xref: /linux-6.15/include/linux/skbuff.h (revision 7a875903)
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/kmemcheck.h>
19 #include <linux/compiler.h>
20 #include <linux/time.h>
21 #include <linux/bug.h>
22 #include <linux/cache.h>
23 
24 #include <linux/atomic.h>
25 #include <asm/types.h>
26 #include <linux/spinlock.h>
27 #include <linux/net.h>
28 #include <linux/textsearch.h>
29 #include <net/checksum.h>
30 #include <linux/rcupdate.h>
31 #include <linux/dmaengine.h>
32 #include <linux/hrtimer.h>
33 #include <linux/dma-mapping.h>
34 #include <linux/netdev_features.h>
35 
36 /* Don't change this without changing skb_csum_unnecessary! */
37 #define CHECKSUM_NONE 0
38 #define CHECKSUM_UNNECESSARY 1
39 #define CHECKSUM_COMPLETE 2
40 #define CHECKSUM_PARTIAL 3
41 
42 #define SKB_DATA_ALIGN(X)	(((X) + (SMP_CACHE_BYTES - 1)) & \
43 				 ~(SMP_CACHE_BYTES - 1))
44 #define SKB_WITH_OVERHEAD(X)	\
45 	((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
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 /* return minimum truesize of one skb containing X bytes of data */
52 #define SKB_TRUESIZE(X) ((X) +						\
53 			 SKB_DATA_ALIGN(sizeof(struct sk_buff)) +	\
54 			 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
55 
56 /* A. Checksumming of received packets by device.
57  *
58  *	NONE: device failed to checksum this packet.
59  *		skb->csum is undefined.
60  *
61  *	UNNECESSARY: device parsed packet and wouldbe verified checksum.
62  *		skb->csum is undefined.
63  *	      It is bad option, but, unfortunately, many of vendors do this.
64  *	      Apparently with secret goal to sell you new device, when you
65  *	      will add new protocol to your host. F.e. IPv6. 8)
66  *
67  *	COMPLETE: the most generic way. Device supplied checksum of _all_
68  *	    the packet as seen by netif_rx in skb->csum.
69  *	    NOTE: Even if device supports only some protocols, but
70  *	    is able to produce some skb->csum, it MUST use COMPLETE,
71  *	    not UNNECESSARY.
72  *
73  *	PARTIAL: identical to the case for output below.  This may occur
74  *	    on a packet received directly from another Linux OS, e.g.,
75  *	    a virtualised Linux kernel on the same host.  The packet can
76  *	    be treated in the same way as UNNECESSARY except that on
77  *	    output (i.e., forwarding) the checksum must be filled in
78  *	    by the OS or the hardware.
79  *
80  * B. Checksumming on output.
81  *
82  *	NONE: skb is checksummed by protocol or csum is not required.
83  *
84  *	PARTIAL: device is required to csum packet as seen by hard_start_xmit
85  *	from skb->csum_start to the end and to record the checksum
86  *	at skb->csum_start + skb->csum_offset.
87  *
88  *	Device must show its capabilities in dev->features, set
89  *	at device setup time.
90  *	NETIF_F_HW_CSUM	- it is clever device, it is able to checksum
91  *			  everything.
92  *	NETIF_F_IP_CSUM - device is dumb. It is able to csum only
93  *			  TCP/UDP over IPv4. Sigh. Vendors like this
94  *			  way by an unknown reason. Though, see comment above
95  *			  about CHECKSUM_UNNECESSARY. 8)
96  *	NETIF_F_IPV6_CSUM about as dumb as the last one but does IPv6 instead.
97  *
98  *	UNNECESSARY: device will do per protocol specific csum. Protocol drivers
99  *	that do not want net to perform the checksum calculation should use
100  *	this flag in their outgoing skbs.
101  *	NETIF_F_FCOE_CRC  this indicates the device can do FCoE FC CRC
102  *			  offload. Correspondingly, the FCoE protocol driver
103  *			  stack should use CHECKSUM_UNNECESSARY.
104  *
105  *	Any questions? No questions, good. 		--ANK
106  */
107 
108 struct net_device;
109 struct scatterlist;
110 struct pipe_inode_info;
111 
112 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
113 struct nf_conntrack {
114 	atomic_t use;
115 };
116 #endif
117 
118 #ifdef CONFIG_BRIDGE_NETFILTER
119 struct nf_bridge_info {
120 	atomic_t		use;
121 	unsigned int		mask;
122 	struct net_device	*physindev;
123 	struct net_device	*physoutdev;
124 	unsigned long		data[32 / sizeof(unsigned long)];
125 };
126 #endif
127 
128 struct sk_buff_head {
129 	/* These two members must be first. */
130 	struct sk_buff	*next;
131 	struct sk_buff	*prev;
132 
133 	__u32		qlen;
134 	spinlock_t	lock;
135 };
136 
137 struct sk_buff;
138 
139 /* To allow 64K frame to be packed as single skb without frag_list we
140  * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
141  * buffers which do not start on a page boundary.
142  *
143  * Since GRO uses frags we allocate at least 16 regardless of page
144  * size.
145  */
146 #if (65536/PAGE_SIZE + 1) < 16
147 #define MAX_SKB_FRAGS 16UL
148 #else
149 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
150 #endif
151 
152 typedef struct skb_frag_struct skb_frag_t;
153 
154 struct skb_frag_struct {
155 	struct {
156 		struct page *p;
157 	} page;
158 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
159 	__u32 page_offset;
160 	__u32 size;
161 #else
162 	__u16 page_offset;
163 	__u16 size;
164 #endif
165 };
166 
167 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
168 {
169 	return frag->size;
170 }
171 
172 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
173 {
174 	frag->size = size;
175 }
176 
177 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
178 {
179 	frag->size += delta;
180 }
181 
182 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
183 {
184 	frag->size -= delta;
185 }
186 
187 #define HAVE_HW_TIME_STAMP
188 
189 /**
190  * struct skb_shared_hwtstamps - hardware time stamps
191  * @hwtstamp:	hardware time stamp transformed into duration
192  *		since arbitrary point in time
193  * @syststamp:	hwtstamp transformed to system time base
194  *
195  * Software time stamps generated by ktime_get_real() are stored in
196  * skb->tstamp. The relation between the different kinds of time
197  * stamps is as follows:
198  *
199  * syststamp and tstamp can be compared against each other in
200  * arbitrary combinations.  The accuracy of a
201  * syststamp/tstamp/"syststamp from other device" comparison is
202  * limited by the accuracy of the transformation into system time
203  * base. This depends on the device driver and its underlying
204  * hardware.
205  *
206  * hwtstamps can only be compared against other hwtstamps from
207  * the same device.
208  *
209  * This structure is attached to packets as part of the
210  * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
211  */
212 struct skb_shared_hwtstamps {
213 	ktime_t	hwtstamp;
214 	ktime_t	syststamp;
215 };
216 
217 /* Definitions for tx_flags in struct skb_shared_info */
218 enum {
219 	/* generate hardware time stamp */
220 	SKBTX_HW_TSTAMP = 1 << 0,
221 
222 	/* generate software time stamp */
223 	SKBTX_SW_TSTAMP = 1 << 1,
224 
225 	/* device driver is going to provide hardware time stamp */
226 	SKBTX_IN_PROGRESS = 1 << 2,
227 
228 	/* device driver supports TX zero-copy buffers */
229 	SKBTX_DEV_ZEROCOPY = 1 << 3,
230 
231 	/* generate wifi status information (where possible) */
232 	SKBTX_WIFI_STATUS = 1 << 4,
233 
234 	/* This indicates at least one fragment might be overwritten
235 	 * (as in vmsplice(), sendfile() ...)
236 	 * If we need to compute a TX checksum, we'll need to copy
237 	 * all frags to avoid possible bad checksum
238 	 */
239 	SKBTX_SHARED_FRAG = 1 << 5,
240 };
241 
242 /*
243  * The callback notifies userspace to release buffers when skb DMA is done in
244  * lower device, the skb last reference should be 0 when calling this.
245  * The zerocopy_success argument is true if zero copy transmit occurred,
246  * false on data copy or out of memory error caused by data copy attempt.
247  * The ctx field is used to track device context.
248  * The desc field is used to track userspace buffer index.
249  */
250 struct ubuf_info {
251 	void (*callback)(struct ubuf_info *, bool zerocopy_success);
252 	void *ctx;
253 	unsigned long desc;
254 };
255 
256 /* This data is invariant across clones and lives at
257  * the end of the header data, ie. at skb->end.
258  */
259 struct skb_shared_info {
260 	unsigned char	nr_frags;
261 	__u8		tx_flags;
262 	unsigned short	gso_size;
263 	/* Warning: this field is not always filled in (UFO)! */
264 	unsigned short	gso_segs;
265 	unsigned short  gso_type;
266 	struct sk_buff	*frag_list;
267 	struct skb_shared_hwtstamps hwtstamps;
268 	__be32          ip6_frag_id;
269 
270 	/*
271 	 * Warning : all fields before dataref are cleared in __alloc_skb()
272 	 */
273 	atomic_t	dataref;
274 
275 	/* Intermediate layers must ensure that destructor_arg
276 	 * remains valid until skb destructor */
277 	void *		destructor_arg;
278 
279 	/* must be last field, see pskb_expand_head() */
280 	skb_frag_t	frags[MAX_SKB_FRAGS];
281 };
282 
283 /* We divide dataref into two halves.  The higher 16 bits hold references
284  * to the payload part of skb->data.  The lower 16 bits hold references to
285  * the entire skb->data.  A clone of a headerless skb holds the length of
286  * the header in skb->hdr_len.
287  *
288  * All users must obey the rule that the skb->data reference count must be
289  * greater than or equal to the payload reference count.
290  *
291  * Holding a reference to the payload part means that the user does not
292  * care about modifications to the header part of skb->data.
293  */
294 #define SKB_DATAREF_SHIFT 16
295 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
296 
297 
298 enum {
299 	SKB_FCLONE_UNAVAILABLE,
300 	SKB_FCLONE_ORIG,
301 	SKB_FCLONE_CLONE,
302 };
303 
304 enum {
305 	SKB_GSO_TCPV4 = 1 << 0,
306 	SKB_GSO_UDP = 1 << 1,
307 
308 	/* This indicates the skb is from an untrusted source. */
309 	SKB_GSO_DODGY = 1 << 2,
310 
311 	/* This indicates the tcp segment has CWR set. */
312 	SKB_GSO_TCP_ECN = 1 << 3,
313 
314 	SKB_GSO_TCPV6 = 1 << 4,
315 
316 	SKB_GSO_FCOE = 1 << 5,
317 
318 	SKB_GSO_GRE = 1 << 6,
319 
320 	SKB_GSO_UDP_TUNNEL = 1 << 7,
321 };
322 
323 #if BITS_PER_LONG > 32
324 #define NET_SKBUFF_DATA_USES_OFFSET 1
325 #endif
326 
327 #ifdef NET_SKBUFF_DATA_USES_OFFSET
328 typedef unsigned int sk_buff_data_t;
329 #else
330 typedef unsigned char *sk_buff_data_t;
331 #endif
332 
333 #if defined(CONFIG_NF_DEFRAG_IPV4) || defined(CONFIG_NF_DEFRAG_IPV4_MODULE) || \
334     defined(CONFIG_NF_DEFRAG_IPV6) || defined(CONFIG_NF_DEFRAG_IPV6_MODULE)
335 #define NET_SKBUFF_NF_DEFRAG_NEEDED 1
336 #endif
337 
338 /**
339  *	struct sk_buff - socket buffer
340  *	@next: Next buffer in list
341  *	@prev: Previous buffer in list
342  *	@tstamp: Time we arrived
343  *	@sk: Socket we are owned by
344  *	@dev: Device we arrived on/are leaving by
345  *	@cb: Control buffer. Free for use by every layer. Put private vars here
346  *	@_skb_refdst: destination entry (with norefcount bit)
347  *	@sp: the security path, used for xfrm
348  *	@len: Length of actual data
349  *	@data_len: Data length
350  *	@mac_len: Length of link layer header
351  *	@hdr_len: writable header length of cloned skb
352  *	@csum: Checksum (must include start/offset pair)
353  *	@csum_start: Offset from skb->head where checksumming should start
354  *	@csum_offset: Offset from csum_start where checksum should be stored
355  *	@priority: Packet queueing priority
356  *	@local_df: allow local fragmentation
357  *	@cloned: Head may be cloned (check refcnt to be sure)
358  *	@ip_summed: Driver fed us an IP checksum
359  *	@nohdr: Payload reference only, must not modify header
360  *	@nfctinfo: Relationship of this skb to the connection
361  *	@pkt_type: Packet class
362  *	@fclone: skbuff clone status
363  *	@ipvs_property: skbuff is owned by ipvs
364  *	@peeked: this packet has been seen already, so stats have been
365  *		done for it, don't do them again
366  *	@nf_trace: netfilter packet trace flag
367  *	@protocol: Packet protocol from driver
368  *	@destructor: Destruct function
369  *	@nfct: Associated connection, if any
370  *	@nfct_reasm: netfilter conntrack re-assembly pointer
371  *	@nf_bridge: Saved data about a bridged frame - see br_netfilter.c
372  *	@skb_iif: ifindex of device we arrived on
373  *	@tc_index: Traffic control index
374  *	@tc_verd: traffic control verdict
375  *	@rxhash: the packet hash computed on receive
376  *	@queue_mapping: Queue mapping for multiqueue devices
377  *	@ndisc_nodetype: router type (from link layer)
378  *	@ooo_okay: allow the mapping of a socket to a queue to be changed
379  *	@l4_rxhash: indicate rxhash is a canonical 4-tuple hash over transport
380  *		ports.
381  *	@wifi_acked_valid: wifi_acked was set
382  *	@wifi_acked: whether frame was acked on wifi or not
383  *	@no_fcs:  Request NIC to treat last 4 bytes as Ethernet FCS
384  *	@dma_cookie: a cookie to one of several possible DMA operations
385  *		done by skb DMA functions
386  *	@secmark: security marking
387  *	@mark: Generic packet mark
388  *	@dropcount: total number of sk_receive_queue overflows
389  *	@vlan_tci: vlan tag control information
390  *	@inner_transport_header: Inner transport layer header (encapsulation)
391  *	@inner_network_header: Network layer header (encapsulation)
392  *	@inner_mac_header: Link layer header (encapsulation)
393  *	@transport_header: Transport layer header
394  *	@network_header: Network layer header
395  *	@mac_header: Link layer header
396  *	@tail: Tail pointer
397  *	@end: End pointer
398  *	@head: Head of buffer
399  *	@data: Data head pointer
400  *	@truesize: Buffer size
401  *	@users: User count - see {datagram,tcp}.c
402  */
403 
404 struct sk_buff {
405 	/* These two members must be first. */
406 	struct sk_buff		*next;
407 	struct sk_buff		*prev;
408 
409 	ktime_t			tstamp;
410 
411 	struct sock		*sk;
412 	struct net_device	*dev;
413 
414 	/*
415 	 * This is the control buffer. It is free to use for every
416 	 * layer. Please put your private variables there. If you
417 	 * want to keep them across layers you have to do a skb_clone()
418 	 * first. This is owned by whoever has the skb queued ATM.
419 	 */
420 	char			cb[48] __aligned(8);
421 
422 	unsigned long		_skb_refdst;
423 #ifdef CONFIG_XFRM
424 	struct	sec_path	*sp;
425 #endif
426 	unsigned int		len,
427 				data_len;
428 	__u16			mac_len,
429 				hdr_len;
430 	union {
431 		__wsum		csum;
432 		struct {
433 			__u16	csum_start;
434 			__u16	csum_offset;
435 		};
436 	};
437 	__u32			priority;
438 	kmemcheck_bitfield_begin(flags1);
439 	__u8			local_df:1,
440 				cloned:1,
441 				ip_summed:2,
442 				nohdr:1,
443 				nfctinfo:3;
444 	__u8			pkt_type:3,
445 				fclone:2,
446 				ipvs_property:1,
447 				peeked:1,
448 				nf_trace:1;
449 	kmemcheck_bitfield_end(flags1);
450 	__be16			protocol;
451 
452 	void			(*destructor)(struct sk_buff *skb);
453 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
454 	struct nf_conntrack	*nfct;
455 #endif
456 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
457 	struct sk_buff		*nfct_reasm;
458 #endif
459 #ifdef CONFIG_BRIDGE_NETFILTER
460 	struct nf_bridge_info	*nf_bridge;
461 #endif
462 
463 	int			skb_iif;
464 
465 	__u32			rxhash;
466 
467 	__u16			vlan_tci;
468 
469 #ifdef CONFIG_NET_SCHED
470 	__u16			tc_index;	/* traffic control index */
471 #ifdef CONFIG_NET_CLS_ACT
472 	__u16			tc_verd;	/* traffic control verdict */
473 #endif
474 #endif
475 
476 	__u16			queue_mapping;
477 	kmemcheck_bitfield_begin(flags2);
478 #ifdef CONFIG_IPV6_NDISC_NODETYPE
479 	__u8			ndisc_nodetype:2;
480 #endif
481 	__u8			pfmemalloc:1;
482 	__u8			ooo_okay:1;
483 	__u8			l4_rxhash:1;
484 	__u8			wifi_acked_valid:1;
485 	__u8			wifi_acked:1;
486 	__u8			no_fcs:1;
487 	__u8			head_frag:1;
488 	/* Encapsulation protocol and NIC drivers should use
489 	 * this flag to indicate to each other if the skb contains
490 	 * encapsulated packet or not and maybe use the inner packet
491 	 * headers if needed
492 	 */
493 	__u8			encapsulation:1;
494 	/* 7/9 bit hole (depending on ndisc_nodetype presence) */
495 	kmemcheck_bitfield_end(flags2);
496 
497 #ifdef CONFIG_NET_DMA
498 	dma_cookie_t		dma_cookie;
499 #endif
500 #ifdef CONFIG_NETWORK_SECMARK
501 	__u32			secmark;
502 #endif
503 	union {
504 		__u32		mark;
505 		__u32		dropcount;
506 		__u32		avail_size;
507 	};
508 
509 	sk_buff_data_t		inner_transport_header;
510 	sk_buff_data_t		inner_network_header;
511 	sk_buff_data_t		inner_mac_header;
512 	sk_buff_data_t		transport_header;
513 	sk_buff_data_t		network_header;
514 	sk_buff_data_t		mac_header;
515 	/* These elements must be at the end, see alloc_skb() for details.  */
516 	sk_buff_data_t		tail;
517 	sk_buff_data_t		end;
518 	unsigned char		*head,
519 				*data;
520 	unsigned int		truesize;
521 	atomic_t		users;
522 };
523 
524 #ifdef __KERNEL__
525 /*
526  *	Handling routines are only of interest to the kernel
527  */
528 #include <linux/slab.h>
529 
530 
531 #define SKB_ALLOC_FCLONE	0x01
532 #define SKB_ALLOC_RX		0x02
533 
534 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
535 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
536 {
537 	return unlikely(skb->pfmemalloc);
538 }
539 
540 /*
541  * skb might have a dst pointer attached, refcounted or not.
542  * _skb_refdst low order bit is set if refcount was _not_ taken
543  */
544 #define SKB_DST_NOREF	1UL
545 #define SKB_DST_PTRMASK	~(SKB_DST_NOREF)
546 
547 /**
548  * skb_dst - returns skb dst_entry
549  * @skb: buffer
550  *
551  * Returns skb dst_entry, regardless of reference taken or not.
552  */
553 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
554 {
555 	/* If refdst was not refcounted, check we still are in a
556 	 * rcu_read_lock section
557 	 */
558 	WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
559 		!rcu_read_lock_held() &&
560 		!rcu_read_lock_bh_held());
561 	return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
562 }
563 
564 /**
565  * skb_dst_set - sets skb dst
566  * @skb: buffer
567  * @dst: dst entry
568  *
569  * Sets skb dst, assuming a reference was taken on dst and should
570  * be released by skb_dst_drop()
571  */
572 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
573 {
574 	skb->_skb_refdst = (unsigned long)dst;
575 }
576 
577 extern void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst);
578 
579 /**
580  * skb_dst_is_noref - Test if skb dst isn't refcounted
581  * @skb: buffer
582  */
583 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
584 {
585 	return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
586 }
587 
588 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
589 {
590 	return (struct rtable *)skb_dst(skb);
591 }
592 
593 extern void kfree_skb(struct sk_buff *skb);
594 extern void skb_tx_error(struct sk_buff *skb);
595 extern void consume_skb(struct sk_buff *skb);
596 extern void	       __kfree_skb(struct sk_buff *skb);
597 extern struct kmem_cache *skbuff_head_cache;
598 
599 extern void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
600 extern bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
601 			     bool *fragstolen, int *delta_truesize);
602 
603 extern struct sk_buff *__alloc_skb(unsigned int size,
604 				   gfp_t priority, int flags, int node);
605 extern struct sk_buff *build_skb(void *data, unsigned int frag_size);
606 static inline struct sk_buff *alloc_skb(unsigned int size,
607 					gfp_t priority)
608 {
609 	return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
610 }
611 
612 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
613 					       gfp_t priority)
614 {
615 	return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
616 }
617 
618 extern struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
619 extern int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
620 extern struct sk_buff *skb_clone(struct sk_buff *skb,
621 				 gfp_t priority);
622 extern struct sk_buff *skb_copy(const struct sk_buff *skb,
623 				gfp_t priority);
624 extern struct sk_buff *__pskb_copy(struct sk_buff *skb,
625 				 int headroom, gfp_t gfp_mask);
626 
627 extern int	       pskb_expand_head(struct sk_buff *skb,
628 					int nhead, int ntail,
629 					gfp_t gfp_mask);
630 extern struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
631 					    unsigned int headroom);
632 extern struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
633 				       int newheadroom, int newtailroom,
634 				       gfp_t priority);
635 extern int	       skb_to_sgvec(struct sk_buff *skb,
636 				    struct scatterlist *sg, int offset,
637 				    int len);
638 extern int	       skb_cow_data(struct sk_buff *skb, int tailbits,
639 				    struct sk_buff **trailer);
640 extern int	       skb_pad(struct sk_buff *skb, int pad);
641 #define dev_kfree_skb(a)	consume_skb(a)
642 
643 extern int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
644 			int getfrag(void *from, char *to, int offset,
645 			int len,int odd, struct sk_buff *skb),
646 			void *from, int length);
647 
648 struct skb_seq_state {
649 	__u32		lower_offset;
650 	__u32		upper_offset;
651 	__u32		frag_idx;
652 	__u32		stepped_offset;
653 	struct sk_buff	*root_skb;
654 	struct sk_buff	*cur_skb;
655 	__u8		*frag_data;
656 };
657 
658 extern void	      skb_prepare_seq_read(struct sk_buff *skb,
659 					   unsigned int from, unsigned int to,
660 					   struct skb_seq_state *st);
661 extern unsigned int   skb_seq_read(unsigned int consumed, const u8 **data,
662 				   struct skb_seq_state *st);
663 extern void	      skb_abort_seq_read(struct skb_seq_state *st);
664 
665 extern unsigned int   skb_find_text(struct sk_buff *skb, unsigned int from,
666 				    unsigned int to, struct ts_config *config,
667 				    struct ts_state *state);
668 
669 extern void __skb_get_rxhash(struct sk_buff *skb);
670 static inline __u32 skb_get_rxhash(struct sk_buff *skb)
671 {
672 	if (!skb->l4_rxhash)
673 		__skb_get_rxhash(skb);
674 
675 	return skb->rxhash;
676 }
677 
678 #ifdef NET_SKBUFF_DATA_USES_OFFSET
679 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
680 {
681 	return skb->head + skb->end;
682 }
683 
684 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
685 {
686 	return skb->end;
687 }
688 #else
689 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
690 {
691 	return skb->end;
692 }
693 
694 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
695 {
696 	return skb->end - skb->head;
697 }
698 #endif
699 
700 /* Internal */
701 #define skb_shinfo(SKB)	((struct skb_shared_info *)(skb_end_pointer(SKB)))
702 
703 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
704 {
705 	return &skb_shinfo(skb)->hwtstamps;
706 }
707 
708 /**
709  *	skb_queue_empty - check if a queue is empty
710  *	@list: queue head
711  *
712  *	Returns true if the queue is empty, false otherwise.
713  */
714 static inline int skb_queue_empty(const struct sk_buff_head *list)
715 {
716 	return list->next == (struct sk_buff *)list;
717 }
718 
719 /**
720  *	skb_queue_is_last - check if skb is the last entry in the queue
721  *	@list: queue head
722  *	@skb: buffer
723  *
724  *	Returns true if @skb is the last buffer on the list.
725  */
726 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
727 				     const struct sk_buff *skb)
728 {
729 	return skb->next == (struct sk_buff *)list;
730 }
731 
732 /**
733  *	skb_queue_is_first - check if skb is the first entry in the queue
734  *	@list: queue head
735  *	@skb: buffer
736  *
737  *	Returns true if @skb is the first buffer on the list.
738  */
739 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
740 				      const struct sk_buff *skb)
741 {
742 	return skb->prev == (struct sk_buff *)list;
743 }
744 
745 /**
746  *	skb_queue_next - return the next packet in the queue
747  *	@list: queue head
748  *	@skb: current buffer
749  *
750  *	Return the next packet in @list after @skb.  It is only valid to
751  *	call this if skb_queue_is_last() evaluates to false.
752  */
753 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
754 					     const struct sk_buff *skb)
755 {
756 	/* This BUG_ON may seem severe, but if we just return then we
757 	 * are going to dereference garbage.
758 	 */
759 	BUG_ON(skb_queue_is_last(list, skb));
760 	return skb->next;
761 }
762 
763 /**
764  *	skb_queue_prev - return the prev packet in the queue
765  *	@list: queue head
766  *	@skb: current buffer
767  *
768  *	Return the prev packet in @list before @skb.  It is only valid to
769  *	call this if skb_queue_is_first() evaluates to false.
770  */
771 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
772 					     const struct sk_buff *skb)
773 {
774 	/* This BUG_ON may seem severe, but if we just return then we
775 	 * are going to dereference garbage.
776 	 */
777 	BUG_ON(skb_queue_is_first(list, skb));
778 	return skb->prev;
779 }
780 
781 /**
782  *	skb_get - reference buffer
783  *	@skb: buffer to reference
784  *
785  *	Makes another reference to a socket buffer and returns a pointer
786  *	to the buffer.
787  */
788 static inline struct sk_buff *skb_get(struct sk_buff *skb)
789 {
790 	atomic_inc(&skb->users);
791 	return skb;
792 }
793 
794 /*
795  * If users == 1, we are the only owner and are can avoid redundant
796  * atomic change.
797  */
798 
799 /**
800  *	skb_cloned - is the buffer a clone
801  *	@skb: buffer to check
802  *
803  *	Returns true if the buffer was generated with skb_clone() and is
804  *	one of multiple shared copies of the buffer. Cloned buffers are
805  *	shared data so must not be written to under normal circumstances.
806  */
807 static inline int skb_cloned(const struct sk_buff *skb)
808 {
809 	return skb->cloned &&
810 	       (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
811 }
812 
813 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
814 {
815 	might_sleep_if(pri & __GFP_WAIT);
816 
817 	if (skb_cloned(skb))
818 		return pskb_expand_head(skb, 0, 0, pri);
819 
820 	return 0;
821 }
822 
823 /**
824  *	skb_header_cloned - is the header a clone
825  *	@skb: buffer to check
826  *
827  *	Returns true if modifying the header part of the buffer requires
828  *	the data to be copied.
829  */
830 static inline int skb_header_cloned(const struct sk_buff *skb)
831 {
832 	int dataref;
833 
834 	if (!skb->cloned)
835 		return 0;
836 
837 	dataref = atomic_read(&skb_shinfo(skb)->dataref);
838 	dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
839 	return dataref != 1;
840 }
841 
842 /**
843  *	skb_header_release - release reference to header
844  *	@skb: buffer to operate on
845  *
846  *	Drop a reference to the header part of the buffer.  This is done
847  *	by acquiring a payload reference.  You must not read from the header
848  *	part of skb->data after this.
849  */
850 static inline void skb_header_release(struct sk_buff *skb)
851 {
852 	BUG_ON(skb->nohdr);
853 	skb->nohdr = 1;
854 	atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
855 }
856 
857 /**
858  *	skb_shared - is the buffer shared
859  *	@skb: buffer to check
860  *
861  *	Returns true if more than one person has a reference to this
862  *	buffer.
863  */
864 static inline int skb_shared(const struct sk_buff *skb)
865 {
866 	return atomic_read(&skb->users) != 1;
867 }
868 
869 /**
870  *	skb_share_check - check if buffer is shared and if so clone it
871  *	@skb: buffer to check
872  *	@pri: priority for memory allocation
873  *
874  *	If the buffer is shared the buffer is cloned and the old copy
875  *	drops a reference. A new clone with a single reference is returned.
876  *	If the buffer is not shared the original buffer is returned. When
877  *	being called from interrupt status or with spinlocks held pri must
878  *	be GFP_ATOMIC.
879  *
880  *	NULL is returned on a memory allocation failure.
881  */
882 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
883 {
884 	might_sleep_if(pri & __GFP_WAIT);
885 	if (skb_shared(skb)) {
886 		struct sk_buff *nskb = skb_clone(skb, pri);
887 
888 		if (likely(nskb))
889 			consume_skb(skb);
890 		else
891 			kfree_skb(skb);
892 		skb = nskb;
893 	}
894 	return skb;
895 }
896 
897 /*
898  *	Copy shared buffers into a new sk_buff. We effectively do COW on
899  *	packets to handle cases where we have a local reader and forward
900  *	and a couple of other messy ones. The normal one is tcpdumping
901  *	a packet thats being forwarded.
902  */
903 
904 /**
905  *	skb_unshare - make a copy of a shared buffer
906  *	@skb: buffer to check
907  *	@pri: priority for memory allocation
908  *
909  *	If the socket buffer is a clone then this function creates a new
910  *	copy of the data, drops a reference count on the old copy and returns
911  *	the new copy with the reference count at 1. If the buffer is not a clone
912  *	the original buffer is returned. When called with a spinlock held or
913  *	from interrupt state @pri must be %GFP_ATOMIC
914  *
915  *	%NULL is returned on a memory allocation failure.
916  */
917 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
918 					  gfp_t pri)
919 {
920 	might_sleep_if(pri & __GFP_WAIT);
921 	if (skb_cloned(skb)) {
922 		struct sk_buff *nskb = skb_copy(skb, pri);
923 		kfree_skb(skb);	/* Free our shared copy */
924 		skb = nskb;
925 	}
926 	return skb;
927 }
928 
929 /**
930  *	skb_peek - peek at the head of an &sk_buff_head
931  *	@list_: list to peek at
932  *
933  *	Peek an &sk_buff. Unlike most other operations you _MUST_
934  *	be careful with this one. A peek leaves the buffer on the
935  *	list and someone else may run off with it. You must hold
936  *	the appropriate locks or have a private queue to do this.
937  *
938  *	Returns %NULL for an empty list or a pointer to the head element.
939  *	The reference count is not incremented and the reference is therefore
940  *	volatile. Use with caution.
941  */
942 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
943 {
944 	struct sk_buff *skb = list_->next;
945 
946 	if (skb == (struct sk_buff *)list_)
947 		skb = NULL;
948 	return skb;
949 }
950 
951 /**
952  *	skb_peek_next - peek skb following the given one from a queue
953  *	@skb: skb to start from
954  *	@list_: list to peek at
955  *
956  *	Returns %NULL when the end of the list is met or a pointer to the
957  *	next element. The reference count is not incremented and the
958  *	reference is therefore volatile. Use with caution.
959  */
960 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
961 		const struct sk_buff_head *list_)
962 {
963 	struct sk_buff *next = skb->next;
964 
965 	if (next == (struct sk_buff *)list_)
966 		next = NULL;
967 	return next;
968 }
969 
970 /**
971  *	skb_peek_tail - peek at the tail of an &sk_buff_head
972  *	@list_: list to peek at
973  *
974  *	Peek an &sk_buff. Unlike most other operations you _MUST_
975  *	be careful with this one. A peek leaves the buffer on the
976  *	list and someone else may run off with it. You must hold
977  *	the appropriate locks or have a private queue to do this.
978  *
979  *	Returns %NULL for an empty list or a pointer to the tail element.
980  *	The reference count is not incremented and the reference is therefore
981  *	volatile. Use with caution.
982  */
983 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
984 {
985 	struct sk_buff *skb = list_->prev;
986 
987 	if (skb == (struct sk_buff *)list_)
988 		skb = NULL;
989 	return skb;
990 
991 }
992 
993 /**
994  *	skb_queue_len	- get queue length
995  *	@list_: list to measure
996  *
997  *	Return the length of an &sk_buff queue.
998  */
999 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1000 {
1001 	return list_->qlen;
1002 }
1003 
1004 /**
1005  *	__skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1006  *	@list: queue to initialize
1007  *
1008  *	This initializes only the list and queue length aspects of
1009  *	an sk_buff_head object.  This allows to initialize the list
1010  *	aspects of an sk_buff_head without reinitializing things like
1011  *	the spinlock.  It can also be used for on-stack sk_buff_head
1012  *	objects where the spinlock is known to not be used.
1013  */
1014 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1015 {
1016 	list->prev = list->next = (struct sk_buff *)list;
1017 	list->qlen = 0;
1018 }
1019 
1020 /*
1021  * This function creates a split out lock class for each invocation;
1022  * this is needed for now since a whole lot of users of the skb-queue
1023  * infrastructure in drivers have different locking usage (in hardirq)
1024  * than the networking core (in softirq only). In the long run either the
1025  * network layer or drivers should need annotation to consolidate the
1026  * main types of usage into 3 classes.
1027  */
1028 static inline void skb_queue_head_init(struct sk_buff_head *list)
1029 {
1030 	spin_lock_init(&list->lock);
1031 	__skb_queue_head_init(list);
1032 }
1033 
1034 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1035 		struct lock_class_key *class)
1036 {
1037 	skb_queue_head_init(list);
1038 	lockdep_set_class(&list->lock, class);
1039 }
1040 
1041 /*
1042  *	Insert an sk_buff on a list.
1043  *
1044  *	The "__skb_xxxx()" functions are the non-atomic ones that
1045  *	can only be called with interrupts disabled.
1046  */
1047 extern void        skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list);
1048 static inline void __skb_insert(struct sk_buff *newsk,
1049 				struct sk_buff *prev, struct sk_buff *next,
1050 				struct sk_buff_head *list)
1051 {
1052 	newsk->next = next;
1053 	newsk->prev = prev;
1054 	next->prev  = prev->next = newsk;
1055 	list->qlen++;
1056 }
1057 
1058 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1059 				      struct sk_buff *prev,
1060 				      struct sk_buff *next)
1061 {
1062 	struct sk_buff *first = list->next;
1063 	struct sk_buff *last = list->prev;
1064 
1065 	first->prev = prev;
1066 	prev->next = first;
1067 
1068 	last->next = next;
1069 	next->prev = last;
1070 }
1071 
1072 /**
1073  *	skb_queue_splice - join two skb lists, this is designed for stacks
1074  *	@list: the new list to add
1075  *	@head: the place to add it in the first list
1076  */
1077 static inline void skb_queue_splice(const struct sk_buff_head *list,
1078 				    struct sk_buff_head *head)
1079 {
1080 	if (!skb_queue_empty(list)) {
1081 		__skb_queue_splice(list, (struct sk_buff *) head, head->next);
1082 		head->qlen += list->qlen;
1083 	}
1084 }
1085 
1086 /**
1087  *	skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1088  *	@list: the new list to add
1089  *	@head: the place to add it in the first list
1090  *
1091  *	The list at @list is reinitialised
1092  */
1093 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1094 					 struct sk_buff_head *head)
1095 {
1096 	if (!skb_queue_empty(list)) {
1097 		__skb_queue_splice(list, (struct sk_buff *) head, head->next);
1098 		head->qlen += list->qlen;
1099 		__skb_queue_head_init(list);
1100 	}
1101 }
1102 
1103 /**
1104  *	skb_queue_splice_tail - join two skb lists, each list being a queue
1105  *	@list: the new list to add
1106  *	@head: the place to add it in the first list
1107  */
1108 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1109 					 struct sk_buff_head *head)
1110 {
1111 	if (!skb_queue_empty(list)) {
1112 		__skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1113 		head->qlen += list->qlen;
1114 	}
1115 }
1116 
1117 /**
1118  *	skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1119  *	@list: the new list to add
1120  *	@head: the place to add it in the first list
1121  *
1122  *	Each of the lists is a queue.
1123  *	The list at @list is reinitialised
1124  */
1125 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1126 					      struct sk_buff_head *head)
1127 {
1128 	if (!skb_queue_empty(list)) {
1129 		__skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1130 		head->qlen += list->qlen;
1131 		__skb_queue_head_init(list);
1132 	}
1133 }
1134 
1135 /**
1136  *	__skb_queue_after - queue a buffer at the list head
1137  *	@list: list to use
1138  *	@prev: place after this buffer
1139  *	@newsk: buffer to queue
1140  *
1141  *	Queue a buffer int the middle of a list. This function takes no locks
1142  *	and you must therefore hold required locks before calling it.
1143  *
1144  *	A buffer cannot be placed on two lists at the same time.
1145  */
1146 static inline void __skb_queue_after(struct sk_buff_head *list,
1147 				     struct sk_buff *prev,
1148 				     struct sk_buff *newsk)
1149 {
1150 	__skb_insert(newsk, prev, prev->next, list);
1151 }
1152 
1153 extern void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1154 		       struct sk_buff_head *list);
1155 
1156 static inline void __skb_queue_before(struct sk_buff_head *list,
1157 				      struct sk_buff *next,
1158 				      struct sk_buff *newsk)
1159 {
1160 	__skb_insert(newsk, next->prev, next, list);
1161 }
1162 
1163 /**
1164  *	__skb_queue_head - queue a buffer at the list head
1165  *	@list: list to use
1166  *	@newsk: buffer to queue
1167  *
1168  *	Queue a buffer at the start of a list. This function takes no locks
1169  *	and you must therefore hold required locks before calling it.
1170  *
1171  *	A buffer cannot be placed on two lists at the same time.
1172  */
1173 extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1174 static inline void __skb_queue_head(struct sk_buff_head *list,
1175 				    struct sk_buff *newsk)
1176 {
1177 	__skb_queue_after(list, (struct sk_buff *)list, newsk);
1178 }
1179 
1180 /**
1181  *	__skb_queue_tail - queue a buffer at the list tail
1182  *	@list: list to use
1183  *	@newsk: buffer to queue
1184  *
1185  *	Queue a buffer at the end of a list. This function takes no locks
1186  *	and you must therefore hold required locks before calling it.
1187  *
1188  *	A buffer cannot be placed on two lists at the same time.
1189  */
1190 extern void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1191 static inline void __skb_queue_tail(struct sk_buff_head *list,
1192 				   struct sk_buff *newsk)
1193 {
1194 	__skb_queue_before(list, (struct sk_buff *)list, newsk);
1195 }
1196 
1197 /*
1198  * remove sk_buff from list. _Must_ be called atomically, and with
1199  * the list known..
1200  */
1201 extern void	   skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1202 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1203 {
1204 	struct sk_buff *next, *prev;
1205 
1206 	list->qlen--;
1207 	next	   = skb->next;
1208 	prev	   = skb->prev;
1209 	skb->next  = skb->prev = NULL;
1210 	next->prev = prev;
1211 	prev->next = next;
1212 }
1213 
1214 /**
1215  *	__skb_dequeue - remove from the head of the queue
1216  *	@list: list to dequeue from
1217  *
1218  *	Remove the head of the list. This function does not take any locks
1219  *	so must be used with appropriate locks held only. The head item is
1220  *	returned or %NULL if the list is empty.
1221  */
1222 extern struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1223 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1224 {
1225 	struct sk_buff *skb = skb_peek(list);
1226 	if (skb)
1227 		__skb_unlink(skb, list);
1228 	return skb;
1229 }
1230 
1231 /**
1232  *	__skb_dequeue_tail - remove from the tail of the queue
1233  *	@list: list to dequeue from
1234  *
1235  *	Remove the tail of the list. This function does not take any locks
1236  *	so must be used with appropriate locks held only. The tail item is
1237  *	returned or %NULL if the list is empty.
1238  */
1239 extern struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1240 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1241 {
1242 	struct sk_buff *skb = skb_peek_tail(list);
1243 	if (skb)
1244 		__skb_unlink(skb, list);
1245 	return skb;
1246 }
1247 
1248 
1249 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1250 {
1251 	return skb->data_len;
1252 }
1253 
1254 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1255 {
1256 	return skb->len - skb->data_len;
1257 }
1258 
1259 static inline int skb_pagelen(const struct sk_buff *skb)
1260 {
1261 	int i, len = 0;
1262 
1263 	for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
1264 		len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1265 	return len + skb_headlen(skb);
1266 }
1267 
1268 /**
1269  * __skb_fill_page_desc - initialise a paged fragment in an skb
1270  * @skb: buffer containing fragment to be initialised
1271  * @i: paged fragment index to initialise
1272  * @page: the page to use for this fragment
1273  * @off: the offset to the data with @page
1274  * @size: the length of the data
1275  *
1276  * Initialises the @i'th fragment of @skb to point to &size bytes at
1277  * offset @off within @page.
1278  *
1279  * Does not take any additional reference on the fragment.
1280  */
1281 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1282 					struct page *page, int off, int size)
1283 {
1284 	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1285 
1286 	/*
1287 	 * Propagate page->pfmemalloc to the skb if we can. The problem is
1288 	 * that not all callers have unique ownership of the page. If
1289 	 * pfmemalloc is set, we check the mapping as a mapping implies
1290 	 * page->index is set (index and pfmemalloc share space).
1291 	 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1292 	 * do not lose pfmemalloc information as the pages would not be
1293 	 * allocated using __GFP_MEMALLOC.
1294 	 */
1295 	if (page->pfmemalloc && !page->mapping)
1296 		skb->pfmemalloc	= true;
1297 	frag->page.p		  = page;
1298 	frag->page_offset	  = off;
1299 	skb_frag_size_set(frag, size);
1300 }
1301 
1302 /**
1303  * skb_fill_page_desc - initialise a paged fragment in an skb
1304  * @skb: buffer containing fragment to be initialised
1305  * @i: paged fragment index to initialise
1306  * @page: the page to use for this fragment
1307  * @off: the offset to the data with @page
1308  * @size: the length of the data
1309  *
1310  * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1311  * @skb to point to &size bytes at offset @off within @page. In
1312  * addition updates @skb such that @i is the last fragment.
1313  *
1314  * Does not take any additional reference on the fragment.
1315  */
1316 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1317 				      struct page *page, int off, int size)
1318 {
1319 	__skb_fill_page_desc(skb, i, page, off, size);
1320 	skb_shinfo(skb)->nr_frags = i + 1;
1321 }
1322 
1323 extern void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page,
1324 			    int off, int size, unsigned int truesize);
1325 
1326 #define SKB_PAGE_ASSERT(skb) 	BUG_ON(skb_shinfo(skb)->nr_frags)
1327 #define SKB_FRAG_ASSERT(skb) 	BUG_ON(skb_has_frag_list(skb))
1328 #define SKB_LINEAR_ASSERT(skb)  BUG_ON(skb_is_nonlinear(skb))
1329 
1330 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1331 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1332 {
1333 	return skb->head + skb->tail;
1334 }
1335 
1336 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1337 {
1338 	skb->tail = skb->data - skb->head;
1339 }
1340 
1341 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1342 {
1343 	skb_reset_tail_pointer(skb);
1344 	skb->tail += offset;
1345 }
1346 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1347 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1348 {
1349 	return skb->tail;
1350 }
1351 
1352 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1353 {
1354 	skb->tail = skb->data;
1355 }
1356 
1357 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1358 {
1359 	skb->tail = skb->data + offset;
1360 }
1361 
1362 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1363 
1364 /*
1365  *	Add data to an sk_buff
1366  */
1367 extern unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1368 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1369 {
1370 	unsigned char *tmp = skb_tail_pointer(skb);
1371 	SKB_LINEAR_ASSERT(skb);
1372 	skb->tail += len;
1373 	skb->len  += len;
1374 	return tmp;
1375 }
1376 
1377 extern unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1378 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1379 {
1380 	skb->data -= len;
1381 	skb->len  += len;
1382 	return skb->data;
1383 }
1384 
1385 extern unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1386 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1387 {
1388 	skb->len -= len;
1389 	BUG_ON(skb->len < skb->data_len);
1390 	return skb->data += len;
1391 }
1392 
1393 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1394 {
1395 	return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1396 }
1397 
1398 extern unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1399 
1400 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1401 {
1402 	if (len > skb_headlen(skb) &&
1403 	    !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1404 		return NULL;
1405 	skb->len -= len;
1406 	return skb->data += len;
1407 }
1408 
1409 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1410 {
1411 	return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1412 }
1413 
1414 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1415 {
1416 	if (likely(len <= skb_headlen(skb)))
1417 		return 1;
1418 	if (unlikely(len > skb->len))
1419 		return 0;
1420 	return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1421 }
1422 
1423 /**
1424  *	skb_headroom - bytes at buffer head
1425  *	@skb: buffer to check
1426  *
1427  *	Return the number of bytes of free space at the head of an &sk_buff.
1428  */
1429 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1430 {
1431 	return skb->data - skb->head;
1432 }
1433 
1434 /**
1435  *	skb_tailroom - bytes at buffer end
1436  *	@skb: buffer to check
1437  *
1438  *	Return the number of bytes of free space at the tail of an sk_buff
1439  */
1440 static inline int skb_tailroom(const struct sk_buff *skb)
1441 {
1442 	return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1443 }
1444 
1445 /**
1446  *	skb_availroom - bytes at buffer end
1447  *	@skb: buffer to check
1448  *
1449  *	Return the number of bytes of free space at the tail of an sk_buff
1450  *	allocated by sk_stream_alloc()
1451  */
1452 static inline int skb_availroom(const struct sk_buff *skb)
1453 {
1454 	return skb_is_nonlinear(skb) ? 0 : skb->avail_size - skb->len;
1455 }
1456 
1457 /**
1458  *	skb_reserve - adjust headroom
1459  *	@skb: buffer to alter
1460  *	@len: bytes to move
1461  *
1462  *	Increase the headroom of an empty &sk_buff by reducing the tail
1463  *	room. This is only allowed for an empty buffer.
1464  */
1465 static inline void skb_reserve(struct sk_buff *skb, int len)
1466 {
1467 	skb->data += len;
1468 	skb->tail += len;
1469 }
1470 
1471 static inline void skb_reset_inner_headers(struct sk_buff *skb)
1472 {
1473 	skb->inner_mac_header = skb->mac_header;
1474 	skb->inner_network_header = skb->network_header;
1475 	skb->inner_transport_header = skb->transport_header;
1476 }
1477 
1478 static inline void skb_reset_mac_len(struct sk_buff *skb)
1479 {
1480 	skb->mac_len = skb->network_header - skb->mac_header;
1481 }
1482 
1483 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1484 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1485 							*skb)
1486 {
1487 	return skb->head + skb->inner_transport_header;
1488 }
1489 
1490 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1491 {
1492 	skb->inner_transport_header = skb->data - skb->head;
1493 }
1494 
1495 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1496 						   const int offset)
1497 {
1498 	skb_reset_inner_transport_header(skb);
1499 	skb->inner_transport_header += offset;
1500 }
1501 
1502 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1503 {
1504 	return skb->head + skb->inner_network_header;
1505 }
1506 
1507 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1508 {
1509 	skb->inner_network_header = skb->data - skb->head;
1510 }
1511 
1512 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1513 						const int offset)
1514 {
1515 	skb_reset_inner_network_header(skb);
1516 	skb->inner_network_header += offset;
1517 }
1518 
1519 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1520 {
1521 	return skb->head + skb->inner_mac_header;
1522 }
1523 
1524 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1525 {
1526 	skb->inner_mac_header = skb->data - skb->head;
1527 }
1528 
1529 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1530 					    const int offset)
1531 {
1532 	skb_reset_inner_mac_header(skb);
1533 	skb->inner_mac_header += offset;
1534 }
1535 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1536 {
1537 	return skb->transport_header != ~0U;
1538 }
1539 
1540 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1541 {
1542 	return skb->head + skb->transport_header;
1543 }
1544 
1545 static inline void skb_reset_transport_header(struct sk_buff *skb)
1546 {
1547 	skb->transport_header = skb->data - skb->head;
1548 }
1549 
1550 static inline void skb_set_transport_header(struct sk_buff *skb,
1551 					    const int offset)
1552 {
1553 	skb_reset_transport_header(skb);
1554 	skb->transport_header += offset;
1555 }
1556 
1557 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1558 {
1559 	return skb->head + skb->network_header;
1560 }
1561 
1562 static inline void skb_reset_network_header(struct sk_buff *skb)
1563 {
1564 	skb->network_header = skb->data - skb->head;
1565 }
1566 
1567 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1568 {
1569 	skb_reset_network_header(skb);
1570 	skb->network_header += offset;
1571 }
1572 
1573 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1574 {
1575 	return skb->head + skb->mac_header;
1576 }
1577 
1578 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1579 {
1580 	return skb->mac_header != ~0U;
1581 }
1582 
1583 static inline void skb_reset_mac_header(struct sk_buff *skb)
1584 {
1585 	skb->mac_header = skb->data - skb->head;
1586 }
1587 
1588 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1589 {
1590 	skb_reset_mac_header(skb);
1591 	skb->mac_header += offset;
1592 }
1593 
1594 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1595 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1596 							*skb)
1597 {
1598 	return skb->inner_transport_header;
1599 }
1600 
1601 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1602 {
1603 	skb->inner_transport_header = skb->data;
1604 }
1605 
1606 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1607 						   const int offset)
1608 {
1609 	skb->inner_transport_header = skb->data + offset;
1610 }
1611 
1612 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1613 {
1614 	return skb->inner_network_header;
1615 }
1616 
1617 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1618 {
1619 	skb->inner_network_header = skb->data;
1620 }
1621 
1622 static inline void skb_set_inner_network_header(struct sk_buff *skb,
1623 						const int offset)
1624 {
1625 	skb->inner_network_header = skb->data + offset;
1626 }
1627 
1628 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1629 {
1630 	return skb->inner_mac_header;
1631 }
1632 
1633 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1634 {
1635 	skb->inner_mac_header = skb->data;
1636 }
1637 
1638 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1639 						const int offset)
1640 {
1641 	skb->inner_mac_header = skb->data + offset;
1642 }
1643 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1644 {
1645 	return skb->transport_header != NULL;
1646 }
1647 
1648 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1649 {
1650 	return skb->transport_header;
1651 }
1652 
1653 static inline void skb_reset_transport_header(struct sk_buff *skb)
1654 {
1655 	skb->transport_header = skb->data;
1656 }
1657 
1658 static inline void skb_set_transport_header(struct sk_buff *skb,
1659 					    const int offset)
1660 {
1661 	skb->transport_header = skb->data + offset;
1662 }
1663 
1664 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1665 {
1666 	return skb->network_header;
1667 }
1668 
1669 static inline void skb_reset_network_header(struct sk_buff *skb)
1670 {
1671 	skb->network_header = skb->data;
1672 }
1673 
1674 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1675 {
1676 	skb->network_header = skb->data + offset;
1677 }
1678 
1679 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1680 {
1681 	return skb->mac_header;
1682 }
1683 
1684 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1685 {
1686 	return skb->mac_header != NULL;
1687 }
1688 
1689 static inline void skb_reset_mac_header(struct sk_buff *skb)
1690 {
1691 	skb->mac_header = skb->data;
1692 }
1693 
1694 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1695 {
1696 	skb->mac_header = skb->data + offset;
1697 }
1698 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1699 
1700 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1701 {
1702 	if (skb_mac_header_was_set(skb)) {
1703 		const unsigned char *old_mac = skb_mac_header(skb);
1704 
1705 		skb_set_mac_header(skb, -skb->mac_len);
1706 		memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1707 	}
1708 }
1709 
1710 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1711 {
1712 	return skb->csum_start - skb_headroom(skb);
1713 }
1714 
1715 static inline int skb_transport_offset(const struct sk_buff *skb)
1716 {
1717 	return skb_transport_header(skb) - skb->data;
1718 }
1719 
1720 static inline u32 skb_network_header_len(const struct sk_buff *skb)
1721 {
1722 	return skb->transport_header - skb->network_header;
1723 }
1724 
1725 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1726 {
1727 	return skb->inner_transport_header - skb->inner_network_header;
1728 }
1729 
1730 static inline int skb_network_offset(const struct sk_buff *skb)
1731 {
1732 	return skb_network_header(skb) - skb->data;
1733 }
1734 
1735 static inline int skb_inner_network_offset(const struct sk_buff *skb)
1736 {
1737 	return skb_inner_network_header(skb) - skb->data;
1738 }
1739 
1740 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1741 {
1742 	return pskb_may_pull(skb, skb_network_offset(skb) + len);
1743 }
1744 
1745 /*
1746  * CPUs often take a performance hit when accessing unaligned memory
1747  * locations. The actual performance hit varies, it can be small if the
1748  * hardware handles it or large if we have to take an exception and fix it
1749  * in software.
1750  *
1751  * Since an ethernet header is 14 bytes network drivers often end up with
1752  * the IP header at an unaligned offset. The IP header can be aligned by
1753  * shifting the start of the packet by 2 bytes. Drivers should do this
1754  * with:
1755  *
1756  * skb_reserve(skb, NET_IP_ALIGN);
1757  *
1758  * The downside to this alignment of the IP header is that the DMA is now
1759  * unaligned. On some architectures the cost of an unaligned DMA is high
1760  * and this cost outweighs the gains made by aligning the IP header.
1761  *
1762  * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1763  * to be overridden.
1764  */
1765 #ifndef NET_IP_ALIGN
1766 #define NET_IP_ALIGN	2
1767 #endif
1768 
1769 /*
1770  * The networking layer reserves some headroom in skb data (via
1771  * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1772  * the header has to grow. In the default case, if the header has to grow
1773  * 32 bytes or less we avoid the reallocation.
1774  *
1775  * Unfortunately this headroom changes the DMA alignment of the resulting
1776  * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1777  * on some architectures. An architecture can override this value,
1778  * perhaps setting it to a cacheline in size (since that will maintain
1779  * cacheline alignment of the DMA). It must be a power of 2.
1780  *
1781  * Various parts of the networking layer expect at least 32 bytes of
1782  * headroom, you should not reduce this.
1783  *
1784  * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
1785  * to reduce average number of cache lines per packet.
1786  * get_rps_cpus() for example only access one 64 bytes aligned block :
1787  * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
1788  */
1789 #ifndef NET_SKB_PAD
1790 #define NET_SKB_PAD	max(32, L1_CACHE_BYTES)
1791 #endif
1792 
1793 extern int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1794 
1795 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1796 {
1797 	if (unlikely(skb_is_nonlinear(skb))) {
1798 		WARN_ON(1);
1799 		return;
1800 	}
1801 	skb->len = len;
1802 	skb_set_tail_pointer(skb, len);
1803 }
1804 
1805 extern void skb_trim(struct sk_buff *skb, unsigned int len);
1806 
1807 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1808 {
1809 	if (skb->data_len)
1810 		return ___pskb_trim(skb, len);
1811 	__skb_trim(skb, len);
1812 	return 0;
1813 }
1814 
1815 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1816 {
1817 	return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1818 }
1819 
1820 /**
1821  *	pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1822  *	@skb: buffer to alter
1823  *	@len: new length
1824  *
1825  *	This is identical to pskb_trim except that the caller knows that
1826  *	the skb is not cloned so we should never get an error due to out-
1827  *	of-memory.
1828  */
1829 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1830 {
1831 	int err = pskb_trim(skb, len);
1832 	BUG_ON(err);
1833 }
1834 
1835 /**
1836  *	skb_orphan - orphan a buffer
1837  *	@skb: buffer to orphan
1838  *
1839  *	If a buffer currently has an owner then we call the owner's
1840  *	destructor function and make the @skb unowned. The buffer continues
1841  *	to exist but is no longer charged to its former owner.
1842  */
1843 static inline void skb_orphan(struct sk_buff *skb)
1844 {
1845 	if (skb->destructor)
1846 		skb->destructor(skb);
1847 	skb->destructor = NULL;
1848 	skb->sk		= NULL;
1849 }
1850 
1851 /**
1852  *	skb_orphan_frags - orphan the frags contained in a buffer
1853  *	@skb: buffer to orphan frags from
1854  *	@gfp_mask: allocation mask for replacement pages
1855  *
1856  *	For each frag in the SKB which needs a destructor (i.e. has an
1857  *	owner) create a copy of that frag and release the original
1858  *	page by calling the destructor.
1859  */
1860 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
1861 {
1862 	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
1863 		return 0;
1864 	return skb_copy_ubufs(skb, gfp_mask);
1865 }
1866 
1867 /**
1868  *	__skb_queue_purge - empty a list
1869  *	@list: list to empty
1870  *
1871  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
1872  *	the list and one reference dropped. This function does not take the
1873  *	list lock and the caller must hold the relevant locks to use it.
1874  */
1875 extern void skb_queue_purge(struct sk_buff_head *list);
1876 static inline void __skb_queue_purge(struct sk_buff_head *list)
1877 {
1878 	struct sk_buff *skb;
1879 	while ((skb = __skb_dequeue(list)) != NULL)
1880 		kfree_skb(skb);
1881 }
1882 
1883 #define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
1884 #define NETDEV_FRAG_PAGE_MAX_SIZE  (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
1885 #define NETDEV_PAGECNT_MAX_BIAS	   NETDEV_FRAG_PAGE_MAX_SIZE
1886 
1887 extern void *netdev_alloc_frag(unsigned int fragsz);
1888 
1889 extern struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
1890 					  unsigned int length,
1891 					  gfp_t gfp_mask);
1892 
1893 /**
1894  *	netdev_alloc_skb - allocate an skbuff for rx on a specific device
1895  *	@dev: network device to receive on
1896  *	@length: length to allocate
1897  *
1898  *	Allocate a new &sk_buff and assign it a usage count of one. The
1899  *	buffer has unspecified headroom built in. Users should allocate
1900  *	the headroom they think they need without accounting for the
1901  *	built in space. The built in space is used for optimisations.
1902  *
1903  *	%NULL is returned if there is no free memory. Although this function
1904  *	allocates memory it can be called from an interrupt.
1905  */
1906 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
1907 					       unsigned int length)
1908 {
1909 	return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
1910 }
1911 
1912 /* legacy helper around __netdev_alloc_skb() */
1913 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
1914 					      gfp_t gfp_mask)
1915 {
1916 	return __netdev_alloc_skb(NULL, length, gfp_mask);
1917 }
1918 
1919 /* legacy helper around netdev_alloc_skb() */
1920 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
1921 {
1922 	return netdev_alloc_skb(NULL, length);
1923 }
1924 
1925 
1926 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
1927 		unsigned int length, gfp_t gfp)
1928 {
1929 	struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
1930 
1931 	if (NET_IP_ALIGN && skb)
1932 		skb_reserve(skb, NET_IP_ALIGN);
1933 	return skb;
1934 }
1935 
1936 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
1937 		unsigned int length)
1938 {
1939 	return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
1940 }
1941 
1942 /*
1943  *	__skb_alloc_page - allocate pages for ps-rx on a skb and preserve pfmemalloc data
1944  *	@gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
1945  *	@skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
1946  *	@order: size of the allocation
1947  *
1948  * 	Allocate a new page.
1949  *
1950  * 	%NULL is returned if there is no free memory.
1951 */
1952 static inline struct page *__skb_alloc_pages(gfp_t gfp_mask,
1953 					      struct sk_buff *skb,
1954 					      unsigned int order)
1955 {
1956 	struct page *page;
1957 
1958 	gfp_mask |= __GFP_COLD;
1959 
1960 	if (!(gfp_mask & __GFP_NOMEMALLOC))
1961 		gfp_mask |= __GFP_MEMALLOC;
1962 
1963 	page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
1964 	if (skb && page && page->pfmemalloc)
1965 		skb->pfmemalloc = true;
1966 
1967 	return page;
1968 }
1969 
1970 /**
1971  *	__skb_alloc_page - allocate a page for ps-rx for a given skb and preserve pfmemalloc data
1972  *	@gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
1973  *	@skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
1974  *
1975  * 	Allocate a new page.
1976  *
1977  * 	%NULL is returned if there is no free memory.
1978  */
1979 static inline struct page *__skb_alloc_page(gfp_t gfp_mask,
1980 					     struct sk_buff *skb)
1981 {
1982 	return __skb_alloc_pages(gfp_mask, skb, 0);
1983 }
1984 
1985 /**
1986  *	skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
1987  *	@page: The page that was allocated from skb_alloc_page
1988  *	@skb: The skb that may need pfmemalloc set
1989  */
1990 static inline void skb_propagate_pfmemalloc(struct page *page,
1991 					     struct sk_buff *skb)
1992 {
1993 	if (page && page->pfmemalloc)
1994 		skb->pfmemalloc = true;
1995 }
1996 
1997 /**
1998  * skb_frag_page - retrieve the page refered to by a paged fragment
1999  * @frag: the paged fragment
2000  *
2001  * Returns the &struct page associated with @frag.
2002  */
2003 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2004 {
2005 	return frag->page.p;
2006 }
2007 
2008 /**
2009  * __skb_frag_ref - take an addition reference on a paged fragment.
2010  * @frag: the paged fragment
2011  *
2012  * Takes an additional reference on the paged fragment @frag.
2013  */
2014 static inline void __skb_frag_ref(skb_frag_t *frag)
2015 {
2016 	get_page(skb_frag_page(frag));
2017 }
2018 
2019 /**
2020  * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2021  * @skb: the buffer
2022  * @f: the fragment offset.
2023  *
2024  * Takes an additional reference on the @f'th paged fragment of @skb.
2025  */
2026 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2027 {
2028 	__skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2029 }
2030 
2031 /**
2032  * __skb_frag_unref - release a reference on a paged fragment.
2033  * @frag: the paged fragment
2034  *
2035  * Releases a reference on the paged fragment @frag.
2036  */
2037 static inline void __skb_frag_unref(skb_frag_t *frag)
2038 {
2039 	put_page(skb_frag_page(frag));
2040 }
2041 
2042 /**
2043  * skb_frag_unref - release a reference on a paged fragment of an skb.
2044  * @skb: the buffer
2045  * @f: the fragment offset
2046  *
2047  * Releases a reference on the @f'th paged fragment of @skb.
2048  */
2049 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2050 {
2051 	__skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2052 }
2053 
2054 /**
2055  * skb_frag_address - gets the address of the data contained in a paged fragment
2056  * @frag: the paged fragment buffer
2057  *
2058  * Returns the address of the data within @frag. The page must already
2059  * be mapped.
2060  */
2061 static inline void *skb_frag_address(const skb_frag_t *frag)
2062 {
2063 	return page_address(skb_frag_page(frag)) + frag->page_offset;
2064 }
2065 
2066 /**
2067  * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2068  * @frag: the paged fragment buffer
2069  *
2070  * Returns the address of the data within @frag. Checks that the page
2071  * is mapped and returns %NULL otherwise.
2072  */
2073 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2074 {
2075 	void *ptr = page_address(skb_frag_page(frag));
2076 	if (unlikely(!ptr))
2077 		return NULL;
2078 
2079 	return ptr + frag->page_offset;
2080 }
2081 
2082 /**
2083  * __skb_frag_set_page - sets the page contained in a paged fragment
2084  * @frag: the paged fragment
2085  * @page: the page to set
2086  *
2087  * Sets the fragment @frag to contain @page.
2088  */
2089 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2090 {
2091 	frag->page.p = page;
2092 }
2093 
2094 /**
2095  * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2096  * @skb: the buffer
2097  * @f: the fragment offset
2098  * @page: the page to set
2099  *
2100  * Sets the @f'th fragment of @skb to contain @page.
2101  */
2102 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2103 				     struct page *page)
2104 {
2105 	__skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2106 }
2107 
2108 /**
2109  * skb_frag_dma_map - maps a paged fragment via the DMA API
2110  * @dev: the device to map the fragment to
2111  * @frag: the paged fragment to map
2112  * @offset: the offset within the fragment (starting at the
2113  *          fragment's own offset)
2114  * @size: the number of bytes to map
2115  * @dir: the direction of the mapping (%PCI_DMA_*)
2116  *
2117  * Maps the page associated with @frag to @device.
2118  */
2119 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2120 					  const skb_frag_t *frag,
2121 					  size_t offset, size_t size,
2122 					  enum dma_data_direction dir)
2123 {
2124 	return dma_map_page(dev, skb_frag_page(frag),
2125 			    frag->page_offset + offset, size, dir);
2126 }
2127 
2128 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2129 					gfp_t gfp_mask)
2130 {
2131 	return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2132 }
2133 
2134 /**
2135  *	skb_clone_writable - is the header of a clone writable
2136  *	@skb: buffer to check
2137  *	@len: length up to which to write
2138  *
2139  *	Returns true if modifying the header part of the cloned buffer
2140  *	does not requires the data to be copied.
2141  */
2142 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2143 {
2144 	return !skb_header_cloned(skb) &&
2145 	       skb_headroom(skb) + len <= skb->hdr_len;
2146 }
2147 
2148 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2149 			    int cloned)
2150 {
2151 	int delta = 0;
2152 
2153 	if (headroom > skb_headroom(skb))
2154 		delta = headroom - skb_headroom(skb);
2155 
2156 	if (delta || cloned)
2157 		return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2158 					GFP_ATOMIC);
2159 	return 0;
2160 }
2161 
2162 /**
2163  *	skb_cow - copy header of skb when it is required
2164  *	@skb: buffer to cow
2165  *	@headroom: needed headroom
2166  *
2167  *	If the skb passed lacks sufficient headroom or its data part
2168  *	is shared, data is reallocated. If reallocation fails, an error
2169  *	is returned and original skb is not changed.
2170  *
2171  *	The result is skb with writable area skb->head...skb->tail
2172  *	and at least @headroom of space at head.
2173  */
2174 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2175 {
2176 	return __skb_cow(skb, headroom, skb_cloned(skb));
2177 }
2178 
2179 /**
2180  *	skb_cow_head - skb_cow but only making the head writable
2181  *	@skb: buffer to cow
2182  *	@headroom: needed headroom
2183  *
2184  *	This function is identical to skb_cow except that we replace the
2185  *	skb_cloned check by skb_header_cloned.  It should be used when
2186  *	you only need to push on some header and do not need to modify
2187  *	the data.
2188  */
2189 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2190 {
2191 	return __skb_cow(skb, headroom, skb_header_cloned(skb));
2192 }
2193 
2194 /**
2195  *	skb_padto	- pad an skbuff up to a minimal size
2196  *	@skb: buffer to pad
2197  *	@len: minimal length
2198  *
2199  *	Pads up a buffer to ensure the trailing bytes exist and are
2200  *	blanked. If the buffer already contains sufficient data it
2201  *	is untouched. Otherwise it is extended. Returns zero on
2202  *	success. The skb is freed on error.
2203  */
2204 
2205 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2206 {
2207 	unsigned int size = skb->len;
2208 	if (likely(size >= len))
2209 		return 0;
2210 	return skb_pad(skb, len - size);
2211 }
2212 
2213 static inline int skb_add_data(struct sk_buff *skb,
2214 			       char __user *from, int copy)
2215 {
2216 	const int off = skb->len;
2217 
2218 	if (skb->ip_summed == CHECKSUM_NONE) {
2219 		int err = 0;
2220 		__wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
2221 							    copy, 0, &err);
2222 		if (!err) {
2223 			skb->csum = csum_block_add(skb->csum, csum, off);
2224 			return 0;
2225 		}
2226 	} else if (!copy_from_user(skb_put(skb, copy), from, copy))
2227 		return 0;
2228 
2229 	__skb_trim(skb, off);
2230 	return -EFAULT;
2231 }
2232 
2233 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2234 				    const struct page *page, int off)
2235 {
2236 	if (i) {
2237 		const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2238 
2239 		return page == skb_frag_page(frag) &&
2240 		       off == frag->page_offset + skb_frag_size(frag);
2241 	}
2242 	return false;
2243 }
2244 
2245 static inline int __skb_linearize(struct sk_buff *skb)
2246 {
2247 	return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2248 }
2249 
2250 /**
2251  *	skb_linearize - convert paged skb to linear one
2252  *	@skb: buffer to linarize
2253  *
2254  *	If there is no free memory -ENOMEM is returned, otherwise zero
2255  *	is returned and the old skb data released.
2256  */
2257 static inline int skb_linearize(struct sk_buff *skb)
2258 {
2259 	return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2260 }
2261 
2262 /**
2263  * skb_has_shared_frag - can any frag be overwritten
2264  * @skb: buffer to test
2265  *
2266  * Return true if the skb has at least one frag that might be modified
2267  * by an external entity (as in vmsplice()/sendfile())
2268  */
2269 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2270 {
2271 	return skb_is_nonlinear(skb) &&
2272 	       skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2273 }
2274 
2275 /**
2276  *	skb_linearize_cow - make sure skb is linear and writable
2277  *	@skb: buffer to process
2278  *
2279  *	If there is no free memory -ENOMEM is returned, otherwise zero
2280  *	is returned and the old skb data released.
2281  */
2282 static inline int skb_linearize_cow(struct sk_buff *skb)
2283 {
2284 	return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2285 	       __skb_linearize(skb) : 0;
2286 }
2287 
2288 /**
2289  *	skb_postpull_rcsum - update checksum for received skb after pull
2290  *	@skb: buffer to update
2291  *	@start: start of data before pull
2292  *	@len: length of data pulled
2293  *
2294  *	After doing a pull on a received packet, you need to call this to
2295  *	update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2296  *	CHECKSUM_NONE so that it can be recomputed from scratch.
2297  */
2298 
2299 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2300 				      const void *start, unsigned int len)
2301 {
2302 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2303 		skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2304 }
2305 
2306 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2307 
2308 /**
2309  *	pskb_trim_rcsum - trim received skb and update checksum
2310  *	@skb: buffer to trim
2311  *	@len: new length
2312  *
2313  *	This is exactly the same as pskb_trim except that it ensures the
2314  *	checksum of received packets are still valid after the operation.
2315  */
2316 
2317 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2318 {
2319 	if (likely(len >= skb->len))
2320 		return 0;
2321 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2322 		skb->ip_summed = CHECKSUM_NONE;
2323 	return __pskb_trim(skb, len);
2324 }
2325 
2326 #define skb_queue_walk(queue, skb) \
2327 		for (skb = (queue)->next;					\
2328 		     skb != (struct sk_buff *)(queue);				\
2329 		     skb = skb->next)
2330 
2331 #define skb_queue_walk_safe(queue, skb, tmp)					\
2332 		for (skb = (queue)->next, tmp = skb->next;			\
2333 		     skb != (struct sk_buff *)(queue);				\
2334 		     skb = tmp, tmp = skb->next)
2335 
2336 #define skb_queue_walk_from(queue, skb)						\
2337 		for (; skb != (struct sk_buff *)(queue);			\
2338 		     skb = skb->next)
2339 
2340 #define skb_queue_walk_from_safe(queue, skb, tmp)				\
2341 		for (tmp = skb->next;						\
2342 		     skb != (struct sk_buff *)(queue);				\
2343 		     skb = tmp, tmp = skb->next)
2344 
2345 #define skb_queue_reverse_walk(queue, skb) \
2346 		for (skb = (queue)->prev;					\
2347 		     skb != (struct sk_buff *)(queue);				\
2348 		     skb = skb->prev)
2349 
2350 #define skb_queue_reverse_walk_safe(queue, skb, tmp)				\
2351 		for (skb = (queue)->prev, tmp = skb->prev;			\
2352 		     skb != (struct sk_buff *)(queue);				\
2353 		     skb = tmp, tmp = skb->prev)
2354 
2355 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp)			\
2356 		for (tmp = skb->prev;						\
2357 		     skb != (struct sk_buff *)(queue);				\
2358 		     skb = tmp, tmp = skb->prev)
2359 
2360 static inline bool skb_has_frag_list(const struct sk_buff *skb)
2361 {
2362 	return skb_shinfo(skb)->frag_list != NULL;
2363 }
2364 
2365 static inline void skb_frag_list_init(struct sk_buff *skb)
2366 {
2367 	skb_shinfo(skb)->frag_list = NULL;
2368 }
2369 
2370 static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2371 {
2372 	frag->next = skb_shinfo(skb)->frag_list;
2373 	skb_shinfo(skb)->frag_list = frag;
2374 }
2375 
2376 #define skb_walk_frags(skb, iter)	\
2377 	for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2378 
2379 extern struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2380 					   int *peeked, int *off, int *err);
2381 extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags,
2382 					 int noblock, int *err);
2383 extern unsigned int    datagram_poll(struct file *file, struct socket *sock,
2384 				     struct poll_table_struct *wait);
2385 extern int	       skb_copy_datagram_iovec(const struct sk_buff *from,
2386 					       int offset, struct iovec *to,
2387 					       int size);
2388 extern int	       skb_copy_and_csum_datagram_iovec(struct sk_buff *skb,
2389 							int hlen,
2390 							struct iovec *iov);
2391 extern int	       skb_copy_datagram_from_iovec(struct sk_buff *skb,
2392 						    int offset,
2393 						    const struct iovec *from,
2394 						    int from_offset,
2395 						    int len);
2396 extern int	       skb_copy_datagram_const_iovec(const struct sk_buff *from,
2397 						     int offset,
2398 						     const struct iovec *to,
2399 						     int to_offset,
2400 						     int size);
2401 extern void	       skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2402 extern void	       skb_free_datagram_locked(struct sock *sk,
2403 						struct sk_buff *skb);
2404 extern int	       skb_kill_datagram(struct sock *sk, struct sk_buff *skb,
2405 					 unsigned int flags);
2406 extern __wsum	       skb_checksum(const struct sk_buff *skb, int offset,
2407 				    int len, __wsum csum);
2408 extern int	       skb_copy_bits(const struct sk_buff *skb, int offset,
2409 				     void *to, int len);
2410 extern int	       skb_store_bits(struct sk_buff *skb, int offset,
2411 				      const void *from, int len);
2412 extern __wsum	       skb_copy_and_csum_bits(const struct sk_buff *skb,
2413 					      int offset, u8 *to, int len,
2414 					      __wsum csum);
2415 extern int             skb_splice_bits(struct sk_buff *skb,
2416 						unsigned int offset,
2417 						struct pipe_inode_info *pipe,
2418 						unsigned int len,
2419 						unsigned int flags);
2420 extern void	       skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
2421 extern void	       skb_split(struct sk_buff *skb,
2422 				 struct sk_buff *skb1, const u32 len);
2423 extern int	       skb_shift(struct sk_buff *tgt, struct sk_buff *skb,
2424 				 int shiftlen);
2425 
2426 extern struct sk_buff *skb_segment(struct sk_buff *skb,
2427 				   netdev_features_t features);
2428 
2429 static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2430 				       int len, void *buffer)
2431 {
2432 	int hlen = skb_headlen(skb);
2433 
2434 	if (hlen - offset >= len)
2435 		return skb->data + offset;
2436 
2437 	if (skb_copy_bits(skb, offset, buffer, len) < 0)
2438 		return NULL;
2439 
2440 	return buffer;
2441 }
2442 
2443 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2444 					     void *to,
2445 					     const unsigned int len)
2446 {
2447 	memcpy(to, skb->data, len);
2448 }
2449 
2450 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2451 						    const int offset, void *to,
2452 						    const unsigned int len)
2453 {
2454 	memcpy(to, skb->data + offset, len);
2455 }
2456 
2457 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2458 					   const void *from,
2459 					   const unsigned int len)
2460 {
2461 	memcpy(skb->data, from, len);
2462 }
2463 
2464 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2465 						  const int offset,
2466 						  const void *from,
2467 						  const unsigned int len)
2468 {
2469 	memcpy(skb->data + offset, from, len);
2470 }
2471 
2472 extern void skb_init(void);
2473 
2474 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2475 {
2476 	return skb->tstamp;
2477 }
2478 
2479 /**
2480  *	skb_get_timestamp - get timestamp from a skb
2481  *	@skb: skb to get stamp from
2482  *	@stamp: pointer to struct timeval to store stamp in
2483  *
2484  *	Timestamps are stored in the skb as offsets to a base timestamp.
2485  *	This function converts the offset back to a struct timeval and stores
2486  *	it in stamp.
2487  */
2488 static inline void skb_get_timestamp(const struct sk_buff *skb,
2489 				     struct timeval *stamp)
2490 {
2491 	*stamp = ktime_to_timeval(skb->tstamp);
2492 }
2493 
2494 static inline void skb_get_timestampns(const struct sk_buff *skb,
2495 				       struct timespec *stamp)
2496 {
2497 	*stamp = ktime_to_timespec(skb->tstamp);
2498 }
2499 
2500 static inline void __net_timestamp(struct sk_buff *skb)
2501 {
2502 	skb->tstamp = ktime_get_real();
2503 }
2504 
2505 static inline ktime_t net_timedelta(ktime_t t)
2506 {
2507 	return ktime_sub(ktime_get_real(), t);
2508 }
2509 
2510 static inline ktime_t net_invalid_timestamp(void)
2511 {
2512 	return ktime_set(0, 0);
2513 }
2514 
2515 extern void skb_timestamping_init(void);
2516 
2517 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2518 
2519 extern void skb_clone_tx_timestamp(struct sk_buff *skb);
2520 extern bool skb_defer_rx_timestamp(struct sk_buff *skb);
2521 
2522 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2523 
2524 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2525 {
2526 }
2527 
2528 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2529 {
2530 	return false;
2531 }
2532 
2533 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2534 
2535 /**
2536  * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2537  *
2538  * PHY drivers may accept clones of transmitted packets for
2539  * timestamping via their phy_driver.txtstamp method. These drivers
2540  * must call this function to return the skb back to the stack, with
2541  * or without a timestamp.
2542  *
2543  * @skb: clone of the the original outgoing packet
2544  * @hwtstamps: hardware time stamps, may be NULL if not available
2545  *
2546  */
2547 void skb_complete_tx_timestamp(struct sk_buff *skb,
2548 			       struct skb_shared_hwtstamps *hwtstamps);
2549 
2550 /**
2551  * skb_tstamp_tx - queue clone of skb with send time stamps
2552  * @orig_skb:	the original outgoing packet
2553  * @hwtstamps:	hardware time stamps, may be NULL if not available
2554  *
2555  * If the skb has a socket associated, then this function clones the
2556  * skb (thus sharing the actual data and optional structures), stores
2557  * the optional hardware time stamping information (if non NULL) or
2558  * generates a software time stamp (otherwise), then queues the clone
2559  * to the error queue of the socket.  Errors are silently ignored.
2560  */
2561 extern void skb_tstamp_tx(struct sk_buff *orig_skb,
2562 			struct skb_shared_hwtstamps *hwtstamps);
2563 
2564 static inline void sw_tx_timestamp(struct sk_buff *skb)
2565 {
2566 	if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2567 	    !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
2568 		skb_tstamp_tx(skb, NULL);
2569 }
2570 
2571 /**
2572  * skb_tx_timestamp() - Driver hook for transmit timestamping
2573  *
2574  * Ethernet MAC Drivers should call this function in their hard_xmit()
2575  * function immediately before giving the sk_buff to the MAC hardware.
2576  *
2577  * @skb: A socket buffer.
2578  */
2579 static inline void skb_tx_timestamp(struct sk_buff *skb)
2580 {
2581 	skb_clone_tx_timestamp(skb);
2582 	sw_tx_timestamp(skb);
2583 }
2584 
2585 /**
2586  * skb_complete_wifi_ack - deliver skb with wifi status
2587  *
2588  * @skb: the original outgoing packet
2589  * @acked: ack status
2590  *
2591  */
2592 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2593 
2594 extern __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2595 extern __sum16 __skb_checksum_complete(struct sk_buff *skb);
2596 
2597 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2598 {
2599 	return skb->ip_summed & CHECKSUM_UNNECESSARY;
2600 }
2601 
2602 /**
2603  *	skb_checksum_complete - Calculate checksum of an entire packet
2604  *	@skb: packet to process
2605  *
2606  *	This function calculates the checksum over the entire packet plus
2607  *	the value of skb->csum.  The latter can be used to supply the
2608  *	checksum of a pseudo header as used by TCP/UDP.  It returns the
2609  *	checksum.
2610  *
2611  *	For protocols that contain complete checksums such as ICMP/TCP/UDP,
2612  *	this function can be used to verify that checksum on received
2613  *	packets.  In that case the function should return zero if the
2614  *	checksum is correct.  In particular, this function will return zero
2615  *	if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2616  *	hardware has already verified the correctness of the checksum.
2617  */
2618 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
2619 {
2620 	return skb_csum_unnecessary(skb) ?
2621 	       0 : __skb_checksum_complete(skb);
2622 }
2623 
2624 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2625 extern void nf_conntrack_destroy(struct nf_conntrack *nfct);
2626 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
2627 {
2628 	if (nfct && atomic_dec_and_test(&nfct->use))
2629 		nf_conntrack_destroy(nfct);
2630 }
2631 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
2632 {
2633 	if (nfct)
2634 		atomic_inc(&nfct->use);
2635 }
2636 #endif
2637 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
2638 static inline void nf_conntrack_get_reasm(struct sk_buff *skb)
2639 {
2640 	if (skb)
2641 		atomic_inc(&skb->users);
2642 }
2643 static inline void nf_conntrack_put_reasm(struct sk_buff *skb)
2644 {
2645 	if (skb)
2646 		kfree_skb(skb);
2647 }
2648 #endif
2649 #ifdef CONFIG_BRIDGE_NETFILTER
2650 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
2651 {
2652 	if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
2653 		kfree(nf_bridge);
2654 }
2655 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
2656 {
2657 	if (nf_bridge)
2658 		atomic_inc(&nf_bridge->use);
2659 }
2660 #endif /* CONFIG_BRIDGE_NETFILTER */
2661 static inline void nf_reset(struct sk_buff *skb)
2662 {
2663 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2664 	nf_conntrack_put(skb->nfct);
2665 	skb->nfct = NULL;
2666 #endif
2667 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
2668 	nf_conntrack_put_reasm(skb->nfct_reasm);
2669 	skb->nfct_reasm = NULL;
2670 #endif
2671 #ifdef CONFIG_BRIDGE_NETFILTER
2672 	nf_bridge_put(skb->nf_bridge);
2673 	skb->nf_bridge = NULL;
2674 #endif
2675 }
2676 
2677 /* Note: This doesn't put any conntrack and bridge info in dst. */
2678 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2679 {
2680 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2681 	dst->nfct = src->nfct;
2682 	nf_conntrack_get(src->nfct);
2683 	dst->nfctinfo = src->nfctinfo;
2684 #endif
2685 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
2686 	dst->nfct_reasm = src->nfct_reasm;
2687 	nf_conntrack_get_reasm(src->nfct_reasm);
2688 #endif
2689 #ifdef CONFIG_BRIDGE_NETFILTER
2690 	dst->nf_bridge  = src->nf_bridge;
2691 	nf_bridge_get(src->nf_bridge);
2692 #endif
2693 }
2694 
2695 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2696 {
2697 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
2698 	nf_conntrack_put(dst->nfct);
2699 #endif
2700 #ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
2701 	nf_conntrack_put_reasm(dst->nfct_reasm);
2702 #endif
2703 #ifdef CONFIG_BRIDGE_NETFILTER
2704 	nf_bridge_put(dst->nf_bridge);
2705 #endif
2706 	__nf_copy(dst, src);
2707 }
2708 
2709 #ifdef CONFIG_NETWORK_SECMARK
2710 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2711 {
2712 	to->secmark = from->secmark;
2713 }
2714 
2715 static inline void skb_init_secmark(struct sk_buff *skb)
2716 {
2717 	skb->secmark = 0;
2718 }
2719 #else
2720 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2721 { }
2722 
2723 static inline void skb_init_secmark(struct sk_buff *skb)
2724 { }
2725 #endif
2726 
2727 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
2728 {
2729 	skb->queue_mapping = queue_mapping;
2730 }
2731 
2732 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
2733 {
2734 	return skb->queue_mapping;
2735 }
2736 
2737 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
2738 {
2739 	to->queue_mapping = from->queue_mapping;
2740 }
2741 
2742 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
2743 {
2744 	skb->queue_mapping = rx_queue + 1;
2745 }
2746 
2747 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
2748 {
2749 	return skb->queue_mapping - 1;
2750 }
2751 
2752 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
2753 {
2754 	return skb->queue_mapping != 0;
2755 }
2756 
2757 extern u16 __skb_tx_hash(const struct net_device *dev,
2758 			 const struct sk_buff *skb,
2759 			 unsigned int num_tx_queues);
2760 
2761 #ifdef CONFIG_XFRM
2762 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
2763 {
2764 	return skb->sp;
2765 }
2766 #else
2767 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
2768 {
2769 	return NULL;
2770 }
2771 #endif
2772 
2773 /* Keeps track of mac header offset relative to skb->head.
2774  * It is useful for TSO of Tunneling protocol. e.g. GRE.
2775  * For non-tunnel skb it points to skb_mac_header() and for
2776  * tunnel skb it points to outer mac header. */
2777 struct skb_gso_cb {
2778 	int mac_offset;
2779 };
2780 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
2781 
2782 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
2783 {
2784 	return (skb_mac_header(inner_skb) - inner_skb->head) -
2785 		SKB_GSO_CB(inner_skb)->mac_offset;
2786 }
2787 
2788 static inline bool skb_is_gso(const struct sk_buff *skb)
2789 {
2790 	return skb_shinfo(skb)->gso_size;
2791 }
2792 
2793 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
2794 {
2795 	return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
2796 }
2797 
2798 extern void __skb_warn_lro_forwarding(const struct sk_buff *skb);
2799 
2800 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
2801 {
2802 	/* LRO sets gso_size but not gso_type, whereas if GSO is really
2803 	 * wanted then gso_type will be set. */
2804 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
2805 
2806 	if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
2807 	    unlikely(shinfo->gso_type == 0)) {
2808 		__skb_warn_lro_forwarding(skb);
2809 		return true;
2810 	}
2811 	return false;
2812 }
2813 
2814 static inline void skb_forward_csum(struct sk_buff *skb)
2815 {
2816 	/* Unfortunately we don't support this one.  Any brave souls? */
2817 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2818 		skb->ip_summed = CHECKSUM_NONE;
2819 }
2820 
2821 /**
2822  * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
2823  * @skb: skb to check
2824  *
2825  * fresh skbs have their ip_summed set to CHECKSUM_NONE.
2826  * Instead of forcing ip_summed to CHECKSUM_NONE, we can
2827  * use this helper, to document places where we make this assertion.
2828  */
2829 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
2830 {
2831 #ifdef DEBUG
2832 	BUG_ON(skb->ip_summed != CHECKSUM_NONE);
2833 #endif
2834 }
2835 
2836 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
2837 
2838 /**
2839  * skb_head_is_locked - Determine if the skb->head is locked down
2840  * @skb: skb to check
2841  *
2842  * The head on skbs build around a head frag can be removed if they are
2843  * not cloned.  This function returns true if the skb head is locked down
2844  * due to either being allocated via kmalloc, or by being a clone with
2845  * multiple references to the head.
2846  */
2847 static inline bool skb_head_is_locked(const struct sk_buff *skb)
2848 {
2849 	return !skb->head_frag || skb_cloned(skb);
2850 }
2851 #endif	/* __KERNEL__ */
2852 #endif	/* _LINUX_SKBUFF_H */
2853