xref: /linux/net/core/skbuff.c (revision 54a8a2220c936a47840c9a3d74910c5a56fae2ed)
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 __read_mostly;
72 static kmem_cache_t *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:%p end:%p dev:%s\n",
92 	       here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
93 	       skb->dev ? skb->dev->name : "<NULL>");
94 	BUG();
95 }
96 
97 /**
98  *	skb_under_panic	- 	private function
99  *	@skb: buffer
100  *	@sz: size
101  *	@here: address
102  *
103  *	Out of line support code for skb_push(). Not user callable.
104  */
105 
106 void skb_under_panic(struct sk_buff *skb, int sz, void *here)
107 {
108 	printk(KERN_EMERG "skb_under_panic: text:%p len:%d put:%d head:%p "
109 	                  "data:%p tail:%p end:%p dev:%s\n",
110 	       here, skb->len, sz, skb->head, skb->data, skb->tail, skb->end,
111 	       skb->dev ? skb->dev->name : "<NULL>");
112 	BUG();
113 }
114 
115 /* 	Allocate a new skbuff. We do this ourselves so we can fill in a few
116  *	'private' fields and also do memory statistics to find all the
117  *	[BEEP] leaks.
118  *
119  */
120 
121 /**
122  *	__alloc_skb	-	allocate a network buffer
123  *	@size: size to allocate
124  *	@gfp_mask: allocation mask
125  *
126  *	Allocate a new &sk_buff. The returned buffer has no headroom and a
127  *	tail room of size bytes. The object has a reference count of one.
128  *	The return is the buffer. On a failure the return is %NULL.
129  *
130  *	Buffers may only be allocated from interrupts using a @gfp_mask of
131  *	%GFP_ATOMIC.
132  */
133 struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
134 			    int fclone)
135 {
136 	struct sk_buff *skb;
137 	u8 *data;
138 
139 	/* Get the HEAD */
140 	if (fclone)
141 		skb = kmem_cache_alloc(skbuff_fclone_cache,
142 				       gfp_mask & ~__GFP_DMA);
143 	else
144 		skb = kmem_cache_alloc(skbuff_head_cache,
145 				       gfp_mask & ~__GFP_DMA);
146 
147 	if (!skb)
148 		goto out;
149 
150 	/* Get the DATA. Size must match skb_add_mtu(). */
151 	size = SKB_DATA_ALIGN(size);
152 	data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
153 	if (!data)
154 		goto nodata;
155 
156 	memset(skb, 0, offsetof(struct sk_buff, truesize));
157 	skb->truesize = size + sizeof(struct sk_buff);
158 	atomic_set(&skb->users, 1);
159 	skb->head = data;
160 	skb->data = data;
161 	skb->tail = data;
162 	skb->end  = data + size;
163 	if (fclone) {
164 		struct sk_buff *child = skb + 1;
165 		atomic_t *fclone_ref = (atomic_t *) (child + 1);
166 
167 		skb->fclone = SKB_FCLONE_ORIG;
168 		atomic_set(fclone_ref, 1);
169 
170 		child->fclone = SKB_FCLONE_UNAVAILABLE;
171 	}
172 	atomic_set(&(skb_shinfo(skb)->dataref), 1);
173 	skb_shinfo(skb)->nr_frags  = 0;
174 	skb_shinfo(skb)->tso_size = 0;
175 	skb_shinfo(skb)->tso_segs = 0;
176 	skb_shinfo(skb)->frag_list = NULL;
177 out:
178 	return skb;
179 nodata:
180 	kmem_cache_free(skbuff_head_cache, skb);
181 	skb = NULL;
182 	goto out;
183 }
184 
185 /**
186  *	alloc_skb_from_cache	-	allocate a network buffer
187  *	@cp: kmem_cache from which to allocate the data area
188  *           (object size must be big enough for @size bytes + skb overheads)
189  *	@size: size to allocate
190  *	@gfp_mask: allocation mask
191  *
192  *	Allocate a new &sk_buff. The returned buffer has no headroom and
193  *	tail room of size bytes. The object has a reference count of one.
194  *	The return is the buffer. On a failure the return is %NULL.
195  *
196  *	Buffers may only be allocated from interrupts using a @gfp_mask of
197  *	%GFP_ATOMIC.
198  */
199 struct sk_buff *alloc_skb_from_cache(kmem_cache_t *cp,
200 				     unsigned int size,
201 				     gfp_t gfp_mask)
202 {
203 	struct sk_buff *skb;
204 	u8 *data;
205 
206 	/* Get the HEAD */
207 	skb = kmem_cache_alloc(skbuff_head_cache,
208 			       gfp_mask & ~__GFP_DMA);
209 	if (!skb)
210 		goto out;
211 
212 	/* Get the DATA. */
213 	size = SKB_DATA_ALIGN(size);
214 	data = kmem_cache_alloc(cp, gfp_mask);
215 	if (!data)
216 		goto nodata;
217 
218 	memset(skb, 0, offsetof(struct sk_buff, truesize));
219 	skb->truesize = size + sizeof(struct sk_buff);
220 	atomic_set(&skb->users, 1);
221 	skb->head = data;
222 	skb->data = data;
223 	skb->tail = data;
224 	skb->end  = data + size;
225 
226 	atomic_set(&(skb_shinfo(skb)->dataref), 1);
227 	skb_shinfo(skb)->nr_frags  = 0;
228 	skb_shinfo(skb)->tso_size = 0;
229 	skb_shinfo(skb)->tso_segs = 0;
230 	skb_shinfo(skb)->frag_list = NULL;
231 out:
232 	return skb;
233 nodata:
234 	kmem_cache_free(skbuff_head_cache, skb);
235 	skb = NULL;
236 	goto out;
237 }
238 
239 
240 static void skb_drop_fraglist(struct sk_buff *skb)
241 {
242 	struct sk_buff *list = skb_shinfo(skb)->frag_list;
243 
244 	skb_shinfo(skb)->frag_list = NULL;
245 
246 	do {
247 		struct sk_buff *this = list;
248 		list = list->next;
249 		kfree_skb(this);
250 	} while (list);
251 }
252 
253 static void skb_clone_fraglist(struct sk_buff *skb)
254 {
255 	struct sk_buff *list;
256 
257 	for (list = skb_shinfo(skb)->frag_list; list; list = list->next)
258 		skb_get(list);
259 }
260 
261 void skb_release_data(struct sk_buff *skb)
262 {
263 	if (!skb->cloned ||
264 	    !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
265 			       &skb_shinfo(skb)->dataref)) {
266 		if (skb_shinfo(skb)->nr_frags) {
267 			int i;
268 			for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
269 				put_page(skb_shinfo(skb)->frags[i].page);
270 		}
271 
272 		if (skb_shinfo(skb)->frag_list)
273 			skb_drop_fraglist(skb);
274 
275 		kfree(skb->head);
276 	}
277 }
278 
279 /*
280  *	Free an skbuff by memory without cleaning the state.
281  */
282 void kfree_skbmem(struct sk_buff *skb)
283 {
284 	struct sk_buff *other;
285 	atomic_t *fclone_ref;
286 
287 	skb_release_data(skb);
288 	switch (skb->fclone) {
289 	case SKB_FCLONE_UNAVAILABLE:
290 		kmem_cache_free(skbuff_head_cache, skb);
291 		break;
292 
293 	case SKB_FCLONE_ORIG:
294 		fclone_ref = (atomic_t *) (skb + 2);
295 		if (atomic_dec_and_test(fclone_ref))
296 			kmem_cache_free(skbuff_fclone_cache, skb);
297 		break;
298 
299 	case SKB_FCLONE_CLONE:
300 		fclone_ref = (atomic_t *) (skb + 1);
301 		other = skb - 1;
302 
303 		/* The clone portion is available for
304 		 * fast-cloning again.
305 		 */
306 		skb->fclone = SKB_FCLONE_UNAVAILABLE;
307 
308 		if (atomic_dec_and_test(fclone_ref))
309 			kmem_cache_free(skbuff_fclone_cache, other);
310 		break;
311 	};
312 }
313 
314 /**
315  *	__kfree_skb - private function
316  *	@skb: buffer
317  *
318  *	Free an sk_buff. Release anything attached to the buffer.
319  *	Clean the state. This is an internal helper function. Users should
320  *	always call kfree_skb
321  */
322 
323 void __kfree_skb(struct sk_buff *skb)
324 {
325 	dst_release(skb->dst);
326 #ifdef CONFIG_XFRM
327 	secpath_put(skb->sp);
328 #endif
329 	if (skb->destructor) {
330 		WARN_ON(in_irq());
331 		skb->destructor(skb);
332 	}
333 #ifdef CONFIG_NETFILTER
334 	nf_conntrack_put(skb->nfct);
335 #ifdef CONFIG_BRIDGE_NETFILTER
336 	nf_bridge_put(skb->nf_bridge);
337 #endif
338 #endif
339 /* XXX: IS this still necessary? - JHS */
340 #ifdef CONFIG_NET_SCHED
341 	skb->tc_index = 0;
342 #ifdef CONFIG_NET_CLS_ACT
343 	skb->tc_verd = 0;
344 #endif
345 #endif
346 
347 	kfree_skbmem(skb);
348 }
349 
350 /**
351  *	skb_clone	-	duplicate an sk_buff
352  *	@skb: buffer to clone
353  *	@gfp_mask: allocation priority
354  *
355  *	Duplicate an &sk_buff. The new one is not owned by a socket. Both
356  *	copies share the same packet data but not structure. The new
357  *	buffer has a reference count of 1. If the allocation fails the
358  *	function returns %NULL otherwise the new buffer is returned.
359  *
360  *	If this function is called from an interrupt gfp_mask() must be
361  *	%GFP_ATOMIC.
362  */
363 
364 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
365 {
366 	struct sk_buff *n;
367 
368 	n = skb + 1;
369 	if (skb->fclone == SKB_FCLONE_ORIG &&
370 	    n->fclone == SKB_FCLONE_UNAVAILABLE) {
371 		atomic_t *fclone_ref = (atomic_t *) (n + 1);
372 		n->fclone = SKB_FCLONE_CLONE;
373 		atomic_inc(fclone_ref);
374 	} else {
375 		n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
376 		if (!n)
377 			return NULL;
378 		n->fclone = SKB_FCLONE_UNAVAILABLE;
379 	}
380 
381 #define C(x) n->x = skb->x
382 
383 	n->next = n->prev = NULL;
384 	n->sk = NULL;
385 	C(tstamp);
386 	C(dev);
387 	C(h);
388 	C(nh);
389 	C(mac);
390 	C(dst);
391 	dst_clone(skb->dst);
392 	C(sp);
393 #ifdef CONFIG_INET
394 	secpath_get(skb->sp);
395 #endif
396 	memcpy(n->cb, skb->cb, sizeof(skb->cb));
397 	C(len);
398 	C(data_len);
399 	C(csum);
400 	C(local_df);
401 	n->cloned = 1;
402 	n->nohdr = 0;
403 	C(pkt_type);
404 	C(ip_summed);
405 	C(priority);
406 	C(protocol);
407 	n->destructor = NULL;
408 #ifdef CONFIG_NETFILTER
409 	C(nfmark);
410 	C(nfct);
411 	nf_conntrack_get(skb->nfct);
412 	C(nfctinfo);
413 #ifdef CONFIG_BRIDGE_NETFILTER
414 	C(nf_bridge);
415 	nf_bridge_get(skb->nf_bridge);
416 #endif
417 #endif /*CONFIG_NETFILTER*/
418 #ifdef CONFIG_NET_SCHED
419 	C(tc_index);
420 #ifdef CONFIG_NET_CLS_ACT
421 	n->tc_verd = SET_TC_VERD(skb->tc_verd,0);
422 	n->tc_verd = CLR_TC_OK2MUNGE(n->tc_verd);
423 	n->tc_verd = CLR_TC_MUNGED(n->tc_verd);
424 	C(input_dev);
425 #endif
426 
427 #endif
428 	C(truesize);
429 	atomic_set(&n->users, 1);
430 	C(head);
431 	C(data);
432 	C(tail);
433 	C(end);
434 
435 	atomic_inc(&(skb_shinfo(skb)->dataref));
436 	skb->cloned = 1;
437 
438 	return n;
439 }
440 
441 static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
442 {
443 	/*
444 	 *	Shift between the two data areas in bytes
445 	 */
446 	unsigned long offset = new->data - old->data;
447 
448 	new->sk		= NULL;
449 	new->dev	= old->dev;
450 	new->priority	= old->priority;
451 	new->protocol	= old->protocol;
452 	new->dst	= dst_clone(old->dst);
453 #ifdef CONFIG_INET
454 	new->sp		= secpath_get(old->sp);
455 #endif
456 	new->h.raw	= old->h.raw + offset;
457 	new->nh.raw	= old->nh.raw + offset;
458 	new->mac.raw	= old->mac.raw + offset;
459 	memcpy(new->cb, old->cb, sizeof(old->cb));
460 	new->local_df	= old->local_df;
461 	new->fclone	= SKB_FCLONE_UNAVAILABLE;
462 	new->pkt_type	= old->pkt_type;
463 	new->tstamp	= old->tstamp;
464 	new->destructor = NULL;
465 #ifdef CONFIG_NETFILTER
466 	new->nfmark	= old->nfmark;
467 	new->nfct	= old->nfct;
468 	nf_conntrack_get(old->nfct);
469 	new->nfctinfo	= old->nfctinfo;
470 #ifdef CONFIG_BRIDGE_NETFILTER
471 	new->nf_bridge	= old->nf_bridge;
472 	nf_bridge_get(old->nf_bridge);
473 #endif
474 #endif
475 #ifdef CONFIG_NET_SCHED
476 #ifdef CONFIG_NET_CLS_ACT
477 	new->tc_verd = old->tc_verd;
478 #endif
479 	new->tc_index	= old->tc_index;
480 #endif
481 	atomic_set(&new->users, 1);
482 	skb_shinfo(new)->tso_size = skb_shinfo(old)->tso_size;
483 	skb_shinfo(new)->tso_segs = skb_shinfo(old)->tso_segs;
484 }
485 
486 /**
487  *	skb_copy	-	create private copy of an sk_buff
488  *	@skb: buffer to copy
489  *	@gfp_mask: allocation priority
490  *
491  *	Make a copy of both an &sk_buff and its data. This is used when the
492  *	caller wishes to modify the data and needs a private copy of the
493  *	data to alter. Returns %NULL on failure or the pointer to the buffer
494  *	on success. The returned buffer has a reference count of 1.
495  *
496  *	As by-product this function converts non-linear &sk_buff to linear
497  *	one, so that &sk_buff becomes completely private and caller is allowed
498  *	to modify all the data of returned buffer. This means that this
499  *	function is not recommended for use in circumstances when only
500  *	header is going to be modified. Use pskb_copy() instead.
501  */
502 
503 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
504 {
505 	int headerlen = skb->data - skb->head;
506 	/*
507 	 *	Allocate the copy buffer
508 	 */
509 	struct sk_buff *n = alloc_skb(skb->end - skb->head + skb->data_len,
510 				      gfp_mask);
511 	if (!n)
512 		return NULL;
513 
514 	/* Set the data pointer */
515 	skb_reserve(n, headerlen);
516 	/* Set the tail pointer and length */
517 	skb_put(n, skb->len);
518 	n->csum	     = skb->csum;
519 	n->ip_summed = skb->ip_summed;
520 
521 	if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
522 		BUG();
523 
524 	copy_skb_header(n, skb);
525 	return n;
526 }
527 
528 
529 /**
530  *	pskb_copy	-	create copy of an sk_buff with private head.
531  *	@skb: buffer to copy
532  *	@gfp_mask: allocation priority
533  *
534  *	Make a copy of both an &sk_buff and part of its data, located
535  *	in header. Fragmented data remain shared. This is used when
536  *	the caller wishes to modify only header of &sk_buff and needs
537  *	private copy of the header to alter. Returns %NULL on failure
538  *	or the pointer to the buffer on success.
539  *	The returned buffer has a reference count of 1.
540  */
541 
542 struct sk_buff *pskb_copy(struct sk_buff *skb, gfp_t gfp_mask)
543 {
544 	/*
545 	 *	Allocate the copy buffer
546 	 */
547 	struct sk_buff *n = alloc_skb(skb->end - skb->head, gfp_mask);
548 
549 	if (!n)
550 		goto out;
551 
552 	/* Set the data pointer */
553 	skb_reserve(n, skb->data - skb->head);
554 	/* Set the tail pointer and length */
555 	skb_put(n, skb_headlen(skb));
556 	/* Copy the bytes */
557 	memcpy(n->data, skb->data, n->len);
558 	n->csum	     = skb->csum;
559 	n->ip_summed = skb->ip_summed;
560 
561 	n->data_len  = skb->data_len;
562 	n->len	     = skb->len;
563 
564 	if (skb_shinfo(skb)->nr_frags) {
565 		int i;
566 
567 		for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
568 			skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
569 			get_page(skb_shinfo(n)->frags[i].page);
570 		}
571 		skb_shinfo(n)->nr_frags = i;
572 	}
573 
574 	if (skb_shinfo(skb)->frag_list) {
575 		skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
576 		skb_clone_fraglist(n);
577 	}
578 
579 	copy_skb_header(n, skb);
580 out:
581 	return n;
582 }
583 
584 /**
585  *	pskb_expand_head - reallocate header of &sk_buff
586  *	@skb: buffer to reallocate
587  *	@nhead: room to add at head
588  *	@ntail: room to add at tail
589  *	@gfp_mask: allocation priority
590  *
591  *	Expands (or creates identical copy, if &nhead and &ntail are zero)
592  *	header of skb. &sk_buff itself is not changed. &sk_buff MUST have
593  *	reference count of 1. Returns zero in the case of success or error,
594  *	if expansion failed. In the last case, &sk_buff is not changed.
595  *
596  *	All the pointers pointing into skb header may change and must be
597  *	reloaded after call to this function.
598  */
599 
600 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
601 		     gfp_t gfp_mask)
602 {
603 	int i;
604 	u8 *data;
605 	int size = nhead + (skb->end - skb->head) + ntail;
606 	long off;
607 
608 	if (skb_shared(skb))
609 		BUG();
610 
611 	size = SKB_DATA_ALIGN(size);
612 
613 	data = kmalloc(size + sizeof(struct skb_shared_info), gfp_mask);
614 	if (!data)
615 		goto nodata;
616 
617 	/* Copy only real data... and, alas, header. This should be
618 	 * optimized for the cases when header is void. */
619 	memcpy(data + nhead, skb->head, skb->tail - skb->head);
620 	memcpy(data + size, skb->end, sizeof(struct skb_shared_info));
621 
622 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
623 		get_page(skb_shinfo(skb)->frags[i].page);
624 
625 	if (skb_shinfo(skb)->frag_list)
626 		skb_clone_fraglist(skb);
627 
628 	skb_release_data(skb);
629 
630 	off = (data + nhead) - skb->head;
631 
632 	skb->head     = data;
633 	skb->end      = data + size;
634 	skb->data    += off;
635 	skb->tail    += off;
636 	skb->mac.raw += off;
637 	skb->h.raw   += off;
638 	skb->nh.raw  += off;
639 	skb->cloned   = 0;
640 	skb->nohdr    = 0;
641 	atomic_set(&skb_shinfo(skb)->dataref, 1);
642 	return 0;
643 
644 nodata:
645 	return -ENOMEM;
646 }
647 
648 /* Make private copy of skb with writable head and some headroom */
649 
650 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
651 {
652 	struct sk_buff *skb2;
653 	int delta = headroom - skb_headroom(skb);
654 
655 	if (delta <= 0)
656 		skb2 = pskb_copy(skb, GFP_ATOMIC);
657 	else {
658 		skb2 = skb_clone(skb, GFP_ATOMIC);
659 		if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
660 					     GFP_ATOMIC)) {
661 			kfree_skb(skb2);
662 			skb2 = NULL;
663 		}
664 	}
665 	return skb2;
666 }
667 
668 
669 /**
670  *	skb_copy_expand	-	copy and expand sk_buff
671  *	@skb: buffer to copy
672  *	@newheadroom: new free bytes at head
673  *	@newtailroom: new free bytes at tail
674  *	@gfp_mask: allocation priority
675  *
676  *	Make a copy of both an &sk_buff and its data and while doing so
677  *	allocate additional space.
678  *
679  *	This is used when the caller wishes to modify the data and needs a
680  *	private copy of the data to alter as well as more space for new fields.
681  *	Returns %NULL on failure or the pointer to the buffer
682  *	on success. The returned buffer has a reference count of 1.
683  *
684  *	You must pass %GFP_ATOMIC as the allocation priority if this function
685  *	is called from an interrupt.
686  *
687  *	BUG ALERT: ip_summed is not copied. Why does this work? Is it used
688  *	only by netfilter in the cases when checksum is recalculated? --ANK
689  */
690 struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
691 				int newheadroom, int newtailroom,
692 				gfp_t gfp_mask)
693 {
694 	/*
695 	 *	Allocate the copy buffer
696 	 */
697 	struct sk_buff *n = alloc_skb(newheadroom + skb->len + newtailroom,
698 				      gfp_mask);
699 	int head_copy_len, head_copy_off;
700 
701 	if (!n)
702 		return NULL;
703 
704 	skb_reserve(n, newheadroom);
705 
706 	/* Set the tail pointer and length */
707 	skb_put(n, skb->len);
708 
709 	head_copy_len = skb_headroom(skb);
710 	head_copy_off = 0;
711 	if (newheadroom <= head_copy_len)
712 		head_copy_len = newheadroom;
713 	else
714 		head_copy_off = newheadroom - head_copy_len;
715 
716 	/* Copy the linear header and data. */
717 	if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
718 			  skb->len + head_copy_len))
719 		BUG();
720 
721 	copy_skb_header(n, skb);
722 
723 	return n;
724 }
725 
726 /**
727  *	skb_pad			-	zero pad the tail of an skb
728  *	@skb: buffer to pad
729  *	@pad: space to pad
730  *
731  *	Ensure that a buffer is followed by a padding area that is zero
732  *	filled. Used by network drivers which may DMA or transfer data
733  *	beyond the buffer end onto the wire.
734  *
735  *	May return NULL in out of memory cases.
736  */
737 
738 struct sk_buff *skb_pad(struct sk_buff *skb, int pad)
739 {
740 	struct sk_buff *nskb;
741 
742 	/* If the skbuff is non linear tailroom is always zero.. */
743 	if (skb_tailroom(skb) >= pad) {
744 		memset(skb->data+skb->len, 0, pad);
745 		return skb;
746 	}
747 
748 	nskb = skb_copy_expand(skb, skb_headroom(skb), skb_tailroom(skb) + pad, GFP_ATOMIC);
749 	kfree_skb(skb);
750 	if (nskb)
751 		memset(nskb->data+nskb->len, 0, pad);
752 	return nskb;
753 }
754 
755 /* Trims skb to length len. It can change skb pointers, if "realloc" is 1.
756  * If realloc==0 and trimming is impossible without change of data,
757  * it is BUG().
758  */
759 
760 int ___pskb_trim(struct sk_buff *skb, unsigned int len, int realloc)
761 {
762 	int offset = skb_headlen(skb);
763 	int nfrags = skb_shinfo(skb)->nr_frags;
764 	int i;
765 
766 	for (i = 0; i < nfrags; i++) {
767 		int end = offset + skb_shinfo(skb)->frags[i].size;
768 		if (end > len) {
769 			if (skb_cloned(skb)) {
770 				if (!realloc)
771 					BUG();
772 				if (pskb_expand_head(skb, 0, 0, GFP_ATOMIC))
773 					return -ENOMEM;
774 			}
775 			if (len <= offset) {
776 				put_page(skb_shinfo(skb)->frags[i].page);
777 				skb_shinfo(skb)->nr_frags--;
778 			} else {
779 				skb_shinfo(skb)->frags[i].size = len - offset;
780 			}
781 		}
782 		offset = end;
783 	}
784 
785 	if (offset < len) {
786 		skb->data_len -= skb->len - len;
787 		skb->len       = len;
788 	} else {
789 		if (len <= skb_headlen(skb)) {
790 			skb->len      = len;
791 			skb->data_len = 0;
792 			skb->tail     = skb->data + len;
793 			if (skb_shinfo(skb)->frag_list && !skb_cloned(skb))
794 				skb_drop_fraglist(skb);
795 		} else {
796 			skb->data_len -= skb->len - len;
797 			skb->len       = len;
798 		}
799 	}
800 
801 	return 0;
802 }
803 
804 /**
805  *	__pskb_pull_tail - advance tail of skb header
806  *	@skb: buffer to reallocate
807  *	@delta: number of bytes to advance tail
808  *
809  *	The function makes a sense only on a fragmented &sk_buff,
810  *	it expands header moving its tail forward and copying necessary
811  *	data from fragmented part.
812  *
813  *	&sk_buff MUST have reference count of 1.
814  *
815  *	Returns %NULL (and &sk_buff does not change) if pull failed
816  *	or value of new tail of skb in the case of success.
817  *
818  *	All the pointers pointing into skb header may change and must be
819  *	reloaded after call to this function.
820  */
821 
822 /* Moves tail of skb head forward, copying data from fragmented part,
823  * when it is necessary.
824  * 1. It may fail due to malloc failure.
825  * 2. It may change skb pointers.
826  *
827  * It is pretty complicated. Luckily, it is called only in exceptional cases.
828  */
829 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
830 {
831 	/* If skb has not enough free space at tail, get new one
832 	 * plus 128 bytes for future expansions. If we have enough
833 	 * room at tail, reallocate without expansion only if skb is cloned.
834 	 */
835 	int i, k, eat = (skb->tail + delta) - skb->end;
836 
837 	if (eat > 0 || skb_cloned(skb)) {
838 		if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
839 				     GFP_ATOMIC))
840 			return NULL;
841 	}
842 
843 	if (skb_copy_bits(skb, skb_headlen(skb), skb->tail, delta))
844 		BUG();
845 
846 	/* Optimization: no fragments, no reasons to preestimate
847 	 * size of pulled pages. Superb.
848 	 */
849 	if (!skb_shinfo(skb)->frag_list)
850 		goto pull_pages;
851 
852 	/* Estimate size of pulled pages. */
853 	eat = delta;
854 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
855 		if (skb_shinfo(skb)->frags[i].size >= eat)
856 			goto pull_pages;
857 		eat -= skb_shinfo(skb)->frags[i].size;
858 	}
859 
860 	/* If we need update frag list, we are in troubles.
861 	 * Certainly, it possible to add an offset to skb data,
862 	 * but taking into account that pulling is expected to
863 	 * be very rare operation, it is worth to fight against
864 	 * further bloating skb head and crucify ourselves here instead.
865 	 * Pure masohism, indeed. 8)8)
866 	 */
867 	if (eat) {
868 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
869 		struct sk_buff *clone = NULL;
870 		struct sk_buff *insp = NULL;
871 
872 		do {
873 			if (!list)
874 				BUG();
875 
876 			if (list->len <= eat) {
877 				/* Eaten as whole. */
878 				eat -= list->len;
879 				list = list->next;
880 				insp = list;
881 			} else {
882 				/* Eaten partially. */
883 
884 				if (skb_shared(list)) {
885 					/* Sucks! We need to fork list. :-( */
886 					clone = skb_clone(list, GFP_ATOMIC);
887 					if (!clone)
888 						return NULL;
889 					insp = list->next;
890 					list = clone;
891 				} else {
892 					/* This may be pulled without
893 					 * problems. */
894 					insp = list;
895 				}
896 				if (!pskb_pull(list, eat)) {
897 					if (clone)
898 						kfree_skb(clone);
899 					return NULL;
900 				}
901 				break;
902 			}
903 		} while (eat);
904 
905 		/* Free pulled out fragments. */
906 		while ((list = skb_shinfo(skb)->frag_list) != insp) {
907 			skb_shinfo(skb)->frag_list = list->next;
908 			kfree_skb(list);
909 		}
910 		/* And insert new clone at head. */
911 		if (clone) {
912 			clone->next = list;
913 			skb_shinfo(skb)->frag_list = clone;
914 		}
915 	}
916 	/* Success! Now we may commit changes to skb data. */
917 
918 pull_pages:
919 	eat = delta;
920 	k = 0;
921 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
922 		if (skb_shinfo(skb)->frags[i].size <= eat) {
923 			put_page(skb_shinfo(skb)->frags[i].page);
924 			eat -= skb_shinfo(skb)->frags[i].size;
925 		} else {
926 			skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
927 			if (eat) {
928 				skb_shinfo(skb)->frags[k].page_offset += eat;
929 				skb_shinfo(skb)->frags[k].size -= eat;
930 				eat = 0;
931 			}
932 			k++;
933 		}
934 	}
935 	skb_shinfo(skb)->nr_frags = k;
936 
937 	skb->tail     += delta;
938 	skb->data_len -= delta;
939 
940 	return skb->tail;
941 }
942 
943 /* Copy some data bits from skb to kernel buffer. */
944 
945 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
946 {
947 	int i, copy;
948 	int start = skb_headlen(skb);
949 
950 	if (offset > (int)skb->len - len)
951 		goto fault;
952 
953 	/* Copy header. */
954 	if ((copy = start - offset) > 0) {
955 		if (copy > len)
956 			copy = len;
957 		memcpy(to, skb->data + offset, copy);
958 		if ((len -= copy) == 0)
959 			return 0;
960 		offset += copy;
961 		to     += copy;
962 	}
963 
964 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
965 		int end;
966 
967 		BUG_TRAP(start <= offset + len);
968 
969 		end = start + skb_shinfo(skb)->frags[i].size;
970 		if ((copy = end - offset) > 0) {
971 			u8 *vaddr;
972 
973 			if (copy > len)
974 				copy = len;
975 
976 			vaddr = kmap_skb_frag(&skb_shinfo(skb)->frags[i]);
977 			memcpy(to,
978 			       vaddr + skb_shinfo(skb)->frags[i].page_offset+
979 			       offset - start, copy);
980 			kunmap_skb_frag(vaddr);
981 
982 			if ((len -= copy) == 0)
983 				return 0;
984 			offset += copy;
985 			to     += copy;
986 		}
987 		start = end;
988 	}
989 
990 	if (skb_shinfo(skb)->frag_list) {
991 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
992 
993 		for (; list; list = list->next) {
994 			int end;
995 
996 			BUG_TRAP(start <= offset + len);
997 
998 			end = start + list->len;
999 			if ((copy = end - offset) > 0) {
1000 				if (copy > len)
1001 					copy = len;
1002 				if (skb_copy_bits(list, offset - start,
1003 						  to, copy))
1004 					goto fault;
1005 				if ((len -= copy) == 0)
1006 					return 0;
1007 				offset += copy;
1008 				to     += copy;
1009 			}
1010 			start = end;
1011 		}
1012 	}
1013 	if (!len)
1014 		return 0;
1015 
1016 fault:
1017 	return -EFAULT;
1018 }
1019 
1020 /**
1021  *	skb_store_bits - store bits from kernel buffer to skb
1022  *	@skb: destination buffer
1023  *	@offset: offset in destination
1024  *	@from: source buffer
1025  *	@len: number of bytes to copy
1026  *
1027  *	Copy the specified number of bytes from the source buffer to the
1028  *	destination skb.  This function handles all the messy bits of
1029  *	traversing fragment lists and such.
1030  */
1031 
1032 int skb_store_bits(const struct sk_buff *skb, int offset, void *from, int len)
1033 {
1034 	int i, copy;
1035 	int start = skb_headlen(skb);
1036 
1037 	if (offset > (int)skb->len - len)
1038 		goto fault;
1039 
1040 	if ((copy = start - offset) > 0) {
1041 		if (copy > len)
1042 			copy = len;
1043 		memcpy(skb->data + offset, from, copy);
1044 		if ((len -= copy) == 0)
1045 			return 0;
1046 		offset += copy;
1047 		from += copy;
1048 	}
1049 
1050 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1051 		skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1052 		int end;
1053 
1054 		BUG_TRAP(start <= offset + len);
1055 
1056 		end = start + frag->size;
1057 		if ((copy = end - offset) > 0) {
1058 			u8 *vaddr;
1059 
1060 			if (copy > len)
1061 				copy = len;
1062 
1063 			vaddr = kmap_skb_frag(frag);
1064 			memcpy(vaddr + frag->page_offset + offset - start,
1065 			       from, copy);
1066 			kunmap_skb_frag(vaddr);
1067 
1068 			if ((len -= copy) == 0)
1069 				return 0;
1070 			offset += copy;
1071 			from += copy;
1072 		}
1073 		start = end;
1074 	}
1075 
1076 	if (skb_shinfo(skb)->frag_list) {
1077 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1078 
1079 		for (; list; list = list->next) {
1080 			int end;
1081 
1082 			BUG_TRAP(start <= offset + len);
1083 
1084 			end = start + list->len;
1085 			if ((copy = end - offset) > 0) {
1086 				if (copy > len)
1087 					copy = len;
1088 				if (skb_store_bits(list, offset - start,
1089 						   from, copy))
1090 					goto fault;
1091 				if ((len -= copy) == 0)
1092 					return 0;
1093 				offset += copy;
1094 				from += copy;
1095 			}
1096 			start = end;
1097 		}
1098 	}
1099 	if (!len)
1100 		return 0;
1101 
1102 fault:
1103 	return -EFAULT;
1104 }
1105 
1106 EXPORT_SYMBOL(skb_store_bits);
1107 
1108 /* Checksum skb data. */
1109 
1110 unsigned int skb_checksum(const struct sk_buff *skb, int offset,
1111 			  int len, unsigned int csum)
1112 {
1113 	int start = skb_headlen(skb);
1114 	int i, copy = start - offset;
1115 	int pos = 0;
1116 
1117 	/* Checksum header. */
1118 	if (copy > 0) {
1119 		if (copy > len)
1120 			copy = len;
1121 		csum = csum_partial(skb->data + offset, copy, csum);
1122 		if ((len -= copy) == 0)
1123 			return csum;
1124 		offset += copy;
1125 		pos	= copy;
1126 	}
1127 
1128 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1129 		int end;
1130 
1131 		BUG_TRAP(start <= offset + len);
1132 
1133 		end = start + skb_shinfo(skb)->frags[i].size;
1134 		if ((copy = end - offset) > 0) {
1135 			unsigned int csum2;
1136 			u8 *vaddr;
1137 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1138 
1139 			if (copy > len)
1140 				copy = len;
1141 			vaddr = kmap_skb_frag(frag);
1142 			csum2 = csum_partial(vaddr + frag->page_offset +
1143 					     offset - start, copy, 0);
1144 			kunmap_skb_frag(vaddr);
1145 			csum = csum_block_add(csum, csum2, pos);
1146 			if (!(len -= copy))
1147 				return csum;
1148 			offset += copy;
1149 			pos    += copy;
1150 		}
1151 		start = end;
1152 	}
1153 
1154 	if (skb_shinfo(skb)->frag_list) {
1155 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1156 
1157 		for (; list; list = list->next) {
1158 			int end;
1159 
1160 			BUG_TRAP(start <= offset + len);
1161 
1162 			end = start + list->len;
1163 			if ((copy = end - offset) > 0) {
1164 				unsigned int csum2;
1165 				if (copy > len)
1166 					copy = len;
1167 				csum2 = skb_checksum(list, offset - start,
1168 						     copy, 0);
1169 				csum = csum_block_add(csum, csum2, pos);
1170 				if ((len -= copy) == 0)
1171 					return csum;
1172 				offset += copy;
1173 				pos    += copy;
1174 			}
1175 			start = end;
1176 		}
1177 	}
1178 	if (len)
1179 		BUG();
1180 
1181 	return csum;
1182 }
1183 
1184 /* Both of above in one bottle. */
1185 
1186 unsigned int skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
1187 				    u8 *to, int len, unsigned int csum)
1188 {
1189 	int start = skb_headlen(skb);
1190 	int i, copy = start - offset;
1191 	int pos = 0;
1192 
1193 	/* Copy header. */
1194 	if (copy > 0) {
1195 		if (copy > len)
1196 			copy = len;
1197 		csum = csum_partial_copy_nocheck(skb->data + offset, to,
1198 						 copy, csum);
1199 		if ((len -= copy) == 0)
1200 			return csum;
1201 		offset += copy;
1202 		to     += copy;
1203 		pos	= copy;
1204 	}
1205 
1206 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1207 		int end;
1208 
1209 		BUG_TRAP(start <= offset + len);
1210 
1211 		end = start + skb_shinfo(skb)->frags[i].size;
1212 		if ((copy = end - offset) > 0) {
1213 			unsigned int csum2;
1214 			u8 *vaddr;
1215 			skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1216 
1217 			if (copy > len)
1218 				copy = len;
1219 			vaddr = kmap_skb_frag(frag);
1220 			csum2 = csum_partial_copy_nocheck(vaddr +
1221 							  frag->page_offset +
1222 							  offset - start, to,
1223 							  copy, 0);
1224 			kunmap_skb_frag(vaddr);
1225 			csum = csum_block_add(csum, csum2, pos);
1226 			if (!(len -= copy))
1227 				return csum;
1228 			offset += copy;
1229 			to     += copy;
1230 			pos    += copy;
1231 		}
1232 		start = end;
1233 	}
1234 
1235 	if (skb_shinfo(skb)->frag_list) {
1236 		struct sk_buff *list = skb_shinfo(skb)->frag_list;
1237 
1238 		for (; list; list = list->next) {
1239 			unsigned int csum2;
1240 			int end;
1241 
1242 			BUG_TRAP(start <= offset + len);
1243 
1244 			end = start + list->len;
1245 			if ((copy = end - offset) > 0) {
1246 				if (copy > len)
1247 					copy = len;
1248 				csum2 = skb_copy_and_csum_bits(list,
1249 							       offset - start,
1250 							       to, copy, 0);
1251 				csum = csum_block_add(csum, csum2, pos);
1252 				if ((len -= copy) == 0)
1253 					return csum;
1254 				offset += copy;
1255 				to     += copy;
1256 				pos    += copy;
1257 			}
1258 			start = end;
1259 		}
1260 	}
1261 	if (len)
1262 		BUG();
1263 	return csum;
1264 }
1265 
1266 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
1267 {
1268 	unsigned int csum;
1269 	long csstart;
1270 
1271 	if (skb->ip_summed == CHECKSUM_HW)
1272 		csstart = skb->h.raw - skb->data;
1273 	else
1274 		csstart = skb_headlen(skb);
1275 
1276 	if (csstart > skb_headlen(skb))
1277 		BUG();
1278 
1279 	memcpy(to, skb->data, csstart);
1280 
1281 	csum = 0;
1282 	if (csstart != skb->len)
1283 		csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
1284 					      skb->len - csstart, 0);
1285 
1286 	if (skb->ip_summed == CHECKSUM_HW) {
1287 		long csstuff = csstart + skb->csum;
1288 
1289 		*((unsigned short *)(to + csstuff)) = csum_fold(csum);
1290 	}
1291 }
1292 
1293 /**
1294  *	skb_dequeue - remove from the head of the queue
1295  *	@list: list to dequeue from
1296  *
1297  *	Remove the head of the list. The list lock is taken so the function
1298  *	may be used safely with other locking list functions. The head item is
1299  *	returned or %NULL if the list is empty.
1300  */
1301 
1302 struct sk_buff *skb_dequeue(struct sk_buff_head *list)
1303 {
1304 	unsigned long flags;
1305 	struct sk_buff *result;
1306 
1307 	spin_lock_irqsave(&list->lock, flags);
1308 	result = __skb_dequeue(list);
1309 	spin_unlock_irqrestore(&list->lock, flags);
1310 	return result;
1311 }
1312 
1313 /**
1314  *	skb_dequeue_tail - remove from the tail of the queue
1315  *	@list: list to dequeue from
1316  *
1317  *	Remove the tail of the list. The list lock is taken so the function
1318  *	may be used safely with other locking list functions. The tail item is
1319  *	returned or %NULL if the list is empty.
1320  */
1321 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
1322 {
1323 	unsigned long flags;
1324 	struct sk_buff *result;
1325 
1326 	spin_lock_irqsave(&list->lock, flags);
1327 	result = __skb_dequeue_tail(list);
1328 	spin_unlock_irqrestore(&list->lock, flags);
1329 	return result;
1330 }
1331 
1332 /**
1333  *	skb_queue_purge - empty a list
1334  *	@list: list to empty
1335  *
1336  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
1337  *	the list and one reference dropped. This function takes the list
1338  *	lock and is atomic with respect to other list locking functions.
1339  */
1340 void skb_queue_purge(struct sk_buff_head *list)
1341 {
1342 	struct sk_buff *skb;
1343 	while ((skb = skb_dequeue(list)) != NULL)
1344 		kfree_skb(skb);
1345 }
1346 
1347 /**
1348  *	skb_queue_head - queue a buffer at the list head
1349  *	@list: list to use
1350  *	@newsk: buffer to queue
1351  *
1352  *	Queue a buffer at the start of the list. This function takes the
1353  *	list lock and can be used safely with other locking &sk_buff functions
1354  *	safely.
1355  *
1356  *	A buffer cannot be placed on two lists at the same time.
1357  */
1358 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
1359 {
1360 	unsigned long flags;
1361 
1362 	spin_lock_irqsave(&list->lock, flags);
1363 	__skb_queue_head(list, newsk);
1364 	spin_unlock_irqrestore(&list->lock, flags);
1365 }
1366 
1367 /**
1368  *	skb_queue_tail - queue a buffer at the list tail
1369  *	@list: list to use
1370  *	@newsk: buffer to queue
1371  *
1372  *	Queue a buffer at the tail of the list. This function takes the
1373  *	list lock and can be used safely with other locking &sk_buff functions
1374  *	safely.
1375  *
1376  *	A buffer cannot be placed on two lists at the same time.
1377  */
1378 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
1379 {
1380 	unsigned long flags;
1381 
1382 	spin_lock_irqsave(&list->lock, flags);
1383 	__skb_queue_tail(list, newsk);
1384 	spin_unlock_irqrestore(&list->lock, flags);
1385 }
1386 
1387 /**
1388  *	skb_unlink	-	remove a buffer from a list
1389  *	@skb: buffer to remove
1390  *	@list: list to use
1391  *
1392  *	Remove a packet from a list. The list locks are taken and this
1393  *	function is atomic with respect to other list locked calls
1394  *
1395  *	You must know what list the SKB is on.
1396  */
1397 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1398 {
1399 	unsigned long flags;
1400 
1401 	spin_lock_irqsave(&list->lock, flags);
1402 	__skb_unlink(skb, list);
1403 	spin_unlock_irqrestore(&list->lock, flags);
1404 }
1405 
1406 /**
1407  *	skb_append	-	append a buffer
1408  *	@old: buffer to insert after
1409  *	@newsk: buffer to insert
1410  *	@list: list to use
1411  *
1412  *	Place a packet after a given packet in a list. The list locks are taken
1413  *	and this function is atomic with respect to other list locked calls.
1414  *	A buffer cannot be placed on two lists at the same time.
1415  */
1416 void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1417 {
1418 	unsigned long flags;
1419 
1420 	spin_lock_irqsave(&list->lock, flags);
1421 	__skb_append(old, newsk, list);
1422 	spin_unlock_irqrestore(&list->lock, flags);
1423 }
1424 
1425 
1426 /**
1427  *	skb_insert	-	insert a buffer
1428  *	@old: buffer to insert before
1429  *	@newsk: buffer to insert
1430  *	@list: list to use
1431  *
1432  *	Place a packet before a given packet in a list. The list locks are
1433  * 	taken and this function is atomic with respect to other list locked
1434  *	calls.
1435  *
1436  *	A buffer cannot be placed on two lists at the same time.
1437  */
1438 void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
1439 {
1440 	unsigned long flags;
1441 
1442 	spin_lock_irqsave(&list->lock, flags);
1443 	__skb_insert(newsk, old->prev, old, list);
1444 	spin_unlock_irqrestore(&list->lock, flags);
1445 }
1446 
1447 #if 0
1448 /*
1449  * 	Tune the memory allocator for a new MTU size.
1450  */
1451 void skb_add_mtu(int mtu)
1452 {
1453 	/* Must match allocation in alloc_skb */
1454 	mtu = SKB_DATA_ALIGN(mtu) + sizeof(struct skb_shared_info);
1455 
1456 	kmem_add_cache_size(mtu);
1457 }
1458 #endif
1459 
1460 static inline void skb_split_inside_header(struct sk_buff *skb,
1461 					   struct sk_buff* skb1,
1462 					   const u32 len, const int pos)
1463 {
1464 	int i;
1465 
1466 	memcpy(skb_put(skb1, pos - len), skb->data + len, pos - len);
1467 
1468 	/* And move data appendix as is. */
1469 	for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1470 		skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
1471 
1472 	skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
1473 	skb_shinfo(skb)->nr_frags  = 0;
1474 	skb1->data_len		   = skb->data_len;
1475 	skb1->len		   += skb1->data_len;
1476 	skb->data_len		   = 0;
1477 	skb->len		   = len;
1478 	skb->tail		   = skb->data + len;
1479 }
1480 
1481 static inline void skb_split_no_header(struct sk_buff *skb,
1482 				       struct sk_buff* skb1,
1483 				       const u32 len, int pos)
1484 {
1485 	int i, k = 0;
1486 	const int nfrags = skb_shinfo(skb)->nr_frags;
1487 
1488 	skb_shinfo(skb)->nr_frags = 0;
1489 	skb1->len		  = skb1->data_len = skb->len - len;
1490 	skb->len		  = len;
1491 	skb->data_len		  = len - pos;
1492 
1493 	for (i = 0; i < nfrags; i++) {
1494 		int size = skb_shinfo(skb)->frags[i].size;
1495 
1496 		if (pos + size > len) {
1497 			skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
1498 
1499 			if (pos < len) {
1500 				/* Split frag.
1501 				 * We have two variants in this case:
1502 				 * 1. Move all the frag to the second
1503 				 *    part, if it is possible. F.e.
1504 				 *    this approach is mandatory for TUX,
1505 				 *    where splitting is expensive.
1506 				 * 2. Split is accurately. We make this.
1507 				 */
1508 				get_page(skb_shinfo(skb)->frags[i].page);
1509 				skb_shinfo(skb1)->frags[0].page_offset += len - pos;
1510 				skb_shinfo(skb1)->frags[0].size -= len - pos;
1511 				skb_shinfo(skb)->frags[i].size	= len - pos;
1512 				skb_shinfo(skb)->nr_frags++;
1513 			}
1514 			k++;
1515 		} else
1516 			skb_shinfo(skb)->nr_frags++;
1517 		pos += size;
1518 	}
1519 	skb_shinfo(skb1)->nr_frags = k;
1520 }
1521 
1522 /**
1523  * skb_split - Split fragmented skb to two parts at length len.
1524  * @skb: the buffer to split
1525  * @skb1: the buffer to receive the second part
1526  * @len: new length for skb
1527  */
1528 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
1529 {
1530 	int pos = skb_headlen(skb);
1531 
1532 	if (len < pos)	/* Split line is inside header. */
1533 		skb_split_inside_header(skb, skb1, len, pos);
1534 	else		/* Second chunk has no header, nothing to copy. */
1535 		skb_split_no_header(skb, skb1, len, pos);
1536 }
1537 
1538 /**
1539  * skb_prepare_seq_read - Prepare a sequential read of skb data
1540  * @skb: the buffer to read
1541  * @from: lower offset of data to be read
1542  * @to: upper offset of data to be read
1543  * @st: state variable
1544  *
1545  * Initializes the specified state variable. Must be called before
1546  * invoking skb_seq_read() for the first time.
1547  */
1548 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1549 			  unsigned int to, struct skb_seq_state *st)
1550 {
1551 	st->lower_offset = from;
1552 	st->upper_offset = to;
1553 	st->root_skb = st->cur_skb = skb;
1554 	st->frag_idx = st->stepped_offset = 0;
1555 	st->frag_data = NULL;
1556 }
1557 
1558 /**
1559  * skb_seq_read - Sequentially read skb data
1560  * @consumed: number of bytes consumed by the caller so far
1561  * @data: destination pointer for data to be returned
1562  * @st: state variable
1563  *
1564  * Reads a block of skb data at &consumed relative to the
1565  * lower offset specified to skb_prepare_seq_read(). Assigns
1566  * the head of the data block to &data and returns the length
1567  * of the block or 0 if the end of the skb data or the upper
1568  * offset has been reached.
1569  *
1570  * The caller is not required to consume all of the data
1571  * returned, i.e. &consumed is typically set to the number
1572  * of bytes already consumed and the next call to
1573  * skb_seq_read() will return the remaining part of the block.
1574  *
1575  * Note: The size of each block of data returned can be arbitary,
1576  *       this limitation is the cost for zerocopy seqeuental
1577  *       reads of potentially non linear data.
1578  *
1579  * Note: Fragment lists within fragments are not implemented
1580  *       at the moment, state->root_skb could be replaced with
1581  *       a stack for this purpose.
1582  */
1583 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1584 			  struct skb_seq_state *st)
1585 {
1586 	unsigned int block_limit, abs_offset = consumed + st->lower_offset;
1587 	skb_frag_t *frag;
1588 
1589 	if (unlikely(abs_offset >= st->upper_offset))
1590 		return 0;
1591 
1592 next_skb:
1593 	block_limit = skb_headlen(st->cur_skb);
1594 
1595 	if (abs_offset < block_limit) {
1596 		*data = st->cur_skb->data + abs_offset;
1597 		return block_limit - abs_offset;
1598 	}
1599 
1600 	if (st->frag_idx == 0 && !st->frag_data)
1601 		st->stepped_offset += skb_headlen(st->cur_skb);
1602 
1603 	while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
1604 		frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
1605 		block_limit = frag->size + st->stepped_offset;
1606 
1607 		if (abs_offset < block_limit) {
1608 			if (!st->frag_data)
1609 				st->frag_data = kmap_skb_frag(frag);
1610 
1611 			*data = (u8 *) st->frag_data + frag->page_offset +
1612 				(abs_offset - st->stepped_offset);
1613 
1614 			return block_limit - abs_offset;
1615 		}
1616 
1617 		if (st->frag_data) {
1618 			kunmap_skb_frag(st->frag_data);
1619 			st->frag_data = NULL;
1620 		}
1621 
1622 		st->frag_idx++;
1623 		st->stepped_offset += frag->size;
1624 	}
1625 
1626 	if (st->cur_skb->next) {
1627 		st->cur_skb = st->cur_skb->next;
1628 		st->frag_idx = 0;
1629 		goto next_skb;
1630 	} else if (st->root_skb == st->cur_skb &&
1631 		   skb_shinfo(st->root_skb)->frag_list) {
1632 		st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
1633 		goto next_skb;
1634 	}
1635 
1636 	return 0;
1637 }
1638 
1639 /**
1640  * skb_abort_seq_read - Abort a sequential read of skb data
1641  * @st: state variable
1642  *
1643  * Must be called if skb_seq_read() was not called until it
1644  * returned 0.
1645  */
1646 void skb_abort_seq_read(struct skb_seq_state *st)
1647 {
1648 	if (st->frag_data)
1649 		kunmap_skb_frag(st->frag_data);
1650 }
1651 
1652 #define TS_SKB_CB(state)	((struct skb_seq_state *) &((state)->cb))
1653 
1654 static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
1655 					  struct ts_config *conf,
1656 					  struct ts_state *state)
1657 {
1658 	return skb_seq_read(offset, text, TS_SKB_CB(state));
1659 }
1660 
1661 static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
1662 {
1663 	skb_abort_seq_read(TS_SKB_CB(state));
1664 }
1665 
1666 /**
1667  * skb_find_text - Find a text pattern in skb data
1668  * @skb: the buffer to look in
1669  * @from: search offset
1670  * @to: search limit
1671  * @config: textsearch configuration
1672  * @state: uninitialized textsearch state variable
1673  *
1674  * Finds a pattern in the skb data according to the specified
1675  * textsearch configuration. Use textsearch_next() to retrieve
1676  * subsequent occurrences of the pattern. Returns the offset
1677  * to the first occurrence or UINT_MAX if no match was found.
1678  */
1679 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1680 			   unsigned int to, struct ts_config *config,
1681 			   struct ts_state *state)
1682 {
1683 	config->get_next_block = skb_ts_get_next_block;
1684 	config->finish = skb_ts_finish;
1685 
1686 	skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
1687 
1688 	return textsearch_find(config, state);
1689 }
1690 
1691 void __init skb_init(void)
1692 {
1693 	skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
1694 					      sizeof(struct sk_buff),
1695 					      0,
1696 					      SLAB_HWCACHE_ALIGN,
1697 					      NULL, NULL);
1698 	if (!skbuff_head_cache)
1699 		panic("cannot create skbuff cache");
1700 
1701 	skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
1702 						(2*sizeof(struct sk_buff)) +
1703 						sizeof(atomic_t),
1704 						0,
1705 						SLAB_HWCACHE_ALIGN,
1706 						NULL, NULL);
1707 	if (!skbuff_fclone_cache)
1708 		panic("cannot create skbuff cache");
1709 }
1710 
1711 EXPORT_SYMBOL(___pskb_trim);
1712 EXPORT_SYMBOL(__kfree_skb);
1713 EXPORT_SYMBOL(__pskb_pull_tail);
1714 EXPORT_SYMBOL(__alloc_skb);
1715 EXPORT_SYMBOL(pskb_copy);
1716 EXPORT_SYMBOL(pskb_expand_head);
1717 EXPORT_SYMBOL(skb_checksum);
1718 EXPORT_SYMBOL(skb_clone);
1719 EXPORT_SYMBOL(skb_clone_fraglist);
1720 EXPORT_SYMBOL(skb_copy);
1721 EXPORT_SYMBOL(skb_copy_and_csum_bits);
1722 EXPORT_SYMBOL(skb_copy_and_csum_dev);
1723 EXPORT_SYMBOL(skb_copy_bits);
1724 EXPORT_SYMBOL(skb_copy_expand);
1725 EXPORT_SYMBOL(skb_over_panic);
1726 EXPORT_SYMBOL(skb_pad);
1727 EXPORT_SYMBOL(skb_realloc_headroom);
1728 EXPORT_SYMBOL(skb_under_panic);
1729 EXPORT_SYMBOL(skb_dequeue);
1730 EXPORT_SYMBOL(skb_dequeue_tail);
1731 EXPORT_SYMBOL(skb_insert);
1732 EXPORT_SYMBOL(skb_queue_purge);
1733 EXPORT_SYMBOL(skb_queue_head);
1734 EXPORT_SYMBOL(skb_queue_tail);
1735 EXPORT_SYMBOL(skb_unlink);
1736 EXPORT_SYMBOL(skb_append);
1737 EXPORT_SYMBOL(skb_split);
1738 EXPORT_SYMBOL(skb_prepare_seq_read);
1739 EXPORT_SYMBOL(skb_seq_read);
1740 EXPORT_SYMBOL(skb_abort_seq_read);
1741 EXPORT_SYMBOL(skb_find_text);
1742