; i < num_frags; i++) {
961 u8 *vaddr;
962 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
963
964 page = alloc_page(gfp_mask);
965 if (!page) {
966 while (head) {
967 struct page *next = (struct page *)page_private(head);
968 put_page(head);
969 head = next;
970 }
971 return -ENOMEM;
972 }
973 vaddr = kmap_atomic(skb_frag_page(f));
974 memcpy(page_address(page),
975 vaddr + f->page_offset, skb_frag_size(f));
976 kunmap_atomic(vaddr);
977 set_page_private(page, (unsigned long)head);
978 head = page;
979 }
980
981
982 for (i = 0; i < num_frags; i++)
983 skb_frag_unref(skb, i);
984
985 uarg->callback(uarg, false);
986
987
988 for (i = num_frags - 1; i >= 0; i--) {
989 __skb_fill_page_desc(skb, i, head, 0,
990 skb_shinfo(skb)->frags[i].size);
991 head = (struct page *)page_private(head);
992 }
993
994 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
995 return 0;
996}
997EXPORT_SYMBOL_GPL(skb_copy_ubufs);
998
999
1000
1001
1002
1003
1004
1005
1006
1007
1008
1009
1010
1011
1012
1013struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
1014{
1015 struct sk_buff_fclones *fclones = container_of(skb,
1016 struct sk_buff_fclones,
1017 skb1);
1018 struct sk_buff *n;
1019
1020 if (skb_orphan_frags(skb, gfp_mask))
1021 return NULL;
1022
1023 if (skb->fclone == SKB_FCLONE_ORIG &&
1024 atomic_read(&fclones->fclone_ref) == 1) {
1025 n = &fclones->skb2;
1026 atomic_set(&fclones->fclone_ref, 2);
1027 } else {
1028 if (skb_pfmemalloc(skb))
1029 gfp_mask |= __GFP_MEMALLOC;
1030
1031 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
1032 if (!n)
1033 return NULL;
1034
1035 kmemcheck_annotate_bitfield(n, flags1);
1036 n->fclone = SKB_FCLONE_UNAVAILABLE;
1037 }
1038
1039 return __skb_clone(n, skb);
1040}
1041EXPORT_SYMBOL(skb_clone);
1042
1043static void skb_headers_offset_update(struct sk_buff *skb, int off)
1044{
1045
1046 if (skb->ip_summed == CHECKSUM_PARTIAL)
1047 skb->csum_start += off;
1048
1049 skb->transport_header += off;
1050 skb->network_header += off;
1051 if (skb_mac_header_was_set(skb))
1052 skb->mac_header += off;
1053 skb->inner_transport_header += off;
1054 skb->inner_network_header += off;
1055 skb->inner_mac_header += off;
1056}
1057
1058static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
1059{
1060 __copy_skb_header(new, old);
1061
1062 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
1063 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
1064 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
1065}
1066
1067static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
1068{
1069 if (skb_pfmemalloc(skb))
1070 return SKB_ALLOC_RX;
1071 return 0;
1072}
1073
1074
1075
1076
1077
1078
1079
1080
1081
1082
1083
1084
1085
1086
1087
1088
1089
1090
1091struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
1092{
1093 int headerlen = skb_headroom(skb);
1094 unsigned int size = skb_end_offset(skb) + skb->data_len;
1095 struct sk_buff *n = __alloc_skb(size, gfp_mask,
1096 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
1097
1098 if (!n)
1099 return NULL;
1100
1101
1102 skb_reserve(n, headerlen);
1103
1104 skb_put(n, skb->len);
1105
1106 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
1107 BUG();
1108
1109 copy_skb_header(n, skb);
1110 return n;
1111}
1112EXPORT_SYMBOL(skb_copy);
1113
1114
1115
1116
1117
1118
1119
1120
1121
1122
1123
1124
1125
1126
1127
1128
1129
1130
1131struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1132 gfp_t gfp_mask, bool fclone)
1133{
1134 unsigned int size = skb_headlen(skb) + headroom;
1135 int flags = skb_alloc_rx_flag(skb) | (fclone ? SKB_ALLOC_FCLONE : 0);
1136 struct sk_buff *n = __alloc_skb(size, gfp_mask, flags, NUMA_NO_NODE);
1137
1138 if (!n)
1139 goto out;
1140
1141
1142 skb_reserve(n, headroom);
1143
1144 skb_put(n, skb_headlen(skb));
1145
1146 skb_copy_from_linear_data(skb, n->data, n->len);
1147
1148 n->truesize += skb->data_len;
1149 n->data_len = skb->data_len;
1150 n->len = skb->len;
1151
1152 if (skb_shinfo(skb)->nr_frags) {
1153 int i;
1154
1155 if (skb_orphan_frags(skb, gfp_mask)) {
1156 kfree_skb(n);
1157 n = NULL;
1158 goto out;
1159 }
1160 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1161 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1162 skb_frag_ref(skb, i);
1163 }
1164 skb_shinfo(n)->nr_frags = i;
1165 }
1166
1167 if (skb_has_frag_list(skb)) {
1168 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1169 skb_clone_fraglist(n);
1170 }
1171
1172 copy_skb_header(n, skb);
1173out:
1174 return n;
1175}
1176EXPORT_SYMBOL(__pskb_copy_fclone);
1177
1178
1179
1180
1181
1182
1183
1184
1185
1186
1187
1188
1189
1190
1191
1192
1193
1194int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1195 gfp_t gfp_mask)
1196{
1197 int i;
1198 u8 *data;
1199 int size = nhead + skb_end_offset(skb) + ntail;
1200 long off;
1201
1202 BUG_ON(nhead < 0);
1203
1204 if (skb_shared(skb))
1205 BUG();
1206
1207 size = SKB_DATA_ALIGN(size);
1208
1209 if (skb_pfmemalloc(skb))
1210 gfp_mask |= __GFP_MEMALLOC;
1211 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1212 gfp_mask, NUMA_NO_NODE, NULL);
1213 if (!data)
1214 goto nodata;
1215 size = SKB_WITH_OVERHEAD(ksize(data));
1216
1217
1218
1219
1220 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1221
1222 memcpy((struct skb_shared_info *)(data + size),
1223 skb_shinfo(skb),
1224 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1225
1226
1227
1228
1229
1230
1231 if (skb_cloned(skb)) {
1232
1233 if (skb_orphan_frags(skb, gfp_mask))
1234 goto nofrags;
1235 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1236 skb_frag_ref(skb, i);
1237
1238 if (skb_has_frag_list(skb))
1239 skb_clone_fraglist(skb);
1240
1241 skb_release_data(skb);
1242 } else {
1243 skb_free_head(skb);
1244 }
1245 off = (data + nhead) - skb->head;
1246
1247 skb->head = data;
1248 skb->head_frag = 0;
1249 skb->data += off;
1250#ifdef NET_SKBUFF_DATA_USES_OFFSET
1251 skb->end = size;
1252 off = nhead;
1253#else
1254 skb->end = skb->head + size;
1255#endif
1256 skb->tail += off;
1257 skb_headers_offset_update(skb, nhead);
1258 skb->cloned = 0;
1259 skb->hdr_len = 0;
1260 skb->nohdr = 0;
1261 atomic_set(&skb_shinfo(skb)->dataref, 1);
1262 return 0;
1263
1264nofrags:
1265 kfree(data);
1266nodata:
1267 return -ENOMEM;
1268}
1269EXPORT_SYMBOL(pskb_expand_head);
1270
1271
1272
1273struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1274{
1275 struct sk_buff *skb2;
1276 int delta = headroom - skb_headroom(skb);
1277
1278 if (delta <= 0)
1279 skb2 = pskb_copy(skb, GFP_ATOMIC);
1280 else {
1281 skb2 = skb_clone(skb, GFP_ATOMIC);
1282 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1283 GFP_ATOMIC)) {
1284 kfree_skb(skb2);
1285 skb2 = NULL;
1286 }
1287 }
1288 return skb2;
1289}
1290EXPORT_SYMBOL(skb_realloc_headroom);
1291
1292
1293
1294
1295
1296
1297
1298
1299
1300
1301
1302
1303
1304
1305
1306
1307
1308
1309
1310struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1311 int newheadroom, int newtailroom,
1312 gfp_t gfp_mask)
1313{
1314
1315
1316
1317 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1318 gfp_mask, skb_alloc_rx_flag(skb),
1319 NUMA_NO_NODE);
1320 int oldheadroom = skb_headroom(skb);
1321 int head_copy_len, head_copy_off;
1322
1323 if (!n)
1324 return NULL;
1325
1326 skb_reserve(n, newheadroom);
1327
1328
1329 skb_put(n, skb->len);
1330
1331 head_copy_len = oldheadroom;
1332 head_copy_off = 0;
1333 if (newheadroom <= head_copy_len)
1334 head_copy_len = newheadroom;
1335 else
1336 head_copy_off = newheadroom - head_copy_len;
1337
1338
1339 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1340 skb->len + head_copy_len))
1341 BUG();
1342
1343 copy_skb_header(n, skb);
1344
1345 skb_headers_offset_update(n, newheadroom - oldheadroom);
1346
1347 return n;
1348}
1349EXPORT_SYMBOL(skb_copy_expand);
1350
1351
1352
1353
1354
1355
1356
1357
1358
1359
1360
1361
1362
1363int skb_pad(struct sk_buff *skb, int pad)
1364{
1365 int err;
1366 int ntail;
1367
1368
1369 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1370 memset(skb->data+skb->len, 0, pad);
1371 return 0;
1372 }
1373
1374 ntail = skb->data_len + pad - (skb->end - skb->tail);
1375 if (likely(skb_cloned(skb) || ntail > 0)) {
1376 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1377 if (unlikely(err))
1378 goto free_skb;
1379 }
1380
1381
1382
1383
1384 err = skb_linearize(skb);
1385 if (unlikely(err))
1386 goto free_skb;
1387
1388 memset(skb->data + skb->len, 0, pad);
1389 return 0;
1390
1391free_skb:
1392 kfree_skb(skb);
1393 return err;
1394}
1395EXPORT_SYMBOL(skb_pad);
1396
1397
1398
1399
1400
1401
1402
1403
1404
1405
1406
1407
1408
1409
1410unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1411{
1412 if (tail != skb) {
1413 skb->data_len += len;
1414 skb->len += len;
1415 }
1416 return skb_put(tail, len);
1417}
1418EXPORT_SYMBOL_GPL(pskb_put);
1419
1420
1421
1422
1423
1424
1425
1426
1427
1428
1429unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1430{
1431 unsigned char *tmp = skb_tail_pointer(skb);
1432 SKB_LINEAR_ASSERT(skb);
1433 skb->tail += len;
1434 skb->len += len;
1435 if (unlikely(skb->tail > skb->end))
1436 skb_over_panic(skb, len, __builtin_return_address(0));
1437 return tmp;
1438}
1439EXPORT_SYMBOL(skb_put);
1440
1441
1442
1443
1444
1445
1446
1447
1448
1449
1450unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1451{
1452 skb->data -= len;
1453 skb->len += len;
1454 if (unlikely(skb->data<skb->head))
1455 skb_under_panic(skb, len, __builtin_return_address(0));
1456 return skb->data;
1457}
1458EXPORT_SYMBOL(skb_push);
1459
1460
1461
1462
1463
1464
1465
1466
1467
1468
1469
1470unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1471{
1472 return skb_pull_inline(skb, len);
1473}
1474EXPORT_SYMBOL(skb_pull);
1475
1476
1477
1478
1479
1480
1481
1482
1483
1484
1485void skb_trim(struct sk_buff *skb, unsigned int len)
1486{
1487 if (skb->len > len)
1488 __skb_trim(skb, len);
1489}
1490EXPORT_SYMBOL(skb_trim);
1491
1492
1493
1494
1495int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1496{
1497 struct sk_buff **fragp;
1498 struct sk_buff *frag;
1499 int offset = skb_headlen(skb);
1500 int nfrags = skb_shinfo(skb)->nr_frags;
1501 int i;
1502 int err;
1503
1504 if (skb_cloned(skb) &&
1505 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1506 return err;
1507
1508 i = 0;
1509 if (offset >= len)
1510 goto drop_pages;
1511
1512 for (; i < nfrags; i++) {
1513 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1514
1515 if (end < len) {
1516 offset = end;
1517 continue;
1518 }
1519
1520 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1521
1522drop_pages:
1523 skb_shinfo(skb)->nr_frags = i;
1524
1525 for (; i < nfrags; i++)
1526 skb_frag_unref(skb, i);
1527
1528 if (skb_has_frag_list(skb))
1529 skb_drop_fraglist(skb);
1530 goto done;
1531 }
1532
1533 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1534 fragp = &frag->next) {
1535 int end = offset + frag->len;
1536
1537 if (skb_shared(frag)) {
1538 struct sk_buff *nfrag;
1539
1540 nfrag = skb_clone(frag, GFP_ATOMIC);
1541 if (unlikely(!nfrag))
1542 return -ENOMEM;
1543
1544 nfrag->next = frag->next;
1545 consume_skb(frag);
1546 frag = nfrag;
1547 *fragp = frag;
1548 }
1549
1550 if (end < len) {
1551 offset = end;
1552 continue;
1553 }
1554
1555 if (end > len &&
1556 unlikely((err = pskb_trim(frag, len - offset))))
1557 return err;
1558
1559 if (frag->next)
1560 skb_drop_list(&frag->next);
1561 break;
1562 }
1563
1564done:
1565 if (len > skb_headlen(skb)) {
1566 skb->data_len -= skb->len - len;
1567 skb->len = len;
1568 } else {
1569 skb->len = len;
1570 skb->data_len = 0;
1571 skb_set_tail_pointer(skb, len);
1572 }
1573
1574 return 0;
1575}
1576EXPORT_SYMBOL(___pskb_trim);
1577
1578
1579
1580
1581
1582
1583
1584
1585
1586
1587
1588
1589
1590
1591
1592
1593
1594
1595
1596
1597
1598
1599
1600
1601
1602
1603unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1604{
1605
1606
1607
1608
1609 int i, k, eat = (skb->tail + delta) - skb->end;
1610
1611 if (eat > 0 || skb_cloned(skb)) {
1612 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1613 GFP_ATOMIC))
1614 return NULL;
1615 }
1616
1617 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1618 BUG();
1619
1620
1621
1622
1623 if (!skb_has_frag_list(skb))
1624 goto pull_pages;
1625
1626
1627 eat = delta;
1628 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1629 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1630
1631 if (size >= eat)
1632 goto pull_pages;
1633 eat -= size;
1634 }
1635
1636
1637
1638
1639
1640
1641
1642
1643 if (eat) {
1644 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1645 struct sk_buff *clone = NULL;
1646 struct sk_buff *insp = NULL;
1647
1648 do {
1649 BUG_ON(!list);
1650
1651 if (list->len <= eat) {
1652
1653 eat -= list->len;
1654 list = list->next;
1655 insp = list;
1656 } else {
1657
1658
1659 if (skb_shared(list)) {
1660
1661 clone = skb_clone(list, GFP_ATOMIC);
1662 if (!clone)
1663 return NULL;
1664 insp = list->next;
1665 list = clone;
1666 } else {
1667
1668
1669 insp = list;
1670 }
1671 if (!pskb_pull(list, eat)) {
1672 kfree_skb(clone);
1673 return NULL;
1674 }
1675 break;
1676 }
1677 } while (eat);
1678
1679
1680 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1681 skb_shinfo(skb)->frag_list = list->next;
1682 kfree_skb(list);
1683 }
1684
1685 if (clone) {
1686 clone->next = list;
1687 skb_shinfo(skb)->frag_list = clone;
1688 }
1689 }
1690
1691
1692pull_pages:
1693 eat = delta;
1694 k = 0;
1695 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1696 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1697
1698 if (size <= eat) {
1699 skb_frag_unref(skb, i);
1700 eat -= size;
1701 } else {
1702 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1703 if (eat) {
1704 skb_shinfo(skb)->frags[k].page_offset += eat;
1705 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1706 eat = 0;
1707 }
1708 k++;
1709 }
1710 }
1711 skb_shinfo(skb)->nr_frags = k;
1712
1713 skb->tail += delta;
1714 skb->data_len -= delta;
1715
1716 return skb_tail_pointer(skb);
1717}
1718EXPORT_SYMBOL(__pskb_pull_tail);
1719
1720
1721
1722
1723
1724
1725
1726
1727
1728
1729
1730
1731
1732
1733
1734
1735int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1736{
1737 int start = skb_headlen(skb);
1738 struct sk_buff *frag_iter;
1739 int i, copy;
1740
1741 if (offset > (int)skb->len - len)
1742 goto fault;
1743
1744
1745 if ((copy = start - offset) > 0) {
1746 if (copy > len)
1747 copy = len;
1748 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1749 if ((len -= copy) == 0)
1750 return 0;
1751 offset += copy;
1752 to += copy;
1753 }
1754
1755 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1756 int end;
1757 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1758
1759 WARN_ON(start > offset + len);
1760
1761 end = start + skb_frag_size(f);
1762 if ((copy = end - offset) > 0) {
1763 u8 *vaddr;
1764
1765 if (copy > len)
1766 copy = len;
1767
1768 vaddr = kmap_atomic(skb_frag_page(f));
1769 memcpy(to,
1770 vaddr + f->page_offset + offset - start,
1771 copy);
1772 kunmap_atomic(vaddr);
1773
1774 if ((len -= copy) == 0)
1775 return 0;
1776 offset += copy;
1777 to += copy;
1778 }
1779 start = end;
1780 }
1781
1782 skb_walk_frags(skb, frag_iter) {
1783 int end;
1784
1785 WARN_ON(start > offset + len);
1786
1787 end = start + frag_iter->len;
1788 if ((copy = end - offset) > 0) {
1789 if (copy > len)
1790 copy = len;
1791 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1792 goto fault;
1793 if ((len -= copy) == 0)
1794 return 0;
1795 offset += copy;
1796 to += copy;
1797 }
1798 start = end;
1799 }
1800
1801 if (!len)
1802 return 0;
1803
1804fault:
1805 return -EFAULT;
1806}
1807EXPORT_SYMBOL(skb_copy_bits);
1808
1809
1810
1811
1812
1813static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1814{
1815 put_page(spd->pages[i]);
1816}
1817
1818static struct page *linear_to_page(struct page *page, unsigned int *len,
1819 unsigned int *offset,
1820 struct sock *sk)
1821{
1822 struct page_frag *pfrag = sk_page_frag(sk);
1823
1824 if (!sk_page_frag_refill(sk, pfrag))
1825 return NULL;
1826
1827 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1828
1829 memcpy(page_address(pfrag->page) + pfrag->offset,
1830 page_address(page) + *offset, *len);
1831 *offset = pfrag->offset;
1832 pfrag->offset += *len;
1833
1834 return pfrag->page;
1835}
1836
1837static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1838 struct page *page,
1839 unsigned int offset)
1840{
1841 return spd->nr_pages &&
1842 spd->pages[spd->nr_pages - 1] == page &&
1843 (spd->partial[spd->nr_pages - 1].offset +
1844 spd->partial[spd->nr_pages - 1].len == offset);
1845}
1846
1847
1848
1849
1850static bool spd_fill_page(struct splice_pipe_desc *spd,
1851 struct pipe_inode_info *pipe, struct page *page,
1852 unsigned int *len, unsigned int offset,
1853 bool linear,
1854 struct sock *sk)
1855{
1856 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1857 return true;
1858
1859 if (linear) {
1860 page = linear_to_page(page, len, &offset, sk);
1861 if (!page)
1862 return true;
1863 }
1864 if (spd_can_coalesce(spd, page, offset)) {
1865 spd->partial[spd->nr_pages - 1].len += *len;
1866 return false;
1867 }
1868 get_page(page);
1869 spd->pages[spd->nr_pages] = page;
1870 spd->partial[spd->nr_pages].len = *len;
1871 spd->partial[spd->nr_pages].offset = offset;
1872 spd->nr_pages++;
1873
1874 return false;
1875}
1876
1877static bool __splice_segment(struct page *page, unsigned int poff,
1878 unsigned int plen, unsigned int *off,
1879 unsigned int *len,
1880 struct splice_pipe_desc *spd, bool linear,
1881 struct sock *sk,
1882 struct pipe_inode_info *pipe)
1883{
1884 if (!*len)
1885 return true;
1886
1887
1888 if (*off >= plen) {
1889 *off -= plen;
1890 return false;
1891 }
1892
1893
1894 poff += *off;
1895 plen -= *off;
1896 *off = 0;
1897
1898 do {
1899 unsigned int flen = min(*len, plen);
1900
1901 if (spd_fill_page(spd, pipe, page, &flen, poff,
1902 linear, sk))
1903 return true;
1904 poff += flen;
1905 plen -= flen;
1906 *len -= flen;
1907 } while (*len && plen);
1908
1909 return false;
1910}
1911
1912
1913
1914
1915
1916static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1917 unsigned int *offset, unsigned int *len,
1918 struct splice_pipe_desc *spd, struct sock *sk)
1919{
1920 int seg;
1921 struct sk_buff *iter;
1922
1923
1924
1925
1926
1927
1928 if (__splice_segment(virt_to_page(skb->data),
1929 (unsigned long) skb->data & (PAGE_SIZE - 1),
1930 skb_headlen(skb),
1931 offset, len, spd,
1932 skb_head_is_locked(skb),
1933 sk, pipe))
1934 return true;
1935
1936
1937
1938
1939 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1940 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1941
1942 if (__splice_segment(skb_frag_page(f),
1943 f->page_offset, skb_frag_size(f),
1944 offset, len, spd, false, sk, pipe))
1945 return true;
1946 }
1947
1948 skb_walk_frags(skb, iter) {
1949 if (*offset >= iter->len) {
1950 *offset -= iter->len;
1951 continue;
1952 }
1953
1954
1955
1956
1957 if (__skb_splice_bits(iter, pipe, offset, len, spd, sk))
1958 return true;
1959 }
1960
1961 return false;
1962}
1963
1964ssize_t skb_socket_splice(struct sock *sk,
1965 struct pipe_inode_info *pipe,
1966 struct splice_pipe_desc *spd)
1967{
1968 int ret;
1969
1970
1971
1972
1973
1974
1975
1976
1977
1978 release_sock(sk);
1979 ret = splice_to_pipe(pipe, spd);
1980 lock_sock(sk);
1981
1982 return ret;
1983}
1984
1985
1986
1987
1988
1989int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
1990 struct pipe_inode_info *pipe, unsigned int tlen,
1991 unsigned int flags,
1992 ssize_t (*splice_cb)(struct sock *,
1993 struct pipe_inode_info *,
1994 struct splice_pipe_desc *))
1995{
1996 struct partial_page partial[MAX_SKB_FRAGS];
1997 struct page *pages[MAX_SKB_FRAGS];
1998 struct splice_pipe_desc spd = {
1999 .pages = pages,
2000 .partial = partial,
2001 .nr_pages_max = MAX_SKB_FRAGS,
2002 .flags = flags,
2003 .ops = &nosteal_pipe_buf_ops,
2004 .spd_release = sock_spd_release,
2005 };
2006 int ret = 0;
2007
2008 __skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk);
2009
2010 if (spd.nr_pages)
2011 ret = splice_cb(sk, pipe, &spd);
2012
2013 return ret;
2014}
2015EXPORT_SYMBOL_GPL(skb_splice_bits);
2016
2017
2018
2019
2020
2021
2022
2023
2024
2025
2026
2027
2028
2029int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
2030{
2031 int start = skb_headlen(skb);
2032 struct sk_buff *frag_iter;
2033 int i, copy;
2034
2035 if (offset > (int)skb->len - len)
2036 goto fault;
2037
2038 if ((copy = start - offset) > 0) {
2039 if (copy > len)
2040 copy = len;
2041 skb_copy_to_linear_data_offset(skb, offset, from, copy);
2042 if ((len -= copy) == 0)
2043 return 0;
2044 offset += copy;
2045 from += copy;
2046 }
2047
2048 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2049 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2050 int end;
2051
2052 WARN_ON(start > offset + len);
2053
2054 end = start + skb_frag_size(frag);
2055 if ((copy = end - offset) > 0) {
2056 u8 *vaddr;
2057
2058 if (copy > len)
2059 copy = len;
2060
2061 vaddr = kmap_atomic(skb_frag_page(frag));
2062 memcpy(vaddr + frag->page_offset + offset - start,
2063 from, copy);
2064 kunmap_atomic(vaddr);
2065
2066 if ((len -= copy) == 0)
2067 return 0;
2068 offset += copy;
2069 from += copy;
2070 }
2071 start = end;
2072 }
2073
2074 skb_walk_frags(skb, frag_iter) {
2075 int end;
2076
2077 WARN_ON(start > offset + len);
2078
2079 end = start + frag_iter->len;
2080 if ((copy = end - offset) > 0) {
2081 if (copy > len)
2082 copy = len;
2083 if (skb_store_bits(frag_iter, offset - start,
2084 from, copy))
2085 goto fault;
2086 if ((len -= copy) == 0)
2087 return 0;
2088 offset += copy;
2089 from += copy;
2090 }
2091 start = end;
2092 }
2093 if (!len)
2094 return 0;
2095
2096fault:
2097 return -EFAULT;
2098}
2099EXPORT_SYMBOL(skb_store_bits);
2100
2101
2102__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2103 __wsum csum, const struct skb_checksum_ops *ops)
2104{
2105 int start = skb_headlen(skb);
2106 int i, copy = start - offset;
2107 struct sk_buff *frag_iter;
2108 int pos = 0;
2109
2110
2111 if (copy > 0) {
2112 if (copy > len)
2113 copy = len;
2114 csum = ops->update(skb->data + offset, copy, csum);
2115 if ((len -= copy) == 0)
2116 return csum;
2117 offset += copy;
2118 pos = copy;
2119 }
2120
2121 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2122 int end;
2123 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2124
2125 WARN_ON(start > offset + len);
2126
2127 end = start + skb_frag_size(frag);
2128 if ((copy = end - offset) > 0) {
2129 __wsum csum2;
2130 u8 *vaddr;
2131
2132 if (copy > len)
2133 copy = len;
2134 vaddr = kmap_atomic(skb_frag_page(frag));
2135 csum2 = ops->update(vaddr + frag->page_offset +
2136 offset - start, copy, 0);
2137 kunmap_atomic(vaddr);
2138 csum = ops->combine(csum, csum2, pos, copy);
2139 if (!(len -= copy))
2140 return csum;
2141 offset += copy;
2142 pos += copy;
2143 }
2144 start = end;
2145 }
2146
2147 skb_walk_frags(skb, frag_iter) {
2148 int end;
2149
2150 WARN_ON(start > offset + len);
2151
2152 end = start + frag_iter->len;
2153 if ((copy = end - offset) > 0) {
2154 __wsum csum2;
2155 if (copy > len)
2156 copy = len;
2157 csum2 = __skb_checksum(frag_iter, offset - start,
2158 copy, 0, ops);
2159 csum = ops->combine(csum, csum2, pos, copy);
2160 if ((len -= copy) == 0)
2161 return csum;
2162 offset += copy;
2163 pos += copy;
2164 }
2165 start = end;
2166 }
2167 BUG_ON(len);
2168
2169 return csum;
2170}
2171EXPORT_SYMBOL(__skb_checksum);
2172
2173__wsum skb_checksum(const struct sk_buff *skb, int offset,
2174 int len, __wsum csum)
2175{
2176 const struct skb_checksum_ops ops = {
2177 .update = csum_partial_ext,
2178 .combine = csum_block_add_ext,
2179 };
2180
2181 return __skb_checksum(skb, offset, len, csum, &ops);
2182}
2183EXPORT_SYMBOL(skb_checksum);
2184
2185
2186
2187__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2188 u8 *to, int len, __wsum csum)
2189{
2190 int start = skb_headlen(skb);
2191 int i, copy = start - offset;
2192 struct sk_buff *frag_iter;
2193 int pos = 0;
2194
2195
2196 if (copy > 0) {
2197 if (copy > len)
2198 copy = len;
2199 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2200 copy, csum);
2201 if ((len -= copy) == 0)
2202 return csum;
2203 offset += copy;
2204 to += copy;
2205 pos = copy;
2206 }
2207
2208 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2209 int end;
2210
2211 WARN_ON(start > offset + len);
2212
2213 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2214 if ((copy = end - offset) > 0) {
2215 __wsum csum2;
2216 u8 *vaddr;
2217 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2218
2219 if (copy > len)
2220 copy = len;
2221 vaddr = kmap_atomic(skb_frag_page(frag));
2222 csum2 = csum_partial_copy_nocheck(vaddr +
2223 frag->page_offset +
2224 offset - start, to,
2225 copy, 0);
2226 kunmap_atomic(vaddr);
2227 csum = csum_block_add(csum, csum2, pos);
2228 if (!(len -= copy))
2229 return csum;
2230 offset += copy;
2231 to += copy;
2232 pos += copy;
2233 }
2234 start = end;
2235 }
2236
2237 skb_walk_frags(skb, frag_iter) {
2238 __wsum csum2;
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 csum2 = skb_copy_and_csum_bits(frag_iter,
2248 offset - start,
2249 to, copy, 0);
2250 csum = csum_block_add(csum, csum2, pos);
2251 if ((len -= copy) == 0)
2252 return csum;
2253 offset += copy;
2254 to += copy;
2255 pos += copy;
2256 }
2257 start = end;
2258 }
2259 BUG_ON(len);
2260 return csum;
2261}
2262EXPORT_SYMBOL(skb_copy_and_csum_bits);
2263
2264
2265
2266
2267
2268
2269
2270
2271unsigned int
2272skb_zerocopy_headlen(const struct sk_buff *from)
2273{
2274 unsigned int hlen = 0;
2275
2276 if (!from->head_frag ||
2277 skb_headlen(from) < L1_CACHE_BYTES ||
2278 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2279 hlen = skb_headlen(from);
2280
2281 if (skb_has_frag_list(from))
2282 hlen = from->len;
2283
2284 return hlen;
2285}
2286EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2287
2288
2289
2290
2291
2292
2293
2294
2295
2296
2297
2298
2299
2300
2301
2302
2303
2304
2305
2306int
2307skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2308{
2309 int i, j = 0;
2310 int plen = 0;
2311 int ret;
2312 struct page *page;
2313 unsigned int offset;
2314
2315 BUG_ON(!from->head_frag && !hlen);
2316
2317
2318 if (len <= skb_tailroom(to))
2319 return skb_copy_bits(from, 0, skb_put(to, len), len);
2320
2321 if (hlen) {
2322 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2323 if (unlikely(ret))
2324 return ret;
2325 len -= hlen;
2326 } else {
2327 plen = min_t(int, skb_headlen(from), len);
2328 if (plen) {
2329 page = virt_to_head_page(from->head);
2330 offset = from->data - (unsigned char *)page_address(page);
2331 __skb_fill_page_desc(to, 0, page, offset, plen);
2332 get_page(page);
2333 j = 1;
2334 len -= plen;
2335 }
2336 }
2337
2338 to->truesize += len + plen;
2339 to->len += len + plen;
2340 to->data_len += len + plen;
2341
2342 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2343 skb_tx_error(from);
2344 return -ENOMEM;
2345 }
2346
2347 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2348 if (!len)
2349 break;
2350 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2351 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2352 len -= skb_shinfo(to)->frags[j].size;
2353 skb_frag_ref(to, j);
2354 j++;
2355 }
2356 skb_shinfo(to)->nr_frags = j;
2357
2358 return 0;
2359}
2360EXPORT_SYMBOL_GPL(skb_zerocopy);
2361
2362void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2363{
2364 __wsum csum;
2365 long csstart;
2366
2367 if (skb->ip_summed == CHECKSUM_PARTIAL)
2368 csstart = skb_checksum_start_offset(skb);
2369 else
2370 csstart = skb_headlen(skb);
2371
2372 BUG_ON(csstart > skb_headlen(skb));
2373
2374 skb_copy_from_linear_data(skb, to, csstart);
2375
2376 csum = 0;
2377 if (csstart != skb->len)
2378 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2379 skb->len - csstart, 0);
2380
2381 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2382 long csstuff = csstart + skb->csum_offset;
2383
2384 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2385 }
2386}
2387EXPORT_SYMBOL(skb_copy_and_csum_dev);
2388
2389
2390
2391
2392
2393
2394
2395
2396
2397
2398struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2399{
2400 unsigned long flags;
2401 struct sk_buff *result;
2402
2403 spin_lock_irqsave(&list->lock, flags);
2404 result = __skb_dequeue(list);
2405 spin_unlock_irqrestore(&list->lock, flags);
2406 return result;
2407}
2408EXPORT_SYMBOL(skb_dequeue);
2409
2410
2411
2412
2413
2414
2415
2416
2417
2418struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2419{
2420 unsigned long flags;
2421 struct sk_buff *result;
2422
2423 spin_lock_irqsave(&list->lock, flags);
2424 result = __skb_dequeue_tail(list);
2425 spin_unlock_irqrestore(&list->lock, flags);
2426 return result;
2427}
2428EXPORT_SYMBOL(skb_dequeue_tail);
2429
2430
2431
2432
2433
2434
2435
2436
2437
2438void skb_queue_purge(struct sk_buff_head *list)
2439{
2440 struct sk_buff *skb;
2441 while ((skb = skb_dequeue(list)) != NULL)
2442 kfree_skb(skb);
2443}
2444EXPORT_SYMBOL(skb_queue_purge);
2445
2446
2447
2448
2449
2450
2451
2452
2453
2454
2455
2456
2457void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2458{
2459 unsigned long flags;
2460
2461 spin_lock_irqsave(&list->lock, flags);
2462 __skb_queue_head(list, newsk);
2463 spin_unlock_irqrestore(&list->lock, flags);
2464}
2465EXPORT_SYMBOL(skb_queue_head);
2466
2467
2468
2469
2470
2471
2472
2473
2474
2475
2476
2477
2478void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2479{
2480 unsigned long flags;
2481
2482 spin_lock_irqsave(&list->lock, flags);
2483 __skb_queue_tail(list, newsk);
2484 spin_unlock_irqrestore(&list->lock, flags);
2485}
2486EXPORT_SYMBOL(skb_queue_tail);
2487
2488
2489
2490
2491
2492
2493
2494
2495
2496
2497
2498void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2499{
2500 unsigned long flags;
2501
2502 spin_lock_irqsave(&list->lock, flags);
2503 __skb_unlink(skb, list);
2504 spin_unlock_irqrestore(&list->lock, flags);
2505}
2506EXPORT_SYMBOL(skb_unlink);
2507
2508
2509
2510
2511
2512
2513
2514
2515
2516
2517
2518void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2519{
2520 unsigned long flags;
2521
2522 spin_lock_irqsave(&list->lock, flags);
2523 __skb_queue_after(list, old, newsk);
2524 spin_unlock_irqrestore(&list->lock, flags);
2525}
2526EXPORT_SYMBOL(skb_append);
2527
2528
2529
2530
2531
2532
2533
2534
2535
2536
2537
2538
2539
2540void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2541{
2542 unsigned long flags;
2543
2544 spin_lock_irqsave(&list->lock, flags);
2545 __skb_insert(newsk, old->prev, old, list);
2546 spin_unlock_irqrestore(&list->lock, flags);
2547}
2548EXPORT_SYMBOL(skb_insert);
2549
2550static inline void skb_split_inside_header(struct sk_buff *skb,
2551 struct sk_buff* skb1,
2552 const u32 len, const int pos)
2553{
2554 int i;
2555
2556 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2557 pos - len);
2558
2559 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2560 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2561
2562 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2563 skb_shinfo(skb)->nr_frags = 0;
2564 skb1->data_len = skb->data_len;
2565 skb1->len += skb1->data_len;
2566 skb->data_len = 0;
2567 skb->len = len;
2568 skb_set_tail_pointer(skb, len);
2569}
2570
2571static inline void skb_split_no_header(struct sk_buff *skb,
2572 struct sk_buff* skb1,
2573 const u32 len, int pos)
2574{
2575 int i, k = 0;
2576 const int nfrags = skb_shinfo(skb)->nr_frags;
2577
2578 skb_shinfo(skb)->nr_frags = 0;
2579 skb1->len = skb1->data_len = skb->len - len;
2580 skb->len = len;
2581 skb->data_len = len - pos;
2582
2583 for (i = 0; i < nfrags; i++) {
2584 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2585
2586 if (pos + size > len) {
2587 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2588
2589 if (pos < len) {
2590
2591
2592
2593
2594
2595
2596
2597
2598 skb_frag_ref(skb, i);
2599 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2600 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2601 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2602 skb_shinfo(skb)->nr_frags++;
2603 }
2604 k++;
2605 } else
2606 skb_shinfo(skb)->nr_frags++;
2607 pos += size;
2608 }
2609 skb_shinfo(skb1)->nr_frags = k;
2610}
2611
2612
2613
2614
2615
2616
2617
2618void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2619{
2620 int pos = skb_headlen(skb);
2621
2622 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2623 if (len < pos)
2624 skb_split_inside_header(skb, skb1, len, pos);
2625 else
2626 skb_split_no_header(skb, skb1, len, pos);
2627}
2628EXPORT_SYMBOL(skb_split);
2629
2630
2631
2632
2633
2634static int skb_prepare_for_shift(struct sk_buff *skb)
2635{
2636 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2637}
2638
2639
2640
2641
2642
2643
2644
2645
2646
2647
2648
2649
2650
2651
2652
2653
2654
2655
2656
2657int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2658{
2659 int from, to, merge, todo;
2660 struct skb_frag_struct *fragfrom, *fragto;
2661
2662 BUG_ON(shiftlen > skb->len);
2663 BUG_ON(skb_headlen(skb));
2664
2665 todo = shiftlen;
2666 from = 0;
2667 to = skb_shinfo(tgt)->nr_frags;
2668 fragfrom = &skb_shinfo(skb)->frags[from];
2669
2670
2671
2672
2673 if (!to ||
2674 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2675 fragfrom->page_offset)) {
2676 merge = -1;
2677 } else {
2678 merge = to - 1;
2679
2680 todo -= skb_frag_size(fragfrom);
2681 if (todo < 0) {
2682 if (skb_prepare_for_shift(skb) ||
2683 skb_prepare_for_shift(tgt))
2684 return 0;
2685
2686
2687 fragfrom = &skb_shinfo(skb)->frags[from];
2688 fragto = &skb_shinfo(tgt)->frags[merge];
2689
2690 skb_frag_size_add(fragto, shiftlen);
2691 skb_frag_size_sub(fragfrom, shiftlen);
2692 fragfrom->page_offset += shiftlen;
2693
2694 goto onlymerged;
2695 }
2696
2697 from++;
2698 }
2699
2700
2701 if ((shiftlen == skb->len) &&
2702 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2703 return 0;
2704
2705 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2706 return 0;
2707
2708 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2709 if (to == MAX_SKB_FRAGS)
2710 return 0;
2711
2712 fragfrom = &skb_shinfo(skb)->frags[from];
2713 fragto = &skb_shinfo(tgt)->frags[to];
2714
2715 if (todo >= skb_frag_size(fragfrom)) {
2716 *fragto = *fragfrom;
2717 todo -= skb_frag_size(fragfrom);
2718 from++;
2719 to++;
2720
2721 } else {
2722 __skb_frag_ref(fragfrom);
2723 fragto->page = fragfrom->page;
2724 fragto->page_offset = fragfrom->page_offset;
2725 skb_frag_size_set(fragto, todo);
2726
2727 fragfrom->page_offset += todo;
2728 skb_frag_size_sub(fragfrom, todo);
2729 todo = 0;
2730
2731 to++;
2732 break;
2733 }
2734 }
2735
2736
2737 skb_shinfo(tgt)->nr_frags = to;
2738
2739 if (merge >= 0) {
2740 fragfrom = &skb_shinfo(skb)->frags[0];
2741 fragto = &skb_shinfo(tgt)->frags[merge];
2742
2743 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2744 __skb_frag_unref(fragfrom);
2745 }
2746
2747
2748 to = 0;
2749 while (from < skb_shinfo(skb)->nr_frags)
2750 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2751 skb_shinfo(skb)->nr_frags = to;
2752
2753 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2754
2755onlymerged:
2756
2757
2758
2759 tgt->ip_summed = CHECKSUM_PARTIAL;
2760 skb->ip_summed = CHECKSUM_PARTIAL;
2761
2762
2763 skb->len -= shiftlen;
2764 skb->data_len -= shiftlen;
2765 skb->truesize -= shiftlen;
2766 tgt->len += shiftlen;
2767 tgt->data_len += shiftlen;
2768 tgt->truesize += shiftlen;
2769
2770 return shiftlen;
2771}
2772
2773
2774
2775
2776
2777
2778
2779
2780
2781
2782
2783void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2784 unsigned int to, struct skb_seq_state *st)
2785{
2786 st->lower_offset = from;
2787 st->upper_offset = to;
2788 st->root_skb = st->cur_skb = skb;
2789 st->frag_idx = st->stepped_offset = 0;
2790 st->frag_data = NULL;
2791}
2792EXPORT_SYMBOL(skb_prepare_seq_read);
2793
2794
2795
2796
2797
2798
2799
2800
2801
2802
2803
2804
2805
2806
2807
2808
2809
2810
2811
2812
2813
2814
2815
2816
2817
2818
2819unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2820 struct skb_seq_state *st)
2821{
2822 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2823 skb_frag_t *frag;
2824
2825 if (unlikely(abs_offset >= st->upper_offset)) {
2826 if (st->frag_data) {
2827 kunmap_atomic(st->frag_data);
2828 st->frag_data = NULL;
2829 }
2830 return 0;
2831 }
2832
2833next_skb:
2834 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2835
2836 if (abs_offset < block_limit && !st->frag_data) {
2837 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2838 return block_limit - abs_offset;
2839 }
2840
2841 if (st->frag_idx == 0 && !st->frag_data)
2842 st->stepped_offset += skb_headlen(st->cur_skb);
2843
2844 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2845 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2846 block_limit = skb_frag_size(frag) + st->stepped_offset;
2847
2848 if (abs_offset < block_limit) {
2849 if (!st->frag_data)
2850 st->frag_data = kmap_atomic(skb_frag_page(frag));
2851
2852 *data = (u8 *) st->frag_data + frag->page_offset +
2853 (abs_offset - st->stepped_offset);
2854
2855 return block_limit - abs_offset;
2856 }
2857
2858 if (st->frag_data) {
2859 kunmap_atomic(st->frag_data);
2860 st->frag_data = NULL;
2861 }
2862
2863 st->frag_idx++;
2864 st->stepped_offset += skb_frag_size(frag);
2865 }
2866
2867 if (st->frag_data) {
2868 kunmap_atomic(st->frag_data);
2869 st->frag_data = NULL;
2870 }
2871
2872 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2873 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2874 st->frag_idx = 0;
2875 goto next_skb;
2876 } else if (st->cur_skb->next) {
2877 st->cur_skb = st->cur_skb->next;
2878 st->frag_idx = 0;
2879 goto next_skb;
2880 }
2881
2882 return 0;
2883}
2884EXPORT_SYMBOL(skb_seq_read);
2885
2886
2887
2888
2889
2890
2891
2892
2893void skb_abort_seq_read(struct skb_seq_state *st)
2894{
2895 if (st->frag_data)
2896 kunmap_atomic(st->frag_data);
2897}
2898EXPORT_SYMBOL(skb_abort_seq_read);
2899
2900#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2901
2902static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2903 struct ts_config *conf,
2904 struct ts_state *state)
2905{
2906 return skb_seq_read(offset, text, TS_SKB_CB(state));
2907}
2908
2909static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2910{
2911 skb_abort_seq_read(TS_SKB_CB(state));
2912}
2913
2914
2915
2916
2917
2918
2919
2920
2921
2922
2923
2924
2925
2926unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2927 unsigned int to, struct ts_config *config)
2928{
2929 struct ts_state state;
2930 unsigned int ret;
2931
2932 config->get_next_block = skb_ts_get_next_block;
2933 config->finish = skb_ts_finish;
2934
2935 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(&state));
2936
2937 ret = textsearch_find(config, &state);
2938 return (ret <= to - from ? ret : UINT_MAX);
2939}
2940EXPORT_SYMBOL(skb_find_text);
2941
2942
2943
2944
2945
2946
2947
2948
2949
2950
2951
2952
2953int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2954 int (*getfrag)(void *from, char *to, int offset,
2955 int len, int odd, struct sk_buff *skb),
2956 void *from, int length)
2957{
2958 int frg_cnt = skb_shinfo(skb)->nr_frags;
2959 int copy;
2960 int offset = 0;
2961 int ret;
2962 struct page_frag *pfrag = ¤t->task_frag;
2963
2964 do {
2965
2966 if (frg_cnt >= MAX_SKB_FRAGS)
2967 return -EMSGSIZE;
2968
2969 if (!sk_page_frag_refill(sk, pfrag))
2970 return -ENOMEM;
2971
2972
2973 copy = min_t(int, length, pfrag->size - pfrag->offset);
2974
2975 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2976 offset, copy, 0, skb);
2977 if (ret < 0)
2978 return -EFAULT;
2979
2980
2981 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2982 copy);
2983 frg_cnt++;
2984 pfrag->offset += copy;
2985 get_page(pfrag->page);
2986
2987 skb->truesize += copy;
2988 atomic_add(copy, &sk->sk_wmem_alloc);
2989 skb->len += copy;
2990 skb->data_len += copy;
2991 offset += copy;
2992 length -= copy;
2993
2994 } while (length > 0);
2995
2996 return 0;
2997}
2998EXPORT_SYMBOL(skb_append_datato_frags);
2999
3000int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
3001 int offset, size_t size)
3002{
3003 int i = skb_shinfo(skb)->nr_frags;
3004
3005 if (skb_can_coalesce(skb, i, page, offset)) {
3006 skb_frag_size_add(&skb_shinfo(skb)->frags[i - 1], size);
3007 } else if (i < MAX_SKB_FRAGS) {
3008 get_page(page);
3009 skb_fill_page_desc(skb, i, page, offset, size);
3010 } else {
3011 return -EMSGSIZE;
3012 }
3013
3014 return 0;
3015}
3016EXPORT_SYMBOL_GPL(skb_append_pagefrags);
3017
3018
3019
3020
3021
3022
3023
3024
3025
3026
3027
3028
3029unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
3030{
3031 unsigned char *data = skb->data;
3032
3033 BUG_ON(len > skb->len);
3034 __skb_pull(skb, len);
3035 skb_postpull_rcsum(skb, data, len);
3036 return skb->data;
3037}
3038EXPORT_SYMBOL_GPL(skb_pull_rcsum);
3039
3040
3041
3042
3043
3044
3045
3046
3047
3048
3049struct sk_buff *skb_segment(struct sk_buff *head_skb,
3050 netdev_features_t features)
3051{
3052 struct sk_buff *segs = NULL;
3053 struct sk_buff *tail = NULL;
3054 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
3055 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
3056 unsigned int mss = skb_shinfo(head_skb)->gso_size;
3057 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
3058 struct sk_buff *frag_skb = head_skb;
3059 unsigned int offset = doffset;
3060 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
3061 unsigned int partial_segs = 0;
3062 unsigned int headroom;
3063 unsigned int len = head_skb->len;
3064 __be16 proto;
3065 bool csum, sg;
3066 int nfrags = skb_shinfo(head_skb)->nr_frags;
3067 int err = -ENOMEM;
3068 int i = 0;
3069 int pos;
3070 int dummy;
3071
3072 __skb_push(head_skb, doffset);
3073 proto = skb_network_protocol(head_skb, &dummy);
3074 if (unlikely(!proto))
3075 return ERR_PTR(-EINVAL);
3076
3077 sg = !!(features & NETIF_F_SG);
3078 csum = !!can_checksum_protocol(features, proto);
3079
3080
3081
3082
3083
3084 if (sg && csum && (features & NETIF_F_GSO_PARTIAL)) {
3085 partial_segs = len / mss;
3086 if (partial_segs > 1)
3087 mss *= partial_segs;
3088 else
3089 partial_segs = 0;
3090 }
3091
3092 headroom = skb_headroom(head_skb);
3093 pos = skb_headlen(head_skb);
3094
3095 do {
3096 struct sk_buff *nskb;
3097 skb_frag_t *nskb_frag;
3098 int hsize;
3099 int size;
3100
3101 len = head_skb->len - offset;
3102 if (len > mss)
3103 len = mss;
3104
3105 hsize = skb_headlen(head_skb) - offset;
3106 if (hsize < 0)
3107 hsize = 0;
3108 if (hsize > len || !sg)
3109 hsize = len;
3110
3111 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
3112 (skb_headlen(list_skb) == len || sg)) {
3113 BUG_ON(skb_headlen(list_skb) > len);
3114
3115 i = 0;
3116 nfrags = skb_shinfo(list_skb)->nr_frags;
3117 frag = skb_shinfo(list_skb)->frags;
3118 frag_skb = list_skb;
3119 pos += skb_headlen(list_skb);
3120
3121 while (pos < offset + len) {
3122 BUG_ON(i >= nfrags);
3123
3124 size = skb_frag_size(frag);
3125 if (pos + size > offset + len)
3126 break;
3127
3128 i++;
3129 pos += size;
3130 frag++;
3131 }
3132
3133 nskb = skb_clone(list_skb, GFP_ATOMIC);
3134 list_skb = list_skb->next;
3135
3136 if (unlikely(!nskb))
3137 goto err;
3138
3139 if (unlikely(pskb_trim(nskb, len))) {
3140 kfree_skb(nskb);
3141 goto err;
3142 }
3143
3144 hsize = skb_end_offset(nskb);
3145 if (skb_cow_head(nskb, doffset + headroom)) {
3146 kfree_skb(nskb);
3147 goto err;
3148 }
3149
3150 nskb->truesize += skb_end_offset(nskb) - hsize;
3151 skb_release_head_state(nskb);
3152 __skb_push(nskb, doffset);
3153 } else {
3154 nskb = __alloc_skb(hsize + doffset + headroom,
3155 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
3156 NUMA_NO_NODE);
3157
3158 if (unlikely(!nskb))
3159 goto err;
3160
3161 skb_reserve(nskb, headroom);
3162 __skb_put(nskb, doffset);
3163 }
3164
3165 if (segs)
3166 tail->next = nskb;
3167 else
3168 segs = nskb;
3169 tail = nskb;
3170
3171 __copy_skb_header(nskb, head_skb);
3172
3173 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
3174 skb_reset_mac_len(nskb);
3175
3176 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
3177 nskb->data - tnl_hlen,
3178 doffset + tnl_hlen);
3179
3180 if (nskb->len == len + doffset)
3181 goto perform_csum_check;
3182
3183 if (!sg) {
3184 if (!nskb->remcsum_offload)
3185 nskb->ip_summed = CHECKSUM_NONE;
3186 SKB_GSO_CB(nskb)->csum =
3187 skb_copy_and_csum_bits(head_skb, offset,
3188 skb_put(nskb, len),
3189 len, 0);
3190 SKB_GSO_CB(nskb)->csum_start =
3191 skb_headroom(nskb) + doffset;
3192 continue;
3193 }
3194
3195 nskb_frag = skb_shinfo(nskb)->frags;
3196
3197 skb_copy_from_linear_data_offset(head_skb, offset,
3198 skb_put(nskb, hsize), hsize);
3199
3200 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
3201 SKBTX_SHARED_FRAG;
3202
3203 while (pos < offset + len) {
3204 if (i >= nfrags) {
3205 BUG_ON(skb_headlen(list_skb));
3206
3207 i = 0;
3208 nfrags = skb_shinfo(list_skb)->nr_frags;
3209 frag = skb_shinfo(list_skb)->frags;
3210 frag_skb = list_skb;
3211
3212 BUG_ON(!nfrags);
3213
3214 list_skb = list_skb->next;
3215 }
3216
3217 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3218 MAX_SKB_FRAGS)) {
3219 net_warn_ratelimited(
3220 "skb_segment: too many frags: %u %u\n",
3221 pos, mss);
3222 goto err;
3223 }
3224
3225 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3226 goto err;
3227
3228 *nskb_frag = *frag;
3229 __skb_frag_ref(nskb_frag);
3230 size = skb_frag_size(nskb_frag);
3231
3232 if (pos < offset) {
3233 nskb_frag->page_offset += offset - pos;
3234 skb_frag_size_sub(nskb_frag, offset - pos);
3235 }
3236
3237 skb_shinfo(nskb)->nr_frags++;
3238
3239 if (pos + size <= offset + len) {
3240 i++;
3241 frag++;
3242 pos += size;
3243 } else {
3244 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3245 goto skip_fraglist;
3246 }
3247
3248 nskb_frag++;
3249 }
3250
3251skip_fraglist:
3252 nskb->data_len = len - hsize;
3253 nskb->len += nskb->data_len;
3254 nskb->truesize += nskb->data_len;
3255
3256perform_csum_check:
3257 if (!csum) {
3258 if (skb_has_shared_frag(nskb)) {
3259 err = __skb_linearize(nskb);
3260 if (err)
3261 goto err;
3262 }
3263 if (!nskb->remcsum_offload)
3264 nskb->ip_summed = CHECKSUM_NONE;
3265 SKB_GSO_CB(nskb)->csum =
3266 skb_checksum(nskb, doffset,
3267 nskb->len - doffset, 0);
3268 SKB_GSO_CB(nskb)->csum_start =
3269 skb_headroom(nskb) + doffset;
3270 }
3271 } while ((offset += len) < head_skb->len);
3272
3273
3274
3275
3276
3277 segs->prev = tail;
3278
3279
3280 if (partial_segs) {
3281 int type = skb_shinfo(head_skb)->gso_type;
3282
3283
3284 type |= SKB_GSO_PARTIAL;
3285 type &= ~SKB_GSO_DODGY;
3286
3287
3288
3289
3290 skb_shinfo(segs)->gso_size = skb_shinfo(head_skb)->gso_size;
3291 skb_shinfo(segs)->gso_segs = partial_segs;
3292 skb_shinfo(segs)->gso_type = type;
3293 SKB_GSO_CB(segs)->data_offset = skb_headroom(segs) + doffset;
3294 }
3295
3296
3297
3298
3299
3300 if (head_skb->destructor == sock_wfree) {
3301 swap(tail->truesize, head_skb->truesize);
3302 swap(tail->destructor, head_skb->destructor);
3303 swap(tail->sk, head_skb->sk);
3304 }
3305 return segs;
3306
3307err:
3308 kfree_skb_list(segs);
3309 return ERR_PTR(err);
3310}
3311EXPORT_SYMBOL_GPL(skb_segment);
3312
3313int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3314{
3315 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3316 unsigned int offset = skb_gro_offset(skb);
3317 unsigned int headlen = skb_headlen(skb);
3318 unsigned int len = skb_gro_len(skb);
3319 struct sk_buff *lp, *p = *head;
3320 unsigned int delta_truesize;
3321
3322 if (unlikely(p->len + len >= 65536))
3323 return -E2BIG;
3324
3325 lp = NAPI_GRO_CB(p)->last;
3326 pinfo = skb_shinfo(lp);
3327
3328 if (headlen <= offset) {
3329 skb_frag_t *frag;
3330 skb_frag_t *frag2;
3331 int i = skbinfo->nr_frags;
3332 int nr_frags = pinfo->nr_frags + i;
3333
3334 if (nr_frags > MAX_SKB_FRAGS)
3335 goto merge;
3336
3337 offset -= headlen;
3338 pinfo->nr_frags = nr_frags;
3339 skbinfo->nr_frags = 0;
3340
3341 frag = pinfo->frags + nr_frags;
3342 frag2 = skbinfo->frags + i;
3343 do {
3344 *--frag = *--frag2;
3345 } while (--i);
3346
3347 frag->page_offset += offset;
3348 skb_frag_size_sub(frag, offset);
3349
3350
3351 delta_truesize = skb->truesize -
3352 SKB_TRUESIZE(skb_end_offset(skb));
3353
3354 skb->truesize -= skb->data_len;
3355 skb->len -= skb->data_len;
3356 skb->data_len = 0;
3357
3358 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3359 goto done;
3360 } else if (skb->head_frag) {
3361 int nr_frags = pinfo->nr_frags;
3362 skb_frag_t *frag = pinfo->frags + nr_frags;
3363 struct page *page = virt_to_head_page(skb->head);
3364 unsigned int first_size = headlen - offset;
3365 unsigned int first_offset;
3366
3367 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3368 goto merge;
3369
3370 first_offset = skb->data -
3371 (unsigned char *)page_address(page) +
3372 offset;
3373
3374 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3375
3376 frag->page.p = page;
3377 frag->page_offset = first_offset;
3378 skb_frag_size_set(frag, first_size);
3379
3380 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3381
3382
3383 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3384 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3385 goto done;
3386 }
3387
3388merge:
3389 delta_truesize = skb->truesize;
3390 if (offset > headlen) {
3391 unsigned int eat = offset - headlen;
3392
3393 skbinfo->frags[0].page_offset += eat;
3394 skb_frag_size_sub(&skbinfo->frags[0], eat);
3395 skb->data_len -= eat;
3396 skb->len -= eat;
3397 offset = headlen;
3398 }
3399
3400 __skb_pull(skb, offset);
3401
3402 if (NAPI_GRO_CB(p)->last == p)
3403 skb_shinfo(p)->frag_list = skb;
3404 else
3405 NAPI_GRO_CB(p)->last->next = skb;
3406 NAPI_GRO_CB(p)->last = skb;
3407 __skb_header_release(skb);
3408 lp = p;
3409
3410done:
3411 NAPI_GRO_CB(p)->count++;
3412 p->data_len += len;
3413 p->truesize += delta_truesize;
3414 p->len += len;
3415 if (lp != p) {
3416 lp->data_len += len;
3417 lp->truesize += delta_truesize;
3418 lp->len += len;
3419 }
3420 NAPI_GRO_CB(skb)->same_flow = 1;
3421 return 0;
3422}
3423
3424void __init skb_init(void)
3425{
3426 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3427 sizeof(struct sk_buff),
3428 0,
3429 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3430 NULL);
3431 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3432 sizeof(struct sk_buff_fclones),
3433 0,
3434 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3435 NULL);
3436}
3437
3438
3439
3440
3441
3442
3443
3444
3445
3446
3447
3448static int
3449__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3450{
3451 int start = skb_headlen(skb);
3452 int i, copy = start - offset;
3453 struct sk_buff *frag_iter;
3454 int elt = 0;
3455
3456 if (copy > 0) {
3457 if (copy > len)
3458 copy = len;
3459 sg_set_buf(sg, skb->data + offset, copy);
3460 elt++;
3461 if ((len -= copy) == 0)
3462 return elt;
3463 offset += copy;
3464 }
3465
3466 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3467 int end;
3468
3469 WARN_ON(start > offset + len);
3470
3471 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3472 if ((copy = end - offset) > 0) {
3473 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3474
3475 if (copy > len)
3476 copy = len;
3477 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3478 frag->page_offset+offset-start);
3479 elt++;
3480 if (!(len -= copy))
3481 return elt;
3482 offset += copy;
3483 }
3484 start = end;
3485 }
3486
3487 skb_walk_frags(skb, frag_iter) {
3488 int end;
3489
3490 WARN_ON(start > offset + len);
3491
3492 end = start + frag_iter->len;
3493 if ((copy = end - offset) > 0) {
3494 if (copy > len)
3495 copy = len;
3496 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3497 copy);
3498 if ((len -= copy) == 0)
3499 return elt;
3500 offset += copy;
3501 }
3502 start = end;
3503 }
3504 BUG_ON(len);
3505 return elt;
3506}
3507
3508
3509
3510
3511
3512
3513
3514
3515
3516
3517
3518
3519
3520
3521
3522
3523
3524
3525
3526
3527int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3528 int offset, int len)
3529{
3530 return __skb_to_sgvec(skb, sg, offset, len);
3531}
3532EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3533
3534int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3535{
3536 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3537
3538 sg_mark_end(&sg[nsg - 1]);
3539
3540 return nsg;
3541}
3542EXPORT_SYMBOL_GPL(skb_to_sgvec);
3543
3544
3545
3546
3547
3548
3549
3550
3551
3552
3553
3554
3555
3556
3557
3558
3559
3560
3561int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3562{
3563 int copyflag;
3564 int elt;
3565 struct sk_buff *skb1, **skb_p;
3566
3567
3568
3569
3570
3571 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3572 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3573 return -ENOMEM;
3574
3575
3576 if (!skb_has_frag_list(skb)) {
3577
3578
3579
3580
3581
3582 if (skb_tailroom(skb) < tailbits &&
3583 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3584 return -ENOMEM;
3585
3586
3587 *trailer = skb;
3588 return 1;
3589 }
3590
3591
3592
3593 elt = 1;
3594 skb_p = &skb_shinfo(skb)->frag_list;
3595 copyflag = 0;
3596
3597 while ((skb1 = *skb_p) != NULL) {
3598 int ntail = 0;
3599
3600
3601
3602
3603
3604 if (skb_shared(skb1))
3605 copyflag = 1;
3606
3607
3608
3609 if (skb1->next == NULL && tailbits) {
3610 if (skb_shinfo(skb1)->nr_frags ||
3611 skb_has_frag_list(skb1) ||
3612 skb_tailroom(skb1) < tailbits)
3613 ntail = tailbits + 128;
3614 }
3615
3616 if (copyflag ||
3617 skb_cloned(skb1) ||
3618 ntail ||
3619 skb_shinfo(skb1)->nr_frags ||
3620 skb_has_frag_list(skb1)) {
3621 struct sk_buff *skb2;
3622
3623
3624 if (ntail == 0)
3625 skb2 = skb_copy(skb1, GFP_ATOMIC);
3626 else
3627 skb2 = skb_copy_expand(skb1,
3628 skb_headroom(skb1),
3629 ntail,
3630 GFP_ATOMIC);
3631 if (unlikely(skb2 == NULL))
3632 return -ENOMEM;
3633
3634 if (skb1->sk)
3635 skb_set_owner_w(skb2, skb1->sk);
3636
3637
3638
3639
3640 skb2->next = skb1->next;
3641 *skb_p = skb2;
3642 kfree_skb(skb1);
3643 skb1 = skb2;
3644 }
3645 elt++;
3646 *trailer = skb1;
3647 skb_p = &skb1->next;
3648 }
3649
3650 return elt;
3651}
3652EXPORT_SYMBOL_GPL(skb_cow_data);
3653
3654static void sock_rmem_free(struct sk_buff *skb)
3655{
3656 struct sock *sk = skb->sk;
3657
3658 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3659}
3660
3661
3662
3663
3664int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3665{
3666 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3667 (unsigned int)sk->sk_rcvbuf)
3668 return -ENOMEM;
3669
3670 skb_orphan(skb);
3671 skb->sk = sk;
3672 skb->destructor = sock_rmem_free;
3673 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3674
3675
3676 skb_dst_force(skb);
3677
3678 skb_queue_tail(&sk->sk_error_queue, skb);
3679 if (!sock_flag(sk, SOCK_DEAD))
3680 sk->sk_data_ready(sk);
3681 return 0;
3682}
3683EXPORT_SYMBOL(sock_queue_err_skb);
3684
3685struct sk_buff *sock_dequeue_err_skb(struct sock *sk)
3686{
3687 struct sk_buff_head *q = &sk->sk_error_queue;
3688 struct sk_buff *skb, *skb_next;
3689 unsigned long flags;
3690 int err = 0;
3691
3692 spin_lock_irqsave(&q->lock, flags);
3693 skb = __skb_dequeue(q);
3694 if (skb && (skb_next = skb_peek(q)))
3695 err = SKB_EXT_ERR(skb_next)->ee.ee_errno;
3696 spin_unlock_irqrestore(&q->lock, flags);
3697
3698 sk->sk_err = err;
3699 if (err)
3700 sk->sk_error_report(sk);
3701
3702 return skb;
3703}
3704EXPORT_SYMBOL(sock_dequeue_err_skb);
3705
3706
3707
3708
3709
3710
3711
3712
3713
3714
3715
3716
3717
3718
3719struct sk_buff *skb_clone_sk(struct sk_buff *skb)
3720{
3721 struct sock *sk = skb->sk;
3722 struct sk_buff *clone;
3723
3724 if (!sk || !atomic_inc_not_zero(&sk->sk_refcnt))
3725 return NULL;
3726
3727 clone = skb_clone(skb, GFP_ATOMIC);
3728 if (!clone) {
3729 sock_put(sk);
3730 return NULL;
3731 }
3732
3733 clone->sk = sk;
3734 clone->destructor = sock_efree;
3735
3736 return clone;
3737}
3738EXPORT_SYMBOL(skb_clone_sk);
3739
3740static void __skb_complete_tx_timestamp(struct sk_buff *skb,
3741 struct sock *sk,
3742 int tstype)
3743{
3744 struct sock_exterr_skb *serr;
3745 int err;
3746
3747 serr = SKB_EXT_ERR(skb);
3748 memset(serr, 0, sizeof(*serr));
3749 serr->ee.ee_errno = ENOMSG;
3750 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3751 serr->ee.ee_info = tstype;
3752 if (sk->sk_tsflags & SOF_TIMESTAMPING_OPT_ID) {
3753 serr->ee.ee_data = skb_shinfo(skb)->tskey;
3754 if (sk->sk_protocol == IPPROTO_TCP &&
3755 sk->sk_type == SOCK_STREAM)
3756 serr->ee.ee_data -= sk->sk_tskey;
3757 }
3758
3759 err = sock_queue_err_skb(sk, skb);
3760
3761 if (err)
3762 kfree_skb(skb);
3763}
3764
3765static bool skb_may_tx_timestamp(struct sock *sk, bool tsonly)
3766{
3767 bool ret;
3768
3769 if (likely(sysctl_tstamp_allow_data || tsonly))
3770 return true;
3771
3772 read_lock_bh(&sk->sk_callback_lock);
3773 ret = sk->sk_socket && sk->sk_socket->file &&
3774 file_ns_capable(sk->sk_socket->file, &init_user_ns, CAP_NET_RAW);
3775 read_unlock_bh(&sk->sk_callback_lock);
3776 return ret;
3777}
3778
3779void skb_complete_tx_timestamp(struct sk_buff *skb,
3780 struct skb_shared_hwtstamps *hwtstamps)
3781{
3782 struct sock *sk = skb->sk;
3783
3784 if (!skb_may_tx_timestamp(sk, false))
3785 return;
3786
3787
3788 sock_hold(sk);
3789
3790 *skb_hwtstamps(skb) = *hwtstamps;
3791 __skb_complete_tx_timestamp(skb, sk, SCM_TSTAMP_SND);
3792
3793 sock_put(sk);
3794}
3795EXPORT_SYMBOL_GPL(skb_complete_tx_timestamp);
3796
3797void __skb_tstamp_tx(struct sk_buff *orig_skb,
3798 struct skb_shared_hwtstamps *hwtstamps,
3799 struct sock *sk, int tstype)
3800{
3801 struct sk_buff *skb;
3802 bool tsonly;
3803
3804 if (!sk)
3805 return;
3806
3807 tsonly = sk->sk_tsflags & SOF_TIMESTAMPING_OPT_TSONLY;
3808 if (!skb_may_tx_timestamp(sk, tsonly))
3809 return;
3810
3811 if (tsonly)
3812 skb = alloc_skb(0, GFP_ATOMIC);
3813 else
3814 skb = skb_clone(orig_skb, GFP_ATOMIC);
3815 if (!skb)
3816 return;
3817
3818 if (tsonly) {
3819 skb_shinfo(skb)->tx_flags = skb_shinfo(orig_skb)->tx_flags;
3820 skb_shinfo(skb)->tskey = skb_shinfo(orig_skb)->tskey;
3821 }
3822
3823 if (hwtstamps)
3824 *skb_hwtstamps(skb) = *hwtstamps;
3825 else
3826 skb->tstamp = ktime_get_real();
3827
3828 __skb_complete_tx_timestamp(skb, sk, tstype);
3829}
3830EXPORT_SYMBOL_GPL(__skb_tstamp_tx);
3831
3832void skb_tstamp_tx(struct sk_buff *orig_skb,
3833 struct skb_shared_hwtstamps *hwtstamps)
3834{
3835 return __skb_tstamp_tx(orig_skb, hwtstamps, orig_skb->sk,
3836 SCM_TSTAMP_SND);
3837}
3838EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3839
3840void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3841{
3842 struct sock *sk = skb->sk;
3843 struct sock_exterr_skb *serr;
3844 int err;
3845
3846 skb->wifi_acked_valid = 1;
3847 skb->wifi_acked = acked;
3848
3849 serr = SKB_EXT_ERR(skb);
3850 memset(serr, 0, sizeof(*serr));
3851 serr->ee.ee_errno = ENOMSG;
3852 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3853
3854
3855 sock_hold(sk);
3856
3857 err = sock_queue_err_skb(sk, skb);
3858 if (err)
3859 kfree_skb(skb);
3860
3861 sock_put(sk);
3862}
3863EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3864
3865
3866
3867
3868
3869
3870
3871
3872
3873
3874
3875
3876
3877bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3878{
3879 if (unlikely(start > skb_headlen(skb)) ||
3880 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3881 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3882 start, off, skb_headlen(skb));
3883 return false;
3884 }
3885 skb->ip_summed = CHECKSUM_PARTIAL;
3886 skb->csum_start = skb_headroom(skb) + start;
3887 skb->csum_offset = off;
3888 skb_set_transport_header(skb, start);
3889 return true;
3890}
3891EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3892
3893static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3894 unsigned int max)
3895{
3896 if (skb_headlen(skb) >= len)
3897 return 0;
3898
3899
3900
3901
3902 if (max > skb->len)
3903 max = skb->len;
3904
3905 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3906 return -ENOMEM;
3907
3908 if (skb_headlen(skb) < len)
3909 return -EPROTO;
3910
3911 return 0;
3912}
3913
3914#define MAX_TCP_HDR_LEN (15 * 4)
3915
3916static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3917 typeof(IPPROTO_IP) proto,
3918 unsigned int off)
3919{
3920 switch (proto) {
3921 int err;
3922
3923 case IPPROTO_TCP:
3924 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3925 off + MAX_TCP_HDR_LEN);
3926 if (!err && !skb_partial_csum_set(skb, off,
3927 offsetof(struct tcphdr,
3928 check)))
3929 err = -EPROTO;
3930 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3931
3932 case IPPROTO_UDP:
3933 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3934 off + sizeof(struct udphdr));
3935 if (!err && !skb_partial_csum_set(skb, off,
3936 offsetof(struct udphdr,
3937 check)))
3938 err = -EPROTO;
3939 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3940 }
3941
3942 return ERR_PTR(-EPROTO);
3943}
3944
3945
3946
3947
3948#define MAX_IP_HDR_LEN 128
3949
3950static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3951{
3952 unsigned int off;
3953 bool fragment;
3954 __sum16 *csum;
3955 int err;
3956
3957 fragment = false;
3958
3959 err = skb_maybe_pull_tail(skb,
3960 sizeof(struct iphdr),
3961 MAX_IP_HDR_LEN);
3962 if (err < 0)
3963 goto out;
3964
3965 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3966 fragment = true;
3967
3968 off = ip_hdrlen(skb);
3969
3970 err = -EPROTO;
3971
3972 if (fragment)
3973 goto out;
3974
3975 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3976 if (IS_ERR(csum))
3977 return PTR_ERR(csum);
3978
3979 if (recalculate)
3980 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3981 ip_hdr(skb)->daddr,
3982 skb->len - off,
3983 ip_hdr(skb)->protocol, 0);
3984 err = 0;
3985
3986out:
3987 return err;
3988}
3989
3990
3991
3992
3993#define MAX_IPV6_HDR_LEN 256
3994
3995#define OPT_HDR(type, skb, off) \
3996 (type *)(skb_network_header(skb) + (off))
3997
3998static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3999{
4000 int err;
4001 u8 nexthdr;
4002 unsigned int off;
4003 unsigned int len;
4004 bool fragment;
4005 bool done;
4006 __sum16 *csum;
4007
4008 fragment = false;
4009 done = false;
4010
4011 off = sizeof(struct ipv6hdr);
4012
4013 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
4014 if (err < 0)
4015 goto out;
4016
4017 nexthdr = ipv6_hdr(skb)->nexthdr;
4018
4019 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
4020 while (off <= len && !done) {
4021 switch (nexthdr) {
4022 case IPPROTO_DSTOPTS:
4023 case IPPROTO_HOPOPTS:
4024 case IPPROTO_ROUTING: {
4025 struct ipv6_opt_hdr *hp;
4026
4027 err = skb_maybe_pull_tail(skb,
4028 off +
4029 sizeof(struct ipv6_opt_hdr),
4030 MAX_IPV6_HDR_LEN);
4031 if (err < 0)
4032 goto out;
4033
4034 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
4035 nexthdr = hp->nexthdr;
4036 off += ipv6_optlen(hp);
4037 break;
4038 }
4039 case IPPROTO_AH: {
4040 struct ip_auth_hdr *hp;
4041
4042 err = skb_maybe_pull_tail(skb,
4043 off +
4044 sizeof(struct ip_auth_hdr),
4045 MAX_IPV6_HDR_LEN);
4046 if (err < 0)
4047 goto out;
4048
4049 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
4050 nexthdr = hp->nexthdr;
4051 off += ipv6_authlen(hp);
4052 break;
4053 }
4054 case IPPROTO_FRAGMENT: {
4055 struct frag_hdr *hp;
4056
4057 err = skb_maybe_pull_tail(skb,
4058 off +
4059 sizeof(struct frag_hdr),
4060 MAX_IPV6_HDR_LEN);
4061 if (err < 0)
4062 goto out;
4063
4064 hp = OPT_HDR(struct frag_hdr, skb, off);
4065
4066 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
4067 fragment = true;
4068
4069 nexthdr = hp->nexthdr;
4070 off += sizeof(struct frag_hdr);
4071 break;
4072 }
4073 default:
4074 done = true;
4075 break;
4076 }
4077 }
4078
4079 err = -EPROTO;
4080
4081 if (!done || fragment)
4082 goto out;
4083
4084 csum = skb_checksum_setup_ip(skb, nexthdr, off);
4085 if (IS_ERR(csum))
4086 return PTR_ERR(csum);
4087
4088 if (recalculate)
4089 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
4090 &ipv6_hdr(skb)->daddr,
4091 skb->len - off, nexthdr, 0);
4092 err = 0;
4093
4094out:
4095 return err;
4096}
4097
4098
4099
4100
4101
4102
4103int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
4104{
4105 int err;
4106
4107 switch (skb->protocol) {
4108 case htons(ETH_P_IP):
4109 err = skb_checksum_setup_ipv4(skb, recalculate);
4110 break;
4111
4112 case htons(ETH_P_IPV6):
4113 err = skb_checksum_setup_ipv6(skb, recalculate);
4114 break;
4115
4116 default:
4117 err = -EPROTO;
4118 break;
4119 }
4120
4121 return err;
4122}
4123EXPORT_SYMBOL(skb_checksum_setup);
4124
4125
4126
4127
4128
4129
4130
4131
4132
4133
4134
4135
4136
4137
4138static struct sk_buff *skb_checksum_maybe_trim(struct sk_buff *skb,
4139 unsigned int transport_len)
4140{
4141 struct sk_buff *skb_chk;
4142 unsigned int len = skb_transport_offset(skb) + transport_len;
4143 int ret;
4144
4145 if (skb->len < len)
4146 return NULL;
4147 else if (skb->len == len)
4148 return skb;
4149
4150 skb_chk = skb_clone(skb, GFP_ATOMIC);
4151 if (!skb_chk)
4152 return NULL;
4153
4154 ret = pskb_trim_rcsum(skb_chk, len);
4155 if (ret) {
4156 kfree_skb(skb_chk);
4157 return NULL;
4158 }
4159
4160 return skb_chk;
4161}
4162
4163
4164
4165
4166
4167
4168
4169
4170
4171
4172
4173
4174
4175
4176
4177
4178struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4179 unsigned int transport_len,
4180 __sum16(*skb_chkf)(struct sk_buff *skb))
4181{
4182 struct sk_buff *skb_chk;
4183 unsigned int offset = skb_transport_offset(skb);
4184 __sum16 ret;
4185
4186 skb_chk = skb_checksum_maybe_trim(skb, transport_len);
4187 if (!skb_chk)
4188 goto err;
4189
4190 if (!pskb_may_pull(skb_chk, offset))
4191 goto err;
4192
4193 skb_pull_rcsum(skb_chk, offset);
4194 ret = skb_chkf(skb_chk);
4195 skb_push_rcsum(skb_chk, offset);
4196
4197 if (ret)
4198 goto err;
4199
4200 return skb_chk;
4201
4202err:
4203 if (skb_chk && skb_chk != skb)
4204 kfree_skb(skb_chk);
4205
4206 return NULL;
4207
4208}
4209EXPORT_SYMBOL(skb_checksum_trimmed);
4210
4211void __skb_warn_lro_forwarding(const struct sk_buff *skb)
4212{
4213 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
4214 skb->dev->name);
4215}
4216EXPORT_SYMBOL(__skb_warn_lro_forwarding);
4217
4218void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
4219{
4220 if (head_stolen) {
4221 skb_release_head_state(skb);
4222 kmem_cache_free(skbuff_head_cache, skb);
4223 } else {
4224 __kfree_skb(skb);
4225 }
4226}
4227EXPORT_SYMBOL(kfree_skb_partial);
4228
4229
4230
4231
4232
4233
4234
4235
4236bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
4237 bool *fragstolen, int *delta_truesize)
4238{
4239 int i, delta, len = from->len;
4240
4241 *fragstolen = false;
4242
4243 if (skb_cloned(to))
4244 return false;
4245
4246 if (len <= skb_tailroom(to)) {
4247 if (len)
4248 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
4249 *delta_truesize = 0;
4250 return true;
4251 }
4252
4253 if (skb_has_frag_list(to) || skb_has_frag_list(from))
4254 return false;
4255
4256 if (skb_headlen(from) != 0) {
4257 struct page *page;
4258 unsigned int offset;
4259
4260 if (skb_shinfo(to)->nr_frags +
4261 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
4262 return false;
4263
4264 if (skb_head_is_locked(from))
4265 return false;
4266
4267 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
4268
4269 page = virt_to_head_page(from->head);
4270 offset = from->data - (unsigned char *)page_address(page);
4271
4272 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
4273 page, offset, skb_headlen(from));
4274 *fragstolen = true;
4275 } else {
4276 if (skb_shinfo(to)->nr_frags +
4277 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
4278 return false;
4279
4280 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
4281 }
4282
4283 WARN_ON_ONCE(delta < len);
4284
4285 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
4286 skb_shinfo(from)->frags,
4287 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
4288 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
4289
4290 if (!skb_cloned(from))
4291 skb_shinfo(from)->nr_frags = 0;
4292
4293
4294
4295
4296 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
4297 skb_frag_ref(from, i);
4298
4299 to->truesize += delta;
4300 to->len += len;
4301 to->data_len += len;
4302
4303 *delta_truesize = delta;
4304 return true;
4305}
4306EXPORT_SYMBOL(skb_try_coalesce);
4307
4308
4309
4310
4311
4312
4313
4314
4315
4316
4317
4318
4319
4320
4321void skb_scrub_packet(struct sk_buff *skb, bool xnet)
4322{
4323 skb->tstamp.tv64 = 0;
4324 skb->pkt_type = PACKET_HOST;
4325 skb->skb_iif = 0;
4326 skb->ignore_df = 0;
4327 skb_dst_drop(skb);
4328 secpath_reset(skb);
4329 nf_reset(skb);
4330 nf_reset_trace(skb);
4331
4332 if (!xnet)
4333 return;
4334
4335 skb_orphan(skb);
4336 skb->mark = 0;
4337}
4338EXPORT_SYMBOL_GPL(skb_scrub_packet);
4339
4340
4341
4342
4343
4344
4345
4346
4347
4348
4349
4350unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
4351{
4352 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4353 unsigned int thlen = 0;
4354
4355 if (skb->encapsulation) {
4356 thlen = skb_inner_transport_header(skb) -
4357 skb_transport_header(skb);
4358
4359 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
4360 thlen += inner_tcp_hdrlen(skb);
4361 } else if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6))) {
4362 thlen = tcp_hdrlen(skb);
4363 }
4364
4365
4366
4367
4368 return thlen + shinfo->gso_size;
4369}
4370EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
4371
4372static struct sk_buff *skb_reorder_vlan_header(struct sk_buff *skb)
4373{
4374 if (skb_cow(skb, skb_headroom(skb)) < 0) {
4375 kfree_skb(skb);
4376 return NULL;
4377 }
4378
4379 memmove(skb->data - ETH_HLEN, skb->data - skb->mac_len - VLAN_HLEN,
4380 2 * ETH_ALEN);
4381 skb->mac_header += VLAN_HLEN;
4382 return skb;
4383}
4384
4385struct sk_buff *skb_vlan_untag(struct sk_buff *skb)
4386{
4387 struct vlan_hdr *vhdr;
4388 u16 vlan_tci;
4389
4390 if (unlikely(skb_vlan_tag_present(skb))) {
4391
4392 return skb;
4393 }
4394
4395 skb = skb_share_check(skb, GFP_ATOMIC);
4396 if (unlikely(!skb))
4397 goto err_free;
4398
4399 if (unlikely(!pskb_may_pull(skb, VLAN_HLEN)))
4400 goto err_free;
4401
4402 vhdr = (struct vlan_hdr *)skb->data;
4403 vlan_tci = ntohs(vhdr->h_vlan_TCI);
4404 __vlan_hwaccel_put_tag(skb, skb->protocol, vlan_tci);
4405
4406 skb_pull_rcsum(skb, VLAN_HLEN);
4407 vlan_set_encap_proto(skb, vhdr);
4408
4409 skb = skb_reorder_vlan_header(skb);
4410 if (unlikely(!skb))
4411 goto err_free;
4412
4413 skb_reset_network_header(skb);
4414 skb_reset_transport_header(skb);
4415 skb_reset_mac_len(skb);
4416
4417 return skb;
4418
4419err_free:
4420 kfree_skb(skb);
4421 return NULL;
4422}
4423EXPORT_SYMBOL(skb_vlan_untag);
4424
4425int skb_ensure_writable(struct sk_buff *skb, int write_len)
4426{
4427 if (!pskb_may_pull(skb, write_len))
4428 return -ENOMEM;
4429
4430 if (!skb_cloned(skb) || skb_clone_writable(skb, write_len))
4431 return 0;
4432
4433 return pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
4434}
4435EXPORT_SYMBOL(skb_ensure_writable);
4436
4437
4438static int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci)
4439{
4440 struct vlan_hdr *vhdr;
4441 unsigned int offset = skb->data - skb_mac_header(skb);
4442 int err;
4443
4444 __skb_push(skb, offset);
4445 err = skb_ensure_writable(skb, VLAN_ETH_HLEN);
4446 if (unlikely(err))
4447 goto pull;
4448
4449 skb_postpull_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4450
4451 vhdr = (struct vlan_hdr *)(skb->data + ETH_HLEN);
4452 *vlan_tci = ntohs(vhdr->h_vlan_TCI);
4453
4454 memmove(skb->data + VLAN_HLEN, skb->data, 2 * ETH_ALEN);
4455 __skb_pull(skb, VLAN_HLEN);
4456
4457 vlan_set_encap_proto(skb, vhdr);
4458 skb->mac_header += VLAN_HLEN;
4459
4460 if (skb_network_offset(skb) < ETH_HLEN)
4461 skb_set_network_header(skb, ETH_HLEN);
4462
4463 skb_reset_mac_len(skb);
4464pull:
4465 __skb_pull(skb, offset);
4466
4467 return err;
4468}
4469
4470int skb_vlan_pop(struct sk_buff *skb)
4471{
4472 u16 vlan_tci;
4473 __be16 vlan_proto;
4474 int err;
4475
4476 if (likely(skb_vlan_tag_present(skb))) {
4477 skb->vlan_tci = 0;
4478 } else {
4479 if (unlikely((skb->protocol != htons(ETH_P_8021Q) &&
4480 skb->protocol != htons(ETH_P_8021AD)) ||
4481 skb->len < VLAN_ETH_HLEN))
4482 return 0;
4483
4484 err = __skb_vlan_pop(skb, &vlan_tci);
4485 if (err)
4486 return err;
4487 }
4488
4489 if (likely((skb->protocol != htons(ETH_P_8021Q) &&
4490 skb->protocol != htons(ETH_P_8021AD)) ||
4491 skb->len < VLAN_ETH_HLEN))
4492 return 0;
4493
4494 vlan_proto = skb->protocol;
4495 err = __skb_vlan_pop(skb, &vlan_tci);
4496 if (unlikely(err))
4497 return err;
4498
4499 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4500 return 0;
4501}
4502EXPORT_SYMBOL(skb_vlan_pop);
4503
4504int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci)
4505{
4506 if (skb_vlan_tag_present(skb)) {
4507 unsigned int offset = skb->data - skb_mac_header(skb);
4508 int err;
4509
4510
4511
4512
4513
4514 __skb_push(skb, offset);
4515 err = __vlan_insert_tag(skb, skb->vlan_proto,
4516 skb_vlan_tag_get(skb));
4517 if (err) {
4518 __skb_pull(skb, offset);
4519 return err;
4520 }
4521
4522 skb->protocol = skb->vlan_proto;
4523 skb->mac_len += VLAN_HLEN;
4524
4525 skb_postpush_rcsum(skb, skb->data + (2 * ETH_ALEN), VLAN_HLEN);
4526 __skb_pull(skb, offset);
4527 }
4528 __vlan_hwaccel_put_tag(skb, vlan_proto, vlan_tci);
4529 return 0;
4530}
4531EXPORT_SYMBOL(skb_vlan_push);
4532
4533
4534
4535
4536
4537
4538
4539
4540
4541
4542
4543
4544struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
4545 unsigned long data_len,
4546 int max_page_order,
4547 int *errcode,
4548 gfp_t gfp_mask)
4549{
4550 int npages = (data_len + (PAGE_SIZE - 1)) >> PAGE_SHIFT;
4551 unsigned long chunk;
4552 struct sk_buff *skb;
4553 struct page *page;
4554 gfp_t gfp_head;
4555 int i;
4556
4557 *errcode = -EMSGSIZE;
4558
4559
4560
4561 if (npages > MAX_SKB_FRAGS)
4562 return NULL;
4563
4564 gfp_head = gfp_mask;
4565 if (gfp_head & __GFP_DIRECT_RECLAIM)
4566 gfp_head |= __GFP_REPEAT;
4567
4568 *errcode = -ENOBUFS;
4569 skb = alloc_skb(header_len, gfp_head);
4570 if (!skb)
4571 return NULL;
4572
4573 skb->truesize += npages << PAGE_SHIFT;
4574
4575 for (i = 0; npages > 0; i++) {
4576 int order = max_page_order;
4577
4578 while (order) {
4579 if (npages >= 1 << order) {
4580 page = alloc_pages((gfp_mask & ~__GFP_DIRECT_RECLAIM) |
4581 __GFP_COMP |
4582 __GFP_NOWARN |
4583 __GFP_NORETRY,
4584 order);
4585 if (page)
4586 goto fill_page;
4587
4588 order = 1;
4589 max_page_order = 0;
4590 }
4591 order--;
4592 }
4593 page = alloc_page(gfp_mask);
4594 if (!page)
4595 goto failure;
4596fill_page:
4597 chunk = min_t(unsigned long, data_len,
4598 PAGE_SIZE << order);
4599 skb_fill_page_desc(skb, i, page, 0, chunk);
4600 data_len -= chunk;
4601 npages -= 1 << order;
4602 }
4603 return skb;
4604
4605failure:
4606 kfree_skb(skb);
4607 return NULL;
4608}
4609EXPORT_SYMBOL(alloc_skb_with_frags);
4610
4611
4612static int pskb_carve_inside_header(struct sk_buff *skb, const u32 off,
4613 const int headlen, gfp_t gfp_mask)
4614{
4615 int i;
4616 int size = skb_end_offset(skb);
4617 int new_hlen = headlen - off;
4618 u8 *data;
4619
4620 size = SKB_DATA_ALIGN(size);
4621
4622 if (skb_pfmemalloc(skb))
4623 gfp_mask |= __GFP_MEMALLOC;
4624 data = kmalloc_reserve(size +
4625 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4626 gfp_mask, NUMA_NO_NODE, NULL);
4627 if (!data)
4628 return -ENOMEM;
4629
4630 size = SKB_WITH_OVERHEAD(ksize(data));
4631
4632
4633 skb_copy_from_linear_data_offset(skb, off, data, new_hlen);
4634 skb->len -= off;
4635
4636 memcpy((struct skb_shared_info *)(data + size),
4637 skb_shinfo(skb),
4638 offsetof(struct skb_shared_info,
4639 frags[skb_shinfo(skb)->nr_frags]));
4640 if (skb_cloned(skb)) {
4641
4642 if (skb_orphan_frags(skb, gfp_mask)) {
4643 kfree(data);
4644 return -ENOMEM;
4645 }
4646 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
4647 skb_frag_ref(skb, i);
4648 if (skb_has_frag_list(skb))
4649 skb_clone_fraglist(skb);
4650 skb_release_data(skb);
4651 } else {
4652
4653
4654
4655 skb_free_head(skb);
4656 }
4657
4658 skb->head = data;
4659 skb->data = data;
4660 skb->head_frag = 0;
4661#ifdef NET_SKBUFF_DATA_USES_OFFSET
4662 skb->end = size;
4663#else
4664 skb->end = skb->head + size;
4665#endif
4666 skb_set_tail_pointer(skb, skb_headlen(skb));
4667 skb_headers_offset_update(skb, 0);
4668 skb->cloned = 0;
4669 skb->hdr_len = 0;
4670 skb->nohdr = 0;
4671 atomic_set(&skb_shinfo(skb)->dataref, 1);
4672
4673 return 0;
4674}
4675
4676static int pskb_carve(struct sk_buff *skb, const u32 off, gfp_t gfp);
4677
4678
4679
4680
4681static int pskb_carve_frag_list(struct sk_buff *skb,
4682 struct skb_shared_info *shinfo, int eat,
4683 gfp_t gfp_mask)
4684{
4685 struct sk_buff *list = shinfo->frag_list;
4686 struct sk_buff *clone = NULL;
4687 struct sk_buff *insp = NULL;
4688
4689 do {
4690 if (!list) {
4691 pr_err("Not enough bytes to eat. Want %d\n", eat);
4692 return -EFAULT;
4693 }
4694 if (list->len <= eat) {
4695
4696 eat -= list->len;
4697 list = list->next;
4698 insp = list;
4699 } else {
4700
4701 if (skb_shared(list)) {
4702 clone = skb_clone(list, gfp_mask);
4703 if (!clone)
4704 return -ENOMEM;
4705 insp = list->next;
4706 list = clone;
4707 } else {
4708
4709 insp = list;
4710 }
4711 if (pskb_carve(list, eat, gfp_mask) < 0) {
4712 kfree_skb(clone);
4713 return -ENOMEM;
4714 }
4715 break;
4716 }
4717 } while (eat);
4718
4719
4720 while ((list = shinfo->frag_list) != insp) {
4721 shinfo->frag_list = list->next;
4722 kfree_skb(list);
4723 }
4724
4725 if (clone) {
4726 clone->next = list;
4727 shinfo->frag_list = clone;
4728 }
4729 return 0;
4730}
4731
4732
4733
4734
4735static int pskb_carve_inside_nonlinear(struct sk_buff *skb, const u32 off,
4736 int pos, gfp_t gfp_mask)
4737{
4738 int i, k = 0;
4739 int size = skb_end_offset(skb);
4740 u8 *data;
4741 const int nfrags = skb_shinfo(skb)->nr_frags;
4742 struct skb_shared_info *shinfo;
4743
4744 size = SKB_DATA_ALIGN(size);
4745
4746 if (skb_pfmemalloc(skb))
4747 gfp_mask |= __GFP_MEMALLOC;
4748 data = kmalloc_reserve(size +
4749 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
4750 gfp_mask, NUMA_NO_NODE, NULL);
4751 if (!data)
4752 return -ENOMEM;
4753
4754 size = SKB_WITH_OVERHEAD(ksize(data));
4755
4756 memcpy((struct skb_shared_info *)(data + size),
4757 skb_shinfo(skb), offsetof(struct skb_shared_info,
4758 frags[skb_shinfo(skb)->nr_frags]));
4759 if (skb_orphan_frags(skb, gfp_mask)) {
4760 kfree(data);
4761 return -ENOMEM;
4762 }
4763 shinfo = (struct skb_shared_info *)(data + size);
4764 for (i = 0; i < nfrags; i++) {
4765 int fsize = skb_frag_size(&skb_shinfo(skb)->frags[i]);
4766
4767 if (pos + fsize > off) {
4768 shinfo->frags[k] = skb_shinfo(skb)->frags[i];
4769
4770 if (pos < off) {
4771
4772
4773
4774
4775
4776
4777
4778
4779 shinfo->frags[0].page_offset += off - pos;
4780 skb_frag_size_sub(&shinfo->frags[0], off - pos);
4781 }
4782 skb_frag_ref(skb, i);
4783 k++;
4784 }
4785 pos += fsize;
4786 }
4787 shinfo->nr_frags = k;
4788 if (skb_has_frag_list(skb))
4789 skb_clone_fraglist(skb);
4790
4791 if (k == 0) {
4792
4793 pskb_carve_frag_list(skb, shinfo, off - pos, gfp_mask);
4794 }
4795 skb_release_data(skb);
4796
4797 skb->head = data;
4798 skb->head_frag = 0;
4799 skb->data = data;
4800#ifdef NET_SKBUFF_DATA_USES_OFFSET
4801 skb->end = size;
4802#else
4803 skb->end = skb->head + size;
4804#endif
4805 skb_reset_tail_pointer(skb);
4806 skb_headers_offset_update(skb, 0);
4807 skb->cloned = 0;
4808 skb->hdr_len = 0;
4809 skb->nohdr = 0;
4810 skb->len -= off;
4811 skb->data_len = skb->len;
4812 atomic_set(&skb_shinfo(skb)->dataref, 1);
4813 return 0;
4814}
4815
4816
4817static int pskb_carve(struct sk_buff *skb, const u32 len, gfp_t gfp)
4818{
4819 int headlen = skb_headlen(skb);
4820
4821 if (len < headlen)
4822 return pskb_carve_inside_header(skb, len, headlen, gfp);
4823 else
4824 return pskb_carve_inside_nonlinear(skb, len, headlen, gfp);
4825}
4826
4827
4828
4829
4830struct sk_buff *pskb_extract(struct sk_buff *skb, int off,
4831 int to_copy, gfp_t gfp)
4832{
4833 struct sk_buff *clone = skb_clone(skb, gfp);
4834
4835 if (!clone)
4836 return NULL;
4837
4838 if (pskb_carve(clone, off, gfp) < 0 ||
4839 pskb_trim(clone, to_copy)) {
4840 kfree_skb(clone);
4841 return NULL;
4842 }
4843 return clone;
4844}
4845EXPORT_SYMBOL(pskb_extract);
4846