Loading...
1/*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
6 *
7 * Fixes:
8 * Alan Cox : Fixed the worst of the load
9 * balancer bugs.
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
22 *
23 * NOTE:
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
28 *
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
33 */
34
35/*
36 * The functions in this file will not compile correctly with gcc 2.4.x
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
66#include <net/protocol.h>
67#include <net/dst.h>
68#include <net/sock.h>
69#include <net/checksum.h>
70#include <net/ip6_checksum.h>
71#include <net/xfrm.h>
72
73#include <asm/uaccess.h>
74#include <trace/events/skb.h>
75#include <linux/highmem.h>
76
77struct kmem_cache *skbuff_head_cache __read_mostly;
78static struct kmem_cache *skbuff_fclone_cache __read_mostly;
79
80/**
81 * skb_panic - private function for out-of-line support
82 * @skb: buffer
83 * @sz: size
84 * @addr: address
85 * @msg: skb_over_panic or skb_under_panic
86 *
87 * Out-of-line support for skb_put() and skb_push().
88 * Called via the wrapper skb_over_panic() or skb_under_panic().
89 * Keep out of line to prevent kernel bloat.
90 * __builtin_return_address is not used because it is not always reliable.
91 */
92static void skb_panic(struct sk_buff *skb, unsigned int sz, void *addr,
93 const char msg[])
94{
95 pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
96 msg, addr, skb->len, sz, skb->head, skb->data,
97 (unsigned long)skb->tail, (unsigned long)skb->end,
98 skb->dev ? skb->dev->name : "<NULL>");
99 BUG();
100}
101
102static void skb_over_panic(struct sk_buff *skb, unsigned int sz, void *addr)
103{
104 skb_panic(skb, sz, addr, __func__);
105}
106
107static void skb_under_panic(struct sk_buff *skb, unsigned int sz, void *addr)
108{
109 skb_panic(skb, sz, addr, __func__);
110}
111
112/*
113 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
114 * the caller if emergency pfmemalloc reserves are being used. If it is and
115 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
116 * may be used. Otherwise, the packet data may be discarded until enough
117 * memory is free
118 */
119#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
120 __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
121
122static void *__kmalloc_reserve(size_t size, gfp_t flags, int node,
123 unsigned long ip, bool *pfmemalloc)
124{
125 void *obj;
126 bool ret_pfmemalloc = false;
127
128 /*
129 * Try a regular allocation, when that fails and we're not entitled
130 * to the reserves, fail.
131 */
132 obj = kmalloc_node_track_caller(size,
133 flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
134 node);
135 if (obj || !(gfp_pfmemalloc_allowed(flags)))
136 goto out;
137
138 /* Try again but now we are using pfmemalloc reserves */
139 ret_pfmemalloc = true;
140 obj = kmalloc_node_track_caller(size, flags, node);
141
142out:
143 if (pfmemalloc)
144 *pfmemalloc = ret_pfmemalloc;
145
146 return obj;
147}
148
149/* Allocate a new skbuff. We do this ourselves so we can fill in a few
150 * 'private' fields and also do memory statistics to find all the
151 * [BEEP] leaks.
152 *
153 */
154
155struct sk_buff *__alloc_skb_head(gfp_t gfp_mask, int node)
156{
157 struct sk_buff *skb;
158
159 /* Get the HEAD */
160 skb = kmem_cache_alloc_node(skbuff_head_cache,
161 gfp_mask & ~__GFP_DMA, node);
162 if (!skb)
163 goto out;
164
165 /*
166 * Only clear those fields we need to clear, not those that we will
167 * actually initialise below. Hence, don't put any more fields after
168 * the tail pointer in struct sk_buff!
169 */
170 memset(skb, 0, offsetof(struct sk_buff, tail));
171 skb->head = NULL;
172 skb->truesize = sizeof(struct sk_buff);
173 atomic_set(&skb->users, 1);
174
175 skb->mac_header = (typeof(skb->mac_header))~0U;
176out:
177 return skb;
178}
179
180/**
181 * __alloc_skb - allocate a network buffer
182 * @size: size to allocate
183 * @gfp_mask: allocation mask
184 * @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
185 * instead of head cache and allocate a cloned (child) skb.
186 * If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
187 * allocations in case the data is required for writeback
188 * @node: numa node to allocate memory on
189 *
190 * Allocate a new &sk_buff. The returned buffer has no headroom and a
191 * tail room of at least size bytes. The object has a reference count
192 * of one. The return is the buffer. On a failure the return is %NULL.
193 *
194 * Buffers may only be allocated from interrupts using a @gfp_mask of
195 * %GFP_ATOMIC.
196 */
197struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
198 int flags, int node)
199{
200 struct kmem_cache *cache;
201 struct skb_shared_info *shinfo;
202 struct sk_buff *skb;
203 u8 *data;
204 bool pfmemalloc;
205
206 cache = (flags & SKB_ALLOC_FCLONE)
207 ? skbuff_fclone_cache : skbuff_head_cache;
208
209 if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
210 gfp_mask |= __GFP_MEMALLOC;
211
212 /* Get the HEAD */
213 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
214 if (!skb)
215 goto out;
216 prefetchw(skb);
217
218 /* We do our best to align skb_shared_info on a separate cache
219 * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
220 * aligned memory blocks, unless SLUB/SLAB debug is enabled.
221 * Both skb->head and skb_shared_info are cache line aligned.
222 */
223 size = SKB_DATA_ALIGN(size);
224 size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
225 data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
226 if (!data)
227 goto nodata;
228 /* kmalloc(size) might give us more room than requested.
229 * Put skb_shared_info exactly at the end of allocated zone,
230 * to allow max possible filling before reallocation.
231 */
232 size = SKB_WITH_OVERHEAD(ksize(data));
233 prefetchw(data + size);
234
235 /*
236 * Only clear those fields we need to clear, not those that we will
237 * actually initialise below. Hence, don't put any more fields after
238 * the tail pointer in struct sk_buff!
239 */
240 memset(skb, 0, offsetof(struct sk_buff, tail));
241 /* Account for allocated memory : skb + skb->head */
242 skb->truesize = SKB_TRUESIZE(size);
243 skb->pfmemalloc = pfmemalloc;
244 atomic_set(&skb->users, 1);
245 skb->head = data;
246 skb->data = data;
247 skb_reset_tail_pointer(skb);
248 skb->end = skb->tail + size;
249 skb->mac_header = (typeof(skb->mac_header))~0U;
250 skb->transport_header = (typeof(skb->transport_header))~0U;
251
252 /* make sure we initialize shinfo sequentially */
253 shinfo = skb_shinfo(skb);
254 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
255 atomic_set(&shinfo->dataref, 1);
256 kmemcheck_annotate_variable(shinfo->destructor_arg);
257
258 if (flags & SKB_ALLOC_FCLONE) {
259 struct sk_buff *child = skb + 1;
260 atomic_t *fclone_ref = (atomic_t *) (child + 1);
261
262 kmemcheck_annotate_bitfield(child, flags1);
263 kmemcheck_annotate_bitfield(child, flags2);
264 skb->fclone = SKB_FCLONE_ORIG;
265 atomic_set(fclone_ref, 1);
266
267 child->fclone = SKB_FCLONE_UNAVAILABLE;
268 child->pfmemalloc = pfmemalloc;
269 }
270out:
271 return skb;
272nodata:
273 kmem_cache_free(cache, skb);
274 skb = NULL;
275 goto out;
276}
277EXPORT_SYMBOL(__alloc_skb);
278
279/**
280 * build_skb - build a network buffer
281 * @data: data buffer provided by caller
282 * @frag_size: size of fragment, or 0 if head was kmalloced
283 *
284 * Allocate a new &sk_buff. Caller provides space holding head and
285 * skb_shared_info. @data must have been allocated by kmalloc() only if
286 * @frag_size is 0, otherwise data should come from the page allocator.
287 * The return is the new skb buffer.
288 * On a failure the return is %NULL, and @data is not freed.
289 * Notes :
290 * Before IO, driver allocates only data buffer where NIC put incoming frame
291 * Driver should add room at head (NET_SKB_PAD) and
292 * MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
293 * After IO, driver calls build_skb(), to allocate sk_buff and populate it
294 * before giving packet to stack.
295 * RX rings only contains data buffers, not full skbs.
296 */
297struct sk_buff *build_skb(void *data, unsigned int frag_size)
298{
299 struct skb_shared_info *shinfo;
300 struct sk_buff *skb;
301 unsigned int size = frag_size ? : ksize(data);
302
303 skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
304 if (!skb)
305 return NULL;
306
307 size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
308
309 memset(skb, 0, offsetof(struct sk_buff, tail));
310 skb->truesize = SKB_TRUESIZE(size);
311 skb->head_frag = frag_size != 0;
312 atomic_set(&skb->users, 1);
313 skb->head = data;
314 skb->data = data;
315 skb_reset_tail_pointer(skb);
316 skb->end = skb->tail + size;
317 skb->mac_header = (typeof(skb->mac_header))~0U;
318 skb->transport_header = (typeof(skb->transport_header))~0U;
319
320 /* make sure we initialize shinfo sequentially */
321 shinfo = skb_shinfo(skb);
322 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
323 atomic_set(&shinfo->dataref, 1);
324 kmemcheck_annotate_variable(shinfo->destructor_arg);
325
326 return skb;
327}
328EXPORT_SYMBOL(build_skb);
329
330struct netdev_alloc_cache {
331 struct page_frag frag;
332 /* we maintain a pagecount bias, so that we dont dirty cache line
333 * containing page->_count every time we allocate a fragment.
334 */
335 unsigned int pagecnt_bias;
336};
337static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
338
339static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
340{
341 struct netdev_alloc_cache *nc;
342 void *data = NULL;
343 int order;
344 unsigned long flags;
345
346 local_irq_save(flags);
347 nc = &__get_cpu_var(netdev_alloc_cache);
348 if (unlikely(!nc->frag.page)) {
349refill:
350 for (order = NETDEV_FRAG_PAGE_MAX_ORDER; ;) {
351 gfp_t gfp = gfp_mask;
352
353 if (order)
354 gfp |= __GFP_COMP | __GFP_NOWARN;
355 nc->frag.page = alloc_pages(gfp, order);
356 if (likely(nc->frag.page))
357 break;
358 if (--order < 0)
359 goto end;
360 }
361 nc->frag.size = PAGE_SIZE << order;
362recycle:
363 atomic_set(&nc->frag.page->_count, NETDEV_PAGECNT_MAX_BIAS);
364 nc->pagecnt_bias = NETDEV_PAGECNT_MAX_BIAS;
365 nc->frag.offset = 0;
366 }
367
368 if (nc->frag.offset + fragsz > nc->frag.size) {
369 /* avoid unnecessary locked operations if possible */
370 if ((atomic_read(&nc->frag.page->_count) == nc->pagecnt_bias) ||
371 atomic_sub_and_test(nc->pagecnt_bias, &nc->frag.page->_count))
372 goto recycle;
373 goto refill;
374 }
375
376 data = page_address(nc->frag.page) + nc->frag.offset;
377 nc->frag.offset += fragsz;
378 nc->pagecnt_bias--;
379end:
380 local_irq_restore(flags);
381 return data;
382}
383
384/**
385 * netdev_alloc_frag - allocate a page fragment
386 * @fragsz: fragment size
387 *
388 * Allocates a frag from a page for receive buffer.
389 * Uses GFP_ATOMIC allocations.
390 */
391void *netdev_alloc_frag(unsigned int fragsz)
392{
393 return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
394}
395EXPORT_SYMBOL(netdev_alloc_frag);
396
397/**
398 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
399 * @dev: network device to receive on
400 * @length: length to allocate
401 * @gfp_mask: get_free_pages mask, passed to alloc_skb
402 *
403 * Allocate a new &sk_buff and assign it a usage count of one. The
404 * buffer has unspecified headroom built in. Users should allocate
405 * the headroom they think they need without accounting for the
406 * built in space. The built in space is used for optimisations.
407 *
408 * %NULL is returned if there is no free memory.
409 */
410struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
411 unsigned int length, gfp_t gfp_mask)
412{
413 struct sk_buff *skb = NULL;
414 unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
415 SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
416
417 if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
418 void *data;
419
420 if (sk_memalloc_socks())
421 gfp_mask |= __GFP_MEMALLOC;
422
423 data = __netdev_alloc_frag(fragsz, gfp_mask);
424
425 if (likely(data)) {
426 skb = build_skb(data, fragsz);
427 if (unlikely(!skb))
428 put_page(virt_to_head_page(data));
429 }
430 } else {
431 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
432 SKB_ALLOC_RX, NUMA_NO_NODE);
433 }
434 if (likely(skb)) {
435 skb_reserve(skb, NET_SKB_PAD);
436 skb->dev = dev;
437 }
438 return skb;
439}
440EXPORT_SYMBOL(__netdev_alloc_skb);
441
442void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
443 int size, unsigned int truesize)
444{
445 skb_fill_page_desc(skb, i, page, off, size);
446 skb->len += size;
447 skb->data_len += size;
448 skb->truesize += truesize;
449}
450EXPORT_SYMBOL(skb_add_rx_frag);
451
452void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
453 unsigned int truesize)
454{
455 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
456
457 skb_frag_size_add(frag, size);
458 skb->len += size;
459 skb->data_len += size;
460 skb->truesize += truesize;
461}
462EXPORT_SYMBOL(skb_coalesce_rx_frag);
463
464static void skb_drop_list(struct sk_buff **listp)
465{
466 kfree_skb_list(*listp);
467 *listp = NULL;
468}
469
470static inline void skb_drop_fraglist(struct sk_buff *skb)
471{
472 skb_drop_list(&skb_shinfo(skb)->frag_list);
473}
474
475static void skb_clone_fraglist(struct sk_buff *skb)
476{
477 struct sk_buff *list;
478
479 skb_walk_frags(skb, list)
480 skb_get(list);
481}
482
483static void skb_free_head(struct sk_buff *skb)
484{
485 if (skb->head_frag)
486 put_page(virt_to_head_page(skb->head));
487 else
488 kfree(skb->head);
489}
490
491static void skb_release_data(struct sk_buff *skb)
492{
493 if (!skb->cloned ||
494 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
495 &skb_shinfo(skb)->dataref)) {
496 if (skb_shinfo(skb)->nr_frags) {
497 int i;
498 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
499 skb_frag_unref(skb, i);
500 }
501
502 /*
503 * If skb buf is from userspace, we need to notify the caller
504 * the lower device DMA has done;
505 */
506 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
507 struct ubuf_info *uarg;
508
509 uarg = skb_shinfo(skb)->destructor_arg;
510 if (uarg->callback)
511 uarg->callback(uarg, true);
512 }
513
514 if (skb_has_frag_list(skb))
515 skb_drop_fraglist(skb);
516
517 skb_free_head(skb);
518 }
519}
520
521/*
522 * Free an skbuff by memory without cleaning the state.
523 */
524static void kfree_skbmem(struct sk_buff *skb)
525{
526 struct sk_buff *other;
527 atomic_t *fclone_ref;
528
529 switch (skb->fclone) {
530 case SKB_FCLONE_UNAVAILABLE:
531 kmem_cache_free(skbuff_head_cache, skb);
532 break;
533
534 case SKB_FCLONE_ORIG:
535 fclone_ref = (atomic_t *) (skb + 2);
536 if (atomic_dec_and_test(fclone_ref))
537 kmem_cache_free(skbuff_fclone_cache, skb);
538 break;
539
540 case SKB_FCLONE_CLONE:
541 fclone_ref = (atomic_t *) (skb + 1);
542 other = skb - 1;
543
544 /* The clone portion is available for
545 * fast-cloning again.
546 */
547 skb->fclone = SKB_FCLONE_UNAVAILABLE;
548
549 if (atomic_dec_and_test(fclone_ref))
550 kmem_cache_free(skbuff_fclone_cache, other);
551 break;
552 }
553}
554
555static void skb_release_head_state(struct sk_buff *skb)
556{
557 skb_dst_drop(skb);
558#ifdef CONFIG_XFRM
559 secpath_put(skb->sp);
560#endif
561 if (skb->destructor) {
562 WARN_ON(in_irq());
563 skb->destructor(skb);
564 }
565#if IS_ENABLED(CONFIG_NF_CONNTRACK)
566 nf_conntrack_put(skb->nfct);
567#endif
568#ifdef CONFIG_BRIDGE_NETFILTER
569 nf_bridge_put(skb->nf_bridge);
570#endif
571/* XXX: IS this still necessary? - JHS */
572#ifdef CONFIG_NET_SCHED
573 skb->tc_index = 0;
574#ifdef CONFIG_NET_CLS_ACT
575 skb->tc_verd = 0;
576#endif
577#endif
578}
579
580/* Free everything but the sk_buff shell. */
581static void skb_release_all(struct sk_buff *skb)
582{
583 skb_release_head_state(skb);
584 if (likely(skb->head))
585 skb_release_data(skb);
586}
587
588/**
589 * __kfree_skb - private function
590 * @skb: buffer
591 *
592 * Free an sk_buff. Release anything attached to the buffer.
593 * Clean the state. This is an internal helper function. Users should
594 * always call kfree_skb
595 */
596
597void __kfree_skb(struct sk_buff *skb)
598{
599 skb_release_all(skb);
600 kfree_skbmem(skb);
601}
602EXPORT_SYMBOL(__kfree_skb);
603
604/**
605 * kfree_skb - free an sk_buff
606 * @skb: buffer to free
607 *
608 * Drop a reference to the buffer and free it if the usage count has
609 * hit zero.
610 */
611void kfree_skb(struct sk_buff *skb)
612{
613 if (unlikely(!skb))
614 return;
615 if (likely(atomic_read(&skb->users) == 1))
616 smp_rmb();
617 else if (likely(!atomic_dec_and_test(&skb->users)))
618 return;
619 trace_kfree_skb(skb, __builtin_return_address(0));
620 __kfree_skb(skb);
621}
622EXPORT_SYMBOL(kfree_skb);
623
624void kfree_skb_list(struct sk_buff *segs)
625{
626 while (segs) {
627 struct sk_buff *next = segs->next;
628
629 kfree_skb(segs);
630 segs = next;
631 }
632}
633EXPORT_SYMBOL(kfree_skb_list);
634
635/**
636 * skb_tx_error - report an sk_buff xmit error
637 * @skb: buffer that triggered an error
638 *
639 * Report xmit error if a device callback is tracking this skb.
640 * skb must be freed afterwards.
641 */
642void skb_tx_error(struct sk_buff *skb)
643{
644 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
645 struct ubuf_info *uarg;
646
647 uarg = skb_shinfo(skb)->destructor_arg;
648 if (uarg->callback)
649 uarg->callback(uarg, false);
650 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
651 }
652}
653EXPORT_SYMBOL(skb_tx_error);
654
655/**
656 * consume_skb - free an skbuff
657 * @skb: buffer to free
658 *
659 * Drop a ref to the buffer and free it if the usage count has hit zero
660 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
661 * is being dropped after a failure and notes that
662 */
663void consume_skb(struct sk_buff *skb)
664{
665 if (unlikely(!skb))
666 return;
667 if (likely(atomic_read(&skb->users) == 1))
668 smp_rmb();
669 else if (likely(!atomic_dec_and_test(&skb->users)))
670 return;
671 trace_consume_skb(skb);
672 __kfree_skb(skb);
673}
674EXPORT_SYMBOL(consume_skb);
675
676static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
677{
678 new->tstamp = old->tstamp;
679 new->dev = old->dev;
680 new->transport_header = old->transport_header;
681 new->network_header = old->network_header;
682 new->mac_header = old->mac_header;
683 new->inner_protocol = old->inner_protocol;
684 new->inner_transport_header = old->inner_transport_header;
685 new->inner_network_header = old->inner_network_header;
686 new->inner_mac_header = old->inner_mac_header;
687 skb_dst_copy(new, old);
688 skb_copy_hash(new, old);
689 new->ooo_okay = old->ooo_okay;
690 new->no_fcs = old->no_fcs;
691 new->encapsulation = old->encapsulation;
692#ifdef CONFIG_XFRM
693 new->sp = secpath_get(old->sp);
694#endif
695 memcpy(new->cb, old->cb, sizeof(old->cb));
696 new->csum = old->csum;
697 new->local_df = old->local_df;
698 new->pkt_type = old->pkt_type;
699 new->ip_summed = old->ip_summed;
700 skb_copy_queue_mapping(new, old);
701 new->priority = old->priority;
702#if IS_ENABLED(CONFIG_IP_VS)
703 new->ipvs_property = old->ipvs_property;
704#endif
705 new->pfmemalloc = old->pfmemalloc;
706 new->protocol = old->protocol;
707 new->mark = old->mark;
708 new->skb_iif = old->skb_iif;
709 __nf_copy(new, old);
710#ifdef CONFIG_NET_SCHED
711 new->tc_index = old->tc_index;
712#ifdef CONFIG_NET_CLS_ACT
713 new->tc_verd = old->tc_verd;
714#endif
715#endif
716 new->vlan_proto = old->vlan_proto;
717 new->vlan_tci = old->vlan_tci;
718
719 skb_copy_secmark(new, old);
720
721#ifdef CONFIG_NET_RX_BUSY_POLL
722 new->napi_id = old->napi_id;
723#endif
724}
725
726/*
727 * You should not add any new code to this function. Add it to
728 * __copy_skb_header above instead.
729 */
730static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
731{
732#define C(x) n->x = skb->x
733
734 n->next = n->prev = NULL;
735 n->sk = NULL;
736 __copy_skb_header(n, skb);
737
738 C(len);
739 C(data_len);
740 C(mac_len);
741 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
742 n->cloned = 1;
743 n->nohdr = 0;
744 n->destructor = NULL;
745 C(tail);
746 C(end);
747 C(head);
748 C(head_frag);
749 C(data);
750 C(truesize);
751 atomic_set(&n->users, 1);
752
753 atomic_inc(&(skb_shinfo(skb)->dataref));
754 skb->cloned = 1;
755
756 return n;
757#undef C
758}
759
760/**
761 * skb_morph - morph one skb into another
762 * @dst: the skb to receive the contents
763 * @src: the skb to supply the contents
764 *
765 * This is identical to skb_clone except that the target skb is
766 * supplied by the user.
767 *
768 * The target skb is returned upon exit.
769 */
770struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
771{
772 skb_release_all(dst);
773 return __skb_clone(dst, src);
774}
775EXPORT_SYMBOL_GPL(skb_morph);
776
777/**
778 * skb_copy_ubufs - copy userspace skb frags buffers to kernel
779 * @skb: the skb to modify
780 * @gfp_mask: allocation priority
781 *
782 * This must be called on SKBTX_DEV_ZEROCOPY skb.
783 * It will copy all frags into kernel and drop the reference
784 * to userspace pages.
785 *
786 * If this function is called from an interrupt gfp_mask() must be
787 * %GFP_ATOMIC.
788 *
789 * Returns 0 on success or a negative error code on failure
790 * to allocate kernel memory to copy to.
791 */
792int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
793{
794 int i;
795 int num_frags = skb_shinfo(skb)->nr_frags;
796 struct page *page, *head = NULL;
797 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
798
799 for (i = 0; i < num_frags; i++) {
800 u8 *vaddr;
801 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
802
803 page = alloc_page(gfp_mask);
804 if (!page) {
805 while (head) {
806 struct page *next = (struct page *)page_private(head);
807 put_page(head);
808 head = next;
809 }
810 return -ENOMEM;
811 }
812 vaddr = kmap_atomic(skb_frag_page(f));
813 memcpy(page_address(page),
814 vaddr + f->page_offset, skb_frag_size(f));
815 kunmap_atomic(vaddr);
816 set_page_private(page, (unsigned long)head);
817 head = page;
818 }
819
820 /* skb frags release userspace buffers */
821 for (i = 0; i < num_frags; i++)
822 skb_frag_unref(skb, i);
823
824 uarg->callback(uarg, false);
825
826 /* skb frags point to kernel buffers */
827 for (i = num_frags - 1; i >= 0; i--) {
828 __skb_fill_page_desc(skb, i, head, 0,
829 skb_shinfo(skb)->frags[i].size);
830 head = (struct page *)page_private(head);
831 }
832
833 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
834 return 0;
835}
836EXPORT_SYMBOL_GPL(skb_copy_ubufs);
837
838/**
839 * skb_clone - duplicate an sk_buff
840 * @skb: buffer to clone
841 * @gfp_mask: allocation priority
842 *
843 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
844 * copies share the same packet data but not structure. The new
845 * buffer has a reference count of 1. If the allocation fails the
846 * function returns %NULL otherwise the new buffer is returned.
847 *
848 * If this function is called from an interrupt gfp_mask() must be
849 * %GFP_ATOMIC.
850 */
851
852struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
853{
854 struct sk_buff *n;
855
856 if (skb_orphan_frags(skb, gfp_mask))
857 return NULL;
858
859 n = skb + 1;
860 if (skb->fclone == SKB_FCLONE_ORIG &&
861 n->fclone == SKB_FCLONE_UNAVAILABLE) {
862 atomic_t *fclone_ref = (atomic_t *) (n + 1);
863 n->fclone = SKB_FCLONE_CLONE;
864 atomic_inc(fclone_ref);
865 } else {
866 if (skb_pfmemalloc(skb))
867 gfp_mask |= __GFP_MEMALLOC;
868
869 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
870 if (!n)
871 return NULL;
872
873 kmemcheck_annotate_bitfield(n, flags1);
874 kmemcheck_annotate_bitfield(n, flags2);
875 n->fclone = SKB_FCLONE_UNAVAILABLE;
876 }
877
878 return __skb_clone(n, skb);
879}
880EXPORT_SYMBOL(skb_clone);
881
882static void skb_headers_offset_update(struct sk_buff *skb, int off)
883{
884 /* Only adjust this if it actually is csum_start rather than csum */
885 if (skb->ip_summed == CHECKSUM_PARTIAL)
886 skb->csum_start += off;
887 /* {transport,network,mac}_header and tail are relative to skb->head */
888 skb->transport_header += off;
889 skb->network_header += off;
890 if (skb_mac_header_was_set(skb))
891 skb->mac_header += off;
892 skb->inner_transport_header += off;
893 skb->inner_network_header += off;
894 skb->inner_mac_header += off;
895}
896
897static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
898{
899 __copy_skb_header(new, old);
900
901 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
902 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
903 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
904}
905
906static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
907{
908 if (skb_pfmemalloc(skb))
909 return SKB_ALLOC_RX;
910 return 0;
911}
912
913/**
914 * skb_copy - create private copy of an sk_buff
915 * @skb: buffer to copy
916 * @gfp_mask: allocation priority
917 *
918 * Make a copy of both an &sk_buff and its data. This is used when the
919 * caller wishes to modify the data and needs a private copy of the
920 * data to alter. Returns %NULL on failure or the pointer to the buffer
921 * on success. The returned buffer has a reference count of 1.
922 *
923 * As by-product this function converts non-linear &sk_buff to linear
924 * one, so that &sk_buff becomes completely private and caller is allowed
925 * to modify all the data of returned buffer. This means that this
926 * function is not recommended for use in circumstances when only
927 * header is going to be modified. Use pskb_copy() instead.
928 */
929
930struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
931{
932 int headerlen = skb_headroom(skb);
933 unsigned int size = skb_end_offset(skb) + skb->data_len;
934 struct sk_buff *n = __alloc_skb(size, gfp_mask,
935 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
936
937 if (!n)
938 return NULL;
939
940 /* Set the data pointer */
941 skb_reserve(n, headerlen);
942 /* Set the tail pointer and length */
943 skb_put(n, skb->len);
944
945 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
946 BUG();
947
948 copy_skb_header(n, skb);
949 return n;
950}
951EXPORT_SYMBOL(skb_copy);
952
953/**
954 * __pskb_copy - create copy of an sk_buff with private head.
955 * @skb: buffer to copy
956 * @headroom: headroom of new skb
957 * @gfp_mask: allocation priority
958 *
959 * Make a copy of both an &sk_buff and part of its data, located
960 * in header. Fragmented data remain shared. This is used when
961 * the caller wishes to modify only header of &sk_buff and needs
962 * private copy of the header to alter. Returns %NULL on failure
963 * or the pointer to the buffer on success.
964 * The returned buffer has a reference count of 1.
965 */
966
967struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
968{
969 unsigned int size = skb_headlen(skb) + headroom;
970 struct sk_buff *n = __alloc_skb(size, gfp_mask,
971 skb_alloc_rx_flag(skb), NUMA_NO_NODE);
972
973 if (!n)
974 goto out;
975
976 /* Set the data pointer */
977 skb_reserve(n, headroom);
978 /* Set the tail pointer and length */
979 skb_put(n, skb_headlen(skb));
980 /* Copy the bytes */
981 skb_copy_from_linear_data(skb, n->data, n->len);
982
983 n->truesize += skb->data_len;
984 n->data_len = skb->data_len;
985 n->len = skb->len;
986
987 if (skb_shinfo(skb)->nr_frags) {
988 int i;
989
990 if (skb_orphan_frags(skb, gfp_mask)) {
991 kfree_skb(n);
992 n = NULL;
993 goto out;
994 }
995 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
996 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
997 skb_frag_ref(skb, i);
998 }
999 skb_shinfo(n)->nr_frags = i;
1000 }
1001
1002 if (skb_has_frag_list(skb)) {
1003 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1004 skb_clone_fraglist(n);
1005 }
1006
1007 copy_skb_header(n, skb);
1008out:
1009 return n;
1010}
1011EXPORT_SYMBOL(__pskb_copy);
1012
1013/**
1014 * pskb_expand_head - reallocate header of &sk_buff
1015 * @skb: buffer to reallocate
1016 * @nhead: room to add at head
1017 * @ntail: room to add at tail
1018 * @gfp_mask: allocation priority
1019 *
1020 * Expands (or creates identical copy, if @nhead and @ntail are zero)
1021 * header of @skb. &sk_buff itself is not changed. &sk_buff MUST have
1022 * reference count of 1. Returns zero in the case of success or error,
1023 * if expansion failed. In the last case, &sk_buff is not changed.
1024 *
1025 * All the pointers pointing into skb header may change and must be
1026 * reloaded after call to this function.
1027 */
1028
1029int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1030 gfp_t gfp_mask)
1031{
1032 int i;
1033 u8 *data;
1034 int size = nhead + skb_end_offset(skb) + ntail;
1035 long off;
1036
1037 BUG_ON(nhead < 0);
1038
1039 if (skb_shared(skb))
1040 BUG();
1041
1042 size = SKB_DATA_ALIGN(size);
1043
1044 if (skb_pfmemalloc(skb))
1045 gfp_mask |= __GFP_MEMALLOC;
1046 data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1047 gfp_mask, NUMA_NO_NODE, NULL);
1048 if (!data)
1049 goto nodata;
1050 size = SKB_WITH_OVERHEAD(ksize(data));
1051
1052 /* Copy only real data... and, alas, header. This should be
1053 * optimized for the cases when header is void.
1054 */
1055 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1056
1057 memcpy((struct skb_shared_info *)(data + size),
1058 skb_shinfo(skb),
1059 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1060
1061 /*
1062 * if shinfo is shared we must drop the old head gracefully, but if it
1063 * is not we can just drop the old head and let the existing refcount
1064 * be since all we did is relocate the values
1065 */
1066 if (skb_cloned(skb)) {
1067 /* copy this zero copy skb frags */
1068 if (skb_orphan_frags(skb, gfp_mask))
1069 goto nofrags;
1070 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1071 skb_frag_ref(skb, i);
1072
1073 if (skb_has_frag_list(skb))
1074 skb_clone_fraglist(skb);
1075
1076 skb_release_data(skb);
1077 } else {
1078 skb_free_head(skb);
1079 }
1080 off = (data + nhead) - skb->head;
1081
1082 skb->head = data;
1083 skb->head_frag = 0;
1084 skb->data += off;
1085#ifdef NET_SKBUFF_DATA_USES_OFFSET
1086 skb->end = size;
1087 off = nhead;
1088#else
1089 skb->end = skb->head + size;
1090#endif
1091 skb->tail += off;
1092 skb_headers_offset_update(skb, nhead);
1093 skb->cloned = 0;
1094 skb->hdr_len = 0;
1095 skb->nohdr = 0;
1096 atomic_set(&skb_shinfo(skb)->dataref, 1);
1097 return 0;
1098
1099nofrags:
1100 kfree(data);
1101nodata:
1102 return -ENOMEM;
1103}
1104EXPORT_SYMBOL(pskb_expand_head);
1105
1106/* Make private copy of skb with writable head and some headroom */
1107
1108struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1109{
1110 struct sk_buff *skb2;
1111 int delta = headroom - skb_headroom(skb);
1112
1113 if (delta <= 0)
1114 skb2 = pskb_copy(skb, GFP_ATOMIC);
1115 else {
1116 skb2 = skb_clone(skb, GFP_ATOMIC);
1117 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1118 GFP_ATOMIC)) {
1119 kfree_skb(skb2);
1120 skb2 = NULL;
1121 }
1122 }
1123 return skb2;
1124}
1125EXPORT_SYMBOL(skb_realloc_headroom);
1126
1127/**
1128 * skb_copy_expand - copy and expand sk_buff
1129 * @skb: buffer to copy
1130 * @newheadroom: new free bytes at head
1131 * @newtailroom: new free bytes at tail
1132 * @gfp_mask: allocation priority
1133 *
1134 * Make a copy of both an &sk_buff and its data and while doing so
1135 * allocate additional space.
1136 *
1137 * This is used when the caller wishes to modify the data and needs a
1138 * private copy of the data to alter as well as more space for new fields.
1139 * Returns %NULL on failure or the pointer to the buffer
1140 * on success. The returned buffer has a reference count of 1.
1141 *
1142 * You must pass %GFP_ATOMIC as the allocation priority if this function
1143 * is called from an interrupt.
1144 */
1145struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1146 int newheadroom, int newtailroom,
1147 gfp_t gfp_mask)
1148{
1149 /*
1150 * Allocate the copy buffer
1151 */
1152 struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1153 gfp_mask, skb_alloc_rx_flag(skb),
1154 NUMA_NO_NODE);
1155 int oldheadroom = skb_headroom(skb);
1156 int head_copy_len, head_copy_off;
1157
1158 if (!n)
1159 return NULL;
1160
1161 skb_reserve(n, newheadroom);
1162
1163 /* Set the tail pointer and length */
1164 skb_put(n, skb->len);
1165
1166 head_copy_len = oldheadroom;
1167 head_copy_off = 0;
1168 if (newheadroom <= head_copy_len)
1169 head_copy_len = newheadroom;
1170 else
1171 head_copy_off = newheadroom - head_copy_len;
1172
1173 /* Copy the linear header and data. */
1174 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1175 skb->len + head_copy_len))
1176 BUG();
1177
1178 copy_skb_header(n, skb);
1179
1180 skb_headers_offset_update(n, newheadroom - oldheadroom);
1181
1182 return n;
1183}
1184EXPORT_SYMBOL(skb_copy_expand);
1185
1186/**
1187 * skb_pad - zero pad the tail of an skb
1188 * @skb: buffer to pad
1189 * @pad: space to pad
1190 *
1191 * Ensure that a buffer is followed by a padding area that is zero
1192 * filled. Used by network drivers which may DMA or transfer data
1193 * beyond the buffer end onto the wire.
1194 *
1195 * May return error in out of memory cases. The skb is freed on error.
1196 */
1197
1198int skb_pad(struct sk_buff *skb, int pad)
1199{
1200 int err;
1201 int ntail;
1202
1203 /* If the skbuff is non linear tailroom is always zero.. */
1204 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1205 memset(skb->data+skb->len, 0, pad);
1206 return 0;
1207 }
1208
1209 ntail = skb->data_len + pad - (skb->end - skb->tail);
1210 if (likely(skb_cloned(skb) || ntail > 0)) {
1211 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1212 if (unlikely(err))
1213 goto free_skb;
1214 }
1215
1216 /* FIXME: The use of this function with non-linear skb's really needs
1217 * to be audited.
1218 */
1219 err = skb_linearize(skb);
1220 if (unlikely(err))
1221 goto free_skb;
1222
1223 memset(skb->data + skb->len, 0, pad);
1224 return 0;
1225
1226free_skb:
1227 kfree_skb(skb);
1228 return err;
1229}
1230EXPORT_SYMBOL(skb_pad);
1231
1232/**
1233 * pskb_put - add data to the tail of a potentially fragmented buffer
1234 * @skb: start of the buffer to use
1235 * @tail: tail fragment of the buffer to use
1236 * @len: amount of data to add
1237 *
1238 * This function extends the used data area of the potentially
1239 * fragmented buffer. @tail must be the last fragment of @skb -- or
1240 * @skb itself. If this would exceed the total buffer size the kernel
1241 * will panic. A pointer to the first byte of the extra data is
1242 * returned.
1243 */
1244
1245unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len)
1246{
1247 if (tail != skb) {
1248 skb->data_len += len;
1249 skb->len += len;
1250 }
1251 return skb_put(tail, len);
1252}
1253EXPORT_SYMBOL_GPL(pskb_put);
1254
1255/**
1256 * skb_put - add data to a buffer
1257 * @skb: buffer to use
1258 * @len: amount of data to add
1259 *
1260 * This function extends the used data area of the buffer. If this would
1261 * exceed the total buffer size the kernel will panic. A pointer to the
1262 * first byte of the extra data is returned.
1263 */
1264unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1265{
1266 unsigned char *tmp = skb_tail_pointer(skb);
1267 SKB_LINEAR_ASSERT(skb);
1268 skb->tail += len;
1269 skb->len += len;
1270 if (unlikely(skb->tail > skb->end))
1271 skb_over_panic(skb, len, __builtin_return_address(0));
1272 return tmp;
1273}
1274EXPORT_SYMBOL(skb_put);
1275
1276/**
1277 * skb_push - add data to the start of a buffer
1278 * @skb: buffer to use
1279 * @len: amount of data to add
1280 *
1281 * This function extends the used data area of the buffer at the buffer
1282 * start. If this would exceed the total buffer headroom the kernel will
1283 * panic. A pointer to the first byte of the extra data is returned.
1284 */
1285unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1286{
1287 skb->data -= len;
1288 skb->len += len;
1289 if (unlikely(skb->data<skb->head))
1290 skb_under_panic(skb, len, __builtin_return_address(0));
1291 return skb->data;
1292}
1293EXPORT_SYMBOL(skb_push);
1294
1295/**
1296 * skb_pull - remove data from the start of a buffer
1297 * @skb: buffer to use
1298 * @len: amount of data to remove
1299 *
1300 * This function removes data from the start of a buffer, returning
1301 * the memory to the headroom. A pointer to the next data in the buffer
1302 * is returned. Once the data has been pulled future pushes will overwrite
1303 * the old data.
1304 */
1305unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1306{
1307 return skb_pull_inline(skb, len);
1308}
1309EXPORT_SYMBOL(skb_pull);
1310
1311/**
1312 * skb_trim - remove end from a buffer
1313 * @skb: buffer to alter
1314 * @len: new length
1315 *
1316 * Cut the length of a buffer down by removing data from the tail. If
1317 * the buffer is already under the length specified it is not modified.
1318 * The skb must be linear.
1319 */
1320void skb_trim(struct sk_buff *skb, unsigned int len)
1321{
1322 if (skb->len > len)
1323 __skb_trim(skb, len);
1324}
1325EXPORT_SYMBOL(skb_trim);
1326
1327/* Trims skb to length len. It can change skb pointers.
1328 */
1329
1330int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1331{
1332 struct sk_buff **fragp;
1333 struct sk_buff *frag;
1334 int offset = skb_headlen(skb);
1335 int nfrags = skb_shinfo(skb)->nr_frags;
1336 int i;
1337 int err;
1338
1339 if (skb_cloned(skb) &&
1340 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1341 return err;
1342
1343 i = 0;
1344 if (offset >= len)
1345 goto drop_pages;
1346
1347 for (; i < nfrags; i++) {
1348 int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1349
1350 if (end < len) {
1351 offset = end;
1352 continue;
1353 }
1354
1355 skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1356
1357drop_pages:
1358 skb_shinfo(skb)->nr_frags = i;
1359
1360 for (; i < nfrags; i++)
1361 skb_frag_unref(skb, i);
1362
1363 if (skb_has_frag_list(skb))
1364 skb_drop_fraglist(skb);
1365 goto done;
1366 }
1367
1368 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1369 fragp = &frag->next) {
1370 int end = offset + frag->len;
1371
1372 if (skb_shared(frag)) {
1373 struct sk_buff *nfrag;
1374
1375 nfrag = skb_clone(frag, GFP_ATOMIC);
1376 if (unlikely(!nfrag))
1377 return -ENOMEM;
1378
1379 nfrag->next = frag->next;
1380 consume_skb(frag);
1381 frag = nfrag;
1382 *fragp = frag;
1383 }
1384
1385 if (end < len) {
1386 offset = end;
1387 continue;
1388 }
1389
1390 if (end > len &&
1391 unlikely((err = pskb_trim(frag, len - offset))))
1392 return err;
1393
1394 if (frag->next)
1395 skb_drop_list(&frag->next);
1396 break;
1397 }
1398
1399done:
1400 if (len > skb_headlen(skb)) {
1401 skb->data_len -= skb->len - len;
1402 skb->len = len;
1403 } else {
1404 skb->len = len;
1405 skb->data_len = 0;
1406 skb_set_tail_pointer(skb, len);
1407 }
1408
1409 return 0;
1410}
1411EXPORT_SYMBOL(___pskb_trim);
1412
1413/**
1414 * __pskb_pull_tail - advance tail of skb header
1415 * @skb: buffer to reallocate
1416 * @delta: number of bytes to advance tail
1417 *
1418 * The function makes a sense only on a fragmented &sk_buff,
1419 * it expands header moving its tail forward and copying necessary
1420 * data from fragmented part.
1421 *
1422 * &sk_buff MUST have reference count of 1.
1423 *
1424 * Returns %NULL (and &sk_buff does not change) if pull failed
1425 * or value of new tail of skb in the case of success.
1426 *
1427 * All the pointers pointing into skb header may change and must be
1428 * reloaded after call to this function.
1429 */
1430
1431/* Moves tail of skb head forward, copying data from fragmented part,
1432 * when it is necessary.
1433 * 1. It may fail due to malloc failure.
1434 * 2. It may change skb pointers.
1435 *
1436 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1437 */
1438unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1439{
1440 /* If skb has not enough free space at tail, get new one
1441 * plus 128 bytes for future expansions. If we have enough
1442 * room at tail, reallocate without expansion only if skb is cloned.
1443 */
1444 int i, k, eat = (skb->tail + delta) - skb->end;
1445
1446 if (eat > 0 || skb_cloned(skb)) {
1447 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1448 GFP_ATOMIC))
1449 return NULL;
1450 }
1451
1452 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1453 BUG();
1454
1455 /* Optimization: no fragments, no reasons to preestimate
1456 * size of pulled pages. Superb.
1457 */
1458 if (!skb_has_frag_list(skb))
1459 goto pull_pages;
1460
1461 /* Estimate size of pulled pages. */
1462 eat = delta;
1463 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1464 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1465
1466 if (size >= eat)
1467 goto pull_pages;
1468 eat -= size;
1469 }
1470
1471 /* If we need update frag list, we are in troubles.
1472 * Certainly, it possible to add an offset to skb data,
1473 * but taking into account that pulling is expected to
1474 * be very rare operation, it is worth to fight against
1475 * further bloating skb head and crucify ourselves here instead.
1476 * Pure masohism, indeed. 8)8)
1477 */
1478 if (eat) {
1479 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1480 struct sk_buff *clone = NULL;
1481 struct sk_buff *insp = NULL;
1482
1483 do {
1484 BUG_ON(!list);
1485
1486 if (list->len <= eat) {
1487 /* Eaten as whole. */
1488 eat -= list->len;
1489 list = list->next;
1490 insp = list;
1491 } else {
1492 /* Eaten partially. */
1493
1494 if (skb_shared(list)) {
1495 /* Sucks! We need to fork list. :-( */
1496 clone = skb_clone(list, GFP_ATOMIC);
1497 if (!clone)
1498 return NULL;
1499 insp = list->next;
1500 list = clone;
1501 } else {
1502 /* This may be pulled without
1503 * problems. */
1504 insp = list;
1505 }
1506 if (!pskb_pull(list, eat)) {
1507 kfree_skb(clone);
1508 return NULL;
1509 }
1510 break;
1511 }
1512 } while (eat);
1513
1514 /* Free pulled out fragments. */
1515 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1516 skb_shinfo(skb)->frag_list = list->next;
1517 kfree_skb(list);
1518 }
1519 /* And insert new clone at head. */
1520 if (clone) {
1521 clone->next = list;
1522 skb_shinfo(skb)->frag_list = clone;
1523 }
1524 }
1525 /* Success! Now we may commit changes to skb data. */
1526
1527pull_pages:
1528 eat = delta;
1529 k = 0;
1530 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1531 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1532
1533 if (size <= eat) {
1534 skb_frag_unref(skb, i);
1535 eat -= size;
1536 } else {
1537 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1538 if (eat) {
1539 skb_shinfo(skb)->frags[k].page_offset += eat;
1540 skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1541 eat = 0;
1542 }
1543 k++;
1544 }
1545 }
1546 skb_shinfo(skb)->nr_frags = k;
1547
1548 skb->tail += delta;
1549 skb->data_len -= delta;
1550
1551 return skb_tail_pointer(skb);
1552}
1553EXPORT_SYMBOL(__pskb_pull_tail);
1554
1555/**
1556 * skb_copy_bits - copy bits from skb to kernel buffer
1557 * @skb: source skb
1558 * @offset: offset in source
1559 * @to: destination buffer
1560 * @len: number of bytes to copy
1561 *
1562 * Copy the specified number of bytes from the source skb to the
1563 * destination buffer.
1564 *
1565 * CAUTION ! :
1566 * If its prototype is ever changed,
1567 * check arch/{*}/net/{*}.S files,
1568 * since it is called from BPF assembly code.
1569 */
1570int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1571{
1572 int start = skb_headlen(skb);
1573 struct sk_buff *frag_iter;
1574 int i, copy;
1575
1576 if (offset > (int)skb->len - len)
1577 goto fault;
1578
1579 /* Copy header. */
1580 if ((copy = start - offset) > 0) {
1581 if (copy > len)
1582 copy = len;
1583 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1584 if ((len -= copy) == 0)
1585 return 0;
1586 offset += copy;
1587 to += copy;
1588 }
1589
1590 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1591 int end;
1592 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1593
1594 WARN_ON(start > offset + len);
1595
1596 end = start + skb_frag_size(f);
1597 if ((copy = end - offset) > 0) {
1598 u8 *vaddr;
1599
1600 if (copy > len)
1601 copy = len;
1602
1603 vaddr = kmap_atomic(skb_frag_page(f));
1604 memcpy(to,
1605 vaddr + f->page_offset + offset - start,
1606 copy);
1607 kunmap_atomic(vaddr);
1608
1609 if ((len -= copy) == 0)
1610 return 0;
1611 offset += copy;
1612 to += copy;
1613 }
1614 start = end;
1615 }
1616
1617 skb_walk_frags(skb, frag_iter) {
1618 int end;
1619
1620 WARN_ON(start > offset + len);
1621
1622 end = start + frag_iter->len;
1623 if ((copy = end - offset) > 0) {
1624 if (copy > len)
1625 copy = len;
1626 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1627 goto fault;
1628 if ((len -= copy) == 0)
1629 return 0;
1630 offset += copy;
1631 to += copy;
1632 }
1633 start = end;
1634 }
1635
1636 if (!len)
1637 return 0;
1638
1639fault:
1640 return -EFAULT;
1641}
1642EXPORT_SYMBOL(skb_copy_bits);
1643
1644/*
1645 * Callback from splice_to_pipe(), if we need to release some pages
1646 * at the end of the spd in case we error'ed out in filling the pipe.
1647 */
1648static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1649{
1650 put_page(spd->pages[i]);
1651}
1652
1653static struct page *linear_to_page(struct page *page, unsigned int *len,
1654 unsigned int *offset,
1655 struct sock *sk)
1656{
1657 struct page_frag *pfrag = sk_page_frag(sk);
1658
1659 if (!sk_page_frag_refill(sk, pfrag))
1660 return NULL;
1661
1662 *len = min_t(unsigned int, *len, pfrag->size - pfrag->offset);
1663
1664 memcpy(page_address(pfrag->page) + pfrag->offset,
1665 page_address(page) + *offset, *len);
1666 *offset = pfrag->offset;
1667 pfrag->offset += *len;
1668
1669 return pfrag->page;
1670}
1671
1672static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1673 struct page *page,
1674 unsigned int offset)
1675{
1676 return spd->nr_pages &&
1677 spd->pages[spd->nr_pages - 1] == page &&
1678 (spd->partial[spd->nr_pages - 1].offset +
1679 spd->partial[spd->nr_pages - 1].len == offset);
1680}
1681
1682/*
1683 * Fill page/offset/length into spd, if it can hold more pages.
1684 */
1685static bool spd_fill_page(struct splice_pipe_desc *spd,
1686 struct pipe_inode_info *pipe, struct page *page,
1687 unsigned int *len, unsigned int offset,
1688 bool linear,
1689 struct sock *sk)
1690{
1691 if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1692 return true;
1693
1694 if (linear) {
1695 page = linear_to_page(page, len, &offset, sk);
1696 if (!page)
1697 return true;
1698 }
1699 if (spd_can_coalesce(spd, page, offset)) {
1700 spd->partial[spd->nr_pages - 1].len += *len;
1701 return false;
1702 }
1703 get_page(page);
1704 spd->pages[spd->nr_pages] = page;
1705 spd->partial[spd->nr_pages].len = *len;
1706 spd->partial[spd->nr_pages].offset = offset;
1707 spd->nr_pages++;
1708
1709 return false;
1710}
1711
1712static bool __splice_segment(struct page *page, unsigned int poff,
1713 unsigned int plen, unsigned int *off,
1714 unsigned int *len,
1715 struct splice_pipe_desc *spd, bool linear,
1716 struct sock *sk,
1717 struct pipe_inode_info *pipe)
1718{
1719 if (!*len)
1720 return true;
1721
1722 /* skip this segment if already processed */
1723 if (*off >= plen) {
1724 *off -= plen;
1725 return false;
1726 }
1727
1728 /* ignore any bits we already processed */
1729 poff += *off;
1730 plen -= *off;
1731 *off = 0;
1732
1733 do {
1734 unsigned int flen = min(*len, plen);
1735
1736 if (spd_fill_page(spd, pipe, page, &flen, poff,
1737 linear, sk))
1738 return true;
1739 poff += flen;
1740 plen -= flen;
1741 *len -= flen;
1742 } while (*len && plen);
1743
1744 return false;
1745}
1746
1747/*
1748 * Map linear and fragment data from the skb to spd. It reports true if the
1749 * pipe is full or if we already spliced the requested length.
1750 */
1751static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1752 unsigned int *offset, unsigned int *len,
1753 struct splice_pipe_desc *spd, struct sock *sk)
1754{
1755 int seg;
1756
1757 /* map the linear part :
1758 * If skb->head_frag is set, this 'linear' part is backed by a
1759 * fragment, and if the head is not shared with any clones then
1760 * we can avoid a copy since we own the head portion of this page.
1761 */
1762 if (__splice_segment(virt_to_page(skb->data),
1763 (unsigned long) skb->data & (PAGE_SIZE - 1),
1764 skb_headlen(skb),
1765 offset, len, spd,
1766 skb_head_is_locked(skb),
1767 sk, pipe))
1768 return true;
1769
1770 /*
1771 * then map the fragments
1772 */
1773 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1774 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1775
1776 if (__splice_segment(skb_frag_page(f),
1777 f->page_offset, skb_frag_size(f),
1778 offset, len, spd, false, sk, pipe))
1779 return true;
1780 }
1781
1782 return false;
1783}
1784
1785/*
1786 * Map data from the skb to a pipe. Should handle both the linear part,
1787 * the fragments, and the frag list. It does NOT handle frag lists within
1788 * the frag list, if such a thing exists. We'd probably need to recurse to
1789 * handle that cleanly.
1790 */
1791int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1792 struct pipe_inode_info *pipe, unsigned int tlen,
1793 unsigned int flags)
1794{
1795 struct partial_page partial[MAX_SKB_FRAGS];
1796 struct page *pages[MAX_SKB_FRAGS];
1797 struct splice_pipe_desc spd = {
1798 .pages = pages,
1799 .partial = partial,
1800 .nr_pages_max = MAX_SKB_FRAGS,
1801 .flags = flags,
1802 .ops = &nosteal_pipe_buf_ops,
1803 .spd_release = sock_spd_release,
1804 };
1805 struct sk_buff *frag_iter;
1806 struct sock *sk = skb->sk;
1807 int ret = 0;
1808
1809 /*
1810 * __skb_splice_bits() only fails if the output has no room left,
1811 * so no point in going over the frag_list for the error case.
1812 */
1813 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1814 goto done;
1815 else if (!tlen)
1816 goto done;
1817
1818 /*
1819 * now see if we have a frag_list to map
1820 */
1821 skb_walk_frags(skb, frag_iter) {
1822 if (!tlen)
1823 break;
1824 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1825 break;
1826 }
1827
1828done:
1829 if (spd.nr_pages) {
1830 /*
1831 * Drop the socket lock, otherwise we have reverse
1832 * locking dependencies between sk_lock and i_mutex
1833 * here as compared to sendfile(). We enter here
1834 * with the socket lock held, and splice_to_pipe() will
1835 * grab the pipe inode lock. For sendfile() emulation,
1836 * we call into ->sendpage() with the i_mutex lock held
1837 * and networking will grab the socket lock.
1838 */
1839 release_sock(sk);
1840 ret = splice_to_pipe(pipe, &spd);
1841 lock_sock(sk);
1842 }
1843
1844 return ret;
1845}
1846
1847/**
1848 * skb_store_bits - store bits from kernel buffer to skb
1849 * @skb: destination buffer
1850 * @offset: offset in destination
1851 * @from: source buffer
1852 * @len: number of bytes to copy
1853 *
1854 * Copy the specified number of bytes from the source buffer to the
1855 * destination skb. This function handles all the messy bits of
1856 * traversing fragment lists and such.
1857 */
1858
1859int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1860{
1861 int start = skb_headlen(skb);
1862 struct sk_buff *frag_iter;
1863 int i, copy;
1864
1865 if (offset > (int)skb->len - len)
1866 goto fault;
1867
1868 if ((copy = start - offset) > 0) {
1869 if (copy > len)
1870 copy = len;
1871 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1872 if ((len -= copy) == 0)
1873 return 0;
1874 offset += copy;
1875 from += copy;
1876 }
1877
1878 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1879 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1880 int end;
1881
1882 WARN_ON(start > offset + len);
1883
1884 end = start + skb_frag_size(frag);
1885 if ((copy = end - offset) > 0) {
1886 u8 *vaddr;
1887
1888 if (copy > len)
1889 copy = len;
1890
1891 vaddr = kmap_atomic(skb_frag_page(frag));
1892 memcpy(vaddr + frag->page_offset + offset - start,
1893 from, copy);
1894 kunmap_atomic(vaddr);
1895
1896 if ((len -= copy) == 0)
1897 return 0;
1898 offset += copy;
1899 from += copy;
1900 }
1901 start = end;
1902 }
1903
1904 skb_walk_frags(skb, frag_iter) {
1905 int end;
1906
1907 WARN_ON(start > offset + len);
1908
1909 end = start + frag_iter->len;
1910 if ((copy = end - offset) > 0) {
1911 if (copy > len)
1912 copy = len;
1913 if (skb_store_bits(frag_iter, offset - start,
1914 from, copy))
1915 goto fault;
1916 if ((len -= copy) == 0)
1917 return 0;
1918 offset += copy;
1919 from += copy;
1920 }
1921 start = end;
1922 }
1923 if (!len)
1924 return 0;
1925
1926fault:
1927 return -EFAULT;
1928}
1929EXPORT_SYMBOL(skb_store_bits);
1930
1931/* Checksum skb data. */
1932__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
1933 __wsum csum, const struct skb_checksum_ops *ops)
1934{
1935 int start = skb_headlen(skb);
1936 int i, copy = start - offset;
1937 struct sk_buff *frag_iter;
1938 int pos = 0;
1939
1940 /* Checksum header. */
1941 if (copy > 0) {
1942 if (copy > len)
1943 copy = len;
1944 csum = ops->update(skb->data + offset, copy, csum);
1945 if ((len -= copy) == 0)
1946 return csum;
1947 offset += copy;
1948 pos = copy;
1949 }
1950
1951 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1952 int end;
1953 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1954
1955 WARN_ON(start > offset + len);
1956
1957 end = start + skb_frag_size(frag);
1958 if ((copy = end - offset) > 0) {
1959 __wsum csum2;
1960 u8 *vaddr;
1961
1962 if (copy > len)
1963 copy = len;
1964 vaddr = kmap_atomic(skb_frag_page(frag));
1965 csum2 = ops->update(vaddr + frag->page_offset +
1966 offset - start, copy, 0);
1967 kunmap_atomic(vaddr);
1968 csum = ops->combine(csum, csum2, pos, copy);
1969 if (!(len -= copy))
1970 return csum;
1971 offset += copy;
1972 pos += copy;
1973 }
1974 start = end;
1975 }
1976
1977 skb_walk_frags(skb, frag_iter) {
1978 int end;
1979
1980 WARN_ON(start > offset + len);
1981
1982 end = start + frag_iter->len;
1983 if ((copy = end - offset) > 0) {
1984 __wsum csum2;
1985 if (copy > len)
1986 copy = len;
1987 csum2 = __skb_checksum(frag_iter, offset - start,
1988 copy, 0, ops);
1989 csum = ops->combine(csum, csum2, pos, copy);
1990 if ((len -= copy) == 0)
1991 return csum;
1992 offset += copy;
1993 pos += copy;
1994 }
1995 start = end;
1996 }
1997 BUG_ON(len);
1998
1999 return csum;
2000}
2001EXPORT_SYMBOL(__skb_checksum);
2002
2003__wsum skb_checksum(const struct sk_buff *skb, int offset,
2004 int len, __wsum csum)
2005{
2006 const struct skb_checksum_ops ops = {
2007 .update = csum_partial_ext,
2008 .combine = csum_block_add_ext,
2009 };
2010
2011 return __skb_checksum(skb, offset, len, csum, &ops);
2012}
2013EXPORT_SYMBOL(skb_checksum);
2014
2015/* Both of above in one bottle. */
2016
2017__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2018 u8 *to, int len, __wsum csum)
2019{
2020 int start = skb_headlen(skb);
2021 int i, copy = start - offset;
2022 struct sk_buff *frag_iter;
2023 int pos = 0;
2024
2025 /* Copy header. */
2026 if (copy > 0) {
2027 if (copy > len)
2028 copy = len;
2029 csum = csum_partial_copy_nocheck(skb->data + offset, to,
2030 copy, csum);
2031 if ((len -= copy) == 0)
2032 return csum;
2033 offset += copy;
2034 to += copy;
2035 pos = copy;
2036 }
2037
2038 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2039 int end;
2040
2041 WARN_ON(start > offset + len);
2042
2043 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2044 if ((copy = end - offset) > 0) {
2045 __wsum csum2;
2046 u8 *vaddr;
2047 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2048
2049 if (copy > len)
2050 copy = len;
2051 vaddr = kmap_atomic(skb_frag_page(frag));
2052 csum2 = csum_partial_copy_nocheck(vaddr +
2053 frag->page_offset +
2054 offset - start, to,
2055 copy, 0);
2056 kunmap_atomic(vaddr);
2057 csum = csum_block_add(csum, csum2, pos);
2058 if (!(len -= copy))
2059 return csum;
2060 offset += copy;
2061 to += copy;
2062 pos += copy;
2063 }
2064 start = end;
2065 }
2066
2067 skb_walk_frags(skb, frag_iter) {
2068 __wsum csum2;
2069 int end;
2070
2071 WARN_ON(start > offset + len);
2072
2073 end = start + frag_iter->len;
2074 if ((copy = end - offset) > 0) {
2075 if (copy > len)
2076 copy = len;
2077 csum2 = skb_copy_and_csum_bits(frag_iter,
2078 offset - start,
2079 to, copy, 0);
2080 csum = csum_block_add(csum, csum2, pos);
2081 if ((len -= copy) == 0)
2082 return csum;
2083 offset += copy;
2084 to += copy;
2085 pos += copy;
2086 }
2087 start = end;
2088 }
2089 BUG_ON(len);
2090 return csum;
2091}
2092EXPORT_SYMBOL(skb_copy_and_csum_bits);
2093
2094 /**
2095 * skb_zerocopy_headlen - Calculate headroom needed for skb_zerocopy()
2096 * @from: source buffer
2097 *
2098 * Calculates the amount of linear headroom needed in the 'to' skb passed
2099 * into skb_zerocopy().
2100 */
2101unsigned int
2102skb_zerocopy_headlen(const struct sk_buff *from)
2103{
2104 unsigned int hlen = 0;
2105
2106 if (!from->head_frag ||
2107 skb_headlen(from) < L1_CACHE_BYTES ||
2108 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
2109 hlen = skb_headlen(from);
2110
2111 if (skb_has_frag_list(from))
2112 hlen = from->len;
2113
2114 return hlen;
2115}
2116EXPORT_SYMBOL_GPL(skb_zerocopy_headlen);
2117
2118/**
2119 * skb_zerocopy - Zero copy skb to skb
2120 * @to: destination buffer
2121 * @from: source buffer
2122 * @len: number of bytes to copy from source buffer
2123 * @hlen: size of linear headroom in destination buffer
2124 *
2125 * Copies up to `len` bytes from `from` to `to` by creating references
2126 * to the frags in the source buffer.
2127 *
2128 * The `hlen` as calculated by skb_zerocopy_headlen() specifies the
2129 * headroom in the `to` buffer.
2130 *
2131 * Return value:
2132 * 0: everything is OK
2133 * -ENOMEM: couldn't orphan frags of @from due to lack of memory
2134 * -EFAULT: skb_copy_bits() found some problem with skb geometry
2135 */
2136int
2137skb_zerocopy(struct sk_buff *to, struct sk_buff *from, int len, int hlen)
2138{
2139 int i, j = 0;
2140 int plen = 0; /* length of skb->head fragment */
2141 int ret;
2142 struct page *page;
2143 unsigned int offset;
2144
2145 BUG_ON(!from->head_frag && !hlen);
2146
2147 /* dont bother with small payloads */
2148 if (len <= skb_tailroom(to))
2149 return skb_copy_bits(from, 0, skb_put(to, len), len);
2150
2151 if (hlen) {
2152 ret = skb_copy_bits(from, 0, skb_put(to, hlen), hlen);
2153 if (unlikely(ret))
2154 return ret;
2155 len -= hlen;
2156 } else {
2157 plen = min_t(int, skb_headlen(from), len);
2158 if (plen) {
2159 page = virt_to_head_page(from->head);
2160 offset = from->data - (unsigned char *)page_address(page);
2161 __skb_fill_page_desc(to, 0, page, offset, plen);
2162 get_page(page);
2163 j = 1;
2164 len -= plen;
2165 }
2166 }
2167
2168 to->truesize += len + plen;
2169 to->len += len + plen;
2170 to->data_len += len + plen;
2171
2172 if (unlikely(skb_orphan_frags(from, GFP_ATOMIC))) {
2173 skb_tx_error(from);
2174 return -ENOMEM;
2175 }
2176
2177 for (i = 0; i < skb_shinfo(from)->nr_frags; i++) {
2178 if (!len)
2179 break;
2180 skb_shinfo(to)->frags[j] = skb_shinfo(from)->frags[i];
2181 skb_shinfo(to)->frags[j].size = min_t(int, skb_shinfo(to)->frags[j].size, len);
2182 len -= skb_shinfo(to)->frags[j].size;
2183 skb_frag_ref(to, j);
2184 j++;
2185 }
2186 skb_shinfo(to)->nr_frags = j;
2187
2188 return 0;
2189}
2190EXPORT_SYMBOL_GPL(skb_zerocopy);
2191
2192void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2193{
2194 __wsum csum;
2195 long csstart;
2196
2197 if (skb->ip_summed == CHECKSUM_PARTIAL)
2198 csstart = skb_checksum_start_offset(skb);
2199 else
2200 csstart = skb_headlen(skb);
2201
2202 BUG_ON(csstart > skb_headlen(skb));
2203
2204 skb_copy_from_linear_data(skb, to, csstart);
2205
2206 csum = 0;
2207 if (csstart != skb->len)
2208 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2209 skb->len - csstart, 0);
2210
2211 if (skb->ip_summed == CHECKSUM_PARTIAL) {
2212 long csstuff = csstart + skb->csum_offset;
2213
2214 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2215 }
2216}
2217EXPORT_SYMBOL(skb_copy_and_csum_dev);
2218
2219/**
2220 * skb_dequeue - remove from the head of the queue
2221 * @list: list to dequeue from
2222 *
2223 * Remove the head of the list. The list lock is taken so the function
2224 * may be used safely with other locking list functions. The head item is
2225 * returned or %NULL if the list is empty.
2226 */
2227
2228struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2229{
2230 unsigned long flags;
2231 struct sk_buff *result;
2232
2233 spin_lock_irqsave(&list->lock, flags);
2234 result = __skb_dequeue(list);
2235 spin_unlock_irqrestore(&list->lock, flags);
2236 return result;
2237}
2238EXPORT_SYMBOL(skb_dequeue);
2239
2240/**
2241 * skb_dequeue_tail - remove from the tail of the queue
2242 * @list: list to dequeue from
2243 *
2244 * Remove the tail of the list. The list lock is taken so the function
2245 * may be used safely with other locking list functions. The tail item is
2246 * returned or %NULL if the list is empty.
2247 */
2248struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2249{
2250 unsigned long flags;
2251 struct sk_buff *result;
2252
2253 spin_lock_irqsave(&list->lock, flags);
2254 result = __skb_dequeue_tail(list);
2255 spin_unlock_irqrestore(&list->lock, flags);
2256 return result;
2257}
2258EXPORT_SYMBOL(skb_dequeue_tail);
2259
2260/**
2261 * skb_queue_purge - empty a list
2262 * @list: list to empty
2263 *
2264 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2265 * the list and one reference dropped. This function takes the list
2266 * lock and is atomic with respect to other list locking functions.
2267 */
2268void skb_queue_purge(struct sk_buff_head *list)
2269{
2270 struct sk_buff *skb;
2271 while ((skb = skb_dequeue(list)) != NULL)
2272 kfree_skb(skb);
2273}
2274EXPORT_SYMBOL(skb_queue_purge);
2275
2276/**
2277 * skb_queue_head - queue a buffer at the list head
2278 * @list: list to use
2279 * @newsk: buffer to queue
2280 *
2281 * Queue a buffer at the start of the list. This function takes the
2282 * list lock and can be used safely with other locking &sk_buff functions
2283 * safely.
2284 *
2285 * A buffer cannot be placed on two lists at the same time.
2286 */
2287void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2288{
2289 unsigned long flags;
2290
2291 spin_lock_irqsave(&list->lock, flags);
2292 __skb_queue_head(list, newsk);
2293 spin_unlock_irqrestore(&list->lock, flags);
2294}
2295EXPORT_SYMBOL(skb_queue_head);
2296
2297/**
2298 * skb_queue_tail - queue a buffer at the list tail
2299 * @list: list to use
2300 * @newsk: buffer to queue
2301 *
2302 * Queue a buffer at the tail of the list. This function takes the
2303 * list lock and can be used safely with other locking &sk_buff functions
2304 * safely.
2305 *
2306 * A buffer cannot be placed on two lists at the same time.
2307 */
2308void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2309{
2310 unsigned long flags;
2311
2312 spin_lock_irqsave(&list->lock, flags);
2313 __skb_queue_tail(list, newsk);
2314 spin_unlock_irqrestore(&list->lock, flags);
2315}
2316EXPORT_SYMBOL(skb_queue_tail);
2317
2318/**
2319 * skb_unlink - remove a buffer from a list
2320 * @skb: buffer to remove
2321 * @list: list to use
2322 *
2323 * Remove a packet from a list. The list locks are taken and this
2324 * function is atomic with respect to other list locked calls
2325 *
2326 * You must know what list the SKB is on.
2327 */
2328void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2329{
2330 unsigned long flags;
2331
2332 spin_lock_irqsave(&list->lock, flags);
2333 __skb_unlink(skb, list);
2334 spin_unlock_irqrestore(&list->lock, flags);
2335}
2336EXPORT_SYMBOL(skb_unlink);
2337
2338/**
2339 * skb_append - append a buffer
2340 * @old: buffer to insert after
2341 * @newsk: buffer to insert
2342 * @list: list to use
2343 *
2344 * Place a packet after a given packet in a list. The list locks are taken
2345 * and this function is atomic with respect to other list locked calls.
2346 * A buffer cannot be placed on two lists at the same time.
2347 */
2348void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2349{
2350 unsigned long flags;
2351
2352 spin_lock_irqsave(&list->lock, flags);
2353 __skb_queue_after(list, old, newsk);
2354 spin_unlock_irqrestore(&list->lock, flags);
2355}
2356EXPORT_SYMBOL(skb_append);
2357
2358/**
2359 * skb_insert - insert a buffer
2360 * @old: buffer to insert before
2361 * @newsk: buffer to insert
2362 * @list: list to use
2363 *
2364 * Place a packet before a given packet in a list. The list locks are
2365 * taken and this function is atomic with respect to other list locked
2366 * calls.
2367 *
2368 * A buffer cannot be placed on two lists at the same time.
2369 */
2370void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2371{
2372 unsigned long flags;
2373
2374 spin_lock_irqsave(&list->lock, flags);
2375 __skb_insert(newsk, old->prev, old, list);
2376 spin_unlock_irqrestore(&list->lock, flags);
2377}
2378EXPORT_SYMBOL(skb_insert);
2379
2380static inline void skb_split_inside_header(struct sk_buff *skb,
2381 struct sk_buff* skb1,
2382 const u32 len, const int pos)
2383{
2384 int i;
2385
2386 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2387 pos - len);
2388 /* And move data appendix as is. */
2389 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2390 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2391
2392 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2393 skb_shinfo(skb)->nr_frags = 0;
2394 skb1->data_len = skb->data_len;
2395 skb1->len += skb1->data_len;
2396 skb->data_len = 0;
2397 skb->len = len;
2398 skb_set_tail_pointer(skb, len);
2399}
2400
2401static inline void skb_split_no_header(struct sk_buff *skb,
2402 struct sk_buff* skb1,
2403 const u32 len, int pos)
2404{
2405 int i, k = 0;
2406 const int nfrags = skb_shinfo(skb)->nr_frags;
2407
2408 skb_shinfo(skb)->nr_frags = 0;
2409 skb1->len = skb1->data_len = skb->len - len;
2410 skb->len = len;
2411 skb->data_len = len - pos;
2412
2413 for (i = 0; i < nfrags; i++) {
2414 int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2415
2416 if (pos + size > len) {
2417 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2418
2419 if (pos < len) {
2420 /* Split frag.
2421 * We have two variants in this case:
2422 * 1. Move all the frag to the second
2423 * part, if it is possible. F.e.
2424 * this approach is mandatory for TUX,
2425 * where splitting is expensive.
2426 * 2. Split is accurately. We make this.
2427 */
2428 skb_frag_ref(skb, i);
2429 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2430 skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2431 skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2432 skb_shinfo(skb)->nr_frags++;
2433 }
2434 k++;
2435 } else
2436 skb_shinfo(skb)->nr_frags++;
2437 pos += size;
2438 }
2439 skb_shinfo(skb1)->nr_frags = k;
2440}
2441
2442/**
2443 * skb_split - Split fragmented skb to two parts at length len.
2444 * @skb: the buffer to split
2445 * @skb1: the buffer to receive the second part
2446 * @len: new length for skb
2447 */
2448void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2449{
2450 int pos = skb_headlen(skb);
2451
2452 skb_shinfo(skb1)->tx_flags = skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2453 if (len < pos) /* Split line is inside header. */
2454 skb_split_inside_header(skb, skb1, len, pos);
2455 else /* Second chunk has no header, nothing to copy. */
2456 skb_split_no_header(skb, skb1, len, pos);
2457}
2458EXPORT_SYMBOL(skb_split);
2459
2460/* Shifting from/to a cloned skb is a no-go.
2461 *
2462 * Caller cannot keep skb_shinfo related pointers past calling here!
2463 */
2464static int skb_prepare_for_shift(struct sk_buff *skb)
2465{
2466 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2467}
2468
2469/**
2470 * skb_shift - Shifts paged data partially from skb to another
2471 * @tgt: buffer into which tail data gets added
2472 * @skb: buffer from which the paged data comes from
2473 * @shiftlen: shift up to this many bytes
2474 *
2475 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2476 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2477 * It's up to caller to free skb if everything was shifted.
2478 *
2479 * If @tgt runs out of frags, the whole operation is aborted.
2480 *
2481 * Skb cannot include anything else but paged data while tgt is allowed
2482 * to have non-paged data as well.
2483 *
2484 * TODO: full sized shift could be optimized but that would need
2485 * specialized skb free'er to handle frags without up-to-date nr_frags.
2486 */
2487int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2488{
2489 int from, to, merge, todo;
2490 struct skb_frag_struct *fragfrom, *fragto;
2491
2492 BUG_ON(shiftlen > skb->len);
2493 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2494
2495 todo = shiftlen;
2496 from = 0;
2497 to = skb_shinfo(tgt)->nr_frags;
2498 fragfrom = &skb_shinfo(skb)->frags[from];
2499
2500 /* Actual merge is delayed until the point when we know we can
2501 * commit all, so that we don't have to undo partial changes
2502 */
2503 if (!to ||
2504 !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2505 fragfrom->page_offset)) {
2506 merge = -1;
2507 } else {
2508 merge = to - 1;
2509
2510 todo -= skb_frag_size(fragfrom);
2511 if (todo < 0) {
2512 if (skb_prepare_for_shift(skb) ||
2513 skb_prepare_for_shift(tgt))
2514 return 0;
2515
2516 /* All previous frag pointers might be stale! */
2517 fragfrom = &skb_shinfo(skb)->frags[from];
2518 fragto = &skb_shinfo(tgt)->frags[merge];
2519
2520 skb_frag_size_add(fragto, shiftlen);
2521 skb_frag_size_sub(fragfrom, shiftlen);
2522 fragfrom->page_offset += shiftlen;
2523
2524 goto onlymerged;
2525 }
2526
2527 from++;
2528 }
2529
2530 /* Skip full, not-fitting skb to avoid expensive operations */
2531 if ((shiftlen == skb->len) &&
2532 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2533 return 0;
2534
2535 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2536 return 0;
2537
2538 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2539 if (to == MAX_SKB_FRAGS)
2540 return 0;
2541
2542 fragfrom = &skb_shinfo(skb)->frags[from];
2543 fragto = &skb_shinfo(tgt)->frags[to];
2544
2545 if (todo >= skb_frag_size(fragfrom)) {
2546 *fragto = *fragfrom;
2547 todo -= skb_frag_size(fragfrom);
2548 from++;
2549 to++;
2550
2551 } else {
2552 __skb_frag_ref(fragfrom);
2553 fragto->page = fragfrom->page;
2554 fragto->page_offset = fragfrom->page_offset;
2555 skb_frag_size_set(fragto, todo);
2556
2557 fragfrom->page_offset += todo;
2558 skb_frag_size_sub(fragfrom, todo);
2559 todo = 0;
2560
2561 to++;
2562 break;
2563 }
2564 }
2565
2566 /* Ready to "commit" this state change to tgt */
2567 skb_shinfo(tgt)->nr_frags = to;
2568
2569 if (merge >= 0) {
2570 fragfrom = &skb_shinfo(skb)->frags[0];
2571 fragto = &skb_shinfo(tgt)->frags[merge];
2572
2573 skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2574 __skb_frag_unref(fragfrom);
2575 }
2576
2577 /* Reposition in the original skb */
2578 to = 0;
2579 while (from < skb_shinfo(skb)->nr_frags)
2580 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2581 skb_shinfo(skb)->nr_frags = to;
2582
2583 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2584
2585onlymerged:
2586 /* Most likely the tgt won't ever need its checksum anymore, skb on
2587 * the other hand might need it if it needs to be resent
2588 */
2589 tgt->ip_summed = CHECKSUM_PARTIAL;
2590 skb->ip_summed = CHECKSUM_PARTIAL;
2591
2592 /* Yak, is it really working this way? Some helper please? */
2593 skb->len -= shiftlen;
2594 skb->data_len -= shiftlen;
2595 skb->truesize -= shiftlen;
2596 tgt->len += shiftlen;
2597 tgt->data_len += shiftlen;
2598 tgt->truesize += shiftlen;
2599
2600 return shiftlen;
2601}
2602
2603/**
2604 * skb_prepare_seq_read - Prepare a sequential read of skb data
2605 * @skb: the buffer to read
2606 * @from: lower offset of data to be read
2607 * @to: upper offset of data to be read
2608 * @st: state variable
2609 *
2610 * Initializes the specified state variable. Must be called before
2611 * invoking skb_seq_read() for the first time.
2612 */
2613void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2614 unsigned int to, struct skb_seq_state *st)
2615{
2616 st->lower_offset = from;
2617 st->upper_offset = to;
2618 st->root_skb = st->cur_skb = skb;
2619 st->frag_idx = st->stepped_offset = 0;
2620 st->frag_data = NULL;
2621}
2622EXPORT_SYMBOL(skb_prepare_seq_read);
2623
2624/**
2625 * skb_seq_read - Sequentially read skb data
2626 * @consumed: number of bytes consumed by the caller so far
2627 * @data: destination pointer for data to be returned
2628 * @st: state variable
2629 *
2630 * Reads a block of skb data at @consumed relative to the
2631 * lower offset specified to skb_prepare_seq_read(). Assigns
2632 * the head of the data block to @data and returns the length
2633 * of the block or 0 if the end of the skb data or the upper
2634 * offset has been reached.
2635 *
2636 * The caller is not required to consume all of the data
2637 * returned, i.e. @consumed is typically set to the number
2638 * of bytes already consumed and the next call to
2639 * skb_seq_read() will return the remaining part of the block.
2640 *
2641 * Note 1: The size of each block of data returned can be arbitrary,
2642 * this limitation is the cost for zerocopy seqeuental
2643 * reads of potentially non linear data.
2644 *
2645 * Note 2: Fragment lists within fragments are not implemented
2646 * at the moment, state->root_skb could be replaced with
2647 * a stack for this purpose.
2648 */
2649unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2650 struct skb_seq_state *st)
2651{
2652 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2653 skb_frag_t *frag;
2654
2655 if (unlikely(abs_offset >= st->upper_offset)) {
2656 if (st->frag_data) {
2657 kunmap_atomic(st->frag_data);
2658 st->frag_data = NULL;
2659 }
2660 return 0;
2661 }
2662
2663next_skb:
2664 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2665
2666 if (abs_offset < block_limit && !st->frag_data) {
2667 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2668 return block_limit - abs_offset;
2669 }
2670
2671 if (st->frag_idx == 0 && !st->frag_data)
2672 st->stepped_offset += skb_headlen(st->cur_skb);
2673
2674 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2675 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2676 block_limit = skb_frag_size(frag) + st->stepped_offset;
2677
2678 if (abs_offset < block_limit) {
2679 if (!st->frag_data)
2680 st->frag_data = kmap_atomic(skb_frag_page(frag));
2681
2682 *data = (u8 *) st->frag_data + frag->page_offset +
2683 (abs_offset - st->stepped_offset);
2684
2685 return block_limit - abs_offset;
2686 }
2687
2688 if (st->frag_data) {
2689 kunmap_atomic(st->frag_data);
2690 st->frag_data = NULL;
2691 }
2692
2693 st->frag_idx++;
2694 st->stepped_offset += skb_frag_size(frag);
2695 }
2696
2697 if (st->frag_data) {
2698 kunmap_atomic(st->frag_data);
2699 st->frag_data = NULL;
2700 }
2701
2702 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2703 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2704 st->frag_idx = 0;
2705 goto next_skb;
2706 } else if (st->cur_skb->next) {
2707 st->cur_skb = st->cur_skb->next;
2708 st->frag_idx = 0;
2709 goto next_skb;
2710 }
2711
2712 return 0;
2713}
2714EXPORT_SYMBOL(skb_seq_read);
2715
2716/**
2717 * skb_abort_seq_read - Abort a sequential read of skb data
2718 * @st: state variable
2719 *
2720 * Must be called if skb_seq_read() was not called until it
2721 * returned 0.
2722 */
2723void skb_abort_seq_read(struct skb_seq_state *st)
2724{
2725 if (st->frag_data)
2726 kunmap_atomic(st->frag_data);
2727}
2728EXPORT_SYMBOL(skb_abort_seq_read);
2729
2730#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2731
2732static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2733 struct ts_config *conf,
2734 struct ts_state *state)
2735{
2736 return skb_seq_read(offset, text, TS_SKB_CB(state));
2737}
2738
2739static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2740{
2741 skb_abort_seq_read(TS_SKB_CB(state));
2742}
2743
2744/**
2745 * skb_find_text - Find a text pattern in skb data
2746 * @skb: the buffer to look in
2747 * @from: search offset
2748 * @to: search limit
2749 * @config: textsearch configuration
2750 * @state: uninitialized textsearch state variable
2751 *
2752 * Finds a pattern in the skb data according to the specified
2753 * textsearch configuration. Use textsearch_next() to retrieve
2754 * subsequent occurrences of the pattern. Returns the offset
2755 * to the first occurrence or UINT_MAX if no match was found.
2756 */
2757unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2758 unsigned int to, struct ts_config *config,
2759 struct ts_state *state)
2760{
2761 unsigned int ret;
2762
2763 config->get_next_block = skb_ts_get_next_block;
2764 config->finish = skb_ts_finish;
2765
2766 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2767
2768 ret = textsearch_find(config, state);
2769 return (ret <= to - from ? ret : UINT_MAX);
2770}
2771EXPORT_SYMBOL(skb_find_text);
2772
2773/**
2774 * skb_append_datato_frags - append the user data to a skb
2775 * @sk: sock structure
2776 * @skb: skb structure to be appened with user data.
2777 * @getfrag: call back function to be used for getting the user data
2778 * @from: pointer to user message iov
2779 * @length: length of the iov message
2780 *
2781 * Description: This procedure append the user data in the fragment part
2782 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2783 */
2784int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2785 int (*getfrag)(void *from, char *to, int offset,
2786 int len, int odd, struct sk_buff *skb),
2787 void *from, int length)
2788{
2789 int frg_cnt = skb_shinfo(skb)->nr_frags;
2790 int copy;
2791 int offset = 0;
2792 int ret;
2793 struct page_frag *pfrag = ¤t->task_frag;
2794
2795 do {
2796 /* Return error if we don't have space for new frag */
2797 if (frg_cnt >= MAX_SKB_FRAGS)
2798 return -EMSGSIZE;
2799
2800 if (!sk_page_frag_refill(sk, pfrag))
2801 return -ENOMEM;
2802
2803 /* copy the user data to page */
2804 copy = min_t(int, length, pfrag->size - pfrag->offset);
2805
2806 ret = getfrag(from, page_address(pfrag->page) + pfrag->offset,
2807 offset, copy, 0, skb);
2808 if (ret < 0)
2809 return -EFAULT;
2810
2811 /* copy was successful so update the size parameters */
2812 skb_fill_page_desc(skb, frg_cnt, pfrag->page, pfrag->offset,
2813 copy);
2814 frg_cnt++;
2815 pfrag->offset += copy;
2816 get_page(pfrag->page);
2817
2818 skb->truesize += copy;
2819 atomic_add(copy, &sk->sk_wmem_alloc);
2820 skb->len += copy;
2821 skb->data_len += copy;
2822 offset += copy;
2823 length -= copy;
2824
2825 } while (length > 0);
2826
2827 return 0;
2828}
2829EXPORT_SYMBOL(skb_append_datato_frags);
2830
2831/**
2832 * skb_pull_rcsum - pull skb and update receive checksum
2833 * @skb: buffer to update
2834 * @len: length of data pulled
2835 *
2836 * This function performs an skb_pull on the packet and updates
2837 * the CHECKSUM_COMPLETE checksum. It should be used on
2838 * receive path processing instead of skb_pull unless you know
2839 * that the checksum difference is zero (e.g., a valid IP header)
2840 * or you are setting ip_summed to CHECKSUM_NONE.
2841 */
2842unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2843{
2844 BUG_ON(len > skb->len);
2845 skb->len -= len;
2846 BUG_ON(skb->len < skb->data_len);
2847 skb_postpull_rcsum(skb, skb->data, len);
2848 return skb->data += len;
2849}
2850EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2851
2852/**
2853 * skb_segment - Perform protocol segmentation on skb.
2854 * @head_skb: buffer to segment
2855 * @features: features for the output path (see dev->features)
2856 *
2857 * This function performs segmentation on the given skb. It returns
2858 * a pointer to the first in a list of new skbs for the segments.
2859 * In case of error it returns ERR_PTR(err).
2860 */
2861struct sk_buff *skb_segment(struct sk_buff *head_skb,
2862 netdev_features_t features)
2863{
2864 struct sk_buff *segs = NULL;
2865 struct sk_buff *tail = NULL;
2866 struct sk_buff *list_skb = skb_shinfo(head_skb)->frag_list;
2867 skb_frag_t *frag = skb_shinfo(head_skb)->frags;
2868 unsigned int mss = skb_shinfo(head_skb)->gso_size;
2869 unsigned int doffset = head_skb->data - skb_mac_header(head_skb);
2870 struct sk_buff *frag_skb = head_skb;
2871 unsigned int offset = doffset;
2872 unsigned int tnl_hlen = skb_tnl_header_len(head_skb);
2873 unsigned int headroom;
2874 unsigned int len;
2875 __be16 proto;
2876 bool csum;
2877 int sg = !!(features & NETIF_F_SG);
2878 int nfrags = skb_shinfo(head_skb)->nr_frags;
2879 int err = -ENOMEM;
2880 int i = 0;
2881 int pos;
2882 int dummy;
2883
2884 proto = skb_network_protocol(head_skb, &dummy);
2885 if (unlikely(!proto))
2886 return ERR_PTR(-EINVAL);
2887
2888 csum = !!can_checksum_protocol(features, proto);
2889 __skb_push(head_skb, doffset);
2890 headroom = skb_headroom(head_skb);
2891 pos = skb_headlen(head_skb);
2892
2893 do {
2894 struct sk_buff *nskb;
2895 skb_frag_t *nskb_frag;
2896 int hsize;
2897 int size;
2898
2899 len = head_skb->len - offset;
2900 if (len > mss)
2901 len = mss;
2902
2903 hsize = skb_headlen(head_skb) - offset;
2904 if (hsize < 0)
2905 hsize = 0;
2906 if (hsize > len || !sg)
2907 hsize = len;
2908
2909 if (!hsize && i >= nfrags && skb_headlen(list_skb) &&
2910 (skb_headlen(list_skb) == len || sg)) {
2911 BUG_ON(skb_headlen(list_skb) > len);
2912
2913 i = 0;
2914 nfrags = skb_shinfo(list_skb)->nr_frags;
2915 frag = skb_shinfo(list_skb)->frags;
2916 frag_skb = list_skb;
2917 pos += skb_headlen(list_skb);
2918
2919 while (pos < offset + len) {
2920 BUG_ON(i >= nfrags);
2921
2922 size = skb_frag_size(frag);
2923 if (pos + size > offset + len)
2924 break;
2925
2926 i++;
2927 pos += size;
2928 frag++;
2929 }
2930
2931 nskb = skb_clone(list_skb, GFP_ATOMIC);
2932 list_skb = list_skb->next;
2933
2934 if (unlikely(!nskb))
2935 goto err;
2936
2937 if (unlikely(pskb_trim(nskb, len))) {
2938 kfree_skb(nskb);
2939 goto err;
2940 }
2941
2942 hsize = skb_end_offset(nskb);
2943 if (skb_cow_head(nskb, doffset + headroom)) {
2944 kfree_skb(nskb);
2945 goto err;
2946 }
2947
2948 nskb->truesize += skb_end_offset(nskb) - hsize;
2949 skb_release_head_state(nskb);
2950 __skb_push(nskb, doffset);
2951 } else {
2952 nskb = __alloc_skb(hsize + doffset + headroom,
2953 GFP_ATOMIC, skb_alloc_rx_flag(head_skb),
2954 NUMA_NO_NODE);
2955
2956 if (unlikely(!nskb))
2957 goto err;
2958
2959 skb_reserve(nskb, headroom);
2960 __skb_put(nskb, doffset);
2961 }
2962
2963 if (segs)
2964 tail->next = nskb;
2965 else
2966 segs = nskb;
2967 tail = nskb;
2968
2969 __copy_skb_header(nskb, head_skb);
2970 nskb->mac_len = head_skb->mac_len;
2971
2972 skb_headers_offset_update(nskb, skb_headroom(nskb) - headroom);
2973
2974 skb_copy_from_linear_data_offset(head_skb, -tnl_hlen,
2975 nskb->data - tnl_hlen,
2976 doffset + tnl_hlen);
2977
2978 if (nskb->len == len + doffset)
2979 goto perform_csum_check;
2980
2981 if (!sg) {
2982 nskb->ip_summed = CHECKSUM_NONE;
2983 nskb->csum = skb_copy_and_csum_bits(head_skb, offset,
2984 skb_put(nskb, len),
2985 len, 0);
2986 continue;
2987 }
2988
2989 nskb_frag = skb_shinfo(nskb)->frags;
2990
2991 skb_copy_from_linear_data_offset(head_skb, offset,
2992 skb_put(nskb, hsize), hsize);
2993
2994 skb_shinfo(nskb)->tx_flags = skb_shinfo(head_skb)->tx_flags &
2995 SKBTX_SHARED_FRAG;
2996
2997 while (pos < offset + len) {
2998 if (i >= nfrags) {
2999 BUG_ON(skb_headlen(list_skb));
3000
3001 i = 0;
3002 nfrags = skb_shinfo(list_skb)->nr_frags;
3003 frag = skb_shinfo(list_skb)->frags;
3004 frag_skb = list_skb;
3005
3006 BUG_ON(!nfrags);
3007
3008 list_skb = list_skb->next;
3009 }
3010
3011 if (unlikely(skb_shinfo(nskb)->nr_frags >=
3012 MAX_SKB_FRAGS)) {
3013 net_warn_ratelimited(
3014 "skb_segment: too many frags: %u %u\n",
3015 pos, mss);
3016 goto err;
3017 }
3018
3019 if (unlikely(skb_orphan_frags(frag_skb, GFP_ATOMIC)))
3020 goto err;
3021
3022 *nskb_frag = *frag;
3023 __skb_frag_ref(nskb_frag);
3024 size = skb_frag_size(nskb_frag);
3025
3026 if (pos < offset) {
3027 nskb_frag->page_offset += offset - pos;
3028 skb_frag_size_sub(nskb_frag, offset - pos);
3029 }
3030
3031 skb_shinfo(nskb)->nr_frags++;
3032
3033 if (pos + size <= offset + len) {
3034 i++;
3035 frag++;
3036 pos += size;
3037 } else {
3038 skb_frag_size_sub(nskb_frag, pos + size - (offset + len));
3039 goto skip_fraglist;
3040 }
3041
3042 nskb_frag++;
3043 }
3044
3045skip_fraglist:
3046 nskb->data_len = len - hsize;
3047 nskb->len += nskb->data_len;
3048 nskb->truesize += nskb->data_len;
3049
3050perform_csum_check:
3051 if (!csum) {
3052 nskb->csum = skb_checksum(nskb, doffset,
3053 nskb->len - doffset, 0);
3054 nskb->ip_summed = CHECKSUM_NONE;
3055 }
3056 } while ((offset += len) < head_skb->len);
3057
3058 return segs;
3059
3060err:
3061 kfree_skb_list(segs);
3062 return ERR_PTR(err);
3063}
3064EXPORT_SYMBOL_GPL(skb_segment);
3065
3066int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
3067{
3068 struct skb_shared_info *pinfo, *skbinfo = skb_shinfo(skb);
3069 unsigned int offset = skb_gro_offset(skb);
3070 unsigned int headlen = skb_headlen(skb);
3071 struct sk_buff *nskb, *lp, *p = *head;
3072 unsigned int len = skb_gro_len(skb);
3073 unsigned int delta_truesize;
3074 unsigned int headroom;
3075
3076 if (unlikely(p->len + len >= 65536))
3077 return -E2BIG;
3078
3079 lp = NAPI_GRO_CB(p)->last;
3080 pinfo = skb_shinfo(lp);
3081
3082 if (headlen <= offset) {
3083 skb_frag_t *frag;
3084 skb_frag_t *frag2;
3085 int i = skbinfo->nr_frags;
3086 int nr_frags = pinfo->nr_frags + i;
3087
3088 if (nr_frags > MAX_SKB_FRAGS)
3089 goto merge;
3090
3091 offset -= headlen;
3092 pinfo->nr_frags = nr_frags;
3093 skbinfo->nr_frags = 0;
3094
3095 frag = pinfo->frags + nr_frags;
3096 frag2 = skbinfo->frags + i;
3097 do {
3098 *--frag = *--frag2;
3099 } while (--i);
3100
3101 frag->page_offset += offset;
3102 skb_frag_size_sub(frag, offset);
3103
3104 /* all fragments truesize : remove (head size + sk_buff) */
3105 delta_truesize = skb->truesize -
3106 SKB_TRUESIZE(skb_end_offset(skb));
3107
3108 skb->truesize -= skb->data_len;
3109 skb->len -= skb->data_len;
3110 skb->data_len = 0;
3111
3112 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3113 goto done;
3114 } else if (skb->head_frag) {
3115 int nr_frags = pinfo->nr_frags;
3116 skb_frag_t *frag = pinfo->frags + nr_frags;
3117 struct page *page = virt_to_head_page(skb->head);
3118 unsigned int first_size = headlen - offset;
3119 unsigned int first_offset;
3120
3121 if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3122 goto merge;
3123
3124 first_offset = skb->data -
3125 (unsigned char *)page_address(page) +
3126 offset;
3127
3128 pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3129
3130 frag->page.p = page;
3131 frag->page_offset = first_offset;
3132 skb_frag_size_set(frag, first_size);
3133
3134 memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3135 /* We dont need to clear skbinfo->nr_frags here */
3136
3137 delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3138 NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3139 goto done;
3140 }
3141 if (pinfo->frag_list)
3142 goto merge;
3143 if (skb_gro_len(p) != pinfo->gso_size)
3144 return -E2BIG;
3145
3146 headroom = skb_headroom(p);
3147 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3148 if (unlikely(!nskb))
3149 return -ENOMEM;
3150
3151 __copy_skb_header(nskb, p);
3152 nskb->mac_len = p->mac_len;
3153
3154 skb_reserve(nskb, headroom);
3155 __skb_put(nskb, skb_gro_offset(p));
3156
3157 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3158 skb_set_network_header(nskb, skb_network_offset(p));
3159 skb_set_transport_header(nskb, skb_transport_offset(p));
3160
3161 __skb_pull(p, skb_gro_offset(p));
3162 memcpy(skb_mac_header(nskb), skb_mac_header(p),
3163 p->data - skb_mac_header(p));
3164
3165 skb_shinfo(nskb)->frag_list = p;
3166 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3167 pinfo->gso_size = 0;
3168 skb_header_release(p);
3169 NAPI_GRO_CB(nskb)->last = p;
3170
3171 nskb->data_len += p->len;
3172 nskb->truesize += p->truesize;
3173 nskb->len += p->len;
3174
3175 *head = nskb;
3176 nskb->next = p->next;
3177 p->next = NULL;
3178
3179 p = nskb;
3180
3181merge:
3182 delta_truesize = skb->truesize;
3183 if (offset > headlen) {
3184 unsigned int eat = offset - headlen;
3185
3186 skbinfo->frags[0].page_offset += eat;
3187 skb_frag_size_sub(&skbinfo->frags[0], eat);
3188 skb->data_len -= eat;
3189 skb->len -= eat;
3190 offset = headlen;
3191 }
3192
3193 __skb_pull(skb, offset);
3194
3195 if (NAPI_GRO_CB(p)->last == p)
3196 skb_shinfo(p)->frag_list = skb;
3197 else
3198 NAPI_GRO_CB(p)->last->next = skb;
3199 NAPI_GRO_CB(p)->last = skb;
3200 skb_header_release(skb);
3201 lp = p;
3202
3203done:
3204 NAPI_GRO_CB(p)->count++;
3205 p->data_len += len;
3206 p->truesize += delta_truesize;
3207 p->len += len;
3208 if (lp != p) {
3209 lp->data_len += len;
3210 lp->truesize += delta_truesize;
3211 lp->len += len;
3212 }
3213 NAPI_GRO_CB(skb)->same_flow = 1;
3214 return 0;
3215}
3216EXPORT_SYMBOL_GPL(skb_gro_receive);
3217
3218void __init skb_init(void)
3219{
3220 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3221 sizeof(struct sk_buff),
3222 0,
3223 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3224 NULL);
3225 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3226 (2*sizeof(struct sk_buff)) +
3227 sizeof(atomic_t),
3228 0,
3229 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3230 NULL);
3231}
3232
3233/**
3234 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3235 * @skb: Socket buffer containing the buffers to be mapped
3236 * @sg: The scatter-gather list to map into
3237 * @offset: The offset into the buffer's contents to start mapping
3238 * @len: Length of buffer space to be mapped
3239 *
3240 * Fill the specified scatter-gather list with mappings/pointers into a
3241 * region of the buffer space attached to a socket buffer.
3242 */
3243static int
3244__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3245{
3246 int start = skb_headlen(skb);
3247 int i, copy = start - offset;
3248 struct sk_buff *frag_iter;
3249 int elt = 0;
3250
3251 if (copy > 0) {
3252 if (copy > len)
3253 copy = len;
3254 sg_set_buf(sg, skb->data + offset, copy);
3255 elt++;
3256 if ((len -= copy) == 0)
3257 return elt;
3258 offset += copy;
3259 }
3260
3261 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3262 int end;
3263
3264 WARN_ON(start > offset + len);
3265
3266 end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3267 if ((copy = end - offset) > 0) {
3268 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3269
3270 if (copy > len)
3271 copy = len;
3272 sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3273 frag->page_offset+offset-start);
3274 elt++;
3275 if (!(len -= copy))
3276 return elt;
3277 offset += copy;
3278 }
3279 start = end;
3280 }
3281
3282 skb_walk_frags(skb, frag_iter) {
3283 int end;
3284
3285 WARN_ON(start > offset + len);
3286
3287 end = start + frag_iter->len;
3288 if ((copy = end - offset) > 0) {
3289 if (copy > len)
3290 copy = len;
3291 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3292 copy);
3293 if ((len -= copy) == 0)
3294 return elt;
3295 offset += copy;
3296 }
3297 start = end;
3298 }
3299 BUG_ON(len);
3300 return elt;
3301}
3302
3303/* As compared with skb_to_sgvec, skb_to_sgvec_nomark only map skb to given
3304 * sglist without mark the sg which contain last skb data as the end.
3305 * So the caller can mannipulate sg list as will when padding new data after
3306 * the first call without calling sg_unmark_end to expend sg list.
3307 *
3308 * Scenario to use skb_to_sgvec_nomark:
3309 * 1. sg_init_table
3310 * 2. skb_to_sgvec_nomark(payload1)
3311 * 3. skb_to_sgvec_nomark(payload2)
3312 *
3313 * This is equivalent to:
3314 * 1. sg_init_table
3315 * 2. skb_to_sgvec(payload1)
3316 * 3. sg_unmark_end
3317 * 4. skb_to_sgvec(payload2)
3318 *
3319 * When mapping mutilple payload conditionally, skb_to_sgvec_nomark
3320 * is more preferable.
3321 */
3322int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
3323 int offset, int len)
3324{
3325 return __skb_to_sgvec(skb, sg, offset, len);
3326}
3327EXPORT_SYMBOL_GPL(skb_to_sgvec_nomark);
3328
3329int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3330{
3331 int nsg = __skb_to_sgvec(skb, sg, offset, len);
3332
3333 sg_mark_end(&sg[nsg - 1]);
3334
3335 return nsg;
3336}
3337EXPORT_SYMBOL_GPL(skb_to_sgvec);
3338
3339/**
3340 * skb_cow_data - Check that a socket buffer's data buffers are writable
3341 * @skb: The socket buffer to check.
3342 * @tailbits: Amount of trailing space to be added
3343 * @trailer: Returned pointer to the skb where the @tailbits space begins
3344 *
3345 * Make sure that the data buffers attached to a socket buffer are
3346 * writable. If they are not, private copies are made of the data buffers
3347 * and the socket buffer is set to use these instead.
3348 *
3349 * If @tailbits is given, make sure that there is space to write @tailbits
3350 * bytes of data beyond current end of socket buffer. @trailer will be
3351 * set to point to the skb in which this space begins.
3352 *
3353 * The number of scatterlist elements required to completely map the
3354 * COW'd and extended socket buffer will be returned.
3355 */
3356int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3357{
3358 int copyflag;
3359 int elt;
3360 struct sk_buff *skb1, **skb_p;
3361
3362 /* If skb is cloned or its head is paged, reallocate
3363 * head pulling out all the pages (pages are considered not writable
3364 * at the moment even if they are anonymous).
3365 */
3366 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3367 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3368 return -ENOMEM;
3369
3370 /* Easy case. Most of packets will go this way. */
3371 if (!skb_has_frag_list(skb)) {
3372 /* A little of trouble, not enough of space for trailer.
3373 * This should not happen, when stack is tuned to generate
3374 * good frames. OK, on miss we reallocate and reserve even more
3375 * space, 128 bytes is fair. */
3376
3377 if (skb_tailroom(skb) < tailbits &&
3378 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3379 return -ENOMEM;
3380
3381 /* Voila! */
3382 *trailer = skb;
3383 return 1;
3384 }
3385
3386 /* Misery. We are in troubles, going to mincer fragments... */
3387
3388 elt = 1;
3389 skb_p = &skb_shinfo(skb)->frag_list;
3390 copyflag = 0;
3391
3392 while ((skb1 = *skb_p) != NULL) {
3393 int ntail = 0;
3394
3395 /* The fragment is partially pulled by someone,
3396 * this can happen on input. Copy it and everything
3397 * after it. */
3398
3399 if (skb_shared(skb1))
3400 copyflag = 1;
3401
3402 /* If the skb is the last, worry about trailer. */
3403
3404 if (skb1->next == NULL && tailbits) {
3405 if (skb_shinfo(skb1)->nr_frags ||
3406 skb_has_frag_list(skb1) ||
3407 skb_tailroom(skb1) < tailbits)
3408 ntail = tailbits + 128;
3409 }
3410
3411 if (copyflag ||
3412 skb_cloned(skb1) ||
3413 ntail ||
3414 skb_shinfo(skb1)->nr_frags ||
3415 skb_has_frag_list(skb1)) {
3416 struct sk_buff *skb2;
3417
3418 /* Fuck, we are miserable poor guys... */
3419 if (ntail == 0)
3420 skb2 = skb_copy(skb1, GFP_ATOMIC);
3421 else
3422 skb2 = skb_copy_expand(skb1,
3423 skb_headroom(skb1),
3424 ntail,
3425 GFP_ATOMIC);
3426 if (unlikely(skb2 == NULL))
3427 return -ENOMEM;
3428
3429 if (skb1->sk)
3430 skb_set_owner_w(skb2, skb1->sk);
3431
3432 /* Looking around. Are we still alive?
3433 * OK, link new skb, drop old one */
3434
3435 skb2->next = skb1->next;
3436 *skb_p = skb2;
3437 kfree_skb(skb1);
3438 skb1 = skb2;
3439 }
3440 elt++;
3441 *trailer = skb1;
3442 skb_p = &skb1->next;
3443 }
3444
3445 return elt;
3446}
3447EXPORT_SYMBOL_GPL(skb_cow_data);
3448
3449static void sock_rmem_free(struct sk_buff *skb)
3450{
3451 struct sock *sk = skb->sk;
3452
3453 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3454}
3455
3456/*
3457 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3458 */
3459int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3460{
3461 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3462 (unsigned int)sk->sk_rcvbuf)
3463 return -ENOMEM;
3464
3465 skb_orphan(skb);
3466 skb->sk = sk;
3467 skb->destructor = sock_rmem_free;
3468 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3469
3470 /* before exiting rcu section, make sure dst is refcounted */
3471 skb_dst_force(skb);
3472
3473 skb_queue_tail(&sk->sk_error_queue, skb);
3474 if (!sock_flag(sk, SOCK_DEAD))
3475 sk->sk_data_ready(sk);
3476 return 0;
3477}
3478EXPORT_SYMBOL(sock_queue_err_skb);
3479
3480void skb_tstamp_tx(struct sk_buff *orig_skb,
3481 struct skb_shared_hwtstamps *hwtstamps)
3482{
3483 struct sock *sk = orig_skb->sk;
3484 struct sock_exterr_skb *serr;
3485 struct sk_buff *skb;
3486 int err;
3487
3488 if (!sk)
3489 return;
3490
3491 if (hwtstamps) {
3492 *skb_hwtstamps(orig_skb) =
3493 *hwtstamps;
3494 } else {
3495 /*
3496 * no hardware time stamps available,
3497 * so keep the shared tx_flags and only
3498 * store software time stamp
3499 */
3500 orig_skb->tstamp = ktime_get_real();
3501 }
3502
3503 skb = skb_clone(orig_skb, GFP_ATOMIC);
3504 if (!skb)
3505 return;
3506
3507 serr = SKB_EXT_ERR(skb);
3508 memset(serr, 0, sizeof(*serr));
3509 serr->ee.ee_errno = ENOMSG;
3510 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3511
3512 err = sock_queue_err_skb(sk, skb);
3513
3514 if (err)
3515 kfree_skb(skb);
3516}
3517EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3518
3519void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3520{
3521 struct sock *sk = skb->sk;
3522 struct sock_exterr_skb *serr;
3523 int err;
3524
3525 skb->wifi_acked_valid = 1;
3526 skb->wifi_acked = acked;
3527
3528 serr = SKB_EXT_ERR(skb);
3529 memset(serr, 0, sizeof(*serr));
3530 serr->ee.ee_errno = ENOMSG;
3531 serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3532
3533 err = sock_queue_err_skb(sk, skb);
3534 if (err)
3535 kfree_skb(skb);
3536}
3537EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3538
3539
3540/**
3541 * skb_partial_csum_set - set up and verify partial csum values for packet
3542 * @skb: the skb to set
3543 * @start: the number of bytes after skb->data to start checksumming.
3544 * @off: the offset from start to place the checksum.
3545 *
3546 * For untrusted partially-checksummed packets, we need to make sure the values
3547 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3548 *
3549 * This function checks and sets those values and skb->ip_summed: if this
3550 * returns false you should drop the packet.
3551 */
3552bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3553{
3554 if (unlikely(start > skb_headlen(skb)) ||
3555 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3556 net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3557 start, off, skb_headlen(skb));
3558 return false;
3559 }
3560 skb->ip_summed = CHECKSUM_PARTIAL;
3561 skb->csum_start = skb_headroom(skb) + start;
3562 skb->csum_offset = off;
3563 skb_set_transport_header(skb, start);
3564 return true;
3565}
3566EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3567
3568static int skb_maybe_pull_tail(struct sk_buff *skb, unsigned int len,
3569 unsigned int max)
3570{
3571 if (skb_headlen(skb) >= len)
3572 return 0;
3573
3574 /* If we need to pullup then pullup to the max, so we
3575 * won't need to do it again.
3576 */
3577 if (max > skb->len)
3578 max = skb->len;
3579
3580 if (__pskb_pull_tail(skb, max - skb_headlen(skb)) == NULL)
3581 return -ENOMEM;
3582
3583 if (skb_headlen(skb) < len)
3584 return -EPROTO;
3585
3586 return 0;
3587}
3588
3589#define MAX_TCP_HDR_LEN (15 * 4)
3590
3591static __sum16 *skb_checksum_setup_ip(struct sk_buff *skb,
3592 typeof(IPPROTO_IP) proto,
3593 unsigned int off)
3594{
3595 switch (proto) {
3596 int err;
3597
3598 case IPPROTO_TCP:
3599 err = skb_maybe_pull_tail(skb, off + sizeof(struct tcphdr),
3600 off + MAX_TCP_HDR_LEN);
3601 if (!err && !skb_partial_csum_set(skb, off,
3602 offsetof(struct tcphdr,
3603 check)))
3604 err = -EPROTO;
3605 return err ? ERR_PTR(err) : &tcp_hdr(skb)->check;
3606
3607 case IPPROTO_UDP:
3608 err = skb_maybe_pull_tail(skb, off + sizeof(struct udphdr),
3609 off + sizeof(struct udphdr));
3610 if (!err && !skb_partial_csum_set(skb, off,
3611 offsetof(struct udphdr,
3612 check)))
3613 err = -EPROTO;
3614 return err ? ERR_PTR(err) : &udp_hdr(skb)->check;
3615 }
3616
3617 return ERR_PTR(-EPROTO);
3618}
3619
3620/* This value should be large enough to cover a tagged ethernet header plus
3621 * maximally sized IP and TCP or UDP headers.
3622 */
3623#define MAX_IP_HDR_LEN 128
3624
3625static int skb_checksum_setup_ipv4(struct sk_buff *skb, bool recalculate)
3626{
3627 unsigned int off;
3628 bool fragment;
3629 __sum16 *csum;
3630 int err;
3631
3632 fragment = false;
3633
3634 err = skb_maybe_pull_tail(skb,
3635 sizeof(struct iphdr),
3636 MAX_IP_HDR_LEN);
3637 if (err < 0)
3638 goto out;
3639
3640 if (ip_hdr(skb)->frag_off & htons(IP_OFFSET | IP_MF))
3641 fragment = true;
3642
3643 off = ip_hdrlen(skb);
3644
3645 err = -EPROTO;
3646
3647 if (fragment)
3648 goto out;
3649
3650 csum = skb_checksum_setup_ip(skb, ip_hdr(skb)->protocol, off);
3651 if (IS_ERR(csum))
3652 return PTR_ERR(csum);
3653
3654 if (recalculate)
3655 *csum = ~csum_tcpudp_magic(ip_hdr(skb)->saddr,
3656 ip_hdr(skb)->daddr,
3657 skb->len - off,
3658 ip_hdr(skb)->protocol, 0);
3659 err = 0;
3660
3661out:
3662 return err;
3663}
3664
3665/* This value should be large enough to cover a tagged ethernet header plus
3666 * an IPv6 header, all options, and a maximal TCP or UDP header.
3667 */
3668#define MAX_IPV6_HDR_LEN 256
3669
3670#define OPT_HDR(type, skb, off) \
3671 (type *)(skb_network_header(skb) + (off))
3672
3673static int skb_checksum_setup_ipv6(struct sk_buff *skb, bool recalculate)
3674{
3675 int err;
3676 u8 nexthdr;
3677 unsigned int off;
3678 unsigned int len;
3679 bool fragment;
3680 bool done;
3681 __sum16 *csum;
3682
3683 fragment = false;
3684 done = false;
3685
3686 off = sizeof(struct ipv6hdr);
3687
3688 err = skb_maybe_pull_tail(skb, off, MAX_IPV6_HDR_LEN);
3689 if (err < 0)
3690 goto out;
3691
3692 nexthdr = ipv6_hdr(skb)->nexthdr;
3693
3694 len = sizeof(struct ipv6hdr) + ntohs(ipv6_hdr(skb)->payload_len);
3695 while (off <= len && !done) {
3696 switch (nexthdr) {
3697 case IPPROTO_DSTOPTS:
3698 case IPPROTO_HOPOPTS:
3699 case IPPROTO_ROUTING: {
3700 struct ipv6_opt_hdr *hp;
3701
3702 err = skb_maybe_pull_tail(skb,
3703 off +
3704 sizeof(struct ipv6_opt_hdr),
3705 MAX_IPV6_HDR_LEN);
3706 if (err < 0)
3707 goto out;
3708
3709 hp = OPT_HDR(struct ipv6_opt_hdr, skb, off);
3710 nexthdr = hp->nexthdr;
3711 off += ipv6_optlen(hp);
3712 break;
3713 }
3714 case IPPROTO_AH: {
3715 struct ip_auth_hdr *hp;
3716
3717 err = skb_maybe_pull_tail(skb,
3718 off +
3719 sizeof(struct ip_auth_hdr),
3720 MAX_IPV6_HDR_LEN);
3721 if (err < 0)
3722 goto out;
3723
3724 hp = OPT_HDR(struct ip_auth_hdr, skb, off);
3725 nexthdr = hp->nexthdr;
3726 off += ipv6_authlen(hp);
3727 break;
3728 }
3729 case IPPROTO_FRAGMENT: {
3730 struct frag_hdr *hp;
3731
3732 err = skb_maybe_pull_tail(skb,
3733 off +
3734 sizeof(struct frag_hdr),
3735 MAX_IPV6_HDR_LEN);
3736 if (err < 0)
3737 goto out;
3738
3739 hp = OPT_HDR(struct frag_hdr, skb, off);
3740
3741 if (hp->frag_off & htons(IP6_OFFSET | IP6_MF))
3742 fragment = true;
3743
3744 nexthdr = hp->nexthdr;
3745 off += sizeof(struct frag_hdr);
3746 break;
3747 }
3748 default:
3749 done = true;
3750 break;
3751 }
3752 }
3753
3754 err = -EPROTO;
3755
3756 if (!done || fragment)
3757 goto out;
3758
3759 csum = skb_checksum_setup_ip(skb, nexthdr, off);
3760 if (IS_ERR(csum))
3761 return PTR_ERR(csum);
3762
3763 if (recalculate)
3764 *csum = ~csum_ipv6_magic(&ipv6_hdr(skb)->saddr,
3765 &ipv6_hdr(skb)->daddr,
3766 skb->len - off, nexthdr, 0);
3767 err = 0;
3768
3769out:
3770 return err;
3771}
3772
3773/**
3774 * skb_checksum_setup - set up partial checksum offset
3775 * @skb: the skb to set up
3776 * @recalculate: if true the pseudo-header checksum will be recalculated
3777 */
3778int skb_checksum_setup(struct sk_buff *skb, bool recalculate)
3779{
3780 int err;
3781
3782 switch (skb->protocol) {
3783 case htons(ETH_P_IP):
3784 err = skb_checksum_setup_ipv4(skb, recalculate);
3785 break;
3786
3787 case htons(ETH_P_IPV6):
3788 err = skb_checksum_setup_ipv6(skb, recalculate);
3789 break;
3790
3791 default:
3792 err = -EPROTO;
3793 break;
3794 }
3795
3796 return err;
3797}
3798EXPORT_SYMBOL(skb_checksum_setup);
3799
3800void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3801{
3802 net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3803 skb->dev->name);
3804}
3805EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3806
3807void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3808{
3809 if (head_stolen) {
3810 skb_release_head_state(skb);
3811 kmem_cache_free(skbuff_head_cache, skb);
3812 } else {
3813 __kfree_skb(skb);
3814 }
3815}
3816EXPORT_SYMBOL(kfree_skb_partial);
3817
3818/**
3819 * skb_try_coalesce - try to merge skb to prior one
3820 * @to: prior buffer
3821 * @from: buffer to add
3822 * @fragstolen: pointer to boolean
3823 * @delta_truesize: how much more was allocated than was requested
3824 */
3825bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3826 bool *fragstolen, int *delta_truesize)
3827{
3828 int i, delta, len = from->len;
3829
3830 *fragstolen = false;
3831
3832 if (skb_cloned(to))
3833 return false;
3834
3835 if (len <= skb_tailroom(to)) {
3836 BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3837 *delta_truesize = 0;
3838 return true;
3839 }
3840
3841 if (skb_has_frag_list(to) || skb_has_frag_list(from))
3842 return false;
3843
3844 if (skb_headlen(from) != 0) {
3845 struct page *page;
3846 unsigned int offset;
3847
3848 if (skb_shinfo(to)->nr_frags +
3849 skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3850 return false;
3851
3852 if (skb_head_is_locked(from))
3853 return false;
3854
3855 delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3856
3857 page = virt_to_head_page(from->head);
3858 offset = from->data - (unsigned char *)page_address(page);
3859
3860 skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3861 page, offset, skb_headlen(from));
3862 *fragstolen = true;
3863 } else {
3864 if (skb_shinfo(to)->nr_frags +
3865 skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3866 return false;
3867
3868 delta = from->truesize - SKB_TRUESIZE(skb_end_offset(from));
3869 }
3870
3871 WARN_ON_ONCE(delta < len);
3872
3873 memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3874 skb_shinfo(from)->frags,
3875 skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3876 skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3877
3878 if (!skb_cloned(from))
3879 skb_shinfo(from)->nr_frags = 0;
3880
3881 /* if the skb is not cloned this does nothing
3882 * since we set nr_frags to 0.
3883 */
3884 for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3885 skb_frag_ref(from, i);
3886
3887 to->truesize += delta;
3888 to->len += len;
3889 to->data_len += len;
3890
3891 *delta_truesize = delta;
3892 return true;
3893}
3894EXPORT_SYMBOL(skb_try_coalesce);
3895
3896/**
3897 * skb_scrub_packet - scrub an skb
3898 *
3899 * @skb: buffer to clean
3900 * @xnet: packet is crossing netns
3901 *
3902 * skb_scrub_packet can be used after encapsulating or decapsulting a packet
3903 * into/from a tunnel. Some information have to be cleared during these
3904 * operations.
3905 * skb_scrub_packet can also be used to clean a skb before injecting it in
3906 * another namespace (@xnet == true). We have to clear all information in the
3907 * skb that could impact namespace isolation.
3908 */
3909void skb_scrub_packet(struct sk_buff *skb, bool xnet)
3910{
3911 if (xnet)
3912 skb_orphan(skb);
3913 skb->tstamp.tv64 = 0;
3914 skb->pkt_type = PACKET_HOST;
3915 skb->skb_iif = 0;
3916 skb->local_df = 0;
3917 skb_dst_drop(skb);
3918 skb->mark = 0;
3919 secpath_reset(skb);
3920 nf_reset(skb);
3921 nf_reset_trace(skb);
3922}
3923EXPORT_SYMBOL_GPL(skb_scrub_packet);
3924
3925/**
3926 * skb_gso_transport_seglen - Return length of individual segments of a gso packet
3927 *
3928 * @skb: GSO skb
3929 *
3930 * skb_gso_transport_seglen is used to determine the real size of the
3931 * individual segments, including Layer4 headers (TCP/UDP).
3932 *
3933 * The MAC/L2 or network (IP, IPv6) headers are not accounted for.
3934 */
3935unsigned int skb_gso_transport_seglen(const struct sk_buff *skb)
3936{
3937 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3938
3939 if (likely(shinfo->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6)))
3940 return tcp_hdrlen(skb) + shinfo->gso_size;
3941
3942 /* UFO sets gso_size to the size of the fragmentation
3943 * payload, i.e. the size of the L4 (UDP) header is already
3944 * accounted for.
3945 */
3946 return shinfo->gso_size;
3947}
3948EXPORT_SYMBOL_GPL(skb_gso_transport_seglen);
1/*
2 * Routines having to do with the 'struct sk_buff' memory handlers.
3 *
4 * Authors: Alan Cox <alan@lxorguk.ukuu.org.uk>
5 * Florian La Roche <rzsfl@rz.uni-sb.de>
6 *
7 * Fixes:
8 * Alan Cox : Fixed the worst of the load
9 * balancer bugs.
10 * Dave Platt : Interrupt stacking fix.
11 * Richard Kooijman : Timestamp fixes.
12 * Alan Cox : Changed buffer format.
13 * Alan Cox : destructor hook for AF_UNIX etc.
14 * Linus Torvalds : Better skb_clone.
15 * Alan Cox : Added skb_copy.
16 * Alan Cox : Added all the changed routines Linus
17 * only put in the headers
18 * Ray VanTassle : Fixed --skb->lock in free
19 * Alan Cox : skb_copy copy arp field
20 * Andi Kleen : slabified it.
21 * Robert Olsson : Removed skb_head_pool
22 *
23 * NOTE:
24 * The __skb_ routines should be called with interrupts
25 * disabled, or you better be *real* sure that the operation is atomic
26 * with respect to whatever list is being frobbed (e.g. via lock_sock()
27 * or via disabling bottom half handlers, etc).
28 *
29 * This program is free software; you can redistribute it and/or
30 * modify it under the terms of the GNU General Public License
31 * as published by the Free Software Foundation; either version
32 * 2 of the License, or (at your option) any later version.
33 */
34
35/*
36 * The functions in this file will not compile correctly with gcc 2.4.x
37 */
38
39#include <linux/module.h>
40#include <linux/types.h>
41#include <linux/kernel.h>
42#include <linux/kmemcheck.h>
43#include <linux/mm.h>
44#include <linux/interrupt.h>
45#include <linux/in.h>
46#include <linux/inet.h>
47#include <linux/slab.h>
48#include <linux/netdevice.h>
49#ifdef CONFIG_NET_CLS_ACT
50#include <net/pkt_sched.h>
51#endif
52#include <linux/string.h>
53#include <linux/skbuff.h>
54#include <linux/splice.h>
55#include <linux/cache.h>
56#include <linux/rtnetlink.h>
57#include <linux/init.h>
58#include <linux/scatterlist.h>
59#include <linux/errqueue.h>
60#include <linux/prefetch.h>
61
62#include <net/protocol.h>
63#include <net/dst.h>
64#include <net/sock.h>
65#include <net/checksum.h>
66#include <net/xfrm.h>
67
68#include <asm/uaccess.h>
69#include <asm/system.h>
70#include <trace/events/skb.h>
71
72#include "kmap_skb.h"
73
74static struct kmem_cache *skbuff_head_cache __read_mostly;
75static struct kmem_cache *skbuff_fclone_cache __read_mostly;
76
77static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
78 struct pipe_buffer *buf)
79{
80 put_page(buf->page);
81}
82
83static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
84 struct pipe_buffer *buf)
85{
86 get_page(buf->page);
87}
88
89static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
90 struct pipe_buffer *buf)
91{
92 return 1;
93}
94
95
96/* Pipe buffer operations for a socket. */
97static const struct pipe_buf_operations sock_pipe_buf_ops = {
98 .can_merge = 0,
99 .map = generic_pipe_buf_map,
100 .unmap = generic_pipe_buf_unmap,
101 .confirm = generic_pipe_buf_confirm,
102 .release = sock_pipe_buf_release,
103 .steal = sock_pipe_buf_steal,
104 .get = sock_pipe_buf_get,
105};
106
107/*
108 * Keep out-of-line to prevent kernel bloat.
109 * __builtin_return_address is not used because it is not always
110 * reliable.
111 */
112
113/**
114 * skb_over_panic - private function
115 * @skb: buffer
116 * @sz: size
117 * @here: address
118 *
119 * Out of line support code for skb_put(). Not user callable.
120 */
121static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
122{
123 printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
124 "data:%p tail:%#lx end:%#lx dev:%s\n",
125 here, skb->len, sz, skb->head, skb->data,
126 (unsigned long)skb->tail, (unsigned long)skb->end,
127 skb->dev ? skb->dev->name : "<NULL>");
128 BUG();
129}
130
131/**
132 * skb_under_panic - private function
133 * @skb: buffer
134 * @sz: size
135 * @here: address
136 *
137 * Out of line support code for skb_push(). Not user callable.
138 */
139
140static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
141{
142 printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
143 "data:%p tail:%#lx end:%#lx dev:%s\n",
144 here, skb->len, sz, skb->head, skb->data,
145 (unsigned long)skb->tail, (unsigned long)skb->end,
146 skb->dev ? skb->dev->name : "<NULL>");
147 BUG();
148}
149
150/* Allocate a new skbuff. We do this ourselves so we can fill in a few
151 * 'private' fields and also do memory statistics to find all the
152 * [BEEP] leaks.
153 *
154 */
155
156/**
157 * __alloc_skb - allocate a network buffer
158 * @size: size to allocate
159 * @gfp_mask: allocation mask
160 * @fclone: allocate from fclone cache instead of head cache
161 * and allocate a cloned (child) skb
162 * @node: numa node to allocate memory on
163 *
164 * Allocate a new &sk_buff. The returned buffer has no headroom and a
165 * tail room of size bytes. The object has a reference count of one.
166 * The return is the buffer. On a failure the return is %NULL.
167 *
168 * Buffers may only be allocated from interrupts using a @gfp_mask of
169 * %GFP_ATOMIC.
170 */
171struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
172 int fclone, int node)
173{
174 struct kmem_cache *cache;
175 struct skb_shared_info *shinfo;
176 struct sk_buff *skb;
177 u8 *data;
178
179 cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
180
181 /* Get the HEAD */
182 skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
183 if (!skb)
184 goto out;
185 prefetchw(skb);
186
187 size = SKB_DATA_ALIGN(size);
188 data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
189 gfp_mask, node);
190 if (!data)
191 goto nodata;
192 prefetchw(data + size);
193
194 /*
195 * Only clear those fields we need to clear, not those that we will
196 * actually initialise below. Hence, don't put any more fields after
197 * the tail pointer in struct sk_buff!
198 */
199 memset(skb, 0, offsetof(struct sk_buff, tail));
200 skb->truesize = size + sizeof(struct sk_buff);
201 atomic_set(&skb->users, 1);
202 skb->head = data;
203 skb->data = data;
204 skb_reset_tail_pointer(skb);
205 skb->end = skb->tail + size;
206#ifdef NET_SKBUFF_DATA_USES_OFFSET
207 skb->mac_header = ~0U;
208#endif
209
210 /* make sure we initialize shinfo sequentially */
211 shinfo = skb_shinfo(skb);
212 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
213 atomic_set(&shinfo->dataref, 1);
214 kmemcheck_annotate_variable(shinfo->destructor_arg);
215
216 if (fclone) {
217 struct sk_buff *child = skb + 1;
218 atomic_t *fclone_ref = (atomic_t *) (child + 1);
219
220 kmemcheck_annotate_bitfield(child, flags1);
221 kmemcheck_annotate_bitfield(child, flags2);
222 skb->fclone = SKB_FCLONE_ORIG;
223 atomic_set(fclone_ref, 1);
224
225 child->fclone = SKB_FCLONE_UNAVAILABLE;
226 }
227out:
228 return skb;
229nodata:
230 kmem_cache_free(cache, skb);
231 skb = NULL;
232 goto out;
233}
234EXPORT_SYMBOL(__alloc_skb);
235
236/**
237 * __netdev_alloc_skb - allocate an skbuff for rx on a specific device
238 * @dev: network device to receive on
239 * @length: length to allocate
240 * @gfp_mask: get_free_pages mask, passed to alloc_skb
241 *
242 * Allocate a new &sk_buff and assign it a usage count of one. The
243 * buffer has unspecified headroom built in. Users should allocate
244 * the headroom they think they need without accounting for the
245 * built in space. The built in space is used for optimisations.
246 *
247 * %NULL is returned if there is no free memory.
248 */
249struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
250 unsigned int length, gfp_t gfp_mask)
251{
252 struct sk_buff *skb;
253
254 skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, NUMA_NO_NODE);
255 if (likely(skb)) {
256 skb_reserve(skb, NET_SKB_PAD);
257 skb->dev = dev;
258 }
259 return skb;
260}
261EXPORT_SYMBOL(__netdev_alloc_skb);
262
263void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
264 int size)
265{
266 skb_fill_page_desc(skb, i, page, off, size);
267 skb->len += size;
268 skb->data_len += size;
269 skb->truesize += size;
270}
271EXPORT_SYMBOL(skb_add_rx_frag);
272
273/**
274 * dev_alloc_skb - allocate an skbuff for receiving
275 * @length: length to allocate
276 *
277 * Allocate a new &sk_buff and assign it a usage count of one. The
278 * buffer has unspecified headroom built in. Users should allocate
279 * the headroom they think they need without accounting for the
280 * built in space. The built in space is used for optimisations.
281 *
282 * %NULL is returned if there is no free memory. Although this function
283 * allocates memory it can be called from an interrupt.
284 */
285struct sk_buff *dev_alloc_skb(unsigned int length)
286{
287 /*
288 * There is more code here than it seems:
289 * __dev_alloc_skb is an inline
290 */
291 return __dev_alloc_skb(length, GFP_ATOMIC);
292}
293EXPORT_SYMBOL(dev_alloc_skb);
294
295static void skb_drop_list(struct sk_buff **listp)
296{
297 struct sk_buff *list = *listp;
298
299 *listp = NULL;
300
301 do {
302 struct sk_buff *this = list;
303 list = list->next;
304 kfree_skb(this);
305 } while (list);
306}
307
308static inline void skb_drop_fraglist(struct sk_buff *skb)
309{
310 skb_drop_list(&skb_shinfo(skb)->frag_list);
311}
312
313static void skb_clone_fraglist(struct sk_buff *skb)
314{
315 struct sk_buff *list;
316
317 skb_walk_frags(skb, list)
318 skb_get(list);
319}
320
321static void skb_release_data(struct sk_buff *skb)
322{
323 if (!skb->cloned ||
324 !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
325 &skb_shinfo(skb)->dataref)) {
326 if (skb_shinfo(skb)->nr_frags) {
327 int i;
328 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
329 put_page(skb_shinfo(skb)->frags[i].page);
330 }
331
332 /*
333 * If skb buf is from userspace, we need to notify the caller
334 * the lower device DMA has done;
335 */
336 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
337 struct ubuf_info *uarg;
338
339 uarg = skb_shinfo(skb)->destructor_arg;
340 if (uarg->callback)
341 uarg->callback(uarg);
342 }
343
344 if (skb_has_frag_list(skb))
345 skb_drop_fraglist(skb);
346
347 kfree(skb->head);
348 }
349}
350
351/*
352 * Free an skbuff by memory without cleaning the state.
353 */
354static void kfree_skbmem(struct sk_buff *skb)
355{
356 struct sk_buff *other;
357 atomic_t *fclone_ref;
358
359 switch (skb->fclone) {
360 case SKB_FCLONE_UNAVAILABLE:
361 kmem_cache_free(skbuff_head_cache, skb);
362 break;
363
364 case SKB_FCLONE_ORIG:
365 fclone_ref = (atomic_t *) (skb + 2);
366 if (atomic_dec_and_test(fclone_ref))
367 kmem_cache_free(skbuff_fclone_cache, skb);
368 break;
369
370 case SKB_FCLONE_CLONE:
371 fclone_ref = (atomic_t *) (skb + 1);
372 other = skb - 1;
373
374 /* The clone portion is available for
375 * fast-cloning again.
376 */
377 skb->fclone = SKB_FCLONE_UNAVAILABLE;
378
379 if (atomic_dec_and_test(fclone_ref))
380 kmem_cache_free(skbuff_fclone_cache, other);
381 break;
382 }
383}
384
385static void skb_release_head_state(struct sk_buff *skb)
386{
387 skb_dst_drop(skb);
388#ifdef CONFIG_XFRM
389 secpath_put(skb->sp);
390#endif
391 if (skb->destructor) {
392 WARN_ON(in_irq());
393 skb->destructor(skb);
394 }
395#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
396 nf_conntrack_put(skb->nfct);
397#endif
398#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
399 nf_conntrack_put_reasm(skb->nfct_reasm);
400#endif
401#ifdef CONFIG_BRIDGE_NETFILTER
402 nf_bridge_put(skb->nf_bridge);
403#endif
404/* XXX: IS this still necessary? - JHS */
405#ifdef CONFIG_NET_SCHED
406 skb->tc_index = 0;
407#ifdef CONFIG_NET_CLS_ACT
408 skb->tc_verd = 0;
409#endif
410#endif
411}
412
413/* Free everything but the sk_buff shell. */
414static void skb_release_all(struct sk_buff *skb)
415{
416 skb_release_head_state(skb);
417 skb_release_data(skb);
418}
419
420/**
421 * __kfree_skb - private function
422 * @skb: buffer
423 *
424 * Free an sk_buff. Release anything attached to the buffer.
425 * Clean the state. This is an internal helper function. Users should
426 * always call kfree_skb
427 */
428
429void __kfree_skb(struct sk_buff *skb)
430{
431 skb_release_all(skb);
432 kfree_skbmem(skb);
433}
434EXPORT_SYMBOL(__kfree_skb);
435
436/**
437 * kfree_skb - free an sk_buff
438 * @skb: buffer to free
439 *
440 * Drop a reference to the buffer and free it if the usage count has
441 * hit zero.
442 */
443void kfree_skb(struct sk_buff *skb)
444{
445 if (unlikely(!skb))
446 return;
447 if (likely(atomic_read(&skb->users) == 1))
448 smp_rmb();
449 else if (likely(!atomic_dec_and_test(&skb->users)))
450 return;
451 trace_kfree_skb(skb, __builtin_return_address(0));
452 __kfree_skb(skb);
453}
454EXPORT_SYMBOL(kfree_skb);
455
456/**
457 * consume_skb - free an skbuff
458 * @skb: buffer to free
459 *
460 * Drop a ref to the buffer and free it if the usage count has hit zero
461 * Functions identically to kfree_skb, but kfree_skb assumes that the frame
462 * is being dropped after a failure and notes that
463 */
464void consume_skb(struct sk_buff *skb)
465{
466 if (unlikely(!skb))
467 return;
468 if (likely(atomic_read(&skb->users) == 1))
469 smp_rmb();
470 else if (likely(!atomic_dec_and_test(&skb->users)))
471 return;
472 trace_consume_skb(skb);
473 __kfree_skb(skb);
474}
475EXPORT_SYMBOL(consume_skb);
476
477/**
478 * skb_recycle_check - check if skb can be reused for receive
479 * @skb: buffer
480 * @skb_size: minimum receive buffer size
481 *
482 * Checks that the skb passed in is not shared or cloned, and
483 * that it is linear and its head portion at least as large as
484 * skb_size so that it can be recycled as a receive buffer.
485 * If these conditions are met, this function does any necessary
486 * reference count dropping and cleans up the skbuff as if it
487 * just came from __alloc_skb().
488 */
489bool skb_recycle_check(struct sk_buff *skb, int skb_size)
490{
491 struct skb_shared_info *shinfo;
492
493 if (irqs_disabled())
494 return false;
495
496 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)
497 return false;
498
499 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE)
500 return false;
501
502 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD);
503 if (skb_end_pointer(skb) - skb->head < skb_size)
504 return false;
505
506 if (skb_shared(skb) || skb_cloned(skb))
507 return false;
508
509 skb_release_head_state(skb);
510
511 shinfo = skb_shinfo(skb);
512 memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
513 atomic_set(&shinfo->dataref, 1);
514
515 memset(skb, 0, offsetof(struct sk_buff, tail));
516 skb->data = skb->head + NET_SKB_PAD;
517 skb_reset_tail_pointer(skb);
518
519 return true;
520}
521EXPORT_SYMBOL(skb_recycle_check);
522
523static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
524{
525 new->tstamp = old->tstamp;
526 new->dev = old->dev;
527 new->transport_header = old->transport_header;
528 new->network_header = old->network_header;
529 new->mac_header = old->mac_header;
530 skb_dst_copy(new, old);
531 new->rxhash = old->rxhash;
532#ifdef CONFIG_XFRM
533 new->sp = secpath_get(old->sp);
534#endif
535 memcpy(new->cb, old->cb, sizeof(old->cb));
536 new->csum = old->csum;
537 new->local_df = old->local_df;
538 new->pkt_type = old->pkt_type;
539 new->ip_summed = old->ip_summed;
540 skb_copy_queue_mapping(new, old);
541 new->priority = old->priority;
542#if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
543 new->ipvs_property = old->ipvs_property;
544#endif
545 new->protocol = old->protocol;
546 new->mark = old->mark;
547 new->skb_iif = old->skb_iif;
548 __nf_copy(new, old);
549#if defined(CONFIG_NETFILTER_XT_TARGET_TRACE) || \
550 defined(CONFIG_NETFILTER_XT_TARGET_TRACE_MODULE)
551 new->nf_trace = old->nf_trace;
552#endif
553#ifdef CONFIG_NET_SCHED
554 new->tc_index = old->tc_index;
555#ifdef CONFIG_NET_CLS_ACT
556 new->tc_verd = old->tc_verd;
557#endif
558#endif
559 new->vlan_tci = old->vlan_tci;
560
561 skb_copy_secmark(new, old);
562}
563
564/*
565 * You should not add any new code to this function. Add it to
566 * __copy_skb_header above instead.
567 */
568static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
569{
570#define C(x) n->x = skb->x
571
572 n->next = n->prev = NULL;
573 n->sk = NULL;
574 __copy_skb_header(n, skb);
575
576 C(len);
577 C(data_len);
578 C(mac_len);
579 n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
580 n->cloned = 1;
581 n->nohdr = 0;
582 n->destructor = NULL;
583 C(tail);
584 C(end);
585 C(head);
586 C(data);
587 C(truesize);
588 atomic_set(&n->users, 1);
589
590 atomic_inc(&(skb_shinfo(skb)->dataref));
591 skb->cloned = 1;
592
593 return n;
594#undef C
595}
596
597/**
598 * skb_morph - morph one skb into another
599 * @dst: the skb to receive the contents
600 * @src: the skb to supply the contents
601 *
602 * This is identical to skb_clone except that the target skb is
603 * supplied by the user.
604 *
605 * The target skb is returned upon exit.
606 */
607struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
608{
609 skb_release_all(dst);
610 return __skb_clone(dst, src);
611}
612EXPORT_SYMBOL_GPL(skb_morph);
613
614/* skb_copy_ubufs - copy userspace skb frags buffers to kernel
615 * @skb: the skb to modify
616 * @gfp_mask: allocation priority
617 *
618 * This must be called on SKBTX_DEV_ZEROCOPY skb.
619 * It will copy all frags into kernel and drop the reference
620 * to userspace pages.
621 *
622 * If this function is called from an interrupt gfp_mask() must be
623 * %GFP_ATOMIC.
624 *
625 * Returns 0 on success or a negative error code on failure
626 * to allocate kernel memory to copy to.
627 */
628int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
629{
630 int i;
631 int num_frags = skb_shinfo(skb)->nr_frags;
632 struct page *page, *head = NULL;
633 struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
634
635 for (i = 0; i < num_frags; i++) {
636 u8 *vaddr;
637 skb_frag_t *f = &skb_shinfo(skb)->frags[i];
638
639 page = alloc_page(GFP_ATOMIC);
640 if (!page) {
641 while (head) {
642 struct page *next = (struct page *)head->private;
643 put_page(head);
644 head = next;
645 }
646 return -ENOMEM;
647 }
648 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
649 memcpy(page_address(page),
650 vaddr + f->page_offset, f->size);
651 kunmap_skb_frag(vaddr);
652 page->private = (unsigned long)head;
653 head = page;
654 }
655
656 /* skb frags release userspace buffers */
657 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
658 put_page(skb_shinfo(skb)->frags[i].page);
659
660 uarg->callback(uarg);
661
662 /* skb frags point to kernel buffers */
663 for (i = skb_shinfo(skb)->nr_frags; i > 0; i--) {
664 skb_shinfo(skb)->frags[i - 1].page_offset = 0;
665 skb_shinfo(skb)->frags[i - 1].page = head;
666 head = (struct page *)head->private;
667 }
668
669 skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
670 return 0;
671}
672
673
674/**
675 * skb_clone - duplicate an sk_buff
676 * @skb: buffer to clone
677 * @gfp_mask: allocation priority
678 *
679 * Duplicate an &sk_buff. The new one is not owned by a socket. Both
680 * copies share the same packet data but not structure. The new
681 * buffer has a reference count of 1. If the allocation fails the
682 * function returns %NULL otherwise the new buffer is returned.
683 *
684 * If this function is called from an interrupt gfp_mask() must be
685 * %GFP_ATOMIC.
686 */
687
688struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
689{
690 struct sk_buff *n;
691
692 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
693 if (skb_copy_ubufs(skb, gfp_mask))
694 return NULL;
695 }
696
697 n = skb + 1;
698 if (skb->fclone == SKB_FCLONE_ORIG &&
699 n->fclone == SKB_FCLONE_UNAVAILABLE) {
700 atomic_t *fclone_ref = (atomic_t *) (n + 1);
701 n->fclone = SKB_FCLONE_CLONE;
702 atomic_inc(fclone_ref);
703 } else {
704 n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
705 if (!n)
706 return NULL;
707
708 kmemcheck_annotate_bitfield(n, flags1);
709 kmemcheck_annotate_bitfield(n, flags2);
710 n->fclone = SKB_FCLONE_UNAVAILABLE;
711 }
712
713 return __skb_clone(n, skb);
714}
715EXPORT_SYMBOL(skb_clone);
716
717static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
718{
719#ifndef NET_SKBUFF_DATA_USES_OFFSET
720 /*
721 * Shift between the two data areas in bytes
722 */
723 unsigned long offset = new->data - old->data;
724#endif
725
726 __copy_skb_header(new, old);
727
728#ifndef NET_SKBUFF_DATA_USES_OFFSET
729 /* {transport,network,mac}_header are relative to skb->head */
730 new->transport_header += offset;
731 new->network_header += offset;
732 if (skb_mac_header_was_set(new))
733 new->mac_header += offset;
734#endif
735 skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
736 skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
737 skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
738}
739
740/**
741 * skb_copy - create private copy of an sk_buff
742 * @skb: buffer to copy
743 * @gfp_mask: allocation priority
744 *
745 * Make a copy of both an &sk_buff and its data. This is used when the
746 * caller wishes to modify the data and needs a private copy of the
747 * data to alter. Returns %NULL on failure or the pointer to the buffer
748 * on success. The returned buffer has a reference count of 1.
749 *
750 * As by-product this function converts non-linear &sk_buff to linear
751 * one, so that &sk_buff becomes completely private and caller is allowed
752 * to modify all the data of returned buffer. This means that this
753 * function is not recommended for use in circumstances when only
754 * header is going to be modified. Use pskb_copy() instead.
755 */
756
757struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
758{
759 int headerlen = skb_headroom(skb);
760 unsigned int size = (skb_end_pointer(skb) - skb->head) + skb->data_len;
761 struct sk_buff *n = alloc_skb(size, gfp_mask);
762
763 if (!n)
764 return NULL;
765
766 /* Set the data pointer */
767 skb_reserve(n, headerlen);
768 /* Set the tail pointer and length */
769 skb_put(n, skb->len);
770
771 if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
772 BUG();
773
774 copy_skb_header(n, skb);
775 return n;
776}
777EXPORT_SYMBOL(skb_copy);
778
779/**
780 * pskb_copy - create copy of an sk_buff with private head.
781 * @skb: buffer to copy
782 * @gfp_mask: allocation priority
783 *
784 * Make a copy of both an &sk_buff and part of its data, located
785 * in header. Fragmented data remain shared. This is used when
786 * the caller wishes to modify only header of &sk_buff and needs
787 * private copy of the header to alter. Returns %NULL on failure
788 * or the pointer to the buffer on success.
789 * The returned buffer has a reference count of 1.
790 */
791
792struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
793{
794 unsigned int size = skb_end_pointer(skb) - skb->head;
795 struct sk_buff *n = alloc_skb(size, gfp_mask);
796
797 if (!n)
798 goto out;
799
800 /* Set the data pointer */
801 skb_reserve(n, skb_headroom(skb));
802 /* Set the tail pointer and length */
803 skb_put(n, skb_headlen(skb));
804 /* Copy the bytes */
805 skb_copy_from_linear_data(skb, n->data, n->len);
806
807 n->truesize += skb->data_len;
808 n->data_len = skb->data_len;
809 n->len = skb->len;
810
811 if (skb_shinfo(skb)->nr_frags) {
812 int i;
813
814 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
815 if (skb_copy_ubufs(skb, gfp_mask)) {
816 kfree_skb(n);
817 n = NULL;
818 goto out;
819 }
820 }
821 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
822 skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
823 get_page(skb_shinfo(n)->frags[i].page);
824 }
825 skb_shinfo(n)->nr_frags = i;
826 }
827
828 if (skb_has_frag_list(skb)) {
829 skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
830 skb_clone_fraglist(n);
831 }
832
833 copy_skb_header(n, skb);
834out:
835 return n;
836}
837EXPORT_SYMBOL(pskb_copy);
838
839/**
840 * pskb_expand_head - reallocate header of &sk_buff
841 * @skb: buffer to reallocate
842 * @nhead: room to add at head
843 * @ntail: room to add at tail
844 * @gfp_mask: allocation priority
845 *
846 * Expands (or creates identical copy, if &nhead and &ntail are zero)
847 * header of skb. &sk_buff itself is not changed. &sk_buff MUST have
848 * reference count of 1. Returns zero in the case of success or error,
849 * if expansion failed. In the last case, &sk_buff is not changed.
850 *
851 * All the pointers pointing into skb header may change and must be
852 * reloaded after call to this function.
853 */
854
855int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
856 gfp_t gfp_mask)
857{
858 int i;
859 u8 *data;
860 int size = nhead + (skb_end_pointer(skb) - skb->head) + ntail;
861 long off;
862 bool fastpath;
863
864 BUG_ON(nhead < 0);
865
866 if (skb_shared(skb))
867 BUG();
868
869 size = SKB_DATA_ALIGN(size);
870
871 /* Check if we can avoid taking references on fragments if we own
872 * the last reference on skb->head. (see skb_release_data())
873 */
874 if (!skb->cloned)
875 fastpath = true;
876 else {
877 int delta = skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1;
878 fastpath = atomic_read(&skb_shinfo(skb)->dataref) == delta;
879 }
880
881 if (fastpath &&
882 size + sizeof(struct skb_shared_info) <= ksize(skb->head)) {
883 memmove(skb->head + size, skb_shinfo(skb),
884 offsetof(struct skb_shared_info,
885 frags[skb_shinfo(skb)->nr_frags]));
886 memmove(skb->head + nhead, skb->head,
887 skb_tail_pointer(skb) - skb->head);
888 off = nhead;
889 goto adjust_others;
890 }
891
892 data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
893 if (!data)
894 goto nodata;
895
896 /* Copy only real data... and, alas, header. This should be
897 * optimized for the cases when header is void.
898 */
899 memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
900
901 memcpy((struct skb_shared_info *)(data + size),
902 skb_shinfo(skb),
903 offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
904
905 if (fastpath) {
906 kfree(skb->head);
907 } else {
908 /* copy this zero copy skb frags */
909 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
910 if (skb_copy_ubufs(skb, gfp_mask))
911 goto nofrags;
912 }
913 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
914 get_page(skb_shinfo(skb)->frags[i].page);
915
916 if (skb_has_frag_list(skb))
917 skb_clone_fraglist(skb);
918
919 skb_release_data(skb);
920 }
921 off = (data + nhead) - skb->head;
922
923 skb->head = data;
924adjust_others:
925 skb->data += off;
926#ifdef NET_SKBUFF_DATA_USES_OFFSET
927 skb->end = size;
928 off = nhead;
929#else
930 skb->end = skb->head + size;
931#endif
932 /* {transport,network,mac}_header and tail are relative to skb->head */
933 skb->tail += off;
934 skb->transport_header += off;
935 skb->network_header += off;
936 if (skb_mac_header_was_set(skb))
937 skb->mac_header += off;
938 /* Only adjust this if it actually is csum_start rather than csum */
939 if (skb->ip_summed == CHECKSUM_PARTIAL)
940 skb->csum_start += nhead;
941 skb->cloned = 0;
942 skb->hdr_len = 0;
943 skb->nohdr = 0;
944 atomic_set(&skb_shinfo(skb)->dataref, 1);
945 return 0;
946
947nofrags:
948 kfree(data);
949nodata:
950 return -ENOMEM;
951}
952EXPORT_SYMBOL(pskb_expand_head);
953
954/* Make private copy of skb with writable head and some headroom */
955
956struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
957{
958 struct sk_buff *skb2;
959 int delta = headroom - skb_headroom(skb);
960
961 if (delta <= 0)
962 skb2 = pskb_copy(skb, GFP_ATOMIC);
963 else {
964 skb2 = skb_clone(skb, GFP_ATOMIC);
965 if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
966 GFP_ATOMIC)) {
967 kfree_skb(skb2);
968 skb2 = NULL;
969 }
970 }
971 return skb2;
972}
973EXPORT_SYMBOL(skb_realloc_headroom);
974
975/**
976 * skb_copy_expand - copy and expand sk_buff
977 * @skb: buffer to copy
978 * @newheadroom: new free bytes at head
979 * @newtailroom: new free bytes at tail
980 * @gfp_mask: allocation priority
981 *
982 * Make a copy of both an &sk_buff and its data and while doing so
983 * allocate additional space.
984 *
985 * This is used when the caller wishes to modify the data and needs a
986 * private copy of the data to alter as well as more space for new fields.
987 * Returns %NULL on failure or the pointer to the buffer
988 * on success. The returned buffer has a reference count of 1.
989 *
990 * You must pass %GFP_ATOMIC as the allocation priority if this function
991 * is called from an interrupt.
992 */
993struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
994 int newheadroom, int newtailroom,
995 gfp_t gfp_mask)
996{
997 /*
998 * Allocate the copy buffer
999 */
1000 struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
1001 gfp_mask);
1002 int oldheadroom = skb_headroom(skb);
1003 int head_copy_len, head_copy_off;
1004 int off;
1005
1006 if (!n)
1007 return NULL;
1008
1009 skb_reserve(n, newheadroom);
1010
1011 /* Set the tail pointer and length */
1012 skb_put(n, skb->len);
1013
1014 head_copy_len = oldheadroom;
1015 head_copy_off = 0;
1016 if (newheadroom <= head_copy_len)
1017 head_copy_len = newheadroom;
1018 else
1019 head_copy_off = newheadroom - head_copy_len;
1020
1021 /* Copy the linear header and data. */
1022 if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1023 skb->len + head_copy_len))
1024 BUG();
1025
1026 copy_skb_header(n, skb);
1027
1028 off = newheadroom - oldheadroom;
1029 if (n->ip_summed == CHECKSUM_PARTIAL)
1030 n->csum_start += off;
1031#ifdef NET_SKBUFF_DATA_USES_OFFSET
1032 n->transport_header += off;
1033 n->network_header += off;
1034 if (skb_mac_header_was_set(skb))
1035 n->mac_header += off;
1036#endif
1037
1038 return n;
1039}
1040EXPORT_SYMBOL(skb_copy_expand);
1041
1042/**
1043 * skb_pad - zero pad the tail of an skb
1044 * @skb: buffer to pad
1045 * @pad: space to pad
1046 *
1047 * Ensure that a buffer is followed by a padding area that is zero
1048 * filled. Used by network drivers which may DMA or transfer data
1049 * beyond the buffer end onto the wire.
1050 *
1051 * May return error in out of memory cases. The skb is freed on error.
1052 */
1053
1054int skb_pad(struct sk_buff *skb, int pad)
1055{
1056 int err;
1057 int ntail;
1058
1059 /* If the skbuff is non linear tailroom is always zero.. */
1060 if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1061 memset(skb->data+skb->len, 0, pad);
1062 return 0;
1063 }
1064
1065 ntail = skb->data_len + pad - (skb->end - skb->tail);
1066 if (likely(skb_cloned(skb) || ntail > 0)) {
1067 err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1068 if (unlikely(err))
1069 goto free_skb;
1070 }
1071
1072 /* FIXME: The use of this function with non-linear skb's really needs
1073 * to be audited.
1074 */
1075 err = skb_linearize(skb);
1076 if (unlikely(err))
1077 goto free_skb;
1078
1079 memset(skb->data + skb->len, 0, pad);
1080 return 0;
1081
1082free_skb:
1083 kfree_skb(skb);
1084 return err;
1085}
1086EXPORT_SYMBOL(skb_pad);
1087
1088/**
1089 * skb_put - add data to a buffer
1090 * @skb: buffer to use
1091 * @len: amount of data to add
1092 *
1093 * This function extends the used data area of the buffer. If this would
1094 * exceed the total buffer size the kernel will panic. A pointer to the
1095 * first byte of the extra data is returned.
1096 */
1097unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1098{
1099 unsigned char *tmp = skb_tail_pointer(skb);
1100 SKB_LINEAR_ASSERT(skb);
1101 skb->tail += len;
1102 skb->len += len;
1103 if (unlikely(skb->tail > skb->end))
1104 skb_over_panic(skb, len, __builtin_return_address(0));
1105 return tmp;
1106}
1107EXPORT_SYMBOL(skb_put);
1108
1109/**
1110 * skb_push - add data to the start of a buffer
1111 * @skb: buffer to use
1112 * @len: amount of data to add
1113 *
1114 * This function extends the used data area of the buffer at the buffer
1115 * start. If this would exceed the total buffer headroom the kernel will
1116 * panic. A pointer to the first byte of the extra data is returned.
1117 */
1118unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1119{
1120 skb->data -= len;
1121 skb->len += len;
1122 if (unlikely(skb->data<skb->head))
1123 skb_under_panic(skb, len, __builtin_return_address(0));
1124 return skb->data;
1125}
1126EXPORT_SYMBOL(skb_push);
1127
1128/**
1129 * skb_pull - remove data from the start of a buffer
1130 * @skb: buffer to use
1131 * @len: amount of data to remove
1132 *
1133 * This function removes data from the start of a buffer, returning
1134 * the memory to the headroom. A pointer to the next data in the buffer
1135 * is returned. Once the data has been pulled future pushes will overwrite
1136 * the old data.
1137 */
1138unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1139{
1140 return skb_pull_inline(skb, len);
1141}
1142EXPORT_SYMBOL(skb_pull);
1143
1144/**
1145 * skb_trim - remove end from a buffer
1146 * @skb: buffer to alter
1147 * @len: new length
1148 *
1149 * Cut the length of a buffer down by removing data from the tail. If
1150 * the buffer is already under the length specified it is not modified.
1151 * The skb must be linear.
1152 */
1153void skb_trim(struct sk_buff *skb, unsigned int len)
1154{
1155 if (skb->len > len)
1156 __skb_trim(skb, len);
1157}
1158EXPORT_SYMBOL(skb_trim);
1159
1160/* Trims skb to length len. It can change skb pointers.
1161 */
1162
1163int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1164{
1165 struct sk_buff **fragp;
1166 struct sk_buff *frag;
1167 int offset = skb_headlen(skb);
1168 int nfrags = skb_shinfo(skb)->nr_frags;
1169 int i;
1170 int err;
1171
1172 if (skb_cloned(skb) &&
1173 unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1174 return err;
1175
1176 i = 0;
1177 if (offset >= len)
1178 goto drop_pages;
1179
1180 for (; i < nfrags; i++) {
1181 int end = offset + skb_shinfo(skb)->frags[i].size;
1182
1183 if (end < len) {
1184 offset = end;
1185 continue;
1186 }
1187
1188 skb_shinfo(skb)->frags[i++].size = len - offset;
1189
1190drop_pages:
1191 skb_shinfo(skb)->nr_frags = i;
1192
1193 for (; i < nfrags; i++)
1194 put_page(skb_shinfo(skb)->frags[i].page);
1195
1196 if (skb_has_frag_list(skb))
1197 skb_drop_fraglist(skb);
1198 goto done;
1199 }
1200
1201 for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1202 fragp = &frag->next) {
1203 int end = offset + frag->len;
1204
1205 if (skb_shared(frag)) {
1206 struct sk_buff *nfrag;
1207
1208 nfrag = skb_clone(frag, GFP_ATOMIC);
1209 if (unlikely(!nfrag))
1210 return -ENOMEM;
1211
1212 nfrag->next = frag->next;
1213 kfree_skb(frag);
1214 frag = nfrag;
1215 *fragp = frag;
1216 }
1217
1218 if (end < len) {
1219 offset = end;
1220 continue;
1221 }
1222
1223 if (end > len &&
1224 unlikely((err = pskb_trim(frag, len - offset))))
1225 return err;
1226
1227 if (frag->next)
1228 skb_drop_list(&frag->next);
1229 break;
1230 }
1231
1232done:
1233 if (len > skb_headlen(skb)) {
1234 skb->data_len -= skb->len - len;
1235 skb->len = len;
1236 } else {
1237 skb->len = len;
1238 skb->data_len = 0;
1239 skb_set_tail_pointer(skb, len);
1240 }
1241
1242 return 0;
1243}
1244EXPORT_SYMBOL(___pskb_trim);
1245
1246/**
1247 * __pskb_pull_tail - advance tail of skb header
1248 * @skb: buffer to reallocate
1249 * @delta: number of bytes to advance tail
1250 *
1251 * The function makes a sense only on a fragmented &sk_buff,
1252 * it expands header moving its tail forward and copying necessary
1253 * data from fragmented part.
1254 *
1255 * &sk_buff MUST have reference count of 1.
1256 *
1257 * Returns %NULL (and &sk_buff does not change) if pull failed
1258 * or value of new tail of skb in the case of success.
1259 *
1260 * All the pointers pointing into skb header may change and must be
1261 * reloaded after call to this function.
1262 */
1263
1264/* Moves tail of skb head forward, copying data from fragmented part,
1265 * when it is necessary.
1266 * 1. It may fail due to malloc failure.
1267 * 2. It may change skb pointers.
1268 *
1269 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1270 */
1271unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1272{
1273 /* If skb has not enough free space at tail, get new one
1274 * plus 128 bytes for future expansions. If we have enough
1275 * room at tail, reallocate without expansion only if skb is cloned.
1276 */
1277 int i, k, eat = (skb->tail + delta) - skb->end;
1278
1279 if (eat > 0 || skb_cloned(skb)) {
1280 if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1281 GFP_ATOMIC))
1282 return NULL;
1283 }
1284
1285 if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1286 BUG();
1287
1288 /* Optimization: no fragments, no reasons to preestimate
1289 * size of pulled pages. Superb.
1290 */
1291 if (!skb_has_frag_list(skb))
1292 goto pull_pages;
1293
1294 /* Estimate size of pulled pages. */
1295 eat = delta;
1296 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1297 if (skb_shinfo(skb)->frags[i].size >= eat)
1298 goto pull_pages;
1299 eat -= skb_shinfo(skb)->frags[i].size;
1300 }
1301
1302 /* If we need update frag list, we are in troubles.
1303 * Certainly, it possible to add an offset to skb data,
1304 * but taking into account that pulling is expected to
1305 * be very rare operation, it is worth to fight against
1306 * further bloating skb head and crucify ourselves here instead.
1307 * Pure masohism, indeed. 8)8)
1308 */
1309 if (eat) {
1310 struct sk_buff *list = skb_shinfo(skb)->frag_list;
1311 struct sk_buff *clone = NULL;
1312 struct sk_buff *insp = NULL;
1313
1314 do {
1315 BUG_ON(!list);
1316
1317 if (list->len <= eat) {
1318 /* Eaten as whole. */
1319 eat -= list->len;
1320 list = list->next;
1321 insp = list;
1322 } else {
1323 /* Eaten partially. */
1324
1325 if (skb_shared(list)) {
1326 /* Sucks! We need to fork list. :-( */
1327 clone = skb_clone(list, GFP_ATOMIC);
1328 if (!clone)
1329 return NULL;
1330 insp = list->next;
1331 list = clone;
1332 } else {
1333 /* This may be pulled without
1334 * problems. */
1335 insp = list;
1336 }
1337 if (!pskb_pull(list, eat)) {
1338 kfree_skb(clone);
1339 return NULL;
1340 }
1341 break;
1342 }
1343 } while (eat);
1344
1345 /* Free pulled out fragments. */
1346 while ((list = skb_shinfo(skb)->frag_list) != insp) {
1347 skb_shinfo(skb)->frag_list = list->next;
1348 kfree_skb(list);
1349 }
1350 /* And insert new clone at head. */
1351 if (clone) {
1352 clone->next = list;
1353 skb_shinfo(skb)->frag_list = clone;
1354 }
1355 }
1356 /* Success! Now we may commit changes to skb data. */
1357
1358pull_pages:
1359 eat = delta;
1360 k = 0;
1361 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1362 if (skb_shinfo(skb)->frags[i].size <= eat) {
1363 put_page(skb_shinfo(skb)->frags[i].page);
1364 eat -= skb_shinfo(skb)->frags[i].size;
1365 } else {
1366 skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1367 if (eat) {
1368 skb_shinfo(skb)->frags[k].page_offset += eat;
1369 skb_shinfo(skb)->frags[k].size -= eat;
1370 eat = 0;
1371 }
1372 k++;
1373 }
1374 }
1375 skb_shinfo(skb)->nr_frags = k;
1376
1377 skb->tail += delta;
1378 skb->data_len -= delta;
1379
1380 return skb_tail_pointer(skb);
1381}
1382EXPORT_SYMBOL(__pskb_pull_tail);
1383
1384/**
1385 * skb_copy_bits - copy bits from skb to kernel buffer
1386 * @skb: source skb
1387 * @offset: offset in source
1388 * @to: destination buffer
1389 * @len: number of bytes to copy
1390 *
1391 * Copy the specified number of bytes from the source skb to the
1392 * destination buffer.
1393 *
1394 * CAUTION ! :
1395 * If its prototype is ever changed,
1396 * check arch/{*}/net/{*}.S files,
1397 * since it is called from BPF assembly code.
1398 */
1399int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1400{
1401 int start = skb_headlen(skb);
1402 struct sk_buff *frag_iter;
1403 int i, copy;
1404
1405 if (offset > (int)skb->len - len)
1406 goto fault;
1407
1408 /* Copy header. */
1409 if ((copy = start - offset) > 0) {
1410 if (copy > len)
1411 copy = len;
1412 skb_copy_from_linear_data_offset(skb, offset, to, copy);
1413 if ((len -= copy) == 0)
1414 return 0;
1415 offset += copy;
1416 to += copy;
1417 }
1418
1419 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1420 int end;
1421
1422 WARN_ON(start > offset + len);
1423
1424 end = start + skb_shinfo(skb)->frags[i].size;
1425 if ((copy = end - offset) > 0) {
1426 u8 *vaddr;
1427
1428 if (copy > len)
1429 copy = len;
1430
1431 vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1432 memcpy(to,
1433 vaddr + skb_shinfo(skb)->frags[i].page_offset+
1434 offset - start, copy);
1435 kunmap_skb_frag(vaddr);
1436
1437 if ((len -= copy) == 0)
1438 return 0;
1439 offset += copy;
1440 to += copy;
1441 }
1442 start = end;
1443 }
1444
1445 skb_walk_frags(skb, frag_iter) {
1446 int end;
1447
1448 WARN_ON(start > offset + len);
1449
1450 end = start + frag_iter->len;
1451 if ((copy = end - offset) > 0) {
1452 if (copy > len)
1453 copy = len;
1454 if (skb_copy_bits(frag_iter, offset - start, to, copy))
1455 goto fault;
1456 if ((len -= copy) == 0)
1457 return 0;
1458 offset += copy;
1459 to += copy;
1460 }
1461 start = end;
1462 }
1463
1464 if (!len)
1465 return 0;
1466
1467fault:
1468 return -EFAULT;
1469}
1470EXPORT_SYMBOL(skb_copy_bits);
1471
1472/*
1473 * Callback from splice_to_pipe(), if we need to release some pages
1474 * at the end of the spd in case we error'ed out in filling the pipe.
1475 */
1476static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1477{
1478 put_page(spd->pages[i]);
1479}
1480
1481static inline struct page *linear_to_page(struct page *page, unsigned int *len,
1482 unsigned int *offset,
1483 struct sk_buff *skb, struct sock *sk)
1484{
1485 struct page *p = sk->sk_sndmsg_page;
1486 unsigned int off;
1487
1488 if (!p) {
1489new_page:
1490 p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1491 if (!p)
1492 return NULL;
1493
1494 off = sk->sk_sndmsg_off = 0;
1495 /* hold one ref to this page until it's full */
1496 } else {
1497 unsigned int mlen;
1498
1499 off = sk->sk_sndmsg_off;
1500 mlen = PAGE_SIZE - off;
1501 if (mlen < 64 && mlen < *len) {
1502 put_page(p);
1503 goto new_page;
1504 }
1505
1506 *len = min_t(unsigned int, *len, mlen);
1507 }
1508
1509 memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1510 sk->sk_sndmsg_off += *len;
1511 *offset = off;
1512 get_page(p);
1513
1514 return p;
1515}
1516
1517/*
1518 * Fill page/offset/length into spd, if it can hold more pages.
1519 */
1520static inline int spd_fill_page(struct splice_pipe_desc *spd,
1521 struct pipe_inode_info *pipe, struct page *page,
1522 unsigned int *len, unsigned int offset,
1523 struct sk_buff *skb, int linear,
1524 struct sock *sk)
1525{
1526 if (unlikely(spd->nr_pages == pipe->buffers))
1527 return 1;
1528
1529 if (linear) {
1530 page = linear_to_page(page, len, &offset, skb, sk);
1531 if (!page)
1532 return 1;
1533 } else
1534 get_page(page);
1535
1536 spd->pages[spd->nr_pages] = page;
1537 spd->partial[spd->nr_pages].len = *len;
1538 spd->partial[spd->nr_pages].offset = offset;
1539 spd->nr_pages++;
1540
1541 return 0;
1542}
1543
1544static inline void __segment_seek(struct page **page, unsigned int *poff,
1545 unsigned int *plen, unsigned int off)
1546{
1547 unsigned long n;
1548
1549 *poff += off;
1550 n = *poff / PAGE_SIZE;
1551 if (n)
1552 *page = nth_page(*page, n);
1553
1554 *poff = *poff % PAGE_SIZE;
1555 *plen -= off;
1556}
1557
1558static inline int __splice_segment(struct page *page, unsigned int poff,
1559 unsigned int plen, unsigned int *off,
1560 unsigned int *len, struct sk_buff *skb,
1561 struct splice_pipe_desc *spd, int linear,
1562 struct sock *sk,
1563 struct pipe_inode_info *pipe)
1564{
1565 if (!*len)
1566 return 1;
1567
1568 /* skip this segment if already processed */
1569 if (*off >= plen) {
1570 *off -= plen;
1571 return 0;
1572 }
1573
1574 /* ignore any bits we already processed */
1575 if (*off) {
1576 __segment_seek(&page, &poff, &plen, *off);
1577 *off = 0;
1578 }
1579
1580 do {
1581 unsigned int flen = min(*len, plen);
1582
1583 /* the linear region may spread across several pages */
1584 flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1585
1586 if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1587 return 1;
1588
1589 __segment_seek(&page, &poff, &plen, flen);
1590 *len -= flen;
1591
1592 } while (*len && plen);
1593
1594 return 0;
1595}
1596
1597/*
1598 * Map linear and fragment data from the skb to spd. It reports failure if the
1599 * pipe is full or if we already spliced the requested length.
1600 */
1601static int __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1602 unsigned int *offset, unsigned int *len,
1603 struct splice_pipe_desc *spd, struct sock *sk)
1604{
1605 int seg;
1606
1607 /*
1608 * map the linear part
1609 */
1610 if (__splice_segment(virt_to_page(skb->data),
1611 (unsigned long) skb->data & (PAGE_SIZE - 1),
1612 skb_headlen(skb),
1613 offset, len, skb, spd, 1, sk, pipe))
1614 return 1;
1615
1616 /*
1617 * then map the fragments
1618 */
1619 for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1620 const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1621
1622 if (__splice_segment(f->page, f->page_offset, f->size,
1623 offset, len, skb, spd, 0, sk, pipe))
1624 return 1;
1625 }
1626
1627 return 0;
1628}
1629
1630/*
1631 * Map data from the skb to a pipe. Should handle both the linear part,
1632 * the fragments, and the frag list. It does NOT handle frag lists within
1633 * the frag list, if such a thing exists. We'd probably need to recurse to
1634 * handle that cleanly.
1635 */
1636int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1637 struct pipe_inode_info *pipe, unsigned int tlen,
1638 unsigned int flags)
1639{
1640 struct partial_page partial[PIPE_DEF_BUFFERS];
1641 struct page *pages[PIPE_DEF_BUFFERS];
1642 struct splice_pipe_desc spd = {
1643 .pages = pages,
1644 .partial = partial,
1645 .flags = flags,
1646 .ops = &sock_pipe_buf_ops,
1647 .spd_release = sock_spd_release,
1648 };
1649 struct sk_buff *frag_iter;
1650 struct sock *sk = skb->sk;
1651 int ret = 0;
1652
1653 if (splice_grow_spd(pipe, &spd))
1654 return -ENOMEM;
1655
1656 /*
1657 * __skb_splice_bits() only fails if the output has no room left,
1658 * so no point in going over the frag_list for the error case.
1659 */
1660 if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1661 goto done;
1662 else if (!tlen)
1663 goto done;
1664
1665 /*
1666 * now see if we have a frag_list to map
1667 */
1668 skb_walk_frags(skb, frag_iter) {
1669 if (!tlen)
1670 break;
1671 if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1672 break;
1673 }
1674
1675done:
1676 if (spd.nr_pages) {
1677 /*
1678 * Drop the socket lock, otherwise we have reverse
1679 * locking dependencies between sk_lock and i_mutex
1680 * here as compared to sendfile(). We enter here
1681 * with the socket lock held, and splice_to_pipe() will
1682 * grab the pipe inode lock. For sendfile() emulation,
1683 * we call into ->sendpage() with the i_mutex lock held
1684 * and networking will grab the socket lock.
1685 */
1686 release_sock(sk);
1687 ret = splice_to_pipe(pipe, &spd);
1688 lock_sock(sk);
1689 }
1690
1691 splice_shrink_spd(pipe, &spd);
1692 return ret;
1693}
1694
1695/**
1696 * skb_store_bits - store bits from kernel buffer to skb
1697 * @skb: destination buffer
1698 * @offset: offset in destination
1699 * @from: source buffer
1700 * @len: number of bytes to copy
1701 *
1702 * Copy the specified number of bytes from the source buffer to the
1703 * destination skb. This function handles all the messy bits of
1704 * traversing fragment lists and such.
1705 */
1706
1707int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1708{
1709 int start = skb_headlen(skb);
1710 struct sk_buff *frag_iter;
1711 int i, copy;
1712
1713 if (offset > (int)skb->len - len)
1714 goto fault;
1715
1716 if ((copy = start - offset) > 0) {
1717 if (copy > len)
1718 copy = len;
1719 skb_copy_to_linear_data_offset(skb, offset, from, copy);
1720 if ((len -= copy) == 0)
1721 return 0;
1722 offset += copy;
1723 from += copy;
1724 }
1725
1726 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1727 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1728 int end;
1729
1730 WARN_ON(start > offset + len);
1731
1732 end = start + frag->size;
1733 if ((copy = end - offset) > 0) {
1734 u8 *vaddr;
1735
1736 if (copy > len)
1737 copy = len;
1738
1739 vaddr = kmap_skb_frag(frag);
1740 memcpy(vaddr + frag->page_offset + offset - start,
1741 from, copy);
1742 kunmap_skb_frag(vaddr);
1743
1744 if ((len -= copy) == 0)
1745 return 0;
1746 offset += copy;
1747 from += copy;
1748 }
1749 start = end;
1750 }
1751
1752 skb_walk_frags(skb, frag_iter) {
1753 int end;
1754
1755 WARN_ON(start > offset + len);
1756
1757 end = start + frag_iter->len;
1758 if ((copy = end - offset) > 0) {
1759 if (copy > len)
1760 copy = len;
1761 if (skb_store_bits(frag_iter, offset - start,
1762 from, copy))
1763 goto fault;
1764 if ((len -= copy) == 0)
1765 return 0;
1766 offset += copy;
1767 from += copy;
1768 }
1769 start = end;
1770 }
1771 if (!len)
1772 return 0;
1773
1774fault:
1775 return -EFAULT;
1776}
1777EXPORT_SYMBOL(skb_store_bits);
1778
1779/* Checksum skb data. */
1780
1781__wsum skb_checksum(const struct sk_buff *skb, int offset,
1782 int len, __wsum csum)
1783{
1784 int start = skb_headlen(skb);
1785 int i, copy = start - offset;
1786 struct sk_buff *frag_iter;
1787 int pos = 0;
1788
1789 /* Checksum header. */
1790 if (copy > 0) {
1791 if (copy > len)
1792 copy = len;
1793 csum = csum_partial(skb->data + offset, copy, csum);
1794 if ((len -= copy) == 0)
1795 return csum;
1796 offset += copy;
1797 pos = copy;
1798 }
1799
1800 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1801 int end;
1802
1803 WARN_ON(start > offset + len);
1804
1805 end = start + skb_shinfo(skb)->frags[i].size;
1806 if ((copy = end - offset) > 0) {
1807 __wsum csum2;
1808 u8 *vaddr;
1809 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1810
1811 if (copy > len)
1812 copy = len;
1813 vaddr = kmap_skb_frag(frag);
1814 csum2 = csum_partial(vaddr + frag->page_offset +
1815 offset - start, copy, 0);
1816 kunmap_skb_frag(vaddr);
1817 csum = csum_block_add(csum, csum2, pos);
1818 if (!(len -= copy))
1819 return csum;
1820 offset += copy;
1821 pos += copy;
1822 }
1823 start = end;
1824 }
1825
1826 skb_walk_frags(skb, frag_iter) {
1827 int end;
1828
1829 WARN_ON(start > offset + len);
1830
1831 end = start + frag_iter->len;
1832 if ((copy = end - offset) > 0) {
1833 __wsum csum2;
1834 if (copy > len)
1835 copy = len;
1836 csum2 = skb_checksum(frag_iter, offset - start,
1837 copy, 0);
1838 csum = csum_block_add(csum, csum2, pos);
1839 if ((len -= copy) == 0)
1840 return csum;
1841 offset += copy;
1842 pos += copy;
1843 }
1844 start = end;
1845 }
1846 BUG_ON(len);
1847
1848 return csum;
1849}
1850EXPORT_SYMBOL(skb_checksum);
1851
1852/* Both of above in one bottle. */
1853
1854__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1855 u8 *to, int len, __wsum csum)
1856{
1857 int start = skb_headlen(skb);
1858 int i, copy = start - offset;
1859 struct sk_buff *frag_iter;
1860 int pos = 0;
1861
1862 /* Copy header. */
1863 if (copy > 0) {
1864 if (copy > len)
1865 copy = len;
1866 csum = csum_partial_copy_nocheck(skb->data + offset, to,
1867 copy, csum);
1868 if ((len -= copy) == 0)
1869 return csum;
1870 offset += copy;
1871 to += copy;
1872 pos = copy;
1873 }
1874
1875 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1876 int end;
1877
1878 WARN_ON(start > offset + len);
1879
1880 end = start + skb_shinfo(skb)->frags[i].size;
1881 if ((copy = end - offset) > 0) {
1882 __wsum csum2;
1883 u8 *vaddr;
1884 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1885
1886 if (copy > len)
1887 copy = len;
1888 vaddr = kmap_skb_frag(frag);
1889 csum2 = csum_partial_copy_nocheck(vaddr +
1890 frag->page_offset +
1891 offset - start, to,
1892 copy, 0);
1893 kunmap_skb_frag(vaddr);
1894 csum = csum_block_add(csum, csum2, pos);
1895 if (!(len -= copy))
1896 return csum;
1897 offset += copy;
1898 to += copy;
1899 pos += copy;
1900 }
1901 start = end;
1902 }
1903
1904 skb_walk_frags(skb, frag_iter) {
1905 __wsum csum2;
1906 int end;
1907
1908 WARN_ON(start > offset + len);
1909
1910 end = start + frag_iter->len;
1911 if ((copy = end - offset) > 0) {
1912 if (copy > len)
1913 copy = len;
1914 csum2 = skb_copy_and_csum_bits(frag_iter,
1915 offset - start,
1916 to, copy, 0);
1917 csum = csum_block_add(csum, csum2, pos);
1918 if ((len -= copy) == 0)
1919 return csum;
1920 offset += copy;
1921 to += copy;
1922 pos += copy;
1923 }
1924 start = end;
1925 }
1926 BUG_ON(len);
1927 return csum;
1928}
1929EXPORT_SYMBOL(skb_copy_and_csum_bits);
1930
1931void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1932{
1933 __wsum csum;
1934 long csstart;
1935
1936 if (skb->ip_summed == CHECKSUM_PARTIAL)
1937 csstart = skb_checksum_start_offset(skb);
1938 else
1939 csstart = skb_headlen(skb);
1940
1941 BUG_ON(csstart > skb_headlen(skb));
1942
1943 skb_copy_from_linear_data(skb, to, csstart);
1944
1945 csum = 0;
1946 if (csstart != skb->len)
1947 csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1948 skb->len - csstart, 0);
1949
1950 if (skb->ip_summed == CHECKSUM_PARTIAL) {
1951 long csstuff = csstart + skb->csum_offset;
1952
1953 *((__sum16 *)(to + csstuff)) = csum_fold(csum);
1954 }
1955}
1956EXPORT_SYMBOL(skb_copy_and_csum_dev);
1957
1958/**
1959 * skb_dequeue - remove from the head of the queue
1960 * @list: list to dequeue from
1961 *
1962 * Remove the head of the list. The list lock is taken so the function
1963 * may be used safely with other locking list functions. The head item is
1964 * returned or %NULL if the list is empty.
1965 */
1966
1967struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1968{
1969 unsigned long flags;
1970 struct sk_buff *result;
1971
1972 spin_lock_irqsave(&list->lock, flags);
1973 result = __skb_dequeue(list);
1974 spin_unlock_irqrestore(&list->lock, flags);
1975 return result;
1976}
1977EXPORT_SYMBOL(skb_dequeue);
1978
1979/**
1980 * skb_dequeue_tail - remove from the tail of the queue
1981 * @list: list to dequeue from
1982 *
1983 * Remove the tail of the list. The list lock is taken so the function
1984 * may be used safely with other locking list functions. The tail item is
1985 * returned or %NULL if the list is empty.
1986 */
1987struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1988{
1989 unsigned long flags;
1990 struct sk_buff *result;
1991
1992 spin_lock_irqsave(&list->lock, flags);
1993 result = __skb_dequeue_tail(list);
1994 spin_unlock_irqrestore(&list->lock, flags);
1995 return result;
1996}
1997EXPORT_SYMBOL(skb_dequeue_tail);
1998
1999/**
2000 * skb_queue_purge - empty a list
2001 * @list: list to empty
2002 *
2003 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2004 * the list and one reference dropped. This function takes the list
2005 * lock and is atomic with respect to other list locking functions.
2006 */
2007void skb_queue_purge(struct sk_buff_head *list)
2008{
2009 struct sk_buff *skb;
2010 while ((skb = skb_dequeue(list)) != NULL)
2011 kfree_skb(skb);
2012}
2013EXPORT_SYMBOL(skb_queue_purge);
2014
2015/**
2016 * skb_queue_head - queue a buffer at the list head
2017 * @list: list to use
2018 * @newsk: buffer to queue
2019 *
2020 * Queue a buffer at the start of the list. This function takes the
2021 * list lock and can be used safely with other locking &sk_buff functions
2022 * safely.
2023 *
2024 * A buffer cannot be placed on two lists at the same time.
2025 */
2026void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2027{
2028 unsigned long flags;
2029
2030 spin_lock_irqsave(&list->lock, flags);
2031 __skb_queue_head(list, newsk);
2032 spin_unlock_irqrestore(&list->lock, flags);
2033}
2034EXPORT_SYMBOL(skb_queue_head);
2035
2036/**
2037 * skb_queue_tail - queue a buffer at the list tail
2038 * @list: list to use
2039 * @newsk: buffer to queue
2040 *
2041 * Queue a buffer at the tail of the list. This function takes the
2042 * list lock and can be used safely with other locking &sk_buff functions
2043 * safely.
2044 *
2045 * A buffer cannot be placed on two lists at the same time.
2046 */
2047void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2048{
2049 unsigned long flags;
2050
2051 spin_lock_irqsave(&list->lock, flags);
2052 __skb_queue_tail(list, newsk);
2053 spin_unlock_irqrestore(&list->lock, flags);
2054}
2055EXPORT_SYMBOL(skb_queue_tail);
2056
2057/**
2058 * skb_unlink - remove a buffer from a list
2059 * @skb: buffer to remove
2060 * @list: list to use
2061 *
2062 * Remove a packet from a list. The list locks are taken and this
2063 * function is atomic with respect to other list locked calls
2064 *
2065 * You must know what list the SKB is on.
2066 */
2067void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2068{
2069 unsigned long flags;
2070
2071 spin_lock_irqsave(&list->lock, flags);
2072 __skb_unlink(skb, list);
2073 spin_unlock_irqrestore(&list->lock, flags);
2074}
2075EXPORT_SYMBOL(skb_unlink);
2076
2077/**
2078 * skb_append - append a buffer
2079 * @old: buffer to insert after
2080 * @newsk: buffer to insert
2081 * @list: list to use
2082 *
2083 * Place a packet after a given packet in a list. The list locks are taken
2084 * and this function is atomic with respect to other list locked calls.
2085 * A buffer cannot be placed on two lists at the same time.
2086 */
2087void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2088{
2089 unsigned long flags;
2090
2091 spin_lock_irqsave(&list->lock, flags);
2092 __skb_queue_after(list, old, newsk);
2093 spin_unlock_irqrestore(&list->lock, flags);
2094}
2095EXPORT_SYMBOL(skb_append);
2096
2097/**
2098 * skb_insert - insert a buffer
2099 * @old: buffer to insert before
2100 * @newsk: buffer to insert
2101 * @list: list to use
2102 *
2103 * Place a packet before a given packet in a list. The list locks are
2104 * taken and this function is atomic with respect to other list locked
2105 * calls.
2106 *
2107 * A buffer cannot be placed on two lists at the same time.
2108 */
2109void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2110{
2111 unsigned long flags;
2112
2113 spin_lock_irqsave(&list->lock, flags);
2114 __skb_insert(newsk, old->prev, old, list);
2115 spin_unlock_irqrestore(&list->lock, flags);
2116}
2117EXPORT_SYMBOL(skb_insert);
2118
2119static inline void skb_split_inside_header(struct sk_buff *skb,
2120 struct sk_buff* skb1,
2121 const u32 len, const int pos)
2122{
2123 int i;
2124
2125 skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2126 pos - len);
2127 /* And move data appendix as is. */
2128 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2129 skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2130
2131 skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2132 skb_shinfo(skb)->nr_frags = 0;
2133 skb1->data_len = skb->data_len;
2134 skb1->len += skb1->data_len;
2135 skb->data_len = 0;
2136 skb->len = len;
2137 skb_set_tail_pointer(skb, len);
2138}
2139
2140static inline void skb_split_no_header(struct sk_buff *skb,
2141 struct sk_buff* skb1,
2142 const u32 len, int pos)
2143{
2144 int i, k = 0;
2145 const int nfrags = skb_shinfo(skb)->nr_frags;
2146
2147 skb_shinfo(skb)->nr_frags = 0;
2148 skb1->len = skb1->data_len = skb->len - len;
2149 skb->len = len;
2150 skb->data_len = len - pos;
2151
2152 for (i = 0; i < nfrags; i++) {
2153 int size = skb_shinfo(skb)->frags[i].size;
2154
2155 if (pos + size > len) {
2156 skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2157
2158 if (pos < len) {
2159 /* Split frag.
2160 * We have two variants in this case:
2161 * 1. Move all the frag to the second
2162 * part, if it is possible. F.e.
2163 * this approach is mandatory for TUX,
2164 * where splitting is expensive.
2165 * 2. Split is accurately. We make this.
2166 */
2167 get_page(skb_shinfo(skb)->frags[i].page);
2168 skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2169 skb_shinfo(skb1)->frags[0].size -= len - pos;
2170 skb_shinfo(skb)->frags[i].size = len - pos;
2171 skb_shinfo(skb)->nr_frags++;
2172 }
2173 k++;
2174 } else
2175 skb_shinfo(skb)->nr_frags++;
2176 pos += size;
2177 }
2178 skb_shinfo(skb1)->nr_frags = k;
2179}
2180
2181/**
2182 * skb_split - Split fragmented skb to two parts at length len.
2183 * @skb: the buffer to split
2184 * @skb1: the buffer to receive the second part
2185 * @len: new length for skb
2186 */
2187void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2188{
2189 int pos = skb_headlen(skb);
2190
2191 if (len < pos) /* Split line is inside header. */
2192 skb_split_inside_header(skb, skb1, len, pos);
2193 else /* Second chunk has no header, nothing to copy. */
2194 skb_split_no_header(skb, skb1, len, pos);
2195}
2196EXPORT_SYMBOL(skb_split);
2197
2198/* Shifting from/to a cloned skb is a no-go.
2199 *
2200 * Caller cannot keep skb_shinfo related pointers past calling here!
2201 */
2202static int skb_prepare_for_shift(struct sk_buff *skb)
2203{
2204 return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2205}
2206
2207/**
2208 * skb_shift - Shifts paged data partially from skb to another
2209 * @tgt: buffer into which tail data gets added
2210 * @skb: buffer from which the paged data comes from
2211 * @shiftlen: shift up to this many bytes
2212 *
2213 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2214 * the length of the skb, from tgt to skb. Returns number bytes shifted.
2215 * It's up to caller to free skb if everything was shifted.
2216 *
2217 * If @tgt runs out of frags, the whole operation is aborted.
2218 *
2219 * Skb cannot include anything else but paged data while tgt is allowed
2220 * to have non-paged data as well.
2221 *
2222 * TODO: full sized shift could be optimized but that would need
2223 * specialized skb free'er to handle frags without up-to-date nr_frags.
2224 */
2225int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2226{
2227 int from, to, merge, todo;
2228 struct skb_frag_struct *fragfrom, *fragto;
2229
2230 BUG_ON(shiftlen > skb->len);
2231 BUG_ON(skb_headlen(skb)); /* Would corrupt stream */
2232
2233 todo = shiftlen;
2234 from = 0;
2235 to = skb_shinfo(tgt)->nr_frags;
2236 fragfrom = &skb_shinfo(skb)->frags[from];
2237
2238 /* Actual merge is delayed until the point when we know we can
2239 * commit all, so that we don't have to undo partial changes
2240 */
2241 if (!to ||
2242 !skb_can_coalesce(tgt, to, fragfrom->page, fragfrom->page_offset)) {
2243 merge = -1;
2244 } else {
2245 merge = to - 1;
2246
2247 todo -= fragfrom->size;
2248 if (todo < 0) {
2249 if (skb_prepare_for_shift(skb) ||
2250 skb_prepare_for_shift(tgt))
2251 return 0;
2252
2253 /* All previous frag pointers might be stale! */
2254 fragfrom = &skb_shinfo(skb)->frags[from];
2255 fragto = &skb_shinfo(tgt)->frags[merge];
2256
2257 fragto->size += shiftlen;
2258 fragfrom->size -= shiftlen;
2259 fragfrom->page_offset += shiftlen;
2260
2261 goto onlymerged;
2262 }
2263
2264 from++;
2265 }
2266
2267 /* Skip full, not-fitting skb to avoid expensive operations */
2268 if ((shiftlen == skb->len) &&
2269 (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2270 return 0;
2271
2272 if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2273 return 0;
2274
2275 while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2276 if (to == MAX_SKB_FRAGS)
2277 return 0;
2278
2279 fragfrom = &skb_shinfo(skb)->frags[from];
2280 fragto = &skb_shinfo(tgt)->frags[to];
2281
2282 if (todo >= fragfrom->size) {
2283 *fragto = *fragfrom;
2284 todo -= fragfrom->size;
2285 from++;
2286 to++;
2287
2288 } else {
2289 get_page(fragfrom->page);
2290 fragto->page = fragfrom->page;
2291 fragto->page_offset = fragfrom->page_offset;
2292 fragto->size = todo;
2293
2294 fragfrom->page_offset += todo;
2295 fragfrom->size -= todo;
2296 todo = 0;
2297
2298 to++;
2299 break;
2300 }
2301 }
2302
2303 /* Ready to "commit" this state change to tgt */
2304 skb_shinfo(tgt)->nr_frags = to;
2305
2306 if (merge >= 0) {
2307 fragfrom = &skb_shinfo(skb)->frags[0];
2308 fragto = &skb_shinfo(tgt)->frags[merge];
2309
2310 fragto->size += fragfrom->size;
2311 put_page(fragfrom->page);
2312 }
2313
2314 /* Reposition in the original skb */
2315 to = 0;
2316 while (from < skb_shinfo(skb)->nr_frags)
2317 skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2318 skb_shinfo(skb)->nr_frags = to;
2319
2320 BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2321
2322onlymerged:
2323 /* Most likely the tgt won't ever need its checksum anymore, skb on
2324 * the other hand might need it if it needs to be resent
2325 */
2326 tgt->ip_summed = CHECKSUM_PARTIAL;
2327 skb->ip_summed = CHECKSUM_PARTIAL;
2328
2329 /* Yak, is it really working this way? Some helper please? */
2330 skb->len -= shiftlen;
2331 skb->data_len -= shiftlen;
2332 skb->truesize -= shiftlen;
2333 tgt->len += shiftlen;
2334 tgt->data_len += shiftlen;
2335 tgt->truesize += shiftlen;
2336
2337 return shiftlen;
2338}
2339
2340/**
2341 * skb_prepare_seq_read - Prepare a sequential read of skb data
2342 * @skb: the buffer to read
2343 * @from: lower offset of data to be read
2344 * @to: upper offset of data to be read
2345 * @st: state variable
2346 *
2347 * Initializes the specified state variable. Must be called before
2348 * invoking skb_seq_read() for the first time.
2349 */
2350void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2351 unsigned int to, struct skb_seq_state *st)
2352{
2353 st->lower_offset = from;
2354 st->upper_offset = to;
2355 st->root_skb = st->cur_skb = skb;
2356 st->frag_idx = st->stepped_offset = 0;
2357 st->frag_data = NULL;
2358}
2359EXPORT_SYMBOL(skb_prepare_seq_read);
2360
2361/**
2362 * skb_seq_read - Sequentially read skb data
2363 * @consumed: number of bytes consumed by the caller so far
2364 * @data: destination pointer for data to be returned
2365 * @st: state variable
2366 *
2367 * Reads a block of skb data at &consumed relative to the
2368 * lower offset specified to skb_prepare_seq_read(). Assigns
2369 * the head of the data block to &data and returns the length
2370 * of the block or 0 if the end of the skb data or the upper
2371 * offset has been reached.
2372 *
2373 * The caller is not required to consume all of the data
2374 * returned, i.e. &consumed is typically set to the number
2375 * of bytes already consumed and the next call to
2376 * skb_seq_read() will return the remaining part of the block.
2377 *
2378 * Note 1: The size of each block of data returned can be arbitrary,
2379 * this limitation is the cost for zerocopy seqeuental
2380 * reads of potentially non linear data.
2381 *
2382 * Note 2: Fragment lists within fragments are not implemented
2383 * at the moment, state->root_skb could be replaced with
2384 * a stack for this purpose.
2385 */
2386unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2387 struct skb_seq_state *st)
2388{
2389 unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2390 skb_frag_t *frag;
2391
2392 if (unlikely(abs_offset >= st->upper_offset))
2393 return 0;
2394
2395next_skb:
2396 block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2397
2398 if (abs_offset < block_limit && !st->frag_data) {
2399 *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2400 return block_limit - abs_offset;
2401 }
2402
2403 if (st->frag_idx == 0 && !st->frag_data)
2404 st->stepped_offset += skb_headlen(st->cur_skb);
2405
2406 while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2407 frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2408 block_limit = frag->size + st->stepped_offset;
2409
2410 if (abs_offset < block_limit) {
2411 if (!st->frag_data)
2412 st->frag_data = kmap_skb_frag(frag);
2413
2414 *data = (u8 *) st->frag_data + frag->page_offset +
2415 (abs_offset - st->stepped_offset);
2416
2417 return block_limit - abs_offset;
2418 }
2419
2420 if (st->frag_data) {
2421 kunmap_skb_frag(st->frag_data);
2422 st->frag_data = NULL;
2423 }
2424
2425 st->frag_idx++;
2426 st->stepped_offset += frag->size;
2427 }
2428
2429 if (st->frag_data) {
2430 kunmap_skb_frag(st->frag_data);
2431 st->frag_data = NULL;
2432 }
2433
2434 if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2435 st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2436 st->frag_idx = 0;
2437 goto next_skb;
2438 } else if (st->cur_skb->next) {
2439 st->cur_skb = st->cur_skb->next;
2440 st->frag_idx = 0;
2441 goto next_skb;
2442 }
2443
2444 return 0;
2445}
2446EXPORT_SYMBOL(skb_seq_read);
2447
2448/**
2449 * skb_abort_seq_read - Abort a sequential read of skb data
2450 * @st: state variable
2451 *
2452 * Must be called if skb_seq_read() was not called until it
2453 * returned 0.
2454 */
2455void skb_abort_seq_read(struct skb_seq_state *st)
2456{
2457 if (st->frag_data)
2458 kunmap_skb_frag(st->frag_data);
2459}
2460EXPORT_SYMBOL(skb_abort_seq_read);
2461
2462#define TS_SKB_CB(state) ((struct skb_seq_state *) &((state)->cb))
2463
2464static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2465 struct ts_config *conf,
2466 struct ts_state *state)
2467{
2468 return skb_seq_read(offset, text, TS_SKB_CB(state));
2469}
2470
2471static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2472{
2473 skb_abort_seq_read(TS_SKB_CB(state));
2474}
2475
2476/**
2477 * skb_find_text - Find a text pattern in skb data
2478 * @skb: the buffer to look in
2479 * @from: search offset
2480 * @to: search limit
2481 * @config: textsearch configuration
2482 * @state: uninitialized textsearch state variable
2483 *
2484 * Finds a pattern in the skb data according to the specified
2485 * textsearch configuration. Use textsearch_next() to retrieve
2486 * subsequent occurrences of the pattern. Returns the offset
2487 * to the first occurrence or UINT_MAX if no match was found.
2488 */
2489unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2490 unsigned int to, struct ts_config *config,
2491 struct ts_state *state)
2492{
2493 unsigned int ret;
2494
2495 config->get_next_block = skb_ts_get_next_block;
2496 config->finish = skb_ts_finish;
2497
2498 skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2499
2500 ret = textsearch_find(config, state);
2501 return (ret <= to - from ? ret : UINT_MAX);
2502}
2503EXPORT_SYMBOL(skb_find_text);
2504
2505/**
2506 * skb_append_datato_frags: - append the user data to a skb
2507 * @sk: sock structure
2508 * @skb: skb structure to be appened with user data.
2509 * @getfrag: call back function to be used for getting the user data
2510 * @from: pointer to user message iov
2511 * @length: length of the iov message
2512 *
2513 * Description: This procedure append the user data in the fragment part
2514 * of the skb if any page alloc fails user this procedure returns -ENOMEM
2515 */
2516int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2517 int (*getfrag)(void *from, char *to, int offset,
2518 int len, int odd, struct sk_buff *skb),
2519 void *from, int length)
2520{
2521 int frg_cnt = 0;
2522 skb_frag_t *frag = NULL;
2523 struct page *page = NULL;
2524 int copy, left;
2525 int offset = 0;
2526 int ret;
2527
2528 do {
2529 /* Return error if we don't have space for new frag */
2530 frg_cnt = skb_shinfo(skb)->nr_frags;
2531 if (frg_cnt >= MAX_SKB_FRAGS)
2532 return -EFAULT;
2533
2534 /* allocate a new page for next frag */
2535 page = alloc_pages(sk->sk_allocation, 0);
2536
2537 /* If alloc_page fails just return failure and caller will
2538 * free previous allocated pages by doing kfree_skb()
2539 */
2540 if (page == NULL)
2541 return -ENOMEM;
2542
2543 /* initialize the next frag */
2544 skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2545 skb->truesize += PAGE_SIZE;
2546 atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2547
2548 /* get the new initialized frag */
2549 frg_cnt = skb_shinfo(skb)->nr_frags;
2550 frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2551
2552 /* copy the user data to page */
2553 left = PAGE_SIZE - frag->page_offset;
2554 copy = (length > left)? left : length;
2555
2556 ret = getfrag(from, (page_address(frag->page) +
2557 frag->page_offset + frag->size),
2558 offset, copy, 0, skb);
2559 if (ret < 0)
2560 return -EFAULT;
2561
2562 /* copy was successful so update the size parameters */
2563 frag->size += copy;
2564 skb->len += copy;
2565 skb->data_len += copy;
2566 offset += copy;
2567 length -= copy;
2568
2569 } while (length > 0);
2570
2571 return 0;
2572}
2573EXPORT_SYMBOL(skb_append_datato_frags);
2574
2575/**
2576 * skb_pull_rcsum - pull skb and update receive checksum
2577 * @skb: buffer to update
2578 * @len: length of data pulled
2579 *
2580 * This function performs an skb_pull on the packet and updates
2581 * the CHECKSUM_COMPLETE checksum. It should be used on
2582 * receive path processing instead of skb_pull unless you know
2583 * that the checksum difference is zero (e.g., a valid IP header)
2584 * or you are setting ip_summed to CHECKSUM_NONE.
2585 */
2586unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2587{
2588 BUG_ON(len > skb->len);
2589 skb->len -= len;
2590 BUG_ON(skb->len < skb->data_len);
2591 skb_postpull_rcsum(skb, skb->data, len);
2592 return skb->data += len;
2593}
2594EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2595
2596/**
2597 * skb_segment - Perform protocol segmentation on skb.
2598 * @skb: buffer to segment
2599 * @features: features for the output path (see dev->features)
2600 *
2601 * This function performs segmentation on the given skb. It returns
2602 * a pointer to the first in a list of new skbs for the segments.
2603 * In case of error it returns ERR_PTR(err).
2604 */
2605struct sk_buff *skb_segment(struct sk_buff *skb, u32 features)
2606{
2607 struct sk_buff *segs = NULL;
2608 struct sk_buff *tail = NULL;
2609 struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2610 unsigned int mss = skb_shinfo(skb)->gso_size;
2611 unsigned int doffset = skb->data - skb_mac_header(skb);
2612 unsigned int offset = doffset;
2613 unsigned int headroom;
2614 unsigned int len;
2615 int sg = !!(features & NETIF_F_SG);
2616 int nfrags = skb_shinfo(skb)->nr_frags;
2617 int err = -ENOMEM;
2618 int i = 0;
2619 int pos;
2620
2621 __skb_push(skb, doffset);
2622 headroom = skb_headroom(skb);
2623 pos = skb_headlen(skb);
2624
2625 do {
2626 struct sk_buff *nskb;
2627 skb_frag_t *frag;
2628 int hsize;
2629 int size;
2630
2631 len = skb->len - offset;
2632 if (len > mss)
2633 len = mss;
2634
2635 hsize = skb_headlen(skb) - offset;
2636 if (hsize < 0)
2637 hsize = 0;
2638 if (hsize > len || !sg)
2639 hsize = len;
2640
2641 if (!hsize && i >= nfrags) {
2642 BUG_ON(fskb->len != len);
2643
2644 pos += len;
2645 nskb = skb_clone(fskb, GFP_ATOMIC);
2646 fskb = fskb->next;
2647
2648 if (unlikely(!nskb))
2649 goto err;
2650
2651 hsize = skb_end_pointer(nskb) - nskb->head;
2652 if (skb_cow_head(nskb, doffset + headroom)) {
2653 kfree_skb(nskb);
2654 goto err;
2655 }
2656
2657 nskb->truesize += skb_end_pointer(nskb) - nskb->head -
2658 hsize;
2659 skb_release_head_state(nskb);
2660 __skb_push(nskb, doffset);
2661 } else {
2662 nskb = alloc_skb(hsize + doffset + headroom,
2663 GFP_ATOMIC);
2664
2665 if (unlikely(!nskb))
2666 goto err;
2667
2668 skb_reserve(nskb, headroom);
2669 __skb_put(nskb, doffset);
2670 }
2671
2672 if (segs)
2673 tail->next = nskb;
2674 else
2675 segs = nskb;
2676 tail = nskb;
2677
2678 __copy_skb_header(nskb, skb);
2679 nskb->mac_len = skb->mac_len;
2680
2681 /* nskb and skb might have different headroom */
2682 if (nskb->ip_summed == CHECKSUM_PARTIAL)
2683 nskb->csum_start += skb_headroom(nskb) - headroom;
2684
2685 skb_reset_mac_header(nskb);
2686 skb_set_network_header(nskb, skb->mac_len);
2687 nskb->transport_header = (nskb->network_header +
2688 skb_network_header_len(skb));
2689 skb_copy_from_linear_data(skb, nskb->data, doffset);
2690
2691 if (fskb != skb_shinfo(skb)->frag_list)
2692 continue;
2693
2694 if (!sg) {
2695 nskb->ip_summed = CHECKSUM_NONE;
2696 nskb->csum = skb_copy_and_csum_bits(skb, offset,
2697 skb_put(nskb, len),
2698 len, 0);
2699 continue;
2700 }
2701
2702 frag = skb_shinfo(nskb)->frags;
2703
2704 skb_copy_from_linear_data_offset(skb, offset,
2705 skb_put(nskb, hsize), hsize);
2706
2707 while (pos < offset + len && i < nfrags) {
2708 *frag = skb_shinfo(skb)->frags[i];
2709 get_page(frag->page);
2710 size = frag->size;
2711
2712 if (pos < offset) {
2713 frag->page_offset += offset - pos;
2714 frag->size -= offset - pos;
2715 }
2716
2717 skb_shinfo(nskb)->nr_frags++;
2718
2719 if (pos + size <= offset + len) {
2720 i++;
2721 pos += size;
2722 } else {
2723 frag->size -= pos + size - (offset + len);
2724 goto skip_fraglist;
2725 }
2726
2727 frag++;
2728 }
2729
2730 if (pos < offset + len) {
2731 struct sk_buff *fskb2 = fskb;
2732
2733 BUG_ON(pos + fskb->len != offset + len);
2734
2735 pos += fskb->len;
2736 fskb = fskb->next;
2737
2738 if (fskb2->next) {
2739 fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2740 if (!fskb2)
2741 goto err;
2742 } else
2743 skb_get(fskb2);
2744
2745 SKB_FRAG_ASSERT(nskb);
2746 skb_shinfo(nskb)->frag_list = fskb2;
2747 }
2748
2749skip_fraglist:
2750 nskb->data_len = len - hsize;
2751 nskb->len += nskb->data_len;
2752 nskb->truesize += nskb->data_len;
2753 } while ((offset += len) < skb->len);
2754
2755 return segs;
2756
2757err:
2758 while ((skb = segs)) {
2759 segs = skb->next;
2760 kfree_skb(skb);
2761 }
2762 return ERR_PTR(err);
2763}
2764EXPORT_SYMBOL_GPL(skb_segment);
2765
2766int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2767{
2768 struct sk_buff *p = *head;
2769 struct sk_buff *nskb;
2770 struct skb_shared_info *skbinfo = skb_shinfo(skb);
2771 struct skb_shared_info *pinfo = skb_shinfo(p);
2772 unsigned int headroom;
2773 unsigned int len = skb_gro_len(skb);
2774 unsigned int offset = skb_gro_offset(skb);
2775 unsigned int headlen = skb_headlen(skb);
2776
2777 if (p->len + len >= 65536)
2778 return -E2BIG;
2779
2780 if (pinfo->frag_list)
2781 goto merge;
2782 else if (headlen <= offset) {
2783 skb_frag_t *frag;
2784 skb_frag_t *frag2;
2785 int i = skbinfo->nr_frags;
2786 int nr_frags = pinfo->nr_frags + i;
2787
2788 offset -= headlen;
2789
2790 if (nr_frags > MAX_SKB_FRAGS)
2791 return -E2BIG;
2792
2793 pinfo->nr_frags = nr_frags;
2794 skbinfo->nr_frags = 0;
2795
2796 frag = pinfo->frags + nr_frags;
2797 frag2 = skbinfo->frags + i;
2798 do {
2799 *--frag = *--frag2;
2800 } while (--i);
2801
2802 frag->page_offset += offset;
2803 frag->size -= offset;
2804
2805 skb->truesize -= skb->data_len;
2806 skb->len -= skb->data_len;
2807 skb->data_len = 0;
2808
2809 NAPI_GRO_CB(skb)->free = 1;
2810 goto done;
2811 } else if (skb_gro_len(p) != pinfo->gso_size)
2812 return -E2BIG;
2813
2814 headroom = skb_headroom(p);
2815 nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
2816 if (unlikely(!nskb))
2817 return -ENOMEM;
2818
2819 __copy_skb_header(nskb, p);
2820 nskb->mac_len = p->mac_len;
2821
2822 skb_reserve(nskb, headroom);
2823 __skb_put(nskb, skb_gro_offset(p));
2824
2825 skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
2826 skb_set_network_header(nskb, skb_network_offset(p));
2827 skb_set_transport_header(nskb, skb_transport_offset(p));
2828
2829 __skb_pull(p, skb_gro_offset(p));
2830 memcpy(skb_mac_header(nskb), skb_mac_header(p),
2831 p->data - skb_mac_header(p));
2832
2833 *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
2834 skb_shinfo(nskb)->frag_list = p;
2835 skb_shinfo(nskb)->gso_size = pinfo->gso_size;
2836 pinfo->gso_size = 0;
2837 skb_header_release(p);
2838 nskb->prev = p;
2839
2840 nskb->data_len += p->len;
2841 nskb->truesize += p->len;
2842 nskb->len += p->len;
2843
2844 *head = nskb;
2845 nskb->next = p->next;
2846 p->next = NULL;
2847
2848 p = nskb;
2849
2850merge:
2851 if (offset > headlen) {
2852 unsigned int eat = offset - headlen;
2853
2854 skbinfo->frags[0].page_offset += eat;
2855 skbinfo->frags[0].size -= eat;
2856 skb->data_len -= eat;
2857 skb->len -= eat;
2858 offset = headlen;
2859 }
2860
2861 __skb_pull(skb, offset);
2862
2863 p->prev->next = skb;
2864 p->prev = skb;
2865 skb_header_release(skb);
2866
2867done:
2868 NAPI_GRO_CB(p)->count++;
2869 p->data_len += len;
2870 p->truesize += len;
2871 p->len += len;
2872
2873 NAPI_GRO_CB(skb)->same_flow = 1;
2874 return 0;
2875}
2876EXPORT_SYMBOL_GPL(skb_gro_receive);
2877
2878void __init skb_init(void)
2879{
2880 skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2881 sizeof(struct sk_buff),
2882 0,
2883 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2884 NULL);
2885 skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2886 (2*sizeof(struct sk_buff)) +
2887 sizeof(atomic_t),
2888 0,
2889 SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2890 NULL);
2891}
2892
2893/**
2894 * skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2895 * @skb: Socket buffer containing the buffers to be mapped
2896 * @sg: The scatter-gather list to map into
2897 * @offset: The offset into the buffer's contents to start mapping
2898 * @len: Length of buffer space to be mapped
2899 *
2900 * Fill the specified scatter-gather list with mappings/pointers into a
2901 * region of the buffer space attached to a socket buffer.
2902 */
2903static int
2904__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2905{
2906 int start = skb_headlen(skb);
2907 int i, copy = start - offset;
2908 struct sk_buff *frag_iter;
2909 int elt = 0;
2910
2911 if (copy > 0) {
2912 if (copy > len)
2913 copy = len;
2914 sg_set_buf(sg, skb->data + offset, copy);
2915 elt++;
2916 if ((len -= copy) == 0)
2917 return elt;
2918 offset += copy;
2919 }
2920
2921 for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2922 int end;
2923
2924 WARN_ON(start > offset + len);
2925
2926 end = start + skb_shinfo(skb)->frags[i].size;
2927 if ((copy = end - offset) > 0) {
2928 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2929
2930 if (copy > len)
2931 copy = len;
2932 sg_set_page(&sg[elt], frag->page, copy,
2933 frag->page_offset+offset-start);
2934 elt++;
2935 if (!(len -= copy))
2936 return elt;
2937 offset += copy;
2938 }
2939 start = end;
2940 }
2941
2942 skb_walk_frags(skb, frag_iter) {
2943 int end;
2944
2945 WARN_ON(start > offset + len);
2946
2947 end = start + frag_iter->len;
2948 if ((copy = end - offset) > 0) {
2949 if (copy > len)
2950 copy = len;
2951 elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
2952 copy);
2953 if ((len -= copy) == 0)
2954 return elt;
2955 offset += copy;
2956 }
2957 start = end;
2958 }
2959 BUG_ON(len);
2960 return elt;
2961}
2962
2963int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2964{
2965 int nsg = __skb_to_sgvec(skb, sg, offset, len);
2966
2967 sg_mark_end(&sg[nsg - 1]);
2968
2969 return nsg;
2970}
2971EXPORT_SYMBOL_GPL(skb_to_sgvec);
2972
2973/**
2974 * skb_cow_data - Check that a socket buffer's data buffers are writable
2975 * @skb: The socket buffer to check.
2976 * @tailbits: Amount of trailing space to be added
2977 * @trailer: Returned pointer to the skb where the @tailbits space begins
2978 *
2979 * Make sure that the data buffers attached to a socket buffer are
2980 * writable. If they are not, private copies are made of the data buffers
2981 * and the socket buffer is set to use these instead.
2982 *
2983 * If @tailbits is given, make sure that there is space to write @tailbits
2984 * bytes of data beyond current end of socket buffer. @trailer will be
2985 * set to point to the skb in which this space begins.
2986 *
2987 * The number of scatterlist elements required to completely map the
2988 * COW'd and extended socket buffer will be returned.
2989 */
2990int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2991{
2992 int copyflag;
2993 int elt;
2994 struct sk_buff *skb1, **skb_p;
2995
2996 /* If skb is cloned or its head is paged, reallocate
2997 * head pulling out all the pages (pages are considered not writable
2998 * at the moment even if they are anonymous).
2999 */
3000 if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3001 __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3002 return -ENOMEM;
3003
3004 /* Easy case. Most of packets will go this way. */
3005 if (!skb_has_frag_list(skb)) {
3006 /* A little of trouble, not enough of space for trailer.
3007 * This should not happen, when stack is tuned to generate
3008 * good frames. OK, on miss we reallocate and reserve even more
3009 * space, 128 bytes is fair. */
3010
3011 if (skb_tailroom(skb) < tailbits &&
3012 pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3013 return -ENOMEM;
3014
3015 /* Voila! */
3016 *trailer = skb;
3017 return 1;
3018 }
3019
3020 /* Misery. We are in troubles, going to mincer fragments... */
3021
3022 elt = 1;
3023 skb_p = &skb_shinfo(skb)->frag_list;
3024 copyflag = 0;
3025
3026 while ((skb1 = *skb_p) != NULL) {
3027 int ntail = 0;
3028
3029 /* The fragment is partially pulled by someone,
3030 * this can happen on input. Copy it and everything
3031 * after it. */
3032
3033 if (skb_shared(skb1))
3034 copyflag = 1;
3035
3036 /* If the skb is the last, worry about trailer. */
3037
3038 if (skb1->next == NULL && tailbits) {
3039 if (skb_shinfo(skb1)->nr_frags ||
3040 skb_has_frag_list(skb1) ||
3041 skb_tailroom(skb1) < tailbits)
3042 ntail = tailbits + 128;
3043 }
3044
3045 if (copyflag ||
3046 skb_cloned(skb1) ||
3047 ntail ||
3048 skb_shinfo(skb1)->nr_frags ||
3049 skb_has_frag_list(skb1)) {
3050 struct sk_buff *skb2;
3051
3052 /* Fuck, we are miserable poor guys... */
3053 if (ntail == 0)
3054 skb2 = skb_copy(skb1, GFP_ATOMIC);
3055 else
3056 skb2 = skb_copy_expand(skb1,
3057 skb_headroom(skb1),
3058 ntail,
3059 GFP_ATOMIC);
3060 if (unlikely(skb2 == NULL))
3061 return -ENOMEM;
3062
3063 if (skb1->sk)
3064 skb_set_owner_w(skb2, skb1->sk);
3065
3066 /* Looking around. Are we still alive?
3067 * OK, link new skb, drop old one */
3068
3069 skb2->next = skb1->next;
3070 *skb_p = skb2;
3071 kfree_skb(skb1);
3072 skb1 = skb2;
3073 }
3074 elt++;
3075 *trailer = skb1;
3076 skb_p = &skb1->next;
3077 }
3078
3079 return elt;
3080}
3081EXPORT_SYMBOL_GPL(skb_cow_data);
3082
3083static void sock_rmem_free(struct sk_buff *skb)
3084{
3085 struct sock *sk = skb->sk;
3086
3087 atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3088}
3089
3090/*
3091 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3092 */
3093int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3094{
3095 if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3096 (unsigned)sk->sk_rcvbuf)
3097 return -ENOMEM;
3098
3099 skb_orphan(skb);
3100 skb->sk = sk;
3101 skb->destructor = sock_rmem_free;
3102 atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3103
3104 /* before exiting rcu section, make sure dst is refcounted */
3105 skb_dst_force(skb);
3106
3107 skb_queue_tail(&sk->sk_error_queue, skb);
3108 if (!sock_flag(sk, SOCK_DEAD))
3109 sk->sk_data_ready(sk, skb->len);
3110 return 0;
3111}
3112EXPORT_SYMBOL(sock_queue_err_skb);
3113
3114void skb_tstamp_tx(struct sk_buff *orig_skb,
3115 struct skb_shared_hwtstamps *hwtstamps)
3116{
3117 struct sock *sk = orig_skb->sk;
3118 struct sock_exterr_skb *serr;
3119 struct sk_buff *skb;
3120 int err;
3121
3122 if (!sk)
3123 return;
3124
3125 skb = skb_clone(orig_skb, GFP_ATOMIC);
3126 if (!skb)
3127 return;
3128
3129 if (hwtstamps) {
3130 *skb_hwtstamps(skb) =
3131 *hwtstamps;
3132 } else {
3133 /*
3134 * no hardware time stamps available,
3135 * so keep the shared tx_flags and only
3136 * store software time stamp
3137 */
3138 skb->tstamp = ktime_get_real();
3139 }
3140
3141 serr = SKB_EXT_ERR(skb);
3142 memset(serr, 0, sizeof(*serr));
3143 serr->ee.ee_errno = ENOMSG;
3144 serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3145
3146 err = sock_queue_err_skb(sk, skb);
3147
3148 if (err)
3149 kfree_skb(skb);
3150}
3151EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3152
3153
3154/**
3155 * skb_partial_csum_set - set up and verify partial csum values for packet
3156 * @skb: the skb to set
3157 * @start: the number of bytes after skb->data to start checksumming.
3158 * @off: the offset from start to place the checksum.
3159 *
3160 * For untrusted partially-checksummed packets, we need to make sure the values
3161 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3162 *
3163 * This function checks and sets those values and skb->ip_summed: if this
3164 * returns false you should drop the packet.
3165 */
3166bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3167{
3168 if (unlikely(start > skb_headlen(skb)) ||
3169 unlikely((int)start + off > skb_headlen(skb) - 2)) {
3170 if (net_ratelimit())
3171 printk(KERN_WARNING
3172 "bad partial csum: csum=%u/%u len=%u\n",
3173 start, off, skb_headlen(skb));
3174 return false;
3175 }
3176 skb->ip_summed = CHECKSUM_PARTIAL;
3177 skb->csum_start = skb_headroom(skb) + start;
3178 skb->csum_offset = off;
3179 return true;
3180}
3181EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3182
3183void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3184{
3185 if (net_ratelimit())
3186 pr_warning("%s: received packets cannot be forwarded"
3187 " while LRO is enabled\n", skb->dev->name);
3188}
3189EXPORT_SYMBOL(__skb_warn_lro_forwarding);