xref: /openbmc/linux/net/core/skbuff.c (revision e868d61272caa648214046a096e5a6bfc068dc8c)
1 /*
2  *	Routines having to do with the 'struct sk_buff' memory handlers.
3  *
4  *	Authors:	Alan Cox <iiitac@pyr.swan.ac.uk>
5  *			Florian La Roche <rzsfl@rz.uni-sb.de>
6  *
7  *	Version:	$Id: skbuff.c,v 1.90 2001/11/07 05:56:19 davem Exp $
8  *
9  *	Fixes:
10  *		Alan Cox	:	Fixed the worst of the load
11  *					balancer bugs.
12  *		Dave Platt	:	Interrupt stacking fix.
13  *	Richard Kooijman	:	Timestamp fixes.
14  *		Alan Cox	:	Changed buffer format.
15  *		Alan Cox	:	destructor hook for AF_UNIX etc.
16  *		Linus Torvalds	:	Better skb_clone.
17  *		Alan Cox	:	Added skb_copy.
18  *		Alan Cox	:	Added all the changed routines Linus
19  *					only put in the headers
20  *		Ray VanTassle	:	Fixed --skb->lock in free
21  *		Alan Cox	:	skb_copy copy arp field
22  *		Andi Kleen	:	slabified it.
23  *		Robert Olsson	:	Removed skb_head_pool
24  *
25  *	NOTE:
26  *		The __skb_ routines should be called with interrupts
27  *	disabled, or you better be *real* sure that the operation is atomic
28  *	with respect to whatever list is being frobbed (e.g. via lock_sock()
29  *	or via disabling bottom half handlers, etc).
30  *
31  *	This program is free software; you can redistribute it and/or
32  *	modify it under the terms of the GNU General Public License
33  *	as published by the Free Software Foundation; either version
34  *	2 of the License, or (at your option) any later version.
35  */
36 
37 /*
38  *	The functions in this file will not compile correctly with gcc 2.4.x
39  */
40 
41 #include <linux/module.h>
42 #include <linux/types.h>
43 #include <linux/kernel.h>
44 #include <linux/mm.h>
45 #include <linux/interrupt.h>
46 #include <linux/in.h>
47 #include <linux/inet.h>
48 #include <linux/slab.h>
49 #include <linux/netdevice.h>
50 #ifdef CONFIG_NET_CLS_ACT
51 #include <net/pkt_sched.h>
52 #endif
53 #include <linux/string.h>
54 #include <linux/skbuff.h>
55 #include <linux/cache.h>
56 #include <linux/rtnetlink.h>
57 #include <linux/init.h>
58 #include <linux/scatterlist.h>
59 
60 #include <net/protocol.h>
61 #include <net/dst.h>
62 #include <net/sock.h>
63 #include <net/checksum.h>
64 #include <net/xfrm.h>
65 
66 #include <asm/uaccess.h>
67 #include <asm/system.h>
68 
69 #include "kmap_skb.h"
70 
71 static struct kmem_cache *skbuff_head_cache __read_mostly;
72 static struct kmem_cache *skbuff_fclone_cache __read_mostly;
73 
74 /*
75  *	Keep out-of-line to prevent kernel bloat.
76  *	__builtin_return_address is not used because it is not always
77  *	reliable.
78  */
79 
80 /**
81  *	skb_over_panic	- 	private function
82  *	@skb: buffer
83  *	@sz: size
84  *	@here: address
85  *
86  *	Out of line support code for skb_put(). Not user callable.
87  */
88 void skb_over_panic(struct sk_buff *skb, int sz, void *here)
89 {
90 	printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
91 			  "data:%p tail:%#lx end:%#lx dev:%s\n",
92 	       here, skb->len, sz, skb->head, skb->data,
93 	       (unsigned long)skb->tail, (unsigned long)skb->end,
94 	       skb->dev ? skb->dev->name : "<NULL>");
95 	BUG();
96 }
97 
98 /**
99  *	skb_under_panic	- 	private function
100  *	@skb: buffer
101  *	@sz: size
102  *	@here: address
103  *
104  *	Out of line support code for skb_push(). Not user callable.
105  */
106 
107 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
108 {
109 	printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
110 			  "data:%p tail:%#lx end:%#lx dev:%s\n",
111 	       here, skb->len, sz, skb->head, skb->data,
112 	       (unsigned long)skb->tail, (unsigned long)skb->end,
113 	       skb->dev ? skb->dev->name : "<NULL>");
114 	BUG();
115 }
116 
117 void skb_truesize_bug(struct sk_buff *skb)
118 {
119 	printk(KERN_ERR "SKB BUG: Invalid truesize (%u) "
120 	       "len=%u, sizeof(sk_buff)=%Zd\n",
121 	       skb->truesize, skb->len, sizeof(struct sk_buff));
122 }
123 EXPORT_SYMBOL(skb_truesize_bug);
124 
125 /* 	Allocate a new skbuff. We do this ourselves so we can fill in a few
126  *	'private' fields and also do memory statistics to find all the
127  *	[BEEP] leaks.
128  *
129  */
130 
131 /**
132  *	__alloc_skb	-	allocate a network buffer
133  *	@size: size to allocate
134  *	@gfp_mask: allocation mask
135  *	@fclone: allocate from fclone cache instead of head cache
136  *		and allocate a cloned (child) skb
137  *	@node: numa node to allocate memory on
138  *
139  *	Allocate a new &sk_buff. The returned buffer has no headroom and a
140  *	tail room of size bytes. The object has a reference count of one.
141  *	The return is the buffer. On a failure the return is %NULL.
142  *
143  *	Buffers may only be allocated from interrupts using a @gfp_mask of
144  *	%GFP_ATOMIC.
145  */
146 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
147 			    int fclone, int node)
148 {
149 	struct kmem_cache *cache;
150 	struct skb_shared_info *shinfo;
151 	struct sk_buff *skb;
152 	u8 *data;
153 
154 	cache = fclone ? skbuff_fclone_cache : skbuff_head_cache;
155 
156 	/* Get the HEAD */
157 	skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
158 	if (!skb)
159 		goto out;
160 
161 	size = SKB_DATA_ALIGN(size);
162 	data = kmalloc_node_track_caller(size + sizeof(struct skb_shared_info),
163 			gfp_mask, node);
164 	if (!data)
165 		goto nodata;
166 
167 	/*
168 	 * See comment in sk_buff definition, just before the 'tail' member
169 	 */
170 	memset(skb, 0, offsetof(struct sk_buff, tail));
171 	skb->truesize = size + sizeof(struct sk_buff);
172 	atomic_set(&skb->users, 1);
173 	skb->head = data;
174 	skb->data = data;
175 	skb_reset_tail_pointer(skb);
176 	skb->end = skb->tail + size;
177 	/* make sure we initialize shinfo sequentially */
178 	shinfo = skb_shinfo(skb);
179 	atomic_set(&shinfo->dataref, 1);
180 	shinfo->nr_frags  = 0;
181 	shinfo->gso_size = 0;
182 	shinfo->gso_segs = 0;
183 	shinfo->gso_type = 0;
184 	shinfo->ip6_frag_id = 0;
185 	shinfo->frag_list = NULL;
186 
187 	if (fclone) {
188 		struct sk_buff *child = skb + 1;
189 		atomic_t *fclone_ref = (atomic_t *) (child + 1);
190 
191 		skb->fclone = SKB_FCLONE_ORIG;
192 		atomic_set(fclone_ref, 1);
193 
194 		child->fclone = SKB_FCLONE_UNAVAILABLE;
195 	}
196 out:
197 	return skb;
198 nodata:
199 	kmem_cache_free(cache, skb);
200 	skb = NULL;
201 	goto out;
202 }
203 
204 /**
205  *	__netdev_alloc_skb - allocate an skbuff for rx on a specific device
206  *	@dev: network device to receive on
207  *	@length: length to allocate
208  *	@gfp_mask: get_free_pages mask, passed to alloc_skb
209  *
210  *	Allocate a new &sk_buff and assign it a usage count of one. The
211  *	buffer has unspecified headroom built in. Users should allocate
212  *	the headroom they think they need without accounting for the
213  *	built in space. The built in space is used for optimisations.
214  *
215  *	%NULL is returned if there is no free memory.
216  */
217 struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
218 		unsigned int length, gfp_t gfp_mask)
219 {
220 	int node = dev->dev.parent ? dev_to_node(dev->dev.parent) : -1;
221 	struct sk_buff *skb;
222 
223 	skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask, 0, node);
224 	if (likely(skb)) {
225 		skb_reserve(skb, NET_SKB_PAD);
226 		skb->dev = dev;
227 	}
228 	return skb;
229 }
230 
231 static void skb_drop_list(struct sk_buff **listp)
232 {
233 	struct sk_buff *list = *listp;
234 
235 	*listp = NULL;
236 
237 	do {
238 		struct sk_buff *this = list;
239 		list = list->next;
240 		kfree_skb(this);
241 	} while (list);
242 }
243 
244 static inline void skb_drop_fraglist(struct sk_buff *skb)
245 {
246 	skb_drop_list(&skb_shinfo(skb)->frag_list);
247 }
248 
249 static void skb_clone_fraglist(struct sk_buff *skb)
250 {
251 	struct sk_buff *list;
252 
253 	for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
254 		skb_get(list);
255 }
256 
257 static void skb_release_data(struct sk_buff *skb)
258 {
259 	if (!skb->cloned ||
260 	    !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
261 			       &skb_shinfo(skb)->dataref)) {
262 		if (skb_shinfo(skb)->nr_frags) {
263 			int i;
264 			for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
265 				put_page(skb_shinfo(skb)->frags[i].page);
266 		}
267 
268 		if (skb_shinfo(skb)->frag_list)
269 			skb_drop_fraglist(skb);
270 
271 		kfree(skb->head);
272 	}
273 }
274 
275 /*
276  *	Free an skbuff by memory without cleaning the state.
277  */
278 void kfree_skbmem(struct sk_buff *skb)
279 {
280 	struct sk_buff *other;
281 	atomic_t *fclone_ref;
282 
283 	skb_release_data(skb);
284 	switch (skb->fclone) {
285 	case SKB_FCLONE_UNAVAILABLE:
286 		kmem_cache_free(skbuff_head_cache, skb);
287 		break;
288 
289 	case SKB_FCLONE_ORIG:
290 		fclone_ref = (atomic_t *) (skb + 2);
291 		if (atomic_dec_and_test(fclone_ref))
292 			kmem_cache_free(skbuff_fclone_cache, skb);
293 		break;
294 
295 	case SKB_FCLONE_CLONE:
296 		fclone_ref = (atomic_t *) (skb + 1);
297 		other = skb - 1;
298 
299 		/* The clone portion is available for
300 		 * fast-cloning again.
301 		 */
302 		skb->fclone = SKB_FCLONE_UNAVAILABLE;
303 
304 		if (atomic_dec_and_test(fclone_ref))
305 			kmem_cache_free(skbuff_fclone_cache, other);
306 		break;
307 	}
308 }
309 
310 /**
311  *	__kfree_skb - private function
312  *	@skb: buffer
313  *
314  *	Free an sk_buff. Release anything attached to the buffer.
315  *	Clean the state. This is an internal helper function. Users should
316  *	always call kfree_skb
317  */
318 
319 void __kfree_skb(struct sk_buff *skb)
320 {
321 	dst_release(skb->dst);
322 #ifdef CONFIG_XFRM
323 	secpath_put(skb->sp);
324 #endif
325 	if (skb->destructor) {
326 		WARN_ON(in_irq());
327 		skb->destructor(skb);
328 	}
329 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
330 	nf_conntrack_put(skb->nfct);
331 	nf_conntrack_put_reasm(skb->nfct_reasm);
332 #endif
333 #ifdef CONFIG_BRIDGE_NETFILTER
334 	nf_bridge_put(skb->nf_bridge);
335 #endif
336 /* XXX: IS this still necessary? - JHS */
337 #ifdef CONFIG_NET_SCHED
338 	skb->tc_index = 0;
339 #ifdef CONFIG_NET_CLS_ACT
340 	skb->tc_verd = 0;
341 #endif
342 #endif
343 
344 	kfree_skbmem(skb);
345 }
346 
347 /**
348  *	kfree_skb - free an sk_buff
349  *	@skb: buffer to free
350  *
351  *	Drop a reference to the buffer and free it if the usage count has
352  *	hit zero.
353  */
354 void kfree_skb(struct sk_buff *skb)
355 {
356 	if (unlikely(!skb))
357 		return;
358 	if (likely(atomic_read(&skb->users) == 1))
359 		smp_rmb();
360 	else if (likely(!atomic_dec_and_test(&skb->users)))
361 		return;
362 	__kfree_skb(skb);
363 }
364 
365 /**
366  *	skb_clone	-	duplicate an sk_buff
367  *	@skb: buffer to clone
368  *	@gfp_mask: allocation priority
369  *
370  *	Duplicate an &sk_buff. The new one is not owned by a socket. Both
371  *	copies share the same packet data but not structure. The new
372  *	buffer has a reference count of 1. If the allocation fails the
373  *	function returns %NULL otherwise the new buffer is returned.
374  *
375  *	If this function is called from an interrupt gfp_mask() must be
376  *	%GFP_ATOMIC.
377  */
378 
379 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
380 {
381 	struct sk_buff *n;
382 
383 	n = skb + 1;
384 	if (skb->fclone == SKB_FCLONE_ORIG &&
385 	    n->fclone == SKB_FCLONE_UNAVAILABLE) {
386 		atomic_t *fclone_ref = (atomic_t *) (n + 1);
387 		n->fclone = SKB_FCLONE_CLONE;
388 		atomic_inc(fclone_ref);
389 	} else {
390 		n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
391 		if (!n)
392 			return NULL;
393 		n->fclone = SKB_FCLONE_UNAVAILABLE;
394 	}
395 
396 #define C(x) n->x = skb->x
397 
398 	n->next = n->prev = NULL;
399 	n->sk = NULL;
400 	C(tstamp);
401 	C(dev);
402 	C(transport_header);
403 	C(network_header);
404 	C(mac_header);
405 	C(dst);
406 	dst_clone(skb->dst);
407 	C(sp);
408 #ifdef CONFIG_INET
409 	secpath_get(skb->sp);
410 #endif
411 	memcpy(n->cb, skb->cb, sizeof(skb->cb));
412 	C(len);
413 	C(data_len);
414 	C(mac_len);
415 	C(csum);
416 	C(local_df);
417 	n->cloned = 1;
418 	n->nohdr = 0;
419 	C(pkt_type);
420 	C(ip_summed);
421 	C(priority);
422 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
423 	C(ipvs_property);
424 #endif
425 	C(protocol);
426 	n->destructor = NULL;
427 	C(mark);
428 	__nf_copy(n, skb);
429 #ifdef CONFIG_NET_SCHED
430 	C(tc_index);
431 #ifdef CONFIG_NET_CLS_ACT
432 	n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
433 	n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
434 	n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
435 	C(iif);
436 #endif
437 	skb_copy_secmark(n, skb);
438 #endif
439 	C(truesize);
440 	atomic_set(&n->users, 1);
441 	C(head);
442 	C(data);
443 	C(tail);
444 	C(end);
445 
446 	atomic_inc(&(skb_shinfo(skb)->dataref));
447 	skb->cloned = 1;
448 
449 	return n;
450 }
451 
452 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
453 {
454 #ifndef NET_SKBUFF_DATA_USES_OFFSET
455 	/*
456 	 *	Shift between the two data areas in bytes
457 	 */
458 	unsigned long offset = new->data - old->data;
459 #endif
460 	new->sk		= NULL;
461 	new->dev	= old->dev;
462 	new->priority	= old->priority;
463 	new->protocol	= old->protocol;
464 	new->dst	= dst_clone(old->dst);
465 #ifdef CONFIG_INET
466 	new->sp		= secpath_get(old->sp);
467 #endif
468 	new->transport_header = old->transport_header;
469 	new->network_header   = old->network_header;
470 	new->mac_header	      = old->mac_header;
471 #ifndef NET_SKBUFF_DATA_USES_OFFSET
472 	/* {transport,network,mac}_header are relative to skb->head */
473 	new->transport_header += offset;
474 	new->network_header   += offset;
475 	new->mac_header	      += offset;
476 #endif
477 	memcpy(new->cb, old->cb, sizeof(old->cb));
478 	new->local_df	= old->local_df;
479 	new->fclone	= SKB_FCLONE_UNAVAILABLE;
480 	new->pkt_type	= old->pkt_type;
481 	new->tstamp	= old->tstamp;
482 	new->destructor = NULL;
483 	new->mark	= old->mark;
484 	__nf_copy(new, old);
485 #if defined(CONFIG_IP_VS) || defined(CONFIG_IP_VS_MODULE)
486 	new->ipvs_property = old->ipvs_property;
487 #endif
488 #ifdef CONFIG_NET_SCHED
489 #ifdef CONFIG_NET_CLS_ACT
490 	new->tc_verd = old->tc_verd;
491 #endif
492 	new->tc_index	= old->tc_index;
493 #endif
494 	skb_copy_secmark(new, old);
495 	atomic_set(&new->users, 1);
496 	skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
497 	skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
498 	skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
499 }
500 
501 /**
502  *	skb_copy	-	create private copy of an sk_buff
503  *	@skb: buffer to copy
504  *	@gfp_mask: allocation priority
505  *
506  *	Make a copy of both an &sk_buff and its data. This is used when the
507  *	caller wishes to modify the data and needs a private copy of the
508  *	data to alter. Returns %NULL on failure or the pointer to the buffer
509  *	on success. The returned buffer has a reference count of 1.
510  *
511  *	As by-product this function converts non-linear &sk_buff to linear
512  *	one, so that &sk_buff becomes completely private and caller is allowed
513  *	to modify all the data of returned buffer. This means that this
514  *	function is not recommended for use in circumstances when only
515  *	header is going to be modified. Use pskb_copy() instead.
516  */
517 
518 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
519 {
520 	int headerlen = skb->data - skb->head;
521 	/*
522 	 *	Allocate the copy buffer
523 	 */
524 	struct sk_buff *n;
525 #ifdef NET_SKBUFF_DATA_USES_OFFSET
526 	n = alloc_skb(skb->end + skb->data_len, gfp_mask);
527 #else
528 	n = alloc_skb(skb->end - skb->head + skb->data_len, gfp_mask);
529 #endif
530 	if (!n)
531 		return NULL;
532 
533 	/* Set the data pointer */
534 	skb_reserve(n, headerlen);
535 	/* Set the tail pointer and length */
536 	skb_put(n, skb->len);
537 	n->csum	     = skb->csum;
538 	n->ip_summed = skb->ip_summed;
539 
540 	if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
541 		BUG();
542 
543 	copy_skb_header(n, skb);
544 	return n;
545 }
546 
547 
548 /**
549  *	pskb_copy	-	create copy of an sk_buff with private head.
550  *	@skb: buffer to copy
551  *	@gfp_mask: allocation priority
552  *
553  *	Make a copy of both an &sk_buff and part of its data, located
554  *	in header. Fragmented data remain shared. This is used when
555  *	the caller wishes to modify only header of &sk_buff and needs
556  *	private copy of the header to alter. Returns %NULL on failure
557  *	or the pointer to the buffer on success.
558  *	The returned buffer has a reference count of 1.
559  */
560 
561 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
562 {
563 	/*
564 	 *	Allocate the copy buffer
565 	 */
566 	struct sk_buff *n;
567 #ifdef NET_SKBUFF_DATA_USES_OFFSET
568 	n = alloc_skb(skb->end, gfp_mask);
569 #else
570 	n = alloc_skb(skb->end - skb->head, gfp_mask);
571 #endif
572 	if (!n)
573 		goto out;
574 
575 	/* Set the data pointer */
576 	skb_reserve(n, skb->data - skb->head);
577 	/* Set the tail pointer and length */
578 	skb_put(n, skb_headlen(skb));
579 	/* Copy the bytes */
580 	skb_copy_from_linear_data(skb, n->data, n->len);
581 	n->csum	     = skb->csum;
582 	n->ip_summed = skb->ip_summed;
583 
584 	n->truesize += skb->data_len;
585 	n->data_len  = skb->data_len;
586 	n->len	     = skb->len;
587 
588 	if (skb_shinfo(skb)->nr_frags) {
589 		int i;
590 
591 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
592 			skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
593 			get_page(skb_shinfo(n)->frags[i].page);
594 		}
595 		skb_shinfo(n)->nr_frags = i;
596 	}
597 
598 	if (skb_shinfo(skb)->frag_list) {
599 		skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
600 		skb_clone_fraglist(n);
601 	}
602 
603 	copy_skb_header(n, skb);
604 out:
605 	return n;
606 }
607 
608 /**
609  *	pskb_expand_head - reallocate header of &sk_buff
610  *	@skb: buffer to reallocate
611  *	@nhead: room to add at head
612  *	@ntail: room to add at tail
613  *	@gfp_mask: allocation priority
614  *
615  *	Expands (or creates identical copy, if &nhead and &ntail are zero)
616  *	header of skb. &sk_buff itself is not changed. &sk_buff MUST have
617  *	reference count of 1. Returns zero in the case of success or error,
618  *	if expansion failed. In the last case, &sk_buff is not changed.
619  *
620  *	All the pointers pointing into skb header may change and must be
621  *	reloaded after call to this function.
622  */
623 
624 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
625 		     gfp_t gfp_mask)
626 {
627 	int i;
628 	u8 *data;
629 #ifdef NET_SKBUFF_DATA_USES_OFFSET
630 	int size = nhead + skb->end + ntail;
631 #else
632 	int size = nhead + (skb->end - skb->head) + ntail;
633 #endif
634 	long off;
635 
636 	if (skb_shared(skb))
637 		BUG();
638 
639 	size = SKB_DATA_ALIGN(size);
640 
641 	data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
642 	if (!data)
643 		goto nodata;
644 
645 	/* Copy only real data... and, alas, header. This should be
646 	 * optimized for the cases when header is void. */
647 	memcpy(data + nhead, skb->head,
648 #ifdef NET_SKBUFF_DATA_USES_OFFSET
649 		skb->tail);
650 #else
651 		skb->tail - skb->head);
652 #endif
653 	memcpy(data + size, skb_end_pointer(skb),
654 	       sizeof(struct skb_shared_info));
655 
656 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
657 		get_page(skb_shinfo(skb)->frags[i].page);
658 
659 	if (skb_shinfo(skb)->frag_list)
660 		skb_clone_fraglist(skb);
661 
662 	skb_release_data(skb);
663 
664 	off = (data + nhead) - skb->head;
665 
666 	skb->head     = data;
667 	skb->data    += off;
668 #ifdef NET_SKBUFF_DATA_USES_OFFSET
669 	skb->end      = size;
670 	off           = nhead;
671 #else
672 	skb->end      = skb->head + size;
673 #endif
674 	/* {transport,network,mac}_header and tail are relative to skb->head */
675 	skb->tail	      += off;
676 	skb->transport_header += off;
677 	skb->network_header   += off;
678 	skb->mac_header	      += off;
679 	skb->cloned   = 0;
680 	skb->nohdr    = 0;
681 	atomic_set(&skb_shinfo(skb)->dataref, 1);
682 	return 0;
683 
684 nodata:
685 	return -ENOMEM;
686 }
687 
688 /* Make private copy of skb with writable head and some headroom */
689 
690 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
691 {
692 	struct sk_buff *skb2;
693 	int delta = headroom - skb_headroom(skb);
694 
695 	if (delta <= 0)
696 		skb2 = pskb_copy(skb, GFP_ATOMIC);
697 	else {
698 		skb2 = skb_clone(skb, GFP_ATOMIC);
699 		if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
700 					     GFP_ATOMIC)) {
701 			kfree_skb(skb2);
702 			skb2 = NULL;
703 		}
704 	}
705 	return skb2;
706 }
707 
708 
709 /**
710  *	skb_copy_expand	-	copy and expand sk_buff
711  *	@skb: buffer to copy
712  *	@newheadroom: new free bytes at head
713  *	@newtailroom: new free bytes at tail
714  *	@gfp_mask: allocation priority
715  *
716  *	Make a copy of both an &sk_buff and its data and while doing so
717  *	allocate additional space.
718  *
719  *	This is used when the caller wishes to modify the data and needs a
720  *	private copy of the data to alter as well as more space for new fields.
721  *	Returns %NULL on failure or the pointer to the buffer
722  *	on success. The returned buffer has a reference count of 1.
723  *
724  *	You must pass %GFP_ATOMIC as the allocation priority if this function
725  *	is called from an interrupt.
726  *
727  *	BUG ALERT: ip_summed is not copied. Why does this work? Is it used
728  *	only by netfilter in the cases when checksum is recalculated? --ANK
729  */
730 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
731 				int newheadroom, int newtailroom,
732 				gfp_t gfp_mask)
733 {
734 	/*
735 	 *	Allocate the copy buffer
736 	 */
737 	struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
738 				      gfp_mask);
739 	int oldheadroom = skb_headroom(skb);
740 	int head_copy_len, head_copy_off;
741 	int off = 0;
742 
743 	if (!n)
744 		return NULL;
745 
746 	skb_reserve(n, newheadroom);
747 
748 	/* Set the tail pointer and length */
749 	skb_put(n, skb->len);
750 
751 	head_copy_len = oldheadroom;
752 	head_copy_off = 0;
753 	if (newheadroom <= head_copy_len)
754 		head_copy_len = newheadroom;
755 	else
756 		head_copy_off = newheadroom - head_copy_len;
757 
758 	/* Copy the linear header and data. */
759 	if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
760 			  skb->len + head_copy_len))
761 		BUG();
762 
763 	copy_skb_header(n, skb);
764 
765 #ifdef NET_SKBUFF_DATA_USES_OFFSET
766 	off                  = newheadroom - oldheadroom;
767 #endif
768 	n->transport_header += off;
769 	n->network_header   += off;
770 	n->mac_header	    += off;
771 
772 	return n;
773 }
774 
775 /**
776  *	skb_pad			-	zero pad the tail of an skb
777  *	@skb: buffer to pad
778  *	@pad: space to pad
779  *
780  *	Ensure that a buffer is followed by a padding area that is zero
781  *	filled. Used by network drivers which may DMA or transfer data
782  *	beyond the buffer end onto the wire.
783  *
784  *	May return error in out of memory cases. The skb is freed on error.
785  */
786 
787 int skb_pad(struct sk_buff *skb, int pad)
788 {
789 	int err;
790 	int ntail;
791 
792 	/* If the skbuff is non linear tailroom is always zero.. */
793 	if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
794 		memset(skb->data+skb->len, 0, pad);
795 		return 0;
796 	}
797 
798 	ntail = skb->data_len + pad - (skb->end - skb->tail);
799 	if (likely(skb_cloned(skb) || ntail > 0)) {
800 		err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
801 		if (unlikely(err))
802 			goto free_skb;
803 	}
804 
805 	/* FIXME: The use of this function with non-linear skb's really needs
806 	 * to be audited.
807 	 */
808 	err = skb_linearize(skb);
809 	if (unlikely(err))
810 		goto free_skb;
811 
812 	memset(skb->data + skb->len, 0, pad);
813 	return 0;
814 
815 free_skb:
816 	kfree_skb(skb);
817 	return err;
818 }
819 
820 /* Trims skb to length len. It can change skb pointers.
821  */
822 
823 int ___pskb_trim(struct sk_buff *skb, unsigned int len)
824 {
825 	struct sk_buff **fragp;
826 	struct sk_buff *frag;
827 	int offset = skb_headlen(skb);
828 	int nfrags = skb_shinfo(skb)->nr_frags;
829 	int i;
830 	int err;
831 
832 	if (skb_cloned(skb) &&
833 	    unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
834 		return err;
835 
836 	i = 0;
837 	if (offset >= len)
838 		goto drop_pages;
839 
840 	for (; i < nfrags; i++) {
841 		int end = offset + skb_shinfo(skb)->frags[i].size;
842 
843 		if (end < len) {
844 			offset = end;
845 			continue;
846 		}
847 
848 		skb_shinfo(skb)->frags[i++].size = len - offset;
849 
850 drop_pages:
851 		skb_shinfo(skb)->nr_frags = i;
852 
853 		for (; i < nfrags; i++)
854 			put_page(skb_shinfo(skb)->frags[i].page);
855 
856 		if (skb_shinfo(skb)->frag_list)
857 			skb_drop_fraglist(skb);
858 		goto done;
859 	}
860 
861 	for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
862 	     fragp = &frag->next) {
863 		int end = offset + frag->len;
864 
865 		if (skb_shared(frag)) {
866 			struct sk_buff *nfrag;
867 
868 			nfrag = skb_clone(frag, GFP_ATOMIC);
869 			if (unlikely(!nfrag))
870 				return -ENOMEM;
871 
872 			nfrag->next = frag->next;
873 			kfree_skb(frag);
874 			frag = nfrag;
875 			*fragp = frag;
876 		}
877 
878 		if (end < len) {
879 			offset = end;
880 			continue;
881 		}
882 
883 		if (end > len &&
884 		    unlikely((err = pskb_trim(frag, len - offset))))
885 			return err;
886 
887 		if (frag->next)
888 			skb_drop_list(&frag->next);
889 		break;
890 	}
891 
892 done:
893 	if (len > skb_headlen(skb)) {
894 		skb->data_len -= skb->len - len;
895 		skb->len       = len;
896 	} else {
897 		skb->len       = len;
898 		skb->data_len  = 0;
899 		skb_set_tail_pointer(skb, len);
900 	}
901 
902 	return 0;
903 }
904 
905 /**
906  *	__pskb_pull_tail - advance tail of skb header
907  *	@skb: buffer to reallocate
908  *	@delta: number of bytes to advance tail
909  *
910  *	The function makes a sense only on a fragmented &sk_buff,
911  *	it expands header moving its tail forward and copying necessary
912  *	data from fragmented part.
913  *
914  *	&sk_buff MUST have reference count of 1.
915  *
916  *	Returns %NULL (and &sk_buff does not change) if pull failed
917  *	or value of new tail of skb in the case of success.
918  *
919  *	All the pointers pointing into skb header may change and must be
920  *	reloaded after call to this function.
921  */
922 
923 /* Moves tail of skb head forward, copying data from fragmented part,
924  * when it is necessary.
925  * 1. It may fail due to malloc failure.
926  * 2. It may change skb pointers.
927  *
928  * It is pretty complicated. Luckily, it is called only in exceptional cases.
929  */
930 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
931 {
932 	/* If skb has not enough free space at tail, get new one
933 	 * plus 128 bytes for future expansions. If we have enough
934 	 * room at tail, reallocate without expansion only if skb is cloned.
935 	 */
936 	int i, k, eat = (skb->tail + delta) - skb->end;
937 
938 	if (eat > 0 || skb_cloned(skb)) {
939 		if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
940 				     GFP_ATOMIC))
941 			return NULL;
942 	}
943 
944 	if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
945 		BUG();
946 
947 	/* Optimization: no fragments, no reasons to preestimate
948 	 * size of pulled pages. Superb.
949 	 */
950 	if (!skb_shinfo(skb)->frag_list)
951 		goto pull_pages;
952 
953 	/* Estimate size of pulled pages. */
954 	eat = delta;
955 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
956 		if (skb_shinfo(skb)->frags[i].size >= eat)
957 			goto pull_pages;
958 		eat -= skb_shinfo(skb)->frags[i].size;
959 	}
960 
961 	/* If we need update frag list, we are in troubles.
962 	 * Certainly, it possible to add an offset to skb data,
963 	 * but taking into account that pulling is expected to
964 	 * be very rare operation, it is worth to fight against
965 	 * further bloating skb head and crucify ourselves here instead.
966 	 * Pure masohism, indeed. 8)8)
967 	 */
968 	if (eat) {
969 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
970 		struct sk_buff *clone = NULL;
971 		struct sk_buff *insp = NULL;
972 
973 		do {
974 			BUG_ON(!list);
975 
976 			if (list->len <= eat) {
977 				/* Eaten as whole. */
978 				eat -= list->len;
979 				list = list->next;
980 				insp = list;
981 			} else {
982 				/* Eaten partially. */
983 
984 				if (skb_shared(list)) {
985 					/* Sucks! We need to fork list. :-( */
986 					clone = skb_clone(list, GFP_ATOMIC);
987 					if (!clone)
988 						return NULL;
989 					insp = list->next;
990 					list = clone;
991 				} else {
992 					/* This may be pulled without
993 					 * problems. */
994 					insp = list;
995 				}
996 				if (!pskb_pull(list, eat)) {
997 					if (clone)
998 						kfree_skb(clone);
999 					return NULL;
1000 				}
1001 				break;
1002 			}
1003 		} while (eat);
1004 
1005 		/* Free pulled out fragments. */
1006 		while ((list = skb_shinfo(skb)->frag_list) != insp) {
1007 			skb_shinfo(skb)->frag_list = list->next;
1008 			kfree_skb(list);
1009 		}
1010 		/* And insert new clone at head. */
1011 		if (clone) {
1012 			clone->next = list;
1013 			skb_shinfo(skb)->frag_list = clone;
1014 		}
1015 	}
1016 	/* Success! Now we may commit changes to skb data. */
1017 
1018 pull_pages:
1019 	eat = delta;
1020 	k = 0;
1021 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1022 		if (skb_shinfo(skb)->frags[i].size <= eat) {
1023 			put_page(skb_shinfo(skb)->frags[i].page);
1024 			eat -= skb_shinfo(skb)->frags[i].size;
1025 		} else {
1026 			skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1027 			if (eat) {
1028 				skb_shinfo(skb)->frags[k].page_offset += eat;
1029 				skb_shinfo(skb)->frags[k].size -= eat;
1030 				eat = 0;
1031 			}
1032 			k++;
1033 		}
1034 	}
1035 	skb_shinfo(skb)->nr_frags = k;
1036 
1037 	skb->tail     += delta;
1038 	skb->data_len -= delta;
1039 
1040 	return skb_tail_pointer(skb);
1041 }
1042 
1043 /* Copy some data bits from skb to kernel buffer. */
1044 
1045 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1046 {
1047 	int i, copy;
1048 	int start = skb_headlen(skb);
1049 
1050 	if (offset > (int)skb->len - len)
1051 		goto fault;
1052 
1053 	/* Copy header. */
1054 	if ((copy = start - offset) > 0) {
1055 		if (copy > len)
1056 			copy = len;
1057 		skb_copy_from_linear_data_offset(skb, offset, to, copy);
1058 		if ((len -= copy) == 0)
1059 			return 0;
1060 		offset += copy;
1061 		to     += copy;
1062 	}
1063 
1064 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1065 		int end;
1066 
1067 		BUG_TRAP(start <= offset + len);
1068 
1069 		end = start + skb_shinfo(skb)->frags[i].size;
1070 		if ((copy = end - offset) > 0) {
1071 			u8 *vaddr;
1072 
1073 			if (copy > len)
1074 				copy = len;
1075 
1076 			vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
1077 			memcpy(to,
1078 			       vaddr + skb_shinfo(skb)->frags[i].page_offset+
1079 			       offset - start, copy);
1080 			kunmap_skb_frag(vaddr);
1081 
1082 			if ((len -= copy) == 0)
1083 				return 0;
1084 			offset += copy;
1085 			to     += copy;
1086 		}
1087 		start = end;
1088 	}
1089 
1090 	if (skb_shinfo(skb)->frag_list) {
1091 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1092 
1093 		for (; list; list = list->next) {
1094 			int end;
1095 
1096 			BUG_TRAP(start <= offset + len);
1097 
1098 			end = start + list->len;
1099 			if ((copy = end - offset) > 0) {
1100 				if (copy > len)
1101 					copy = len;
1102 				if (skb_copy_bits(list, offset - start,
1103 						  to, copy))
1104 					goto fault;
1105 				if ((len -= copy) == 0)
1106 					return 0;
1107 				offset += copy;
1108 				to     += copy;
1109 			}
1110 			start = end;
1111 		}
1112 	}
1113 	if (!len)
1114 		return 0;
1115 
1116 fault:
1117 	return -EFAULT;
1118 }
1119 
1120 /**
1121  *	skb_store_bits - store bits from kernel buffer to skb
1122  *	@skb: destination buffer
1123  *	@offset: offset in destination
1124  *	@from: source buffer
1125  *	@len: number of bytes to copy
1126  *
1127  *	Copy the specified number of bytes from the source buffer to the
1128  *	destination skb.  This function handles all the messy bits of
1129  *	traversing fragment lists and such.
1130  */
1131 
1132 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1133 {
1134 	int i, copy;
1135 	int start = skb_headlen(skb);
1136 
1137 	if (offset > (int)skb->len - len)
1138 		goto fault;
1139 
1140 	if ((copy = start - offset) > 0) {
1141 		if (copy > len)
1142 			copy = len;
1143 		skb_copy_to_linear_data_offset(skb, offset, from, copy);
1144 		if ((len -= copy) == 0)
1145 			return 0;
1146 		offset += copy;
1147 		from += copy;
1148 	}
1149 
1150 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1151 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1152 		int end;
1153 
1154 		BUG_TRAP(start <= offset + len);
1155 
1156 		end = start + frag->size;
1157 		if ((copy = end - offset) > 0) {
1158 			u8 *vaddr;
1159 
1160 			if (copy > len)
1161 				copy = len;
1162 
1163 			vaddr = kmap_skb_frag(frag);
1164 			memcpy(vaddr + frag->page_offset + offset - start,
1165 			       from, copy);
1166 			kunmap_skb_frag(vaddr);
1167 
1168 			if ((len -= copy) == 0)
1169 				return 0;
1170 			offset += copy;
1171 			from += copy;
1172 		}
1173 		start = end;
1174 	}
1175 
1176 	if (skb_shinfo(skb)->frag_list) {
1177 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1178 
1179 		for (; list; list = list->next) {
1180 			int end;
1181 
1182 			BUG_TRAP(start <= offset + len);
1183 
1184 			end = start + list->len;
1185 			if ((copy = end - offset) > 0) {
1186 				if (copy > len)
1187 					copy = len;
1188 				if (skb_store_bits(list, offset - start,
1189 						   from, copy))
1190 					goto fault;
1191 				if ((len -= copy) == 0)
1192 					return 0;
1193 				offset += copy;
1194 				from += copy;
1195 			}
1196 			start = end;
1197 		}
1198 	}
1199 	if (!len)
1200 		return 0;
1201 
1202 fault:
1203 	return -EFAULT;
1204 }
1205 
1206 EXPORT_SYMBOL(skb_store_bits);
1207 
1208 /* Checksum skb data. */
1209 
1210 __wsum skb_checksum(const struct sk_buff *skb, int offset,
1211 			  int len, __wsum csum)
1212 {
1213 	int start = skb_headlen(skb);
1214 	int i, copy = start - offset;
1215 	int pos = 0;
1216 
1217 	/* Checksum header. */
1218 	if (copy > 0) {
1219 		if (copy > len)
1220 			copy = len;
1221 		csum = csum_partial(skb->data + offset, copy, csum);
1222 		if ((len -= copy) == 0)
1223 			return csum;
1224 		offset += copy;
1225 		pos	= copy;
1226 	}
1227 
1228 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1229 		int end;
1230 
1231 		BUG_TRAP(start <= offset + len);
1232 
1233 		end = start + skb_shinfo(skb)->frags[i].size;
1234 		if ((copy = end - offset) > 0) {
1235 			__wsum csum2;
1236 			u8 *vaddr;
1237 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1238 
1239 			if (copy > len)
1240 				copy = len;
1241 			vaddr = kmap_skb_frag(frag);
1242 			csum2 = csum_partial(vaddr + frag->page_offset +
1243 					     offset - start, copy, 0);
1244 			kunmap_skb_frag(vaddr);
1245 			csum = csum_block_add(csum, csum2, pos);
1246 			if (!(len -= copy))
1247 				return csum;
1248 			offset += copy;
1249 			pos    += copy;
1250 		}
1251 		start = end;
1252 	}
1253 
1254 	if (skb_shinfo(skb)->frag_list) {
1255 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1256 
1257 		for (; list; list = list->next) {
1258 			int end;
1259 
1260 			BUG_TRAP(start <= offset + len);
1261 
1262 			end = start + list->len;
1263 			if ((copy = end - offset) > 0) {
1264 				__wsum csum2;
1265 				if (copy > len)
1266 					copy = len;
1267 				csum2 = skb_checksum(list, offset - start,
1268 						     copy, 0);
1269 				csum = csum_block_add(csum, csum2, pos);
1270 				if ((len -= copy) == 0)
1271 					return csum;
1272 				offset += copy;
1273 				pos    += copy;
1274 			}
1275 			start = end;
1276 		}
1277 	}
1278 	BUG_ON(len);
1279 
1280 	return csum;
1281 }
1282 
1283 /* Both of above in one bottle. */
1284 
1285 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1286 				    u8 *to, int len, __wsum csum)
1287 {
1288 	int start = skb_headlen(skb);
1289 	int i, copy = start - offset;
1290 	int pos = 0;
1291 
1292 	/* Copy header. */
1293 	if (copy > 0) {
1294 		if (copy > len)
1295 			copy = len;
1296 		csum = csum_partial_copy_nocheck(skb->data + offset, to,
1297 						 copy, csum);
1298 		if ((len -= copy) == 0)
1299 			return csum;
1300 		offset += copy;
1301 		to     += copy;
1302 		pos	= copy;
1303 	}
1304 
1305 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1306 		int end;
1307 
1308 		BUG_TRAP(start <= offset + len);
1309 
1310 		end = start + skb_shinfo(skb)->frags[i].size;
1311 		if ((copy = end - offset) > 0) {
1312 			__wsum csum2;
1313 			u8 *vaddr;
1314 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1315 
1316 			if (copy > len)
1317 				copy = len;
1318 			vaddr = kmap_skb_frag(frag);
1319 			csum2 = csum_partial_copy_nocheck(vaddr +
1320 							  frag->page_offset +
1321 							  offset - start, to,
1322 							  copy, 0);
1323 			kunmap_skb_frag(vaddr);
1324 			csum = csum_block_add(csum, csum2, pos);
1325 			if (!(len -= copy))
1326 				return csum;
1327 			offset += copy;
1328 			to     += copy;
1329 			pos    += copy;
1330 		}
1331 		start = end;
1332 	}
1333 
1334 	if (skb_shinfo(skb)->frag_list) {
1335 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1336 
1337 		for (; list; list = list->next) {
1338 			__wsum csum2;
1339 			int end;
1340 
1341 			BUG_TRAP(start <= offset + len);
1342 
1343 			end = start + list->len;
1344 			if ((copy = end - offset) > 0) {
1345 				if (copy > len)
1346 					copy = len;
1347 				csum2 = skb_copy_and_csum_bits(list,
1348 							       offset - start,
1349 							       to, copy, 0);
1350 				csum = csum_block_add(csum, csum2, pos);
1351 				if ((len -= copy) == 0)
1352 					return csum;
1353 				offset += copy;
1354 				to     += copy;
1355 				pos    += copy;
1356 			}
1357 			start = end;
1358 		}
1359 	}
1360 	BUG_ON(len);
1361 	return csum;
1362 }
1363 
1364 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1365 {
1366 	__wsum csum;
1367 	long csstart;
1368 
1369 	if (skb->ip_summed == CHECKSUM_PARTIAL)
1370 		csstart = skb->csum_start - skb_headroom(skb);
1371 	else
1372 		csstart = skb_headlen(skb);
1373 
1374 	BUG_ON(csstart > skb_headlen(skb));
1375 
1376 	skb_copy_from_linear_data(skb, to, csstart);
1377 
1378 	csum = 0;
1379 	if (csstart != skb->len)
1380 		csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1381 					      skb->len - csstart, 0);
1382 
1383 	if (skb->ip_summed == CHECKSUM_PARTIAL) {
1384 		long csstuff = csstart + skb->csum_offset;
1385 
1386 		*((__sum16 *)(to + csstuff)) = csum_fold(csum);
1387 	}
1388 }
1389 
1390 /**
1391  *	skb_dequeue - remove from the head of the queue
1392  *	@list: list to dequeue from
1393  *
1394  *	Remove the head of the list. The list lock is taken so the function
1395  *	may be used safely with other locking list functions. The head item is
1396  *	returned or %NULL if the list is empty.
1397  */
1398 
1399 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1400 {
1401 	unsigned long flags;
1402 	struct sk_buff *result;
1403 
1404 	spin_lock_irqsave(&list->lock, flags);
1405 	result = __skb_dequeue(list);
1406 	spin_unlock_irqrestore(&list->lock, flags);
1407 	return result;
1408 }
1409 
1410 /**
1411  *	skb_dequeue_tail - remove from the tail of the queue
1412  *	@list: list to dequeue from
1413  *
1414  *	Remove the tail of the list. The list lock is taken so the function
1415  *	may be used safely with other locking list functions. The tail item is
1416  *	returned or %NULL if the list is empty.
1417  */
1418 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1419 {
1420 	unsigned long flags;
1421 	struct sk_buff *result;
1422 
1423 	spin_lock_irqsave(&list->lock, flags);
1424 	result = __skb_dequeue_tail(list);
1425 	spin_unlock_irqrestore(&list->lock, flags);
1426 	return result;
1427 }
1428 
1429 /**
1430  *	skb_queue_purge - empty a list
1431  *	@list: list to empty
1432  *
1433  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
1434  *	the list and one reference dropped. This function takes the list
1435  *	lock and is atomic with respect to other list locking functions.
1436  */
1437 void skb_queue_purge(struct sk_buff_head *list)
1438 {
1439 	struct sk_buff *skb;
1440 	while ((skb = skb_dequeue(list)) != NULL)
1441 		kfree_skb(skb);
1442 }
1443 
1444 /**
1445  *	skb_queue_head - queue a buffer at the list head
1446  *	@list: list to use
1447  *	@newsk: buffer to queue
1448  *
1449  *	Queue a buffer at the start of the list. This function takes the
1450  *	list lock and can be used safely with other locking &sk_buff functions
1451  *	safely.
1452  *
1453  *	A buffer cannot be placed on two lists at the same time.
1454  */
1455 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1456 {
1457 	unsigned long flags;
1458 
1459 	spin_lock_irqsave(&list->lock, flags);
1460 	__skb_queue_head(list, newsk);
1461 	spin_unlock_irqrestore(&list->lock, flags);
1462 }
1463 
1464 /**
1465  *	skb_queue_tail - queue a buffer at the list tail
1466  *	@list: list to use
1467  *	@newsk: buffer to queue
1468  *
1469  *	Queue a buffer at the tail of the list. This function takes the
1470  *	list lock and can be used safely with other locking &sk_buff functions
1471  *	safely.
1472  *
1473  *	A buffer cannot be placed on two lists at the same time.
1474  */
1475 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1476 {
1477 	unsigned long flags;
1478 
1479 	spin_lock_irqsave(&list->lock, flags);
1480 	__skb_queue_tail(list, newsk);
1481 	spin_unlock_irqrestore(&list->lock, flags);
1482 }
1483 
1484 /**
1485  *	skb_unlink	-	remove a buffer from a list
1486  *	@skb: buffer to remove
1487  *	@list: list to use
1488  *
1489  *	Remove a packet from a list. The list locks are taken and this
1490  *	function is atomic with respect to other list locked calls
1491  *
1492  *	You must know what list the SKB is on.
1493  */
1494 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1495 {
1496 	unsigned long flags;
1497 
1498 	spin_lock_irqsave(&list->lock, flags);
1499 	__skb_unlink(skb, list);
1500 	spin_unlock_irqrestore(&list->lock, flags);
1501 }
1502 
1503 /**
1504  *	skb_append	-	append a buffer
1505  *	@old: buffer to insert after
1506  *	@newsk: buffer to insert
1507  *	@list: list to use
1508  *
1509  *	Place a packet after a given packet in a list. The list locks are taken
1510  *	and this function is atomic with respect to other list locked calls.
1511  *	A buffer cannot be placed on two lists at the same time.
1512  */
1513 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1514 {
1515 	unsigned long flags;
1516 
1517 	spin_lock_irqsave(&list->lock, flags);
1518 	__skb_append(old, newsk, list);
1519 	spin_unlock_irqrestore(&list->lock, flags);
1520 }
1521 
1522 
1523 /**
1524  *	skb_insert	-	insert a buffer
1525  *	@old: buffer to insert before
1526  *	@newsk: buffer to insert
1527  *	@list: list to use
1528  *
1529  *	Place a packet before a given packet in a list. The list locks are
1530  * 	taken and this function is atomic with respect to other list locked
1531  *	calls.
1532  *
1533  *	A buffer cannot be placed on two lists at the same time.
1534  */
1535 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1536 {
1537 	unsigned long flags;
1538 
1539 	spin_lock_irqsave(&list->lock, flags);
1540 	__skb_insert(newsk, old->prev, old, list);
1541 	spin_unlock_irqrestore(&list->lock, flags);
1542 }
1543 
1544 static inline void skb_split_inside_header(struct sk_buff *skb,
1545 					   struct sk_buff* skb1,
1546 					   const u32 len, const int pos)
1547 {
1548 	int i;
1549 
1550 	skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
1551 					 pos - len);
1552 	/* And move data appendix as is. */
1553 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1554 		skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1555 
1556 	skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1557 	skb_shinfo(skb)->nr_frags  = 0;
1558 	skb1->data_len		   = skb->data_len;
1559 	skb1->len		   += skb1->data_len;
1560 	skb->data_len		   = 0;
1561 	skb->len		   = len;
1562 	skb_set_tail_pointer(skb, len);
1563 }
1564 
1565 static inline void skb_split_no_header(struct sk_buff *skb,
1566 				       struct sk_buff* skb1,
1567 				       const u32 len, int pos)
1568 {
1569 	int i, k = 0;
1570 	const int nfrags = skb_shinfo(skb)->nr_frags;
1571 
1572 	skb_shinfo(skb)->nr_frags = 0;
1573 	skb1->len		  = skb1->data_len = skb->len - len;
1574 	skb->len		  = len;
1575 	skb->data_len		  = len - pos;
1576 
1577 	for (i = 0; i < nfrags; i++) {
1578 		int size = skb_shinfo(skb)->frags[i].size;
1579 
1580 		if (pos + size > len) {
1581 			skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1582 
1583 			if (pos < len) {
1584 				/* Split frag.
1585 				 * We have two variants in this case:
1586 				 * 1. Move all the frag to the second
1587 				 *    part, if it is possible. F.e.
1588 				 *    this approach is mandatory for TUX,
1589 				 *    where splitting is expensive.
1590 				 * 2. Split is accurately. We make this.
1591 				 */
1592 				get_page(skb_shinfo(skb)->frags[i].page);
1593 				skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1594 				skb_shinfo(skb1)->frags[0].size -= len - pos;
1595 				skb_shinfo(skb)->frags[i].size	= len - pos;
1596 				skb_shinfo(skb)->nr_frags++;
1597 			}
1598 			k++;
1599 		} else
1600 			skb_shinfo(skb)->nr_frags++;
1601 		pos += size;
1602 	}
1603 	skb_shinfo(skb1)->nr_frags = k;
1604 }
1605 
1606 /**
1607  * skb_split - Split fragmented skb to two parts at length len.
1608  * @skb: the buffer to split
1609  * @skb1: the buffer to receive the second part
1610  * @len: new length for skb
1611  */
1612 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1613 {
1614 	int pos = skb_headlen(skb);
1615 
1616 	if (len < pos)	/* Split line is inside header. */
1617 		skb_split_inside_header(skb, skb1, len, pos);
1618 	else		/* Second chunk has no header, nothing to copy. */
1619 		skb_split_no_header(skb, skb1, len, pos);
1620 }
1621 
1622 /**
1623  * skb_prepare_seq_read - Prepare a sequential read of skb data
1624  * @skb: the buffer to read
1625  * @from: lower offset of data to be read
1626  * @to: upper offset of data to be read
1627  * @st: state variable
1628  *
1629  * Initializes the specified state variable. Must be called before
1630  * invoking skb_seq_read() for the first time.
1631  */
1632 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1633 			  unsigned int to, struct skb_seq_state *st)
1634 {
1635 	st->lower_offset = from;
1636 	st->upper_offset = to;
1637 	st->root_skb = st->cur_skb = skb;
1638 	st->frag_idx = st->stepped_offset = 0;
1639 	st->frag_data = NULL;
1640 }
1641 
1642 /**
1643  * skb_seq_read - Sequentially read skb data
1644  * @consumed: number of bytes consumed by the caller so far
1645  * @data: destination pointer for data to be returned
1646  * @st: state variable
1647  *
1648  * Reads a block of skb data at &consumed relative to the
1649  * lower offset specified to skb_prepare_seq_read(). Assigns
1650  * the head of the data block to &data and returns the length
1651  * of the block or 0 if the end of the skb data or the upper
1652  * offset has been reached.
1653  *
1654  * The caller is not required to consume all of the data
1655  * returned, i.e. &consumed is typically set to the number
1656  * of bytes already consumed and the next call to
1657  * skb_seq_read() will return the remaining part of the block.
1658  *
1659  * Note: The size of each block of data returned can be arbitary,
1660  *       this limitation is the cost for zerocopy seqeuental
1661  *       reads of potentially non linear data.
1662  *
1663  * Note: Fragment lists within fragments are not implemented
1664  *       at the moment, state->root_skb could be replaced with
1665  *       a stack for this purpose.
1666  */
1667 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1668 			  struct skb_seq_state *st)
1669 {
1670 	unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1671 	skb_frag_t *frag;
1672 
1673 	if (unlikely(abs_offset >= st->upper_offset))
1674 		return 0;
1675 
1676 next_skb:
1677 	block_limit = skb_headlen(st->cur_skb);
1678 
1679 	if (abs_offset < block_limit) {
1680 		*data = st->cur_skb->data + abs_offset;
1681 		return block_limit - abs_offset;
1682 	}
1683 
1684 	if (st->frag_idx == 0 && !st->frag_data)
1685 		st->stepped_offset += skb_headlen(st->cur_skb);
1686 
1687 	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1688 		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1689 		block_limit = frag->size + st->stepped_offset;
1690 
1691 		if (abs_offset < block_limit) {
1692 			if (!st->frag_data)
1693 				st->frag_data = kmap_skb_frag(frag);
1694 
1695 			*data = (u8 *) st->frag_data + frag->page_offset +
1696 				(abs_offset - st->stepped_offset);
1697 
1698 			return block_limit - abs_offset;
1699 		}
1700 
1701 		if (st->frag_data) {
1702 			kunmap_skb_frag(st->frag_data);
1703 			st->frag_data = NULL;
1704 		}
1705 
1706 		st->frag_idx++;
1707 		st->stepped_offset += frag->size;
1708 	}
1709 
1710 	if (st->cur_skb->next) {
1711 		st->cur_skb = st->cur_skb->next;
1712 		st->frag_idx = 0;
1713 		goto next_skb;
1714 	} else if (st->root_skb == st->cur_skb &&
1715 		   skb_shinfo(st->root_skb)->frag_list) {
1716 		st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1717 		goto next_skb;
1718 	}
1719 
1720 	return 0;
1721 }
1722 
1723 /**
1724  * skb_abort_seq_read - Abort a sequential read of skb data
1725  * @st: state variable
1726  *
1727  * Must be called if skb_seq_read() was not called until it
1728  * returned 0.
1729  */
1730 void skb_abort_seq_read(struct skb_seq_state *st)
1731 {
1732 	if (st->frag_data)
1733 		kunmap_skb_frag(st->frag_data);
1734 }
1735 
1736 #define TS_SKB_CB(state)	((struct skb_seq_state *) &((state)->cb))
1737 
1738 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1739 					  struct ts_config *conf,
1740 					  struct ts_state *state)
1741 {
1742 	return skb_seq_read(offset, text, TS_SKB_CB(state));
1743 }
1744 
1745 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1746 {
1747 	skb_abort_seq_read(TS_SKB_CB(state));
1748 }
1749 
1750 /**
1751  * skb_find_text - Find a text pattern in skb data
1752  * @skb: the buffer to look in
1753  * @from: search offset
1754  * @to: search limit
1755  * @config: textsearch configuration
1756  * @state: uninitialized textsearch state variable
1757  *
1758  * Finds a pattern in the skb data according to the specified
1759  * textsearch configuration. Use textsearch_next() to retrieve
1760  * subsequent occurrences of the pattern. Returns the offset
1761  * to the first occurrence or UINT_MAX if no match was found.
1762  */
1763 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1764 			   unsigned int to, struct ts_config *config,
1765 			   struct ts_state *state)
1766 {
1767 	unsigned int ret;
1768 
1769 	config->get_next_block = skb_ts_get_next_block;
1770 	config->finish = skb_ts_finish;
1771 
1772 	skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1773 
1774 	ret = textsearch_find(config, state);
1775 	return (ret <= to - from ? ret : UINT_MAX);
1776 }
1777 
1778 /**
1779  * skb_append_datato_frags: - append the user data to a skb
1780  * @sk: sock  structure
1781  * @skb: skb structure to be appened with user data.
1782  * @getfrag: call back function to be used for getting the user data
1783  * @from: pointer to user message iov
1784  * @length: length of the iov message
1785  *
1786  * Description: This procedure append the user data in the fragment part
1787  * of the skb if any page alloc fails user this procedure returns  -ENOMEM
1788  */
1789 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1790 			int (*getfrag)(void *from, char *to, int offset,
1791 					int len, int odd, struct sk_buff *skb),
1792 			void *from, int length)
1793 {
1794 	int frg_cnt = 0;
1795 	skb_frag_t *frag = NULL;
1796 	struct page *page = NULL;
1797 	int copy, left;
1798 	int offset = 0;
1799 	int ret;
1800 
1801 	do {
1802 		/* Return error if we don't have space for new frag */
1803 		frg_cnt = skb_shinfo(skb)->nr_frags;
1804 		if (frg_cnt >= MAX_SKB_FRAGS)
1805 			return -EFAULT;
1806 
1807 		/* allocate a new page for next frag */
1808 		page = alloc_pages(sk->sk_allocation, 0);
1809 
1810 		/* If alloc_page fails just return failure and caller will
1811 		 * free previous allocated pages by doing kfree_skb()
1812 		 */
1813 		if (page == NULL)
1814 			return -ENOMEM;
1815 
1816 		/* initialize the next frag */
1817 		sk->sk_sndmsg_page = page;
1818 		sk->sk_sndmsg_off = 0;
1819 		skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
1820 		skb->truesize += PAGE_SIZE;
1821 		atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
1822 
1823 		/* get the new initialized frag */
1824 		frg_cnt = skb_shinfo(skb)->nr_frags;
1825 		frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
1826 
1827 		/* copy the user data to page */
1828 		left = PAGE_SIZE - frag->page_offset;
1829 		copy = (length > left)? left : length;
1830 
1831 		ret = getfrag(from, (page_address(frag->page) +
1832 			    frag->page_offset + frag->size),
1833 			    offset, copy, 0, skb);
1834 		if (ret < 0)
1835 			return -EFAULT;
1836 
1837 		/* copy was successful so update the size parameters */
1838 		sk->sk_sndmsg_off += copy;
1839 		frag->size += copy;
1840 		skb->len += copy;
1841 		skb->data_len += copy;
1842 		offset += copy;
1843 		length -= copy;
1844 
1845 	} while (length > 0);
1846 
1847 	return 0;
1848 }
1849 
1850 /**
1851  *	skb_pull_rcsum - pull skb and update receive checksum
1852  *	@skb: buffer to update
1853  *	@start: start of data before pull
1854  *	@len: length of data pulled
1855  *
1856  *	This function performs an skb_pull on the packet and updates
1857  *	update the CHECKSUM_COMPLETE checksum.  It should be used on
1858  *	receive path processing instead of skb_pull unless you know
1859  *	that the checksum difference is zero (e.g., a valid IP header)
1860  *	or you are setting ip_summed to CHECKSUM_NONE.
1861  */
1862 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
1863 {
1864 	BUG_ON(len > skb->len);
1865 	skb->len -= len;
1866 	BUG_ON(skb->len < skb->data_len);
1867 	skb_postpull_rcsum(skb, skb->data, len);
1868 	return skb->data += len;
1869 }
1870 
1871 EXPORT_SYMBOL_GPL(skb_pull_rcsum);
1872 
1873 /**
1874  *	skb_segment - Perform protocol segmentation on skb.
1875  *	@skb: buffer to segment
1876  *	@features: features for the output path (see dev->features)
1877  *
1878  *	This function performs segmentation on the given skb.  It returns
1879  *	the segment at the given position.  It returns NULL if there are
1880  *	no more segments to generate, or when an error is encountered.
1881  */
1882 struct sk_buff *skb_segment(struct sk_buff *skb, int features)
1883 {
1884 	struct sk_buff *segs = NULL;
1885 	struct sk_buff *tail = NULL;
1886 	unsigned int mss = skb_shinfo(skb)->gso_size;
1887 	unsigned int doffset = skb->data - skb_mac_header(skb);
1888 	unsigned int offset = doffset;
1889 	unsigned int headroom;
1890 	unsigned int len;
1891 	int sg = features & NETIF_F_SG;
1892 	int nfrags = skb_shinfo(skb)->nr_frags;
1893 	int err = -ENOMEM;
1894 	int i = 0;
1895 	int pos;
1896 
1897 	__skb_push(skb, doffset);
1898 	headroom = skb_headroom(skb);
1899 	pos = skb_headlen(skb);
1900 
1901 	do {
1902 		struct sk_buff *nskb;
1903 		skb_frag_t *frag;
1904 		int hsize;
1905 		int k;
1906 		int size;
1907 
1908 		len = skb->len - offset;
1909 		if (len > mss)
1910 			len = mss;
1911 
1912 		hsize = skb_headlen(skb) - offset;
1913 		if (hsize < 0)
1914 			hsize = 0;
1915 		if (hsize > len || !sg)
1916 			hsize = len;
1917 
1918 		nskb = alloc_skb(hsize + doffset + headroom, GFP_ATOMIC);
1919 		if (unlikely(!nskb))
1920 			goto err;
1921 
1922 		if (segs)
1923 			tail->next = nskb;
1924 		else
1925 			segs = nskb;
1926 		tail = nskb;
1927 
1928 		nskb->dev = skb->dev;
1929 		nskb->priority = skb->priority;
1930 		nskb->protocol = skb->protocol;
1931 		nskb->dst = dst_clone(skb->dst);
1932 		memcpy(nskb->cb, skb->cb, sizeof(skb->cb));
1933 		nskb->pkt_type = skb->pkt_type;
1934 		nskb->mac_len = skb->mac_len;
1935 
1936 		skb_reserve(nskb, headroom);
1937 		skb_reset_mac_header(nskb);
1938 		skb_set_network_header(nskb, skb->mac_len);
1939 		nskb->transport_header = (nskb->network_header +
1940 					  skb_network_header_len(skb));
1941 		skb_copy_from_linear_data(skb, skb_put(nskb, doffset),
1942 					  doffset);
1943 		if (!sg) {
1944 			nskb->csum = skb_copy_and_csum_bits(skb, offset,
1945 							    skb_put(nskb, len),
1946 							    len, 0);
1947 			continue;
1948 		}
1949 
1950 		frag = skb_shinfo(nskb)->frags;
1951 		k = 0;
1952 
1953 		nskb->ip_summed = CHECKSUM_PARTIAL;
1954 		nskb->csum = skb->csum;
1955 		skb_copy_from_linear_data_offset(skb, offset,
1956 						 skb_put(nskb, hsize), hsize);
1957 
1958 		while (pos < offset + len) {
1959 			BUG_ON(i >= nfrags);
1960 
1961 			*frag = skb_shinfo(skb)->frags[i];
1962 			get_page(frag->page);
1963 			size = frag->size;
1964 
1965 			if (pos < offset) {
1966 				frag->page_offset += offset - pos;
1967 				frag->size -= offset - pos;
1968 			}
1969 
1970 			k++;
1971 
1972 			if (pos + size <= offset + len) {
1973 				i++;
1974 				pos += size;
1975 			} else {
1976 				frag->size -= pos + size - (offset + len);
1977 				break;
1978 			}
1979 
1980 			frag++;
1981 		}
1982 
1983 		skb_shinfo(nskb)->nr_frags = k;
1984 		nskb->data_len = len - hsize;
1985 		nskb->len += nskb->data_len;
1986 		nskb->truesize += nskb->data_len;
1987 	} while ((offset += len) < skb->len);
1988 
1989 	return segs;
1990 
1991 err:
1992 	while ((skb = segs)) {
1993 		segs = skb->next;
1994 		kfree_skb(skb);
1995 	}
1996 	return ERR_PTR(err);
1997 }
1998 
1999 EXPORT_SYMBOL_GPL(skb_segment);
2000 
2001 void __init skb_init(void)
2002 {
2003 	skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
2004 					      sizeof(struct sk_buff),
2005 					      0,
2006 					      SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2007 					      NULL, NULL);
2008 	skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
2009 						(2*sizeof(struct sk_buff)) +
2010 						sizeof(atomic_t),
2011 						0,
2012 						SLAB_HWCACHE_ALIGN|SLAB_PANIC,
2013 						NULL, NULL);
2014 }
2015 
2016 /**
2017  *	skb_to_sgvec - Fill a scatter-gather list from a socket buffer
2018  *	@skb: Socket buffer containing the buffers to be mapped
2019  *	@sg: The scatter-gather list to map into
2020  *	@offset: The offset into the buffer's contents to start mapping
2021  *	@len: Length of buffer space to be mapped
2022  *
2023  *	Fill the specified scatter-gather list with mappings/pointers into a
2024  *	region of the buffer space attached to a socket buffer.
2025  */
2026 int
2027 skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
2028 {
2029 	int start = skb_headlen(skb);
2030 	int i, copy = start - offset;
2031 	int elt = 0;
2032 
2033 	if (copy > 0) {
2034 		if (copy > len)
2035 			copy = len;
2036 		sg[elt].page = virt_to_page(skb->data + offset);
2037 		sg[elt].offset = (unsigned long)(skb->data + offset) % PAGE_SIZE;
2038 		sg[elt].length = copy;
2039 		elt++;
2040 		if ((len -= copy) == 0)
2041 			return elt;
2042 		offset += copy;
2043 	}
2044 
2045 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2046 		int end;
2047 
2048 		BUG_TRAP(start <= offset + len);
2049 
2050 		end = start + skb_shinfo(skb)->frags[i].size;
2051 		if ((copy = end - offset) > 0) {
2052 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2053 
2054 			if (copy > len)
2055 				copy = len;
2056 			sg[elt].page = frag->page;
2057 			sg[elt].offset = frag->page_offset+offset-start;
2058 			sg[elt].length = copy;
2059 			elt++;
2060 			if (!(len -= copy))
2061 				return elt;
2062 			offset += copy;
2063 		}
2064 		start = end;
2065 	}
2066 
2067 	if (skb_shinfo(skb)->frag_list) {
2068 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
2069 
2070 		for (; list; list = list->next) {
2071 			int end;
2072 
2073 			BUG_TRAP(start <= offset + len);
2074 
2075 			end = start + list->len;
2076 			if ((copy = end - offset) > 0) {
2077 				if (copy > len)
2078 					copy = len;
2079 				elt += skb_to_sgvec(list, sg+elt, offset - start, copy);
2080 				if ((len -= copy) == 0)
2081 					return elt;
2082 				offset += copy;
2083 			}
2084 			start = end;
2085 		}
2086 	}
2087 	BUG_ON(len);
2088 	return elt;
2089 }
2090 
2091 /**
2092  *	skb_cow_data - Check that a socket buffer's data buffers are writable
2093  *	@skb: The socket buffer to check.
2094  *	@tailbits: Amount of trailing space to be added
2095  *	@trailer: Returned pointer to the skb where the @tailbits space begins
2096  *
2097  *	Make sure that the data buffers attached to a socket buffer are
2098  *	writable. If they are not, private copies are made of the data buffers
2099  *	and the socket buffer is set to use these instead.
2100  *
2101  *	If @tailbits is given, make sure that there is space to write @tailbits
2102  *	bytes of data beyond current end of socket buffer.  @trailer will be
2103  *	set to point to the skb in which this space begins.
2104  *
2105  *	The number of scatterlist elements required to completely map the
2106  *	COW'd and extended socket buffer will be returned.
2107  */
2108 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
2109 {
2110 	int copyflag;
2111 	int elt;
2112 	struct sk_buff *skb1, **skb_p;
2113 
2114 	/* If skb is cloned or its head is paged, reallocate
2115 	 * head pulling out all the pages (pages are considered not writable
2116 	 * at the moment even if they are anonymous).
2117 	 */
2118 	if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
2119 	    __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
2120 		return -ENOMEM;
2121 
2122 	/* Easy case. Most of packets will go this way. */
2123 	if (!skb_shinfo(skb)->frag_list) {
2124 		/* A little of trouble, not enough of space for trailer.
2125 		 * This should not happen, when stack is tuned to generate
2126 		 * good frames. OK, on miss we reallocate and reserve even more
2127 		 * space, 128 bytes is fair. */
2128 
2129 		if (skb_tailroom(skb) < tailbits &&
2130 		    pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
2131 			return -ENOMEM;
2132 
2133 		/* Voila! */
2134 		*trailer = skb;
2135 		return 1;
2136 	}
2137 
2138 	/* Misery. We are in troubles, going to mincer fragments... */
2139 
2140 	elt = 1;
2141 	skb_p = &skb_shinfo(skb)->frag_list;
2142 	copyflag = 0;
2143 
2144 	while ((skb1 = *skb_p) != NULL) {
2145 		int ntail = 0;
2146 
2147 		/* The fragment is partially pulled by someone,
2148 		 * this can happen on input. Copy it and everything
2149 		 * after it. */
2150 
2151 		if (skb_shared(skb1))
2152 			copyflag = 1;
2153 
2154 		/* If the skb is the last, worry about trailer. */
2155 
2156 		if (skb1->next == NULL && tailbits) {
2157 			if (skb_shinfo(skb1)->nr_frags ||
2158 			    skb_shinfo(skb1)->frag_list ||
2159 			    skb_tailroom(skb1) < tailbits)
2160 				ntail = tailbits + 128;
2161 		}
2162 
2163 		if (copyflag ||
2164 		    skb_cloned(skb1) ||
2165 		    ntail ||
2166 		    skb_shinfo(skb1)->nr_frags ||
2167 		    skb_shinfo(skb1)->frag_list) {
2168 			struct sk_buff *skb2;
2169 
2170 			/* Fuck, we are miserable poor guys... */
2171 			if (ntail == 0)
2172 				skb2 = skb_copy(skb1, GFP_ATOMIC);
2173 			else
2174 				skb2 = skb_copy_expand(skb1,
2175 						       skb_headroom(skb1),
2176 						       ntail,
2177 						       GFP_ATOMIC);
2178 			if (unlikely(skb2 == NULL))
2179 				return -ENOMEM;
2180 
2181 			if (skb1->sk)
2182 				skb_set_owner_w(skb2, skb1->sk);
2183 
2184 			/* Looking around. Are we still alive?
2185 			 * OK, link new skb, drop old one */
2186 
2187 			skb2->next = skb1->next;
2188 			*skb_p = skb2;
2189 			kfree_skb(skb1);
2190 			skb1 = skb2;
2191 		}
2192 		elt++;
2193 		*trailer = skb1;
2194 		skb_p = &skb1->next;
2195 	}
2196 
2197 	return elt;
2198 }
2199 
2200 EXPORT_SYMBOL(___pskb_trim);
2201 EXPORT_SYMBOL(__kfree_skb);
2202 EXPORT_SYMBOL(kfree_skb);
2203 EXPORT_SYMBOL(__pskb_pull_tail);
2204 EXPORT_SYMBOL(__alloc_skb);
2205 EXPORT_SYMBOL(__netdev_alloc_skb);
2206 EXPORT_SYMBOL(pskb_copy);
2207 EXPORT_SYMBOL(pskb_expand_head);
2208 EXPORT_SYMBOL(skb_checksum);
2209 EXPORT_SYMBOL(skb_clone);
2210 EXPORT_SYMBOL(skb_clone_fraglist);
2211 EXPORT_SYMBOL(skb_copy);
2212 EXPORT_SYMBOL(skb_copy_and_csum_bits);
2213 EXPORT_SYMBOL(skb_copy_and_csum_dev);
2214 EXPORT_SYMBOL(skb_copy_bits);
2215 EXPORT_SYMBOL(skb_copy_expand);
2216 EXPORT_SYMBOL(skb_over_panic);
2217 EXPORT_SYMBOL(skb_pad);
2218 EXPORT_SYMBOL(skb_realloc_headroom);
2219 EXPORT_SYMBOL(skb_under_panic);
2220 EXPORT_SYMBOL(skb_dequeue);
2221 EXPORT_SYMBOL(skb_dequeue_tail);
2222 EXPORT_SYMBOL(skb_insert);
2223 EXPORT_SYMBOL(skb_queue_purge);
2224 EXPORT_SYMBOL(skb_queue_head);
2225 EXPORT_SYMBOL(skb_queue_tail);
2226 EXPORT_SYMBOL(skb_unlink);
2227 EXPORT_SYMBOL(skb_append);
2228 EXPORT_SYMBOL(skb_split);
2229 EXPORT_SYMBOL(skb_prepare_seq_read);
2230 EXPORT_SYMBOL(skb_seq_read);
2231 EXPORT_SYMBOL(skb_abort_seq_read);
2232 EXPORT_SYMBOL(skb_find_text);
2233 EXPORT_SYMBOL(skb_append_datato_frags);
2234 
2235 EXPORT_SYMBOL_GPL(skb_to_sgvec);
2236 EXPORT_SYMBOL_GPL(skb_cow_data);
2237