xref: /linux-6.15/include/linux/skbuff.h (revision 42fda663)
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) - sizeof(struct skb_shared_info)) & \
45 	 ~(SMP_CACHE_BYTES - 1))
46 #define SKB_MAX_ORDER(X, ORDER) \
47 	SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
48 #define SKB_MAX_HEAD(X)		(SKB_MAX_ORDER((X), 0))
49 #define SKB_MAX_ALLOC		(SKB_MAX_ORDER(0, 2))
50 
51 /* A. Checksumming of received packets by device.
52  *
53  *	NONE: device failed to checksum this packet.
54  *		skb->csum is undefined.
55  *
56  *	UNNECESSARY: device parsed packet and wouldbe verified checksum.
57  *		skb->csum is undefined.
58  *	      It is bad option, but, unfortunately, many of vendors do this.
59  *	      Apparently with secret goal to sell you new device, when you
60  *	      will add new protocol to your host. F.e. IPv6. 8)
61  *
62  *	COMPLETE: the most generic way. Device supplied checksum of _all_
63  *	    the packet as seen by netif_rx in skb->csum.
64  *	    NOTE: Even if device supports only some protocols, but
65  *	    is able to produce some skb->csum, it MUST use COMPLETE,
66  *	    not UNNECESSARY.
67  *
68  *	PARTIAL: identical to the case for output below.  This may occur
69  *	    on a packet received directly from another Linux OS, e.g.,
70  *	    a virtualised Linux kernel on the same host.  The packet can
71  *	    be treated in the same way as UNNECESSARY except that on
72  *	    output (i.e., forwarding) the checksum must be filled in
73  *	    by the OS or the hardware.
74  *
75  * B. Checksumming on output.
76  *
77  *	NONE: skb is checksummed by protocol or csum is not required.
78  *
79  *	PARTIAL: device is required to csum packet as seen by hard_start_xmit
80  *	from skb->csum_start to the end and to record the checksum
81  *	at skb->csum_start + skb->csum_offset.
82  *
83  *	Device must show its capabilities in dev->features, set
84  *	at device setup time.
85  *	NETIF_F_HW_CSUM	- it is clever device, it is able to checksum
86  *			  everything.
87  *	NETIF_F_NO_CSUM - loopback or reliable single hop media.
88  *	NETIF_F_IP_CSUM - device is dumb. It is able to csum only
89  *			  TCP/UDP over IPv4. Sigh. Vendors like this
90  *			  way by an unknown reason. Though, see comment above
91  *			  about CHECKSUM_UNNECESSARY. 8)
92  *	NETIF_F_IPV6_CSUM about as dumb as the last one but does IPv6 instead.
93  *
94  *	Any questions? No questions, good. 		--ANK
95  */
96 
97 struct net_device;
98 struct scatterlist;
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 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
112 	struct net_device *netoutdev;
113 #endif
114 	unsigned int mask;
115 	unsigned long data[32 / sizeof(unsigned long)];
116 };
117 #endif
118 
119 struct sk_buff_head {
120 	/* These two members must be first. */
121 	struct sk_buff	*next;
122 	struct sk_buff	*prev;
123 
124 	__u32		qlen;
125 	spinlock_t	lock;
126 };
127 
128 struct sk_buff;
129 
130 /* To allow 64K frame to be packed as single skb without frag_list */
131 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 2)
132 
133 typedef struct skb_frag_struct skb_frag_t;
134 
135 struct skb_frag_struct {
136 	struct page *page;
137 	__u32 page_offset;
138 	__u32 size;
139 };
140 
141 /* This data is invariant across clones and lives at
142  * the end of the header data, ie. at skb->end.
143  */
144 struct skb_shared_info {
145 	atomic_t	dataref;
146 	unsigned short	nr_frags;
147 	unsigned short	gso_size;
148 	/* Warning: this field is not always filled in (UFO)! */
149 	unsigned short	gso_segs;
150 	unsigned short  gso_type;
151 	__be32          ip6_frag_id;
152 	struct sk_buff	*frag_list;
153 	skb_frag_t	frags[MAX_SKB_FRAGS];
154 };
155 
156 /* We divide dataref into two halves.  The higher 16 bits hold references
157  * to the payload part of skb->data.  The lower 16 bits hold references to
158  * the entire skb->data.  A clone of a headerless skb holds the length of
159  * the header in skb->hdr_len.
160  *
161  * All users must obey the rule that the skb->data reference count must be
162  * greater than or equal to the payload reference count.
163  *
164  * Holding a reference to the payload part means that the user does not
165  * care about modifications to the header part of skb->data.
166  */
167 #define SKB_DATAREF_SHIFT 16
168 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
169 
170 
171 enum {
172 	SKB_FCLONE_UNAVAILABLE,
173 	SKB_FCLONE_ORIG,
174 	SKB_FCLONE_CLONE,
175 };
176 
177 enum {
178 	SKB_GSO_TCPV4 = 1 << 0,
179 	SKB_GSO_UDP = 1 << 1,
180 
181 	/* This indicates the skb is from an untrusted source. */
182 	SKB_GSO_DODGY = 1 << 2,
183 
184 	/* This indicates the tcp segment has CWR set. */
185 	SKB_GSO_TCP_ECN = 1 << 3,
186 
187 	SKB_GSO_TCPV6 = 1 << 4,
188 };
189 
190 #if BITS_PER_LONG > 32
191 #define NET_SKBUFF_DATA_USES_OFFSET 1
192 #endif
193 
194 #ifdef NET_SKBUFF_DATA_USES_OFFSET
195 typedef unsigned int sk_buff_data_t;
196 #else
197 typedef unsigned char *sk_buff_data_t;
198 #endif
199 
200 /**
201  *	struct sk_buff - socket buffer
202  *	@next: Next buffer in list
203  *	@prev: Previous buffer in list
204  *	@sk: Socket we are owned by
205  *	@tstamp: Time we arrived
206  *	@dev: Device we arrived on/are leaving by
207  *	@transport_header: Transport layer header
208  *	@network_header: Network layer header
209  *	@mac_header: Link layer header
210  *	@dst: destination entry
211  *	@sp: the security path, used for xfrm
212  *	@cb: Control buffer. Free for use by every layer. Put private vars here
213  *	@len: Length of actual data
214  *	@data_len: Data length
215  *	@mac_len: Length of link layer header
216  *	@hdr_len: writable header length of cloned skb
217  *	@csum: Checksum (must include start/offset pair)
218  *	@csum_start: Offset from skb->head where checksumming should start
219  *	@csum_offset: Offset from csum_start where checksum should be stored
220  *	@local_df: allow local fragmentation
221  *	@cloned: Head may be cloned (check refcnt to be sure)
222  *	@nohdr: Payload reference only, must not modify header
223  *	@pkt_type: Packet class
224  *	@fclone: skbuff clone status
225  *	@ip_summed: Driver fed us an IP checksum
226  *	@priority: Packet queueing priority
227  *	@users: User count - see {datagram,tcp}.c
228  *	@protocol: Packet protocol from driver
229  *	@truesize: Buffer size
230  *	@head: Head of buffer
231  *	@data: Data head pointer
232  *	@tail: Tail pointer
233  *	@end: End pointer
234  *	@destructor: Destruct function
235  *	@mark: Generic packet mark
236  *	@nfct: Associated connection, if any
237  *	@ipvs_property: skbuff is owned by ipvs
238  *	@nf_trace: netfilter packet trace flag
239  *	@nfctinfo: Relationship of this skb to the connection
240  *	@nfct_reasm: netfilter conntrack re-assembly pointer
241  *	@nf_bridge: Saved data about a bridged frame - see br_netfilter.c
242  *	@iif: ifindex of device we arrived on
243  *	@queue_mapping: Queue mapping for multiqueue devices
244  *	@tc_index: Traffic control index
245  *	@tc_verd: traffic control verdict
246  *	@dma_cookie: a cookie to one of several possible DMA operations
247  *		done by skb DMA functions
248  *	@secmark: security marking
249  */
250 
251 struct sk_buff {
252 	/* These two members must be first. */
253 	struct sk_buff		*next;
254 	struct sk_buff		*prev;
255 
256 	struct sock		*sk;
257 	ktime_t			tstamp;
258 	struct net_device	*dev;
259 
260 	struct  dst_entry	*dst;
261 	struct	sec_path	*sp;
262 
263 	/*
264 	 * This is the control buffer. It is free to use for every
265 	 * layer. Please put your private variables there. If you
266 	 * want to keep them across layers you have to do a skb_clone()
267 	 * first. This is owned by whoever has the skb queued ATM.
268 	 */
269 	char			cb[48];
270 
271 	unsigned int		len,
272 				data_len;
273 	__u16			mac_len,
274 				hdr_len;
275 	union {
276 		__wsum		csum;
277 		struct {
278 			__u16	csum_start;
279 			__u16	csum_offset;
280 		};
281 	};
282 	__u32			priority;
283 	__u8			local_df:1,
284 				cloned:1,
285 				ip_summed:2,
286 				nohdr:1,
287 				nfctinfo:3;
288 	__u8			pkt_type:3,
289 				fclone:2,
290 				ipvs_property: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 	__u16			queue_mapping;
305 
306 #ifdef CONFIG_NET_SCHED
307 	__u16			tc_index;	/* traffic control index */
308 #ifdef CONFIG_NET_CLS_ACT
309 	__u16			tc_verd;	/* traffic control verdict */
310 #endif
311 #endif
312 	/* 2 byte hole */
313 
314 #ifdef CONFIG_NET_DMA
315 	dma_cookie_t		dma_cookie;
316 #endif
317 #ifdef CONFIG_NETWORK_SECMARK
318 	__u32			secmark;
319 #endif
320 
321 	__u32			mark;
322 
323 	sk_buff_data_t		transport_header;
324 	sk_buff_data_t		network_header;
325 	sk_buff_data_t		mac_header;
326 	/* These elements must be at the end, see alloc_skb() for details.  */
327 	sk_buff_data_t		tail;
328 	sk_buff_data_t		end;
329 	unsigned char		*head,
330 				*data;
331 	unsigned int		truesize;
332 	atomic_t		users;
333 };
334 
335 #ifdef __KERNEL__
336 /*
337  *	Handling routines are only of interest to the kernel
338  */
339 #include <linux/slab.h>
340 
341 #include <asm/system.h>
342 
343 extern void kfree_skb(struct sk_buff *skb);
344 extern void	       __kfree_skb(struct sk_buff *skb);
345 extern struct sk_buff *__alloc_skb(unsigned int size,
346 				   gfp_t priority, int fclone, int node);
347 static inline struct sk_buff *alloc_skb(unsigned int size,
348 					gfp_t priority)
349 {
350 	return __alloc_skb(size, priority, 0, -1);
351 }
352 
353 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
354 					       gfp_t priority)
355 {
356 	return __alloc_skb(size, priority, 1, -1);
357 }
358 
359 extern void	       kfree_skbmem(struct sk_buff *skb);
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 	if (unlikely((int)skb->truesize < sizeof(struct sk_buff) + skb->len))
391 		skb_truesize_bug(skb);
392 }
393 
394 extern int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
395 			int getfrag(void *from, char *to, int offset,
396 			int len,int odd, struct sk_buff *skb),
397 			void *from, int length);
398 
399 struct skb_seq_state
400 {
401 	__u32		lower_offset;
402 	__u32		upper_offset;
403 	__u32		frag_idx;
404 	__u32		stepped_offset;
405 	struct sk_buff	*root_skb;
406 	struct sk_buff	*cur_skb;
407 	__u8		*frag_data;
408 };
409 
410 extern void	      skb_prepare_seq_read(struct sk_buff *skb,
411 					   unsigned int from, unsigned int to,
412 					   struct skb_seq_state *st);
413 extern unsigned int   skb_seq_read(unsigned int consumed, const u8 **data,
414 				   struct skb_seq_state *st);
415 extern void	      skb_abort_seq_read(struct skb_seq_state *st);
416 
417 extern unsigned int   skb_find_text(struct sk_buff *skb, unsigned int from,
418 				    unsigned int to, struct ts_config *config,
419 				    struct ts_state *state);
420 
421 #ifdef NET_SKBUFF_DATA_USES_OFFSET
422 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
423 {
424 	return skb->head + skb->end;
425 }
426 #else
427 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
428 {
429 	return skb->end;
430 }
431 #endif
432 
433 /* Internal */
434 #define skb_shinfo(SKB)	((struct skb_shared_info *)(skb_end_pointer(SKB)))
435 
436 /**
437  *	skb_queue_empty - check if a queue is empty
438  *	@list: queue head
439  *
440  *	Returns true if the queue is empty, false otherwise.
441  */
442 static inline int skb_queue_empty(const struct sk_buff_head *list)
443 {
444 	return list->next == (struct sk_buff *)list;
445 }
446 
447 /**
448  *	skb_get - reference buffer
449  *	@skb: buffer to reference
450  *
451  *	Makes another reference to a socket buffer and returns a pointer
452  *	to the buffer.
453  */
454 static inline struct sk_buff *skb_get(struct sk_buff *skb)
455 {
456 	atomic_inc(&skb->users);
457 	return skb;
458 }
459 
460 /*
461  * If users == 1, we are the only owner and are can avoid redundant
462  * atomic change.
463  */
464 
465 /**
466  *	skb_cloned - is the buffer a clone
467  *	@skb: buffer to check
468  *
469  *	Returns true if the buffer was generated with skb_clone() and is
470  *	one of multiple shared copies of the buffer. Cloned buffers are
471  *	shared data so must not be written to under normal circumstances.
472  */
473 static inline int skb_cloned(const struct sk_buff *skb)
474 {
475 	return skb->cloned &&
476 	       (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
477 }
478 
479 /**
480  *	skb_header_cloned - is the header a clone
481  *	@skb: buffer to check
482  *
483  *	Returns true if modifying the header part of the buffer requires
484  *	the data to be copied.
485  */
486 static inline int skb_header_cloned(const struct sk_buff *skb)
487 {
488 	int dataref;
489 
490 	if (!skb->cloned)
491 		return 0;
492 
493 	dataref = atomic_read(&skb_shinfo(skb)->dataref);
494 	dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
495 	return dataref != 1;
496 }
497 
498 /**
499  *	skb_header_release - release reference to header
500  *	@skb: buffer to operate on
501  *
502  *	Drop a reference to the header part of the buffer.  This is done
503  *	by acquiring a payload reference.  You must not read from the header
504  *	part of skb->data after this.
505  */
506 static inline void skb_header_release(struct sk_buff *skb)
507 {
508 	BUG_ON(skb->nohdr);
509 	skb->nohdr = 1;
510 	atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
511 }
512 
513 /**
514  *	skb_shared - is the buffer shared
515  *	@skb: buffer to check
516  *
517  *	Returns true if more than one person has a reference to this
518  *	buffer.
519  */
520 static inline int skb_shared(const struct sk_buff *skb)
521 {
522 	return atomic_read(&skb->users) != 1;
523 }
524 
525 /**
526  *	skb_share_check - check if buffer is shared and if so clone it
527  *	@skb: buffer to check
528  *	@pri: priority for memory allocation
529  *
530  *	If the buffer is shared the buffer is cloned and the old copy
531  *	drops a reference. A new clone with a single reference is returned.
532  *	If the buffer is not shared the original buffer is returned. When
533  *	being called from interrupt status or with spinlocks held pri must
534  *	be GFP_ATOMIC.
535  *
536  *	NULL is returned on a memory allocation failure.
537  */
538 static inline struct sk_buff *skb_share_check(struct sk_buff *skb,
539 					      gfp_t pri)
540 {
541 	might_sleep_if(pri & __GFP_WAIT);
542 	if (skb_shared(skb)) {
543 		struct sk_buff *nskb = skb_clone(skb, pri);
544 		kfree_skb(skb);
545 		skb = nskb;
546 	}
547 	return skb;
548 }
549 
550 /*
551  *	Copy shared buffers into a new sk_buff. We effectively do COW on
552  *	packets to handle cases where we have a local reader and forward
553  *	and a couple of other messy ones. The normal one is tcpdumping
554  *	a packet thats being forwarded.
555  */
556 
557 /**
558  *	skb_unshare - make a copy of a shared buffer
559  *	@skb: buffer to check
560  *	@pri: priority for memory allocation
561  *
562  *	If the socket buffer is a clone then this function creates a new
563  *	copy of the data, drops a reference count on the old copy and returns
564  *	the new copy with the reference count at 1. If the buffer is not a clone
565  *	the original buffer is returned. When called with a spinlock held or
566  *	from interrupt state @pri must be %GFP_ATOMIC
567  *
568  *	%NULL is returned on a memory allocation failure.
569  */
570 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
571 					  gfp_t pri)
572 {
573 	might_sleep_if(pri & __GFP_WAIT);
574 	if (skb_cloned(skb)) {
575 		struct sk_buff *nskb = skb_copy(skb, pri);
576 		kfree_skb(skb);	/* Free our shared copy */
577 		skb = nskb;
578 	}
579 	return skb;
580 }
581 
582 /**
583  *	skb_peek
584  *	@list_: list to peek at
585  *
586  *	Peek an &sk_buff. Unlike most other operations you _MUST_
587  *	be careful with this one. A peek leaves the buffer on the
588  *	list and someone else may run off with it. You must hold
589  *	the appropriate locks or have a private queue to do this.
590  *
591  *	Returns %NULL for an empty list or a pointer to the head element.
592  *	The reference count is not incremented and the reference is therefore
593  *	volatile. Use with caution.
594  */
595 static inline struct sk_buff *skb_peek(struct sk_buff_head *list_)
596 {
597 	struct sk_buff *list = ((struct sk_buff *)list_)->next;
598 	if (list == (struct sk_buff *)list_)
599 		list = NULL;
600 	return list;
601 }
602 
603 /**
604  *	skb_peek_tail
605  *	@list_: list to peek at
606  *
607  *	Peek an &sk_buff. Unlike most other operations you _MUST_
608  *	be careful with this one. A peek leaves the buffer on the
609  *	list and someone else may run off with it. You must hold
610  *	the appropriate locks or have a private queue to do this.
611  *
612  *	Returns %NULL for an empty list or a pointer to the tail element.
613  *	The reference count is not incremented and the reference is therefore
614  *	volatile. Use with caution.
615  */
616 static inline struct sk_buff *skb_peek_tail(struct sk_buff_head *list_)
617 {
618 	struct sk_buff *list = ((struct sk_buff *)list_)->prev;
619 	if (list == (struct sk_buff *)list_)
620 		list = NULL;
621 	return list;
622 }
623 
624 /**
625  *	skb_queue_len	- get queue length
626  *	@list_: list to measure
627  *
628  *	Return the length of an &sk_buff queue.
629  */
630 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
631 {
632 	return list_->qlen;
633 }
634 
635 /*
636  * This function creates a split out lock class for each invocation;
637  * this is needed for now since a whole lot of users of the skb-queue
638  * infrastructure in drivers have different locking usage (in hardirq)
639  * than the networking core (in softirq only). In the long run either the
640  * network layer or drivers should need annotation to consolidate the
641  * main types of usage into 3 classes.
642  */
643 static inline void skb_queue_head_init(struct sk_buff_head *list)
644 {
645 	spin_lock_init(&list->lock);
646 	list->prev = list->next = (struct sk_buff *)list;
647 	list->qlen = 0;
648 }
649 
650 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
651 		struct lock_class_key *class)
652 {
653 	skb_queue_head_init(list);
654 	lockdep_set_class(&list->lock, class);
655 }
656 
657 /*
658  *	Insert an sk_buff at the start of a list.
659  *
660  *	The "__skb_xxxx()" functions are the non-atomic ones that
661  *	can only be called with interrupts disabled.
662  */
663 
664 /**
665  *	__skb_queue_after - queue a buffer at the list head
666  *	@list: list to use
667  *	@prev: place after this buffer
668  *	@newsk: buffer to queue
669  *
670  *	Queue a buffer int the middle of a list. This function takes no locks
671  *	and you must therefore hold required locks before calling it.
672  *
673  *	A buffer cannot be placed on two lists at the same time.
674  */
675 static inline void __skb_queue_after(struct sk_buff_head *list,
676 				     struct sk_buff *prev,
677 				     struct sk_buff *newsk)
678 {
679 	struct sk_buff *next;
680 	list->qlen++;
681 
682 	next = prev->next;
683 	newsk->next = next;
684 	newsk->prev = prev;
685 	next->prev  = prev->next = newsk;
686 }
687 
688 /**
689  *	__skb_queue_head - queue a buffer at the list head
690  *	@list: list to use
691  *	@newsk: buffer to queue
692  *
693  *	Queue a buffer at the start of a list. This function takes no locks
694  *	and you must therefore hold required locks before calling it.
695  *
696  *	A buffer cannot be placed on two lists at the same time.
697  */
698 extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
699 static inline void __skb_queue_head(struct sk_buff_head *list,
700 				    struct sk_buff *newsk)
701 {
702 	__skb_queue_after(list, (struct sk_buff *)list, newsk);
703 }
704 
705 /**
706  *	__skb_queue_tail - queue a buffer at the list tail
707  *	@list: list to use
708  *	@newsk: buffer to queue
709  *
710  *	Queue a buffer at the end of a list. This function takes no locks
711  *	and you must therefore hold required locks before calling it.
712  *
713  *	A buffer cannot be placed on two lists at the same time.
714  */
715 extern void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
716 static inline void __skb_queue_tail(struct sk_buff_head *list,
717 				   struct sk_buff *newsk)
718 {
719 	struct sk_buff *prev, *next;
720 
721 	list->qlen++;
722 	next = (struct sk_buff *)list;
723 	prev = next->prev;
724 	newsk->next = next;
725 	newsk->prev = prev;
726 	next->prev  = prev->next = newsk;
727 }
728 
729 
730 /**
731  *	__skb_dequeue - remove from the head of the queue
732  *	@list: list to dequeue from
733  *
734  *	Remove the head of the list. This function does not take any locks
735  *	so must be used with appropriate locks held only. The head item is
736  *	returned or %NULL if the list is empty.
737  */
738 extern struct sk_buff *skb_dequeue(struct sk_buff_head *list);
739 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
740 {
741 	struct sk_buff *next, *prev, *result;
742 
743 	prev = (struct sk_buff *) list;
744 	next = prev->next;
745 	result = NULL;
746 	if (next != prev) {
747 		result	     = next;
748 		next	     = next->next;
749 		list->qlen--;
750 		next->prev   = prev;
751 		prev->next   = next;
752 		result->next = result->prev = NULL;
753 	}
754 	return result;
755 }
756 
757 
758 /*
759  *	Insert a packet on a list.
760  */
761 extern void        skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
762 static inline void __skb_insert(struct sk_buff *newsk,
763 				struct sk_buff *prev, struct sk_buff *next,
764 				struct sk_buff_head *list)
765 {
766 	newsk->next = next;
767 	newsk->prev = prev;
768 	next->prev  = prev->next = newsk;
769 	list->qlen++;
770 }
771 
772 /*
773  *	Place a packet after a given packet in a list.
774  */
775 extern void	   skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
776 static inline void __skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
777 {
778 	__skb_insert(newsk, old, old->next, list);
779 }
780 
781 /*
782  * remove sk_buff from list. _Must_ be called atomically, and with
783  * the list known..
784  */
785 extern void	   skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
786 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
787 {
788 	struct sk_buff *next, *prev;
789 
790 	list->qlen--;
791 	next	   = skb->next;
792 	prev	   = skb->prev;
793 	skb->next  = skb->prev = NULL;
794 	next->prev = prev;
795 	prev->next = next;
796 }
797 
798 
799 /* XXX: more streamlined implementation */
800 
801 /**
802  *	__skb_dequeue_tail - remove from the tail of the queue
803  *	@list: list to dequeue from
804  *
805  *	Remove the tail of the list. This function does not take any locks
806  *	so must be used with appropriate locks held only. The tail item is
807  *	returned or %NULL if the list is empty.
808  */
809 extern struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
810 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
811 {
812 	struct sk_buff *skb = skb_peek_tail(list);
813 	if (skb)
814 		__skb_unlink(skb, list);
815 	return skb;
816 }
817 
818 
819 static inline int skb_is_nonlinear(const struct sk_buff *skb)
820 {
821 	return skb->data_len;
822 }
823 
824 static inline unsigned int skb_headlen(const struct sk_buff *skb)
825 {
826 	return skb->len - skb->data_len;
827 }
828 
829 static inline int skb_pagelen(const struct sk_buff *skb)
830 {
831 	int i, len = 0;
832 
833 	for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
834 		len += skb_shinfo(skb)->frags[i].size;
835 	return len + skb_headlen(skb);
836 }
837 
838 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
839 				      struct page *page, int off, int size)
840 {
841 	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
842 
843 	frag->page		  = page;
844 	frag->page_offset	  = off;
845 	frag->size		  = size;
846 	skb_shinfo(skb)->nr_frags = i + 1;
847 }
848 
849 #define SKB_PAGE_ASSERT(skb) 	BUG_ON(skb_shinfo(skb)->nr_frags)
850 #define SKB_FRAG_ASSERT(skb) 	BUG_ON(skb_shinfo(skb)->frag_list)
851 #define SKB_LINEAR_ASSERT(skb)  BUG_ON(skb_is_nonlinear(skb))
852 
853 #ifdef NET_SKBUFF_DATA_USES_OFFSET
854 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
855 {
856 	return skb->head + skb->tail;
857 }
858 
859 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
860 {
861 	skb->tail = skb->data - skb->head;
862 }
863 
864 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
865 {
866 	skb_reset_tail_pointer(skb);
867 	skb->tail += offset;
868 }
869 #else /* NET_SKBUFF_DATA_USES_OFFSET */
870 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
871 {
872 	return skb->tail;
873 }
874 
875 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
876 {
877 	skb->tail = skb->data;
878 }
879 
880 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
881 {
882 	skb->tail = skb->data + offset;
883 }
884 
885 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
886 
887 /*
888  *	Add data to an sk_buff
889  */
890 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
891 {
892 	unsigned char *tmp = skb_tail_pointer(skb);
893 	SKB_LINEAR_ASSERT(skb);
894 	skb->tail += len;
895 	skb->len  += len;
896 	return tmp;
897 }
898 
899 /**
900  *	skb_put - add data to a buffer
901  *	@skb: buffer to use
902  *	@len: amount of data to add
903  *
904  *	This function extends the used data area of the buffer. If this would
905  *	exceed the total buffer size the kernel will panic. A pointer to the
906  *	first byte of the extra data is returned.
907  */
908 static inline unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
909 {
910 	unsigned char *tmp = skb_tail_pointer(skb);
911 	SKB_LINEAR_ASSERT(skb);
912 	skb->tail += len;
913 	skb->len  += len;
914 	if (unlikely(skb->tail > skb->end))
915 		skb_over_panic(skb, len, current_text_addr());
916 	return tmp;
917 }
918 
919 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
920 {
921 	skb->data -= len;
922 	skb->len  += len;
923 	return skb->data;
924 }
925 
926 /**
927  *	skb_push - add data to the start of a buffer
928  *	@skb: buffer to use
929  *	@len: amount of data to add
930  *
931  *	This function extends the used data area of the buffer at the buffer
932  *	start. If this would exceed the total buffer headroom the kernel will
933  *	panic. A pointer to the first byte of the extra data is returned.
934  */
935 static inline unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
936 {
937 	skb->data -= len;
938 	skb->len  += len;
939 	if (unlikely(skb->data<skb->head))
940 		skb_under_panic(skb, len, current_text_addr());
941 	return skb->data;
942 }
943 
944 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
945 {
946 	skb->len -= len;
947 	BUG_ON(skb->len < skb->data_len);
948 	return skb->data += len;
949 }
950 
951 /**
952  *	skb_pull - remove data from the start of a buffer
953  *	@skb: buffer to use
954  *	@len: amount of data to remove
955  *
956  *	This function removes data from the start of a buffer, returning
957  *	the memory to the headroom. A pointer to the next data in the buffer
958  *	is returned. Once the data has been pulled future pushes will overwrite
959  *	the old data.
960  */
961 static inline unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
962 {
963 	return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
964 }
965 
966 extern unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
967 
968 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
969 {
970 	if (len > skb_headlen(skb) &&
971 	    !__pskb_pull_tail(skb, len-skb_headlen(skb)))
972 		return NULL;
973 	skb->len -= len;
974 	return skb->data += len;
975 }
976 
977 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
978 {
979 	return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
980 }
981 
982 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
983 {
984 	if (likely(len <= skb_headlen(skb)))
985 		return 1;
986 	if (unlikely(len > skb->len))
987 		return 0;
988 	return __pskb_pull_tail(skb, len-skb_headlen(skb)) != NULL;
989 }
990 
991 /**
992  *	skb_headroom - bytes at buffer head
993  *	@skb: buffer to check
994  *
995  *	Return the number of bytes of free space at the head of an &sk_buff.
996  */
997 static inline int skb_headroom(const struct sk_buff *skb)
998 {
999 	return skb->data - skb->head;
1000 }
1001 
1002 /**
1003  *	skb_tailroom - bytes at buffer end
1004  *	@skb: buffer to check
1005  *
1006  *	Return the number of bytes of free space at the tail of an sk_buff
1007  */
1008 static inline int skb_tailroom(const struct sk_buff *skb)
1009 {
1010 	return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1011 }
1012 
1013 /**
1014  *	skb_reserve - adjust headroom
1015  *	@skb: buffer to alter
1016  *	@len: bytes to move
1017  *
1018  *	Increase the headroom of an empty &sk_buff by reducing the tail
1019  *	room. This is only allowed for an empty buffer.
1020  */
1021 static inline void skb_reserve(struct sk_buff *skb, int len)
1022 {
1023 	skb->data += len;
1024 	skb->tail += len;
1025 }
1026 
1027 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1028 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1029 {
1030 	return skb->head + skb->transport_header;
1031 }
1032 
1033 static inline void skb_reset_transport_header(struct sk_buff *skb)
1034 {
1035 	skb->transport_header = skb->data - skb->head;
1036 }
1037 
1038 static inline void skb_set_transport_header(struct sk_buff *skb,
1039 					    const int offset)
1040 {
1041 	skb_reset_transport_header(skb);
1042 	skb->transport_header += offset;
1043 }
1044 
1045 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1046 {
1047 	return skb->head + skb->network_header;
1048 }
1049 
1050 static inline void skb_reset_network_header(struct sk_buff *skb)
1051 {
1052 	skb->network_header = skb->data - skb->head;
1053 }
1054 
1055 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1056 {
1057 	skb_reset_network_header(skb);
1058 	skb->network_header += offset;
1059 }
1060 
1061 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1062 {
1063 	return skb->head + skb->mac_header;
1064 }
1065 
1066 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1067 {
1068 	return skb->mac_header != ~0U;
1069 }
1070 
1071 static inline void skb_reset_mac_header(struct sk_buff *skb)
1072 {
1073 	skb->mac_header = skb->data - skb->head;
1074 }
1075 
1076 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1077 {
1078 	skb_reset_mac_header(skb);
1079 	skb->mac_header += offset;
1080 }
1081 
1082 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1083 
1084 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1085 {
1086 	return skb->transport_header;
1087 }
1088 
1089 static inline void skb_reset_transport_header(struct sk_buff *skb)
1090 {
1091 	skb->transport_header = skb->data;
1092 }
1093 
1094 static inline void skb_set_transport_header(struct sk_buff *skb,
1095 					    const int offset)
1096 {
1097 	skb->transport_header = skb->data + offset;
1098 }
1099 
1100 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1101 {
1102 	return skb->network_header;
1103 }
1104 
1105 static inline void skb_reset_network_header(struct sk_buff *skb)
1106 {
1107 	skb->network_header = skb->data;
1108 }
1109 
1110 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1111 {
1112 	skb->network_header = skb->data + offset;
1113 }
1114 
1115 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1116 {
1117 	return skb->mac_header;
1118 }
1119 
1120 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1121 {
1122 	return skb->mac_header != NULL;
1123 }
1124 
1125 static inline void skb_reset_mac_header(struct sk_buff *skb)
1126 {
1127 	skb->mac_header = skb->data;
1128 }
1129 
1130 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1131 {
1132 	skb->mac_header = skb->data + offset;
1133 }
1134 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1135 
1136 static inline int skb_transport_offset(const struct sk_buff *skb)
1137 {
1138 	return skb_transport_header(skb) - skb->data;
1139 }
1140 
1141 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1142 {
1143 	return skb->transport_header - skb->network_header;
1144 }
1145 
1146 static inline int skb_network_offset(const struct sk_buff *skb)
1147 {
1148 	return skb_network_header(skb) - skb->data;
1149 }
1150 
1151 /*
1152  * CPUs often take a performance hit when accessing unaligned memory
1153  * locations. The actual performance hit varies, it can be small if the
1154  * hardware handles it or large if we have to take an exception and fix it
1155  * in software.
1156  *
1157  * Since an ethernet header is 14 bytes network drivers often end up with
1158  * the IP header at an unaligned offset. The IP header can be aligned by
1159  * shifting the start of the packet by 2 bytes. Drivers should do this
1160  * with:
1161  *
1162  * skb_reserve(NET_IP_ALIGN);
1163  *
1164  * The downside to this alignment of the IP header is that the DMA is now
1165  * unaligned. On some architectures the cost of an unaligned DMA is high
1166  * and this cost outweighs the gains made by aligning the IP header.
1167  *
1168  * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1169  * to be overridden.
1170  */
1171 #ifndef NET_IP_ALIGN
1172 #define NET_IP_ALIGN	2
1173 #endif
1174 
1175 /*
1176  * The networking layer reserves some headroom in skb data (via
1177  * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1178  * the header has to grow. In the default case, if the header has to grow
1179  * 16 bytes or less we avoid the reallocation.
1180  *
1181  * Unfortunately this headroom changes the DMA alignment of the resulting
1182  * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1183  * on some architectures. An architecture can override this value,
1184  * perhaps setting it to a cacheline in size (since that will maintain
1185  * cacheline alignment of the DMA). It must be a power of 2.
1186  *
1187  * Various parts of the networking layer expect at least 16 bytes of
1188  * headroom, you should not reduce this.
1189  */
1190 #ifndef NET_SKB_PAD
1191 #define NET_SKB_PAD	16
1192 #endif
1193 
1194 extern int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1195 
1196 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1197 {
1198 	if (unlikely(skb->data_len)) {
1199 		WARN_ON(1);
1200 		return;
1201 	}
1202 	skb->len = len;
1203 	skb_set_tail_pointer(skb, len);
1204 }
1205 
1206 /**
1207  *	skb_trim - remove end from a buffer
1208  *	@skb: buffer to alter
1209  *	@len: new length
1210  *
1211  *	Cut the length of a buffer down by removing data from the tail. If
1212  *	the buffer is already under the length specified it is not modified.
1213  *	The skb must be linear.
1214  */
1215 static inline void skb_trim(struct sk_buff *skb, unsigned int len)
1216 {
1217 	if (skb->len > len)
1218 		__skb_trim(skb, len);
1219 }
1220 
1221 
1222 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1223 {
1224 	if (skb->data_len)
1225 		return ___pskb_trim(skb, len);
1226 	__skb_trim(skb, len);
1227 	return 0;
1228 }
1229 
1230 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1231 {
1232 	return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1233 }
1234 
1235 /**
1236  *	pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1237  *	@skb: buffer to alter
1238  *	@len: new length
1239  *
1240  *	This is identical to pskb_trim except that the caller knows that
1241  *	the skb is not cloned so we should never get an error due to out-
1242  *	of-memory.
1243  */
1244 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1245 {
1246 	int err = pskb_trim(skb, len);
1247 	BUG_ON(err);
1248 }
1249 
1250 /**
1251  *	skb_orphan - orphan a buffer
1252  *	@skb: buffer to orphan
1253  *
1254  *	If a buffer currently has an owner then we call the owner's
1255  *	destructor function and make the @skb unowned. The buffer continues
1256  *	to exist but is no longer charged to its former owner.
1257  */
1258 static inline void skb_orphan(struct sk_buff *skb)
1259 {
1260 	if (skb->destructor)
1261 		skb->destructor(skb);
1262 	skb->destructor = NULL;
1263 	skb->sk		= NULL;
1264 }
1265 
1266 /**
1267  *	__skb_queue_purge - empty a list
1268  *	@list: list to empty
1269  *
1270  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
1271  *	the list and one reference dropped. This function does not take the
1272  *	list lock and the caller must hold the relevant locks to use it.
1273  */
1274 extern void skb_queue_purge(struct sk_buff_head *list);
1275 static inline void __skb_queue_purge(struct sk_buff_head *list)
1276 {
1277 	struct sk_buff *skb;
1278 	while ((skb = __skb_dequeue(list)) != NULL)
1279 		kfree_skb(skb);
1280 }
1281 
1282 /**
1283  *	__dev_alloc_skb - allocate an skbuff for receiving
1284  *	@length: length to allocate
1285  *	@gfp_mask: get_free_pages mask, passed to alloc_skb
1286  *
1287  *	Allocate a new &sk_buff and assign it a usage count of one. The
1288  *	buffer has unspecified headroom built in. Users should allocate
1289  *	the headroom they think they need without accounting for the
1290  *	built in space. The built in space is used for optimisations.
1291  *
1292  *	%NULL is returned if there is no free memory.
1293  */
1294 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
1295 					      gfp_t gfp_mask)
1296 {
1297 	struct sk_buff *skb = alloc_skb(length + NET_SKB_PAD, gfp_mask);
1298 	if (likely(skb))
1299 		skb_reserve(skb, NET_SKB_PAD);
1300 	return skb;
1301 }
1302 
1303 /**
1304  *	dev_alloc_skb - allocate an skbuff for receiving
1305  *	@length: length to allocate
1306  *
1307  *	Allocate a new &sk_buff and assign it a usage count of one. The
1308  *	buffer has unspecified headroom built in. Users should allocate
1309  *	the headroom they think they need without accounting for the
1310  *	built in space. The built in space is used for optimisations.
1311  *
1312  *	%NULL is returned if there is no free memory. Although this function
1313  *	allocates memory it can be called from an interrupt.
1314  */
1315 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
1316 {
1317 	return __dev_alloc_skb(length, GFP_ATOMIC);
1318 }
1319 
1320 extern struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
1321 		unsigned int length, gfp_t gfp_mask);
1322 
1323 /**
1324  *	netdev_alloc_skb - allocate an skbuff for rx on a specific device
1325  *	@dev: network device to receive on
1326  *	@length: length to allocate
1327  *
1328  *	Allocate a new &sk_buff and assign it a usage count of one. The
1329  *	buffer has unspecified headroom built in. Users should allocate
1330  *	the headroom they think they need without accounting for the
1331  *	built in space. The built in space is used for optimisations.
1332  *
1333  *	%NULL is returned if there is no free memory. Although this function
1334  *	allocates memory it can be called from an interrupt.
1335  */
1336 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
1337 		unsigned int length)
1338 {
1339 	return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
1340 }
1341 
1342 /**
1343  *	skb_clone_writable - is the header of a clone writable
1344  *	@skb: buffer to check
1345  *	@len: length up to which to write
1346  *
1347  *	Returns true if modifying the header part of the cloned buffer
1348  *	does not requires the data to be copied.
1349  */
1350 static inline int skb_clone_writable(struct sk_buff *skb, int len)
1351 {
1352 	return !skb_header_cloned(skb) &&
1353 	       skb_headroom(skb) + len <= skb->hdr_len;
1354 }
1355 
1356 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
1357 			    int cloned)
1358 {
1359 	int delta = 0;
1360 
1361 	if (headroom < NET_SKB_PAD)
1362 		headroom = NET_SKB_PAD;
1363 	if (headroom > skb_headroom(skb))
1364 		delta = headroom - skb_headroom(skb);
1365 
1366 	if (delta || cloned)
1367 		return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
1368 					GFP_ATOMIC);
1369 	return 0;
1370 }
1371 
1372 /**
1373  *	skb_cow - copy header of skb when it is required
1374  *	@skb: buffer to cow
1375  *	@headroom: needed headroom
1376  *
1377  *	If the skb passed lacks sufficient headroom or its data part
1378  *	is shared, data is reallocated. If reallocation fails, an error
1379  *	is returned and original skb is not changed.
1380  *
1381  *	The result is skb with writable area skb->head...skb->tail
1382  *	and at least @headroom of space at head.
1383  */
1384 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
1385 {
1386 	return __skb_cow(skb, headroom, skb_cloned(skb));
1387 }
1388 
1389 /**
1390  *	skb_cow_head - skb_cow but only making the head writable
1391  *	@skb: buffer to cow
1392  *	@headroom: needed headroom
1393  *
1394  *	This function is identical to skb_cow except that we replace the
1395  *	skb_cloned check by skb_header_cloned.  It should be used when
1396  *	you only need to push on some header and do not need to modify
1397  *	the data.
1398  */
1399 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
1400 {
1401 	return __skb_cow(skb, headroom, skb_header_cloned(skb));
1402 }
1403 
1404 /**
1405  *	skb_padto	- pad an skbuff up to a minimal size
1406  *	@skb: buffer to pad
1407  *	@len: minimal length
1408  *
1409  *	Pads up a buffer to ensure the trailing bytes exist and are
1410  *	blanked. If the buffer already contains sufficient data it
1411  *	is untouched. Otherwise it is extended. Returns zero on
1412  *	success. The skb is freed on error.
1413  */
1414 
1415 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
1416 {
1417 	unsigned int size = skb->len;
1418 	if (likely(size >= len))
1419 		return 0;
1420 	return skb_pad(skb, len-size);
1421 }
1422 
1423 static inline int skb_add_data(struct sk_buff *skb,
1424 			       char __user *from, int copy)
1425 {
1426 	const int off = skb->len;
1427 
1428 	if (skb->ip_summed == CHECKSUM_NONE) {
1429 		int err = 0;
1430 		__wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
1431 							    copy, 0, &err);
1432 		if (!err) {
1433 			skb->csum = csum_block_add(skb->csum, csum, off);
1434 			return 0;
1435 		}
1436 	} else if (!copy_from_user(skb_put(skb, copy), from, copy))
1437 		return 0;
1438 
1439 	__skb_trim(skb, off);
1440 	return -EFAULT;
1441 }
1442 
1443 static inline int skb_can_coalesce(struct sk_buff *skb, int i,
1444 				   struct page *page, int off)
1445 {
1446 	if (i) {
1447 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1448 
1449 		return page == frag->page &&
1450 		       off == frag->page_offset + frag->size;
1451 	}
1452 	return 0;
1453 }
1454 
1455 static inline int __skb_linearize(struct sk_buff *skb)
1456 {
1457 	return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
1458 }
1459 
1460 /**
1461  *	skb_linearize - convert paged skb to linear one
1462  *	@skb: buffer to linarize
1463  *
1464  *	If there is no free memory -ENOMEM is returned, otherwise zero
1465  *	is returned and the old skb data released.
1466  */
1467 static inline int skb_linearize(struct sk_buff *skb)
1468 {
1469 	return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
1470 }
1471 
1472 /**
1473  *	skb_linearize_cow - make sure skb is linear and writable
1474  *	@skb: buffer to process
1475  *
1476  *	If there is no free memory -ENOMEM is returned, otherwise zero
1477  *	is returned and the old skb data released.
1478  */
1479 static inline int skb_linearize_cow(struct sk_buff *skb)
1480 {
1481 	return skb_is_nonlinear(skb) || skb_cloned(skb) ?
1482 	       __skb_linearize(skb) : 0;
1483 }
1484 
1485 /**
1486  *	skb_postpull_rcsum - update checksum for received skb after pull
1487  *	@skb: buffer to update
1488  *	@start: start of data before pull
1489  *	@len: length of data pulled
1490  *
1491  *	After doing a pull on a received packet, you need to call this to
1492  *	update the CHECKSUM_COMPLETE checksum, or set ip_summed to
1493  *	CHECKSUM_NONE so that it can be recomputed from scratch.
1494  */
1495 
1496 static inline void skb_postpull_rcsum(struct sk_buff *skb,
1497 				      const void *start, unsigned int len)
1498 {
1499 	if (skb->ip_summed == CHECKSUM_COMPLETE)
1500 		skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
1501 }
1502 
1503 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
1504 
1505 /**
1506  *	pskb_trim_rcsum - trim received skb and update checksum
1507  *	@skb: buffer to trim
1508  *	@len: new length
1509  *
1510  *	This is exactly the same as pskb_trim except that it ensures the
1511  *	checksum of received packets are still valid after the operation.
1512  */
1513 
1514 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
1515 {
1516 	if (likely(len >= skb->len))
1517 		return 0;
1518 	if (skb->ip_summed == CHECKSUM_COMPLETE)
1519 		skb->ip_summed = CHECKSUM_NONE;
1520 	return __pskb_trim(skb, len);
1521 }
1522 
1523 #define skb_queue_walk(queue, skb) \
1524 		for (skb = (queue)->next;					\
1525 		     prefetch(skb->next), (skb != (struct sk_buff *)(queue));	\
1526 		     skb = skb->next)
1527 
1528 #define skb_queue_walk_safe(queue, skb, tmp)					\
1529 		for (skb = (queue)->next, tmp = skb->next;			\
1530 		     skb != (struct sk_buff *)(queue);				\
1531 		     skb = tmp, tmp = skb->next)
1532 
1533 #define skb_queue_reverse_walk(queue, skb) \
1534 		for (skb = (queue)->prev;					\
1535 		     prefetch(skb->prev), (skb != (struct sk_buff *)(queue));	\
1536 		     skb = skb->prev)
1537 
1538 
1539 extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
1540 					 int noblock, int *err);
1541 extern unsigned int    datagram_poll(struct file *file, struct socket *sock,
1542 				     struct poll_table_struct *wait);
1543 extern int	       skb_copy_datagram_iovec(const struct sk_buff *from,
1544 					       int offset, struct iovec *to,
1545 					       int size);
1546 extern int	       skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
1547 							int hlen,
1548 							struct iovec *iov);
1549 extern void	       skb_free_datagram(struct sock *sk, struct sk_buff *skb);
1550 extern void	       skb_kill_datagram(struct sock *sk, struct sk_buff *skb,
1551 					 unsigned int flags);
1552 extern __wsum	       skb_checksum(const struct sk_buff *skb, int offset,
1553 				    int len, __wsum csum);
1554 extern int	       skb_copy_bits(const struct sk_buff *skb, int offset,
1555 				     void *to, int len);
1556 extern int	       skb_store_bits(struct sk_buff *skb, int offset,
1557 				      const void *from, int len);
1558 extern __wsum	       skb_copy_and_csum_bits(const struct sk_buff *skb,
1559 					      int offset, u8 *to, int len,
1560 					      __wsum csum);
1561 extern void	       skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
1562 extern void	       skb_split(struct sk_buff *skb,
1563 				 struct sk_buff *skb1, const u32 len);
1564 
1565 extern struct sk_buff *skb_segment(struct sk_buff *skb, int features);
1566 
1567 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
1568 				       int len, void *buffer)
1569 {
1570 	int hlen = skb_headlen(skb);
1571 
1572 	if (hlen - offset >= len)
1573 		return skb->data + offset;
1574 
1575 	if (skb_copy_bits(skb, offset, buffer, len) < 0)
1576 		return NULL;
1577 
1578 	return buffer;
1579 }
1580 
1581 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
1582 					     void *to,
1583 					     const unsigned int len)
1584 {
1585 	memcpy(to, skb->data, len);
1586 }
1587 
1588 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
1589 						    const int offset, void *to,
1590 						    const unsigned int len)
1591 {
1592 	memcpy(to, skb->data + offset, len);
1593 }
1594 
1595 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
1596 					   const void *from,
1597 					   const unsigned int len)
1598 {
1599 	memcpy(skb->data, from, len);
1600 }
1601 
1602 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
1603 						  const int offset,
1604 						  const void *from,
1605 						  const unsigned int len)
1606 {
1607 	memcpy(skb->data + offset, from, len);
1608 }
1609 
1610 extern void skb_init(void);
1611 
1612 /**
1613  *	skb_get_timestamp - get timestamp from a skb
1614  *	@skb: skb to get stamp from
1615  *	@stamp: pointer to struct timeval to store stamp in
1616  *
1617  *	Timestamps are stored in the skb as offsets to a base timestamp.
1618  *	This function converts the offset back to a struct timeval and stores
1619  *	it in stamp.
1620  */
1621 static inline void skb_get_timestamp(const struct sk_buff *skb, struct timeval *stamp)
1622 {
1623 	*stamp = ktime_to_timeval(skb->tstamp);
1624 }
1625 
1626 static inline void __net_timestamp(struct sk_buff *skb)
1627 {
1628 	skb->tstamp = ktime_get_real();
1629 }
1630 
1631 static inline ktime_t net_timedelta(ktime_t t)
1632 {
1633 	return ktime_sub(ktime_get_real(), t);
1634 }
1635 
1636 static inline ktime_t net_invalid_timestamp(void)
1637 {
1638 	return ktime_set(0, 0);
1639 }
1640 
1641 extern __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
1642 extern __sum16 __skb_checksum_complete(struct sk_buff *skb);
1643 
1644 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
1645 {
1646 	return skb->ip_summed & CHECKSUM_UNNECESSARY;
1647 }
1648 
1649 /**
1650  *	skb_checksum_complete - Calculate checksum of an entire packet
1651  *	@skb: packet to process
1652  *
1653  *	This function calculates the checksum over the entire packet plus
1654  *	the value of skb->csum.  The latter can be used to supply the
1655  *	checksum of a pseudo header as used by TCP/UDP.  It returns the
1656  *	checksum.
1657  *
1658  *	For protocols that contain complete checksums such as ICMP/TCP/UDP,
1659  *	this function can be used to verify that checksum on received
1660  *	packets.  In that case the function should return zero if the
1661  *	checksum is correct.  In particular, this function will return zero
1662  *	if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
1663  *	hardware has already verified the correctness of the checksum.
1664  */
1665 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
1666 {
1667 	return skb_csum_unnecessary(skb) ?
1668 	       0 : __skb_checksum_complete(skb);
1669 }
1670 
1671 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1672 extern void nf_conntrack_destroy(struct nf_conntrack *nfct);
1673 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
1674 {
1675 	if (nfct && atomic_dec_and_test(&nfct->use))
1676 		nf_conntrack_destroy(nfct);
1677 }
1678 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
1679 {
1680 	if (nfct)
1681 		atomic_inc(&nfct->use);
1682 }
1683 static inline void nf_conntrack_get_reasm(struct sk_buff *skb)
1684 {
1685 	if (skb)
1686 		atomic_inc(&skb->users);
1687 }
1688 static inline void nf_conntrack_put_reasm(struct sk_buff *skb)
1689 {
1690 	if (skb)
1691 		kfree_skb(skb);
1692 }
1693 #endif
1694 #ifdef CONFIG_BRIDGE_NETFILTER
1695 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
1696 {
1697 	if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
1698 		kfree(nf_bridge);
1699 }
1700 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
1701 {
1702 	if (nf_bridge)
1703 		atomic_inc(&nf_bridge->use);
1704 }
1705 #endif /* CONFIG_BRIDGE_NETFILTER */
1706 static inline void nf_reset(struct sk_buff *skb)
1707 {
1708 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1709 	nf_conntrack_put(skb->nfct);
1710 	skb->nfct = NULL;
1711 	nf_conntrack_put_reasm(skb->nfct_reasm);
1712 	skb->nfct_reasm = NULL;
1713 #endif
1714 #ifdef CONFIG_BRIDGE_NETFILTER
1715 	nf_bridge_put(skb->nf_bridge);
1716 	skb->nf_bridge = NULL;
1717 #endif
1718 }
1719 
1720 /* Note: This doesn't put any conntrack and bridge info in dst. */
1721 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
1722 {
1723 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1724 	dst->nfct = src->nfct;
1725 	nf_conntrack_get(src->nfct);
1726 	dst->nfctinfo = src->nfctinfo;
1727 	dst->nfct_reasm = src->nfct_reasm;
1728 	nf_conntrack_get_reasm(src->nfct_reasm);
1729 #endif
1730 #ifdef CONFIG_BRIDGE_NETFILTER
1731 	dst->nf_bridge  = src->nf_bridge;
1732 	nf_bridge_get(src->nf_bridge);
1733 #endif
1734 }
1735 
1736 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
1737 {
1738 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1739 	nf_conntrack_put(dst->nfct);
1740 	nf_conntrack_put_reasm(dst->nfct_reasm);
1741 #endif
1742 #ifdef CONFIG_BRIDGE_NETFILTER
1743 	nf_bridge_put(dst->nf_bridge);
1744 #endif
1745 	__nf_copy(dst, src);
1746 }
1747 
1748 #ifdef CONFIG_NETWORK_SECMARK
1749 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
1750 {
1751 	to->secmark = from->secmark;
1752 }
1753 
1754 static inline void skb_init_secmark(struct sk_buff *skb)
1755 {
1756 	skb->secmark = 0;
1757 }
1758 #else
1759 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
1760 { }
1761 
1762 static inline void skb_init_secmark(struct sk_buff *skb)
1763 { }
1764 #endif
1765 
1766 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
1767 {
1768 #ifdef CONFIG_NETDEVICES_MULTIQUEUE
1769 	skb->queue_mapping = queue_mapping;
1770 #endif
1771 }
1772 
1773 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
1774 {
1775 #ifdef CONFIG_NETDEVICES_MULTIQUEUE
1776 	to->queue_mapping = from->queue_mapping;
1777 #endif
1778 }
1779 
1780 static inline int skb_is_gso(const struct sk_buff *skb)
1781 {
1782 	return skb_shinfo(skb)->gso_size;
1783 }
1784 
1785 static inline int skb_is_gso_v6(const struct sk_buff *skb)
1786 {
1787 	return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
1788 }
1789 
1790 static inline void skb_forward_csum(struct sk_buff *skb)
1791 {
1792 	/* Unfortunately we don't support this one.  Any brave souls? */
1793 	if (skb->ip_summed == CHECKSUM_COMPLETE)
1794 		skb->ip_summed = CHECKSUM_NONE;
1795 }
1796 
1797 #endif	/* __KERNEL__ */
1798 #endif	/* _LINUX_SKBUFF_H */
1799