xref: /linux-6.15/include/linux/skbuff.h (revision 6e5c8381)
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 #include <linux/rbtree.h>
24 #include <linux/socket.h>
25 
26 #include <linux/atomic.h>
27 #include <asm/types.h>
28 #include <linux/spinlock.h>
29 #include <linux/net.h>
30 #include <linux/textsearch.h>
31 #include <net/checksum.h>
32 #include <linux/rcupdate.h>
33 #include <linux/hrtimer.h>
34 #include <linux/dma-mapping.h>
35 #include <linux/netdev_features.h>
36 #include <linux/sched.h>
37 #include <net/flow_dissector.h>
38 #include <linux/splice.h>
39 #include <linux/in6.h>
40 #include <linux/if_packet.h>
41 #include <net/flow.h>
42 
43 /* The interface for checksum offload between the stack and networking drivers
44  * is as follows...
45  *
46  * A. IP checksum related features
47  *
48  * Drivers advertise checksum offload capabilities in the features of a device.
49  * From the stack's point of view these are capabilities offered by the driver,
50  * a driver typically only advertises features that it is capable of offloading
51  * to its device.
52  *
53  * The checksum related features are:
54  *
55  *	NETIF_F_HW_CSUM	- The driver (or its device) is able to compute one
56  *			  IP (one's complement) checksum for any combination
57  *			  of protocols or protocol layering. The checksum is
58  *			  computed and set in a packet per the CHECKSUM_PARTIAL
59  *			  interface (see below).
60  *
61  *	NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
62  *			  TCP or UDP packets over IPv4. These are specifically
63  *			  unencapsulated packets of the form IPv4|TCP or
64  *			  IPv4|UDP where the Protocol field in the IPv4 header
65  *			  is TCP or UDP. The IPv4 header may contain IP options
66  *			  This feature cannot be set in features for a device
67  *			  with NETIF_F_HW_CSUM also set. This feature is being
68  *			  DEPRECATED (see below).
69  *
70  *	NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
71  *			  TCP or UDP packets over IPv6. These are specifically
72  *			  unencapsulated packets of the form IPv6|TCP or
73  *			  IPv4|UDP where the Next Header field in the IPv6
74  *			  header is either TCP or UDP. IPv6 extension headers
75  *			  are not supported with this feature. This feature
76  *			  cannot be set in features for a device with
77  *			  NETIF_F_HW_CSUM also set. This feature is being
78  *			  DEPRECATED (see below).
79  *
80  *	NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
81  *			 This flag is used only used to disable the RX checksum
82  *			 feature for a device. The stack will accept receive
83  *			 checksum indication in packets received on a device
84  *			 regardless of whether NETIF_F_RXCSUM is set.
85  *
86  * B. Checksumming of received packets by device. Indication of checksum
87  *    verification is in set skb->ip_summed. Possible values are:
88  *
89  * CHECKSUM_NONE:
90  *
91  *   Device did not checksum this packet e.g. due to lack of capabilities.
92  *   The packet contains full (though not verified) checksum in packet but
93  *   not in skb->csum. Thus, skb->csum is undefined in this case.
94  *
95  * CHECKSUM_UNNECESSARY:
96  *
97  *   The hardware you're dealing with doesn't calculate the full checksum
98  *   (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
99  *   for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
100  *   if their checksums are okay. skb->csum is still undefined in this case
101  *   though. A driver or device must never modify the checksum field in the
102  *   packet even if checksum is verified.
103  *
104  *   CHECKSUM_UNNECESSARY is applicable to following protocols:
105  *     TCP: IPv6 and IPv4.
106  *     UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
107  *       zero UDP checksum for either IPv4 or IPv6, the networking stack
108  *       may perform further validation in this case.
109  *     GRE: only if the checksum is present in the header.
110  *     SCTP: indicates the CRC in SCTP header has been validated.
111  *
112  *   skb->csum_level indicates the number of consecutive checksums found in
113  *   the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
114  *   For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
115  *   and a device is able to verify the checksums for UDP (possibly zero),
116  *   GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
117  *   two. If the device were only able to verify the UDP checksum and not
118  *   GRE, either because it doesn't support GRE checksum of because GRE
119  *   checksum is bad, skb->csum_level would be set to zero (TCP checksum is
120  *   not considered in this case).
121  *
122  * CHECKSUM_COMPLETE:
123  *
124  *   This is the most generic way. The device supplied checksum of the _whole_
125  *   packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
126  *   hardware doesn't need to parse L3/L4 headers to implement this.
127  *
128  *   Note: Even if device supports only some protocols, but is able to produce
129  *   skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
130  *
131  * CHECKSUM_PARTIAL:
132  *
133  *   A checksum is set up to be offloaded to a device as described in the
134  *   output description for CHECKSUM_PARTIAL. This may occur on a packet
135  *   received directly from another Linux OS, e.g., a virtualized Linux kernel
136  *   on the same host, or it may be set in the input path in GRO or remote
137  *   checksum offload. For the purposes of checksum verification, the checksum
138  *   referred to by skb->csum_start + skb->csum_offset and any preceding
139  *   checksums in the packet are considered verified. Any checksums in the
140  *   packet that are after the checksum being offloaded are not considered to
141  *   be verified.
142  *
143  * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
144  *    in the skb->ip_summed for a packet. Values are:
145  *
146  * CHECKSUM_PARTIAL:
147  *
148  *   The driver is required to checksum the packet as seen by hard_start_xmit()
149  *   from skb->csum_start up to the end, and to record/write the checksum at
150  *   offset skb->csum_start + skb->csum_offset. A driver may verify that the
151  *   csum_start and csum_offset values are valid values given the length and
152  *   offset of the packet, however they should not attempt to validate that the
153  *   checksum refers to a legitimate transport layer checksum-- it is the
154  *   purview of the stack to validate that csum_start and csum_offset are set
155  *   correctly.
156  *
157  *   When the stack requests checksum offload for a packet, the driver MUST
158  *   ensure that the checksum is set correctly. A driver can either offload the
159  *   checksum calculation to the device, or call skb_checksum_help (in the case
160  *   that the device does not support offload for a particular checksum).
161  *
162  *   NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
163  *   NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
164  *   checksum offload capability. If a	device has limited checksum capabilities
165  *   (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as
166  *   described above) a helper function can be called to resolve
167  *   CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper
168  *   function takes a spec argument that describes the protocol layer that is
169  *   supported for checksum offload and can be called for each packet. If a
170  *   packet does not match the specification for offload, skb_checksum_help
171  *   is called to resolve the checksum.
172  *
173  * CHECKSUM_NONE:
174  *
175  *   The skb was already checksummed by the protocol, or a checksum is not
176  *   required.
177  *
178  * CHECKSUM_UNNECESSARY:
179  *
180  *   This has the same meaning on as CHECKSUM_NONE for checksum offload on
181  *   output.
182  *
183  * CHECKSUM_COMPLETE:
184  *   Not used in checksum output. If a driver observes a packet with this value
185  *   set in skbuff, if should treat as CHECKSUM_NONE being set.
186  *
187  * D. Non-IP checksum (CRC) offloads
188  *
189  *   NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
190  *     offloading the SCTP CRC in a packet. To perform this offload the stack
191  *     will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
192  *     accordingly. Note the there is no indication in the skbuff that the
193  *     CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports
194  *     both IP checksum offload and SCTP CRC offload must verify which offload
195  *     is configured for a packet presumably by inspecting packet headers.
196  *
197  *   NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
198  *     offloading the FCOE CRC in a packet. To perform this offload the stack
199  *     will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
200  *     accordingly. Note the there is no indication in the skbuff that the
201  *     CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
202  *     both IP checksum offload and FCOE CRC offload must verify which offload
203  *     is configured for a packet presumably by inspecting packet headers.
204  *
205  * E. Checksumming on output with GSO.
206  *
207  * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
208  * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
209  * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
210  * part of the GSO operation is implied. If a checksum is being offloaded
211  * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
212  * are set to refer to the outermost checksum being offload (two offloaded
213  * checksums are possible with UDP encapsulation).
214  */
215 
216 /* Don't change this without changing skb_csum_unnecessary! */
217 #define CHECKSUM_NONE		0
218 #define CHECKSUM_UNNECESSARY	1
219 #define CHECKSUM_COMPLETE	2
220 #define CHECKSUM_PARTIAL	3
221 
222 /* Maximum value in skb->csum_level */
223 #define SKB_MAX_CSUM_LEVEL	3
224 
225 #define SKB_DATA_ALIGN(X)	ALIGN(X, SMP_CACHE_BYTES)
226 #define SKB_WITH_OVERHEAD(X)	\
227 	((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
228 #define SKB_MAX_ORDER(X, ORDER) \
229 	SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
230 #define SKB_MAX_HEAD(X)		(SKB_MAX_ORDER((X), 0))
231 #define SKB_MAX_ALLOC		(SKB_MAX_ORDER(0, 2))
232 
233 /* return minimum truesize of one skb containing X bytes of data */
234 #define SKB_TRUESIZE(X) ((X) +						\
235 			 SKB_DATA_ALIGN(sizeof(struct sk_buff)) +	\
236 			 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
237 
238 struct net_device;
239 struct scatterlist;
240 struct pipe_inode_info;
241 struct iov_iter;
242 struct napi_struct;
243 
244 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
245 struct nf_conntrack {
246 	atomic_t use;
247 };
248 #endif
249 
250 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
251 struct nf_bridge_info {
252 	atomic_t		use;
253 	enum {
254 		BRNF_PROTO_UNCHANGED,
255 		BRNF_PROTO_8021Q,
256 		BRNF_PROTO_PPPOE
257 	} orig_proto:8;
258 	u8			pkt_otherhost:1;
259 	u8			in_prerouting:1;
260 	u8			bridged_dnat:1;
261 	__u16			frag_max_size;
262 	struct net_device	*physindev;
263 
264 	/* always valid & non-NULL from FORWARD on, for physdev match */
265 	struct net_device	*physoutdev;
266 	union {
267 		/* prerouting: detect dnat in orig/reply direction */
268 		__be32          ipv4_daddr;
269 		struct in6_addr ipv6_daddr;
270 
271 		/* after prerouting + nat detected: store original source
272 		 * mac since neigh resolution overwrites it, only used while
273 		 * skb is out in neigh layer.
274 		 */
275 		char neigh_header[8];
276 	};
277 };
278 #endif
279 
280 struct sk_buff_head {
281 	/* These two members must be first. */
282 	struct sk_buff	*next;
283 	struct sk_buff	*prev;
284 
285 	__u32		qlen;
286 	spinlock_t	lock;
287 };
288 
289 struct sk_buff;
290 
291 /* To allow 64K frame to be packed as single skb without frag_list we
292  * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
293  * buffers which do not start on a page boundary.
294  *
295  * Since GRO uses frags we allocate at least 16 regardless of page
296  * size.
297  */
298 #if (65536/PAGE_SIZE + 1) < 16
299 #define MAX_SKB_FRAGS 16UL
300 #else
301 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
302 #endif
303 extern int sysctl_max_skb_frags;
304 
305 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
306  * segment using its current segmentation instead.
307  */
308 #define GSO_BY_FRAGS	0xFFFF
309 
310 typedef struct skb_frag_struct skb_frag_t;
311 
312 struct skb_frag_struct {
313 	struct {
314 		struct page *p;
315 	} page;
316 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
317 	__u32 page_offset;
318 	__u32 size;
319 #else
320 	__u16 page_offset;
321 	__u16 size;
322 #endif
323 };
324 
325 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
326 {
327 	return frag->size;
328 }
329 
330 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
331 {
332 	frag->size = size;
333 }
334 
335 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
336 {
337 	frag->size += delta;
338 }
339 
340 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
341 {
342 	frag->size -= delta;
343 }
344 
345 #define HAVE_HW_TIME_STAMP
346 
347 /**
348  * struct skb_shared_hwtstamps - hardware time stamps
349  * @hwtstamp:	hardware time stamp transformed into duration
350  *		since arbitrary point in time
351  *
352  * Software time stamps generated by ktime_get_real() are stored in
353  * skb->tstamp.
354  *
355  * hwtstamps can only be compared against other hwtstamps from
356  * the same device.
357  *
358  * This structure is attached to packets as part of the
359  * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
360  */
361 struct skb_shared_hwtstamps {
362 	ktime_t	hwtstamp;
363 };
364 
365 /* Definitions for tx_flags in struct skb_shared_info */
366 enum {
367 	/* generate hardware time stamp */
368 	SKBTX_HW_TSTAMP = 1 << 0,
369 
370 	/* generate software time stamp when queueing packet to NIC */
371 	SKBTX_SW_TSTAMP = 1 << 1,
372 
373 	/* device driver is going to provide hardware time stamp */
374 	SKBTX_IN_PROGRESS = 1 << 2,
375 
376 	/* device driver supports TX zero-copy buffers */
377 	SKBTX_DEV_ZEROCOPY = 1 << 3,
378 
379 	/* generate wifi status information (where possible) */
380 	SKBTX_WIFI_STATUS = 1 << 4,
381 
382 	/* This indicates at least one fragment might be overwritten
383 	 * (as in vmsplice(), sendfile() ...)
384 	 * If we need to compute a TX checksum, we'll need to copy
385 	 * all frags to avoid possible bad checksum
386 	 */
387 	SKBTX_SHARED_FRAG = 1 << 5,
388 
389 	/* generate software time stamp when entering packet scheduling */
390 	SKBTX_SCHED_TSTAMP = 1 << 6,
391 };
392 
393 #define SKBTX_ANY_SW_TSTAMP	(SKBTX_SW_TSTAMP    | \
394 				 SKBTX_SCHED_TSTAMP)
395 #define SKBTX_ANY_TSTAMP	(SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
396 
397 /*
398  * The callback notifies userspace to release buffers when skb DMA is done in
399  * lower device, the skb last reference should be 0 when calling this.
400  * The zerocopy_success argument is true if zero copy transmit occurred,
401  * false on data copy or out of memory error caused by data copy attempt.
402  * The ctx field is used to track device context.
403  * The desc field is used to track userspace buffer index.
404  */
405 struct ubuf_info {
406 	void (*callback)(struct ubuf_info *, bool zerocopy_success);
407 	void *ctx;
408 	unsigned long desc;
409 };
410 
411 /* This data is invariant across clones and lives at
412  * the end of the header data, ie. at skb->end.
413  */
414 struct skb_shared_info {
415 	unsigned char	nr_frags;
416 	__u8		tx_flags;
417 	unsigned short	gso_size;
418 	/* Warning: this field is not always filled in (UFO)! */
419 	unsigned short	gso_segs;
420 	unsigned short  gso_type;
421 	struct sk_buff	*frag_list;
422 	struct skb_shared_hwtstamps hwtstamps;
423 	u32		tskey;
424 	__be32          ip6_frag_id;
425 
426 	/*
427 	 * Warning : all fields before dataref are cleared in __alloc_skb()
428 	 */
429 	atomic_t	dataref;
430 
431 	/* Intermediate layers must ensure that destructor_arg
432 	 * remains valid until skb destructor */
433 	void *		destructor_arg;
434 
435 	/* must be last field, see pskb_expand_head() */
436 	skb_frag_t	frags[MAX_SKB_FRAGS];
437 };
438 
439 /* We divide dataref into two halves.  The higher 16 bits hold references
440  * to the payload part of skb->data.  The lower 16 bits hold references to
441  * the entire skb->data.  A clone of a headerless skb holds the length of
442  * the header in skb->hdr_len.
443  *
444  * All users must obey the rule that the skb->data reference count must be
445  * greater than or equal to the payload reference count.
446  *
447  * Holding a reference to the payload part means that the user does not
448  * care about modifications to the header part of skb->data.
449  */
450 #define SKB_DATAREF_SHIFT 16
451 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
452 
453 
454 enum {
455 	SKB_FCLONE_UNAVAILABLE,	/* skb has no fclone (from head_cache) */
456 	SKB_FCLONE_ORIG,	/* orig skb (from fclone_cache) */
457 	SKB_FCLONE_CLONE,	/* companion fclone skb (from fclone_cache) */
458 };
459 
460 enum {
461 	SKB_GSO_TCPV4 = 1 << 0,
462 	SKB_GSO_UDP = 1 << 1,
463 
464 	/* This indicates the skb is from an untrusted source. */
465 	SKB_GSO_DODGY = 1 << 2,
466 
467 	/* This indicates the tcp segment has CWR set. */
468 	SKB_GSO_TCP_ECN = 1 << 3,
469 
470 	SKB_GSO_TCP_FIXEDID = 1 << 4,
471 
472 	SKB_GSO_TCPV6 = 1 << 5,
473 
474 	SKB_GSO_FCOE = 1 << 6,
475 
476 	SKB_GSO_GRE = 1 << 7,
477 
478 	SKB_GSO_GRE_CSUM = 1 << 8,
479 
480 	SKB_GSO_IPXIP4 = 1 << 9,
481 
482 	SKB_GSO_IPXIP6 = 1 << 10,
483 
484 	SKB_GSO_UDP_TUNNEL = 1 << 11,
485 
486 	SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
487 
488 	SKB_GSO_PARTIAL = 1 << 13,
489 
490 	SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
491 
492 	SKB_GSO_SCTP = 1 << 15,
493 };
494 
495 #if BITS_PER_LONG > 32
496 #define NET_SKBUFF_DATA_USES_OFFSET 1
497 #endif
498 
499 #ifdef NET_SKBUFF_DATA_USES_OFFSET
500 typedef unsigned int sk_buff_data_t;
501 #else
502 typedef unsigned char *sk_buff_data_t;
503 #endif
504 
505 /**
506  * struct skb_mstamp - multi resolution time stamps
507  * @stamp_us: timestamp in us resolution
508  * @stamp_jiffies: timestamp in jiffies
509  */
510 struct skb_mstamp {
511 	union {
512 		u64		v64;
513 		struct {
514 			u32	stamp_us;
515 			u32	stamp_jiffies;
516 		};
517 	};
518 };
519 
520 /**
521  * skb_mstamp_get - get current timestamp
522  * @cl: place to store timestamps
523  */
524 static inline void skb_mstamp_get(struct skb_mstamp *cl)
525 {
526 	u64 val = local_clock();
527 
528 	do_div(val, NSEC_PER_USEC);
529 	cl->stamp_us = (u32)val;
530 	cl->stamp_jiffies = (u32)jiffies;
531 }
532 
533 /**
534  * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
535  * @t1: pointer to newest sample
536  * @t0: pointer to oldest sample
537  */
538 static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
539 				      const struct skb_mstamp *t0)
540 {
541 	s32 delta_us = t1->stamp_us - t0->stamp_us;
542 	u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
543 
544 	/* If delta_us is negative, this might be because interval is too big,
545 	 * or local_clock() drift is too big : fallback using jiffies.
546 	 */
547 	if (delta_us <= 0 ||
548 	    delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
549 
550 		delta_us = jiffies_to_usecs(delta_jiffies);
551 
552 	return delta_us;
553 }
554 
555 static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
556 				    const struct skb_mstamp *t0)
557 {
558 	s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
559 
560 	if (!diff)
561 		diff = t1->stamp_us - t0->stamp_us;
562 	return diff > 0;
563 }
564 
565 /**
566  *	struct sk_buff - socket buffer
567  *	@next: Next buffer in list
568  *	@prev: Previous buffer in list
569  *	@tstamp: Time we arrived/left
570  *	@rbnode: RB tree node, alternative to next/prev for netem/tcp
571  *	@sk: Socket we are owned by
572  *	@dev: Device we arrived on/are leaving by
573  *	@cb: Control buffer. Free for use by every layer. Put private vars here
574  *	@_skb_refdst: destination entry (with norefcount bit)
575  *	@sp: the security path, used for xfrm
576  *	@len: Length of actual data
577  *	@data_len: Data length
578  *	@mac_len: Length of link layer header
579  *	@hdr_len: writable header length of cloned skb
580  *	@csum: Checksum (must include start/offset pair)
581  *	@csum_start: Offset from skb->head where checksumming should start
582  *	@csum_offset: Offset from csum_start where checksum should be stored
583  *	@priority: Packet queueing priority
584  *	@ignore_df: allow local fragmentation
585  *	@cloned: Head may be cloned (check refcnt to be sure)
586  *	@ip_summed: Driver fed us an IP checksum
587  *	@nohdr: Payload reference only, must not modify header
588  *	@pkt_type: Packet class
589  *	@fclone: skbuff clone status
590  *	@ipvs_property: skbuff is owned by ipvs
591  *	@tc_skip_classify: do not classify packet. set by IFB device
592  *	@tc_at_ingress: used within tc_classify to distinguish in/egress
593  *	@tc_redirected: packet was redirected by a tc action
594  *	@tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
595  *	@peeked: this packet has been seen already, so stats have been
596  *		done for it, don't do them again
597  *	@nf_trace: netfilter packet trace flag
598  *	@protocol: Packet protocol from driver
599  *	@destructor: Destruct function
600  *	@_nfct: Associated connection, if any (with nfctinfo bits)
601  *	@nf_bridge: Saved data about a bridged frame - see br_netfilter.c
602  *	@skb_iif: ifindex of device we arrived on
603  *	@tc_index: Traffic control index
604  *	@hash: the packet hash
605  *	@queue_mapping: Queue mapping for multiqueue devices
606  *	@xmit_more: More SKBs are pending for this queue
607  *	@ndisc_nodetype: router type (from link layer)
608  *	@ooo_okay: allow the mapping of a socket to a queue to be changed
609  *	@l4_hash: indicate hash is a canonical 4-tuple hash over transport
610  *		ports.
611  *	@sw_hash: indicates hash was computed in software stack
612  *	@wifi_acked_valid: wifi_acked was set
613  *	@wifi_acked: whether frame was acked on wifi or not
614  *	@no_fcs:  Request NIC to treat last 4 bytes as Ethernet FCS
615  *	@dst_pending_confirm: need to confirm neighbour
616   *	@napi_id: id of the NAPI struct this skb came from
617  *	@secmark: security marking
618  *	@mark: Generic packet mark
619  *	@vlan_proto: vlan encapsulation protocol
620  *	@vlan_tci: vlan tag control information
621  *	@inner_protocol: Protocol (encapsulation)
622  *	@inner_transport_header: Inner transport layer header (encapsulation)
623  *	@inner_network_header: Network layer header (encapsulation)
624  *	@inner_mac_header: Link layer header (encapsulation)
625  *	@transport_header: Transport layer header
626  *	@network_header: Network layer header
627  *	@mac_header: Link layer header
628  *	@tail: Tail pointer
629  *	@end: End pointer
630  *	@head: Head of buffer
631  *	@data: Data head pointer
632  *	@truesize: Buffer size
633  *	@users: User count - see {datagram,tcp}.c
634  */
635 
636 struct sk_buff {
637 	union {
638 		struct {
639 			/* These two members must be first. */
640 			struct sk_buff		*next;
641 			struct sk_buff		*prev;
642 
643 			union {
644 				ktime_t		tstamp;
645 				struct skb_mstamp skb_mstamp;
646 			};
647 		};
648 		struct rb_node	rbnode; /* used in netem & tcp stack */
649 	};
650 	struct sock		*sk;
651 
652 	union {
653 		struct net_device	*dev;
654 		/* Some protocols might use this space to store information,
655 		 * while device pointer would be NULL.
656 		 * UDP receive path is one user.
657 		 */
658 		unsigned long		dev_scratch;
659 	};
660 	/*
661 	 * This is the control buffer. It is free to use for every
662 	 * layer. Please put your private variables there. If you
663 	 * want to keep them across layers you have to do a skb_clone()
664 	 * first. This is owned by whoever has the skb queued ATM.
665 	 */
666 	char			cb[48] __aligned(8);
667 
668 	unsigned long		_skb_refdst;
669 	void			(*destructor)(struct sk_buff *skb);
670 #ifdef CONFIG_XFRM
671 	struct	sec_path	*sp;
672 #endif
673 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
674 	unsigned long		 _nfct;
675 #endif
676 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
677 	struct nf_bridge_info	*nf_bridge;
678 #endif
679 	unsigned int		len,
680 				data_len;
681 	__u16			mac_len,
682 				hdr_len;
683 
684 	/* Following fields are _not_ copied in __copy_skb_header()
685 	 * Note that queue_mapping is here mostly to fill a hole.
686 	 */
687 	kmemcheck_bitfield_begin(flags1);
688 	__u16			queue_mapping;
689 
690 /* if you move cloned around you also must adapt those constants */
691 #ifdef __BIG_ENDIAN_BITFIELD
692 #define CLONED_MASK	(1 << 7)
693 #else
694 #define CLONED_MASK	1
695 #endif
696 #define CLONED_OFFSET()		offsetof(struct sk_buff, __cloned_offset)
697 
698 	__u8			__cloned_offset[0];
699 	__u8			cloned:1,
700 				nohdr:1,
701 				fclone:2,
702 				peeked:1,
703 				head_frag:1,
704 				xmit_more:1,
705 				__unused:1; /* one bit hole */
706 	kmemcheck_bitfield_end(flags1);
707 
708 	/* fields enclosed in headers_start/headers_end are copied
709 	 * using a single memcpy() in __copy_skb_header()
710 	 */
711 	/* private: */
712 	__u32			headers_start[0];
713 	/* public: */
714 
715 /* if you move pkt_type around you also must adapt those constants */
716 #ifdef __BIG_ENDIAN_BITFIELD
717 #define PKT_TYPE_MAX	(7 << 5)
718 #else
719 #define PKT_TYPE_MAX	7
720 #endif
721 #define PKT_TYPE_OFFSET()	offsetof(struct sk_buff, __pkt_type_offset)
722 
723 	__u8			__pkt_type_offset[0];
724 	__u8			pkt_type:3;
725 	__u8			pfmemalloc:1;
726 	__u8			ignore_df:1;
727 
728 	__u8			nf_trace:1;
729 	__u8			ip_summed:2;
730 	__u8			ooo_okay:1;
731 	__u8			l4_hash:1;
732 	__u8			sw_hash:1;
733 	__u8			wifi_acked_valid:1;
734 	__u8			wifi_acked:1;
735 
736 	__u8			no_fcs:1;
737 	/* Indicates the inner headers are valid in the skbuff. */
738 	__u8			encapsulation:1;
739 	__u8			encap_hdr_csum:1;
740 	__u8			csum_valid:1;
741 	__u8			csum_complete_sw:1;
742 	__u8			csum_level:2;
743 	__u8			csum_bad:1;
744 
745 	__u8			dst_pending_confirm:1;
746 #ifdef CONFIG_IPV6_NDISC_NODETYPE
747 	__u8			ndisc_nodetype:2;
748 #endif
749 	__u8			ipvs_property:1;
750 	__u8			inner_protocol_type:1;
751 	__u8			remcsum_offload:1;
752 #ifdef CONFIG_NET_SWITCHDEV
753 	__u8			offload_fwd_mark:1;
754 #endif
755 #ifdef CONFIG_NET_CLS_ACT
756 	__u8			tc_skip_classify:1;
757 	__u8			tc_at_ingress:1;
758 	__u8			tc_redirected:1;
759 	__u8			tc_from_ingress:1;
760 #endif
761 
762 #ifdef CONFIG_NET_SCHED
763 	__u16			tc_index;	/* traffic control index */
764 #endif
765 
766 	union {
767 		__wsum		csum;
768 		struct {
769 			__u16	csum_start;
770 			__u16	csum_offset;
771 		};
772 	};
773 	__u32			priority;
774 	int			skb_iif;
775 	__u32			hash;
776 	__be16			vlan_proto;
777 	__u16			vlan_tci;
778 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
779 	union {
780 		unsigned int	napi_id;
781 		unsigned int	sender_cpu;
782 	};
783 #endif
784 #ifdef CONFIG_NETWORK_SECMARK
785 	__u32		secmark;
786 #endif
787 
788 	union {
789 		__u32		mark;
790 		__u32		reserved_tailroom;
791 	};
792 
793 	union {
794 		__be16		inner_protocol;
795 		__u8		inner_ipproto;
796 	};
797 
798 	__u16			inner_transport_header;
799 	__u16			inner_network_header;
800 	__u16			inner_mac_header;
801 
802 	__be16			protocol;
803 	__u16			transport_header;
804 	__u16			network_header;
805 	__u16			mac_header;
806 
807 	/* private: */
808 	__u32			headers_end[0];
809 	/* public: */
810 
811 	/* These elements must be at the end, see alloc_skb() for details.  */
812 	sk_buff_data_t		tail;
813 	sk_buff_data_t		end;
814 	unsigned char		*head,
815 				*data;
816 	unsigned int		truesize;
817 	atomic_t		users;
818 };
819 
820 #ifdef __KERNEL__
821 /*
822  *	Handling routines are only of interest to the kernel
823  */
824 #include <linux/slab.h>
825 
826 
827 #define SKB_ALLOC_FCLONE	0x01
828 #define SKB_ALLOC_RX		0x02
829 #define SKB_ALLOC_NAPI		0x04
830 
831 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
832 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
833 {
834 	return unlikely(skb->pfmemalloc);
835 }
836 
837 /*
838  * skb might have a dst pointer attached, refcounted or not.
839  * _skb_refdst low order bit is set if refcount was _not_ taken
840  */
841 #define SKB_DST_NOREF	1UL
842 #define SKB_DST_PTRMASK	~(SKB_DST_NOREF)
843 
844 #define SKB_NFCT_PTRMASK	~(7UL)
845 /**
846  * skb_dst - returns skb dst_entry
847  * @skb: buffer
848  *
849  * Returns skb dst_entry, regardless of reference taken or not.
850  */
851 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
852 {
853 	/* If refdst was not refcounted, check we still are in a
854 	 * rcu_read_lock section
855 	 */
856 	WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
857 		!rcu_read_lock_held() &&
858 		!rcu_read_lock_bh_held());
859 	return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
860 }
861 
862 /**
863  * skb_dst_set - sets skb dst
864  * @skb: buffer
865  * @dst: dst entry
866  *
867  * Sets skb dst, assuming a reference was taken on dst and should
868  * be released by skb_dst_drop()
869  */
870 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
871 {
872 	skb->_skb_refdst = (unsigned long)dst;
873 }
874 
875 /**
876  * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
877  * @skb: buffer
878  * @dst: dst entry
879  *
880  * Sets skb dst, assuming a reference was not taken on dst.
881  * If dst entry is cached, we do not take reference and dst_release
882  * will be avoided by refdst_drop. If dst entry is not cached, we take
883  * reference, so that last dst_release can destroy the dst immediately.
884  */
885 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
886 {
887 	WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
888 	skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
889 }
890 
891 /**
892  * skb_dst_is_noref - Test if skb dst isn't refcounted
893  * @skb: buffer
894  */
895 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
896 {
897 	return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
898 }
899 
900 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
901 {
902 	return (struct rtable *)skb_dst(skb);
903 }
904 
905 /* For mangling skb->pkt_type from user space side from applications
906  * such as nft, tc, etc, we only allow a conservative subset of
907  * possible pkt_types to be set.
908 */
909 static inline bool skb_pkt_type_ok(u32 ptype)
910 {
911 	return ptype <= PACKET_OTHERHOST;
912 }
913 
914 void kfree_skb(struct sk_buff *skb);
915 void kfree_skb_list(struct sk_buff *segs);
916 void skb_tx_error(struct sk_buff *skb);
917 void consume_skb(struct sk_buff *skb);
918 void  __kfree_skb(struct sk_buff *skb);
919 extern struct kmem_cache *skbuff_head_cache;
920 
921 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
922 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
923 		      bool *fragstolen, int *delta_truesize);
924 
925 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
926 			    int node);
927 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
928 struct sk_buff *build_skb(void *data, unsigned int frag_size);
929 static inline struct sk_buff *alloc_skb(unsigned int size,
930 					gfp_t priority)
931 {
932 	return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
933 }
934 
935 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
936 				     unsigned long data_len,
937 				     int max_page_order,
938 				     int *errcode,
939 				     gfp_t gfp_mask);
940 
941 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
942 struct sk_buff_fclones {
943 	struct sk_buff	skb1;
944 
945 	struct sk_buff	skb2;
946 
947 	atomic_t	fclone_ref;
948 };
949 
950 /**
951  *	skb_fclone_busy - check if fclone is busy
952  *	@sk: socket
953  *	@skb: buffer
954  *
955  * Returns true if skb is a fast clone, and its clone is not freed.
956  * Some drivers call skb_orphan() in their ndo_start_xmit(),
957  * so we also check that this didnt happen.
958  */
959 static inline bool skb_fclone_busy(const struct sock *sk,
960 				   const struct sk_buff *skb)
961 {
962 	const struct sk_buff_fclones *fclones;
963 
964 	fclones = container_of(skb, struct sk_buff_fclones, skb1);
965 
966 	return skb->fclone == SKB_FCLONE_ORIG &&
967 	       atomic_read(&fclones->fclone_ref) > 1 &&
968 	       fclones->skb2.sk == sk;
969 }
970 
971 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
972 					       gfp_t priority)
973 {
974 	return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
975 }
976 
977 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
978 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
979 {
980 	return __alloc_skb_head(priority, -1);
981 }
982 
983 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
984 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
985 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
986 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
987 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
988 				   gfp_t gfp_mask, bool fclone);
989 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
990 					  gfp_t gfp_mask)
991 {
992 	return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
993 }
994 
995 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
996 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
997 				     unsigned int headroom);
998 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
999 				int newtailroom, gfp_t priority);
1000 int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1001 			int offset, int len);
1002 int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
1003 		 int len);
1004 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1005 int skb_pad(struct sk_buff *skb, int pad);
1006 #define dev_kfree_skb(a)	consume_skb(a)
1007 
1008 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1009 			    int getfrag(void *from, char *to, int offset,
1010 					int len, int odd, struct sk_buff *skb),
1011 			    void *from, int length);
1012 
1013 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1014 			 int offset, size_t size);
1015 
1016 struct skb_seq_state {
1017 	__u32		lower_offset;
1018 	__u32		upper_offset;
1019 	__u32		frag_idx;
1020 	__u32		stepped_offset;
1021 	struct sk_buff	*root_skb;
1022 	struct sk_buff	*cur_skb;
1023 	__u8		*frag_data;
1024 };
1025 
1026 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1027 			  unsigned int to, struct skb_seq_state *st);
1028 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1029 			  struct skb_seq_state *st);
1030 void skb_abort_seq_read(struct skb_seq_state *st);
1031 
1032 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1033 			   unsigned int to, struct ts_config *config);
1034 
1035 /*
1036  * Packet hash types specify the type of hash in skb_set_hash.
1037  *
1038  * Hash types refer to the protocol layer addresses which are used to
1039  * construct a packet's hash. The hashes are used to differentiate or identify
1040  * flows of the protocol layer for the hash type. Hash types are either
1041  * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1042  *
1043  * Properties of hashes:
1044  *
1045  * 1) Two packets in different flows have different hash values
1046  * 2) Two packets in the same flow should have the same hash value
1047  *
1048  * A hash at a higher layer is considered to be more specific. A driver should
1049  * set the most specific hash possible.
1050  *
1051  * A driver cannot indicate a more specific hash than the layer at which a hash
1052  * was computed. For instance an L3 hash cannot be set as an L4 hash.
1053  *
1054  * A driver may indicate a hash level which is less specific than the
1055  * actual layer the hash was computed on. For instance, a hash computed
1056  * at L4 may be considered an L3 hash. This should only be done if the
1057  * driver can't unambiguously determine that the HW computed the hash at
1058  * the higher layer. Note that the "should" in the second property above
1059  * permits this.
1060  */
1061 enum pkt_hash_types {
1062 	PKT_HASH_TYPE_NONE,	/* Undefined type */
1063 	PKT_HASH_TYPE_L2,	/* Input: src_MAC, dest_MAC */
1064 	PKT_HASH_TYPE_L3,	/* Input: src_IP, dst_IP */
1065 	PKT_HASH_TYPE_L4,	/* Input: src_IP, dst_IP, src_port, dst_port */
1066 };
1067 
1068 static inline void skb_clear_hash(struct sk_buff *skb)
1069 {
1070 	skb->hash = 0;
1071 	skb->sw_hash = 0;
1072 	skb->l4_hash = 0;
1073 }
1074 
1075 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1076 {
1077 	if (!skb->l4_hash)
1078 		skb_clear_hash(skb);
1079 }
1080 
1081 static inline void
1082 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1083 {
1084 	skb->l4_hash = is_l4;
1085 	skb->sw_hash = is_sw;
1086 	skb->hash = hash;
1087 }
1088 
1089 static inline void
1090 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1091 {
1092 	/* Used by drivers to set hash from HW */
1093 	__skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1094 }
1095 
1096 static inline void
1097 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1098 {
1099 	__skb_set_hash(skb, hash, true, is_l4);
1100 }
1101 
1102 void __skb_get_hash(struct sk_buff *skb);
1103 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1104 u32 skb_get_poff(const struct sk_buff *skb);
1105 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1106 		   const struct flow_keys *keys, int hlen);
1107 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1108 			    void *data, int hlen_proto);
1109 
1110 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1111 					int thoff, u8 ip_proto)
1112 {
1113 	return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1114 }
1115 
1116 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1117 			     const struct flow_dissector_key *key,
1118 			     unsigned int key_count);
1119 
1120 bool __skb_flow_dissect(const struct sk_buff *skb,
1121 			struct flow_dissector *flow_dissector,
1122 			void *target_container,
1123 			void *data, __be16 proto, int nhoff, int hlen,
1124 			unsigned int flags);
1125 
1126 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1127 				    struct flow_dissector *flow_dissector,
1128 				    void *target_container, unsigned int flags)
1129 {
1130 	return __skb_flow_dissect(skb, flow_dissector, target_container,
1131 				  NULL, 0, 0, 0, flags);
1132 }
1133 
1134 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1135 					      struct flow_keys *flow,
1136 					      unsigned int flags)
1137 {
1138 	memset(flow, 0, sizeof(*flow));
1139 	return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1140 				  NULL, 0, 0, 0, flags);
1141 }
1142 
1143 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1144 						  void *data, __be16 proto,
1145 						  int nhoff, int hlen,
1146 						  unsigned int flags)
1147 {
1148 	memset(flow, 0, sizeof(*flow));
1149 	return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1150 				  data, proto, nhoff, hlen, flags);
1151 }
1152 
1153 static inline __u32 skb_get_hash(struct sk_buff *skb)
1154 {
1155 	if (!skb->l4_hash && !skb->sw_hash)
1156 		__skb_get_hash(skb);
1157 
1158 	return skb->hash;
1159 }
1160 
1161 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1162 
1163 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1164 {
1165 	if (!skb->l4_hash && !skb->sw_hash) {
1166 		struct flow_keys keys;
1167 		__u32 hash = __get_hash_from_flowi6(fl6, &keys);
1168 
1169 		__skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1170 	}
1171 
1172 	return skb->hash;
1173 }
1174 
1175 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1176 
1177 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1178 {
1179 	if (!skb->l4_hash && !skb->sw_hash) {
1180 		struct flow_keys keys;
1181 		__u32 hash = __get_hash_from_flowi4(fl4, &keys);
1182 
1183 		__skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1184 	}
1185 
1186 	return skb->hash;
1187 }
1188 
1189 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1190 
1191 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1192 {
1193 	return skb->hash;
1194 }
1195 
1196 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1197 {
1198 	to->hash = from->hash;
1199 	to->sw_hash = from->sw_hash;
1200 	to->l4_hash = from->l4_hash;
1201 };
1202 
1203 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1204 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1205 {
1206 	return skb->head + skb->end;
1207 }
1208 
1209 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1210 {
1211 	return skb->end;
1212 }
1213 #else
1214 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1215 {
1216 	return skb->end;
1217 }
1218 
1219 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1220 {
1221 	return skb->end - skb->head;
1222 }
1223 #endif
1224 
1225 /* Internal */
1226 #define skb_shinfo(SKB)	((struct skb_shared_info *)(skb_end_pointer(SKB)))
1227 
1228 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1229 {
1230 	return &skb_shinfo(skb)->hwtstamps;
1231 }
1232 
1233 /**
1234  *	skb_queue_empty - check if a queue is empty
1235  *	@list: queue head
1236  *
1237  *	Returns true if the queue is empty, false otherwise.
1238  */
1239 static inline int skb_queue_empty(const struct sk_buff_head *list)
1240 {
1241 	return list->next == (const struct sk_buff *) list;
1242 }
1243 
1244 /**
1245  *	skb_queue_is_last - check if skb is the last entry in the queue
1246  *	@list: queue head
1247  *	@skb: buffer
1248  *
1249  *	Returns true if @skb is the last buffer on the list.
1250  */
1251 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1252 				     const struct sk_buff *skb)
1253 {
1254 	return skb->next == (const struct sk_buff *) list;
1255 }
1256 
1257 /**
1258  *	skb_queue_is_first - check if skb is the first entry in the queue
1259  *	@list: queue head
1260  *	@skb: buffer
1261  *
1262  *	Returns true if @skb is the first buffer on the list.
1263  */
1264 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1265 				      const struct sk_buff *skb)
1266 {
1267 	return skb->prev == (const struct sk_buff *) list;
1268 }
1269 
1270 /**
1271  *	skb_queue_next - return the next packet in the queue
1272  *	@list: queue head
1273  *	@skb: current buffer
1274  *
1275  *	Return the next packet in @list after @skb.  It is only valid to
1276  *	call this if skb_queue_is_last() evaluates to false.
1277  */
1278 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1279 					     const struct sk_buff *skb)
1280 {
1281 	/* This BUG_ON may seem severe, but if we just return then we
1282 	 * are going to dereference garbage.
1283 	 */
1284 	BUG_ON(skb_queue_is_last(list, skb));
1285 	return skb->next;
1286 }
1287 
1288 /**
1289  *	skb_queue_prev - return the prev packet in the queue
1290  *	@list: queue head
1291  *	@skb: current buffer
1292  *
1293  *	Return the prev packet in @list before @skb.  It is only valid to
1294  *	call this if skb_queue_is_first() evaluates to false.
1295  */
1296 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1297 					     const struct sk_buff *skb)
1298 {
1299 	/* This BUG_ON may seem severe, but if we just return then we
1300 	 * are going to dereference garbage.
1301 	 */
1302 	BUG_ON(skb_queue_is_first(list, skb));
1303 	return skb->prev;
1304 }
1305 
1306 /**
1307  *	skb_get - reference buffer
1308  *	@skb: buffer to reference
1309  *
1310  *	Makes another reference to a socket buffer and returns a pointer
1311  *	to the buffer.
1312  */
1313 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1314 {
1315 	atomic_inc(&skb->users);
1316 	return skb;
1317 }
1318 
1319 /*
1320  * If users == 1, we are the only owner and are can avoid redundant
1321  * atomic change.
1322  */
1323 
1324 /**
1325  *	skb_cloned - is the buffer a clone
1326  *	@skb: buffer to check
1327  *
1328  *	Returns true if the buffer was generated with skb_clone() and is
1329  *	one of multiple shared copies of the buffer. Cloned buffers are
1330  *	shared data so must not be written to under normal circumstances.
1331  */
1332 static inline int skb_cloned(const struct sk_buff *skb)
1333 {
1334 	return skb->cloned &&
1335 	       (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1336 }
1337 
1338 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1339 {
1340 	might_sleep_if(gfpflags_allow_blocking(pri));
1341 
1342 	if (skb_cloned(skb))
1343 		return pskb_expand_head(skb, 0, 0, pri);
1344 
1345 	return 0;
1346 }
1347 
1348 /**
1349  *	skb_header_cloned - is the header a clone
1350  *	@skb: buffer to check
1351  *
1352  *	Returns true if modifying the header part of the buffer requires
1353  *	the data to be copied.
1354  */
1355 static inline int skb_header_cloned(const struct sk_buff *skb)
1356 {
1357 	int dataref;
1358 
1359 	if (!skb->cloned)
1360 		return 0;
1361 
1362 	dataref = atomic_read(&skb_shinfo(skb)->dataref);
1363 	dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1364 	return dataref != 1;
1365 }
1366 
1367 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1368 {
1369 	might_sleep_if(gfpflags_allow_blocking(pri));
1370 
1371 	if (skb_header_cloned(skb))
1372 		return pskb_expand_head(skb, 0, 0, pri);
1373 
1374 	return 0;
1375 }
1376 
1377 /**
1378  *	skb_header_release - release reference to header
1379  *	@skb: buffer to operate on
1380  *
1381  *	Drop a reference to the header part of the buffer.  This is done
1382  *	by acquiring a payload reference.  You must not read from the header
1383  *	part of skb->data after this.
1384  *	Note : Check if you can use __skb_header_release() instead.
1385  */
1386 static inline void skb_header_release(struct sk_buff *skb)
1387 {
1388 	BUG_ON(skb->nohdr);
1389 	skb->nohdr = 1;
1390 	atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1391 }
1392 
1393 /**
1394  *	__skb_header_release - release reference to header
1395  *	@skb: buffer to operate on
1396  *
1397  *	Variant of skb_header_release() assuming skb is private to caller.
1398  *	We can avoid one atomic operation.
1399  */
1400 static inline void __skb_header_release(struct sk_buff *skb)
1401 {
1402 	skb->nohdr = 1;
1403 	atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1404 }
1405 
1406 
1407 /**
1408  *	skb_shared - is the buffer shared
1409  *	@skb: buffer to check
1410  *
1411  *	Returns true if more than one person has a reference to this
1412  *	buffer.
1413  */
1414 static inline int skb_shared(const struct sk_buff *skb)
1415 {
1416 	return atomic_read(&skb->users) != 1;
1417 }
1418 
1419 /**
1420  *	skb_share_check - check if buffer is shared and if so clone it
1421  *	@skb: buffer to check
1422  *	@pri: priority for memory allocation
1423  *
1424  *	If the buffer is shared the buffer is cloned and the old copy
1425  *	drops a reference. A new clone with a single reference is returned.
1426  *	If the buffer is not shared the original buffer is returned. When
1427  *	being called from interrupt status or with spinlocks held pri must
1428  *	be GFP_ATOMIC.
1429  *
1430  *	NULL is returned on a memory allocation failure.
1431  */
1432 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1433 {
1434 	might_sleep_if(gfpflags_allow_blocking(pri));
1435 	if (skb_shared(skb)) {
1436 		struct sk_buff *nskb = skb_clone(skb, pri);
1437 
1438 		if (likely(nskb))
1439 			consume_skb(skb);
1440 		else
1441 			kfree_skb(skb);
1442 		skb = nskb;
1443 	}
1444 	return skb;
1445 }
1446 
1447 /*
1448  *	Copy shared buffers into a new sk_buff. We effectively do COW on
1449  *	packets to handle cases where we have a local reader and forward
1450  *	and a couple of other messy ones. The normal one is tcpdumping
1451  *	a packet thats being forwarded.
1452  */
1453 
1454 /**
1455  *	skb_unshare - make a copy of a shared buffer
1456  *	@skb: buffer to check
1457  *	@pri: priority for memory allocation
1458  *
1459  *	If the socket buffer is a clone then this function creates a new
1460  *	copy of the data, drops a reference count on the old copy and returns
1461  *	the new copy with the reference count at 1. If the buffer is not a clone
1462  *	the original buffer is returned. When called with a spinlock held or
1463  *	from interrupt state @pri must be %GFP_ATOMIC
1464  *
1465  *	%NULL is returned on a memory allocation failure.
1466  */
1467 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1468 					  gfp_t pri)
1469 {
1470 	might_sleep_if(gfpflags_allow_blocking(pri));
1471 	if (skb_cloned(skb)) {
1472 		struct sk_buff *nskb = skb_copy(skb, pri);
1473 
1474 		/* Free our shared copy */
1475 		if (likely(nskb))
1476 			consume_skb(skb);
1477 		else
1478 			kfree_skb(skb);
1479 		skb = nskb;
1480 	}
1481 	return skb;
1482 }
1483 
1484 /**
1485  *	skb_peek - peek at the head of an &sk_buff_head
1486  *	@list_: list to peek at
1487  *
1488  *	Peek an &sk_buff. Unlike most other operations you _MUST_
1489  *	be careful with this one. A peek leaves the buffer on the
1490  *	list and someone else may run off with it. You must hold
1491  *	the appropriate locks or have a private queue to do this.
1492  *
1493  *	Returns %NULL for an empty list or a pointer to the head element.
1494  *	The reference count is not incremented and the reference is therefore
1495  *	volatile. Use with caution.
1496  */
1497 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1498 {
1499 	struct sk_buff *skb = list_->next;
1500 
1501 	if (skb == (struct sk_buff *)list_)
1502 		skb = NULL;
1503 	return skb;
1504 }
1505 
1506 /**
1507  *	skb_peek_next - peek skb following the given one from a queue
1508  *	@skb: skb to start from
1509  *	@list_: list to peek at
1510  *
1511  *	Returns %NULL when the end of the list is met or a pointer to the
1512  *	next element. The reference count is not incremented and the
1513  *	reference is therefore volatile. Use with caution.
1514  */
1515 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1516 		const struct sk_buff_head *list_)
1517 {
1518 	struct sk_buff *next = skb->next;
1519 
1520 	if (next == (struct sk_buff *)list_)
1521 		next = NULL;
1522 	return next;
1523 }
1524 
1525 /**
1526  *	skb_peek_tail - peek at the tail of an &sk_buff_head
1527  *	@list_: list to peek at
1528  *
1529  *	Peek an &sk_buff. Unlike most other operations you _MUST_
1530  *	be careful with this one. A peek leaves the buffer on the
1531  *	list and someone else may run off with it. You must hold
1532  *	the appropriate locks or have a private queue to do this.
1533  *
1534  *	Returns %NULL for an empty list or a pointer to the tail element.
1535  *	The reference count is not incremented and the reference is therefore
1536  *	volatile. Use with caution.
1537  */
1538 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1539 {
1540 	struct sk_buff *skb = list_->prev;
1541 
1542 	if (skb == (struct sk_buff *)list_)
1543 		skb = NULL;
1544 	return skb;
1545 
1546 }
1547 
1548 /**
1549  *	skb_queue_len	- get queue length
1550  *	@list_: list to measure
1551  *
1552  *	Return the length of an &sk_buff queue.
1553  */
1554 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1555 {
1556 	return list_->qlen;
1557 }
1558 
1559 /**
1560  *	__skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1561  *	@list: queue to initialize
1562  *
1563  *	This initializes only the list and queue length aspects of
1564  *	an sk_buff_head object.  This allows to initialize the list
1565  *	aspects of an sk_buff_head without reinitializing things like
1566  *	the spinlock.  It can also be used for on-stack sk_buff_head
1567  *	objects where the spinlock is known to not be used.
1568  */
1569 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1570 {
1571 	list->prev = list->next = (struct sk_buff *)list;
1572 	list->qlen = 0;
1573 }
1574 
1575 /*
1576  * This function creates a split out lock class for each invocation;
1577  * this is needed for now since a whole lot of users of the skb-queue
1578  * infrastructure in drivers have different locking usage (in hardirq)
1579  * than the networking core (in softirq only). In the long run either the
1580  * network layer or drivers should need annotation to consolidate the
1581  * main types of usage into 3 classes.
1582  */
1583 static inline void skb_queue_head_init(struct sk_buff_head *list)
1584 {
1585 	spin_lock_init(&list->lock);
1586 	__skb_queue_head_init(list);
1587 }
1588 
1589 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1590 		struct lock_class_key *class)
1591 {
1592 	skb_queue_head_init(list);
1593 	lockdep_set_class(&list->lock, class);
1594 }
1595 
1596 /*
1597  *	Insert an sk_buff on a list.
1598  *
1599  *	The "__skb_xxxx()" functions are the non-atomic ones that
1600  *	can only be called with interrupts disabled.
1601  */
1602 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1603 		struct sk_buff_head *list);
1604 static inline void __skb_insert(struct sk_buff *newsk,
1605 				struct sk_buff *prev, struct sk_buff *next,
1606 				struct sk_buff_head *list)
1607 {
1608 	newsk->next = next;
1609 	newsk->prev = prev;
1610 	next->prev  = prev->next = newsk;
1611 	list->qlen++;
1612 }
1613 
1614 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1615 				      struct sk_buff *prev,
1616 				      struct sk_buff *next)
1617 {
1618 	struct sk_buff *first = list->next;
1619 	struct sk_buff *last = list->prev;
1620 
1621 	first->prev = prev;
1622 	prev->next = first;
1623 
1624 	last->next = next;
1625 	next->prev = last;
1626 }
1627 
1628 /**
1629  *	skb_queue_splice - join two skb lists, this is designed for stacks
1630  *	@list: the new list to add
1631  *	@head: the place to add it in the first list
1632  */
1633 static inline void skb_queue_splice(const struct sk_buff_head *list,
1634 				    struct sk_buff_head *head)
1635 {
1636 	if (!skb_queue_empty(list)) {
1637 		__skb_queue_splice(list, (struct sk_buff *) head, head->next);
1638 		head->qlen += list->qlen;
1639 	}
1640 }
1641 
1642 /**
1643  *	skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1644  *	@list: the new list to add
1645  *	@head: the place to add it in the first list
1646  *
1647  *	The list at @list is reinitialised
1648  */
1649 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1650 					 struct sk_buff_head *head)
1651 {
1652 	if (!skb_queue_empty(list)) {
1653 		__skb_queue_splice(list, (struct sk_buff *) head, head->next);
1654 		head->qlen += list->qlen;
1655 		__skb_queue_head_init(list);
1656 	}
1657 }
1658 
1659 /**
1660  *	skb_queue_splice_tail - join two skb lists, each list being a queue
1661  *	@list: the new list to add
1662  *	@head: the place to add it in the first list
1663  */
1664 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1665 					 struct sk_buff_head *head)
1666 {
1667 	if (!skb_queue_empty(list)) {
1668 		__skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1669 		head->qlen += list->qlen;
1670 	}
1671 }
1672 
1673 /**
1674  *	skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1675  *	@list: the new list to add
1676  *	@head: the place to add it in the first list
1677  *
1678  *	Each of the lists is a queue.
1679  *	The list at @list is reinitialised
1680  */
1681 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1682 					      struct sk_buff_head *head)
1683 {
1684 	if (!skb_queue_empty(list)) {
1685 		__skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1686 		head->qlen += list->qlen;
1687 		__skb_queue_head_init(list);
1688 	}
1689 }
1690 
1691 /**
1692  *	__skb_queue_after - queue a buffer at the list head
1693  *	@list: list to use
1694  *	@prev: place after this buffer
1695  *	@newsk: buffer to queue
1696  *
1697  *	Queue a buffer int the middle of a list. This function takes no locks
1698  *	and you must therefore hold required locks before calling it.
1699  *
1700  *	A buffer cannot be placed on two lists at the same time.
1701  */
1702 static inline void __skb_queue_after(struct sk_buff_head *list,
1703 				     struct sk_buff *prev,
1704 				     struct sk_buff *newsk)
1705 {
1706 	__skb_insert(newsk, prev, prev->next, list);
1707 }
1708 
1709 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1710 		struct sk_buff_head *list);
1711 
1712 static inline void __skb_queue_before(struct sk_buff_head *list,
1713 				      struct sk_buff *next,
1714 				      struct sk_buff *newsk)
1715 {
1716 	__skb_insert(newsk, next->prev, next, list);
1717 }
1718 
1719 /**
1720  *	__skb_queue_head - queue a buffer at the list head
1721  *	@list: list to use
1722  *	@newsk: buffer to queue
1723  *
1724  *	Queue a buffer at the start of a list. This function takes no locks
1725  *	and you must therefore hold required locks before calling it.
1726  *
1727  *	A buffer cannot be placed on two lists at the same time.
1728  */
1729 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1730 static inline void __skb_queue_head(struct sk_buff_head *list,
1731 				    struct sk_buff *newsk)
1732 {
1733 	__skb_queue_after(list, (struct sk_buff *)list, newsk);
1734 }
1735 
1736 /**
1737  *	__skb_queue_tail - queue a buffer at the list tail
1738  *	@list: list to use
1739  *	@newsk: buffer to queue
1740  *
1741  *	Queue a buffer at the end of a list. This function takes no locks
1742  *	and you must therefore hold required locks before calling it.
1743  *
1744  *	A buffer cannot be placed on two lists at the same time.
1745  */
1746 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1747 static inline void __skb_queue_tail(struct sk_buff_head *list,
1748 				   struct sk_buff *newsk)
1749 {
1750 	__skb_queue_before(list, (struct sk_buff *)list, newsk);
1751 }
1752 
1753 /*
1754  * remove sk_buff from list. _Must_ be called atomically, and with
1755  * the list known..
1756  */
1757 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1758 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1759 {
1760 	struct sk_buff *next, *prev;
1761 
1762 	list->qlen--;
1763 	next	   = skb->next;
1764 	prev	   = skb->prev;
1765 	skb->next  = skb->prev = NULL;
1766 	next->prev = prev;
1767 	prev->next = next;
1768 }
1769 
1770 /**
1771  *	__skb_dequeue - remove from the head of the queue
1772  *	@list: list to dequeue from
1773  *
1774  *	Remove the head of the list. This function does not take any locks
1775  *	so must be used with appropriate locks held only. The head item is
1776  *	returned or %NULL if the list is empty.
1777  */
1778 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1779 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1780 {
1781 	struct sk_buff *skb = skb_peek(list);
1782 	if (skb)
1783 		__skb_unlink(skb, list);
1784 	return skb;
1785 }
1786 
1787 /**
1788  *	__skb_dequeue_tail - remove from the tail of the queue
1789  *	@list: list to dequeue from
1790  *
1791  *	Remove the tail of the list. This function does not take any locks
1792  *	so must be used with appropriate locks held only. The tail item is
1793  *	returned or %NULL if the list is empty.
1794  */
1795 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1796 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1797 {
1798 	struct sk_buff *skb = skb_peek_tail(list);
1799 	if (skb)
1800 		__skb_unlink(skb, list);
1801 	return skb;
1802 }
1803 
1804 
1805 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1806 {
1807 	return skb->data_len;
1808 }
1809 
1810 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1811 {
1812 	return skb->len - skb->data_len;
1813 }
1814 
1815 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1816 {
1817 	unsigned int i, len = 0;
1818 
1819 	for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1820 		len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1821 	return len + skb_headlen(skb);
1822 }
1823 
1824 /**
1825  * __skb_fill_page_desc - initialise a paged fragment in an skb
1826  * @skb: buffer containing fragment to be initialised
1827  * @i: paged fragment index to initialise
1828  * @page: the page to use for this fragment
1829  * @off: the offset to the data with @page
1830  * @size: the length of the data
1831  *
1832  * Initialises the @i'th fragment of @skb to point to &size bytes at
1833  * offset @off within @page.
1834  *
1835  * Does not take any additional reference on the fragment.
1836  */
1837 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1838 					struct page *page, int off, int size)
1839 {
1840 	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1841 
1842 	/*
1843 	 * Propagate page pfmemalloc to the skb if we can. The problem is
1844 	 * that not all callers have unique ownership of the page but rely
1845 	 * on page_is_pfmemalloc doing the right thing(tm).
1846 	 */
1847 	frag->page.p		  = page;
1848 	frag->page_offset	  = off;
1849 	skb_frag_size_set(frag, size);
1850 
1851 	page = compound_head(page);
1852 	if (page_is_pfmemalloc(page))
1853 		skb->pfmemalloc	= true;
1854 }
1855 
1856 /**
1857  * skb_fill_page_desc - initialise a paged fragment in an skb
1858  * @skb: buffer containing fragment to be initialised
1859  * @i: paged fragment index to initialise
1860  * @page: the page to use for this fragment
1861  * @off: the offset to the data with @page
1862  * @size: the length of the data
1863  *
1864  * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1865  * @skb to point to @size bytes at offset @off within @page. In
1866  * addition updates @skb such that @i is the last fragment.
1867  *
1868  * Does not take any additional reference on the fragment.
1869  */
1870 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1871 				      struct page *page, int off, int size)
1872 {
1873 	__skb_fill_page_desc(skb, i, page, off, size);
1874 	skb_shinfo(skb)->nr_frags = i + 1;
1875 }
1876 
1877 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1878 		     int size, unsigned int truesize);
1879 
1880 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1881 			  unsigned int truesize);
1882 
1883 #define SKB_PAGE_ASSERT(skb) 	BUG_ON(skb_shinfo(skb)->nr_frags)
1884 #define SKB_FRAG_ASSERT(skb) 	BUG_ON(skb_has_frag_list(skb))
1885 #define SKB_LINEAR_ASSERT(skb)  BUG_ON(skb_is_nonlinear(skb))
1886 
1887 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1888 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1889 {
1890 	return skb->head + skb->tail;
1891 }
1892 
1893 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1894 {
1895 	skb->tail = skb->data - skb->head;
1896 }
1897 
1898 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1899 {
1900 	skb_reset_tail_pointer(skb);
1901 	skb->tail += offset;
1902 }
1903 
1904 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1905 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1906 {
1907 	return skb->tail;
1908 }
1909 
1910 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1911 {
1912 	skb->tail = skb->data;
1913 }
1914 
1915 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1916 {
1917 	skb->tail = skb->data + offset;
1918 }
1919 
1920 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1921 
1922 /*
1923  *	Add data to an sk_buff
1924  */
1925 unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1926 unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1927 static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1928 {
1929 	unsigned char *tmp = skb_tail_pointer(skb);
1930 	SKB_LINEAR_ASSERT(skb);
1931 	skb->tail += len;
1932 	skb->len  += len;
1933 	return tmp;
1934 }
1935 
1936 unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1937 static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1938 {
1939 	skb->data -= len;
1940 	skb->len  += len;
1941 	return skb->data;
1942 }
1943 
1944 unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1945 static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1946 {
1947 	skb->len -= len;
1948 	BUG_ON(skb->len < skb->data_len);
1949 	return skb->data += len;
1950 }
1951 
1952 static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1953 {
1954 	return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1955 }
1956 
1957 unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1958 
1959 static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1960 {
1961 	if (len > skb_headlen(skb) &&
1962 	    !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1963 		return NULL;
1964 	skb->len -= len;
1965 	return skb->data += len;
1966 }
1967 
1968 static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1969 {
1970 	return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1971 }
1972 
1973 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1974 {
1975 	if (likely(len <= skb_headlen(skb)))
1976 		return 1;
1977 	if (unlikely(len > skb->len))
1978 		return 0;
1979 	return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1980 }
1981 
1982 void skb_condense(struct sk_buff *skb);
1983 
1984 /**
1985  *	skb_headroom - bytes at buffer head
1986  *	@skb: buffer to check
1987  *
1988  *	Return the number of bytes of free space at the head of an &sk_buff.
1989  */
1990 static inline unsigned int skb_headroom(const struct sk_buff *skb)
1991 {
1992 	return skb->data - skb->head;
1993 }
1994 
1995 /**
1996  *	skb_tailroom - bytes at buffer end
1997  *	@skb: buffer to check
1998  *
1999  *	Return the number of bytes of free space at the tail of an sk_buff
2000  */
2001 static inline int skb_tailroom(const struct sk_buff *skb)
2002 {
2003 	return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2004 }
2005 
2006 /**
2007  *	skb_availroom - bytes at buffer end
2008  *	@skb: buffer to check
2009  *
2010  *	Return the number of bytes of free space at the tail of an sk_buff
2011  *	allocated by sk_stream_alloc()
2012  */
2013 static inline int skb_availroom(const struct sk_buff *skb)
2014 {
2015 	if (skb_is_nonlinear(skb))
2016 		return 0;
2017 
2018 	return skb->end - skb->tail - skb->reserved_tailroom;
2019 }
2020 
2021 /**
2022  *	skb_reserve - adjust headroom
2023  *	@skb: buffer to alter
2024  *	@len: bytes to move
2025  *
2026  *	Increase the headroom of an empty &sk_buff by reducing the tail
2027  *	room. This is only allowed for an empty buffer.
2028  */
2029 static inline void skb_reserve(struct sk_buff *skb, int len)
2030 {
2031 	skb->data += len;
2032 	skb->tail += len;
2033 }
2034 
2035 /**
2036  *	skb_tailroom_reserve - adjust reserved_tailroom
2037  *	@skb: buffer to alter
2038  *	@mtu: maximum amount of headlen permitted
2039  *	@needed_tailroom: minimum amount of reserved_tailroom
2040  *
2041  *	Set reserved_tailroom so that headlen can be as large as possible but
2042  *	not larger than mtu and tailroom cannot be smaller than
2043  *	needed_tailroom.
2044  *	The required headroom should already have been reserved before using
2045  *	this function.
2046  */
2047 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2048 					unsigned int needed_tailroom)
2049 {
2050 	SKB_LINEAR_ASSERT(skb);
2051 	if (mtu < skb_tailroom(skb) - needed_tailroom)
2052 		/* use at most mtu */
2053 		skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2054 	else
2055 		/* use up to all available space */
2056 		skb->reserved_tailroom = needed_tailroom;
2057 }
2058 
2059 #define ENCAP_TYPE_ETHER	0
2060 #define ENCAP_TYPE_IPPROTO	1
2061 
2062 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2063 					  __be16 protocol)
2064 {
2065 	skb->inner_protocol = protocol;
2066 	skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2067 }
2068 
2069 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2070 					 __u8 ipproto)
2071 {
2072 	skb->inner_ipproto = ipproto;
2073 	skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2074 }
2075 
2076 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2077 {
2078 	skb->inner_mac_header = skb->mac_header;
2079 	skb->inner_network_header = skb->network_header;
2080 	skb->inner_transport_header = skb->transport_header;
2081 }
2082 
2083 static inline void skb_reset_mac_len(struct sk_buff *skb)
2084 {
2085 	skb->mac_len = skb->network_header - skb->mac_header;
2086 }
2087 
2088 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2089 							*skb)
2090 {
2091 	return skb->head + skb->inner_transport_header;
2092 }
2093 
2094 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2095 {
2096 	return skb_inner_transport_header(skb) - skb->data;
2097 }
2098 
2099 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2100 {
2101 	skb->inner_transport_header = skb->data - skb->head;
2102 }
2103 
2104 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2105 						   const int offset)
2106 {
2107 	skb_reset_inner_transport_header(skb);
2108 	skb->inner_transport_header += offset;
2109 }
2110 
2111 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2112 {
2113 	return skb->head + skb->inner_network_header;
2114 }
2115 
2116 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2117 {
2118 	skb->inner_network_header = skb->data - skb->head;
2119 }
2120 
2121 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2122 						const int offset)
2123 {
2124 	skb_reset_inner_network_header(skb);
2125 	skb->inner_network_header += offset;
2126 }
2127 
2128 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2129 {
2130 	return skb->head + skb->inner_mac_header;
2131 }
2132 
2133 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2134 {
2135 	skb->inner_mac_header = skb->data - skb->head;
2136 }
2137 
2138 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2139 					    const int offset)
2140 {
2141 	skb_reset_inner_mac_header(skb);
2142 	skb->inner_mac_header += offset;
2143 }
2144 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2145 {
2146 	return skb->transport_header != (typeof(skb->transport_header))~0U;
2147 }
2148 
2149 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2150 {
2151 	return skb->head + skb->transport_header;
2152 }
2153 
2154 static inline void skb_reset_transport_header(struct sk_buff *skb)
2155 {
2156 	skb->transport_header = skb->data - skb->head;
2157 }
2158 
2159 static inline void skb_set_transport_header(struct sk_buff *skb,
2160 					    const int offset)
2161 {
2162 	skb_reset_transport_header(skb);
2163 	skb->transport_header += offset;
2164 }
2165 
2166 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2167 {
2168 	return skb->head + skb->network_header;
2169 }
2170 
2171 static inline void skb_reset_network_header(struct sk_buff *skb)
2172 {
2173 	skb->network_header = skb->data - skb->head;
2174 }
2175 
2176 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2177 {
2178 	skb_reset_network_header(skb);
2179 	skb->network_header += offset;
2180 }
2181 
2182 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2183 {
2184 	return skb->head + skb->mac_header;
2185 }
2186 
2187 static inline int skb_mac_offset(const struct sk_buff *skb)
2188 {
2189 	return skb_mac_header(skb) - skb->data;
2190 }
2191 
2192 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2193 {
2194 	return skb->mac_header != (typeof(skb->mac_header))~0U;
2195 }
2196 
2197 static inline void skb_reset_mac_header(struct sk_buff *skb)
2198 {
2199 	skb->mac_header = skb->data - skb->head;
2200 }
2201 
2202 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2203 {
2204 	skb_reset_mac_header(skb);
2205 	skb->mac_header += offset;
2206 }
2207 
2208 static inline void skb_pop_mac_header(struct sk_buff *skb)
2209 {
2210 	skb->mac_header = skb->network_header;
2211 }
2212 
2213 static inline void skb_probe_transport_header(struct sk_buff *skb,
2214 					      const int offset_hint)
2215 {
2216 	struct flow_keys keys;
2217 
2218 	if (skb_transport_header_was_set(skb))
2219 		return;
2220 	else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2221 		skb_set_transport_header(skb, keys.control.thoff);
2222 	else
2223 		skb_set_transport_header(skb, offset_hint);
2224 }
2225 
2226 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2227 {
2228 	if (skb_mac_header_was_set(skb)) {
2229 		const unsigned char *old_mac = skb_mac_header(skb);
2230 
2231 		skb_set_mac_header(skb, -skb->mac_len);
2232 		memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2233 	}
2234 }
2235 
2236 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2237 {
2238 	return skb->csum_start - skb_headroom(skb);
2239 }
2240 
2241 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2242 {
2243 	return skb->head + skb->csum_start;
2244 }
2245 
2246 static inline int skb_transport_offset(const struct sk_buff *skb)
2247 {
2248 	return skb_transport_header(skb) - skb->data;
2249 }
2250 
2251 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2252 {
2253 	return skb->transport_header - skb->network_header;
2254 }
2255 
2256 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2257 {
2258 	return skb->inner_transport_header - skb->inner_network_header;
2259 }
2260 
2261 static inline int skb_network_offset(const struct sk_buff *skb)
2262 {
2263 	return skb_network_header(skb) - skb->data;
2264 }
2265 
2266 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2267 {
2268 	return skb_inner_network_header(skb) - skb->data;
2269 }
2270 
2271 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2272 {
2273 	return pskb_may_pull(skb, skb_network_offset(skb) + len);
2274 }
2275 
2276 /*
2277  * CPUs often take a performance hit when accessing unaligned memory
2278  * locations. The actual performance hit varies, it can be small if the
2279  * hardware handles it or large if we have to take an exception and fix it
2280  * in software.
2281  *
2282  * Since an ethernet header is 14 bytes network drivers often end up with
2283  * the IP header at an unaligned offset. The IP header can be aligned by
2284  * shifting the start of the packet by 2 bytes. Drivers should do this
2285  * with:
2286  *
2287  * skb_reserve(skb, NET_IP_ALIGN);
2288  *
2289  * The downside to this alignment of the IP header is that the DMA is now
2290  * unaligned. On some architectures the cost of an unaligned DMA is high
2291  * and this cost outweighs the gains made by aligning the IP header.
2292  *
2293  * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2294  * to be overridden.
2295  */
2296 #ifndef NET_IP_ALIGN
2297 #define NET_IP_ALIGN	2
2298 #endif
2299 
2300 /*
2301  * The networking layer reserves some headroom in skb data (via
2302  * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2303  * the header has to grow. In the default case, if the header has to grow
2304  * 32 bytes or less we avoid the reallocation.
2305  *
2306  * Unfortunately this headroom changes the DMA alignment of the resulting
2307  * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2308  * on some architectures. An architecture can override this value,
2309  * perhaps setting it to a cacheline in size (since that will maintain
2310  * cacheline alignment of the DMA). It must be a power of 2.
2311  *
2312  * Various parts of the networking layer expect at least 32 bytes of
2313  * headroom, you should not reduce this.
2314  *
2315  * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2316  * to reduce average number of cache lines per packet.
2317  * get_rps_cpus() for example only access one 64 bytes aligned block :
2318  * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2319  */
2320 #ifndef NET_SKB_PAD
2321 #define NET_SKB_PAD	max(32, L1_CACHE_BYTES)
2322 #endif
2323 
2324 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2325 
2326 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2327 {
2328 	if (unlikely(skb_is_nonlinear(skb))) {
2329 		WARN_ON(1);
2330 		return;
2331 	}
2332 	skb->len = len;
2333 	skb_set_tail_pointer(skb, len);
2334 }
2335 
2336 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2337 {
2338 	__skb_set_length(skb, len);
2339 }
2340 
2341 void skb_trim(struct sk_buff *skb, unsigned int len);
2342 
2343 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2344 {
2345 	if (skb->data_len)
2346 		return ___pskb_trim(skb, len);
2347 	__skb_trim(skb, len);
2348 	return 0;
2349 }
2350 
2351 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2352 {
2353 	return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2354 }
2355 
2356 /**
2357  *	pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2358  *	@skb: buffer to alter
2359  *	@len: new length
2360  *
2361  *	This is identical to pskb_trim except that the caller knows that
2362  *	the skb is not cloned so we should never get an error due to out-
2363  *	of-memory.
2364  */
2365 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2366 {
2367 	int err = pskb_trim(skb, len);
2368 	BUG_ON(err);
2369 }
2370 
2371 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2372 {
2373 	unsigned int diff = len - skb->len;
2374 
2375 	if (skb_tailroom(skb) < diff) {
2376 		int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2377 					   GFP_ATOMIC);
2378 		if (ret)
2379 			return ret;
2380 	}
2381 	__skb_set_length(skb, len);
2382 	return 0;
2383 }
2384 
2385 /**
2386  *	skb_orphan - orphan a buffer
2387  *	@skb: buffer to orphan
2388  *
2389  *	If a buffer currently has an owner then we call the owner's
2390  *	destructor function and make the @skb unowned. The buffer continues
2391  *	to exist but is no longer charged to its former owner.
2392  */
2393 static inline void skb_orphan(struct sk_buff *skb)
2394 {
2395 	if (skb->destructor) {
2396 		skb->destructor(skb);
2397 		skb->destructor = NULL;
2398 		skb->sk		= NULL;
2399 	} else {
2400 		BUG_ON(skb->sk);
2401 	}
2402 }
2403 
2404 /**
2405  *	skb_orphan_frags - orphan the frags contained in a buffer
2406  *	@skb: buffer to orphan frags from
2407  *	@gfp_mask: allocation mask for replacement pages
2408  *
2409  *	For each frag in the SKB which needs a destructor (i.e. has an
2410  *	owner) create a copy of that frag and release the original
2411  *	page by calling the destructor.
2412  */
2413 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2414 {
2415 	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2416 		return 0;
2417 	return skb_copy_ubufs(skb, gfp_mask);
2418 }
2419 
2420 /**
2421  *	__skb_queue_purge - empty a list
2422  *	@list: list to empty
2423  *
2424  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
2425  *	the list and one reference dropped. This function does not take the
2426  *	list lock and the caller must hold the relevant locks to use it.
2427  */
2428 void skb_queue_purge(struct sk_buff_head *list);
2429 static inline void __skb_queue_purge(struct sk_buff_head *list)
2430 {
2431 	struct sk_buff *skb;
2432 	while ((skb = __skb_dequeue(list)) != NULL)
2433 		kfree_skb(skb);
2434 }
2435 
2436 void skb_rbtree_purge(struct rb_root *root);
2437 
2438 void *netdev_alloc_frag(unsigned int fragsz);
2439 
2440 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2441 				   gfp_t gfp_mask);
2442 
2443 /**
2444  *	netdev_alloc_skb - allocate an skbuff for rx on a specific device
2445  *	@dev: network device to receive on
2446  *	@length: length to allocate
2447  *
2448  *	Allocate a new &sk_buff and assign it a usage count of one. The
2449  *	buffer has unspecified headroom built in. Users should allocate
2450  *	the headroom they think they need without accounting for the
2451  *	built in space. The built in space is used for optimisations.
2452  *
2453  *	%NULL is returned if there is no free memory. Although this function
2454  *	allocates memory it can be called from an interrupt.
2455  */
2456 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2457 					       unsigned int length)
2458 {
2459 	return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2460 }
2461 
2462 /* legacy helper around __netdev_alloc_skb() */
2463 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2464 					      gfp_t gfp_mask)
2465 {
2466 	return __netdev_alloc_skb(NULL, length, gfp_mask);
2467 }
2468 
2469 /* legacy helper around netdev_alloc_skb() */
2470 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2471 {
2472 	return netdev_alloc_skb(NULL, length);
2473 }
2474 
2475 
2476 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2477 		unsigned int length, gfp_t gfp)
2478 {
2479 	struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2480 
2481 	if (NET_IP_ALIGN && skb)
2482 		skb_reserve(skb, NET_IP_ALIGN);
2483 	return skb;
2484 }
2485 
2486 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2487 		unsigned int length)
2488 {
2489 	return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2490 }
2491 
2492 static inline void skb_free_frag(void *addr)
2493 {
2494 	page_frag_free(addr);
2495 }
2496 
2497 void *napi_alloc_frag(unsigned int fragsz);
2498 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2499 				 unsigned int length, gfp_t gfp_mask);
2500 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2501 					     unsigned int length)
2502 {
2503 	return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2504 }
2505 void napi_consume_skb(struct sk_buff *skb, int budget);
2506 
2507 void __kfree_skb_flush(void);
2508 void __kfree_skb_defer(struct sk_buff *skb);
2509 
2510 /**
2511  * __dev_alloc_pages - allocate page for network Rx
2512  * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2513  * @order: size of the allocation
2514  *
2515  * Allocate a new page.
2516  *
2517  * %NULL is returned if there is no free memory.
2518 */
2519 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2520 					     unsigned int order)
2521 {
2522 	/* This piece of code contains several assumptions.
2523 	 * 1.  This is for device Rx, therefor a cold page is preferred.
2524 	 * 2.  The expectation is the user wants a compound page.
2525 	 * 3.  If requesting a order 0 page it will not be compound
2526 	 *     due to the check to see if order has a value in prep_new_page
2527 	 * 4.  __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2528 	 *     code in gfp_to_alloc_flags that should be enforcing this.
2529 	 */
2530 	gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2531 
2532 	return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2533 }
2534 
2535 static inline struct page *dev_alloc_pages(unsigned int order)
2536 {
2537 	return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2538 }
2539 
2540 /**
2541  * __dev_alloc_page - allocate a page for network Rx
2542  * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2543  *
2544  * Allocate a new page.
2545  *
2546  * %NULL is returned if there is no free memory.
2547  */
2548 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2549 {
2550 	return __dev_alloc_pages(gfp_mask, 0);
2551 }
2552 
2553 static inline struct page *dev_alloc_page(void)
2554 {
2555 	return dev_alloc_pages(0);
2556 }
2557 
2558 /**
2559  *	skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2560  *	@page: The page that was allocated from skb_alloc_page
2561  *	@skb: The skb that may need pfmemalloc set
2562  */
2563 static inline void skb_propagate_pfmemalloc(struct page *page,
2564 					     struct sk_buff *skb)
2565 {
2566 	if (page_is_pfmemalloc(page))
2567 		skb->pfmemalloc = true;
2568 }
2569 
2570 /**
2571  * skb_frag_page - retrieve the page referred to by a paged fragment
2572  * @frag: the paged fragment
2573  *
2574  * Returns the &struct page associated with @frag.
2575  */
2576 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2577 {
2578 	return frag->page.p;
2579 }
2580 
2581 /**
2582  * __skb_frag_ref - take an addition reference on a paged fragment.
2583  * @frag: the paged fragment
2584  *
2585  * Takes an additional reference on the paged fragment @frag.
2586  */
2587 static inline void __skb_frag_ref(skb_frag_t *frag)
2588 {
2589 	get_page(skb_frag_page(frag));
2590 }
2591 
2592 /**
2593  * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2594  * @skb: the buffer
2595  * @f: the fragment offset.
2596  *
2597  * Takes an additional reference on the @f'th paged fragment of @skb.
2598  */
2599 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2600 {
2601 	__skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2602 }
2603 
2604 /**
2605  * __skb_frag_unref - release a reference on a paged fragment.
2606  * @frag: the paged fragment
2607  *
2608  * Releases a reference on the paged fragment @frag.
2609  */
2610 static inline void __skb_frag_unref(skb_frag_t *frag)
2611 {
2612 	put_page(skb_frag_page(frag));
2613 }
2614 
2615 /**
2616  * skb_frag_unref - release a reference on a paged fragment of an skb.
2617  * @skb: the buffer
2618  * @f: the fragment offset
2619  *
2620  * Releases a reference on the @f'th paged fragment of @skb.
2621  */
2622 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2623 {
2624 	__skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2625 }
2626 
2627 /**
2628  * skb_frag_address - gets the address of the data contained in a paged fragment
2629  * @frag: the paged fragment buffer
2630  *
2631  * Returns the address of the data within @frag. The page must already
2632  * be mapped.
2633  */
2634 static inline void *skb_frag_address(const skb_frag_t *frag)
2635 {
2636 	return page_address(skb_frag_page(frag)) + frag->page_offset;
2637 }
2638 
2639 /**
2640  * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2641  * @frag: the paged fragment buffer
2642  *
2643  * Returns the address of the data within @frag. Checks that the page
2644  * is mapped and returns %NULL otherwise.
2645  */
2646 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2647 {
2648 	void *ptr = page_address(skb_frag_page(frag));
2649 	if (unlikely(!ptr))
2650 		return NULL;
2651 
2652 	return ptr + frag->page_offset;
2653 }
2654 
2655 /**
2656  * __skb_frag_set_page - sets the page contained in a paged fragment
2657  * @frag: the paged fragment
2658  * @page: the page to set
2659  *
2660  * Sets the fragment @frag to contain @page.
2661  */
2662 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2663 {
2664 	frag->page.p = page;
2665 }
2666 
2667 /**
2668  * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2669  * @skb: the buffer
2670  * @f: the fragment offset
2671  * @page: the page to set
2672  *
2673  * Sets the @f'th fragment of @skb to contain @page.
2674  */
2675 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2676 				     struct page *page)
2677 {
2678 	__skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2679 }
2680 
2681 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2682 
2683 /**
2684  * skb_frag_dma_map - maps a paged fragment via the DMA API
2685  * @dev: the device to map the fragment to
2686  * @frag: the paged fragment to map
2687  * @offset: the offset within the fragment (starting at the
2688  *          fragment's own offset)
2689  * @size: the number of bytes to map
2690  * @dir: the direction of the mapping (%PCI_DMA_*)
2691  *
2692  * Maps the page associated with @frag to @device.
2693  */
2694 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2695 					  const skb_frag_t *frag,
2696 					  size_t offset, size_t size,
2697 					  enum dma_data_direction dir)
2698 {
2699 	return dma_map_page(dev, skb_frag_page(frag),
2700 			    frag->page_offset + offset, size, dir);
2701 }
2702 
2703 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2704 					gfp_t gfp_mask)
2705 {
2706 	return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2707 }
2708 
2709 
2710 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2711 						  gfp_t gfp_mask)
2712 {
2713 	return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2714 }
2715 
2716 
2717 /**
2718  *	skb_clone_writable - is the header of a clone writable
2719  *	@skb: buffer to check
2720  *	@len: length up to which to write
2721  *
2722  *	Returns true if modifying the header part of the cloned buffer
2723  *	does not requires the data to be copied.
2724  */
2725 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2726 {
2727 	return !skb_header_cloned(skb) &&
2728 	       skb_headroom(skb) + len <= skb->hdr_len;
2729 }
2730 
2731 static inline int skb_try_make_writable(struct sk_buff *skb,
2732 					unsigned int write_len)
2733 {
2734 	return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2735 	       pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2736 }
2737 
2738 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2739 			    int cloned)
2740 {
2741 	int delta = 0;
2742 
2743 	if (headroom > skb_headroom(skb))
2744 		delta = headroom - skb_headroom(skb);
2745 
2746 	if (delta || cloned)
2747 		return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2748 					GFP_ATOMIC);
2749 	return 0;
2750 }
2751 
2752 /**
2753  *	skb_cow - copy header of skb when it is required
2754  *	@skb: buffer to cow
2755  *	@headroom: needed headroom
2756  *
2757  *	If the skb passed lacks sufficient headroom or its data part
2758  *	is shared, data is reallocated. If reallocation fails, an error
2759  *	is returned and original skb is not changed.
2760  *
2761  *	The result is skb with writable area skb->head...skb->tail
2762  *	and at least @headroom of space at head.
2763  */
2764 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2765 {
2766 	return __skb_cow(skb, headroom, skb_cloned(skb));
2767 }
2768 
2769 /**
2770  *	skb_cow_head - skb_cow but only making the head writable
2771  *	@skb: buffer to cow
2772  *	@headroom: needed headroom
2773  *
2774  *	This function is identical to skb_cow except that we replace the
2775  *	skb_cloned check by skb_header_cloned.  It should be used when
2776  *	you only need to push on some header and do not need to modify
2777  *	the data.
2778  */
2779 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2780 {
2781 	return __skb_cow(skb, headroom, skb_header_cloned(skb));
2782 }
2783 
2784 /**
2785  *	skb_padto	- pad an skbuff up to a minimal size
2786  *	@skb: buffer to pad
2787  *	@len: minimal length
2788  *
2789  *	Pads up a buffer to ensure the trailing bytes exist and are
2790  *	blanked. If the buffer already contains sufficient data it
2791  *	is untouched. Otherwise it is extended. Returns zero on
2792  *	success. The skb is freed on error.
2793  */
2794 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2795 {
2796 	unsigned int size = skb->len;
2797 	if (likely(size >= len))
2798 		return 0;
2799 	return skb_pad(skb, len - size);
2800 }
2801 
2802 /**
2803  *	skb_put_padto - increase size and pad an skbuff up to a minimal size
2804  *	@skb: buffer to pad
2805  *	@len: minimal length
2806  *
2807  *	Pads up a buffer to ensure the trailing bytes exist and are
2808  *	blanked. If the buffer already contains sufficient data it
2809  *	is untouched. Otherwise it is extended. Returns zero on
2810  *	success. The skb is freed on error.
2811  */
2812 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2813 {
2814 	unsigned int size = skb->len;
2815 
2816 	if (unlikely(size < len)) {
2817 		len -= size;
2818 		if (skb_pad(skb, len))
2819 			return -ENOMEM;
2820 		__skb_put(skb, len);
2821 	}
2822 	return 0;
2823 }
2824 
2825 static inline int skb_add_data(struct sk_buff *skb,
2826 			       struct iov_iter *from, int copy)
2827 {
2828 	const int off = skb->len;
2829 
2830 	if (skb->ip_summed == CHECKSUM_NONE) {
2831 		__wsum csum = 0;
2832 		if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2833 					         &csum, from)) {
2834 			skb->csum = csum_block_add(skb->csum, csum, off);
2835 			return 0;
2836 		}
2837 	} else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
2838 		return 0;
2839 
2840 	__skb_trim(skb, off);
2841 	return -EFAULT;
2842 }
2843 
2844 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2845 				    const struct page *page, int off)
2846 {
2847 	if (i) {
2848 		const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2849 
2850 		return page == skb_frag_page(frag) &&
2851 		       off == frag->page_offset + skb_frag_size(frag);
2852 	}
2853 	return false;
2854 }
2855 
2856 static inline int __skb_linearize(struct sk_buff *skb)
2857 {
2858 	return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2859 }
2860 
2861 /**
2862  *	skb_linearize - convert paged skb to linear one
2863  *	@skb: buffer to linarize
2864  *
2865  *	If there is no free memory -ENOMEM is returned, otherwise zero
2866  *	is returned and the old skb data released.
2867  */
2868 static inline int skb_linearize(struct sk_buff *skb)
2869 {
2870 	return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2871 }
2872 
2873 /**
2874  * skb_has_shared_frag - can any frag be overwritten
2875  * @skb: buffer to test
2876  *
2877  * Return true if the skb has at least one frag that might be modified
2878  * by an external entity (as in vmsplice()/sendfile())
2879  */
2880 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2881 {
2882 	return skb_is_nonlinear(skb) &&
2883 	       skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2884 }
2885 
2886 /**
2887  *	skb_linearize_cow - make sure skb is linear and writable
2888  *	@skb: buffer to process
2889  *
2890  *	If there is no free memory -ENOMEM is returned, otherwise zero
2891  *	is returned and the old skb data released.
2892  */
2893 static inline int skb_linearize_cow(struct sk_buff *skb)
2894 {
2895 	return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2896 	       __skb_linearize(skb) : 0;
2897 }
2898 
2899 static __always_inline void
2900 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2901 		     unsigned int off)
2902 {
2903 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2904 		skb->csum = csum_block_sub(skb->csum,
2905 					   csum_partial(start, len, 0), off);
2906 	else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2907 		 skb_checksum_start_offset(skb) < 0)
2908 		skb->ip_summed = CHECKSUM_NONE;
2909 }
2910 
2911 /**
2912  *	skb_postpull_rcsum - update checksum for received skb after pull
2913  *	@skb: buffer to update
2914  *	@start: start of data before pull
2915  *	@len: length of data pulled
2916  *
2917  *	After doing a pull on a received packet, you need to call this to
2918  *	update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2919  *	CHECKSUM_NONE so that it can be recomputed from scratch.
2920  */
2921 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2922 				      const void *start, unsigned int len)
2923 {
2924 	__skb_postpull_rcsum(skb, start, len, 0);
2925 }
2926 
2927 static __always_inline void
2928 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2929 		     unsigned int off)
2930 {
2931 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2932 		skb->csum = csum_block_add(skb->csum,
2933 					   csum_partial(start, len, 0), off);
2934 }
2935 
2936 /**
2937  *	skb_postpush_rcsum - update checksum for received skb after push
2938  *	@skb: buffer to update
2939  *	@start: start of data after push
2940  *	@len: length of data pushed
2941  *
2942  *	After doing a push on a received packet, you need to call this to
2943  *	update the CHECKSUM_COMPLETE checksum.
2944  */
2945 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2946 				      const void *start, unsigned int len)
2947 {
2948 	__skb_postpush_rcsum(skb, start, len, 0);
2949 }
2950 
2951 unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2952 
2953 /**
2954  *	skb_push_rcsum - push skb and update receive checksum
2955  *	@skb: buffer to update
2956  *	@len: length of data pulled
2957  *
2958  *	This function performs an skb_push on the packet and updates
2959  *	the CHECKSUM_COMPLETE checksum.  It should be used on
2960  *	receive path processing instead of skb_push unless you know
2961  *	that the checksum difference is zero (e.g., a valid IP header)
2962  *	or you are setting ip_summed to CHECKSUM_NONE.
2963  */
2964 static inline unsigned char *skb_push_rcsum(struct sk_buff *skb,
2965 					    unsigned int len)
2966 {
2967 	skb_push(skb, len);
2968 	skb_postpush_rcsum(skb, skb->data, len);
2969 	return skb->data;
2970 }
2971 
2972 /**
2973  *	pskb_trim_rcsum - trim received skb and update checksum
2974  *	@skb: buffer to trim
2975  *	@len: new length
2976  *
2977  *	This is exactly the same as pskb_trim except that it ensures the
2978  *	checksum of received packets are still valid after the operation.
2979  */
2980 
2981 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2982 {
2983 	if (likely(len >= skb->len))
2984 		return 0;
2985 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2986 		skb->ip_summed = CHECKSUM_NONE;
2987 	return __pskb_trim(skb, len);
2988 }
2989 
2990 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2991 {
2992 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2993 		skb->ip_summed = CHECKSUM_NONE;
2994 	__skb_trim(skb, len);
2995 	return 0;
2996 }
2997 
2998 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
2999 {
3000 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3001 		skb->ip_summed = CHECKSUM_NONE;
3002 	return __skb_grow(skb, len);
3003 }
3004 
3005 #define skb_queue_walk(queue, skb) \
3006 		for (skb = (queue)->next;					\
3007 		     skb != (struct sk_buff *)(queue);				\
3008 		     skb = skb->next)
3009 
3010 #define skb_queue_walk_safe(queue, skb, tmp)					\
3011 		for (skb = (queue)->next, tmp = skb->next;			\
3012 		     skb != (struct sk_buff *)(queue);				\
3013 		     skb = tmp, tmp = skb->next)
3014 
3015 #define skb_queue_walk_from(queue, skb)						\
3016 		for (; skb != (struct sk_buff *)(queue);			\
3017 		     skb = skb->next)
3018 
3019 #define skb_queue_walk_from_safe(queue, skb, tmp)				\
3020 		for (tmp = skb->next;						\
3021 		     skb != (struct sk_buff *)(queue);				\
3022 		     skb = tmp, tmp = skb->next)
3023 
3024 #define skb_queue_reverse_walk(queue, skb) \
3025 		for (skb = (queue)->prev;					\
3026 		     skb != (struct sk_buff *)(queue);				\
3027 		     skb = skb->prev)
3028 
3029 #define skb_queue_reverse_walk_safe(queue, skb, tmp)				\
3030 		for (skb = (queue)->prev, tmp = skb->prev;			\
3031 		     skb != (struct sk_buff *)(queue);				\
3032 		     skb = tmp, tmp = skb->prev)
3033 
3034 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp)			\
3035 		for (tmp = skb->prev;						\
3036 		     skb != (struct sk_buff *)(queue);				\
3037 		     skb = tmp, tmp = skb->prev)
3038 
3039 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3040 {
3041 	return skb_shinfo(skb)->frag_list != NULL;
3042 }
3043 
3044 static inline void skb_frag_list_init(struct sk_buff *skb)
3045 {
3046 	skb_shinfo(skb)->frag_list = NULL;
3047 }
3048 
3049 #define skb_walk_frags(skb, iter)	\
3050 	for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3051 
3052 
3053 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3054 				const struct sk_buff *skb);
3055 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3056 					void (*destructor)(struct sock *sk,
3057 							   struct sk_buff *skb),
3058 					int *peeked, int *off, int *err,
3059 					struct sk_buff **last);
3060 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3061 				    void (*destructor)(struct sock *sk,
3062 						       struct sk_buff *skb),
3063 				    int *peeked, int *off, int *err);
3064 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3065 				  int *err);
3066 unsigned int datagram_poll(struct file *file, struct socket *sock,
3067 			   struct poll_table_struct *wait);
3068 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3069 			   struct iov_iter *to, int size);
3070 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3071 					struct msghdr *msg, int size)
3072 {
3073 	return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3074 }
3075 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3076 				   struct msghdr *msg);
3077 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3078 				 struct iov_iter *from, int len);
3079 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3080 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3081 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3082 static inline void skb_free_datagram_locked(struct sock *sk,
3083 					    struct sk_buff *skb)
3084 {
3085 	__skb_free_datagram_locked(sk, skb, 0);
3086 }
3087 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3088 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3089 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3090 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3091 			      int len, __wsum csum);
3092 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3093 		    struct pipe_inode_info *pipe, unsigned int len,
3094 		    unsigned int flags);
3095 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3096 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3097 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3098 		 int len, int hlen);
3099 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3100 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3101 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3102 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3103 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3104 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3105 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3106 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3107 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3108 int skb_vlan_pop(struct sk_buff *skb);
3109 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3110 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3111 			     gfp_t gfp);
3112 
3113 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3114 {
3115 	return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3116 }
3117 
3118 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3119 {
3120 	return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3121 }
3122 
3123 struct skb_checksum_ops {
3124 	__wsum (*update)(const void *mem, int len, __wsum wsum);
3125 	__wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3126 };
3127 
3128 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3129 		      __wsum csum, const struct skb_checksum_ops *ops);
3130 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3131 		    __wsum csum);
3132 
3133 static inline void * __must_check
3134 __skb_header_pointer(const struct sk_buff *skb, int offset,
3135 		     int len, void *data, int hlen, void *buffer)
3136 {
3137 	if (hlen - offset >= len)
3138 		return data + offset;
3139 
3140 	if (!skb ||
3141 	    skb_copy_bits(skb, offset, buffer, len) < 0)
3142 		return NULL;
3143 
3144 	return buffer;
3145 }
3146 
3147 static inline void * __must_check
3148 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3149 {
3150 	return __skb_header_pointer(skb, offset, len, skb->data,
3151 				    skb_headlen(skb), buffer);
3152 }
3153 
3154 /**
3155  *	skb_needs_linearize - check if we need to linearize a given skb
3156  *			      depending on the given device features.
3157  *	@skb: socket buffer to check
3158  *	@features: net device features
3159  *
3160  *	Returns true if either:
3161  *	1. skb has frag_list and the device doesn't support FRAGLIST, or
3162  *	2. skb is fragmented and the device does not support SG.
3163  */
3164 static inline bool skb_needs_linearize(struct sk_buff *skb,
3165 				       netdev_features_t features)
3166 {
3167 	return skb_is_nonlinear(skb) &&
3168 	       ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3169 		(skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3170 }
3171 
3172 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3173 					     void *to,
3174 					     const unsigned int len)
3175 {
3176 	memcpy(to, skb->data, len);
3177 }
3178 
3179 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3180 						    const int offset, void *to,
3181 						    const unsigned int len)
3182 {
3183 	memcpy(to, skb->data + offset, len);
3184 }
3185 
3186 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3187 					   const void *from,
3188 					   const unsigned int len)
3189 {
3190 	memcpy(skb->data, from, len);
3191 }
3192 
3193 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3194 						  const int offset,
3195 						  const void *from,
3196 						  const unsigned int len)
3197 {
3198 	memcpy(skb->data + offset, from, len);
3199 }
3200 
3201 void skb_init(void);
3202 
3203 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3204 {
3205 	return skb->tstamp;
3206 }
3207 
3208 /**
3209  *	skb_get_timestamp - get timestamp from a skb
3210  *	@skb: skb to get stamp from
3211  *	@stamp: pointer to struct timeval to store stamp in
3212  *
3213  *	Timestamps are stored in the skb as offsets to a base timestamp.
3214  *	This function converts the offset back to a struct timeval and stores
3215  *	it in stamp.
3216  */
3217 static inline void skb_get_timestamp(const struct sk_buff *skb,
3218 				     struct timeval *stamp)
3219 {
3220 	*stamp = ktime_to_timeval(skb->tstamp);
3221 }
3222 
3223 static inline void skb_get_timestampns(const struct sk_buff *skb,
3224 				       struct timespec *stamp)
3225 {
3226 	*stamp = ktime_to_timespec(skb->tstamp);
3227 }
3228 
3229 static inline void __net_timestamp(struct sk_buff *skb)
3230 {
3231 	skb->tstamp = ktime_get_real();
3232 }
3233 
3234 static inline ktime_t net_timedelta(ktime_t t)
3235 {
3236 	return ktime_sub(ktime_get_real(), t);
3237 }
3238 
3239 static inline ktime_t net_invalid_timestamp(void)
3240 {
3241 	return 0;
3242 }
3243 
3244 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3245 
3246 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3247 
3248 void skb_clone_tx_timestamp(struct sk_buff *skb);
3249 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3250 
3251 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3252 
3253 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3254 {
3255 }
3256 
3257 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3258 {
3259 	return false;
3260 }
3261 
3262 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3263 
3264 /**
3265  * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3266  *
3267  * PHY drivers may accept clones of transmitted packets for
3268  * timestamping via their phy_driver.txtstamp method. These drivers
3269  * must call this function to return the skb back to the stack with a
3270  * timestamp.
3271  *
3272  * @skb: clone of the the original outgoing packet
3273  * @hwtstamps: hardware time stamps
3274  *
3275  */
3276 void skb_complete_tx_timestamp(struct sk_buff *skb,
3277 			       struct skb_shared_hwtstamps *hwtstamps);
3278 
3279 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3280 		     struct skb_shared_hwtstamps *hwtstamps,
3281 		     struct sock *sk, int tstype);
3282 
3283 /**
3284  * skb_tstamp_tx - queue clone of skb with send time stamps
3285  * @orig_skb:	the original outgoing packet
3286  * @hwtstamps:	hardware time stamps, may be NULL if not available
3287  *
3288  * If the skb has a socket associated, then this function clones the
3289  * skb (thus sharing the actual data and optional structures), stores
3290  * the optional hardware time stamping information (if non NULL) or
3291  * generates a software time stamp (otherwise), then queues the clone
3292  * to the error queue of the socket.  Errors are silently ignored.
3293  */
3294 void skb_tstamp_tx(struct sk_buff *orig_skb,
3295 		   struct skb_shared_hwtstamps *hwtstamps);
3296 
3297 static inline void sw_tx_timestamp(struct sk_buff *skb)
3298 {
3299 	if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3300 	    !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
3301 		skb_tstamp_tx(skb, NULL);
3302 }
3303 
3304 /**
3305  * skb_tx_timestamp() - Driver hook for transmit timestamping
3306  *
3307  * Ethernet MAC Drivers should call this function in their hard_xmit()
3308  * function immediately before giving the sk_buff to the MAC hardware.
3309  *
3310  * Specifically, one should make absolutely sure that this function is
3311  * called before TX completion of this packet can trigger.  Otherwise
3312  * the packet could potentially already be freed.
3313  *
3314  * @skb: A socket buffer.
3315  */
3316 static inline void skb_tx_timestamp(struct sk_buff *skb)
3317 {
3318 	skb_clone_tx_timestamp(skb);
3319 	sw_tx_timestamp(skb);
3320 }
3321 
3322 /**
3323  * skb_complete_wifi_ack - deliver skb with wifi status
3324  *
3325  * @skb: the original outgoing packet
3326  * @acked: ack status
3327  *
3328  */
3329 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3330 
3331 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3332 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3333 
3334 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3335 {
3336 	return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3337 		skb->csum_valid ||
3338 		(skb->ip_summed == CHECKSUM_PARTIAL &&
3339 		 skb_checksum_start_offset(skb) >= 0));
3340 }
3341 
3342 /**
3343  *	skb_checksum_complete - Calculate checksum of an entire packet
3344  *	@skb: packet to process
3345  *
3346  *	This function calculates the checksum over the entire packet plus
3347  *	the value of skb->csum.  The latter can be used to supply the
3348  *	checksum of a pseudo header as used by TCP/UDP.  It returns the
3349  *	checksum.
3350  *
3351  *	For protocols that contain complete checksums such as ICMP/TCP/UDP,
3352  *	this function can be used to verify that checksum on received
3353  *	packets.  In that case the function should return zero if the
3354  *	checksum is correct.  In particular, this function will return zero
3355  *	if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3356  *	hardware has already verified the correctness of the checksum.
3357  */
3358 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3359 {
3360 	return skb_csum_unnecessary(skb) ?
3361 	       0 : __skb_checksum_complete(skb);
3362 }
3363 
3364 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3365 {
3366 	if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3367 		if (skb->csum_level == 0)
3368 			skb->ip_summed = CHECKSUM_NONE;
3369 		else
3370 			skb->csum_level--;
3371 	}
3372 }
3373 
3374 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3375 {
3376 	if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3377 		if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3378 			skb->csum_level++;
3379 	} else if (skb->ip_summed == CHECKSUM_NONE) {
3380 		skb->ip_summed = CHECKSUM_UNNECESSARY;
3381 		skb->csum_level = 0;
3382 	}
3383 }
3384 
3385 static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3386 {
3387 	/* Mark current checksum as bad (typically called from GRO
3388 	 * path). In the case that ip_summed is CHECKSUM_NONE
3389 	 * this must be the first checksum encountered in the packet.
3390 	 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3391 	 * checksum after the last one validated. For UDP, a zero
3392 	 * checksum can not be marked as bad.
3393 	 */
3394 
3395 	if (skb->ip_summed == CHECKSUM_NONE ||
3396 	    skb->ip_summed == CHECKSUM_UNNECESSARY)
3397 		skb->csum_bad = 1;
3398 }
3399 
3400 /* Check if we need to perform checksum complete validation.
3401  *
3402  * Returns true if checksum complete is needed, false otherwise
3403  * (either checksum is unnecessary or zero checksum is allowed).
3404  */
3405 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3406 						  bool zero_okay,
3407 						  __sum16 check)
3408 {
3409 	if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3410 		skb->csum_valid = 1;
3411 		__skb_decr_checksum_unnecessary(skb);
3412 		return false;
3413 	}
3414 
3415 	return true;
3416 }
3417 
3418 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3419  * in checksum_init.
3420  */
3421 #define CHECKSUM_BREAK 76
3422 
3423 /* Unset checksum-complete
3424  *
3425  * Unset checksum complete can be done when packet is being modified
3426  * (uncompressed for instance) and checksum-complete value is
3427  * invalidated.
3428  */
3429 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3430 {
3431 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3432 		skb->ip_summed = CHECKSUM_NONE;
3433 }
3434 
3435 /* Validate (init) checksum based on checksum complete.
3436  *
3437  * Return values:
3438  *   0: checksum is validated or try to in skb_checksum_complete. In the latter
3439  *	case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3440  *	checksum is stored in skb->csum for use in __skb_checksum_complete
3441  *   non-zero: value of invalid checksum
3442  *
3443  */
3444 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3445 						       bool complete,
3446 						       __wsum psum)
3447 {
3448 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
3449 		if (!csum_fold(csum_add(psum, skb->csum))) {
3450 			skb->csum_valid = 1;
3451 			return 0;
3452 		}
3453 	} else if (skb->csum_bad) {
3454 		/* ip_summed == CHECKSUM_NONE in this case */
3455 		return (__force __sum16)1;
3456 	}
3457 
3458 	skb->csum = psum;
3459 
3460 	if (complete || skb->len <= CHECKSUM_BREAK) {
3461 		__sum16 csum;
3462 
3463 		csum = __skb_checksum_complete(skb);
3464 		skb->csum_valid = !csum;
3465 		return csum;
3466 	}
3467 
3468 	return 0;
3469 }
3470 
3471 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3472 {
3473 	return 0;
3474 }
3475 
3476 /* Perform checksum validate (init). Note that this is a macro since we only
3477  * want to calculate the pseudo header which is an input function if necessary.
3478  * First we try to validate without any computation (checksum unnecessary) and
3479  * then calculate based on checksum complete calling the function to compute
3480  * pseudo header.
3481  *
3482  * Return values:
3483  *   0: checksum is validated or try to in skb_checksum_complete
3484  *   non-zero: value of invalid checksum
3485  */
3486 #define __skb_checksum_validate(skb, proto, complete,			\
3487 				zero_okay, check, compute_pseudo)	\
3488 ({									\
3489 	__sum16 __ret = 0;						\
3490 	skb->csum_valid = 0;						\
3491 	if (__skb_checksum_validate_needed(skb, zero_okay, check))	\
3492 		__ret = __skb_checksum_validate_complete(skb,		\
3493 				complete, compute_pseudo(skb, proto));	\
3494 	__ret;								\
3495 })
3496 
3497 #define skb_checksum_init(skb, proto, compute_pseudo)			\
3498 	__skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3499 
3500 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo)	\
3501 	__skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3502 
3503 #define skb_checksum_validate(skb, proto, compute_pseudo)		\
3504 	__skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3505 
3506 #define skb_checksum_validate_zero_check(skb, proto, check,		\
3507 					 compute_pseudo)		\
3508 	__skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3509 
3510 #define skb_checksum_simple_validate(skb)				\
3511 	__skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3512 
3513 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3514 {
3515 	return (skb->ip_summed == CHECKSUM_NONE &&
3516 		skb->csum_valid && !skb->csum_bad);
3517 }
3518 
3519 static inline void __skb_checksum_convert(struct sk_buff *skb,
3520 					  __sum16 check, __wsum pseudo)
3521 {
3522 	skb->csum = ~pseudo;
3523 	skb->ip_summed = CHECKSUM_COMPLETE;
3524 }
3525 
3526 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo)	\
3527 do {									\
3528 	if (__skb_checksum_convert_check(skb))				\
3529 		__skb_checksum_convert(skb, check,			\
3530 				       compute_pseudo(skb, proto));	\
3531 } while (0)
3532 
3533 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3534 					      u16 start, u16 offset)
3535 {
3536 	skb->ip_summed = CHECKSUM_PARTIAL;
3537 	skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3538 	skb->csum_offset = offset - start;
3539 }
3540 
3541 /* Update skbuf and packet to reflect the remote checksum offload operation.
3542  * When called, ptr indicates the starting point for skb->csum when
3543  * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3544  * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3545  */
3546 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3547 				       int start, int offset, bool nopartial)
3548 {
3549 	__wsum delta;
3550 
3551 	if (!nopartial) {
3552 		skb_remcsum_adjust_partial(skb, ptr, start, offset);
3553 		return;
3554 	}
3555 
3556 	 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3557 		__skb_checksum_complete(skb);
3558 		skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3559 	}
3560 
3561 	delta = remcsum_adjust(ptr, skb->csum, start, offset);
3562 
3563 	/* Adjust skb->csum since we changed the packet */
3564 	skb->csum = csum_add(skb->csum, delta);
3565 }
3566 
3567 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3568 {
3569 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3570 	return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3571 #else
3572 	return NULL;
3573 #endif
3574 }
3575 
3576 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3577 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3578 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3579 {
3580 	if (nfct && atomic_dec_and_test(&nfct->use))
3581 		nf_conntrack_destroy(nfct);
3582 }
3583 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3584 {
3585 	if (nfct)
3586 		atomic_inc(&nfct->use);
3587 }
3588 #endif
3589 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3590 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3591 {
3592 	if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3593 		kfree(nf_bridge);
3594 }
3595 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3596 {
3597 	if (nf_bridge)
3598 		atomic_inc(&nf_bridge->use);
3599 }
3600 #endif /* CONFIG_BRIDGE_NETFILTER */
3601 static inline void nf_reset(struct sk_buff *skb)
3602 {
3603 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3604 	nf_conntrack_put(skb_nfct(skb));
3605 	skb->_nfct = 0;
3606 #endif
3607 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3608 	nf_bridge_put(skb->nf_bridge);
3609 	skb->nf_bridge = NULL;
3610 #endif
3611 }
3612 
3613 static inline void nf_reset_trace(struct sk_buff *skb)
3614 {
3615 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3616 	skb->nf_trace = 0;
3617 #endif
3618 }
3619 
3620 /* Note: This doesn't put any conntrack and bridge info in dst. */
3621 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3622 			     bool copy)
3623 {
3624 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3625 	dst->_nfct = src->_nfct;
3626 	nf_conntrack_get(skb_nfct(src));
3627 #endif
3628 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3629 	dst->nf_bridge  = src->nf_bridge;
3630 	nf_bridge_get(src->nf_bridge);
3631 #endif
3632 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3633 	if (copy)
3634 		dst->nf_trace = src->nf_trace;
3635 #endif
3636 }
3637 
3638 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3639 {
3640 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3641 	nf_conntrack_put(skb_nfct(dst));
3642 #endif
3643 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3644 	nf_bridge_put(dst->nf_bridge);
3645 #endif
3646 	__nf_copy(dst, src, true);
3647 }
3648 
3649 #ifdef CONFIG_NETWORK_SECMARK
3650 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3651 {
3652 	to->secmark = from->secmark;
3653 }
3654 
3655 static inline void skb_init_secmark(struct sk_buff *skb)
3656 {
3657 	skb->secmark = 0;
3658 }
3659 #else
3660 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3661 { }
3662 
3663 static inline void skb_init_secmark(struct sk_buff *skb)
3664 { }
3665 #endif
3666 
3667 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3668 {
3669 	return !skb->destructor &&
3670 #if IS_ENABLED(CONFIG_XFRM)
3671 		!skb->sp &&
3672 #endif
3673 		!skb_nfct(skb) &&
3674 		!skb->_skb_refdst &&
3675 		!skb_has_frag_list(skb);
3676 }
3677 
3678 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3679 {
3680 	skb->queue_mapping = queue_mapping;
3681 }
3682 
3683 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3684 {
3685 	return skb->queue_mapping;
3686 }
3687 
3688 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3689 {
3690 	to->queue_mapping = from->queue_mapping;
3691 }
3692 
3693 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3694 {
3695 	skb->queue_mapping = rx_queue + 1;
3696 }
3697 
3698 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3699 {
3700 	return skb->queue_mapping - 1;
3701 }
3702 
3703 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3704 {
3705 	return skb->queue_mapping != 0;
3706 }
3707 
3708 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3709 {
3710 	skb->dst_pending_confirm = val;
3711 }
3712 
3713 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3714 {
3715 	return skb->dst_pending_confirm != 0;
3716 }
3717 
3718 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3719 {
3720 #ifdef CONFIG_XFRM
3721 	return skb->sp;
3722 #else
3723 	return NULL;
3724 #endif
3725 }
3726 
3727 /* Keeps track of mac header offset relative to skb->head.
3728  * It is useful for TSO of Tunneling protocol. e.g. GRE.
3729  * For non-tunnel skb it points to skb_mac_header() and for
3730  * tunnel skb it points to outer mac header.
3731  * Keeps track of level of encapsulation of network headers.
3732  */
3733 struct skb_gso_cb {
3734 	union {
3735 		int	mac_offset;
3736 		int	data_offset;
3737 	};
3738 	int	encap_level;
3739 	__wsum	csum;
3740 	__u16	csum_start;
3741 };
3742 #define SKB_SGO_CB_OFFSET	32
3743 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3744 
3745 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3746 {
3747 	return (skb_mac_header(inner_skb) - inner_skb->head) -
3748 		SKB_GSO_CB(inner_skb)->mac_offset;
3749 }
3750 
3751 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3752 {
3753 	int new_headroom, headroom;
3754 	int ret;
3755 
3756 	headroom = skb_headroom(skb);
3757 	ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3758 	if (ret)
3759 		return ret;
3760 
3761 	new_headroom = skb_headroom(skb);
3762 	SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3763 	return 0;
3764 }
3765 
3766 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3767 {
3768 	/* Do not update partial checksums if remote checksum is enabled. */
3769 	if (skb->remcsum_offload)
3770 		return;
3771 
3772 	SKB_GSO_CB(skb)->csum = res;
3773 	SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3774 }
3775 
3776 /* Compute the checksum for a gso segment. First compute the checksum value
3777  * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3778  * then add in skb->csum (checksum from csum_start to end of packet).
3779  * skb->csum and csum_start are then updated to reflect the checksum of the
3780  * resultant packet starting from the transport header-- the resultant checksum
3781  * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3782  * header.
3783  */
3784 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3785 {
3786 	unsigned char *csum_start = skb_transport_header(skb);
3787 	int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3788 	__wsum partial = SKB_GSO_CB(skb)->csum;
3789 
3790 	SKB_GSO_CB(skb)->csum = res;
3791 	SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3792 
3793 	return csum_fold(csum_partial(csum_start, plen, partial));
3794 }
3795 
3796 static inline bool skb_is_gso(const struct sk_buff *skb)
3797 {
3798 	return skb_shinfo(skb)->gso_size;
3799 }
3800 
3801 /* Note: Should be called only if skb_is_gso(skb) is true */
3802 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3803 {
3804 	return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3805 }
3806 
3807 static inline void skb_gso_reset(struct sk_buff *skb)
3808 {
3809 	skb_shinfo(skb)->gso_size = 0;
3810 	skb_shinfo(skb)->gso_segs = 0;
3811 	skb_shinfo(skb)->gso_type = 0;
3812 }
3813 
3814 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3815 
3816 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3817 {
3818 	/* LRO sets gso_size but not gso_type, whereas if GSO is really
3819 	 * wanted then gso_type will be set. */
3820 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
3821 
3822 	if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3823 	    unlikely(shinfo->gso_type == 0)) {
3824 		__skb_warn_lro_forwarding(skb);
3825 		return true;
3826 	}
3827 	return false;
3828 }
3829 
3830 static inline void skb_forward_csum(struct sk_buff *skb)
3831 {
3832 	/* Unfortunately we don't support this one.  Any brave souls? */
3833 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3834 		skb->ip_summed = CHECKSUM_NONE;
3835 }
3836 
3837 /**
3838  * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3839  * @skb: skb to check
3840  *
3841  * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3842  * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3843  * use this helper, to document places where we make this assertion.
3844  */
3845 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3846 {
3847 #ifdef DEBUG
3848 	BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3849 #endif
3850 }
3851 
3852 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3853 
3854 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3855 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3856 				     unsigned int transport_len,
3857 				     __sum16(*skb_chkf)(struct sk_buff *skb));
3858 
3859 /**
3860  * skb_head_is_locked - Determine if the skb->head is locked down
3861  * @skb: skb to check
3862  *
3863  * The head on skbs build around a head frag can be removed if they are
3864  * not cloned.  This function returns true if the skb head is locked down
3865  * due to either being allocated via kmalloc, or by being a clone with
3866  * multiple references to the head.
3867  */
3868 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3869 {
3870 	return !skb->head_frag || skb_cloned(skb);
3871 }
3872 
3873 /**
3874  * skb_gso_network_seglen - Return length of individual segments of a gso packet
3875  *
3876  * @skb: GSO skb
3877  *
3878  * skb_gso_network_seglen is used to determine the real size of the
3879  * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3880  *
3881  * The MAC/L2 header is not accounted for.
3882  */
3883 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3884 {
3885 	unsigned int hdr_len = skb_transport_header(skb) -
3886 			       skb_network_header(skb);
3887 	return hdr_len + skb_gso_transport_seglen(skb);
3888 }
3889 
3890 /* Local Checksum Offload.
3891  * Compute outer checksum based on the assumption that the
3892  * inner checksum will be offloaded later.
3893  * See Documentation/networking/checksum-offloads.txt for
3894  * explanation of how this works.
3895  * Fill in outer checksum adjustment (e.g. with sum of outer
3896  * pseudo-header) before calling.
3897  * Also ensure that inner checksum is in linear data area.
3898  */
3899 static inline __wsum lco_csum(struct sk_buff *skb)
3900 {
3901 	unsigned char *csum_start = skb_checksum_start(skb);
3902 	unsigned char *l4_hdr = skb_transport_header(skb);
3903 	__wsum partial;
3904 
3905 	/* Start with complement of inner checksum adjustment */
3906 	partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3907 						    skb->csum_offset));
3908 
3909 	/* Add in checksum of our headers (incl. outer checksum
3910 	 * adjustment filled in by caller) and return result.
3911 	 */
3912 	return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3913 }
3914 
3915 #endif	/* __KERNEL__ */
3916 #endif	/* _LINUX_SKBUFF_H */
3917