xref: /linux-6.15/include/linux/skbuff.h (revision 40efcb05)
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 
99 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
100 struct nf_conntrack {
101 	atomic_t use;
102 };
103 #endif
104 
105 #ifdef CONFIG_BRIDGE_NETFILTER
106 struct nf_bridge_info {
107 	atomic_t use;
108 	struct net_device *physindev;
109 	struct net_device *physoutdev;
110 #if defined(CONFIG_VLAN_8021Q) || defined(CONFIG_VLAN_8021Q_MODULE)
111 	struct net_device *netoutdev;
112 #endif
113 	unsigned int mask;
114 	unsigned long data[32 / sizeof(unsigned long)];
115 };
116 #endif
117 
118 struct sk_buff_head {
119 	/* These two members must be first. */
120 	struct sk_buff	*next;
121 	struct sk_buff	*prev;
122 
123 	__u32		qlen;
124 	spinlock_t	lock;
125 };
126 
127 struct sk_buff;
128 
129 /* To allow 64K frame to be packed as single skb without frag_list */
130 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 2)
131 
132 typedef struct skb_frag_struct skb_frag_t;
133 
134 struct skb_frag_struct {
135 	struct page *page;
136 	__u32 page_offset;
137 	__u32 size;
138 };
139 
140 /* This data is invariant across clones and lives at
141  * the end of the header data, ie. at skb->end.
142  */
143 struct skb_shared_info {
144 	atomic_t	dataref;
145 	unsigned short	nr_frags;
146 	unsigned short	gso_size;
147 	/* Warning: this field is not always filled in (UFO)! */
148 	unsigned short	gso_segs;
149 	unsigned short  gso_type;
150 	__be32          ip6_frag_id;
151 	struct sk_buff	*frag_list;
152 	skb_frag_t	frags[MAX_SKB_FRAGS];
153 };
154 
155 /* We divide dataref into two halves.  The higher 16 bits hold references
156  * to the payload part of skb->data.  The lower 16 bits hold references to
157  * the entire skb->data.  A clone of a headerless skb holds the length of
158  * the header in skb->hdr_len.
159  *
160  * All users must obey the rule that the skb->data reference count must be
161  * greater than or equal to the payload reference count.
162  *
163  * Holding a reference to the payload part means that the user does not
164  * care about modifications to the header part of skb->data.
165  */
166 #define SKB_DATAREF_SHIFT 16
167 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
168 
169 
170 enum {
171 	SKB_FCLONE_UNAVAILABLE,
172 	SKB_FCLONE_ORIG,
173 	SKB_FCLONE_CLONE,
174 };
175 
176 enum {
177 	SKB_GSO_TCPV4 = 1 << 0,
178 	SKB_GSO_UDP = 1 << 1,
179 
180 	/* This indicates the skb is from an untrusted source. */
181 	SKB_GSO_DODGY = 1 << 2,
182 
183 	/* This indicates the tcp segment has CWR set. */
184 	SKB_GSO_TCP_ECN = 1 << 3,
185 
186 	SKB_GSO_TCPV6 = 1 << 4,
187 };
188 
189 #if BITS_PER_LONG > 32
190 #define NET_SKBUFF_DATA_USES_OFFSET 1
191 #endif
192 
193 #ifdef NET_SKBUFF_DATA_USES_OFFSET
194 typedef unsigned int sk_buff_data_t;
195 #else
196 typedef unsigned char *sk_buff_data_t;
197 #endif
198 
199 /**
200  *	struct sk_buff - socket buffer
201  *	@next: Next buffer in list
202  *	@prev: Previous buffer in list
203  *	@sk: Socket we are owned by
204  *	@tstamp: Time we arrived
205  *	@dev: Device we arrived on/are leaving by
206  *	@transport_header: Transport layer header
207  *	@network_header: Network layer header
208  *	@mac_header: Link layer header
209  *	@dst: destination entry
210  *	@sp: the security path, used for xfrm
211  *	@cb: Control buffer. Free for use by every layer. Put private vars here
212  *	@len: Length of actual data
213  *	@data_len: Data length
214  *	@mac_len: Length of link layer header
215  *	@hdr_len: writable header length of cloned skb
216  *	@csum: Checksum (must include start/offset pair)
217  *	@csum_start: Offset from skb->head where checksumming should start
218  *	@csum_offset: Offset from csum_start where checksum should be stored
219  *	@local_df: allow local fragmentation
220  *	@cloned: Head may be cloned (check refcnt to be sure)
221  *	@nohdr: Payload reference only, must not modify header
222  *	@pkt_type: Packet class
223  *	@fclone: skbuff clone status
224  *	@ip_summed: Driver fed us an IP checksum
225  *	@priority: Packet queueing priority
226  *	@users: User count - see {datagram,tcp}.c
227  *	@protocol: Packet protocol from driver
228  *	@truesize: Buffer size
229  *	@head: Head of buffer
230  *	@data: Data head pointer
231  *	@tail: Tail pointer
232  *	@end: End pointer
233  *	@destructor: Destruct function
234  *	@mark: Generic packet mark
235  *	@nfct: Associated connection, if any
236  *	@ipvs_property: skbuff is owned by ipvs
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 				nf_trace:1;
291 	__be16			protocol;
292 
293 	void			(*destructor)(struct sk_buff *skb);
294 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
295 	struct nf_conntrack	*nfct;
296 	struct sk_buff		*nfct_reasm;
297 #endif
298 #ifdef CONFIG_BRIDGE_NETFILTER
299 	struct nf_bridge_info	*nf_bridge;
300 #endif
301 
302 	int			iif;
303 #ifdef CONFIG_NETDEVICES_MULTIQUEUE
304 	__u16			queue_mapping;
305 #endif
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 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
360 extern struct sk_buff *skb_clone(struct sk_buff *skb,
361 				 gfp_t priority);
362 extern struct sk_buff *skb_copy(const struct sk_buff *skb,
363 				gfp_t priority);
364 extern struct sk_buff *pskb_copy(struct sk_buff *skb,
365 				 gfp_t gfp_mask);
366 extern int	       pskb_expand_head(struct sk_buff *skb,
367 					int nhead, int ntail,
368 					gfp_t gfp_mask);
369 extern struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
370 					    unsigned int headroom);
371 extern struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
372 				       int newheadroom, int newtailroom,
373 				       gfp_t priority);
374 extern int	       skb_to_sgvec(struct sk_buff *skb,
375 				    struct scatterlist *sg, int offset,
376 				    int len);
377 extern int	       skb_cow_data(struct sk_buff *skb, int tailbits,
378 				    struct sk_buff **trailer);
379 extern int	       skb_pad(struct sk_buff *skb, int pad);
380 #define dev_kfree_skb(a)	kfree_skb(a)
381 extern void	      skb_over_panic(struct sk_buff *skb, int len,
382 				     void *here);
383 extern void	      skb_under_panic(struct sk_buff *skb, int len,
384 				      void *here);
385 extern void	      skb_truesize_bug(struct sk_buff *skb);
386 
387 static inline void skb_truesize_check(struct sk_buff *skb)
388 {
389 	int len = sizeof(struct sk_buff) + skb->len;
390 
391 	if (unlikely((int)skb->truesize < len))
392 		skb_truesize_bug(skb);
393 }
394 
395 extern int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
396 			int getfrag(void *from, char *to, int offset,
397 			int len,int odd, struct sk_buff *skb),
398 			void *from, int length);
399 
400 struct skb_seq_state
401 {
402 	__u32		lower_offset;
403 	__u32		upper_offset;
404 	__u32		frag_idx;
405 	__u32		stepped_offset;
406 	struct sk_buff	*root_skb;
407 	struct sk_buff	*cur_skb;
408 	__u8		*frag_data;
409 };
410 
411 extern void	      skb_prepare_seq_read(struct sk_buff *skb,
412 					   unsigned int from, unsigned int to,
413 					   struct skb_seq_state *st);
414 extern unsigned int   skb_seq_read(unsigned int consumed, const u8 **data,
415 				   struct skb_seq_state *st);
416 extern void	      skb_abort_seq_read(struct skb_seq_state *st);
417 
418 extern unsigned int   skb_find_text(struct sk_buff *skb, unsigned int from,
419 				    unsigned int to, struct ts_config *config,
420 				    struct ts_state *state);
421 
422 #ifdef NET_SKBUFF_DATA_USES_OFFSET
423 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
424 {
425 	return skb->head + skb->end;
426 }
427 #else
428 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
429 {
430 	return skb->end;
431 }
432 #endif
433 
434 /* Internal */
435 #define skb_shinfo(SKB)	((struct skb_shared_info *)(skb_end_pointer(SKB)))
436 
437 /**
438  *	skb_queue_empty - check if a queue is empty
439  *	@list: queue head
440  *
441  *	Returns true if the queue is empty, false otherwise.
442  */
443 static inline int skb_queue_empty(const struct sk_buff_head *list)
444 {
445 	return list->next == (struct sk_buff *)list;
446 }
447 
448 /**
449  *	skb_get - reference buffer
450  *	@skb: buffer to reference
451  *
452  *	Makes another reference to a socket buffer and returns a pointer
453  *	to the buffer.
454  */
455 static inline struct sk_buff *skb_get(struct sk_buff *skb)
456 {
457 	atomic_inc(&skb->users);
458 	return skb;
459 }
460 
461 /*
462  * If users == 1, we are the only owner and are can avoid redundant
463  * atomic change.
464  */
465 
466 /**
467  *	skb_cloned - is the buffer a clone
468  *	@skb: buffer to check
469  *
470  *	Returns true if the buffer was generated with skb_clone() and is
471  *	one of multiple shared copies of the buffer. Cloned buffers are
472  *	shared data so must not be written to under normal circumstances.
473  */
474 static inline int skb_cloned(const struct sk_buff *skb)
475 {
476 	return skb->cloned &&
477 	       (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
478 }
479 
480 /**
481  *	skb_header_cloned - is the header a clone
482  *	@skb: buffer to check
483  *
484  *	Returns true if modifying the header part of the buffer requires
485  *	the data to be copied.
486  */
487 static inline int skb_header_cloned(const struct sk_buff *skb)
488 {
489 	int dataref;
490 
491 	if (!skb->cloned)
492 		return 0;
493 
494 	dataref = atomic_read(&skb_shinfo(skb)->dataref);
495 	dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
496 	return dataref != 1;
497 }
498 
499 /**
500  *	skb_header_release - release reference to header
501  *	@skb: buffer to operate on
502  *
503  *	Drop a reference to the header part of the buffer.  This is done
504  *	by acquiring a payload reference.  You must not read from the header
505  *	part of skb->data after this.
506  */
507 static inline void skb_header_release(struct sk_buff *skb)
508 {
509 	BUG_ON(skb->nohdr);
510 	skb->nohdr = 1;
511 	atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
512 }
513 
514 /**
515  *	skb_shared - is the buffer shared
516  *	@skb: buffer to check
517  *
518  *	Returns true if more than one person has a reference to this
519  *	buffer.
520  */
521 static inline int skb_shared(const struct sk_buff *skb)
522 {
523 	return atomic_read(&skb->users) != 1;
524 }
525 
526 /**
527  *	skb_share_check - check if buffer is shared and if so clone it
528  *	@skb: buffer to check
529  *	@pri: priority for memory allocation
530  *
531  *	If the buffer is shared the buffer is cloned and the old copy
532  *	drops a reference. A new clone with a single reference is returned.
533  *	If the buffer is not shared the original buffer is returned. When
534  *	being called from interrupt status or with spinlocks held pri must
535  *	be GFP_ATOMIC.
536  *
537  *	NULL is returned on a memory allocation failure.
538  */
539 static inline struct sk_buff *skb_share_check(struct sk_buff *skb,
540 					      gfp_t pri)
541 {
542 	might_sleep_if(pri & __GFP_WAIT);
543 	if (skb_shared(skb)) {
544 		struct sk_buff *nskb = skb_clone(skb, pri);
545 		kfree_skb(skb);
546 		skb = nskb;
547 	}
548 	return skb;
549 }
550 
551 /*
552  *	Copy shared buffers into a new sk_buff. We effectively do COW on
553  *	packets to handle cases where we have a local reader and forward
554  *	and a couple of other messy ones. The normal one is tcpdumping
555  *	a packet thats being forwarded.
556  */
557 
558 /**
559  *	skb_unshare - make a copy of a shared buffer
560  *	@skb: buffer to check
561  *	@pri: priority for memory allocation
562  *
563  *	If the socket buffer is a clone then this function creates a new
564  *	copy of the data, drops a reference count on the old copy and returns
565  *	the new copy with the reference count at 1. If the buffer is not a clone
566  *	the original buffer is returned. When called with a spinlock held or
567  *	from interrupt state @pri must be %GFP_ATOMIC
568  *
569  *	%NULL is returned on a memory allocation failure.
570  */
571 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
572 					  gfp_t pri)
573 {
574 	might_sleep_if(pri & __GFP_WAIT);
575 	if (skb_cloned(skb)) {
576 		struct sk_buff *nskb = skb_copy(skb, pri);
577 		kfree_skb(skb);	/* Free our shared copy */
578 		skb = nskb;
579 	}
580 	return skb;
581 }
582 
583 /**
584  *	skb_peek
585  *	@list_: list to peek at
586  *
587  *	Peek an &sk_buff. Unlike most other operations you _MUST_
588  *	be careful with this one. A peek leaves the buffer on the
589  *	list and someone else may run off with it. You must hold
590  *	the appropriate locks or have a private queue to do this.
591  *
592  *	Returns %NULL for an empty list or a pointer to the head element.
593  *	The reference count is not incremented and the reference is therefore
594  *	volatile. Use with caution.
595  */
596 static inline struct sk_buff *skb_peek(struct sk_buff_head *list_)
597 {
598 	struct sk_buff *list = ((struct sk_buff *)list_)->next;
599 	if (list == (struct sk_buff *)list_)
600 		list = NULL;
601 	return list;
602 }
603 
604 /**
605  *	skb_peek_tail
606  *	@list_: list to peek at
607  *
608  *	Peek an &sk_buff. Unlike most other operations you _MUST_
609  *	be careful with this one. A peek leaves the buffer on the
610  *	list and someone else may run off with it. You must hold
611  *	the appropriate locks or have a private queue to do this.
612  *
613  *	Returns %NULL for an empty list or a pointer to the tail element.
614  *	The reference count is not incremented and the reference is therefore
615  *	volatile. Use with caution.
616  */
617 static inline struct sk_buff *skb_peek_tail(struct sk_buff_head *list_)
618 {
619 	struct sk_buff *list = ((struct sk_buff *)list_)->prev;
620 	if (list == (struct sk_buff *)list_)
621 		list = NULL;
622 	return list;
623 }
624 
625 /**
626  *	skb_queue_len	- get queue length
627  *	@list_: list to measure
628  *
629  *	Return the length of an &sk_buff queue.
630  */
631 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
632 {
633 	return list_->qlen;
634 }
635 
636 /*
637  * This function creates a split out lock class for each invocation;
638  * this is needed for now since a whole lot of users of the skb-queue
639  * infrastructure in drivers have different locking usage (in hardirq)
640  * than the networking core (in softirq only). In the long run either the
641  * network layer or drivers should need annotation to consolidate the
642  * main types of usage into 3 classes.
643  */
644 static inline void skb_queue_head_init(struct sk_buff_head *list)
645 {
646 	spin_lock_init(&list->lock);
647 	list->prev = list->next = (struct sk_buff *)list;
648 	list->qlen = 0;
649 }
650 
651 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
652 		struct lock_class_key *class)
653 {
654 	skb_queue_head_init(list);
655 	lockdep_set_class(&list->lock, class);
656 }
657 
658 /*
659  *	Insert an sk_buff at the start of a list.
660  *
661  *	The "__skb_xxxx()" functions are the non-atomic ones that
662  *	can only be called with interrupts disabled.
663  */
664 
665 /**
666  *	__skb_queue_after - queue a buffer at the list head
667  *	@list: list to use
668  *	@prev: place after this buffer
669  *	@newsk: buffer to queue
670  *
671  *	Queue a buffer int the middle of a list. This function takes no locks
672  *	and you must therefore hold required locks before calling it.
673  *
674  *	A buffer cannot be placed on two lists at the same time.
675  */
676 static inline void __skb_queue_after(struct sk_buff_head *list,
677 				     struct sk_buff *prev,
678 				     struct sk_buff *newsk)
679 {
680 	struct sk_buff *next;
681 	list->qlen++;
682 
683 	next = prev->next;
684 	newsk->next = next;
685 	newsk->prev = prev;
686 	next->prev  = prev->next = newsk;
687 }
688 
689 /**
690  *	__skb_queue_head - queue a buffer at the list head
691  *	@list: list to use
692  *	@newsk: buffer to queue
693  *
694  *	Queue a buffer at the start of a list. This function takes no locks
695  *	and you must therefore hold required locks before calling it.
696  *
697  *	A buffer cannot be placed on two lists at the same time.
698  */
699 extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
700 static inline void __skb_queue_head(struct sk_buff_head *list,
701 				    struct sk_buff *newsk)
702 {
703 	__skb_queue_after(list, (struct sk_buff *)list, newsk);
704 }
705 
706 /**
707  *	__skb_queue_tail - queue a buffer at the list tail
708  *	@list: list to use
709  *	@newsk: buffer to queue
710  *
711  *	Queue a buffer at the end of a list. This function takes no locks
712  *	and you must therefore hold required locks before calling it.
713  *
714  *	A buffer cannot be placed on two lists at the same time.
715  */
716 extern void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
717 static inline void __skb_queue_tail(struct sk_buff_head *list,
718 				   struct sk_buff *newsk)
719 {
720 	struct sk_buff *prev, *next;
721 
722 	list->qlen++;
723 	next = (struct sk_buff *)list;
724 	prev = next->prev;
725 	newsk->next = next;
726 	newsk->prev = prev;
727 	next->prev  = prev->next = newsk;
728 }
729 
730 
731 /**
732  *	__skb_dequeue - remove from the head of the queue
733  *	@list: list to dequeue from
734  *
735  *	Remove the head of the list. This function does not take any locks
736  *	so must be used with appropriate locks held only. The head item is
737  *	returned or %NULL if the list is empty.
738  */
739 extern struct sk_buff *skb_dequeue(struct sk_buff_head *list);
740 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
741 {
742 	struct sk_buff *next, *prev, *result;
743 
744 	prev = (struct sk_buff *) list;
745 	next = prev->next;
746 	result = NULL;
747 	if (next != prev) {
748 		result	     = next;
749 		next	     = next->next;
750 		list->qlen--;
751 		next->prev   = prev;
752 		prev->next   = next;
753 		result->next = result->prev = NULL;
754 	}
755 	return result;
756 }
757 
758 
759 /*
760  *	Insert a packet on a list.
761  */
762 extern void        skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
763 static inline void __skb_insert(struct sk_buff *newsk,
764 				struct sk_buff *prev, struct sk_buff *next,
765 				struct sk_buff_head *list)
766 {
767 	newsk->next = next;
768 	newsk->prev = prev;
769 	next->prev  = prev->next = newsk;
770 	list->qlen++;
771 }
772 
773 /*
774  *	Place a packet after a given packet in a list.
775  */
776 extern void	   skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
777 static inline void __skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
778 {
779 	__skb_insert(newsk, old, old->next, list);
780 }
781 
782 /*
783  * remove sk_buff from list. _Must_ be called atomically, and with
784  * the list known..
785  */
786 extern void	   skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
787 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
788 {
789 	struct sk_buff *next, *prev;
790 
791 	list->qlen--;
792 	next	   = skb->next;
793 	prev	   = skb->prev;
794 	skb->next  = skb->prev = NULL;
795 	next->prev = prev;
796 	prev->next = next;
797 }
798 
799 
800 /* XXX: more streamlined implementation */
801 
802 /**
803  *	__skb_dequeue_tail - remove from the tail of the queue
804  *	@list: list to dequeue from
805  *
806  *	Remove the tail of the list. This function does not take any locks
807  *	so must be used with appropriate locks held only. The tail item is
808  *	returned or %NULL if the list is empty.
809  */
810 extern struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
811 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
812 {
813 	struct sk_buff *skb = skb_peek_tail(list);
814 	if (skb)
815 		__skb_unlink(skb, list);
816 	return skb;
817 }
818 
819 
820 static inline int skb_is_nonlinear(const struct sk_buff *skb)
821 {
822 	return skb->data_len;
823 }
824 
825 static inline unsigned int skb_headlen(const struct sk_buff *skb)
826 {
827 	return skb->len - skb->data_len;
828 }
829 
830 static inline int skb_pagelen(const struct sk_buff *skb)
831 {
832 	int i, len = 0;
833 
834 	for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
835 		len += skb_shinfo(skb)->frags[i].size;
836 	return len + skb_headlen(skb);
837 }
838 
839 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
840 				      struct page *page, int off, int size)
841 {
842 	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
843 
844 	frag->page		  = page;
845 	frag->page_offset	  = off;
846 	frag->size		  = size;
847 	skb_shinfo(skb)->nr_frags = i + 1;
848 }
849 
850 #define SKB_PAGE_ASSERT(skb) 	BUG_ON(skb_shinfo(skb)->nr_frags)
851 #define SKB_FRAG_ASSERT(skb) 	BUG_ON(skb_shinfo(skb)->frag_list)
852 #define SKB_LINEAR_ASSERT(skb)  BUG_ON(skb_is_nonlinear(skb))
853 
854 #ifdef NET_SKBUFF_DATA_USES_OFFSET
855 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
856 {
857 	return skb->head + skb->tail;
858 }
859 
860 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
861 {
862 	skb->tail = skb->data - skb->head;
863 }
864 
865 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
866 {
867 	skb_reset_tail_pointer(skb);
868 	skb->tail += offset;
869 }
870 #else /* NET_SKBUFF_DATA_USES_OFFSET */
871 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
872 {
873 	return skb->tail;
874 }
875 
876 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
877 {
878 	skb->tail = skb->data;
879 }
880 
881 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
882 {
883 	skb->tail = skb->data + offset;
884 }
885 
886 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
887 
888 /*
889  *	Add data to an sk_buff
890  */
891 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
892 {
893 	unsigned char *tmp = skb_tail_pointer(skb);
894 	SKB_LINEAR_ASSERT(skb);
895 	skb->tail += len;
896 	skb->len  += len;
897 	return tmp;
898 }
899 
900 /**
901  *	skb_put - add data to a buffer
902  *	@skb: buffer to use
903  *	@len: amount of data to add
904  *
905  *	This function extends the used data area of the buffer. If this would
906  *	exceed the total buffer size the kernel will panic. A pointer to the
907  *	first byte of the extra data is returned.
908  */
909 static inline unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
910 {
911 	unsigned char *tmp = skb_tail_pointer(skb);
912 	SKB_LINEAR_ASSERT(skb);
913 	skb->tail += len;
914 	skb->len  += len;
915 	if (unlikely(skb->tail > skb->end))
916 		skb_over_panic(skb, len, current_text_addr());
917 	return tmp;
918 }
919 
920 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
921 {
922 	skb->data -= len;
923 	skb->len  += len;
924 	return skb->data;
925 }
926 
927 /**
928  *	skb_push - add data to the start of a buffer
929  *	@skb: buffer to use
930  *	@len: amount of data to add
931  *
932  *	This function extends the used data area of the buffer at the buffer
933  *	start. If this would exceed the total buffer headroom the kernel will
934  *	panic. A pointer to the first byte of the extra data is returned.
935  */
936 static inline unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
937 {
938 	skb->data -= len;
939 	skb->len  += len;
940 	if (unlikely(skb->data<skb->head))
941 		skb_under_panic(skb, len, current_text_addr());
942 	return skb->data;
943 }
944 
945 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
946 {
947 	skb->len -= len;
948 	BUG_ON(skb->len < skb->data_len);
949 	return skb->data += len;
950 }
951 
952 /**
953  *	skb_pull - remove data from the start of a buffer
954  *	@skb: buffer to use
955  *	@len: amount of data to remove
956  *
957  *	This function removes data from the start of a buffer, returning
958  *	the memory to the headroom. A pointer to the next data in the buffer
959  *	is returned. Once the data has been pulled future pushes will overwrite
960  *	the old data.
961  */
962 static inline unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
963 {
964 	return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
965 }
966 
967 extern unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
968 
969 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
970 {
971 	if (len > skb_headlen(skb) &&
972 	    !__pskb_pull_tail(skb, len-skb_headlen(skb)))
973 		return NULL;
974 	skb->len -= len;
975 	return skb->data += len;
976 }
977 
978 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
979 {
980 	return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
981 }
982 
983 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
984 {
985 	if (likely(len <= skb_headlen(skb)))
986 		return 1;
987 	if (unlikely(len > skb->len))
988 		return 0;
989 	return __pskb_pull_tail(skb, len-skb_headlen(skb)) != NULL;
990 }
991 
992 /**
993  *	skb_headroom - bytes at buffer head
994  *	@skb: buffer to check
995  *
996  *	Return the number of bytes of free space at the head of an &sk_buff.
997  */
998 static inline unsigned int skb_headroom(const struct sk_buff *skb)
999 {
1000 	return skb->data - skb->head;
1001 }
1002 
1003 /**
1004  *	skb_tailroom - bytes at buffer end
1005  *	@skb: buffer to check
1006  *
1007  *	Return the number of bytes of free space at the tail of an sk_buff
1008  */
1009 static inline int skb_tailroom(const struct sk_buff *skb)
1010 {
1011 	return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1012 }
1013 
1014 /**
1015  *	skb_reserve - adjust headroom
1016  *	@skb: buffer to alter
1017  *	@len: bytes to move
1018  *
1019  *	Increase the headroom of an empty &sk_buff by reducing the tail
1020  *	room. This is only allowed for an empty buffer.
1021  */
1022 static inline void skb_reserve(struct sk_buff *skb, int len)
1023 {
1024 	skb->data += len;
1025 	skb->tail += len;
1026 }
1027 
1028 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1029 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1030 {
1031 	return skb->head + skb->transport_header;
1032 }
1033 
1034 static inline void skb_reset_transport_header(struct sk_buff *skb)
1035 {
1036 	skb->transport_header = skb->data - skb->head;
1037 }
1038 
1039 static inline void skb_set_transport_header(struct sk_buff *skb,
1040 					    const int offset)
1041 {
1042 	skb_reset_transport_header(skb);
1043 	skb->transport_header += offset;
1044 }
1045 
1046 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1047 {
1048 	return skb->head + skb->network_header;
1049 }
1050 
1051 static inline void skb_reset_network_header(struct sk_buff *skb)
1052 {
1053 	skb->network_header = skb->data - skb->head;
1054 }
1055 
1056 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1057 {
1058 	skb_reset_network_header(skb);
1059 	skb->network_header += offset;
1060 }
1061 
1062 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1063 {
1064 	return skb->head + skb->mac_header;
1065 }
1066 
1067 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1068 {
1069 	return skb->mac_header != ~0U;
1070 }
1071 
1072 static inline void skb_reset_mac_header(struct sk_buff *skb)
1073 {
1074 	skb->mac_header = skb->data - skb->head;
1075 }
1076 
1077 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1078 {
1079 	skb_reset_mac_header(skb);
1080 	skb->mac_header += offset;
1081 }
1082 
1083 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1084 
1085 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1086 {
1087 	return skb->transport_header;
1088 }
1089 
1090 static inline void skb_reset_transport_header(struct sk_buff *skb)
1091 {
1092 	skb->transport_header = skb->data;
1093 }
1094 
1095 static inline void skb_set_transport_header(struct sk_buff *skb,
1096 					    const int offset)
1097 {
1098 	skb->transport_header = skb->data + offset;
1099 }
1100 
1101 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1102 {
1103 	return skb->network_header;
1104 }
1105 
1106 static inline void skb_reset_network_header(struct sk_buff *skb)
1107 {
1108 	skb->network_header = skb->data;
1109 }
1110 
1111 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1112 {
1113 	skb->network_header = skb->data + offset;
1114 }
1115 
1116 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1117 {
1118 	return skb->mac_header;
1119 }
1120 
1121 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1122 {
1123 	return skb->mac_header != NULL;
1124 }
1125 
1126 static inline void skb_reset_mac_header(struct sk_buff *skb)
1127 {
1128 	skb->mac_header = skb->data;
1129 }
1130 
1131 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1132 {
1133 	skb->mac_header = skb->data + offset;
1134 }
1135 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1136 
1137 static inline int skb_transport_offset(const struct sk_buff *skb)
1138 {
1139 	return skb_transport_header(skb) - skb->data;
1140 }
1141 
1142 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1143 {
1144 	return skb->transport_header - skb->network_header;
1145 }
1146 
1147 static inline int skb_network_offset(const struct sk_buff *skb)
1148 {
1149 	return skb_network_header(skb) - skb->data;
1150 }
1151 
1152 /*
1153  * CPUs often take a performance hit when accessing unaligned memory
1154  * locations. The actual performance hit varies, it can be small if the
1155  * hardware handles it or large if we have to take an exception and fix it
1156  * in software.
1157  *
1158  * Since an ethernet header is 14 bytes network drivers often end up with
1159  * the IP header at an unaligned offset. The IP header can be aligned by
1160  * shifting the start of the packet by 2 bytes. Drivers should do this
1161  * with:
1162  *
1163  * skb_reserve(NET_IP_ALIGN);
1164  *
1165  * The downside to this alignment of the IP header is that the DMA is now
1166  * unaligned. On some architectures the cost of an unaligned DMA is high
1167  * and this cost outweighs the gains made by aligning the IP header.
1168  *
1169  * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1170  * to be overridden.
1171  */
1172 #ifndef NET_IP_ALIGN
1173 #define NET_IP_ALIGN	2
1174 #endif
1175 
1176 /*
1177  * The networking layer reserves some headroom in skb data (via
1178  * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1179  * the header has to grow. In the default case, if the header has to grow
1180  * 16 bytes or less we avoid the reallocation.
1181  *
1182  * Unfortunately this headroom changes the DMA alignment of the resulting
1183  * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1184  * on some architectures. An architecture can override this value,
1185  * perhaps setting it to a cacheline in size (since that will maintain
1186  * cacheline alignment of the DMA). It must be a power of 2.
1187  *
1188  * Various parts of the networking layer expect at least 16 bytes of
1189  * headroom, you should not reduce this.
1190  */
1191 #ifndef NET_SKB_PAD
1192 #define NET_SKB_PAD	16
1193 #endif
1194 
1195 extern int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1196 
1197 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1198 {
1199 	if (unlikely(skb->data_len)) {
1200 		WARN_ON(1);
1201 		return;
1202 	}
1203 	skb->len = len;
1204 	skb_set_tail_pointer(skb, len);
1205 }
1206 
1207 /**
1208  *	skb_trim - remove end from a buffer
1209  *	@skb: buffer to alter
1210  *	@len: new length
1211  *
1212  *	Cut the length of a buffer down by removing data from the tail. If
1213  *	the buffer is already under the length specified it is not modified.
1214  *	The skb must be linear.
1215  */
1216 static inline void skb_trim(struct sk_buff *skb, unsigned int len)
1217 {
1218 	if (skb->len > len)
1219 		__skb_trim(skb, len);
1220 }
1221 
1222 
1223 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1224 {
1225 	if (skb->data_len)
1226 		return ___pskb_trim(skb, len);
1227 	__skb_trim(skb, len);
1228 	return 0;
1229 }
1230 
1231 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1232 {
1233 	return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1234 }
1235 
1236 /**
1237  *	pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1238  *	@skb: buffer to alter
1239  *	@len: new length
1240  *
1241  *	This is identical to pskb_trim except that the caller knows that
1242  *	the skb is not cloned so we should never get an error due to out-
1243  *	of-memory.
1244  */
1245 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1246 {
1247 	int err = pskb_trim(skb, len);
1248 	BUG_ON(err);
1249 }
1250 
1251 /**
1252  *	skb_orphan - orphan a buffer
1253  *	@skb: buffer to orphan
1254  *
1255  *	If a buffer currently has an owner then we call the owner's
1256  *	destructor function and make the @skb unowned. The buffer continues
1257  *	to exist but is no longer charged to its former owner.
1258  */
1259 static inline void skb_orphan(struct sk_buff *skb)
1260 {
1261 	if (skb->destructor)
1262 		skb->destructor(skb);
1263 	skb->destructor = NULL;
1264 	skb->sk		= NULL;
1265 }
1266 
1267 /**
1268  *	__skb_queue_purge - empty a list
1269  *	@list: list to empty
1270  *
1271  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
1272  *	the list and one reference dropped. This function does not take the
1273  *	list lock and the caller must hold the relevant locks to use it.
1274  */
1275 extern void skb_queue_purge(struct sk_buff_head *list);
1276 static inline void __skb_queue_purge(struct sk_buff_head *list)
1277 {
1278 	struct sk_buff *skb;
1279 	while ((skb = __skb_dequeue(list)) != NULL)
1280 		kfree_skb(skb);
1281 }
1282 
1283 /**
1284  *	__dev_alloc_skb - allocate an skbuff for receiving
1285  *	@length: length to allocate
1286  *	@gfp_mask: get_free_pages mask, passed to alloc_skb
1287  *
1288  *	Allocate a new &sk_buff and assign it a usage count of one. The
1289  *	buffer has unspecified headroom built in. Users should allocate
1290  *	the headroom they think they need without accounting for the
1291  *	built in space. The built in space is used for optimisations.
1292  *
1293  *	%NULL is returned if there is no free memory.
1294  */
1295 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
1296 					      gfp_t gfp_mask)
1297 {
1298 	struct sk_buff *skb = alloc_skb(length + NET_SKB_PAD, gfp_mask);
1299 	if (likely(skb))
1300 		skb_reserve(skb, NET_SKB_PAD);
1301 	return skb;
1302 }
1303 
1304 /**
1305  *	dev_alloc_skb - allocate an skbuff for receiving
1306  *	@length: length to allocate
1307  *
1308  *	Allocate a new &sk_buff and assign it a usage count of one. The
1309  *	buffer has unspecified headroom built in. Users should allocate
1310  *	the headroom they think they need without accounting for the
1311  *	built in space. The built in space is used for optimisations.
1312  *
1313  *	%NULL is returned if there is no free memory. Although this function
1314  *	allocates memory it can be called from an interrupt.
1315  */
1316 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
1317 {
1318 	return __dev_alloc_skb(length, GFP_ATOMIC);
1319 }
1320 
1321 extern struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
1322 		unsigned int length, gfp_t gfp_mask);
1323 
1324 /**
1325  *	netdev_alloc_skb - allocate an skbuff for rx on a specific device
1326  *	@dev: network device to receive on
1327  *	@length: length to allocate
1328  *
1329  *	Allocate a new &sk_buff and assign it a usage count of one. The
1330  *	buffer has unspecified headroom built in. Users should allocate
1331  *	the headroom they think they need without accounting for the
1332  *	built in space. The built in space is used for optimisations.
1333  *
1334  *	%NULL is returned if there is no free memory. Although this function
1335  *	allocates memory it can be called from an interrupt.
1336  */
1337 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
1338 		unsigned int length)
1339 {
1340 	return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
1341 }
1342 
1343 /**
1344  *	skb_clone_writable - is the header of a clone writable
1345  *	@skb: buffer to check
1346  *	@len: length up to which to write
1347  *
1348  *	Returns true if modifying the header part of the cloned buffer
1349  *	does not requires the data to be copied.
1350  */
1351 static inline int skb_clone_writable(struct sk_buff *skb, unsigned int len)
1352 {
1353 	return !skb_header_cloned(skb) &&
1354 	       skb_headroom(skb) + len <= skb->hdr_len;
1355 }
1356 
1357 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
1358 			    int cloned)
1359 {
1360 	int delta = 0;
1361 
1362 	if (headroom < NET_SKB_PAD)
1363 		headroom = NET_SKB_PAD;
1364 	if (headroom > skb_headroom(skb))
1365 		delta = headroom - skb_headroom(skb);
1366 
1367 	if (delta || cloned)
1368 		return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
1369 					GFP_ATOMIC);
1370 	return 0;
1371 }
1372 
1373 /**
1374  *	skb_cow - copy header of skb when it is required
1375  *	@skb: buffer to cow
1376  *	@headroom: needed headroom
1377  *
1378  *	If the skb passed lacks sufficient headroom or its data part
1379  *	is shared, data is reallocated. If reallocation fails, an error
1380  *	is returned and original skb is not changed.
1381  *
1382  *	The result is skb with writable area skb->head...skb->tail
1383  *	and at least @headroom of space at head.
1384  */
1385 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
1386 {
1387 	return __skb_cow(skb, headroom, skb_cloned(skb));
1388 }
1389 
1390 /**
1391  *	skb_cow_head - skb_cow but only making the head writable
1392  *	@skb: buffer to cow
1393  *	@headroom: needed headroom
1394  *
1395  *	This function is identical to skb_cow except that we replace the
1396  *	skb_cloned check by skb_header_cloned.  It should be used when
1397  *	you only need to push on some header and do not need to modify
1398  *	the data.
1399  */
1400 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
1401 {
1402 	return __skb_cow(skb, headroom, skb_header_cloned(skb));
1403 }
1404 
1405 /**
1406  *	skb_padto	- pad an skbuff up to a minimal size
1407  *	@skb: buffer to pad
1408  *	@len: minimal length
1409  *
1410  *	Pads up a buffer to ensure the trailing bytes exist and are
1411  *	blanked. If the buffer already contains sufficient data it
1412  *	is untouched. Otherwise it is extended. Returns zero on
1413  *	success. The skb is freed on error.
1414  */
1415 
1416 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
1417 {
1418 	unsigned int size = skb->len;
1419 	if (likely(size >= len))
1420 		return 0;
1421 	return skb_pad(skb, len-size);
1422 }
1423 
1424 static inline int skb_add_data(struct sk_buff *skb,
1425 			       char __user *from, int copy)
1426 {
1427 	const int off = skb->len;
1428 
1429 	if (skb->ip_summed == CHECKSUM_NONE) {
1430 		int err = 0;
1431 		__wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
1432 							    copy, 0, &err);
1433 		if (!err) {
1434 			skb->csum = csum_block_add(skb->csum, csum, off);
1435 			return 0;
1436 		}
1437 	} else if (!copy_from_user(skb_put(skb, copy), from, copy))
1438 		return 0;
1439 
1440 	__skb_trim(skb, off);
1441 	return -EFAULT;
1442 }
1443 
1444 static inline int skb_can_coalesce(struct sk_buff *skb, int i,
1445 				   struct page *page, int off)
1446 {
1447 	if (i) {
1448 		struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1449 
1450 		return page == frag->page &&
1451 		       off == frag->page_offset + frag->size;
1452 	}
1453 	return 0;
1454 }
1455 
1456 static inline int __skb_linearize(struct sk_buff *skb)
1457 {
1458 	return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
1459 }
1460 
1461 /**
1462  *	skb_linearize - convert paged skb to linear one
1463  *	@skb: buffer to linarize
1464  *
1465  *	If there is no free memory -ENOMEM is returned, otherwise zero
1466  *	is returned and the old skb data released.
1467  */
1468 static inline int skb_linearize(struct sk_buff *skb)
1469 {
1470 	return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
1471 }
1472 
1473 /**
1474  *	skb_linearize_cow - make sure skb is linear and writable
1475  *	@skb: buffer to process
1476  *
1477  *	If there is no free memory -ENOMEM is returned, otherwise zero
1478  *	is returned and the old skb data released.
1479  */
1480 static inline int skb_linearize_cow(struct sk_buff *skb)
1481 {
1482 	return skb_is_nonlinear(skb) || skb_cloned(skb) ?
1483 	       __skb_linearize(skb) : 0;
1484 }
1485 
1486 /**
1487  *	skb_postpull_rcsum - update checksum for received skb after pull
1488  *	@skb: buffer to update
1489  *	@start: start of data before pull
1490  *	@len: length of data pulled
1491  *
1492  *	After doing a pull on a received packet, you need to call this to
1493  *	update the CHECKSUM_COMPLETE checksum, or set ip_summed to
1494  *	CHECKSUM_NONE so that it can be recomputed from scratch.
1495  */
1496 
1497 static inline void skb_postpull_rcsum(struct sk_buff *skb,
1498 				      const void *start, unsigned int len)
1499 {
1500 	if (skb->ip_summed == CHECKSUM_COMPLETE)
1501 		skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
1502 }
1503 
1504 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
1505 
1506 /**
1507  *	pskb_trim_rcsum - trim received skb and update checksum
1508  *	@skb: buffer to trim
1509  *	@len: new length
1510  *
1511  *	This is exactly the same as pskb_trim except that it ensures the
1512  *	checksum of received packets are still valid after the operation.
1513  */
1514 
1515 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
1516 {
1517 	if (likely(len >= skb->len))
1518 		return 0;
1519 	if (skb->ip_summed == CHECKSUM_COMPLETE)
1520 		skb->ip_summed = CHECKSUM_NONE;
1521 	return __pskb_trim(skb, len);
1522 }
1523 
1524 #define skb_queue_walk(queue, skb) \
1525 		for (skb = (queue)->next;					\
1526 		     prefetch(skb->next), (skb != (struct sk_buff *)(queue));	\
1527 		     skb = skb->next)
1528 
1529 #define skb_queue_walk_safe(queue, skb, tmp)					\
1530 		for (skb = (queue)->next, tmp = skb->next;			\
1531 		     skb != (struct sk_buff *)(queue);				\
1532 		     skb = tmp, tmp = skb->next)
1533 
1534 #define skb_queue_reverse_walk(queue, skb) \
1535 		for (skb = (queue)->prev;					\
1536 		     prefetch(skb->prev), (skb != (struct sk_buff *)(queue));	\
1537 		     skb = skb->prev)
1538 
1539 
1540 extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
1541 					 int noblock, int *err);
1542 extern unsigned int    datagram_poll(struct file *file, struct socket *sock,
1543 				     struct poll_table_struct *wait);
1544 extern int	       skb_copy_datagram_iovec(const struct sk_buff *from,
1545 					       int offset, struct iovec *to,
1546 					       int size);
1547 extern int	       skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
1548 							int hlen,
1549 							struct iovec *iov);
1550 extern void	       skb_free_datagram(struct sock *sk, struct sk_buff *skb);
1551 extern void	       skb_kill_datagram(struct sock *sk, struct sk_buff *skb,
1552 					 unsigned int flags);
1553 extern __wsum	       skb_checksum(const struct sk_buff *skb, int offset,
1554 				    int len, __wsum csum);
1555 extern int	       skb_copy_bits(const struct sk_buff *skb, int offset,
1556 				     void *to, int len);
1557 extern int	       skb_store_bits(struct sk_buff *skb, int offset,
1558 				      const void *from, int len);
1559 extern __wsum	       skb_copy_and_csum_bits(const struct sk_buff *skb,
1560 					      int offset, u8 *to, int len,
1561 					      __wsum csum);
1562 extern void	       skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
1563 extern void	       skb_split(struct sk_buff *skb,
1564 				 struct sk_buff *skb1, const u32 len);
1565 
1566 extern struct sk_buff *skb_segment(struct sk_buff *skb, int features);
1567 
1568 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
1569 				       int len, void *buffer)
1570 {
1571 	int hlen = skb_headlen(skb);
1572 
1573 	if (hlen - offset >= len)
1574 		return skb->data + offset;
1575 
1576 	if (skb_copy_bits(skb, offset, buffer, len) < 0)
1577 		return NULL;
1578 
1579 	return buffer;
1580 }
1581 
1582 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
1583 					     void *to,
1584 					     const unsigned int len)
1585 {
1586 	memcpy(to, skb->data, len);
1587 }
1588 
1589 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
1590 						    const int offset, void *to,
1591 						    const unsigned int len)
1592 {
1593 	memcpy(to, skb->data + offset, len);
1594 }
1595 
1596 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
1597 					   const void *from,
1598 					   const unsigned int len)
1599 {
1600 	memcpy(skb->data, from, len);
1601 }
1602 
1603 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
1604 						  const int offset,
1605 						  const void *from,
1606 						  const unsigned int len)
1607 {
1608 	memcpy(skb->data + offset, from, len);
1609 }
1610 
1611 extern void skb_init(void);
1612 
1613 /**
1614  *	skb_get_timestamp - get timestamp from a skb
1615  *	@skb: skb to get stamp from
1616  *	@stamp: pointer to struct timeval to store stamp in
1617  *
1618  *	Timestamps are stored in the skb as offsets to a base timestamp.
1619  *	This function converts the offset back to a struct timeval and stores
1620  *	it in stamp.
1621  */
1622 static inline void skb_get_timestamp(const struct sk_buff *skb, struct timeval *stamp)
1623 {
1624 	*stamp = ktime_to_timeval(skb->tstamp);
1625 }
1626 
1627 static inline void __net_timestamp(struct sk_buff *skb)
1628 {
1629 	skb->tstamp = ktime_get_real();
1630 }
1631 
1632 static inline ktime_t net_timedelta(ktime_t t)
1633 {
1634 	return ktime_sub(ktime_get_real(), t);
1635 }
1636 
1637 static inline ktime_t net_invalid_timestamp(void)
1638 {
1639 	return ktime_set(0, 0);
1640 }
1641 
1642 extern __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
1643 extern __sum16 __skb_checksum_complete(struct sk_buff *skb);
1644 
1645 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
1646 {
1647 	return skb->ip_summed & CHECKSUM_UNNECESSARY;
1648 }
1649 
1650 /**
1651  *	skb_checksum_complete - Calculate checksum of an entire packet
1652  *	@skb: packet to process
1653  *
1654  *	This function calculates the checksum over the entire packet plus
1655  *	the value of skb->csum.  The latter can be used to supply the
1656  *	checksum of a pseudo header as used by TCP/UDP.  It returns the
1657  *	checksum.
1658  *
1659  *	For protocols that contain complete checksums such as ICMP/TCP/UDP,
1660  *	this function can be used to verify that checksum on received
1661  *	packets.  In that case the function should return zero if the
1662  *	checksum is correct.  In particular, this function will return zero
1663  *	if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
1664  *	hardware has already verified the correctness of the checksum.
1665  */
1666 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
1667 {
1668 	return skb_csum_unnecessary(skb) ?
1669 	       0 : __skb_checksum_complete(skb);
1670 }
1671 
1672 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1673 extern void nf_conntrack_destroy(struct nf_conntrack *nfct);
1674 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
1675 {
1676 	if (nfct && atomic_dec_and_test(&nfct->use))
1677 		nf_conntrack_destroy(nfct);
1678 }
1679 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
1680 {
1681 	if (nfct)
1682 		atomic_inc(&nfct->use);
1683 }
1684 static inline void nf_conntrack_get_reasm(struct sk_buff *skb)
1685 {
1686 	if (skb)
1687 		atomic_inc(&skb->users);
1688 }
1689 static inline void nf_conntrack_put_reasm(struct sk_buff *skb)
1690 {
1691 	if (skb)
1692 		kfree_skb(skb);
1693 }
1694 #endif
1695 #ifdef CONFIG_BRIDGE_NETFILTER
1696 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
1697 {
1698 	if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
1699 		kfree(nf_bridge);
1700 }
1701 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
1702 {
1703 	if (nf_bridge)
1704 		atomic_inc(&nf_bridge->use);
1705 }
1706 #endif /* CONFIG_BRIDGE_NETFILTER */
1707 static inline void nf_reset(struct sk_buff *skb)
1708 {
1709 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1710 	nf_conntrack_put(skb->nfct);
1711 	skb->nfct = NULL;
1712 	nf_conntrack_put_reasm(skb->nfct_reasm);
1713 	skb->nfct_reasm = NULL;
1714 #endif
1715 #ifdef CONFIG_BRIDGE_NETFILTER
1716 	nf_bridge_put(skb->nf_bridge);
1717 	skb->nf_bridge = NULL;
1718 #endif
1719 }
1720 
1721 /* Note: This doesn't put any conntrack and bridge info in dst. */
1722 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
1723 {
1724 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1725 	dst->nfct = src->nfct;
1726 	nf_conntrack_get(src->nfct);
1727 	dst->nfctinfo = src->nfctinfo;
1728 	dst->nfct_reasm = src->nfct_reasm;
1729 	nf_conntrack_get_reasm(src->nfct_reasm);
1730 #endif
1731 #ifdef CONFIG_BRIDGE_NETFILTER
1732 	dst->nf_bridge  = src->nf_bridge;
1733 	nf_bridge_get(src->nf_bridge);
1734 #endif
1735 }
1736 
1737 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
1738 {
1739 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1740 	nf_conntrack_put(dst->nfct);
1741 	nf_conntrack_put_reasm(dst->nfct_reasm);
1742 #endif
1743 #ifdef CONFIG_BRIDGE_NETFILTER
1744 	nf_bridge_put(dst->nf_bridge);
1745 #endif
1746 	__nf_copy(dst, src);
1747 }
1748 
1749 #ifdef CONFIG_NETWORK_SECMARK
1750 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
1751 {
1752 	to->secmark = from->secmark;
1753 }
1754 
1755 static inline void skb_init_secmark(struct sk_buff *skb)
1756 {
1757 	skb->secmark = 0;
1758 }
1759 #else
1760 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
1761 { }
1762 
1763 static inline void skb_init_secmark(struct sk_buff *skb)
1764 { }
1765 #endif
1766 
1767 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
1768 {
1769 #ifdef CONFIG_NETDEVICES_MULTIQUEUE
1770 	skb->queue_mapping = queue_mapping;
1771 #endif
1772 }
1773 
1774 static inline u16 skb_get_queue_mapping(struct sk_buff *skb)
1775 {
1776 #ifdef CONFIG_NETDEVICES_MULTIQUEUE
1777 	return skb->queue_mapping;
1778 #else
1779 	return 0;
1780 #endif
1781 }
1782 
1783 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
1784 {
1785 #ifdef CONFIG_NETDEVICES_MULTIQUEUE
1786 	to->queue_mapping = from->queue_mapping;
1787 #endif
1788 }
1789 
1790 static inline int skb_is_gso(const struct sk_buff *skb)
1791 {
1792 	return skb_shinfo(skb)->gso_size;
1793 }
1794 
1795 static inline int skb_is_gso_v6(const struct sk_buff *skb)
1796 {
1797 	return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
1798 }
1799 
1800 static inline void skb_forward_csum(struct sk_buff *skb)
1801 {
1802 	/* Unfortunately we don't support this one.  Any brave souls? */
1803 	if (skb->ip_summed == CHECKSUM_COMPLETE)
1804 		skb->ip_summed = CHECKSUM_NONE;
1805 }
1806 
1807 #endif	/* __KERNEL__ */
1808 #endif	/* _LINUX_SKBUFF_H */
1809