xref: /linux/net/core/skbuff.c (revision d67b569f5f620c0fb95d5212642746b7ba9d29e4)
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/config.h>
42 #include <linux/module.h>
43 #include <linux/types.h>
44 #include <linux/kernel.h>
45 #include <linux/sched.h>
46 #include <linux/mm.h>
47 #include <linux/interrupt.h>
48 #include <linux/in.h>
49 #include <linux/inet.h>
50 #include <linux/slab.h>
51 #include <linux/netdevice.h>
52 #ifdef CONFIG_NET_CLS_ACT
53 #include <net/pkt_sched.h>
54 #endif
55 #include <linux/string.h>
56 #include <linux/skbuff.h>
57 #include <linux/cache.h>
58 #include <linux/rtnetlink.h>
59 #include <linux/init.h>
60 #include <linux/highmem.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 
71 static kmem_cache_t *skbuff_head_cache;
72 
73 /*
74  *	Keep out-of-line to prevent kernel bloat.
75  *	__builtin_return_address is not used because it is not always
76  *	reliable.
77  */
78 
79 /**
80  *	skb_over_panic	- 	private function
81  *	@skb: buffer
82  *	@sz: size
83  *	@here: address
84  *
85  *	Out of line support code for skb_put(). Not user callable.
86  */
87 void skb_over_panic(struct sk_buff *skb, int sz, void *here)
88 {
89 	printk(KERN_EMERG "skb_over_panic: text:%p len:%d put:%d head:%p "
90 	                  "data:%p tail:%p end:%p dev:%s\n",
91 	       here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
92 	       skb->dev ? skb->dev->name : "<NULL>");
93 	BUG();
94 }
95 
96 /**
97  *	skb_under_panic	- 	private function
98  *	@skb: buffer
99  *	@sz: size
100  *	@here: address
101  *
102  *	Out of line support code for skb_push(). Not user callable.
103  */
104 
105 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
106 {
107 	printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
108 	                  "data:%p tail:%p end:%p dev:%s\n",
109 	       here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
110 	       skb->dev ? skb->dev->name : "<NULL>");
111 	BUG();
112 }
113 
114 /* 	Allocate a new skbuff. We do this ourselves so we can fill in a few
115  *	'private' fields and also do memory statistics to find all the
116  *	[BEEP] leaks.
117  *
118  */
119 
120 /**
121  *	alloc_skb	-	allocate a network buffer
122  *	@size: size to allocate
123  *	@gfp_mask: allocation mask
124  *
125  *	Allocate a new &sk_buff. The returned buffer has no headroom and a
126  *	tail room of size bytes. The object has a reference count of one.
127  *	The return is the buffer. On a failure the return is %NULL.
128  *
129  *	Buffers may only be allocated from interrupts using a @gfp_mask of
130  *	%GFP_ATOMIC.
131  */
132 struct sk_buff *alloc_skb(unsigned int size, int gfp_mask)
133 {
134 	struct sk_buff *skb;
135 	u8 *data;
136 
137 	/* Get the HEAD */
138 	skb = kmem_cache_alloc(skbuff_head_cache,
139 			       gfp_mask & ~__GFP_DMA);
140 	if (!skb)
141 		goto out;
142 
143 	/* Get the DATA. Size must match skb_add_mtu(). */
144 	size = SKB_DATA_ALIGN(size);
145 	data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
146 	if (!data)
147 		goto nodata;
148 
149 	memset(skb, 0, offsetof(struct sk_buff, truesize));
150 	skb->truesize = size + sizeof(struct sk_buff);
151 	atomic_set(&skb->users, 1);
152 	skb->head = data;
153 	skb->data = data;
154 	skb->tail = data;
155 	skb->end  = data + size;
156 
157 	atomic_set(&(skb_shinfo(skb)->dataref), 1);
158 	skb_shinfo(skb)->nr_frags  = 0;
159 	skb_shinfo(skb)->tso_size = 0;
160 	skb_shinfo(skb)->tso_segs = 0;
161 	skb_shinfo(skb)->frag_list = NULL;
162 out:
163 	return skb;
164 nodata:
165 	kmem_cache_free(skbuff_head_cache, skb);
166 	skb = NULL;
167 	goto out;
168 }
169 
170 /**
171  *	alloc_skb_from_cache	-	allocate a network buffer
172  *	@cp: kmem_cache from which to allocate the data area
173  *           (object size must be big enough for @size bytes + skb overheads)
174  *	@size: size to allocate
175  *	@gfp_mask: allocation mask
176  *
177  *	Allocate a new &sk_buff. The returned buffer has no headroom and
178  *	tail room of size bytes. The object has a reference count of one.
179  *	The return is the buffer. On a failure the return is %NULL.
180  *
181  *	Buffers may only be allocated from interrupts using a @gfp_mask of
182  *	%GFP_ATOMIC.
183  */
184 struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
185 				     unsigned int size, int gfp_mask)
186 {
187 	struct sk_buff *skb;
188 	u8 *data;
189 
190 	/* Get the HEAD */
191 	skb = kmem_cache_alloc(skbuff_head_cache,
192 			       gfp_mask & ~__GFP_DMA);
193 	if (!skb)
194 		goto out;
195 
196 	/* Get the DATA. */
197 	size = SKB_DATA_ALIGN(size);
198 	data = kmem_cache_alloc(cp, gfp_mask);
199 	if (!data)
200 		goto nodata;
201 
202 	memset(skb, 0, offsetof(struct sk_buff, truesize));
203 	skb->truesize = size + sizeof(struct sk_buff);
204 	atomic_set(&skb->users, 1);
205 	skb->head = data;
206 	skb->data = data;
207 	skb->tail = data;
208 	skb->end  = data + size;
209 
210 	atomic_set(&(skb_shinfo(skb)->dataref), 1);
211 	skb_shinfo(skb)->nr_frags  = 0;
212 	skb_shinfo(skb)->tso_size = 0;
213 	skb_shinfo(skb)->tso_segs = 0;
214 	skb_shinfo(skb)->frag_list = NULL;
215 out:
216 	return skb;
217 nodata:
218 	kmem_cache_free(skbuff_head_cache, skb);
219 	skb = NULL;
220 	goto out;
221 }
222 
223 
224 static void skb_drop_fraglist(struct sk_buff *skb)
225 {
226 	struct sk_buff *list = skb_shinfo(skb)->frag_list;
227 
228 	skb_shinfo(skb)->frag_list = NULL;
229 
230 	do {
231 		struct sk_buff *this = list;
232 		list = list->next;
233 		kfree_skb(this);
234 	} while (list);
235 }
236 
237 static void skb_clone_fraglist(struct sk_buff *skb)
238 {
239 	struct sk_buff *list;
240 
241 	for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
242 		skb_get(list);
243 }
244 
245 void skb_release_data(struct sk_buff *skb)
246 {
247 	if (!skb->cloned ||
248 	    !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
249 			       &skb_shinfo(skb)->dataref)) {
250 		if (skb_shinfo(skb)->nr_frags) {
251 			int i;
252 			for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
253 				put_page(skb_shinfo(skb)->frags[i].page);
254 		}
255 
256 		if (skb_shinfo(skb)->frag_list)
257 			skb_drop_fraglist(skb);
258 
259 		kfree(skb->head);
260 	}
261 }
262 
263 /*
264  *	Free an skbuff by memory without cleaning the state.
265  */
266 void kfree_skbmem(struct sk_buff *skb)
267 {
268 	skb_release_data(skb);
269 	kmem_cache_free(skbuff_head_cache, skb);
270 }
271 
272 /**
273  *	__kfree_skb - private function
274  *	@skb: buffer
275  *
276  *	Free an sk_buff. Release anything attached to the buffer.
277  *	Clean the state. This is an internal helper function. Users should
278  *	always call kfree_skb
279  */
280 
281 void __kfree_skb(struct sk_buff *skb)
282 {
283 	BUG_ON(skb->list != NULL);
284 
285 	dst_release(skb->dst);
286 #ifdef CONFIG_XFRM
287 	secpath_put(skb->sp);
288 #endif
289 	if (skb->destructor) {
290 		WARN_ON(in_irq());
291 		skb->destructor(skb);
292 	}
293 #ifdef CONFIG_NETFILTER
294 	nf_conntrack_put(skb->nfct);
295 #ifdef CONFIG_BRIDGE_NETFILTER
296 	nf_bridge_put(skb->nf_bridge);
297 #endif
298 #endif
299 /* XXX: IS this still necessary? - JHS */
300 #ifdef CONFIG_NET_SCHED
301 	skb->tc_index = 0;
302 #ifdef CONFIG_NET_CLS_ACT
303 	skb->tc_verd = 0;
304 	skb->tc_classid = 0;
305 #endif
306 #endif
307 
308 	kfree_skbmem(skb);
309 }
310 
311 /**
312  *	skb_clone	-	duplicate an sk_buff
313  *	@skb: buffer to clone
314  *	@gfp_mask: allocation priority
315  *
316  *	Duplicate an &sk_buff. The new one is not owned by a socket. Both
317  *	copies share the same packet data but not structure. The new
318  *	buffer has a reference count of 1. If the allocation fails the
319  *	function returns %NULL otherwise the new buffer is returned.
320  *
321  *	If this function is called from an interrupt gfp_mask() must be
322  *	%GFP_ATOMIC.
323  */
324 
325 struct sk_buff *skb_clone(struct sk_buff *skb, int gfp_mask)
326 {
327 	struct sk_buff *n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
328 
329 	if (!n)
330 		return NULL;
331 
332 #define C(x) n->x = skb->x
333 
334 	n->next = n->prev = NULL;
335 	n->list = NULL;
336 	n->sk = NULL;
337 	C(stamp);
338 	C(dev);
339 	C(real_dev);
340 	C(h);
341 	C(nh);
342 	C(mac);
343 	C(dst);
344 	dst_clone(skb->dst);
345 	C(sp);
346 #ifdef CONFIG_INET
347 	secpath_get(skb->sp);
348 #endif
349 	memcpy(n->cb, skb->cb, sizeof(skb->cb));
350 	C(len);
351 	C(data_len);
352 	C(csum);
353 	C(local_df);
354 	n->cloned = 1;
355 	n->nohdr = 0;
356 	C(pkt_type);
357 	C(ip_summed);
358 	C(priority);
359 	C(protocol);
360 	n->destructor = NULL;
361 #ifdef CONFIG_NETFILTER
362 	C(nfmark);
363 	C(nfcache);
364 	C(nfct);
365 	nf_conntrack_get(skb->nfct);
366 	C(nfctinfo);
367 #ifdef CONFIG_BRIDGE_NETFILTER
368 	C(nf_bridge);
369 	nf_bridge_get(skb->nf_bridge);
370 #endif
371 #endif /*CONFIG_NETFILTER*/
372 #if defined(CONFIG_HIPPI)
373 	C(private);
374 #endif
375 #ifdef CONFIG_NET_SCHED
376 	C(tc_index);
377 #ifdef CONFIG_NET_CLS_ACT
378 	n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
379 	n->tc_verd = CLR_TC_OK2MUNGE(skb->tc_verd);
380 	n->tc_verd = CLR_TC_MUNGED(skb->tc_verd);
381 	C(input_dev);
382 	C(tc_classid);
383 #endif
384 
385 #endif
386 	C(truesize);
387 	atomic_set(&n->users, 1);
388 	C(head);
389 	C(data);
390 	C(tail);
391 	C(end);
392 
393 	atomic_inc(&(skb_shinfo(skb)->dataref));
394 	skb->cloned = 1;
395 
396 	return n;
397 }
398 
399 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
400 {
401 	/*
402 	 *	Shift between the two data areas in bytes
403 	 */
404 	unsigned long offset = new->data - old->data;
405 
406 	new->list	= NULL;
407 	new->sk		= NULL;
408 	new->dev	= old->dev;
409 	new->real_dev	= old->real_dev;
410 	new->priority	= old->priority;
411 	new->protocol	= old->protocol;
412 	new->dst	= dst_clone(old->dst);
413 #ifdef CONFIG_INET
414 	new->sp		= secpath_get(old->sp);
415 #endif
416 	new->h.raw	= old->h.raw + offset;
417 	new->nh.raw	= old->nh.raw + offset;
418 	new->mac.raw	= old->mac.raw + offset;
419 	memcpy(new->cb, old->cb, sizeof(old->cb));
420 	new->local_df	= old->local_df;
421 	new->pkt_type	= old->pkt_type;
422 	new->stamp	= old->stamp;
423 	new->destructor = NULL;
424 #ifdef CONFIG_NETFILTER
425 	new->nfmark	= old->nfmark;
426 	new->nfcache	= old->nfcache;
427 	new->nfct	= old->nfct;
428 	nf_conntrack_get(old->nfct);
429 	new->nfctinfo	= old->nfctinfo;
430 #ifdef CONFIG_BRIDGE_NETFILTER
431 	new->nf_bridge	= old->nf_bridge;
432 	nf_bridge_get(old->nf_bridge);
433 #endif
434 #endif
435 #ifdef CONFIG_NET_SCHED
436 #ifdef CONFIG_NET_CLS_ACT
437 	new->tc_verd = old->tc_verd;
438 #endif
439 	new->tc_index	= old->tc_index;
440 #endif
441 	atomic_set(&new->users, 1);
442 	skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
443 	skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
444 }
445 
446 /**
447  *	skb_copy	-	create private copy of an sk_buff
448  *	@skb: buffer to copy
449  *	@gfp_mask: allocation priority
450  *
451  *	Make a copy of both an &sk_buff and its data. This is used when the
452  *	caller wishes to modify the data and needs a private copy of the
453  *	data to alter. Returns %NULL on failure or the pointer to the buffer
454  *	on success. The returned buffer has a reference count of 1.
455  *
456  *	As by-product this function converts non-linear &sk_buff to linear
457  *	one, so that &sk_buff becomes completely private and caller is allowed
458  *	to modify all the data of returned buffer. This means that this
459  *	function is not recommended for use in circumstances when only
460  *	header is going to be modified. Use pskb_copy() instead.
461  */
462 
463 struct sk_buff *skb_copy(const struct sk_buff *skb, int gfp_mask)
464 {
465 	int headerlen = skb->data - skb->head;
466 	/*
467 	 *	Allocate the copy buffer
468 	 */
469 	struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
470 				      gfp_mask);
471 	if (!n)
472 		return NULL;
473 
474 	/* Set the data pointer */
475 	skb_reserve(n, headerlen);
476 	/* Set the tail pointer and length */
477 	skb_put(n, skb->len);
478 	n->csum	     = skb->csum;
479 	n->ip_summed = skb->ip_summed;
480 
481 	if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
482 		BUG();
483 
484 	copy_skb_header(n, skb);
485 	return n;
486 }
487 
488 
489 /**
490  *	pskb_copy	-	create copy of an sk_buff with private head.
491  *	@skb: buffer to copy
492  *	@gfp_mask: allocation priority
493  *
494  *	Make a copy of both an &sk_buff and part of its data, located
495  *	in header. Fragmented data remain shared. This is used when
496  *	the caller wishes to modify only header of &sk_buff and needs
497  *	private copy of the header to alter. Returns %NULL on failure
498  *	or the pointer to the buffer on success.
499  *	The returned buffer has a reference count of 1.
500  */
501 
502 struct sk_buff *pskb_copy(struct sk_buff *skb, int gfp_mask)
503 {
504 	/*
505 	 *	Allocate the copy buffer
506 	 */
507 	struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
508 
509 	if (!n)
510 		goto out;
511 
512 	/* Set the data pointer */
513 	skb_reserve(n, skb->data - skb->head);
514 	/* Set the tail pointer and length */
515 	skb_put(n, skb_headlen(skb));
516 	/* Copy the bytes */
517 	memcpy(n->data, skb->data, n->len);
518 	n->csum	     = skb->csum;
519 	n->ip_summed = skb->ip_summed;
520 
521 	n->data_len  = skb->data_len;
522 	n->len	     = skb->len;
523 
524 	if (skb_shinfo(skb)->nr_frags) {
525 		int i;
526 
527 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
528 			skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
529 			get_page(skb_shinfo(n)->frags[i].page);
530 		}
531 		skb_shinfo(n)->nr_frags = i;
532 	}
533 
534 	if (skb_shinfo(skb)->frag_list) {
535 		skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
536 		skb_clone_fraglist(n);
537 	}
538 
539 	copy_skb_header(n, skb);
540 out:
541 	return n;
542 }
543 
544 /**
545  *	pskb_expand_head - reallocate header of &sk_buff
546  *	@skb: buffer to reallocate
547  *	@nhead: room to add at head
548  *	@ntail: room to add at tail
549  *	@gfp_mask: allocation priority
550  *
551  *	Expands (or creates identical copy, if &nhead and &ntail are zero)
552  *	header of skb. &sk_buff itself is not changed. &sk_buff MUST have
553  *	reference count of 1. Returns zero in the case of success or error,
554  *	if expansion failed. In the last case, &sk_buff is not changed.
555  *
556  *	All the pointers pointing into skb header may change and must be
557  *	reloaded after call to this function.
558  */
559 
560 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, int gfp_mask)
561 {
562 	int i;
563 	u8 *data;
564 	int size = nhead + (skb->end - skb->head) + ntail;
565 	long off;
566 
567 	if (skb_shared(skb))
568 		BUG();
569 
570 	size = SKB_DATA_ALIGN(size);
571 
572 	data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
573 	if (!data)
574 		goto nodata;
575 
576 	/* Copy only real data... and, alas, header. This should be
577 	 * optimized for the cases when header is void. */
578 	memcpy(data + nhead, skb->head, skb->tail - skb->head);
579 	memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
580 
581 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
582 		get_page(skb_shinfo(skb)->frags[i].page);
583 
584 	if (skb_shinfo(skb)->frag_list)
585 		skb_clone_fraglist(skb);
586 
587 	skb_release_data(skb);
588 
589 	off = (data + nhead) - skb->head;
590 
591 	skb->head     = data;
592 	skb->end      = data + size;
593 	skb->data    += off;
594 	skb->tail    += off;
595 	skb->mac.raw += off;
596 	skb->h.raw   += off;
597 	skb->nh.raw  += off;
598 	skb->cloned   = 0;
599 	skb->nohdr    = 0;
600 	atomic_set(&skb_shinfo(skb)->dataref, 1);
601 	return 0;
602 
603 nodata:
604 	return -ENOMEM;
605 }
606 
607 /* Make private copy of skb with writable head and some headroom */
608 
609 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
610 {
611 	struct sk_buff *skb2;
612 	int delta = headroom - skb_headroom(skb);
613 
614 	if (delta <= 0)
615 		skb2 = pskb_copy(skb, GFP_ATOMIC);
616 	else {
617 		skb2 = skb_clone(skb, GFP_ATOMIC);
618 		if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
619 					     GFP_ATOMIC)) {
620 			kfree_skb(skb2);
621 			skb2 = NULL;
622 		}
623 	}
624 	return skb2;
625 }
626 
627 
628 /**
629  *	skb_copy_expand	-	copy and expand sk_buff
630  *	@skb: buffer to copy
631  *	@newheadroom: new free bytes at head
632  *	@newtailroom: new free bytes at tail
633  *	@gfp_mask: allocation priority
634  *
635  *	Make a copy of both an &sk_buff and its data and while doing so
636  *	allocate additional space.
637  *
638  *	This is used when the caller wishes to modify the data and needs a
639  *	private copy of the data to alter as well as more space for new fields.
640  *	Returns %NULL on failure or the pointer to the buffer
641  *	on success. The returned buffer has a reference count of 1.
642  *
643  *	You must pass %GFP_ATOMIC as the allocation priority if this function
644  *	is called from an interrupt.
645  *
646  *	BUG ALERT: ip_summed is not copied. Why does this work? Is it used
647  *	only by netfilter in the cases when checksum is recalculated? --ANK
648  */
649 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
650 				int newheadroom, int newtailroom, int gfp_mask)
651 {
652 	/*
653 	 *	Allocate the copy buffer
654 	 */
655 	struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
656 				      gfp_mask);
657 	int head_copy_len, head_copy_off;
658 
659 	if (!n)
660 		return NULL;
661 
662 	skb_reserve(n, newheadroom);
663 
664 	/* Set the tail pointer and length */
665 	skb_put(n, skb->len);
666 
667 	head_copy_len = skb_headroom(skb);
668 	head_copy_off = 0;
669 	if (newheadroom <= head_copy_len)
670 		head_copy_len = newheadroom;
671 	else
672 		head_copy_off = newheadroom - head_copy_len;
673 
674 	/* Copy the linear header and data. */
675 	if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
676 			  skb->len + head_copy_len))
677 		BUG();
678 
679 	copy_skb_header(n, skb);
680 
681 	return n;
682 }
683 
684 /**
685  *	skb_pad			-	zero pad the tail of an skb
686  *	@skb: buffer to pad
687  *	@pad: space to pad
688  *
689  *	Ensure that a buffer is followed by a padding area that is zero
690  *	filled. Used by network drivers which may DMA or transfer data
691  *	beyond the buffer end onto the wire.
692  *
693  *	May return NULL in out of memory cases.
694  */
695 
696 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
697 {
698 	struct sk_buff *nskb;
699 
700 	/* If the skbuff is non linear tailroom is always zero.. */
701 	if (skb_tailroom(skb) >= pad) {
702 		memset(skb->data+skb->len, 0, pad);
703 		return skb;
704 	}
705 
706 	nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
707 	kfree_skb(skb);
708 	if (nskb)
709 		memset(nskb->data+nskb->len, 0, pad);
710 	return nskb;
711 }
712 
713 /* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
714  * If realloc==0 and trimming is impossible without change of data,
715  * it is BUG().
716  */
717 
718 int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
719 {
720 	int offset = skb_headlen(skb);
721 	int nfrags = skb_shinfo(skb)->nr_frags;
722 	int i;
723 
724 	for (i = 0; i < nfrags; i++) {
725 		int end = offset + skb_shinfo(skb)->frags[i].size;
726 		if (end > len) {
727 			if (skb_cloned(skb)) {
728 				if (!realloc)
729 					BUG();
730 				if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
731 					return -ENOMEM;
732 			}
733 			if (len <= offset) {
734 				put_page(skb_shinfo(skb)->frags[i].page);
735 				skb_shinfo(skb)->nr_frags--;
736 			} else {
737 				skb_shinfo(skb)->frags[i].size = len - offset;
738 			}
739 		}
740 		offset = end;
741 	}
742 
743 	if (offset < len) {
744 		skb->data_len -= skb->len - len;
745 		skb->len       = len;
746 	} else {
747 		if (len <= skb_headlen(skb)) {
748 			skb->len      = len;
749 			skb->data_len = 0;
750 			skb->tail     = skb->data + len;
751 			if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
752 				skb_drop_fraglist(skb);
753 		} else {
754 			skb->data_len -= skb->len - len;
755 			skb->len       = len;
756 		}
757 	}
758 
759 	return 0;
760 }
761 
762 /**
763  *	__pskb_pull_tail - advance tail of skb header
764  *	@skb: buffer to reallocate
765  *	@delta: number of bytes to advance tail
766  *
767  *	The function makes a sense only on a fragmented &sk_buff,
768  *	it expands header moving its tail forward and copying necessary
769  *	data from fragmented part.
770  *
771  *	&sk_buff MUST have reference count of 1.
772  *
773  *	Returns %NULL (and &sk_buff does not change) if pull failed
774  *	or value of new tail of skb in the case of success.
775  *
776  *	All the pointers pointing into skb header may change and must be
777  *	reloaded after call to this function.
778  */
779 
780 /* Moves tail of skb head forward, copying data from fragmented part,
781  * when it is necessary.
782  * 1. It may fail due to malloc failure.
783  * 2. It may change skb pointers.
784  *
785  * It is pretty complicated. Luckily, it is called only in exceptional cases.
786  */
787 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
788 {
789 	/* If skb has not enough free space at tail, get new one
790 	 * plus 128 bytes for future expansions. If we have enough
791 	 * room at tail, reallocate without expansion only if skb is cloned.
792 	 */
793 	int i, k, eat = (skb->tail + delta) - skb->end;
794 
795 	if (eat > 0 || skb_cloned(skb)) {
796 		if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
797 				     GFP_ATOMIC))
798 			return NULL;
799 	}
800 
801 	if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
802 		BUG();
803 
804 	/* Optimization: no fragments, no reasons to preestimate
805 	 * size of pulled pages. Superb.
806 	 */
807 	if (!skb_shinfo(skb)->frag_list)
808 		goto pull_pages;
809 
810 	/* Estimate size of pulled pages. */
811 	eat = delta;
812 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
813 		if (skb_shinfo(skb)->frags[i].size >= eat)
814 			goto pull_pages;
815 		eat -= skb_shinfo(skb)->frags[i].size;
816 	}
817 
818 	/* If we need update frag list, we are in troubles.
819 	 * Certainly, it possible to add an offset to skb data,
820 	 * but taking into account that pulling is expected to
821 	 * be very rare operation, it is worth to fight against
822 	 * further bloating skb head and crucify ourselves here instead.
823 	 * Pure masohism, indeed. 8)8)
824 	 */
825 	if (eat) {
826 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
827 		struct sk_buff *clone = NULL;
828 		struct sk_buff *insp = NULL;
829 
830 		do {
831 			if (!list)
832 				BUG();
833 
834 			if (list->len <= eat) {
835 				/* Eaten as whole. */
836 				eat -= list->len;
837 				list = list->next;
838 				insp = list;
839 			} else {
840 				/* Eaten partially. */
841 
842 				if (skb_shared(list)) {
843 					/* Sucks! We need to fork list. :-( */
844 					clone = skb_clone(list, GFP_ATOMIC);
845 					if (!clone)
846 						return NULL;
847 					insp = list->next;
848 					list = clone;
849 				} else {
850 					/* This may be pulled without
851 					 * problems. */
852 					insp = list;
853 				}
854 				if (!pskb_pull(list, eat)) {
855 					if (clone)
856 						kfree_skb(clone);
857 					return NULL;
858 				}
859 				break;
860 			}
861 		} while (eat);
862 
863 		/* Free pulled out fragments. */
864 		while ((list = skb_shinfo(skb)->frag_list) != insp) {
865 			skb_shinfo(skb)->frag_list = list->next;
866 			kfree_skb(list);
867 		}
868 		/* And insert new clone at head. */
869 		if (clone) {
870 			clone->next = list;
871 			skb_shinfo(skb)->frag_list = clone;
872 		}
873 	}
874 	/* Success! Now we may commit changes to skb data. */
875 
876 pull_pages:
877 	eat = delta;
878 	k = 0;
879 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
880 		if (skb_shinfo(skb)->frags[i].size <= eat) {
881 			put_page(skb_shinfo(skb)->frags[i].page);
882 			eat -= skb_shinfo(skb)->frags[i].size;
883 		} else {
884 			skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
885 			if (eat) {
886 				skb_shinfo(skb)->frags[k].page_offset += eat;
887 				skb_shinfo(skb)->frags[k].size -= eat;
888 				eat = 0;
889 			}
890 			k++;
891 		}
892 	}
893 	skb_shinfo(skb)->nr_frags = k;
894 
895 	skb->tail     += delta;
896 	skb->data_len -= delta;
897 
898 	return skb->tail;
899 }
900 
901 /* Copy some data bits from skb to kernel buffer. */
902 
903 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
904 {
905 	int i, copy;
906 	int start = skb_headlen(skb);
907 
908 	if (offset > (int)skb->len - len)
909 		goto fault;
910 
911 	/* Copy header. */
912 	if ((copy = start - offset) > 0) {
913 		if (copy > len)
914 			copy = len;
915 		memcpy(to, skb->data + offset, copy);
916 		if ((len -= copy) == 0)
917 			return 0;
918 		offset += copy;
919 		to     += copy;
920 	}
921 
922 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
923 		int end;
924 
925 		BUG_TRAP(start <= offset + len);
926 
927 		end = start + skb_shinfo(skb)->frags[i].size;
928 		if ((copy = end - offset) > 0) {
929 			u8 *vaddr;
930 
931 			if (copy > len)
932 				copy = len;
933 
934 			vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
935 			memcpy(to,
936 			       vaddr + skb_shinfo(skb)->frags[i].page_offset+
937 			       offset - start, copy);
938 			kunmap_skb_frag(vaddr);
939 
940 			if ((len -= copy) == 0)
941 				return 0;
942 			offset += copy;
943 			to     += copy;
944 		}
945 		start = end;
946 	}
947 
948 	if (skb_shinfo(skb)->frag_list) {
949 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
950 
951 		for (; list; list = list->next) {
952 			int end;
953 
954 			BUG_TRAP(start <= offset + len);
955 
956 			end = start + list->len;
957 			if ((copy = end - offset) > 0) {
958 				if (copy > len)
959 					copy = len;
960 				if (skb_copy_bits(list, offset - start,
961 						  to, copy))
962 					goto fault;
963 				if ((len -= copy) == 0)
964 					return 0;
965 				offset += copy;
966 				to     += copy;
967 			}
968 			start = end;
969 		}
970 	}
971 	if (!len)
972 		return 0;
973 
974 fault:
975 	return -EFAULT;
976 }
977 
978 /**
979  *	skb_store_bits - store bits from kernel buffer to skb
980  *	@skb: destination buffer
981  *	@offset: offset in destination
982  *	@from: source buffer
983  *	@len: number of bytes to copy
984  *
985  *	Copy the specified number of bytes from the source buffer to the
986  *	destination skb.  This function handles all the messy bits of
987  *	traversing fragment lists and such.
988  */
989 
990 int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
991 {
992 	int i, copy;
993 	int start = skb_headlen(skb);
994 
995 	if (offset > (int)skb->len - len)
996 		goto fault;
997 
998 	if ((copy = start - offset) > 0) {
999 		if (copy > len)
1000 			copy = len;
1001 		memcpy(skb->data + offset, from, copy);
1002 		if ((len -= copy) == 0)
1003 			return 0;
1004 		offset += copy;
1005 		from += copy;
1006 	}
1007 
1008 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1009 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1010 		int end;
1011 
1012 		BUG_TRAP(start <= offset + len);
1013 
1014 		end = start + frag->size;
1015 		if ((copy = end - offset) > 0) {
1016 			u8 *vaddr;
1017 
1018 			if (copy > len)
1019 				copy = len;
1020 
1021 			vaddr = kmap_skb_frag(frag);
1022 			memcpy(vaddr + frag->page_offset + offset - start,
1023 			       from, copy);
1024 			kunmap_skb_frag(vaddr);
1025 
1026 			if ((len -= copy) == 0)
1027 				return 0;
1028 			offset += copy;
1029 			from += copy;
1030 		}
1031 		start = end;
1032 	}
1033 
1034 	if (skb_shinfo(skb)->frag_list) {
1035 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1036 
1037 		for (; list; list = list->next) {
1038 			int end;
1039 
1040 			BUG_TRAP(start <= offset + len);
1041 
1042 			end = start + list->len;
1043 			if ((copy = end - offset) > 0) {
1044 				if (copy > len)
1045 					copy = len;
1046 				if (skb_store_bits(list, offset - start,
1047 						   from, copy))
1048 					goto fault;
1049 				if ((len -= copy) == 0)
1050 					return 0;
1051 				offset += copy;
1052 				from += copy;
1053 			}
1054 			start = end;
1055 		}
1056 	}
1057 	if (!len)
1058 		return 0;
1059 
1060 fault:
1061 	return -EFAULT;
1062 }
1063 
1064 EXPORT_SYMBOL(skb_store_bits);
1065 
1066 /* Checksum skb data. */
1067 
1068 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1069 			  int len, unsigned int csum)
1070 {
1071 	int start = skb_headlen(skb);
1072 	int i, copy = start - offset;
1073 	int pos = 0;
1074 
1075 	/* Checksum header. */
1076 	if (copy > 0) {
1077 		if (copy > len)
1078 			copy = len;
1079 		csum = csum_partial(skb->data + offset, copy, csum);
1080 		if ((len -= copy) == 0)
1081 			return csum;
1082 		offset += copy;
1083 		pos	= copy;
1084 	}
1085 
1086 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1087 		int end;
1088 
1089 		BUG_TRAP(start <= offset + len);
1090 
1091 		end = start + skb_shinfo(skb)->frags[i].size;
1092 		if ((copy = end - offset) > 0) {
1093 			unsigned int csum2;
1094 			u8 *vaddr;
1095 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1096 
1097 			if (copy > len)
1098 				copy = len;
1099 			vaddr = kmap_skb_frag(frag);
1100 			csum2 = csum_partial(vaddr + frag->page_offset +
1101 					     offset - start, copy, 0);
1102 			kunmap_skb_frag(vaddr);
1103 			csum = csum_block_add(csum, csum2, pos);
1104 			if (!(len -= copy))
1105 				return csum;
1106 			offset += copy;
1107 			pos    += copy;
1108 		}
1109 		start = end;
1110 	}
1111 
1112 	if (skb_shinfo(skb)->frag_list) {
1113 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1114 
1115 		for (; list; list = list->next) {
1116 			int end;
1117 
1118 			BUG_TRAP(start <= offset + len);
1119 
1120 			end = start + list->len;
1121 			if ((copy = end - offset) > 0) {
1122 				unsigned int csum2;
1123 				if (copy > len)
1124 					copy = len;
1125 				csum2 = skb_checksum(list, offset - start,
1126 						     copy, 0);
1127 				csum = csum_block_add(csum, csum2, pos);
1128 				if ((len -= copy) == 0)
1129 					return csum;
1130 				offset += copy;
1131 				pos    += copy;
1132 			}
1133 			start = end;
1134 		}
1135 	}
1136 	if (len)
1137 		BUG();
1138 
1139 	return csum;
1140 }
1141 
1142 /* Both of above in one bottle. */
1143 
1144 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1145 				    u8 *to, int len, unsigned int csum)
1146 {
1147 	int start = skb_headlen(skb);
1148 	int i, copy = start - offset;
1149 	int pos = 0;
1150 
1151 	/* Copy header. */
1152 	if (copy > 0) {
1153 		if (copy > len)
1154 			copy = len;
1155 		csum = csum_partial_copy_nocheck(skb->data + offset, to,
1156 						 copy, csum);
1157 		if ((len -= copy) == 0)
1158 			return csum;
1159 		offset += copy;
1160 		to     += copy;
1161 		pos	= copy;
1162 	}
1163 
1164 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1165 		int end;
1166 
1167 		BUG_TRAP(start <= offset + len);
1168 
1169 		end = start + skb_shinfo(skb)->frags[i].size;
1170 		if ((copy = end - offset) > 0) {
1171 			unsigned int csum2;
1172 			u8 *vaddr;
1173 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1174 
1175 			if (copy > len)
1176 				copy = len;
1177 			vaddr = kmap_skb_frag(frag);
1178 			csum2 = csum_partial_copy_nocheck(vaddr +
1179 							  frag->page_offset +
1180 							  offset - start, to,
1181 							  copy, 0);
1182 			kunmap_skb_frag(vaddr);
1183 			csum = csum_block_add(csum, csum2, pos);
1184 			if (!(len -= copy))
1185 				return csum;
1186 			offset += copy;
1187 			to     += copy;
1188 			pos    += copy;
1189 		}
1190 		start = end;
1191 	}
1192 
1193 	if (skb_shinfo(skb)->frag_list) {
1194 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1195 
1196 		for (; list; list = list->next) {
1197 			unsigned int csum2;
1198 			int end;
1199 
1200 			BUG_TRAP(start <= offset + len);
1201 
1202 			end = start + list->len;
1203 			if ((copy = end - offset) > 0) {
1204 				if (copy > len)
1205 					copy = len;
1206 				csum2 = skb_copy_and_csum_bits(list,
1207 							       offset - start,
1208 							       to, copy, 0);
1209 				csum = csum_block_add(csum, csum2, pos);
1210 				if ((len -= copy) == 0)
1211 					return csum;
1212 				offset += copy;
1213 				to     += copy;
1214 				pos    += copy;
1215 			}
1216 			start = end;
1217 		}
1218 	}
1219 	if (len)
1220 		BUG();
1221 	return csum;
1222 }
1223 
1224 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1225 {
1226 	unsigned int csum;
1227 	long csstart;
1228 
1229 	if (skb->ip_summed == CHECKSUM_HW)
1230 		csstart = skb->h.raw - skb->data;
1231 	else
1232 		csstart = skb_headlen(skb);
1233 
1234 	if (csstart > skb_headlen(skb))
1235 		BUG();
1236 
1237 	memcpy(to, skb->data, csstart);
1238 
1239 	csum = 0;
1240 	if (csstart != skb->len)
1241 		csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1242 					      skb->len - csstart, 0);
1243 
1244 	if (skb->ip_summed == CHECKSUM_HW) {
1245 		long csstuff = csstart + skb->csum;
1246 
1247 		*((unsigned short *)(to + csstuff)) = csum_fold(csum);
1248 	}
1249 }
1250 
1251 /**
1252  *	skb_dequeue - remove from the head of the queue
1253  *	@list: list to dequeue from
1254  *
1255  *	Remove the head of the list. The list lock is taken so the function
1256  *	may be used safely with other locking list functions. The head item is
1257  *	returned or %NULL if the list is empty.
1258  */
1259 
1260 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1261 {
1262 	unsigned long flags;
1263 	struct sk_buff *result;
1264 
1265 	spin_lock_irqsave(&list->lock, flags);
1266 	result = __skb_dequeue(list);
1267 	spin_unlock_irqrestore(&list->lock, flags);
1268 	return result;
1269 }
1270 
1271 /**
1272  *	skb_dequeue_tail - remove from the tail of the queue
1273  *	@list: list to dequeue from
1274  *
1275  *	Remove the tail of the list. The list lock is taken so the function
1276  *	may be used safely with other locking list functions. The tail item is
1277  *	returned or %NULL if the list is empty.
1278  */
1279 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1280 {
1281 	unsigned long flags;
1282 	struct sk_buff *result;
1283 
1284 	spin_lock_irqsave(&list->lock, flags);
1285 	result = __skb_dequeue_tail(list);
1286 	spin_unlock_irqrestore(&list->lock, flags);
1287 	return result;
1288 }
1289 
1290 /**
1291  *	skb_queue_purge - empty a list
1292  *	@list: list to empty
1293  *
1294  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
1295  *	the list and one reference dropped. This function takes the list
1296  *	lock and is atomic with respect to other list locking functions.
1297  */
1298 void skb_queue_purge(struct sk_buff_head *list)
1299 {
1300 	struct sk_buff *skb;
1301 	while ((skb = skb_dequeue(list)) != NULL)
1302 		kfree_skb(skb);
1303 }
1304 
1305 /**
1306  *	skb_queue_head - queue a buffer at the list head
1307  *	@list: list to use
1308  *	@newsk: buffer to queue
1309  *
1310  *	Queue a buffer at the start of the list. This function takes the
1311  *	list lock and can be used safely with other locking &sk_buff functions
1312  *	safely.
1313  *
1314  *	A buffer cannot be placed on two lists at the same time.
1315  */
1316 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1317 {
1318 	unsigned long flags;
1319 
1320 	spin_lock_irqsave(&list->lock, flags);
1321 	__skb_queue_head(list, newsk);
1322 	spin_unlock_irqrestore(&list->lock, flags);
1323 }
1324 
1325 /**
1326  *	skb_queue_tail - queue a buffer at the list tail
1327  *	@list: list to use
1328  *	@newsk: buffer to queue
1329  *
1330  *	Queue a buffer at the tail of the list. This function takes the
1331  *	list lock and can be used safely with other locking &sk_buff functions
1332  *	safely.
1333  *
1334  *	A buffer cannot be placed on two lists at the same time.
1335  */
1336 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1337 {
1338 	unsigned long flags;
1339 
1340 	spin_lock_irqsave(&list->lock, flags);
1341 	__skb_queue_tail(list, newsk);
1342 	spin_unlock_irqrestore(&list->lock, flags);
1343 }
1344 /**
1345  *	skb_unlink	-	remove a buffer from a list
1346  *	@skb: buffer to remove
1347  *
1348  *	Place a packet after a given packet in a list. The list locks are taken
1349  *	and this function is atomic with respect to other list locked calls
1350  *
1351  *	Works even without knowing the list it is sitting on, which can be
1352  *	handy at times. It also means that THE LIST MUST EXIST when you
1353  *	unlink. Thus a list must have its contents unlinked before it is
1354  *	destroyed.
1355  */
1356 void skb_unlink(struct sk_buff *skb)
1357 {
1358 	struct sk_buff_head *list = skb->list;
1359 
1360 	if (list) {
1361 		unsigned long flags;
1362 
1363 		spin_lock_irqsave(&list->lock, flags);
1364 		if (skb->list == list)
1365 			__skb_unlink(skb, skb->list);
1366 		spin_unlock_irqrestore(&list->lock, flags);
1367 	}
1368 }
1369 
1370 
1371 /**
1372  *	skb_append	-	append a buffer
1373  *	@old: buffer to insert after
1374  *	@newsk: buffer to insert
1375  *
1376  *	Place a packet after a given packet in a list. The list locks are taken
1377  *	and this function is atomic with respect to other list locked calls.
1378  *	A buffer cannot be placed on two lists at the same time.
1379  */
1380 
1381 void skb_append(struct sk_buff *old, struct sk_buff *newsk)
1382 {
1383 	unsigned long flags;
1384 
1385 	spin_lock_irqsave(&old->list->lock, flags);
1386 	__skb_append(old, newsk);
1387 	spin_unlock_irqrestore(&old->list->lock, flags);
1388 }
1389 
1390 
1391 /**
1392  *	skb_insert	-	insert a buffer
1393  *	@old: buffer to insert before
1394  *	@newsk: buffer to insert
1395  *
1396  *	Place a packet before a given packet in a list. The list locks are taken
1397  *	and this function is atomic with respect to other list locked calls
1398  *	A buffer cannot be placed on two lists at the same time.
1399  */
1400 
1401 void skb_insert(struct sk_buff *old, struct sk_buff *newsk)
1402 {
1403 	unsigned long flags;
1404 
1405 	spin_lock_irqsave(&old->list->lock, flags);
1406 	__skb_insert(newsk, old->prev, old, old->list);
1407 	spin_unlock_irqrestore(&old->list->lock, flags);
1408 }
1409 
1410 #if 0
1411 /*
1412  * 	Tune the memory allocator for a new MTU size.
1413  */
1414 void skb_add_mtu(int mtu)
1415 {
1416 	/* Must match allocation in alloc_skb */
1417 	mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1418 
1419 	kmem_add_cache_size(mtu);
1420 }
1421 #endif
1422 
1423 static inline void skb_split_inside_header(struct sk_buff *skb,
1424 					   struct sk_buff* skb1,
1425 					   const u32 len, const int pos)
1426 {
1427 	int i;
1428 
1429 	memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1430 
1431 	/* And move data appendix as is. */
1432 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1433 		skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1434 
1435 	skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1436 	skb_shinfo(skb)->nr_frags  = 0;
1437 	skb1->data_len		   = skb->data_len;
1438 	skb1->len		   += skb1->data_len;
1439 	skb->data_len		   = 0;
1440 	skb->len		   = len;
1441 	skb->tail		   = skb->data + len;
1442 }
1443 
1444 static inline void skb_split_no_header(struct sk_buff *skb,
1445 				       struct sk_buff* skb1,
1446 				       const u32 len, int pos)
1447 {
1448 	int i, k = 0;
1449 	const int nfrags = skb_shinfo(skb)->nr_frags;
1450 
1451 	skb_shinfo(skb)->nr_frags = 0;
1452 	skb1->len		  = skb1->data_len = skb->len - len;
1453 	skb->len		  = len;
1454 	skb->data_len		  = len - pos;
1455 
1456 	for (i = 0; i < nfrags; i++) {
1457 		int size = skb_shinfo(skb)->frags[i].size;
1458 
1459 		if (pos + size > len) {
1460 			skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1461 
1462 			if (pos < len) {
1463 				/* Split frag.
1464 				 * We have two variants in this case:
1465 				 * 1. Move all the frag to the second
1466 				 *    part, if it is possible. F.e.
1467 				 *    this approach is mandatory for TUX,
1468 				 *    where splitting is expensive.
1469 				 * 2. Split is accurately. We make this.
1470 				 */
1471 				get_page(skb_shinfo(skb)->frags[i].page);
1472 				skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1473 				skb_shinfo(skb1)->frags[0].size -= len - pos;
1474 				skb_shinfo(skb)->frags[i].size	= len - pos;
1475 				skb_shinfo(skb)->nr_frags++;
1476 			}
1477 			k++;
1478 		} else
1479 			skb_shinfo(skb)->nr_frags++;
1480 		pos += size;
1481 	}
1482 	skb_shinfo(skb1)->nr_frags = k;
1483 }
1484 
1485 /**
1486  * skb_split - Split fragmented skb to two parts at length len.
1487  * @skb: the buffer to split
1488  * @skb1: the buffer to receive the second part
1489  * @len: new length for skb
1490  */
1491 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1492 {
1493 	int pos = skb_headlen(skb);
1494 
1495 	if (len < pos)	/* Split line is inside header. */
1496 		skb_split_inside_header(skb, skb1, len, pos);
1497 	else		/* Second chunk has no header, nothing to copy. */
1498 		skb_split_no_header(skb, skb1, len, pos);
1499 }
1500 
1501 /**
1502  * skb_prepare_seq_read - Prepare a sequential read of skb data
1503  * @skb: the buffer to read
1504  * @from: lower offset of data to be read
1505  * @to: upper offset of data to be read
1506  * @st: state variable
1507  *
1508  * Initializes the specified state variable. Must be called before
1509  * invoking skb_seq_read() for the first time.
1510  */
1511 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1512 			  unsigned int to, struct skb_seq_state *st)
1513 {
1514 	st->lower_offset = from;
1515 	st->upper_offset = to;
1516 	st->root_skb = st->cur_skb = skb;
1517 	st->frag_idx = st->stepped_offset = 0;
1518 	st->frag_data = NULL;
1519 }
1520 
1521 /**
1522  * skb_seq_read - Sequentially read skb data
1523  * @consumed: number of bytes consumed by the caller so far
1524  * @data: destination pointer for data to be returned
1525  * @st: state variable
1526  *
1527  * Reads a block of skb data at &consumed relative to the
1528  * lower offset specified to skb_prepare_seq_read(). Assigns
1529  * the head of the data block to &data and returns the length
1530  * of the block or 0 if the end of the skb data or the upper
1531  * offset has been reached.
1532  *
1533  * The caller is not required to consume all of the data
1534  * returned, i.e. &consumed is typically set to the number
1535  * of bytes already consumed and the next call to
1536  * skb_seq_read() will return the remaining part of the block.
1537  *
1538  * Note: The size of each block of data returned can be arbitary,
1539  *       this limitation is the cost for zerocopy seqeuental
1540  *       reads of potentially non linear data.
1541  *
1542  * Note: Fragment lists within fragments are not implemented
1543  *       at the moment, state->root_skb could be replaced with
1544  *       a stack for this purpose.
1545  */
1546 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1547 			  struct skb_seq_state *st)
1548 {
1549 	unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1550 	skb_frag_t *frag;
1551 
1552 	if (unlikely(abs_offset >= st->upper_offset))
1553 		return 0;
1554 
1555 next_skb:
1556 	block_limit = skb_headlen(st->cur_skb);
1557 
1558 	if (abs_offset < block_limit) {
1559 		*data = st->cur_skb->data + abs_offset;
1560 		return block_limit - abs_offset;
1561 	}
1562 
1563 	if (st->frag_idx == 0 && !st->frag_data)
1564 		st->stepped_offset += skb_headlen(st->cur_skb);
1565 
1566 	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1567 		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1568 		block_limit = frag->size + st->stepped_offset;
1569 
1570 		if (abs_offset < block_limit) {
1571 			if (!st->frag_data)
1572 				st->frag_data = kmap_skb_frag(frag);
1573 
1574 			*data = (u8 *) st->frag_data + frag->page_offset +
1575 				(abs_offset - st->stepped_offset);
1576 
1577 			return block_limit - abs_offset;
1578 		}
1579 
1580 		if (st->frag_data) {
1581 			kunmap_skb_frag(st->frag_data);
1582 			st->frag_data = NULL;
1583 		}
1584 
1585 		st->frag_idx++;
1586 		st->stepped_offset += frag->size;
1587 	}
1588 
1589 	if (st->cur_skb->next) {
1590 		st->cur_skb = st->cur_skb->next;
1591 		st->frag_idx = 0;
1592 		goto next_skb;
1593 	} else if (st->root_skb == st->cur_skb &&
1594 		   skb_shinfo(st->root_skb)->frag_list) {
1595 		st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1596 		goto next_skb;
1597 	}
1598 
1599 	return 0;
1600 }
1601 
1602 /**
1603  * skb_abort_seq_read - Abort a sequential read of skb data
1604  * @st: state variable
1605  *
1606  * Must be called if skb_seq_read() was not called until it
1607  * returned 0.
1608  */
1609 void skb_abort_seq_read(struct skb_seq_state *st)
1610 {
1611 	if (st->frag_data)
1612 		kunmap_skb_frag(st->frag_data);
1613 }
1614 
1615 #define TS_SKB_CB(state)	((struct skb_seq_state *) &((state)->cb))
1616 
1617 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1618 					  struct ts_config *conf,
1619 					  struct ts_state *state)
1620 {
1621 	return skb_seq_read(offset, text, TS_SKB_CB(state));
1622 }
1623 
1624 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1625 {
1626 	skb_abort_seq_read(TS_SKB_CB(state));
1627 }
1628 
1629 /**
1630  * skb_find_text - Find a text pattern in skb data
1631  * @skb: the buffer to look in
1632  * @from: search offset
1633  * @to: search limit
1634  * @config: textsearch configuration
1635  * @state: uninitialized textsearch state variable
1636  *
1637  * Finds a pattern in the skb data according to the specified
1638  * textsearch configuration. Use textsearch_next() to retrieve
1639  * subsequent occurrences of the pattern. Returns the offset
1640  * to the first occurrence or UINT_MAX if no match was found.
1641  */
1642 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1643 			   unsigned int to, struct ts_config *config,
1644 			   struct ts_state *state)
1645 {
1646 	config->get_next_block = skb_ts_get_next_block;
1647 	config->finish = skb_ts_finish;
1648 
1649 	skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1650 
1651 	return textsearch_find(config, state);
1652 }
1653 
1654 void __init skb_init(void)
1655 {
1656 	skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1657 					      sizeof(struct sk_buff),
1658 					      0,
1659 					      SLAB_HWCACHE_ALIGN,
1660 					      NULL, NULL);
1661 	if (!skbuff_head_cache)
1662 		panic("cannot create skbuff cache");
1663 }
1664 
1665 EXPORT_SYMBOL(___pskb_trim);
1666 EXPORT_SYMBOL(__kfree_skb);
1667 EXPORT_SYMBOL(__pskb_pull_tail);
1668 EXPORT_SYMBOL(alloc_skb);
1669 EXPORT_SYMBOL(pskb_copy);
1670 EXPORT_SYMBOL(pskb_expand_head);
1671 EXPORT_SYMBOL(skb_checksum);
1672 EXPORT_SYMBOL(skb_clone);
1673 EXPORT_SYMBOL(skb_clone_fraglist);
1674 EXPORT_SYMBOL(skb_copy);
1675 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1676 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1677 EXPORT_SYMBOL(skb_copy_bits);
1678 EXPORT_SYMBOL(skb_copy_expand);
1679 EXPORT_SYMBOL(skb_over_panic);
1680 EXPORT_SYMBOL(skb_pad);
1681 EXPORT_SYMBOL(skb_realloc_headroom);
1682 EXPORT_SYMBOL(skb_under_panic);
1683 EXPORT_SYMBOL(skb_dequeue);
1684 EXPORT_SYMBOL(skb_dequeue_tail);
1685 EXPORT_SYMBOL(skb_insert);
1686 EXPORT_SYMBOL(skb_queue_purge);
1687 EXPORT_SYMBOL(skb_queue_head);
1688 EXPORT_SYMBOL(skb_queue_tail);
1689 EXPORT_SYMBOL(skb_unlink);
1690 EXPORT_SYMBOL(skb_append);
1691 EXPORT_SYMBOL(skb_split);
1692 EXPORT_SYMBOL(skb_prepare_seq_read);
1693 EXPORT_SYMBOL(skb_seq_read);
1694 EXPORT_SYMBOL(skb_abort_seq_read);
1695 EXPORT_SYMBOL(skb_find_text);
1696