xref: /linux-6.15/include/linux/skbuff.h (revision a1e58bbd)
1 /*
2  *	Definitions for the 'struct sk_buff' memory handlers.
3  *
4  *	Authors:
5  *		Alan Cox, <[email protected]>
6  *		Florian La Roche, <[email protected]>
7  *
8  *	This program is free software; you can redistribute it and/or
9  *	modify it under the terms of the GNU General Public License
10  *	as published by the Free Software Foundation; either version
11  *	2 of the License, or (at your option) any later version.
12  */
13 
14 #ifndef _LINUX_SKBUFF_H
15 #define _LINUX_SKBUFF_H
16 
17 #include <linux/kernel.h>
18 #include <linux/compiler.h>
19 #include <linux/time.h>
20 #include <linux/cache.h>
21 
22 #include <asm/atomic.h>
23 #include <asm/types.h>
24 #include <linux/spinlock.h>
25 #include <linux/net.h>
26 #include <linux/textsearch.h>
27 #include <net/checksum.h>
28 #include <linux/rcupdate.h>
29 #include <linux/dmaengine.h>
30 #include <linux/hrtimer.h>
31 
32 #define HAVE_ALLOC_SKB		/* For the drivers to know */
33 #define HAVE_ALIGNABLE_SKB	/* Ditto 8)		   */
34 
35 /* Don't change this without changing skb_csum_unnecessary! */
36 #define CHECKSUM_NONE 0
37 #define CHECKSUM_UNNECESSARY 1
38 #define CHECKSUM_COMPLETE 2
39 #define CHECKSUM_PARTIAL 3
40 
41 #define SKB_DATA_ALIGN(X)	(((X) + (SMP_CACHE_BYTES - 1)) & \
42 				 ~(SMP_CACHE_BYTES - 1))
43 #define SKB_WITH_OVERHEAD(X)	\
44 	((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
45 #define SKB_MAX_ORDER(X, ORDER) \
46 	SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
47 #define SKB_MAX_HEAD(X)		(SKB_MAX_ORDER((X), 0))
48 #define SKB_MAX_ALLOC		(SKB_MAX_ORDER(0, 2))
49 
50 /* A. Checksumming of received packets by device.
51  *
52  *	NONE: device failed to checksum this packet.
53  *		skb->csum is undefined.
54  *
55  *	UNNECESSARY: device parsed packet and wouldbe verified checksum.
56  *		skb->csum is undefined.
57  *	      It is bad option, but, unfortunately, many of vendors do this.
58  *	      Apparently with secret goal to sell you new device, when you
59  *	      will add new protocol to your host. F.e. IPv6. 8)
60  *
61  *	COMPLETE: the most generic way. Device supplied checksum of _all_
62  *	    the packet as seen by netif_rx in skb->csum.
63  *	    NOTE: Even if device supports only some protocols, but
64  *	    is able to produce some skb->csum, it MUST use COMPLETE,
65  *	    not UNNECESSARY.
66  *
67  *	PARTIAL: identical to the case for output below.  This may occur
68  *	    on a packet received directly from another Linux OS, e.g.,
69  *	    a virtualised Linux kernel on the same host.  The packet can
70  *	    be treated in the same way as UNNECESSARY except that on
71  *	    output (i.e., forwarding) the checksum must be filled in
72  *	    by the OS or the hardware.
73  *
74  * B. Checksumming on output.
75  *
76  *	NONE: skb is checksummed by protocol or csum is not required.
77  *
78  *	PARTIAL: device is required to csum packet as seen by hard_start_xmit
79  *	from skb->csum_start to the end and to record the checksum
80  *	at skb->csum_start + skb->csum_offset.
81  *
82  *	Device must show its capabilities in dev->features, set
83  *	at device setup time.
84  *	NETIF_F_HW_CSUM	- it is clever device, it is able to checksum
85  *			  everything.
86  *	NETIF_F_NO_CSUM - loopback or reliable single hop media.
87  *	NETIF_F_IP_CSUM - device is dumb. It is able to csum only
88  *			  TCP/UDP over IPv4. Sigh. Vendors like this
89  *			  way by an unknown reason. Though, see comment above
90  *			  about CHECKSUM_UNNECESSARY. 8)
91  *	NETIF_F_IPV6_CSUM about as dumb as the last one but does IPv6 instead.
92  *
93  *	Any questions? No questions, good. 		--ANK
94  */
95 
96 struct net_device;
97 struct scatterlist;
98 struct pipe_inode_info;
99 
100 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
101 struct nf_conntrack {
102 	atomic_t use;
103 };
104 #endif
105 
106 #ifdef CONFIG_BRIDGE_NETFILTER
107 struct nf_bridge_info {
108 	atomic_t use;
109 	struct net_device *physindev;
110 	struct net_device *physoutdev;
111 	unsigned int mask;
112 	unsigned long data[32 / sizeof(unsigned long)];
113 };
114 #endif
115 
116 struct sk_buff_head {
117 	/* These two members must be first. */
118 	struct sk_buff	*next;
119 	struct sk_buff	*prev;
120 
121 	__u32		qlen;
122 	spinlock_t	lock;
123 };
124 
125 struct sk_buff;
126 
127 /* To allow 64K frame to be packed as single skb without frag_list */
128 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 2)
129 
130 typedef struct skb_frag_struct skb_frag_t;
131 
132 struct skb_frag_struct {
133 	struct page *page;
134 	__u32 page_offset;
135 	__u32 size;
136 };
137 
138 /* This data is invariant across clones and lives at
139  * the end of the header data, ie. at skb->end.
140  */
141 struct skb_shared_info {
142 	atomic_t	dataref;
143 	unsigned short	nr_frags;
144 	unsigned short	gso_size;
145 	/* Warning: this field is not always filled in (UFO)! */
146 	unsigned short	gso_segs;
147 	unsigned short  gso_type;
148 	__be32          ip6_frag_id;
149 	struct sk_buff	*frag_list;
150 	skb_frag_t	frags[MAX_SKB_FRAGS];
151 };
152 
153 /* We divide dataref into two halves.  The higher 16 bits hold references
154  * to the payload part of skb->data.  The lower 16 bits hold references to
155  * the entire skb->data.  A clone of a headerless skb holds the length of
156  * the header in skb->hdr_len.
157  *
158  * All users must obey the rule that the skb->data reference count must be
159  * greater than or equal to the payload reference count.
160  *
161  * Holding a reference to the payload part means that the user does not
162  * care about modifications to the header part of skb->data.
163  */
164 #define SKB_DATAREF_SHIFT 16
165 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
166 
167 
168 enum {
169 	SKB_FCLONE_UNAVAILABLE,
170 	SKB_FCLONE_ORIG,
171 	SKB_FCLONE_CLONE,
172 };
173 
174 enum {
175 	SKB_GSO_TCPV4 = 1 << 0,
176 	SKB_GSO_UDP = 1 << 1,
177 
178 	/* This indicates the skb is from an untrusted source. */
179 	SKB_GSO_DODGY = 1 << 2,
180 
181 	/* This indicates the tcp segment has CWR set. */
182 	SKB_GSO_TCP_ECN = 1 << 3,
183 
184 	SKB_GSO_TCPV6 = 1 << 4,
185 };
186 
187 #if BITS_PER_LONG > 32
188 #define NET_SKBUFF_DATA_USES_OFFSET 1
189 #endif
190 
191 #ifdef NET_SKBUFF_DATA_USES_OFFSET
192 typedef unsigned int sk_buff_data_t;
193 #else
194 typedef unsigned char *sk_buff_data_t;
195 #endif
196 
197 /**
198  *	struct sk_buff - socket buffer
199  *	@next: Next buffer in list
200  *	@prev: Previous buffer in list
201  *	@sk: Socket we are owned by
202  *	@tstamp: Time we arrived
203  *	@dev: Device we arrived on/are leaving by
204  *	@transport_header: Transport layer header
205  *	@network_header: Network layer header
206  *	@mac_header: Link layer header
207  *	@dst: destination entry
208  *	@sp: the security path, used for xfrm
209  *	@cb: Control buffer. Free for use by every layer. Put private vars here
210  *	@len: Length of actual data
211  *	@data_len: Data length
212  *	@mac_len: Length of link layer header
213  *	@hdr_len: writable header length of cloned skb
214  *	@csum: Checksum (must include start/offset pair)
215  *	@csum_start: Offset from skb->head where checksumming should start
216  *	@csum_offset: Offset from csum_start where checksum should be stored
217  *	@local_df: allow local fragmentation
218  *	@cloned: Head may be cloned (check refcnt to be sure)
219  *	@nohdr: Payload reference only, must not modify header
220  *	@pkt_type: Packet class
221  *	@fclone: skbuff clone status
222  *	@ip_summed: Driver fed us an IP checksum
223  *	@priority: Packet queueing priority
224  *	@users: User count - see {datagram,tcp}.c
225  *	@protocol: Packet protocol from driver
226  *	@truesize: Buffer size
227  *	@head: Head of buffer
228  *	@data: Data head pointer
229  *	@tail: Tail pointer
230  *	@end: End pointer
231  *	@destructor: Destruct function
232  *	@mark: Generic packet mark
233  *	@nfct: Associated connection, if any
234  *	@ipvs_property: skbuff is owned by ipvs
235  *	@peeked: this packet has been seen already, so stats have been
236  *		done for it, don't do them again
237  *	@nf_trace: netfilter packet trace flag
238  *	@nfctinfo: Relationship of this skb to the connection
239  *	@nfct_reasm: netfilter conntrack re-assembly pointer
240  *	@nf_bridge: Saved data about a bridged frame - see br_netfilter.c
241  *	@iif: ifindex of device we arrived on
242  *	@queue_mapping: Queue mapping for multiqueue devices
243  *	@tc_index: Traffic control index
244  *	@tc_verd: traffic control verdict
245  *	@dma_cookie: a cookie to one of several possible DMA operations
246  *		done by skb DMA functions
247  *	@secmark: security marking
248  */
249 
250 struct sk_buff {
251 	/* These two members must be first. */
252 	struct sk_buff		*next;
253 	struct sk_buff		*prev;
254 
255 	struct sock		*sk;
256 	ktime_t			tstamp;
257 	struct net_device	*dev;
258 
259 	struct  dst_entry	*dst;
260 	struct	sec_path	*sp;
261 
262 	/*
263 	 * This is the control buffer. It is free to use for every
264 	 * layer. Please put your private variables there. If you
265 	 * want to keep them across layers you have to do a skb_clone()
266 	 * first. This is owned by whoever has the skb queued ATM.
267 	 */
268 	char			cb[48];
269 
270 	unsigned int		len,
271 				data_len;
272 	__u16			mac_len,
273 				hdr_len;
274 	union {
275 		__wsum		csum;
276 		struct {
277 			__u16	csum_start;
278 			__u16	csum_offset;
279 		};
280 	};
281 	__u32			priority;
282 	__u8			local_df:1,
283 				cloned:1,
284 				ip_summed:2,
285 				nohdr:1,
286 				nfctinfo:3;
287 	__u8			pkt_type:3,
288 				fclone:2,
289 				ipvs_property:1,
290 				peeked:1,
291 				nf_trace:1;
292 	__be16			protocol;
293 
294 	void			(*destructor)(struct sk_buff *skb);
295 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
296 	struct nf_conntrack	*nfct;
297 	struct sk_buff		*nfct_reasm;
298 #endif
299 #ifdef CONFIG_BRIDGE_NETFILTER
300 	struct nf_bridge_info	*nf_bridge;
301 #endif
302 
303 	int			iif;
304 #ifdef CONFIG_NETDEVICES_MULTIQUEUE
305 	__u16			queue_mapping;
306 #endif
307 #ifdef CONFIG_NET_SCHED
308 	__u16			tc_index;	/* traffic control index */
309 #ifdef CONFIG_NET_CLS_ACT
310 	__u16			tc_verd;	/* traffic control verdict */
311 #endif
312 #endif
313 	/* 2 byte hole */
314 
315 #ifdef CONFIG_NET_DMA
316 	dma_cookie_t		dma_cookie;
317 #endif
318 #ifdef CONFIG_NETWORK_SECMARK
319 	__u32			secmark;
320 #endif
321 
322 	__u32			mark;
323 
324 	sk_buff_data_t		transport_header;
325 	sk_buff_data_t		network_header;
326 	sk_buff_data_t		mac_header;
327 	/* These elements must be at the end, see alloc_skb() for details.  */
328 	sk_buff_data_t		tail;
329 	sk_buff_data_t		end;
330 	unsigned char		*head,
331 				*data;
332 	unsigned int		truesize;
333 	atomic_t		users;
334 };
335 
336 #ifdef __KERNEL__
337 /*
338  *	Handling routines are only of interest to the kernel
339  */
340 #include <linux/slab.h>
341 
342 #include <asm/system.h>
343 
344 extern void kfree_skb(struct sk_buff *skb);
345 extern void	       __kfree_skb(struct sk_buff *skb);
346 extern struct sk_buff *__alloc_skb(unsigned int size,
347 				   gfp_t priority, int fclone, int node);
348 static inline struct sk_buff *alloc_skb(unsigned int size,
349 					gfp_t priority)
350 {
351 	return __alloc_skb(size, priority, 0, -1);
352 }
353 
354 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
355 					       gfp_t priority)
356 {
357 	return __alloc_skb(size, priority, 1, -1);
358 }
359 
360 extern struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
361 extern struct sk_buff *skb_clone(struct sk_buff *skb,
362 				 gfp_t priority);
363 extern struct sk_buff *skb_copy(const struct sk_buff *skb,
364 				gfp_t priority);
365 extern struct sk_buff *pskb_copy(struct sk_buff *skb,
366 				 gfp_t gfp_mask);
367 extern int	       pskb_expand_head(struct sk_buff *skb,
368 					int nhead, int ntail,
369 					gfp_t gfp_mask);
370 extern struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
371 					    unsigned int headroom);
372 extern struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
373 				       int newheadroom, int newtailroom,
374 				       gfp_t priority);
375 extern int	       skb_to_sgvec(struct sk_buff *skb,
376 				    struct scatterlist *sg, int offset,
377 				    int len);
378 extern int	       skb_cow_data(struct sk_buff *skb, int tailbits,
379 				    struct sk_buff **trailer);
380 extern int	       skb_pad(struct sk_buff *skb, int pad);
381 #define dev_kfree_skb(a)	kfree_skb(a)
382 extern void	      skb_over_panic(struct sk_buff *skb, int len,
383 				     void *here);
384 extern void	      skb_under_panic(struct sk_buff *skb, int len,
385 				      void *here);
386 extern void	      skb_truesize_bug(struct sk_buff *skb);
387 
388 static inline void skb_truesize_check(struct sk_buff *skb)
389 {
390 	int len = sizeof(struct sk_buff) + skb->len;
391 
392 	if (unlikely((int)skb->truesize < len))
393 		skb_truesize_bug(skb);
394 }
395 
396 extern int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
397 			int getfrag(void *from, char *to, int offset,
398 			int len,int odd, struct sk_buff *skb),
399 			void *from, int length);
400 
401 struct skb_seq_state
402 {
403 	__u32		lower_offset;
404 	__u32		upper_offset;
405 	__u32		frag_idx;
406 	__u32		stepped_offset;
407 	struct sk_buff	*root_skb;
408 	struct sk_buff	*cur_skb;
409 	__u8		*frag_data;
410 };
411 
412 extern void	      skb_prepare_seq_read(struct sk_buff *skb,
413 					   unsigned int from, unsigned int to,
414 					   struct skb_seq_state *st);
415 extern unsigned int   skb_seq_read(unsigned int consumed, const u8 **data,
416 				   struct skb_seq_state *st);
417 extern void	      skb_abort_seq_read(struct skb_seq_state *st);
418 
419 extern unsigned int   skb_find_text(struct sk_buff *skb, unsigned int from,
420 				    unsigned int to, struct ts_config *config,
421 				    struct ts_state *state);
422 
423 #ifdef NET_SKBUFF_DATA_USES_OFFSET
424 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
425 {
426 	return skb->head + skb->end;
427 }
428 #else
429 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
430 {
431 	return skb->end;
432 }
433 #endif
434 
435 /* Internal */
436 #define skb_shinfo(SKB)	((struct skb_shared_info *)(skb_end_pointer(SKB)))
437 
438 /**
439  *	skb_queue_empty - check if a queue is empty
440  *	@list: queue head
441  *
442  *	Returns true if the queue is empty, false otherwise.
443  */
444 static inline int skb_queue_empty(const struct sk_buff_head *list)
445 {
446 	return list->next == (struct sk_buff *)list;
447 }
448 
449 /**
450  *	skb_get - reference buffer
451  *	@skb: buffer to reference
452  *
453  *	Makes another reference to a socket buffer and returns a pointer
454  *	to the buffer.
455  */
456 static inline struct sk_buff *skb_get(struct sk_buff *skb)
457 {
458 	atomic_inc(&skb->users);
459 	return skb;
460 }
461 
462 /*
463  * If users == 1, we are the only owner and are can avoid redundant
464  * atomic change.
465  */
466 
467 /**
468  *	skb_cloned - is the buffer a clone
469  *	@skb: buffer to check
470  *
471  *	Returns true if the buffer was generated with skb_clone() and is
472  *	one of multiple shared copies of the buffer. Cloned buffers are
473  *	shared data so must not be written to under normal circumstances.
474  */
475 static inline int skb_cloned(const struct sk_buff *skb)
476 {
477 	return skb->cloned &&
478 	       (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
479 }
480 
481 /**
482  *	skb_header_cloned - is the header a clone
483  *	@skb: buffer to check
484  *
485  *	Returns true if modifying the header part of the buffer requires
486  *	the data to be copied.
487  */
488 static inline int skb_header_cloned(const struct sk_buff *skb)
489 {
490 	int dataref;
491 
492 	if (!skb->cloned)
493 		return 0;
494 
495 	dataref = atomic_read(&skb_shinfo(skb)->dataref);
496 	dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
497 	return dataref != 1;
498 }
499 
500 /**
501  *	skb_header_release - release reference to header
502  *	@skb: buffer to operate on
503  *
504  *	Drop a reference to the header part of the buffer.  This is done
505  *	by acquiring a payload reference.  You must not read from the header
506  *	part of skb->data after this.
507  */
508 static inline void skb_header_release(struct sk_buff *skb)
509 {
510 	BUG_ON(skb->nohdr);
511 	skb->nohdr = 1;
512 	atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
513 }
514 
515 /**
516  *	skb_shared - is the buffer shared
517  *	@skb: buffer to check
518  *
519  *	Returns true if more than one person has a reference to this
520  *	buffer.
521  */
522 static inline int skb_shared(const struct sk_buff *skb)
523 {
524 	return atomic_read(&skb->users) != 1;
525 }
526 
527 /**
528  *	skb_share_check - check if buffer is shared and if so clone it
529  *	@skb: buffer to check
530  *	@pri: priority for memory allocation
531  *
532  *	If the buffer is shared the buffer is cloned and the old copy
533  *	drops a reference. A new clone with a single reference is returned.
534  *	If the buffer is not shared the original buffer is returned. When
535  *	being called from interrupt status or with spinlocks held pri must
536  *	be GFP_ATOMIC.
537  *
538  *	NULL is returned on a memory allocation failure.
539  */
540 static inline struct sk_buff *skb_share_check(struct sk_buff *skb,
541 					      gfp_t pri)
542 {
543 	might_sleep_if(pri & __GFP_WAIT);
544 	if (skb_shared(skb)) {
545 		struct sk_buff *nskb = skb_clone(skb, pri);
546 		kfree_skb(skb);
547 		skb = nskb;
548 	}
549 	return skb;
550 }
551 
552 /*
553  *	Copy shared buffers into a new sk_buff. We effectively do COW on
554  *	packets to handle cases where we have a local reader and forward
555  *	and a couple of other messy ones. The normal one is tcpdumping
556  *	a packet thats being forwarded.
557  */
558 
559 /**
560  *	skb_unshare - make a copy of a shared buffer
561  *	@skb: buffer to check
562  *	@pri: priority for memory allocation
563  *
564  *	If the socket buffer is a clone then this function creates a new
565  *	copy of the data, drops a reference count on the old copy and returns
566  *	the new copy with the reference count at 1. If the buffer is not a clone
567  *	the original buffer is returned. When called with a spinlock held or
568  *	from interrupt state @pri must be %GFP_ATOMIC
569  *
570  *	%NULL is returned on a memory allocation failure.
571  */
572 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
573 					  gfp_t pri)
574 {
575 	might_sleep_if(pri & __GFP_WAIT);
576 	if (skb_cloned(skb)) {
577 		struct sk_buff *nskb = skb_copy(skb, pri);
578 		kfree_skb(skb);	/* Free our shared copy */
579 		skb = nskb;
580 	}
581 	return skb;
582 }
583 
584 /**
585  *	skb_peek
586  *	@list_: list to peek at
587  *
588  *	Peek an &sk_buff. Unlike most other operations you _MUST_
589  *	be careful with this one. A peek leaves the buffer on the
590  *	list and someone else may run off with it. You must hold
591  *	the appropriate locks or have a private queue to do this.
592  *
593  *	Returns %NULL for an empty list or a pointer to the head element.
594  *	The reference count is not incremented and the reference is therefore
595  *	volatile. Use with caution.
596  */
597 static inline struct sk_buff *skb_peek(struct sk_buff_head *list_)
598 {
599 	struct sk_buff *list = ((struct sk_buff *)list_)->next;
600 	if (list == (struct sk_buff *)list_)
601 		list = NULL;
602 	return list;
603 }
604 
605 /**
606  *	skb_peek_tail
607  *	@list_: list to peek at
608  *
609  *	Peek an &sk_buff. Unlike most other operations you _MUST_
610  *	be careful with this one. A peek leaves the buffer on the
611  *	list and someone else may run off with it. You must hold
612  *	the appropriate locks or have a private queue to do this.
613  *
614  *	Returns %NULL for an empty list or a pointer to the tail element.
615  *	The reference count is not incremented and the reference is therefore
616  *	volatile. Use with caution.
617  */
618 static inline struct sk_buff *skb_peek_tail(struct sk_buff_head *list_)
619 {
620 	struct sk_buff *list = ((struct sk_buff *)list_)->prev;
621 	if (list == (struct sk_buff *)list_)
622 		list = NULL;
623 	return list;
624 }
625 
626 /**
627  *	skb_queue_len	- get queue length
628  *	@list_: list to measure
629  *
630  *	Return the length of an &sk_buff queue.
631  */
632 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
633 {
634 	return list_->qlen;
635 }
636 
637 /*
638  * This function creates a split out lock class for each invocation;
639  * this is needed for now since a whole lot of users of the skb-queue
640  * infrastructure in drivers have different locking usage (in hardirq)
641  * than the networking core (in softirq only). In the long run either the
642  * network layer or drivers should need annotation to consolidate the
643  * main types of usage into 3 classes.
644  */
645 static inline void skb_queue_head_init(struct sk_buff_head *list)
646 {
647 	spin_lock_init(&list->lock);
648 	list->prev = list->next = (struct sk_buff *)list;
649 	list->qlen = 0;
650 }
651 
652 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
653 		struct lock_class_key *class)
654 {
655 	skb_queue_head_init(list);
656 	lockdep_set_class(&list->lock, class);
657 }
658 
659 /*
660  *	Insert an sk_buff at the start of a list.
661  *
662  *	The "__skb_xxxx()" functions are the non-atomic ones that
663  *	can only be called with interrupts disabled.
664  */
665 
666 /**
667  *	__skb_queue_after - queue a buffer at the list head
668  *	@list: list to use
669  *	@prev: place after this buffer
670  *	@newsk: buffer to queue
671  *
672  *	Queue a buffer int the middle of a list. This function takes no locks
673  *	and you must therefore hold required locks before calling it.
674  *
675  *	A buffer cannot be placed on two lists at the same time.
676  */
677 static inline void __skb_queue_after(struct sk_buff_head *list,
678 				     struct sk_buff *prev,
679 				     struct sk_buff *newsk)
680 {
681 	struct sk_buff *next;
682 	list->qlen++;
683 
684 	next = prev->next;
685 	newsk->next = next;
686 	newsk->prev = prev;
687 	next->prev  = prev->next = newsk;
688 }
689 
690 /**
691  *	__skb_queue_head - queue a buffer at the list head
692  *	@list: list to use
693  *	@newsk: buffer to queue
694  *
695  *	Queue a buffer at the start of a list. This function takes no locks
696  *	and you must therefore hold required locks before calling it.
697  *
698  *	A buffer cannot be placed on two lists at the same time.
699  */
700 extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
701 static inline void __skb_queue_head(struct sk_buff_head *list,
702 				    struct sk_buff *newsk)
703 {
704 	__skb_queue_after(list, (struct sk_buff *)list, newsk);
705 }
706 
707 /**
708  *	__skb_queue_tail - queue a buffer at the list tail
709  *	@list: list to use
710  *	@newsk: buffer to queue
711  *
712  *	Queue a buffer at the end of a list. This function takes no locks
713  *	and you must therefore hold required locks before calling it.
714  *
715  *	A buffer cannot be placed on two lists at the same time.
716  */
717 extern void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
718 static inline void __skb_queue_tail(struct sk_buff_head *list,
719 				   struct sk_buff *newsk)
720 {
721 	struct sk_buff *prev, *next;
722 
723 	list->qlen++;
724 	next = (struct sk_buff *)list;
725 	prev = next->prev;
726 	newsk->next = next;
727 	newsk->prev = prev;
728 	next->prev  = prev->next = newsk;
729 }
730 
731 
732 /**
733  *	__skb_dequeue - remove from the head of the queue
734  *	@list: list to dequeue from
735  *
736  *	Remove the head of the list. This function does not take any locks
737  *	so must be used with appropriate locks held only. The head item is
738  *	returned or %NULL if the list is empty.
739  */
740 extern struct sk_buff *skb_dequeue(struct sk_buff_head *list);
741 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
742 {
743 	struct sk_buff *next, *prev, *result;
744 
745 	prev = (struct sk_buff *) list;
746 	next = prev->next;
747 	result = NULL;
748 	if (next != prev) {
749 		result	     = next;
750 		next	     = next->next;
751 		list->qlen--;
752 		next->prev   = prev;
753 		prev->next   = next;
754 		result->next = result->prev = NULL;
755 	}
756 	return result;
757 }
758 
759 
760 /*
761  *	Insert a packet on a list.
762  */
763 extern void        skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
764 static inline void __skb_insert(struct sk_buff *newsk,
765 				struct sk_buff *prev, struct sk_buff *next,
766 				struct sk_buff_head *list)
767 {
768 	newsk->next = next;
769 	newsk->prev = prev;
770 	next->prev  = prev->next = newsk;
771 	list->qlen++;
772 }
773 
774 /*
775  *	Place a packet after a given packet in a list.
776  */
777 extern void	   skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
778 static inline void __skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
779 {
780 	__skb_insert(newsk, old, old->next, list);
781 }
782 
783 /*
784  * remove sk_buff from list. _Must_ be called atomically, and with
785  * the list known..
786  */
787 extern void	   skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
788 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
789 {
790 	struct sk_buff *next, *prev;
791 
792 	list->qlen--;
793 	next	   = skb->next;
794 	prev	   = skb->prev;
795 	skb->next  = skb->prev = NULL;
796 	next->prev = prev;
797 	prev->next = next;
798 }
799 
800 
801 /* XXX: more streamlined implementation */
802 
803 /**
804  *	__skb_dequeue_tail - remove from the tail of the queue
805  *	@list: list to dequeue from
806  *
807  *	Remove the tail of the list. This function does not take any locks
808  *	so must be used with appropriate locks held only. The tail item is
809  *	returned or %NULL if the list is empty.
810  */
811 extern struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
812 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
813 {
814 	struct sk_buff *skb = skb_peek_tail(list);
815 	if (skb)
816 		__skb_unlink(skb, list);
817 	return skb;
818 }
819 
820 
821 static inline int skb_is_nonlinear(const struct sk_buff *skb)
822 {
823 	return skb->data_len;
824 }
825 
826 static inline unsigned int skb_headlen(const struct sk_buff *skb)
827 {
828 	return skb->len - skb->data_len;
829 }
830 
831 static inline int skb_pagelen(const struct sk_buff *skb)
832 {
833 	int i, len = 0;
834 
835 	for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
836 		len += skb_shinfo(skb)->frags[i].size;
837 	return len + skb_headlen(skb);
838 }
839 
840 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
841 				      struct page *page, int off, int size)
842 {
843 	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
844 
845 	frag->page		  = page;
846 	frag->page_offset	  = off;
847 	frag->size		  = size;
848 	skb_shinfo(skb)->nr_frags = i + 1;
849 }
850 
851 #define SKB_PAGE_ASSERT(skb) 	BUG_ON(skb_shinfo(skb)->nr_frags)
852 #define SKB_FRAG_ASSERT(skb) 	BUG_ON(skb_shinfo(skb)->frag_list)
853 #define SKB_LINEAR_ASSERT(skb)  BUG_ON(skb_is_nonlinear(skb))
854 
855 #ifdef NET_SKBUFF_DATA_USES_OFFSET
856 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
857 {
858 	return skb->head + skb->tail;
859 }
860 
861 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
862 {
863 	skb->tail = skb->data - skb->head;
864 }
865 
866 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
867 {
868 	skb_reset_tail_pointer(skb);
869 	skb->tail += offset;
870 }
871 #else /* NET_SKBUFF_DATA_USES_OFFSET */
872 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
873 {
874 	return skb->tail;
875 }
876 
877 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
878 {
879 	skb->tail = skb->data;
880 }
881 
882 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
883 {
884 	skb->tail = skb->data + offset;
885 }
886 
887 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
888 
889 /*
890  *	Add data to an sk_buff
891  */
892 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
893 {
894 	unsigned char *tmp = skb_tail_pointer(skb);
895 	SKB_LINEAR_ASSERT(skb);
896 	skb->tail += len;
897 	skb->len  += len;
898 	return tmp;
899 }
900 
901 /**
902  *	skb_put - add data to a buffer
903  *	@skb: buffer to use
904  *	@len: amount of data to add
905  *
906  *	This function extends the used data area of the buffer. If this would
907  *	exceed the total buffer size the kernel will panic. A pointer to the
908  *	first byte of the extra data is returned.
909  */
910 static inline unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
911 {
912 	unsigned char *tmp = skb_tail_pointer(skb);
913 	SKB_LINEAR_ASSERT(skb);
914 	skb->tail += len;
915 	skb->len  += len;
916 	if (unlikely(skb->tail > skb->end))
917 		skb_over_panic(skb, len, current_text_addr());
918 	return tmp;
919 }
920 
921 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
922 {
923 	skb->data -= len;
924 	skb->len  += len;
925 	return skb->data;
926 }
927 
928 /**
929  *	skb_push - add data to the start of a buffer
930  *	@skb: buffer to use
931  *	@len: amount of data to add
932  *
933  *	This function extends the used data area of the buffer at the buffer
934  *	start. If this would exceed the total buffer headroom the kernel will
935  *	panic. A pointer to the first byte of the extra data is returned.
936  */
937 static inline unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
938 {
939 	skb->data -= len;
940 	skb->len  += len;
941 	if (unlikely(skb->data<skb->head))
942 		skb_under_panic(skb, len, current_text_addr());
943 	return skb->data;
944 }
945 
946 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
947 {
948 	skb->len -= len;
949 	BUG_ON(skb->len < skb->data_len);
950 	return skb->data += len;
951 }
952 
953 /**
954  *	skb_pull - remove data from the start of a buffer
955  *	@skb: buffer to use
956  *	@len: amount of data to remove
957  *
958  *	This function removes data from the start of a buffer, returning
959  *	the memory to the headroom. A pointer to the next data in the buffer
960  *	is returned. Once the data has been pulled future pushes will overwrite
961  *	the old data.
962  */
963 static inline unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
964 {
965 	return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
966 }
967 
968 extern unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
969 
970 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
971 {
972 	if (len > skb_headlen(skb) &&
973 	    !__pskb_pull_tail(skb, len-skb_headlen(skb)))
974 		return NULL;
975 	skb->len -= len;
976 	return skb->data += len;
977 }
978 
979 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
980 {
981 	return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
982 }
983 
984 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
985 {
986 	if (likely(len <= skb_headlen(skb)))
987 		return 1;
988 	if (unlikely(len > skb->len))
989 		return 0;
990 	return __pskb_pull_tail(skb, len-skb_headlen(skb)) != NULL;
991 }
992 
993 /**
994  *	skb_headroom - bytes at buffer head
995  *	@skb: buffer to check
996  *
997  *	Return the number of bytes of free space at the head of an &sk_buff.
998  */
999 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1000 {
1001 	return skb->data - skb->head;
1002 }
1003 
1004 /**
1005  *	skb_tailroom - bytes at buffer end
1006  *	@skb: buffer to check
1007  *
1008  *	Return the number of bytes of free space at the tail of an sk_buff
1009  */
1010 static inline int skb_tailroom(const struct sk_buff *skb)
1011 {
1012 	return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1013 }
1014 
1015 /**
1016  *	skb_reserve - adjust headroom
1017  *	@skb: buffer to alter
1018  *	@len: bytes to move
1019  *
1020  *	Increase the headroom of an empty &sk_buff by reducing the tail
1021  *	room. This is only allowed for an empty buffer.
1022  */
1023 static inline void skb_reserve(struct sk_buff *skb, int len)
1024 {
1025 	skb->data += len;
1026 	skb->tail += len;
1027 }
1028 
1029 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1030 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1031 {
1032 	return skb->head + skb->transport_header;
1033 }
1034 
1035 static inline void skb_reset_transport_header(struct sk_buff *skb)
1036 {
1037 	skb->transport_header = skb->data - skb->head;
1038 }
1039 
1040 static inline void skb_set_transport_header(struct sk_buff *skb,
1041 					    const int offset)
1042 {
1043 	skb_reset_transport_header(skb);
1044 	skb->transport_header += offset;
1045 }
1046 
1047 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1048 {
1049 	return skb->head + skb->network_header;
1050 }
1051 
1052 static inline void skb_reset_network_header(struct sk_buff *skb)
1053 {
1054 	skb->network_header = skb->data - skb->head;
1055 }
1056 
1057 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1058 {
1059 	skb_reset_network_header(skb);
1060 	skb->network_header += offset;
1061 }
1062 
1063 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1064 {
1065 	return skb->head + skb->mac_header;
1066 }
1067 
1068 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1069 {
1070 	return skb->mac_header != ~0U;
1071 }
1072 
1073 static inline void skb_reset_mac_header(struct sk_buff *skb)
1074 {
1075 	skb->mac_header = skb->data - skb->head;
1076 }
1077 
1078 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1079 {
1080 	skb_reset_mac_header(skb);
1081 	skb->mac_header += offset;
1082 }
1083 
1084 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1085 
1086 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1087 {
1088 	return skb->transport_header;
1089 }
1090 
1091 static inline void skb_reset_transport_header(struct sk_buff *skb)
1092 {
1093 	skb->transport_header = skb->data;
1094 }
1095 
1096 static inline void skb_set_transport_header(struct sk_buff *skb,
1097 					    const int offset)
1098 {
1099 	skb->transport_header = skb->data + offset;
1100 }
1101 
1102 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1103 {
1104 	return skb->network_header;
1105 }
1106 
1107 static inline void skb_reset_network_header(struct sk_buff *skb)
1108 {
1109 	skb->network_header = skb->data;
1110 }
1111 
1112 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1113 {
1114 	skb->network_header = skb->data + offset;
1115 }
1116 
1117 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1118 {
1119 	return skb->mac_header;
1120 }
1121 
1122 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1123 {
1124 	return skb->mac_header != NULL;
1125 }
1126 
1127 static inline void skb_reset_mac_header(struct sk_buff *skb)
1128 {
1129 	skb->mac_header = skb->data;
1130 }
1131 
1132 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1133 {
1134 	skb->mac_header = skb->data + offset;
1135 }
1136 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1137 
1138 static inline int skb_transport_offset(const struct sk_buff *skb)
1139 {
1140 	return skb_transport_header(skb) - skb->data;
1141 }
1142 
1143 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1144 {
1145 	return skb->transport_header - skb->network_header;
1146 }
1147 
1148 static inline int skb_network_offset(const struct sk_buff *skb)
1149 {
1150 	return skb_network_header(skb) - skb->data;
1151 }
1152 
1153 /*
1154  * CPUs often take a performance hit when accessing unaligned memory
1155  * locations. The actual performance hit varies, it can be small if the
1156  * hardware handles it or large if we have to take an exception and fix it
1157  * in software.
1158  *
1159  * Since an ethernet header is 14 bytes network drivers often end up with
1160  * the IP header at an unaligned offset. The IP header can be aligned by
1161  * shifting the start of the packet by 2 bytes. Drivers should do this
1162  * with:
1163  *
1164  * skb_reserve(NET_IP_ALIGN);
1165  *
1166  * The downside to this alignment of the IP header is that the DMA is now
1167  * unaligned. On some architectures the cost of an unaligned DMA is high
1168  * and this cost outweighs the gains made by aligning the IP header.
1169  *
1170  * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1171  * to be overridden.
1172  */
1173 #ifndef NET_IP_ALIGN
1174 #define NET_IP_ALIGN	2
1175 #endif
1176 
1177 /*
1178  * The networking layer reserves some headroom in skb data (via
1179  * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1180  * the header has to grow. In the default case, if the header has to grow
1181  * 16 bytes or less we avoid the reallocation.
1182  *
1183  * Unfortunately this headroom changes the DMA alignment of the resulting
1184  * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1185  * on some architectures. An architecture can override this value,
1186  * perhaps setting it to a cacheline in size (since that will maintain
1187  * cacheline alignment of the DMA). It must be a power of 2.
1188  *
1189  * Various parts of the networking layer expect at least 16 bytes of
1190  * headroom, you should not reduce this.
1191  */
1192 #ifndef NET_SKB_PAD
1193 #define NET_SKB_PAD	16
1194 #endif
1195 
1196 extern int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1197 
1198 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1199 {
1200 	if (unlikely(skb->data_len)) {
1201 		WARN_ON(1);
1202 		return;
1203 	}
1204 	skb->len = len;
1205 	skb_set_tail_pointer(skb, len);
1206 }
1207 
1208 /**
1209  *	skb_trim - remove end from a buffer
1210  *	@skb: buffer to alter
1211  *	@len: new length
1212  *
1213  *	Cut the length of a buffer down by removing data from the tail. If
1214  *	the buffer is already under the length specified it is not modified.
1215  *	The skb must be linear.
1216  */
1217 static inline void skb_trim(struct sk_buff *skb, unsigned int len)
1218 {
1219 	if (skb->len > len)
1220 		__skb_trim(skb, len);
1221 }
1222 
1223 
1224 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1225 {
1226 	if (skb->data_len)
1227 		return ___pskb_trim(skb, len);
1228 	__skb_trim(skb, len);
1229 	return 0;
1230 }
1231 
1232 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1233 {
1234 	return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1235 }
1236 
1237 /**
1238  *	pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1239  *	@skb: buffer to alter
1240  *	@len: new length
1241  *
1242  *	This is identical to pskb_trim except that the caller knows that
1243  *	the skb is not cloned so we should never get an error due to out-
1244  *	of-memory.
1245  */
1246 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1247 {
1248 	int err = pskb_trim(skb, len);
1249 	BUG_ON(err);
1250 }
1251 
1252 /**
1253  *	skb_orphan - orphan a buffer
1254  *	@skb: buffer to orphan
1255  *
1256  *	If a buffer currently has an owner then we call the owner's
1257  *	destructor function and make the @skb unowned. The buffer continues
1258  *	to exist but is no longer charged to its former owner.
1259  */
1260 static inline void skb_orphan(struct sk_buff *skb)
1261 {
1262 	if (skb->destructor)
1263 		skb->destructor(skb);
1264 	skb->destructor = NULL;
1265 	skb->sk		= NULL;
1266 }
1267 
1268 /**
1269  *	__skb_queue_purge - empty a list
1270  *	@list: list to empty
1271  *
1272  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
1273  *	the list and one reference dropped. This function does not take the
1274  *	list lock and the caller must hold the relevant locks to use it.
1275  */
1276 extern void skb_queue_purge(struct sk_buff_head *list);
1277 static inline void __skb_queue_purge(struct sk_buff_head *list)
1278 {
1279 	struct sk_buff *skb;
1280 	while ((skb = __skb_dequeue(list)) != NULL)
1281 		kfree_skb(skb);
1282 }
1283 
1284 /**
1285  *	__dev_alloc_skb - allocate an skbuff for receiving
1286  *	@length: length to allocate
1287  *	@gfp_mask: get_free_pages mask, passed to alloc_skb
1288  *
1289  *	Allocate a new &sk_buff and assign it a usage count of one. The
1290  *	buffer has unspecified headroom built in. Users should allocate
1291  *	the headroom they think they need without accounting for the
1292  *	built in space. The built in space is used for optimisations.
1293  *
1294  *	%NULL is returned if there is no free memory.
1295  */
1296 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
1297 					      gfp_t gfp_mask)
1298 {
1299 	struct sk_buff *skb = alloc_skb(length + NET_SKB_PAD, gfp_mask);
1300 	if (likely(skb))
1301 		skb_reserve(skb, NET_SKB_PAD);
1302 	return skb;
1303 }
1304 
1305 /**
1306  *	dev_alloc_skb - allocate an skbuff for receiving
1307  *	@length: length to allocate
1308  *
1309  *	Allocate a new &sk_buff and assign it a usage count of one. The
1310  *	buffer has unspecified headroom built in. Users should allocate
1311  *	the headroom they think they need without accounting for the
1312  *	built in space. The built in space is used for optimisations.
1313  *
1314  *	%NULL is returned if there is no free memory. Although this function
1315  *	allocates memory it can be called from an interrupt.
1316  */
1317 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
1318 {
1319 	return __dev_alloc_skb(length, GFP_ATOMIC);
1320 }
1321 
1322 extern struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
1323 		unsigned int length, gfp_t gfp_mask);
1324 
1325 /**
1326  *	netdev_alloc_skb - allocate an skbuff for rx on a specific device
1327  *	@dev: network device to receive on
1328  *	@length: length to allocate
1329  *
1330  *	Allocate a new &sk_buff and assign it a usage count of one. The
1331  *	buffer has unspecified headroom built in. Users should allocate
1332  *	the headroom they think they need without accounting for the
1333  *	built in space. The built in space is used for optimisations.
1334  *
1335  *	%NULL is returned if there is no free memory. Although this function
1336  *	allocates memory it can be called from an interrupt.
1337  */
1338 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
1339 		unsigned int length)
1340 {
1341 	return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
1342 }
1343 
1344 /**
1345  *	skb_clone_writable - is the header of a clone writable
1346  *	@skb: buffer to check
1347  *	@len: length up to which to write
1348  *
1349  *	Returns true if modifying the header part of the cloned buffer
1350  *	does not requires the data to be copied.
1351  */
1352 static inline int skb_clone_writable(struct sk_buff *skb, unsigned int len)
1353 {
1354 	return !skb_header_cloned(skb) &&
1355 	       skb_headroom(skb) + len <= skb->hdr_len;
1356 }
1357 
1358 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
1359 			    int cloned)
1360 {
1361 	int delta = 0;
1362 
1363 	if (headroom < NET_SKB_PAD)
1364 		headroom = NET_SKB_PAD;
1365 	if (headroom > skb_headroom(skb))
1366 		delta = headroom - skb_headroom(skb);
1367 
1368 	if (delta || cloned)
1369 		return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
1370 					GFP_ATOMIC);
1371 	return 0;
1372 }
1373 
1374 /**
1375  *	skb_cow - copy header of skb when it is required
1376  *	@skb: buffer to cow
1377  *	@headroom: needed headroom
1378  *
1379  *	If the skb passed lacks sufficient headroom or its data part
1380  *	is shared, data is reallocated. If reallocation fails, an error
1381  *	is returned and original skb is not changed.
1382  *
1383  *	The result is skb with writable area skb->head...skb->tail
1384  *	and at least @headroom of space at head.
1385  */
1386 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
1387 {
1388 	return __skb_cow(skb, headroom, skb_cloned(skb));
1389 }
1390 
1391 /**
1392  *	skb_cow_head - skb_cow but only making the head writable
1393  *	@skb: buffer to cow
1394  *	@headroom: needed headroom
1395  *
1396  *	This function is identical to skb_cow except that we replace the
1397  *	skb_cloned check by skb_header_cloned.  It should be used when
1398  *	you only need to push on some header and do not need to modify
1399  *	the data.
1400  */
1401 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
1402 {
1403 	return __skb_cow(skb, headroom, skb_header_cloned(skb));
1404 }
1405 
1406 /**
1407  *	skb_padto	- pad an skbuff up to a minimal size
1408  *	@skb: buffer to pad
1409  *	@len: minimal length
1410  *
1411  *	Pads up a buffer to ensure the trailing bytes exist and are
1412  *	blanked. If the buffer already contains sufficient data it
1413  *	is untouched. Otherwise it is extended. Returns zero on
1414  *	success. The skb is freed on error.
1415  */
1416 
1417 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
1418 {
1419 	unsigned int size = skb->len;
1420 	if (likely(size >= len))
1421 		return 0;
1422 	return skb_pad(skb, len-size);
1423 }
1424 
1425 static inline int skb_add_data(struct sk_buff *skb,
1426 			       char __user *from, int copy)
1427 {
1428 	const int off = skb->len;
1429 
1430 	if (skb->ip_summed == CHECKSUM_NONE) {
1431 		int err = 0;
1432 		__wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
1433 							    copy, 0, &err);
1434 		if (!err) {
1435 			skb->csum = csum_block_add(skb->csum, csum, off);
1436 			return 0;
1437 		}
1438 	} else if (!copy_from_user(skb_put(skb, copy), from, copy))
1439 		return 0;
1440 
1441 	__skb_trim(skb, off);
1442 	return -EFAULT;
1443 }
1444 
1445 static inline int skb_can_coalesce(struct sk_buff *skb, int i,
1446 				   struct page *page, int off)
1447 {
1448 	if (i) {
1449 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1450 
1451 		return page == frag->page &&
1452 		       off == frag->page_offset + frag->size;
1453 	}
1454 	return 0;
1455 }
1456 
1457 static inline int __skb_linearize(struct sk_buff *skb)
1458 {
1459 	return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
1460 }
1461 
1462 /**
1463  *	skb_linearize - convert paged skb to linear one
1464  *	@skb: buffer to linarize
1465  *
1466  *	If there is no free memory -ENOMEM is returned, otherwise zero
1467  *	is returned and the old skb data released.
1468  */
1469 static inline int skb_linearize(struct sk_buff *skb)
1470 {
1471 	return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
1472 }
1473 
1474 /**
1475  *	skb_linearize_cow - make sure skb is linear and writable
1476  *	@skb: buffer to process
1477  *
1478  *	If there is no free memory -ENOMEM is returned, otherwise zero
1479  *	is returned and the old skb data released.
1480  */
1481 static inline int skb_linearize_cow(struct sk_buff *skb)
1482 {
1483 	return skb_is_nonlinear(skb) || skb_cloned(skb) ?
1484 	       __skb_linearize(skb) : 0;
1485 }
1486 
1487 /**
1488  *	skb_postpull_rcsum - update checksum for received skb after pull
1489  *	@skb: buffer to update
1490  *	@start: start of data before pull
1491  *	@len: length of data pulled
1492  *
1493  *	After doing a pull on a received packet, you need to call this to
1494  *	update the CHECKSUM_COMPLETE checksum, or set ip_summed to
1495  *	CHECKSUM_NONE so that it can be recomputed from scratch.
1496  */
1497 
1498 static inline void skb_postpull_rcsum(struct sk_buff *skb,
1499 				      const void *start, unsigned int len)
1500 {
1501 	if (skb->ip_summed == CHECKSUM_COMPLETE)
1502 		skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
1503 }
1504 
1505 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
1506 
1507 /**
1508  *	pskb_trim_rcsum - trim received skb and update checksum
1509  *	@skb: buffer to trim
1510  *	@len: new length
1511  *
1512  *	This is exactly the same as pskb_trim except that it ensures the
1513  *	checksum of received packets are still valid after the operation.
1514  */
1515 
1516 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
1517 {
1518 	if (likely(len >= skb->len))
1519 		return 0;
1520 	if (skb->ip_summed == CHECKSUM_COMPLETE)
1521 		skb->ip_summed = CHECKSUM_NONE;
1522 	return __pskb_trim(skb, len);
1523 }
1524 
1525 #define skb_queue_walk(queue, skb) \
1526 		for (skb = (queue)->next;					\
1527 		     prefetch(skb->next), (skb != (struct sk_buff *)(queue));	\
1528 		     skb = skb->next)
1529 
1530 #define skb_queue_walk_safe(queue, skb, tmp)					\
1531 		for (skb = (queue)->next, tmp = skb->next;			\
1532 		     skb != (struct sk_buff *)(queue);				\
1533 		     skb = tmp, tmp = skb->next)
1534 
1535 #define skb_queue_reverse_walk(queue, skb) \
1536 		for (skb = (queue)->prev;					\
1537 		     prefetch(skb->prev), (skb != (struct sk_buff *)(queue));	\
1538 		     skb = skb->prev)
1539 
1540 
1541 extern struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
1542 					   int *peeked, int *err);
1543 extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
1544 					 int noblock, int *err);
1545 extern unsigned int    datagram_poll(struct file *file, struct socket *sock,
1546 				     struct poll_table_struct *wait);
1547 extern int	       skb_copy_datagram_iovec(const struct sk_buff *from,
1548 					       int offset, struct iovec *to,
1549 					       int size);
1550 extern int	       skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
1551 							int hlen,
1552 							struct iovec *iov);
1553 extern void	       skb_free_datagram(struct sock *sk, struct sk_buff *skb);
1554 extern int	       skb_kill_datagram(struct sock *sk, struct sk_buff *skb,
1555 					 unsigned int flags);
1556 extern __wsum	       skb_checksum(const struct sk_buff *skb, int offset,
1557 				    int len, __wsum csum);
1558 extern int	       skb_copy_bits(const struct sk_buff *skb, int offset,
1559 				     void *to, int len);
1560 extern int	       skb_store_bits(struct sk_buff *skb, int offset,
1561 				      const void *from, int len);
1562 extern __wsum	       skb_copy_and_csum_bits(const struct sk_buff *skb,
1563 					      int offset, u8 *to, int len,
1564 					      __wsum csum);
1565 extern int             skb_splice_bits(struct sk_buff *skb,
1566 						unsigned int offset,
1567 						struct pipe_inode_info *pipe,
1568 						unsigned int len,
1569 						unsigned int flags);
1570 extern void	       skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
1571 extern void	       skb_split(struct sk_buff *skb,
1572 				 struct sk_buff *skb1, const u32 len);
1573 
1574 extern struct sk_buff *skb_segment(struct sk_buff *skb, int features);
1575 
1576 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
1577 				       int len, void *buffer)
1578 {
1579 	int hlen = skb_headlen(skb);
1580 
1581 	if (hlen - offset >= len)
1582 		return skb->data + offset;
1583 
1584 	if (skb_copy_bits(skb, offset, buffer, len) < 0)
1585 		return NULL;
1586 
1587 	return buffer;
1588 }
1589 
1590 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
1591 					     void *to,
1592 					     const unsigned int len)
1593 {
1594 	memcpy(to, skb->data, len);
1595 }
1596 
1597 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
1598 						    const int offset, void *to,
1599 						    const unsigned int len)
1600 {
1601 	memcpy(to, skb->data + offset, len);
1602 }
1603 
1604 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
1605 					   const void *from,
1606 					   const unsigned int len)
1607 {
1608 	memcpy(skb->data, from, len);
1609 }
1610 
1611 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
1612 						  const int offset,
1613 						  const void *from,
1614 						  const unsigned int len)
1615 {
1616 	memcpy(skb->data + offset, from, len);
1617 }
1618 
1619 extern void skb_init(void);
1620 
1621 /**
1622  *	skb_get_timestamp - get timestamp from a skb
1623  *	@skb: skb to get stamp from
1624  *	@stamp: pointer to struct timeval to store stamp in
1625  *
1626  *	Timestamps are stored in the skb as offsets to a base timestamp.
1627  *	This function converts the offset back to a struct timeval and stores
1628  *	it in stamp.
1629  */
1630 static inline void skb_get_timestamp(const struct sk_buff *skb, struct timeval *stamp)
1631 {
1632 	*stamp = ktime_to_timeval(skb->tstamp);
1633 }
1634 
1635 static inline void __net_timestamp(struct sk_buff *skb)
1636 {
1637 	skb->tstamp = ktime_get_real();
1638 }
1639 
1640 static inline ktime_t net_timedelta(ktime_t t)
1641 {
1642 	return ktime_sub(ktime_get_real(), t);
1643 }
1644 
1645 static inline ktime_t net_invalid_timestamp(void)
1646 {
1647 	return ktime_set(0, 0);
1648 }
1649 
1650 extern __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
1651 extern __sum16 __skb_checksum_complete(struct sk_buff *skb);
1652 
1653 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
1654 {
1655 	return skb->ip_summed & CHECKSUM_UNNECESSARY;
1656 }
1657 
1658 /**
1659  *	skb_checksum_complete - Calculate checksum of an entire packet
1660  *	@skb: packet to process
1661  *
1662  *	This function calculates the checksum over the entire packet plus
1663  *	the value of skb->csum.  The latter can be used to supply the
1664  *	checksum of a pseudo header as used by TCP/UDP.  It returns the
1665  *	checksum.
1666  *
1667  *	For protocols that contain complete checksums such as ICMP/TCP/UDP,
1668  *	this function can be used to verify that checksum on received
1669  *	packets.  In that case the function should return zero if the
1670  *	checksum is correct.  In particular, this function will return zero
1671  *	if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
1672  *	hardware has already verified the correctness of the checksum.
1673  */
1674 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
1675 {
1676 	return skb_csum_unnecessary(skb) ?
1677 	       0 : __skb_checksum_complete(skb);
1678 }
1679 
1680 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1681 extern void nf_conntrack_destroy(struct nf_conntrack *nfct);
1682 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
1683 {
1684 	if (nfct && atomic_dec_and_test(&nfct->use))
1685 		nf_conntrack_destroy(nfct);
1686 }
1687 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
1688 {
1689 	if (nfct)
1690 		atomic_inc(&nfct->use);
1691 }
1692 static inline void nf_conntrack_get_reasm(struct sk_buff *skb)
1693 {
1694 	if (skb)
1695 		atomic_inc(&skb->users);
1696 }
1697 static inline void nf_conntrack_put_reasm(struct sk_buff *skb)
1698 {
1699 	if (skb)
1700 		kfree_skb(skb);
1701 }
1702 #endif
1703 #ifdef CONFIG_BRIDGE_NETFILTER
1704 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
1705 {
1706 	if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
1707 		kfree(nf_bridge);
1708 }
1709 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
1710 {
1711 	if (nf_bridge)
1712 		atomic_inc(&nf_bridge->use);
1713 }
1714 #endif /* CONFIG_BRIDGE_NETFILTER */
1715 static inline void nf_reset(struct sk_buff *skb)
1716 {
1717 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1718 	nf_conntrack_put(skb->nfct);
1719 	skb->nfct = NULL;
1720 	nf_conntrack_put_reasm(skb->nfct_reasm);
1721 	skb->nfct_reasm = NULL;
1722 #endif
1723 #ifdef CONFIG_BRIDGE_NETFILTER
1724 	nf_bridge_put(skb->nf_bridge);
1725 	skb->nf_bridge = NULL;
1726 #endif
1727 }
1728 
1729 /* Note: This doesn't put any conntrack and bridge info in dst. */
1730 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
1731 {
1732 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1733 	dst->nfct = src->nfct;
1734 	nf_conntrack_get(src->nfct);
1735 	dst->nfctinfo = src->nfctinfo;
1736 	dst->nfct_reasm = src->nfct_reasm;
1737 	nf_conntrack_get_reasm(src->nfct_reasm);
1738 #endif
1739 #ifdef CONFIG_BRIDGE_NETFILTER
1740 	dst->nf_bridge  = src->nf_bridge;
1741 	nf_bridge_get(src->nf_bridge);
1742 #endif
1743 }
1744 
1745 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
1746 {
1747 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1748 	nf_conntrack_put(dst->nfct);
1749 	nf_conntrack_put_reasm(dst->nfct_reasm);
1750 #endif
1751 #ifdef CONFIG_BRIDGE_NETFILTER
1752 	nf_bridge_put(dst->nf_bridge);
1753 #endif
1754 	__nf_copy(dst, src);
1755 }
1756 
1757 #ifdef CONFIG_NETWORK_SECMARK
1758 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
1759 {
1760 	to->secmark = from->secmark;
1761 }
1762 
1763 static inline void skb_init_secmark(struct sk_buff *skb)
1764 {
1765 	skb->secmark = 0;
1766 }
1767 #else
1768 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
1769 { }
1770 
1771 static inline void skb_init_secmark(struct sk_buff *skb)
1772 { }
1773 #endif
1774 
1775 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
1776 {
1777 #ifdef CONFIG_NETDEVICES_MULTIQUEUE
1778 	skb->queue_mapping = queue_mapping;
1779 #endif
1780 }
1781 
1782 static inline u16 skb_get_queue_mapping(struct sk_buff *skb)
1783 {
1784 #ifdef CONFIG_NETDEVICES_MULTIQUEUE
1785 	return skb->queue_mapping;
1786 #else
1787 	return 0;
1788 #endif
1789 }
1790 
1791 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
1792 {
1793 #ifdef CONFIG_NETDEVICES_MULTIQUEUE
1794 	to->queue_mapping = from->queue_mapping;
1795 #endif
1796 }
1797 
1798 static inline int skb_is_gso(const struct sk_buff *skb)
1799 {
1800 	return skb_shinfo(skb)->gso_size;
1801 }
1802 
1803 static inline int skb_is_gso_v6(const struct sk_buff *skb)
1804 {
1805 	return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
1806 }
1807 
1808 static inline void skb_forward_csum(struct sk_buff *skb)
1809 {
1810 	/* Unfortunately we don't support this one.  Any brave souls? */
1811 	if (skb->ip_summed == CHECKSUM_COMPLETE)
1812 		skb->ip_summed = CHECKSUM_NONE;
1813 }
1814 
1815 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
1816 #endif	/* __KERNEL__ */
1817 #endif	/* _LINUX_SKBUFF_H */
1818