xref: /linux-6.15/include/linux/skbuff.h (revision 6d20e6b2)
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 <linux/sched/clock.h>
38 #include <net/flow_dissector.h>
39 #include <linux/splice.h>
40 #include <linux/in6.h>
41 #include <linux/if_packet.h>
42 #include <net/flow.h>
43 
44 /* The interface for checksum offload between the stack and networking drivers
45  * is as follows...
46  *
47  * A. IP checksum related features
48  *
49  * Drivers advertise checksum offload capabilities in the features of a device.
50  * From the stack's point of view these are capabilities offered by the driver,
51  * a driver typically only advertises features that it is capable of offloading
52  * to its device.
53  *
54  * The checksum related features are:
55  *
56  *	NETIF_F_HW_CSUM	- The driver (or its device) is able to compute one
57  *			  IP (one's complement) checksum for any combination
58  *			  of protocols or protocol layering. The checksum is
59  *			  computed and set in a packet per the CHECKSUM_PARTIAL
60  *			  interface (see below).
61  *
62  *	NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
63  *			  TCP or UDP packets over IPv4. These are specifically
64  *			  unencapsulated packets of the form IPv4|TCP or
65  *			  IPv4|UDP where the Protocol field in the IPv4 header
66  *			  is TCP or UDP. The IPv4 header may contain IP options
67  *			  This feature cannot be set in features for a device
68  *			  with NETIF_F_HW_CSUM also set. This feature is being
69  *			  DEPRECATED (see below).
70  *
71  *	NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
72  *			  TCP or UDP packets over IPv6. These are specifically
73  *			  unencapsulated packets of the form IPv6|TCP or
74  *			  IPv4|UDP where the Next Header field in the IPv6
75  *			  header is either TCP or UDP. IPv6 extension headers
76  *			  are not supported with this feature. This feature
77  *			  cannot be set in features for a device with
78  *			  NETIF_F_HW_CSUM also set. This feature is being
79  *			  DEPRECATED (see below).
80  *
81  *	NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
82  *			 This flag is used only used to disable the RX checksum
83  *			 feature for a device. The stack will accept receive
84  *			 checksum indication in packets received on a device
85  *			 regardless of whether NETIF_F_RXCSUM is set.
86  *
87  * B. Checksumming of received packets by device. Indication of checksum
88  *    verification is in set skb->ip_summed. Possible values are:
89  *
90  * CHECKSUM_NONE:
91  *
92  *   Device did not checksum this packet e.g. due to lack of capabilities.
93  *   The packet contains full (though not verified) checksum in packet but
94  *   not in skb->csum. Thus, skb->csum is undefined in this case.
95  *
96  * CHECKSUM_UNNECESSARY:
97  *
98  *   The hardware you're dealing with doesn't calculate the full checksum
99  *   (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
100  *   for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
101  *   if their checksums are okay. skb->csum is still undefined in this case
102  *   though. A driver or device must never modify the checksum field in the
103  *   packet even if checksum is verified.
104  *
105  *   CHECKSUM_UNNECESSARY is applicable to following protocols:
106  *     TCP: IPv6 and IPv4.
107  *     UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
108  *       zero UDP checksum for either IPv4 or IPv6, the networking stack
109  *       may perform further validation in this case.
110  *     GRE: only if the checksum is present in the header.
111  *     SCTP: indicates the CRC in SCTP header has been validated.
112  *     FCOE: indicates the CRC in FC frame has been validated.
113  *
114  *   skb->csum_level indicates the number of consecutive checksums found in
115  *   the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
116  *   For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
117  *   and a device is able to verify the checksums for UDP (possibly zero),
118  *   GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
119  *   two. If the device were only able to verify the UDP checksum and not
120  *   GRE, either because it doesn't support GRE checksum of because GRE
121  *   checksum is bad, skb->csum_level would be set to zero (TCP checksum is
122  *   not considered in this case).
123  *
124  * CHECKSUM_COMPLETE:
125  *
126  *   This is the most generic way. The device supplied checksum of the _whole_
127  *   packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
128  *   hardware doesn't need to parse L3/L4 headers to implement this.
129  *
130  *   Notes:
131  *   - Even if device supports only some protocols, but is able to produce
132  *     skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
133  *   - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
134  *
135  * CHECKSUM_PARTIAL:
136  *
137  *   A checksum is set up to be offloaded to a device as described in the
138  *   output description for CHECKSUM_PARTIAL. This may occur on a packet
139  *   received directly from another Linux OS, e.g., a virtualized Linux kernel
140  *   on the same host, or it may be set in the input path in GRO or remote
141  *   checksum offload. For the purposes of checksum verification, the checksum
142  *   referred to by skb->csum_start + skb->csum_offset and any preceding
143  *   checksums in the packet are considered verified. Any checksums in the
144  *   packet that are after the checksum being offloaded are not considered to
145  *   be verified.
146  *
147  * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
148  *    in the skb->ip_summed for a packet. Values are:
149  *
150  * CHECKSUM_PARTIAL:
151  *
152  *   The driver is required to checksum the packet as seen by hard_start_xmit()
153  *   from skb->csum_start up to the end, and to record/write the checksum at
154  *   offset skb->csum_start + skb->csum_offset. A driver may verify that the
155  *   csum_start and csum_offset values are valid values given the length and
156  *   offset of the packet, however they should not attempt to validate that the
157  *   checksum refers to a legitimate transport layer checksum-- it is the
158  *   purview of the stack to validate that csum_start and csum_offset are set
159  *   correctly.
160  *
161  *   When the stack requests checksum offload for a packet, the driver MUST
162  *   ensure that the checksum is set correctly. A driver can either offload the
163  *   checksum calculation to the device, or call skb_checksum_help (in the case
164  *   that the device does not support offload for a particular checksum).
165  *
166  *   NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
167  *   NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
168  *   checksum offload capability.
169  *   skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
170  *   on network device checksumming capabilities: if a packet does not match
171  *   them, skb_checksum_help or skb_crc32c_help (depending on the value of
172  *   csum_not_inet, see item D.) is called to resolve the checksum.
173  *
174  * CHECKSUM_NONE:
175  *
176  *   The skb was already checksummed by the protocol, or a checksum is not
177  *   required.
178  *
179  * CHECKSUM_UNNECESSARY:
180  *
181  *   This has the same meaning on as CHECKSUM_NONE for checksum offload on
182  *   output.
183  *
184  * CHECKSUM_COMPLETE:
185  *   Not used in checksum output. If a driver observes a packet with this value
186  *   set in skbuff, if should treat as CHECKSUM_NONE being set.
187  *
188  * D. Non-IP checksum (CRC) offloads
189  *
190  *   NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
191  *     offloading the SCTP CRC in a packet. To perform this offload the stack
192  *     will set set csum_start and csum_offset accordingly, set ip_summed to
193  *     CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
194  *     the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
195  *     A driver that supports both IP checksum offload and SCTP CRC32c offload
196  *     must verify which offload is configured for a packet by testing the
197  *     value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
198  *     CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
199  *
200  *   NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
201  *     offloading the FCOE CRC in a packet. To perform this offload the stack
202  *     will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
203  *     accordingly. Note the there is no indication in the skbuff that the
204  *     CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
205  *     both IP checksum offload and FCOE CRC offload must verify which offload
206  *     is configured for a packet presumably by inspecting packet headers.
207  *
208  * E. Checksumming on output with GSO.
209  *
210  * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
211  * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
212  * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
213  * part of the GSO operation is implied. If a checksum is being offloaded
214  * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
215  * are set to refer to the outermost checksum being offload (two offloaded
216  * checksums are possible with UDP encapsulation).
217  */
218 
219 /* Don't change this without changing skb_csum_unnecessary! */
220 #define CHECKSUM_NONE		0
221 #define CHECKSUM_UNNECESSARY	1
222 #define CHECKSUM_COMPLETE	2
223 #define CHECKSUM_PARTIAL	3
224 
225 /* Maximum value in skb->csum_level */
226 #define SKB_MAX_CSUM_LEVEL	3
227 
228 #define SKB_DATA_ALIGN(X)	ALIGN(X, SMP_CACHE_BYTES)
229 #define SKB_WITH_OVERHEAD(X)	\
230 	((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
231 #define SKB_MAX_ORDER(X, ORDER) \
232 	SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
233 #define SKB_MAX_HEAD(X)		(SKB_MAX_ORDER((X), 0))
234 #define SKB_MAX_ALLOC		(SKB_MAX_ORDER(0, 2))
235 
236 /* return minimum truesize of one skb containing X bytes of data */
237 #define SKB_TRUESIZE(X) ((X) +						\
238 			 SKB_DATA_ALIGN(sizeof(struct sk_buff)) +	\
239 			 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
240 
241 struct net_device;
242 struct scatterlist;
243 struct pipe_inode_info;
244 struct iov_iter;
245 struct napi_struct;
246 
247 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
248 struct nf_conntrack {
249 	atomic_t use;
250 };
251 #endif
252 
253 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
254 struct nf_bridge_info {
255 	refcount_t		use;
256 	enum {
257 		BRNF_PROTO_UNCHANGED,
258 		BRNF_PROTO_8021Q,
259 		BRNF_PROTO_PPPOE
260 	} orig_proto:8;
261 	u8			pkt_otherhost:1;
262 	u8			in_prerouting:1;
263 	u8			bridged_dnat:1;
264 	__u16			frag_max_size;
265 	struct net_device	*physindev;
266 
267 	/* always valid & non-NULL from FORWARD on, for physdev match */
268 	struct net_device	*physoutdev;
269 	union {
270 		/* prerouting: detect dnat in orig/reply direction */
271 		__be32          ipv4_daddr;
272 		struct in6_addr ipv6_daddr;
273 
274 		/* after prerouting + nat detected: store original source
275 		 * mac since neigh resolution overwrites it, only used while
276 		 * skb is out in neigh layer.
277 		 */
278 		char neigh_header[8];
279 	};
280 };
281 #endif
282 
283 struct sk_buff_head {
284 	/* These two members must be first. */
285 	struct sk_buff	*next;
286 	struct sk_buff	*prev;
287 
288 	__u32		qlen;
289 	spinlock_t	lock;
290 };
291 
292 struct sk_buff;
293 
294 /* To allow 64K frame to be packed as single skb without frag_list we
295  * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
296  * buffers which do not start on a page boundary.
297  *
298  * Since GRO uses frags we allocate at least 16 regardless of page
299  * size.
300  */
301 #if (65536/PAGE_SIZE + 1) < 16
302 #define MAX_SKB_FRAGS 16UL
303 #else
304 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
305 #endif
306 extern int sysctl_max_skb_frags;
307 
308 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
309  * segment using its current segmentation instead.
310  */
311 #define GSO_BY_FRAGS	0xFFFF
312 
313 typedef struct skb_frag_struct skb_frag_t;
314 
315 struct skb_frag_struct {
316 	struct {
317 		struct page *p;
318 	} page;
319 #if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
320 	__u32 page_offset;
321 	__u32 size;
322 #else
323 	__u16 page_offset;
324 	__u16 size;
325 #endif
326 };
327 
328 static inline unsigned int skb_frag_size(const skb_frag_t *frag)
329 {
330 	return frag->size;
331 }
332 
333 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
334 {
335 	frag->size = size;
336 }
337 
338 static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
339 {
340 	frag->size += delta;
341 }
342 
343 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
344 {
345 	frag->size -= delta;
346 }
347 
348 #define HAVE_HW_TIME_STAMP
349 
350 /**
351  * struct skb_shared_hwtstamps - hardware time stamps
352  * @hwtstamp:	hardware time stamp transformed into duration
353  *		since arbitrary point in time
354  *
355  * Software time stamps generated by ktime_get_real() are stored in
356  * skb->tstamp.
357  *
358  * hwtstamps can only be compared against other hwtstamps from
359  * the same device.
360  *
361  * This structure is attached to packets as part of the
362  * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
363  */
364 struct skb_shared_hwtstamps {
365 	ktime_t	hwtstamp;
366 };
367 
368 /* Definitions for tx_flags in struct skb_shared_info */
369 enum {
370 	/* generate hardware time stamp */
371 	SKBTX_HW_TSTAMP = 1 << 0,
372 
373 	/* generate software time stamp when queueing packet to NIC */
374 	SKBTX_SW_TSTAMP = 1 << 1,
375 
376 	/* device driver is going to provide hardware time stamp */
377 	SKBTX_IN_PROGRESS = 1 << 2,
378 
379 	/* device driver supports TX zero-copy buffers */
380 	SKBTX_DEV_ZEROCOPY = 1 << 3,
381 
382 	/* generate wifi status information (where possible) */
383 	SKBTX_WIFI_STATUS = 1 << 4,
384 
385 	/* This indicates at least one fragment might be overwritten
386 	 * (as in vmsplice(), sendfile() ...)
387 	 * If we need to compute a TX checksum, we'll need to copy
388 	 * all frags to avoid possible bad checksum
389 	 */
390 	SKBTX_SHARED_FRAG = 1 << 5,
391 
392 	/* generate software time stamp when entering packet scheduling */
393 	SKBTX_SCHED_TSTAMP = 1 << 6,
394 };
395 
396 #define SKBTX_ANY_SW_TSTAMP	(SKBTX_SW_TSTAMP    | \
397 				 SKBTX_SCHED_TSTAMP)
398 #define SKBTX_ANY_TSTAMP	(SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
399 
400 /*
401  * The callback notifies userspace to release buffers when skb DMA is done in
402  * lower device, the skb last reference should be 0 when calling this.
403  * The zerocopy_success argument is true if zero copy transmit occurred,
404  * false on data copy or out of memory error caused by data copy attempt.
405  * The ctx field is used to track device context.
406  * The desc field is used to track userspace buffer index.
407  */
408 struct ubuf_info {
409 	void (*callback)(struct ubuf_info *, bool zerocopy_success);
410 	void *ctx;
411 	unsigned long desc;
412 };
413 
414 /* This data is invariant across clones and lives at
415  * the end of the header data, ie. at skb->end.
416  */
417 struct skb_shared_info {
418 	unsigned short	_unused;
419 	unsigned char	nr_frags;
420 	__u8		tx_flags;
421 	unsigned short	gso_size;
422 	/* Warning: this field is not always filled in (UFO)! */
423 	unsigned short	gso_segs;
424 	struct sk_buff	*frag_list;
425 	struct skb_shared_hwtstamps hwtstamps;
426 	unsigned int	gso_type;
427 	u32		tskey;
428 	__be32          ip6_frag_id;
429 
430 	/*
431 	 * Warning : all fields before dataref are cleared in __alloc_skb()
432 	 */
433 	atomic_t	dataref;
434 
435 	/* Intermediate layers must ensure that destructor_arg
436 	 * remains valid until skb destructor */
437 	void *		destructor_arg;
438 
439 	/* must be last field, see pskb_expand_head() */
440 	skb_frag_t	frags[MAX_SKB_FRAGS];
441 };
442 
443 /* We divide dataref into two halves.  The higher 16 bits hold references
444  * to the payload part of skb->data.  The lower 16 bits hold references to
445  * the entire skb->data.  A clone of a headerless skb holds the length of
446  * the header in skb->hdr_len.
447  *
448  * All users must obey the rule that the skb->data reference count must be
449  * greater than or equal to the payload reference count.
450  *
451  * Holding a reference to the payload part means that the user does not
452  * care about modifications to the header part of skb->data.
453  */
454 #define SKB_DATAREF_SHIFT 16
455 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
456 
457 
458 enum {
459 	SKB_FCLONE_UNAVAILABLE,	/* skb has no fclone (from head_cache) */
460 	SKB_FCLONE_ORIG,	/* orig skb (from fclone_cache) */
461 	SKB_FCLONE_CLONE,	/* companion fclone skb (from fclone_cache) */
462 };
463 
464 enum {
465 	SKB_GSO_TCPV4 = 1 << 0,
466 	SKB_GSO_UDP = 1 << 1,
467 
468 	/* This indicates the skb is from an untrusted source. */
469 	SKB_GSO_DODGY = 1 << 2,
470 
471 	/* This indicates the tcp segment has CWR set. */
472 	SKB_GSO_TCP_ECN = 1 << 3,
473 
474 	SKB_GSO_TCP_FIXEDID = 1 << 4,
475 
476 	SKB_GSO_TCPV6 = 1 << 5,
477 
478 	SKB_GSO_FCOE = 1 << 6,
479 
480 	SKB_GSO_GRE = 1 << 7,
481 
482 	SKB_GSO_GRE_CSUM = 1 << 8,
483 
484 	SKB_GSO_IPXIP4 = 1 << 9,
485 
486 	SKB_GSO_IPXIP6 = 1 << 10,
487 
488 	SKB_GSO_UDP_TUNNEL = 1 << 11,
489 
490 	SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
491 
492 	SKB_GSO_PARTIAL = 1 << 13,
493 
494 	SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
495 
496 	SKB_GSO_SCTP = 1 << 15,
497 
498 	SKB_GSO_ESP = 1 << 16,
499 };
500 
501 #if BITS_PER_LONG > 32
502 #define NET_SKBUFF_DATA_USES_OFFSET 1
503 #endif
504 
505 #ifdef NET_SKBUFF_DATA_USES_OFFSET
506 typedef unsigned int sk_buff_data_t;
507 #else
508 typedef unsigned char *sk_buff_data_t;
509 #endif
510 
511 /**
512  *	struct sk_buff - socket buffer
513  *	@next: Next buffer in list
514  *	@prev: Previous buffer in list
515  *	@tstamp: Time we arrived/left
516  *	@rbnode: RB tree node, alternative to next/prev for netem/tcp
517  *	@sk: Socket we are owned by
518  *	@dev: Device we arrived on/are leaving by
519  *	@cb: Control buffer. Free for use by every layer. Put private vars here
520  *	@_skb_refdst: destination entry (with norefcount bit)
521  *	@sp: the security path, used for xfrm
522  *	@len: Length of actual data
523  *	@data_len: Data length
524  *	@mac_len: Length of link layer header
525  *	@hdr_len: writable header length of cloned skb
526  *	@csum: Checksum (must include start/offset pair)
527  *	@csum_start: Offset from skb->head where checksumming should start
528  *	@csum_offset: Offset from csum_start where checksum should be stored
529  *	@priority: Packet queueing priority
530  *	@ignore_df: allow local fragmentation
531  *	@cloned: Head may be cloned (check refcnt to be sure)
532  *	@ip_summed: Driver fed us an IP checksum
533  *	@nohdr: Payload reference only, must not modify header
534  *	@pkt_type: Packet class
535  *	@fclone: skbuff clone status
536  *	@ipvs_property: skbuff is owned by ipvs
537  *	@tc_skip_classify: do not classify packet. set by IFB device
538  *	@tc_at_ingress: used within tc_classify to distinguish in/egress
539  *	@tc_redirected: packet was redirected by a tc action
540  *	@tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
541  *	@peeked: this packet has been seen already, so stats have been
542  *		done for it, don't do them again
543  *	@nf_trace: netfilter packet trace flag
544  *	@protocol: Packet protocol from driver
545  *	@destructor: Destruct function
546  *	@_nfct: Associated connection, if any (with nfctinfo bits)
547  *	@nf_bridge: Saved data about a bridged frame - see br_netfilter.c
548  *	@skb_iif: ifindex of device we arrived on
549  *	@tc_index: Traffic control index
550  *	@hash: the packet hash
551  *	@queue_mapping: Queue mapping for multiqueue devices
552  *	@xmit_more: More SKBs are pending for this queue
553  *	@ndisc_nodetype: router type (from link layer)
554  *	@ooo_okay: allow the mapping of a socket to a queue to be changed
555  *	@l4_hash: indicate hash is a canonical 4-tuple hash over transport
556  *		ports.
557  *	@sw_hash: indicates hash was computed in software stack
558  *	@wifi_acked_valid: wifi_acked was set
559  *	@wifi_acked: whether frame was acked on wifi or not
560  *	@no_fcs:  Request NIC to treat last 4 bytes as Ethernet FCS
561  *	@csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
562  *	@dst_pending_confirm: need to confirm neighbour
563   *	@napi_id: id of the NAPI struct this skb came from
564  *	@secmark: security marking
565  *	@mark: Generic packet mark
566  *	@vlan_proto: vlan encapsulation protocol
567  *	@vlan_tci: vlan tag control information
568  *	@inner_protocol: Protocol (encapsulation)
569  *	@inner_transport_header: Inner transport layer header (encapsulation)
570  *	@inner_network_header: Network layer header (encapsulation)
571  *	@inner_mac_header: Link layer header (encapsulation)
572  *	@transport_header: Transport layer header
573  *	@network_header: Network layer header
574  *	@mac_header: Link layer header
575  *	@tail: Tail pointer
576  *	@end: End pointer
577  *	@head: Head of buffer
578  *	@data: Data head pointer
579  *	@truesize: Buffer size
580  *	@users: User count - see {datagram,tcp}.c
581  */
582 
583 struct sk_buff {
584 	union {
585 		struct {
586 			/* These two members must be first. */
587 			struct sk_buff		*next;
588 			struct sk_buff		*prev;
589 
590 			union {
591 				ktime_t		tstamp;
592 				u64		skb_mstamp;
593 			};
594 		};
595 		struct rb_node	rbnode; /* used in netem & tcp stack */
596 	};
597 	struct sock		*sk;
598 
599 	union {
600 		struct net_device	*dev;
601 		/* Some protocols might use this space to store information,
602 		 * while device pointer would be NULL.
603 		 * UDP receive path is one user.
604 		 */
605 		unsigned long		dev_scratch;
606 	};
607 	/*
608 	 * This is the control buffer. It is free to use for every
609 	 * layer. Please put your private variables there. If you
610 	 * want to keep them across layers you have to do a skb_clone()
611 	 * first. This is owned by whoever has the skb queued ATM.
612 	 */
613 	char			cb[48] __aligned(8);
614 
615 	unsigned long		_skb_refdst;
616 	void			(*destructor)(struct sk_buff *skb);
617 #ifdef CONFIG_XFRM
618 	struct	sec_path	*sp;
619 #endif
620 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
621 	unsigned long		 _nfct;
622 #endif
623 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
624 	struct nf_bridge_info	*nf_bridge;
625 #endif
626 	unsigned int		len,
627 				data_len;
628 	__u16			mac_len,
629 				hdr_len;
630 
631 	/* Following fields are _not_ copied in __copy_skb_header()
632 	 * Note that queue_mapping is here mostly to fill a hole.
633 	 */
634 	kmemcheck_bitfield_begin(flags1);
635 	__u16			queue_mapping;
636 
637 /* if you move cloned around you also must adapt those constants */
638 #ifdef __BIG_ENDIAN_BITFIELD
639 #define CLONED_MASK	(1 << 7)
640 #else
641 #define CLONED_MASK	1
642 #endif
643 #define CLONED_OFFSET()		offsetof(struct sk_buff, __cloned_offset)
644 
645 	__u8			__cloned_offset[0];
646 	__u8			cloned:1,
647 				nohdr:1,
648 				fclone:2,
649 				peeked:1,
650 				head_frag:1,
651 				xmit_more:1,
652 				__unused:1; /* one bit hole */
653 	kmemcheck_bitfield_end(flags1);
654 
655 	/* fields enclosed in headers_start/headers_end are copied
656 	 * using a single memcpy() in __copy_skb_header()
657 	 */
658 	/* private: */
659 	__u32			headers_start[0];
660 	/* public: */
661 
662 /* if you move pkt_type around you also must adapt those constants */
663 #ifdef __BIG_ENDIAN_BITFIELD
664 #define PKT_TYPE_MAX	(7 << 5)
665 #else
666 #define PKT_TYPE_MAX	7
667 #endif
668 #define PKT_TYPE_OFFSET()	offsetof(struct sk_buff, __pkt_type_offset)
669 
670 	__u8			__pkt_type_offset[0];
671 	__u8			pkt_type:3;
672 	__u8			pfmemalloc:1;
673 	__u8			ignore_df:1;
674 
675 	__u8			nf_trace:1;
676 	__u8			ip_summed:2;
677 	__u8			ooo_okay:1;
678 	__u8			l4_hash:1;
679 	__u8			sw_hash:1;
680 	__u8			wifi_acked_valid:1;
681 	__u8			wifi_acked:1;
682 
683 	__u8			no_fcs:1;
684 	/* Indicates the inner headers are valid in the skbuff. */
685 	__u8			encapsulation:1;
686 	__u8			encap_hdr_csum:1;
687 	__u8			csum_valid:1;
688 	__u8			csum_complete_sw:1;
689 	__u8			csum_level:2;
690 	__u8			csum_not_inet:1;
691 
692 	__u8			dst_pending_confirm:1;
693 #ifdef CONFIG_IPV6_NDISC_NODETYPE
694 	__u8			ndisc_nodetype:2;
695 #endif
696 	__u8			ipvs_property:1;
697 	__u8			inner_protocol_type:1;
698 	__u8			remcsum_offload:1;
699 #ifdef CONFIG_NET_SWITCHDEV
700 	__u8			offload_fwd_mark:1;
701 #endif
702 #ifdef CONFIG_NET_CLS_ACT
703 	__u8			tc_skip_classify:1;
704 	__u8			tc_at_ingress:1;
705 	__u8			tc_redirected:1;
706 	__u8			tc_from_ingress:1;
707 #endif
708 
709 #ifdef CONFIG_NET_SCHED
710 	__u16			tc_index;	/* traffic control index */
711 #endif
712 
713 	union {
714 		__wsum		csum;
715 		struct {
716 			__u16	csum_start;
717 			__u16	csum_offset;
718 		};
719 	};
720 	__u32			priority;
721 	int			skb_iif;
722 	__u32			hash;
723 	__be16			vlan_proto;
724 	__u16			vlan_tci;
725 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
726 	union {
727 		unsigned int	napi_id;
728 		unsigned int	sender_cpu;
729 	};
730 #endif
731 #ifdef CONFIG_NETWORK_SECMARK
732 	__u32		secmark;
733 #endif
734 
735 	union {
736 		__u32		mark;
737 		__u32		reserved_tailroom;
738 	};
739 
740 	union {
741 		__be16		inner_protocol;
742 		__u8		inner_ipproto;
743 	};
744 
745 	__u16			inner_transport_header;
746 	__u16			inner_network_header;
747 	__u16			inner_mac_header;
748 
749 	__be16			protocol;
750 	__u16			transport_header;
751 	__u16			network_header;
752 	__u16			mac_header;
753 
754 	/* private: */
755 	__u32			headers_end[0];
756 	/* public: */
757 
758 	/* These elements must be at the end, see alloc_skb() for details.  */
759 	sk_buff_data_t		tail;
760 	sk_buff_data_t		end;
761 	unsigned char		*head,
762 				*data;
763 	unsigned int		truesize;
764 	refcount_t		users;
765 };
766 
767 #ifdef __KERNEL__
768 /*
769  *	Handling routines are only of interest to the kernel
770  */
771 #include <linux/slab.h>
772 
773 
774 #define SKB_ALLOC_FCLONE	0x01
775 #define SKB_ALLOC_RX		0x02
776 #define SKB_ALLOC_NAPI		0x04
777 
778 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
779 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
780 {
781 	return unlikely(skb->pfmemalloc);
782 }
783 
784 /*
785  * skb might have a dst pointer attached, refcounted or not.
786  * _skb_refdst low order bit is set if refcount was _not_ taken
787  */
788 #define SKB_DST_NOREF	1UL
789 #define SKB_DST_PTRMASK	~(SKB_DST_NOREF)
790 
791 #define SKB_NFCT_PTRMASK	~(7UL)
792 /**
793  * skb_dst - returns skb dst_entry
794  * @skb: buffer
795  *
796  * Returns skb dst_entry, regardless of reference taken or not.
797  */
798 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
799 {
800 	/* If refdst was not refcounted, check we still are in a
801 	 * rcu_read_lock section
802 	 */
803 	WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
804 		!rcu_read_lock_held() &&
805 		!rcu_read_lock_bh_held());
806 	return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
807 }
808 
809 /**
810  * skb_dst_set - sets skb dst
811  * @skb: buffer
812  * @dst: dst entry
813  *
814  * Sets skb dst, assuming a reference was taken on dst and should
815  * be released by skb_dst_drop()
816  */
817 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
818 {
819 	skb->_skb_refdst = (unsigned long)dst;
820 }
821 
822 /**
823  * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
824  * @skb: buffer
825  * @dst: dst entry
826  *
827  * Sets skb dst, assuming a reference was not taken on dst.
828  * If dst entry is cached, we do not take reference and dst_release
829  * will be avoided by refdst_drop. If dst entry is not cached, we take
830  * reference, so that last dst_release can destroy the dst immediately.
831  */
832 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
833 {
834 	WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
835 	skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
836 }
837 
838 /**
839  * skb_dst_is_noref - Test if skb dst isn't refcounted
840  * @skb: buffer
841  */
842 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
843 {
844 	return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
845 }
846 
847 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
848 {
849 	return (struct rtable *)skb_dst(skb);
850 }
851 
852 /* For mangling skb->pkt_type from user space side from applications
853  * such as nft, tc, etc, we only allow a conservative subset of
854  * possible pkt_types to be set.
855 */
856 static inline bool skb_pkt_type_ok(u32 ptype)
857 {
858 	return ptype <= PACKET_OTHERHOST;
859 }
860 
861 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
862 {
863 #ifdef CONFIG_NET_RX_BUSY_POLL
864 	return skb->napi_id;
865 #else
866 	return 0;
867 #endif
868 }
869 
870 /* decrement the reference count and return true if we can free the skb */
871 static inline bool skb_unref(struct sk_buff *skb)
872 {
873 	if (unlikely(!skb))
874 		return false;
875 	if (likely(refcount_read(&skb->users) == 1))
876 		smp_rmb();
877 	else if (likely(!refcount_dec_and_test(&skb->users)))
878 		return false;
879 
880 	return true;
881 }
882 
883 void skb_release_head_state(struct sk_buff *skb);
884 void kfree_skb(struct sk_buff *skb);
885 void kfree_skb_list(struct sk_buff *segs);
886 void skb_tx_error(struct sk_buff *skb);
887 void consume_skb(struct sk_buff *skb);
888 void consume_stateless_skb(struct sk_buff *skb);
889 void  __kfree_skb(struct sk_buff *skb);
890 extern struct kmem_cache *skbuff_head_cache;
891 
892 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
893 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
894 		      bool *fragstolen, int *delta_truesize);
895 
896 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
897 			    int node);
898 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
899 struct sk_buff *build_skb(void *data, unsigned int frag_size);
900 static inline struct sk_buff *alloc_skb(unsigned int size,
901 					gfp_t priority)
902 {
903 	return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
904 }
905 
906 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
907 				     unsigned long data_len,
908 				     int max_page_order,
909 				     int *errcode,
910 				     gfp_t gfp_mask);
911 
912 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
913 struct sk_buff_fclones {
914 	struct sk_buff	skb1;
915 
916 	struct sk_buff	skb2;
917 
918 	refcount_t	fclone_ref;
919 };
920 
921 /**
922  *	skb_fclone_busy - check if fclone is busy
923  *	@sk: socket
924  *	@skb: buffer
925  *
926  * Returns true if skb is a fast clone, and its clone is not freed.
927  * Some drivers call skb_orphan() in their ndo_start_xmit(),
928  * so we also check that this didnt happen.
929  */
930 static inline bool skb_fclone_busy(const struct sock *sk,
931 				   const struct sk_buff *skb)
932 {
933 	const struct sk_buff_fclones *fclones;
934 
935 	fclones = container_of(skb, struct sk_buff_fclones, skb1);
936 
937 	return skb->fclone == SKB_FCLONE_ORIG &&
938 	       refcount_read(&fclones->fclone_ref) > 1 &&
939 	       fclones->skb2.sk == sk;
940 }
941 
942 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
943 					       gfp_t priority)
944 {
945 	return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
946 }
947 
948 struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
949 static inline struct sk_buff *alloc_skb_head(gfp_t priority)
950 {
951 	return __alloc_skb_head(priority, -1);
952 }
953 
954 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
955 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
956 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
957 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
958 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
959 				   gfp_t gfp_mask, bool fclone);
960 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
961 					  gfp_t gfp_mask)
962 {
963 	return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
964 }
965 
966 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
967 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
968 				     unsigned int headroom);
969 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
970 				int newtailroom, gfp_t priority);
971 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
972 				     int offset, int len);
973 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
974 			      int offset, int len);
975 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
976 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
977 
978 /**
979  *	skb_pad			-	zero pad the tail of an skb
980  *	@skb: buffer to pad
981  *	@pad: space to pad
982  *
983  *	Ensure that a buffer is followed by a padding area that is zero
984  *	filled. Used by network drivers which may DMA or transfer data
985  *	beyond the buffer end onto the wire.
986  *
987  *	May return error in out of memory cases. The skb is freed on error.
988  */
989 static inline int skb_pad(struct sk_buff *skb, int pad)
990 {
991 	return __skb_pad(skb, pad, true);
992 }
993 #define dev_kfree_skb(a)	consume_skb(a)
994 
995 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
996 			    int getfrag(void *from, char *to, int offset,
997 					int len, int odd, struct sk_buff *skb),
998 			    void *from, int length);
999 
1000 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1001 			 int offset, size_t size);
1002 
1003 struct skb_seq_state {
1004 	__u32		lower_offset;
1005 	__u32		upper_offset;
1006 	__u32		frag_idx;
1007 	__u32		stepped_offset;
1008 	struct sk_buff	*root_skb;
1009 	struct sk_buff	*cur_skb;
1010 	__u8		*frag_data;
1011 };
1012 
1013 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1014 			  unsigned int to, struct skb_seq_state *st);
1015 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1016 			  struct skb_seq_state *st);
1017 void skb_abort_seq_read(struct skb_seq_state *st);
1018 
1019 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1020 			   unsigned int to, struct ts_config *config);
1021 
1022 /*
1023  * Packet hash types specify the type of hash in skb_set_hash.
1024  *
1025  * Hash types refer to the protocol layer addresses which are used to
1026  * construct a packet's hash. The hashes are used to differentiate or identify
1027  * flows of the protocol layer for the hash type. Hash types are either
1028  * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1029  *
1030  * Properties of hashes:
1031  *
1032  * 1) Two packets in different flows have different hash values
1033  * 2) Two packets in the same flow should have the same hash value
1034  *
1035  * A hash at a higher layer is considered to be more specific. A driver should
1036  * set the most specific hash possible.
1037  *
1038  * A driver cannot indicate a more specific hash than the layer at which a hash
1039  * was computed. For instance an L3 hash cannot be set as an L4 hash.
1040  *
1041  * A driver may indicate a hash level which is less specific than the
1042  * actual layer the hash was computed on. For instance, a hash computed
1043  * at L4 may be considered an L3 hash. This should only be done if the
1044  * driver can't unambiguously determine that the HW computed the hash at
1045  * the higher layer. Note that the "should" in the second property above
1046  * permits this.
1047  */
1048 enum pkt_hash_types {
1049 	PKT_HASH_TYPE_NONE,	/* Undefined type */
1050 	PKT_HASH_TYPE_L2,	/* Input: src_MAC, dest_MAC */
1051 	PKT_HASH_TYPE_L3,	/* Input: src_IP, dst_IP */
1052 	PKT_HASH_TYPE_L4,	/* Input: src_IP, dst_IP, src_port, dst_port */
1053 };
1054 
1055 static inline void skb_clear_hash(struct sk_buff *skb)
1056 {
1057 	skb->hash = 0;
1058 	skb->sw_hash = 0;
1059 	skb->l4_hash = 0;
1060 }
1061 
1062 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1063 {
1064 	if (!skb->l4_hash)
1065 		skb_clear_hash(skb);
1066 }
1067 
1068 static inline void
1069 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1070 {
1071 	skb->l4_hash = is_l4;
1072 	skb->sw_hash = is_sw;
1073 	skb->hash = hash;
1074 }
1075 
1076 static inline void
1077 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1078 {
1079 	/* Used by drivers to set hash from HW */
1080 	__skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1081 }
1082 
1083 static inline void
1084 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1085 {
1086 	__skb_set_hash(skb, hash, true, is_l4);
1087 }
1088 
1089 void __skb_get_hash(struct sk_buff *skb);
1090 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1091 u32 skb_get_poff(const struct sk_buff *skb);
1092 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1093 		   const struct flow_keys *keys, int hlen);
1094 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1095 			    void *data, int hlen_proto);
1096 
1097 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1098 					int thoff, u8 ip_proto)
1099 {
1100 	return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1101 }
1102 
1103 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1104 			     const struct flow_dissector_key *key,
1105 			     unsigned int key_count);
1106 
1107 bool __skb_flow_dissect(const struct sk_buff *skb,
1108 			struct flow_dissector *flow_dissector,
1109 			void *target_container,
1110 			void *data, __be16 proto, int nhoff, int hlen,
1111 			unsigned int flags);
1112 
1113 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1114 				    struct flow_dissector *flow_dissector,
1115 				    void *target_container, unsigned int flags)
1116 {
1117 	return __skb_flow_dissect(skb, flow_dissector, target_container,
1118 				  NULL, 0, 0, 0, flags);
1119 }
1120 
1121 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1122 					      struct flow_keys *flow,
1123 					      unsigned int flags)
1124 {
1125 	memset(flow, 0, sizeof(*flow));
1126 	return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1127 				  NULL, 0, 0, 0, flags);
1128 }
1129 
1130 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1131 						  void *data, __be16 proto,
1132 						  int nhoff, int hlen,
1133 						  unsigned int flags)
1134 {
1135 	memset(flow, 0, sizeof(*flow));
1136 	return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1137 				  data, proto, nhoff, hlen, flags);
1138 }
1139 
1140 static inline __u32 skb_get_hash(struct sk_buff *skb)
1141 {
1142 	if (!skb->l4_hash && !skb->sw_hash)
1143 		__skb_get_hash(skb);
1144 
1145 	return skb->hash;
1146 }
1147 
1148 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1149 
1150 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1151 {
1152 	if (!skb->l4_hash && !skb->sw_hash) {
1153 		struct flow_keys keys;
1154 		__u32 hash = __get_hash_from_flowi6(fl6, &keys);
1155 
1156 		__skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1157 	}
1158 
1159 	return skb->hash;
1160 }
1161 
1162 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1163 
1164 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1165 {
1166 	if (!skb->l4_hash && !skb->sw_hash) {
1167 		struct flow_keys keys;
1168 		__u32 hash = __get_hash_from_flowi4(fl4, &keys);
1169 
1170 		__skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1171 	}
1172 
1173 	return skb->hash;
1174 }
1175 
1176 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1177 
1178 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1179 {
1180 	return skb->hash;
1181 }
1182 
1183 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1184 {
1185 	to->hash = from->hash;
1186 	to->sw_hash = from->sw_hash;
1187 	to->l4_hash = from->l4_hash;
1188 };
1189 
1190 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1191 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1192 {
1193 	return skb->head + skb->end;
1194 }
1195 
1196 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1197 {
1198 	return skb->end;
1199 }
1200 #else
1201 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1202 {
1203 	return skb->end;
1204 }
1205 
1206 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1207 {
1208 	return skb->end - skb->head;
1209 }
1210 #endif
1211 
1212 /* Internal */
1213 #define skb_shinfo(SKB)	((struct skb_shared_info *)(skb_end_pointer(SKB)))
1214 
1215 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1216 {
1217 	return &skb_shinfo(skb)->hwtstamps;
1218 }
1219 
1220 /**
1221  *	skb_queue_empty - check if a queue is empty
1222  *	@list: queue head
1223  *
1224  *	Returns true if the queue is empty, false otherwise.
1225  */
1226 static inline int skb_queue_empty(const struct sk_buff_head *list)
1227 {
1228 	return list->next == (const struct sk_buff *) list;
1229 }
1230 
1231 /**
1232  *	skb_queue_is_last - check if skb is the last entry in the queue
1233  *	@list: queue head
1234  *	@skb: buffer
1235  *
1236  *	Returns true if @skb is the last buffer on the list.
1237  */
1238 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1239 				     const struct sk_buff *skb)
1240 {
1241 	return skb->next == (const struct sk_buff *) list;
1242 }
1243 
1244 /**
1245  *	skb_queue_is_first - check if skb is the first entry in the queue
1246  *	@list: queue head
1247  *	@skb: buffer
1248  *
1249  *	Returns true if @skb is the first buffer on the list.
1250  */
1251 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1252 				      const struct sk_buff *skb)
1253 {
1254 	return skb->prev == (const struct sk_buff *) list;
1255 }
1256 
1257 /**
1258  *	skb_queue_next - return the next packet in the queue
1259  *	@list: queue head
1260  *	@skb: current buffer
1261  *
1262  *	Return the next packet in @list after @skb.  It is only valid to
1263  *	call this if skb_queue_is_last() evaluates to false.
1264  */
1265 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1266 					     const struct sk_buff *skb)
1267 {
1268 	/* This BUG_ON may seem severe, but if we just return then we
1269 	 * are going to dereference garbage.
1270 	 */
1271 	BUG_ON(skb_queue_is_last(list, skb));
1272 	return skb->next;
1273 }
1274 
1275 /**
1276  *	skb_queue_prev - return the prev packet in the queue
1277  *	@list: queue head
1278  *	@skb: current buffer
1279  *
1280  *	Return the prev packet in @list before @skb.  It is only valid to
1281  *	call this if skb_queue_is_first() evaluates to false.
1282  */
1283 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1284 					     const struct sk_buff *skb)
1285 {
1286 	/* This BUG_ON may seem severe, but if we just return then we
1287 	 * are going to dereference garbage.
1288 	 */
1289 	BUG_ON(skb_queue_is_first(list, skb));
1290 	return skb->prev;
1291 }
1292 
1293 /**
1294  *	skb_get - reference buffer
1295  *	@skb: buffer to reference
1296  *
1297  *	Makes another reference to a socket buffer and returns a pointer
1298  *	to the buffer.
1299  */
1300 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1301 {
1302 	refcount_inc(&skb->users);
1303 	return skb;
1304 }
1305 
1306 /*
1307  * If users == 1, we are the only owner and are can avoid redundant
1308  * atomic change.
1309  */
1310 
1311 /**
1312  *	skb_cloned - is the buffer a clone
1313  *	@skb: buffer to check
1314  *
1315  *	Returns true if the buffer was generated with skb_clone() and is
1316  *	one of multiple shared copies of the buffer. Cloned buffers are
1317  *	shared data so must not be written to under normal circumstances.
1318  */
1319 static inline int skb_cloned(const struct sk_buff *skb)
1320 {
1321 	return skb->cloned &&
1322 	       (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1323 }
1324 
1325 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1326 {
1327 	might_sleep_if(gfpflags_allow_blocking(pri));
1328 
1329 	if (skb_cloned(skb))
1330 		return pskb_expand_head(skb, 0, 0, pri);
1331 
1332 	return 0;
1333 }
1334 
1335 /**
1336  *	skb_header_cloned - is the header a clone
1337  *	@skb: buffer to check
1338  *
1339  *	Returns true if modifying the header part of the buffer requires
1340  *	the data to be copied.
1341  */
1342 static inline int skb_header_cloned(const struct sk_buff *skb)
1343 {
1344 	int dataref;
1345 
1346 	if (!skb->cloned)
1347 		return 0;
1348 
1349 	dataref = atomic_read(&skb_shinfo(skb)->dataref);
1350 	dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1351 	return dataref != 1;
1352 }
1353 
1354 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1355 {
1356 	might_sleep_if(gfpflags_allow_blocking(pri));
1357 
1358 	if (skb_header_cloned(skb))
1359 		return pskb_expand_head(skb, 0, 0, pri);
1360 
1361 	return 0;
1362 }
1363 
1364 /**
1365  *	skb_header_release - release reference to header
1366  *	@skb: buffer to operate on
1367  *
1368  *	Drop a reference to the header part of the buffer.  This is done
1369  *	by acquiring a payload reference.  You must not read from the header
1370  *	part of skb->data after this.
1371  *	Note : Check if you can use __skb_header_release() instead.
1372  */
1373 static inline void skb_header_release(struct sk_buff *skb)
1374 {
1375 	BUG_ON(skb->nohdr);
1376 	skb->nohdr = 1;
1377 	atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1378 }
1379 
1380 /**
1381  *	__skb_header_release - release reference to header
1382  *	@skb: buffer to operate on
1383  *
1384  *	Variant of skb_header_release() assuming skb is private to caller.
1385  *	We can avoid one atomic operation.
1386  */
1387 static inline void __skb_header_release(struct sk_buff *skb)
1388 {
1389 	skb->nohdr = 1;
1390 	atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1391 }
1392 
1393 
1394 /**
1395  *	skb_shared - is the buffer shared
1396  *	@skb: buffer to check
1397  *
1398  *	Returns true if more than one person has a reference to this
1399  *	buffer.
1400  */
1401 static inline int skb_shared(const struct sk_buff *skb)
1402 {
1403 	return refcount_read(&skb->users) != 1;
1404 }
1405 
1406 /**
1407  *	skb_share_check - check if buffer is shared and if so clone it
1408  *	@skb: buffer to check
1409  *	@pri: priority for memory allocation
1410  *
1411  *	If the buffer is shared the buffer is cloned and the old copy
1412  *	drops a reference. A new clone with a single reference is returned.
1413  *	If the buffer is not shared the original buffer is returned. When
1414  *	being called from interrupt status or with spinlocks held pri must
1415  *	be GFP_ATOMIC.
1416  *
1417  *	NULL is returned on a memory allocation failure.
1418  */
1419 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1420 {
1421 	might_sleep_if(gfpflags_allow_blocking(pri));
1422 	if (skb_shared(skb)) {
1423 		struct sk_buff *nskb = skb_clone(skb, pri);
1424 
1425 		if (likely(nskb))
1426 			consume_skb(skb);
1427 		else
1428 			kfree_skb(skb);
1429 		skb = nskb;
1430 	}
1431 	return skb;
1432 }
1433 
1434 /*
1435  *	Copy shared buffers into a new sk_buff. We effectively do COW on
1436  *	packets to handle cases where we have a local reader and forward
1437  *	and a couple of other messy ones. The normal one is tcpdumping
1438  *	a packet thats being forwarded.
1439  */
1440 
1441 /**
1442  *	skb_unshare - make a copy of a shared buffer
1443  *	@skb: buffer to check
1444  *	@pri: priority for memory allocation
1445  *
1446  *	If the socket buffer is a clone then this function creates a new
1447  *	copy of the data, drops a reference count on the old copy and returns
1448  *	the new copy with the reference count at 1. If the buffer is not a clone
1449  *	the original buffer is returned. When called with a spinlock held or
1450  *	from interrupt state @pri must be %GFP_ATOMIC
1451  *
1452  *	%NULL is returned on a memory allocation failure.
1453  */
1454 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1455 					  gfp_t pri)
1456 {
1457 	might_sleep_if(gfpflags_allow_blocking(pri));
1458 	if (skb_cloned(skb)) {
1459 		struct sk_buff *nskb = skb_copy(skb, pri);
1460 
1461 		/* Free our shared copy */
1462 		if (likely(nskb))
1463 			consume_skb(skb);
1464 		else
1465 			kfree_skb(skb);
1466 		skb = nskb;
1467 	}
1468 	return skb;
1469 }
1470 
1471 /**
1472  *	skb_peek - peek at the head of an &sk_buff_head
1473  *	@list_: list to peek at
1474  *
1475  *	Peek an &sk_buff. Unlike most other operations you _MUST_
1476  *	be careful with this one. A peek leaves the buffer on the
1477  *	list and someone else may run off with it. You must hold
1478  *	the appropriate locks or have a private queue to do this.
1479  *
1480  *	Returns %NULL for an empty list or a pointer to the head element.
1481  *	The reference count is not incremented and the reference is therefore
1482  *	volatile. Use with caution.
1483  */
1484 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1485 {
1486 	struct sk_buff *skb = list_->next;
1487 
1488 	if (skb == (struct sk_buff *)list_)
1489 		skb = NULL;
1490 	return skb;
1491 }
1492 
1493 /**
1494  *	skb_peek_next - peek skb following the given one from a queue
1495  *	@skb: skb to start from
1496  *	@list_: list to peek at
1497  *
1498  *	Returns %NULL when the end of the list is met or a pointer to the
1499  *	next element. The reference count is not incremented and the
1500  *	reference is therefore volatile. Use with caution.
1501  */
1502 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1503 		const struct sk_buff_head *list_)
1504 {
1505 	struct sk_buff *next = skb->next;
1506 
1507 	if (next == (struct sk_buff *)list_)
1508 		next = NULL;
1509 	return next;
1510 }
1511 
1512 /**
1513  *	skb_peek_tail - peek at the tail of an &sk_buff_head
1514  *	@list_: list to peek at
1515  *
1516  *	Peek an &sk_buff. Unlike most other operations you _MUST_
1517  *	be careful with this one. A peek leaves the buffer on the
1518  *	list and someone else may run off with it. You must hold
1519  *	the appropriate locks or have a private queue to do this.
1520  *
1521  *	Returns %NULL for an empty list or a pointer to the tail element.
1522  *	The reference count is not incremented and the reference is therefore
1523  *	volatile. Use with caution.
1524  */
1525 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1526 {
1527 	struct sk_buff *skb = list_->prev;
1528 
1529 	if (skb == (struct sk_buff *)list_)
1530 		skb = NULL;
1531 	return skb;
1532 
1533 }
1534 
1535 /**
1536  *	skb_queue_len	- get queue length
1537  *	@list_: list to measure
1538  *
1539  *	Return the length of an &sk_buff queue.
1540  */
1541 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1542 {
1543 	return list_->qlen;
1544 }
1545 
1546 /**
1547  *	__skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1548  *	@list: queue to initialize
1549  *
1550  *	This initializes only the list and queue length aspects of
1551  *	an sk_buff_head object.  This allows to initialize the list
1552  *	aspects of an sk_buff_head without reinitializing things like
1553  *	the spinlock.  It can also be used for on-stack sk_buff_head
1554  *	objects where the spinlock is known to not be used.
1555  */
1556 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1557 {
1558 	list->prev = list->next = (struct sk_buff *)list;
1559 	list->qlen = 0;
1560 }
1561 
1562 /*
1563  * This function creates a split out lock class for each invocation;
1564  * this is needed for now since a whole lot of users of the skb-queue
1565  * infrastructure in drivers have different locking usage (in hardirq)
1566  * than the networking core (in softirq only). In the long run either the
1567  * network layer or drivers should need annotation to consolidate the
1568  * main types of usage into 3 classes.
1569  */
1570 static inline void skb_queue_head_init(struct sk_buff_head *list)
1571 {
1572 	spin_lock_init(&list->lock);
1573 	__skb_queue_head_init(list);
1574 }
1575 
1576 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1577 		struct lock_class_key *class)
1578 {
1579 	skb_queue_head_init(list);
1580 	lockdep_set_class(&list->lock, class);
1581 }
1582 
1583 /*
1584  *	Insert an sk_buff on a list.
1585  *
1586  *	The "__skb_xxxx()" functions are the non-atomic ones that
1587  *	can only be called with interrupts disabled.
1588  */
1589 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1590 		struct sk_buff_head *list);
1591 static inline void __skb_insert(struct sk_buff *newsk,
1592 				struct sk_buff *prev, struct sk_buff *next,
1593 				struct sk_buff_head *list)
1594 {
1595 	newsk->next = next;
1596 	newsk->prev = prev;
1597 	next->prev  = prev->next = newsk;
1598 	list->qlen++;
1599 }
1600 
1601 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1602 				      struct sk_buff *prev,
1603 				      struct sk_buff *next)
1604 {
1605 	struct sk_buff *first = list->next;
1606 	struct sk_buff *last = list->prev;
1607 
1608 	first->prev = prev;
1609 	prev->next = first;
1610 
1611 	last->next = next;
1612 	next->prev = last;
1613 }
1614 
1615 /**
1616  *	skb_queue_splice - join two skb lists, this is designed for stacks
1617  *	@list: the new list to add
1618  *	@head: the place to add it in the first list
1619  */
1620 static inline void skb_queue_splice(const struct sk_buff_head *list,
1621 				    struct sk_buff_head *head)
1622 {
1623 	if (!skb_queue_empty(list)) {
1624 		__skb_queue_splice(list, (struct sk_buff *) head, head->next);
1625 		head->qlen += list->qlen;
1626 	}
1627 }
1628 
1629 /**
1630  *	skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1631  *	@list: the new list to add
1632  *	@head: the place to add it in the first list
1633  *
1634  *	The list at @list is reinitialised
1635  */
1636 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1637 					 struct sk_buff_head *head)
1638 {
1639 	if (!skb_queue_empty(list)) {
1640 		__skb_queue_splice(list, (struct sk_buff *) head, head->next);
1641 		head->qlen += list->qlen;
1642 		__skb_queue_head_init(list);
1643 	}
1644 }
1645 
1646 /**
1647  *	skb_queue_splice_tail - join two skb lists, each list being a queue
1648  *	@list: the new list to add
1649  *	@head: the place to add it in the first list
1650  */
1651 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1652 					 struct sk_buff_head *head)
1653 {
1654 	if (!skb_queue_empty(list)) {
1655 		__skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1656 		head->qlen += list->qlen;
1657 	}
1658 }
1659 
1660 /**
1661  *	skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1662  *	@list: the new list to add
1663  *	@head: the place to add it in the first list
1664  *
1665  *	Each of the lists is a queue.
1666  *	The list at @list is reinitialised
1667  */
1668 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1669 					      struct sk_buff_head *head)
1670 {
1671 	if (!skb_queue_empty(list)) {
1672 		__skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1673 		head->qlen += list->qlen;
1674 		__skb_queue_head_init(list);
1675 	}
1676 }
1677 
1678 /**
1679  *	__skb_queue_after - queue a buffer at the list head
1680  *	@list: list to use
1681  *	@prev: place after this buffer
1682  *	@newsk: buffer to queue
1683  *
1684  *	Queue a buffer int the middle of a list. This function takes no locks
1685  *	and you must therefore hold required locks before calling it.
1686  *
1687  *	A buffer cannot be placed on two lists at the same time.
1688  */
1689 static inline void __skb_queue_after(struct sk_buff_head *list,
1690 				     struct sk_buff *prev,
1691 				     struct sk_buff *newsk)
1692 {
1693 	__skb_insert(newsk, prev, prev->next, list);
1694 }
1695 
1696 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1697 		struct sk_buff_head *list);
1698 
1699 static inline void __skb_queue_before(struct sk_buff_head *list,
1700 				      struct sk_buff *next,
1701 				      struct sk_buff *newsk)
1702 {
1703 	__skb_insert(newsk, next->prev, next, list);
1704 }
1705 
1706 /**
1707  *	__skb_queue_head - queue a buffer at the list head
1708  *	@list: list to use
1709  *	@newsk: buffer to queue
1710  *
1711  *	Queue a buffer at the start of a list. This function takes no locks
1712  *	and you must therefore hold required locks before calling it.
1713  *
1714  *	A buffer cannot be placed on two lists at the same time.
1715  */
1716 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1717 static inline void __skb_queue_head(struct sk_buff_head *list,
1718 				    struct sk_buff *newsk)
1719 {
1720 	__skb_queue_after(list, (struct sk_buff *)list, newsk);
1721 }
1722 
1723 /**
1724  *	__skb_queue_tail - queue a buffer at the list tail
1725  *	@list: list to use
1726  *	@newsk: buffer to queue
1727  *
1728  *	Queue a buffer at the end of a list. This function takes no locks
1729  *	and you must therefore hold required locks before calling it.
1730  *
1731  *	A buffer cannot be placed on two lists at the same time.
1732  */
1733 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1734 static inline void __skb_queue_tail(struct sk_buff_head *list,
1735 				   struct sk_buff *newsk)
1736 {
1737 	__skb_queue_before(list, (struct sk_buff *)list, newsk);
1738 }
1739 
1740 /*
1741  * remove sk_buff from list. _Must_ be called atomically, and with
1742  * the list known..
1743  */
1744 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1745 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1746 {
1747 	struct sk_buff *next, *prev;
1748 
1749 	list->qlen--;
1750 	next	   = skb->next;
1751 	prev	   = skb->prev;
1752 	skb->next  = skb->prev = NULL;
1753 	next->prev = prev;
1754 	prev->next = next;
1755 }
1756 
1757 /**
1758  *	__skb_dequeue - remove from the head of the queue
1759  *	@list: list to dequeue from
1760  *
1761  *	Remove the head of the list. This function does not take any locks
1762  *	so must be used with appropriate locks held only. The head item is
1763  *	returned or %NULL if the list is empty.
1764  */
1765 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1766 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1767 {
1768 	struct sk_buff *skb = skb_peek(list);
1769 	if (skb)
1770 		__skb_unlink(skb, list);
1771 	return skb;
1772 }
1773 
1774 /**
1775  *	__skb_dequeue_tail - remove from the tail of the queue
1776  *	@list: list to dequeue from
1777  *
1778  *	Remove the tail of the list. This function does not take any locks
1779  *	so must be used with appropriate locks held only. The tail item is
1780  *	returned or %NULL if the list is empty.
1781  */
1782 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1783 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1784 {
1785 	struct sk_buff *skb = skb_peek_tail(list);
1786 	if (skb)
1787 		__skb_unlink(skb, list);
1788 	return skb;
1789 }
1790 
1791 
1792 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1793 {
1794 	return skb->data_len;
1795 }
1796 
1797 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1798 {
1799 	return skb->len - skb->data_len;
1800 }
1801 
1802 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1803 {
1804 	unsigned int i, len = 0;
1805 
1806 	for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1807 		len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1808 	return len + skb_headlen(skb);
1809 }
1810 
1811 /**
1812  * __skb_fill_page_desc - initialise a paged fragment in an skb
1813  * @skb: buffer containing fragment to be initialised
1814  * @i: paged fragment index to initialise
1815  * @page: the page to use for this fragment
1816  * @off: the offset to the data with @page
1817  * @size: the length of the data
1818  *
1819  * Initialises the @i'th fragment of @skb to point to &size bytes at
1820  * offset @off within @page.
1821  *
1822  * Does not take any additional reference on the fragment.
1823  */
1824 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1825 					struct page *page, int off, int size)
1826 {
1827 	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1828 
1829 	/*
1830 	 * Propagate page pfmemalloc to the skb if we can. The problem is
1831 	 * that not all callers have unique ownership of the page but rely
1832 	 * on page_is_pfmemalloc doing the right thing(tm).
1833 	 */
1834 	frag->page.p		  = page;
1835 	frag->page_offset	  = off;
1836 	skb_frag_size_set(frag, size);
1837 
1838 	page = compound_head(page);
1839 	if (page_is_pfmemalloc(page))
1840 		skb->pfmemalloc	= true;
1841 }
1842 
1843 /**
1844  * skb_fill_page_desc - initialise a paged fragment in an skb
1845  * @skb: buffer containing fragment to be initialised
1846  * @i: paged fragment index to initialise
1847  * @page: the page to use for this fragment
1848  * @off: the offset to the data with @page
1849  * @size: the length of the data
1850  *
1851  * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1852  * @skb to point to @size bytes at offset @off within @page. In
1853  * addition updates @skb such that @i is the last fragment.
1854  *
1855  * Does not take any additional reference on the fragment.
1856  */
1857 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1858 				      struct page *page, int off, int size)
1859 {
1860 	__skb_fill_page_desc(skb, i, page, off, size);
1861 	skb_shinfo(skb)->nr_frags = i + 1;
1862 }
1863 
1864 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1865 		     int size, unsigned int truesize);
1866 
1867 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1868 			  unsigned int truesize);
1869 
1870 #define SKB_PAGE_ASSERT(skb) 	BUG_ON(skb_shinfo(skb)->nr_frags)
1871 #define SKB_FRAG_ASSERT(skb) 	BUG_ON(skb_has_frag_list(skb))
1872 #define SKB_LINEAR_ASSERT(skb)  BUG_ON(skb_is_nonlinear(skb))
1873 
1874 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1875 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1876 {
1877 	return skb->head + skb->tail;
1878 }
1879 
1880 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1881 {
1882 	skb->tail = skb->data - skb->head;
1883 }
1884 
1885 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1886 {
1887 	skb_reset_tail_pointer(skb);
1888 	skb->tail += offset;
1889 }
1890 
1891 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1892 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1893 {
1894 	return skb->tail;
1895 }
1896 
1897 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1898 {
1899 	skb->tail = skb->data;
1900 }
1901 
1902 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1903 {
1904 	skb->tail = skb->data + offset;
1905 }
1906 
1907 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1908 
1909 /*
1910  *	Add data to an sk_buff
1911  */
1912 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1913 void *skb_put(struct sk_buff *skb, unsigned int len);
1914 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
1915 {
1916 	void *tmp = skb_tail_pointer(skb);
1917 	SKB_LINEAR_ASSERT(skb);
1918 	skb->tail += len;
1919 	skb->len  += len;
1920 	return tmp;
1921 }
1922 
1923 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
1924 {
1925 	void *tmp = __skb_put(skb, len);
1926 
1927 	memset(tmp, 0, len);
1928 	return tmp;
1929 }
1930 
1931 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
1932 				   unsigned int len)
1933 {
1934 	void *tmp = __skb_put(skb, len);
1935 
1936 	memcpy(tmp, data, len);
1937 	return tmp;
1938 }
1939 
1940 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
1941 {
1942 	*(u8 *)__skb_put(skb, 1) = val;
1943 }
1944 
1945 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
1946 {
1947 	void *tmp = skb_put(skb, len);
1948 
1949 	memset(tmp, 0, len);
1950 
1951 	return tmp;
1952 }
1953 
1954 static inline void *skb_put_data(struct sk_buff *skb, const void *data,
1955 				 unsigned int len)
1956 {
1957 	void *tmp = skb_put(skb, len);
1958 
1959 	memcpy(tmp, data, len);
1960 
1961 	return tmp;
1962 }
1963 
1964 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
1965 {
1966 	*(u8 *)skb_put(skb, 1) = val;
1967 }
1968 
1969 void *skb_push(struct sk_buff *skb, unsigned int len);
1970 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
1971 {
1972 	skb->data -= len;
1973 	skb->len  += len;
1974 	return skb->data;
1975 }
1976 
1977 void *skb_pull(struct sk_buff *skb, unsigned int len);
1978 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
1979 {
1980 	skb->len -= len;
1981 	BUG_ON(skb->len < skb->data_len);
1982 	return skb->data += len;
1983 }
1984 
1985 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1986 {
1987 	return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1988 }
1989 
1990 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
1991 
1992 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
1993 {
1994 	if (len > skb_headlen(skb) &&
1995 	    !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1996 		return NULL;
1997 	skb->len -= len;
1998 	return skb->data += len;
1999 }
2000 
2001 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2002 {
2003 	return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2004 }
2005 
2006 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2007 {
2008 	if (likely(len <= skb_headlen(skb)))
2009 		return 1;
2010 	if (unlikely(len > skb->len))
2011 		return 0;
2012 	return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2013 }
2014 
2015 void skb_condense(struct sk_buff *skb);
2016 
2017 /**
2018  *	skb_headroom - bytes at buffer head
2019  *	@skb: buffer to check
2020  *
2021  *	Return the number of bytes of free space at the head of an &sk_buff.
2022  */
2023 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2024 {
2025 	return skb->data - skb->head;
2026 }
2027 
2028 /**
2029  *	skb_tailroom - bytes at buffer end
2030  *	@skb: buffer to check
2031  *
2032  *	Return the number of bytes of free space at the tail of an sk_buff
2033  */
2034 static inline int skb_tailroom(const struct sk_buff *skb)
2035 {
2036 	return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2037 }
2038 
2039 /**
2040  *	skb_availroom - bytes at buffer end
2041  *	@skb: buffer to check
2042  *
2043  *	Return the number of bytes of free space at the tail of an sk_buff
2044  *	allocated by sk_stream_alloc()
2045  */
2046 static inline int skb_availroom(const struct sk_buff *skb)
2047 {
2048 	if (skb_is_nonlinear(skb))
2049 		return 0;
2050 
2051 	return skb->end - skb->tail - skb->reserved_tailroom;
2052 }
2053 
2054 /**
2055  *	skb_reserve - adjust headroom
2056  *	@skb: buffer to alter
2057  *	@len: bytes to move
2058  *
2059  *	Increase the headroom of an empty &sk_buff by reducing the tail
2060  *	room. This is only allowed for an empty buffer.
2061  */
2062 static inline void skb_reserve(struct sk_buff *skb, int len)
2063 {
2064 	skb->data += len;
2065 	skb->tail += len;
2066 }
2067 
2068 /**
2069  *	skb_tailroom_reserve - adjust reserved_tailroom
2070  *	@skb: buffer to alter
2071  *	@mtu: maximum amount of headlen permitted
2072  *	@needed_tailroom: minimum amount of reserved_tailroom
2073  *
2074  *	Set reserved_tailroom so that headlen can be as large as possible but
2075  *	not larger than mtu and tailroom cannot be smaller than
2076  *	needed_tailroom.
2077  *	The required headroom should already have been reserved before using
2078  *	this function.
2079  */
2080 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2081 					unsigned int needed_tailroom)
2082 {
2083 	SKB_LINEAR_ASSERT(skb);
2084 	if (mtu < skb_tailroom(skb) - needed_tailroom)
2085 		/* use at most mtu */
2086 		skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2087 	else
2088 		/* use up to all available space */
2089 		skb->reserved_tailroom = needed_tailroom;
2090 }
2091 
2092 #define ENCAP_TYPE_ETHER	0
2093 #define ENCAP_TYPE_IPPROTO	1
2094 
2095 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2096 					  __be16 protocol)
2097 {
2098 	skb->inner_protocol = protocol;
2099 	skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2100 }
2101 
2102 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2103 					 __u8 ipproto)
2104 {
2105 	skb->inner_ipproto = ipproto;
2106 	skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2107 }
2108 
2109 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2110 {
2111 	skb->inner_mac_header = skb->mac_header;
2112 	skb->inner_network_header = skb->network_header;
2113 	skb->inner_transport_header = skb->transport_header;
2114 }
2115 
2116 static inline void skb_reset_mac_len(struct sk_buff *skb)
2117 {
2118 	skb->mac_len = skb->network_header - skb->mac_header;
2119 }
2120 
2121 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2122 							*skb)
2123 {
2124 	return skb->head + skb->inner_transport_header;
2125 }
2126 
2127 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2128 {
2129 	return skb_inner_transport_header(skb) - skb->data;
2130 }
2131 
2132 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2133 {
2134 	skb->inner_transport_header = skb->data - skb->head;
2135 }
2136 
2137 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2138 						   const int offset)
2139 {
2140 	skb_reset_inner_transport_header(skb);
2141 	skb->inner_transport_header += offset;
2142 }
2143 
2144 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2145 {
2146 	return skb->head + skb->inner_network_header;
2147 }
2148 
2149 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2150 {
2151 	skb->inner_network_header = skb->data - skb->head;
2152 }
2153 
2154 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2155 						const int offset)
2156 {
2157 	skb_reset_inner_network_header(skb);
2158 	skb->inner_network_header += offset;
2159 }
2160 
2161 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2162 {
2163 	return skb->head + skb->inner_mac_header;
2164 }
2165 
2166 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2167 {
2168 	skb->inner_mac_header = skb->data - skb->head;
2169 }
2170 
2171 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2172 					    const int offset)
2173 {
2174 	skb_reset_inner_mac_header(skb);
2175 	skb->inner_mac_header += offset;
2176 }
2177 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2178 {
2179 	return skb->transport_header != (typeof(skb->transport_header))~0U;
2180 }
2181 
2182 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2183 {
2184 	return skb->head + skb->transport_header;
2185 }
2186 
2187 static inline void skb_reset_transport_header(struct sk_buff *skb)
2188 {
2189 	skb->transport_header = skb->data - skb->head;
2190 }
2191 
2192 static inline void skb_set_transport_header(struct sk_buff *skb,
2193 					    const int offset)
2194 {
2195 	skb_reset_transport_header(skb);
2196 	skb->transport_header += offset;
2197 }
2198 
2199 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2200 {
2201 	return skb->head + skb->network_header;
2202 }
2203 
2204 static inline void skb_reset_network_header(struct sk_buff *skb)
2205 {
2206 	skb->network_header = skb->data - skb->head;
2207 }
2208 
2209 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2210 {
2211 	skb_reset_network_header(skb);
2212 	skb->network_header += offset;
2213 }
2214 
2215 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2216 {
2217 	return skb->head + skb->mac_header;
2218 }
2219 
2220 static inline int skb_mac_offset(const struct sk_buff *skb)
2221 {
2222 	return skb_mac_header(skb) - skb->data;
2223 }
2224 
2225 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2226 {
2227 	return skb->network_header - skb->mac_header;
2228 }
2229 
2230 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2231 {
2232 	return skb->mac_header != (typeof(skb->mac_header))~0U;
2233 }
2234 
2235 static inline void skb_reset_mac_header(struct sk_buff *skb)
2236 {
2237 	skb->mac_header = skb->data - skb->head;
2238 }
2239 
2240 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2241 {
2242 	skb_reset_mac_header(skb);
2243 	skb->mac_header += offset;
2244 }
2245 
2246 static inline void skb_pop_mac_header(struct sk_buff *skb)
2247 {
2248 	skb->mac_header = skb->network_header;
2249 }
2250 
2251 static inline void skb_probe_transport_header(struct sk_buff *skb,
2252 					      const int offset_hint)
2253 {
2254 	struct flow_keys keys;
2255 
2256 	if (skb_transport_header_was_set(skb))
2257 		return;
2258 	else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2259 		skb_set_transport_header(skb, keys.control.thoff);
2260 	else
2261 		skb_set_transport_header(skb, offset_hint);
2262 }
2263 
2264 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2265 {
2266 	if (skb_mac_header_was_set(skb)) {
2267 		const unsigned char *old_mac = skb_mac_header(skb);
2268 
2269 		skb_set_mac_header(skb, -skb->mac_len);
2270 		memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2271 	}
2272 }
2273 
2274 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2275 {
2276 	return skb->csum_start - skb_headroom(skb);
2277 }
2278 
2279 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2280 {
2281 	return skb->head + skb->csum_start;
2282 }
2283 
2284 static inline int skb_transport_offset(const struct sk_buff *skb)
2285 {
2286 	return skb_transport_header(skb) - skb->data;
2287 }
2288 
2289 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2290 {
2291 	return skb->transport_header - skb->network_header;
2292 }
2293 
2294 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2295 {
2296 	return skb->inner_transport_header - skb->inner_network_header;
2297 }
2298 
2299 static inline int skb_network_offset(const struct sk_buff *skb)
2300 {
2301 	return skb_network_header(skb) - skb->data;
2302 }
2303 
2304 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2305 {
2306 	return skb_inner_network_header(skb) - skb->data;
2307 }
2308 
2309 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2310 {
2311 	return pskb_may_pull(skb, skb_network_offset(skb) + len);
2312 }
2313 
2314 /*
2315  * CPUs often take a performance hit when accessing unaligned memory
2316  * locations. The actual performance hit varies, it can be small if the
2317  * hardware handles it or large if we have to take an exception and fix it
2318  * in software.
2319  *
2320  * Since an ethernet header is 14 bytes network drivers often end up with
2321  * the IP header at an unaligned offset. The IP header can be aligned by
2322  * shifting the start of the packet by 2 bytes. Drivers should do this
2323  * with:
2324  *
2325  * skb_reserve(skb, NET_IP_ALIGN);
2326  *
2327  * The downside to this alignment of the IP header is that the DMA is now
2328  * unaligned. On some architectures the cost of an unaligned DMA is high
2329  * and this cost outweighs the gains made by aligning the IP header.
2330  *
2331  * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2332  * to be overridden.
2333  */
2334 #ifndef NET_IP_ALIGN
2335 #define NET_IP_ALIGN	2
2336 #endif
2337 
2338 /*
2339  * The networking layer reserves some headroom in skb data (via
2340  * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2341  * the header has to grow. In the default case, if the header has to grow
2342  * 32 bytes or less we avoid the reallocation.
2343  *
2344  * Unfortunately this headroom changes the DMA alignment of the resulting
2345  * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2346  * on some architectures. An architecture can override this value,
2347  * perhaps setting it to a cacheline in size (since that will maintain
2348  * cacheline alignment of the DMA). It must be a power of 2.
2349  *
2350  * Various parts of the networking layer expect at least 32 bytes of
2351  * headroom, you should not reduce this.
2352  *
2353  * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2354  * to reduce average number of cache lines per packet.
2355  * get_rps_cpus() for example only access one 64 bytes aligned block :
2356  * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2357  */
2358 #ifndef NET_SKB_PAD
2359 #define NET_SKB_PAD	max(32, L1_CACHE_BYTES)
2360 #endif
2361 
2362 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2363 
2364 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2365 {
2366 	if (unlikely(skb_is_nonlinear(skb))) {
2367 		WARN_ON(1);
2368 		return;
2369 	}
2370 	skb->len = len;
2371 	skb_set_tail_pointer(skb, len);
2372 }
2373 
2374 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2375 {
2376 	__skb_set_length(skb, len);
2377 }
2378 
2379 void skb_trim(struct sk_buff *skb, unsigned int len);
2380 
2381 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2382 {
2383 	if (skb->data_len)
2384 		return ___pskb_trim(skb, len);
2385 	__skb_trim(skb, len);
2386 	return 0;
2387 }
2388 
2389 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2390 {
2391 	return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2392 }
2393 
2394 /**
2395  *	pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2396  *	@skb: buffer to alter
2397  *	@len: new length
2398  *
2399  *	This is identical to pskb_trim except that the caller knows that
2400  *	the skb is not cloned so we should never get an error due to out-
2401  *	of-memory.
2402  */
2403 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2404 {
2405 	int err = pskb_trim(skb, len);
2406 	BUG_ON(err);
2407 }
2408 
2409 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2410 {
2411 	unsigned int diff = len - skb->len;
2412 
2413 	if (skb_tailroom(skb) < diff) {
2414 		int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2415 					   GFP_ATOMIC);
2416 		if (ret)
2417 			return ret;
2418 	}
2419 	__skb_set_length(skb, len);
2420 	return 0;
2421 }
2422 
2423 /**
2424  *	skb_orphan - orphan a buffer
2425  *	@skb: buffer to orphan
2426  *
2427  *	If a buffer currently has an owner then we call the owner's
2428  *	destructor function and make the @skb unowned. The buffer continues
2429  *	to exist but is no longer charged to its former owner.
2430  */
2431 static inline void skb_orphan(struct sk_buff *skb)
2432 {
2433 	if (skb->destructor) {
2434 		skb->destructor(skb);
2435 		skb->destructor = NULL;
2436 		skb->sk		= NULL;
2437 	} else {
2438 		BUG_ON(skb->sk);
2439 	}
2440 }
2441 
2442 /**
2443  *	skb_orphan_frags - orphan the frags contained in a buffer
2444  *	@skb: buffer to orphan frags from
2445  *	@gfp_mask: allocation mask for replacement pages
2446  *
2447  *	For each frag in the SKB which needs a destructor (i.e. has an
2448  *	owner) create a copy of that frag and release the original
2449  *	page by calling the destructor.
2450  */
2451 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2452 {
2453 	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2454 		return 0;
2455 	return skb_copy_ubufs(skb, gfp_mask);
2456 }
2457 
2458 /**
2459  *	__skb_queue_purge - empty a list
2460  *	@list: list to empty
2461  *
2462  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
2463  *	the list and one reference dropped. This function does not take the
2464  *	list lock and the caller must hold the relevant locks to use it.
2465  */
2466 void skb_queue_purge(struct sk_buff_head *list);
2467 static inline void __skb_queue_purge(struct sk_buff_head *list)
2468 {
2469 	struct sk_buff *skb;
2470 	while ((skb = __skb_dequeue(list)) != NULL)
2471 		kfree_skb(skb);
2472 }
2473 
2474 void skb_rbtree_purge(struct rb_root *root);
2475 
2476 void *netdev_alloc_frag(unsigned int fragsz);
2477 
2478 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2479 				   gfp_t gfp_mask);
2480 
2481 /**
2482  *	netdev_alloc_skb - allocate an skbuff for rx on a specific device
2483  *	@dev: network device to receive on
2484  *	@length: length to allocate
2485  *
2486  *	Allocate a new &sk_buff and assign it a usage count of one. The
2487  *	buffer has unspecified headroom built in. Users should allocate
2488  *	the headroom they think they need without accounting for the
2489  *	built in space. The built in space is used for optimisations.
2490  *
2491  *	%NULL is returned if there is no free memory. Although this function
2492  *	allocates memory it can be called from an interrupt.
2493  */
2494 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2495 					       unsigned int length)
2496 {
2497 	return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2498 }
2499 
2500 /* legacy helper around __netdev_alloc_skb() */
2501 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2502 					      gfp_t gfp_mask)
2503 {
2504 	return __netdev_alloc_skb(NULL, length, gfp_mask);
2505 }
2506 
2507 /* legacy helper around netdev_alloc_skb() */
2508 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2509 {
2510 	return netdev_alloc_skb(NULL, length);
2511 }
2512 
2513 
2514 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2515 		unsigned int length, gfp_t gfp)
2516 {
2517 	struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2518 
2519 	if (NET_IP_ALIGN && skb)
2520 		skb_reserve(skb, NET_IP_ALIGN);
2521 	return skb;
2522 }
2523 
2524 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2525 		unsigned int length)
2526 {
2527 	return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2528 }
2529 
2530 static inline void skb_free_frag(void *addr)
2531 {
2532 	page_frag_free(addr);
2533 }
2534 
2535 void *napi_alloc_frag(unsigned int fragsz);
2536 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2537 				 unsigned int length, gfp_t gfp_mask);
2538 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2539 					     unsigned int length)
2540 {
2541 	return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2542 }
2543 void napi_consume_skb(struct sk_buff *skb, int budget);
2544 
2545 void __kfree_skb_flush(void);
2546 void __kfree_skb_defer(struct sk_buff *skb);
2547 
2548 /**
2549  * __dev_alloc_pages - allocate page for network Rx
2550  * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2551  * @order: size of the allocation
2552  *
2553  * Allocate a new page.
2554  *
2555  * %NULL is returned if there is no free memory.
2556 */
2557 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2558 					     unsigned int order)
2559 {
2560 	/* This piece of code contains several assumptions.
2561 	 * 1.  This is for device Rx, therefor a cold page is preferred.
2562 	 * 2.  The expectation is the user wants a compound page.
2563 	 * 3.  If requesting a order 0 page it will not be compound
2564 	 *     due to the check to see if order has a value in prep_new_page
2565 	 * 4.  __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2566 	 *     code in gfp_to_alloc_flags that should be enforcing this.
2567 	 */
2568 	gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2569 
2570 	return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2571 }
2572 
2573 static inline struct page *dev_alloc_pages(unsigned int order)
2574 {
2575 	return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2576 }
2577 
2578 /**
2579  * __dev_alloc_page - allocate a page for network Rx
2580  * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2581  *
2582  * Allocate a new page.
2583  *
2584  * %NULL is returned if there is no free memory.
2585  */
2586 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2587 {
2588 	return __dev_alloc_pages(gfp_mask, 0);
2589 }
2590 
2591 static inline struct page *dev_alloc_page(void)
2592 {
2593 	return dev_alloc_pages(0);
2594 }
2595 
2596 /**
2597  *	skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2598  *	@page: The page that was allocated from skb_alloc_page
2599  *	@skb: The skb that may need pfmemalloc set
2600  */
2601 static inline void skb_propagate_pfmemalloc(struct page *page,
2602 					     struct sk_buff *skb)
2603 {
2604 	if (page_is_pfmemalloc(page))
2605 		skb->pfmemalloc = true;
2606 }
2607 
2608 /**
2609  * skb_frag_page - retrieve the page referred to by a paged fragment
2610  * @frag: the paged fragment
2611  *
2612  * Returns the &struct page associated with @frag.
2613  */
2614 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2615 {
2616 	return frag->page.p;
2617 }
2618 
2619 /**
2620  * __skb_frag_ref - take an addition reference on a paged fragment.
2621  * @frag: the paged fragment
2622  *
2623  * Takes an additional reference on the paged fragment @frag.
2624  */
2625 static inline void __skb_frag_ref(skb_frag_t *frag)
2626 {
2627 	get_page(skb_frag_page(frag));
2628 }
2629 
2630 /**
2631  * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2632  * @skb: the buffer
2633  * @f: the fragment offset.
2634  *
2635  * Takes an additional reference on the @f'th paged fragment of @skb.
2636  */
2637 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2638 {
2639 	__skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2640 }
2641 
2642 /**
2643  * __skb_frag_unref - release a reference on a paged fragment.
2644  * @frag: the paged fragment
2645  *
2646  * Releases a reference on the paged fragment @frag.
2647  */
2648 static inline void __skb_frag_unref(skb_frag_t *frag)
2649 {
2650 	put_page(skb_frag_page(frag));
2651 }
2652 
2653 /**
2654  * skb_frag_unref - release a reference on a paged fragment of an skb.
2655  * @skb: the buffer
2656  * @f: the fragment offset
2657  *
2658  * Releases a reference on the @f'th paged fragment of @skb.
2659  */
2660 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2661 {
2662 	__skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2663 }
2664 
2665 /**
2666  * skb_frag_address - gets the address of the data contained in a paged fragment
2667  * @frag: the paged fragment buffer
2668  *
2669  * Returns the address of the data within @frag. The page must already
2670  * be mapped.
2671  */
2672 static inline void *skb_frag_address(const skb_frag_t *frag)
2673 {
2674 	return page_address(skb_frag_page(frag)) + frag->page_offset;
2675 }
2676 
2677 /**
2678  * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2679  * @frag: the paged fragment buffer
2680  *
2681  * Returns the address of the data within @frag. Checks that the page
2682  * is mapped and returns %NULL otherwise.
2683  */
2684 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2685 {
2686 	void *ptr = page_address(skb_frag_page(frag));
2687 	if (unlikely(!ptr))
2688 		return NULL;
2689 
2690 	return ptr + frag->page_offset;
2691 }
2692 
2693 /**
2694  * __skb_frag_set_page - sets the page contained in a paged fragment
2695  * @frag: the paged fragment
2696  * @page: the page to set
2697  *
2698  * Sets the fragment @frag to contain @page.
2699  */
2700 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2701 {
2702 	frag->page.p = page;
2703 }
2704 
2705 /**
2706  * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2707  * @skb: the buffer
2708  * @f: the fragment offset
2709  * @page: the page to set
2710  *
2711  * Sets the @f'th fragment of @skb to contain @page.
2712  */
2713 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2714 				     struct page *page)
2715 {
2716 	__skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2717 }
2718 
2719 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2720 
2721 /**
2722  * skb_frag_dma_map - maps a paged fragment via the DMA API
2723  * @dev: the device to map the fragment to
2724  * @frag: the paged fragment to map
2725  * @offset: the offset within the fragment (starting at the
2726  *          fragment's own offset)
2727  * @size: the number of bytes to map
2728  * @dir: the direction of the mapping (``PCI_DMA_*``)
2729  *
2730  * Maps the page associated with @frag to @device.
2731  */
2732 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2733 					  const skb_frag_t *frag,
2734 					  size_t offset, size_t size,
2735 					  enum dma_data_direction dir)
2736 {
2737 	return dma_map_page(dev, skb_frag_page(frag),
2738 			    frag->page_offset + offset, size, dir);
2739 }
2740 
2741 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2742 					gfp_t gfp_mask)
2743 {
2744 	return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2745 }
2746 
2747 
2748 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2749 						  gfp_t gfp_mask)
2750 {
2751 	return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2752 }
2753 
2754 
2755 /**
2756  *	skb_clone_writable - is the header of a clone writable
2757  *	@skb: buffer to check
2758  *	@len: length up to which to write
2759  *
2760  *	Returns true if modifying the header part of the cloned buffer
2761  *	does not requires the data to be copied.
2762  */
2763 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2764 {
2765 	return !skb_header_cloned(skb) &&
2766 	       skb_headroom(skb) + len <= skb->hdr_len;
2767 }
2768 
2769 static inline int skb_try_make_writable(struct sk_buff *skb,
2770 					unsigned int write_len)
2771 {
2772 	return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2773 	       pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2774 }
2775 
2776 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2777 			    int cloned)
2778 {
2779 	int delta = 0;
2780 
2781 	if (headroom > skb_headroom(skb))
2782 		delta = headroom - skb_headroom(skb);
2783 
2784 	if (delta || cloned)
2785 		return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2786 					GFP_ATOMIC);
2787 	return 0;
2788 }
2789 
2790 /**
2791  *	skb_cow - copy header of skb when it is required
2792  *	@skb: buffer to cow
2793  *	@headroom: needed headroom
2794  *
2795  *	If the skb passed lacks sufficient headroom or its data part
2796  *	is shared, data is reallocated. If reallocation fails, an error
2797  *	is returned and original skb is not changed.
2798  *
2799  *	The result is skb with writable area skb->head...skb->tail
2800  *	and at least @headroom of space at head.
2801  */
2802 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2803 {
2804 	return __skb_cow(skb, headroom, skb_cloned(skb));
2805 }
2806 
2807 /**
2808  *	skb_cow_head - skb_cow but only making the head writable
2809  *	@skb: buffer to cow
2810  *	@headroom: needed headroom
2811  *
2812  *	This function is identical to skb_cow except that we replace the
2813  *	skb_cloned check by skb_header_cloned.  It should be used when
2814  *	you only need to push on some header and do not need to modify
2815  *	the data.
2816  */
2817 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2818 {
2819 	return __skb_cow(skb, headroom, skb_header_cloned(skb));
2820 }
2821 
2822 /**
2823  *	skb_padto	- pad an skbuff up to a minimal size
2824  *	@skb: buffer to pad
2825  *	@len: minimal length
2826  *
2827  *	Pads up a buffer to ensure the trailing bytes exist and are
2828  *	blanked. If the buffer already contains sufficient data it
2829  *	is untouched. Otherwise it is extended. Returns zero on
2830  *	success. The skb is freed on error.
2831  */
2832 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2833 {
2834 	unsigned int size = skb->len;
2835 	if (likely(size >= len))
2836 		return 0;
2837 	return skb_pad(skb, len - size);
2838 }
2839 
2840 /**
2841  *	skb_put_padto - increase size and pad an skbuff up to a minimal size
2842  *	@skb: buffer to pad
2843  *	@len: minimal length
2844  *	@free_on_error: free buffer on error
2845  *
2846  *	Pads up a buffer to ensure the trailing bytes exist and are
2847  *	blanked. If the buffer already contains sufficient data it
2848  *	is untouched. Otherwise it is extended. Returns zero on
2849  *	success. The skb is freed on error if @free_on_error is true.
2850  */
2851 static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
2852 				  bool free_on_error)
2853 {
2854 	unsigned int size = skb->len;
2855 
2856 	if (unlikely(size < len)) {
2857 		len -= size;
2858 		if (__skb_pad(skb, len, free_on_error))
2859 			return -ENOMEM;
2860 		__skb_put(skb, len);
2861 	}
2862 	return 0;
2863 }
2864 
2865 /**
2866  *	skb_put_padto - increase size and pad an skbuff up to a minimal size
2867  *	@skb: buffer to pad
2868  *	@len: minimal length
2869  *
2870  *	Pads up a buffer to ensure the trailing bytes exist and are
2871  *	blanked. If the buffer already contains sufficient data it
2872  *	is untouched. Otherwise it is extended. Returns zero on
2873  *	success. The skb is freed on error.
2874  */
2875 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2876 {
2877 	return __skb_put_padto(skb, len, true);
2878 }
2879 
2880 static inline int skb_add_data(struct sk_buff *skb,
2881 			       struct iov_iter *from, int copy)
2882 {
2883 	const int off = skb->len;
2884 
2885 	if (skb->ip_summed == CHECKSUM_NONE) {
2886 		__wsum csum = 0;
2887 		if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2888 					         &csum, from)) {
2889 			skb->csum = csum_block_add(skb->csum, csum, off);
2890 			return 0;
2891 		}
2892 	} else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
2893 		return 0;
2894 
2895 	__skb_trim(skb, off);
2896 	return -EFAULT;
2897 }
2898 
2899 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2900 				    const struct page *page, int off)
2901 {
2902 	if (i) {
2903 		const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2904 
2905 		return page == skb_frag_page(frag) &&
2906 		       off == frag->page_offset + skb_frag_size(frag);
2907 	}
2908 	return false;
2909 }
2910 
2911 static inline int __skb_linearize(struct sk_buff *skb)
2912 {
2913 	return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2914 }
2915 
2916 /**
2917  *	skb_linearize - convert paged skb to linear one
2918  *	@skb: buffer to linarize
2919  *
2920  *	If there is no free memory -ENOMEM is returned, otherwise zero
2921  *	is returned and the old skb data released.
2922  */
2923 static inline int skb_linearize(struct sk_buff *skb)
2924 {
2925 	return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2926 }
2927 
2928 /**
2929  * skb_has_shared_frag - can any frag be overwritten
2930  * @skb: buffer to test
2931  *
2932  * Return true if the skb has at least one frag that might be modified
2933  * by an external entity (as in vmsplice()/sendfile())
2934  */
2935 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2936 {
2937 	return skb_is_nonlinear(skb) &&
2938 	       skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2939 }
2940 
2941 /**
2942  *	skb_linearize_cow - make sure skb is linear and writable
2943  *	@skb: buffer to process
2944  *
2945  *	If there is no free memory -ENOMEM is returned, otherwise zero
2946  *	is returned and the old skb data released.
2947  */
2948 static inline int skb_linearize_cow(struct sk_buff *skb)
2949 {
2950 	return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2951 	       __skb_linearize(skb) : 0;
2952 }
2953 
2954 static __always_inline void
2955 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2956 		     unsigned int off)
2957 {
2958 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2959 		skb->csum = csum_block_sub(skb->csum,
2960 					   csum_partial(start, len, 0), off);
2961 	else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2962 		 skb_checksum_start_offset(skb) < 0)
2963 		skb->ip_summed = CHECKSUM_NONE;
2964 }
2965 
2966 /**
2967  *	skb_postpull_rcsum - update checksum for received skb after pull
2968  *	@skb: buffer to update
2969  *	@start: start of data before pull
2970  *	@len: length of data pulled
2971  *
2972  *	After doing a pull on a received packet, you need to call this to
2973  *	update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2974  *	CHECKSUM_NONE so that it can be recomputed from scratch.
2975  */
2976 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2977 				      const void *start, unsigned int len)
2978 {
2979 	__skb_postpull_rcsum(skb, start, len, 0);
2980 }
2981 
2982 static __always_inline void
2983 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2984 		     unsigned int off)
2985 {
2986 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2987 		skb->csum = csum_block_add(skb->csum,
2988 					   csum_partial(start, len, 0), off);
2989 }
2990 
2991 /**
2992  *	skb_postpush_rcsum - update checksum for received skb after push
2993  *	@skb: buffer to update
2994  *	@start: start of data after push
2995  *	@len: length of data pushed
2996  *
2997  *	After doing a push on a received packet, you need to call this to
2998  *	update the CHECKSUM_COMPLETE checksum.
2999  */
3000 static inline void skb_postpush_rcsum(struct sk_buff *skb,
3001 				      const void *start, unsigned int len)
3002 {
3003 	__skb_postpush_rcsum(skb, start, len, 0);
3004 }
3005 
3006 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3007 
3008 /**
3009  *	skb_push_rcsum - push skb and update receive checksum
3010  *	@skb: buffer to update
3011  *	@len: length of data pulled
3012  *
3013  *	This function performs an skb_push on the packet and updates
3014  *	the CHECKSUM_COMPLETE checksum.  It should be used on
3015  *	receive path processing instead of skb_push unless you know
3016  *	that the checksum difference is zero (e.g., a valid IP header)
3017  *	or you are setting ip_summed to CHECKSUM_NONE.
3018  */
3019 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3020 {
3021 	skb_push(skb, len);
3022 	skb_postpush_rcsum(skb, skb->data, len);
3023 	return skb->data;
3024 }
3025 
3026 /**
3027  *	pskb_trim_rcsum - trim received skb and update checksum
3028  *	@skb: buffer to trim
3029  *	@len: new length
3030  *
3031  *	This is exactly the same as pskb_trim except that it ensures the
3032  *	checksum of received packets are still valid after the operation.
3033  */
3034 
3035 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3036 {
3037 	if (likely(len >= skb->len))
3038 		return 0;
3039 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3040 		skb->ip_summed = CHECKSUM_NONE;
3041 	return __pskb_trim(skb, len);
3042 }
3043 
3044 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3045 {
3046 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3047 		skb->ip_summed = CHECKSUM_NONE;
3048 	__skb_trim(skb, len);
3049 	return 0;
3050 }
3051 
3052 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3053 {
3054 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3055 		skb->ip_summed = CHECKSUM_NONE;
3056 	return __skb_grow(skb, len);
3057 }
3058 
3059 #define skb_queue_walk(queue, skb) \
3060 		for (skb = (queue)->next;					\
3061 		     skb != (struct sk_buff *)(queue);				\
3062 		     skb = skb->next)
3063 
3064 #define skb_queue_walk_safe(queue, skb, tmp)					\
3065 		for (skb = (queue)->next, tmp = skb->next;			\
3066 		     skb != (struct sk_buff *)(queue);				\
3067 		     skb = tmp, tmp = skb->next)
3068 
3069 #define skb_queue_walk_from(queue, skb)						\
3070 		for (; skb != (struct sk_buff *)(queue);			\
3071 		     skb = skb->next)
3072 
3073 #define skb_queue_walk_from_safe(queue, skb, tmp)				\
3074 		for (tmp = skb->next;						\
3075 		     skb != (struct sk_buff *)(queue);				\
3076 		     skb = tmp, tmp = skb->next)
3077 
3078 #define skb_queue_reverse_walk(queue, skb) \
3079 		for (skb = (queue)->prev;					\
3080 		     skb != (struct sk_buff *)(queue);				\
3081 		     skb = skb->prev)
3082 
3083 #define skb_queue_reverse_walk_safe(queue, skb, tmp)				\
3084 		for (skb = (queue)->prev, tmp = skb->prev;			\
3085 		     skb != (struct sk_buff *)(queue);				\
3086 		     skb = tmp, tmp = skb->prev)
3087 
3088 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp)			\
3089 		for (tmp = skb->prev;						\
3090 		     skb != (struct sk_buff *)(queue);				\
3091 		     skb = tmp, tmp = skb->prev)
3092 
3093 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3094 {
3095 	return skb_shinfo(skb)->frag_list != NULL;
3096 }
3097 
3098 static inline void skb_frag_list_init(struct sk_buff *skb)
3099 {
3100 	skb_shinfo(skb)->frag_list = NULL;
3101 }
3102 
3103 #define skb_walk_frags(skb, iter)	\
3104 	for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3105 
3106 
3107 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3108 				const struct sk_buff *skb);
3109 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3110 					  struct sk_buff_head *queue,
3111 					  unsigned int flags,
3112 					  void (*destructor)(struct sock *sk,
3113 							   struct sk_buff *skb),
3114 					  int *peeked, int *off, int *err,
3115 					  struct sk_buff **last);
3116 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3117 					void (*destructor)(struct sock *sk,
3118 							   struct sk_buff *skb),
3119 					int *peeked, int *off, int *err,
3120 					struct sk_buff **last);
3121 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3122 				    void (*destructor)(struct sock *sk,
3123 						       struct sk_buff *skb),
3124 				    int *peeked, int *off, int *err);
3125 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3126 				  int *err);
3127 unsigned int datagram_poll(struct file *file, struct socket *sock,
3128 			   struct poll_table_struct *wait);
3129 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3130 			   struct iov_iter *to, int size);
3131 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3132 					struct msghdr *msg, int size)
3133 {
3134 	return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3135 }
3136 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3137 				   struct msghdr *msg);
3138 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3139 				 struct iov_iter *from, int len);
3140 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3141 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3142 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3143 static inline void skb_free_datagram_locked(struct sock *sk,
3144 					    struct sk_buff *skb)
3145 {
3146 	__skb_free_datagram_locked(sk, skb, 0);
3147 }
3148 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3149 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3150 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3151 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3152 			      int len, __wsum csum);
3153 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3154 		    struct pipe_inode_info *pipe, unsigned int len,
3155 		    unsigned int flags);
3156 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3157 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3158 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3159 		 int len, int hlen);
3160 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3161 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3162 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3163 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3164 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3165 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3166 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3167 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3168 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3169 int skb_vlan_pop(struct sk_buff *skb);
3170 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3171 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3172 			     gfp_t gfp);
3173 
3174 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3175 {
3176 	return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3177 }
3178 
3179 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3180 {
3181 	return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3182 }
3183 
3184 struct skb_checksum_ops {
3185 	__wsum (*update)(const void *mem, int len, __wsum wsum);
3186 	__wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3187 };
3188 
3189 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3190 
3191 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3192 		      __wsum csum, const struct skb_checksum_ops *ops);
3193 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3194 		    __wsum csum);
3195 
3196 static inline void * __must_check
3197 __skb_header_pointer(const struct sk_buff *skb, int offset,
3198 		     int len, void *data, int hlen, void *buffer)
3199 {
3200 	if (hlen - offset >= len)
3201 		return data + offset;
3202 
3203 	if (!skb ||
3204 	    skb_copy_bits(skb, offset, buffer, len) < 0)
3205 		return NULL;
3206 
3207 	return buffer;
3208 }
3209 
3210 static inline void * __must_check
3211 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3212 {
3213 	return __skb_header_pointer(skb, offset, len, skb->data,
3214 				    skb_headlen(skb), buffer);
3215 }
3216 
3217 /**
3218  *	skb_needs_linearize - check if we need to linearize a given skb
3219  *			      depending on the given device features.
3220  *	@skb: socket buffer to check
3221  *	@features: net device features
3222  *
3223  *	Returns true if either:
3224  *	1. skb has frag_list and the device doesn't support FRAGLIST, or
3225  *	2. skb is fragmented and the device does not support SG.
3226  */
3227 static inline bool skb_needs_linearize(struct sk_buff *skb,
3228 				       netdev_features_t features)
3229 {
3230 	return skb_is_nonlinear(skb) &&
3231 	       ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3232 		(skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3233 }
3234 
3235 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3236 					     void *to,
3237 					     const unsigned int len)
3238 {
3239 	memcpy(to, skb->data, len);
3240 }
3241 
3242 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3243 						    const int offset, void *to,
3244 						    const unsigned int len)
3245 {
3246 	memcpy(to, skb->data + offset, len);
3247 }
3248 
3249 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3250 					   const void *from,
3251 					   const unsigned int len)
3252 {
3253 	memcpy(skb->data, from, len);
3254 }
3255 
3256 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3257 						  const int offset,
3258 						  const void *from,
3259 						  const unsigned int len)
3260 {
3261 	memcpy(skb->data + offset, from, len);
3262 }
3263 
3264 void skb_init(void);
3265 
3266 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3267 {
3268 	return skb->tstamp;
3269 }
3270 
3271 /**
3272  *	skb_get_timestamp - get timestamp from a skb
3273  *	@skb: skb to get stamp from
3274  *	@stamp: pointer to struct timeval to store stamp in
3275  *
3276  *	Timestamps are stored in the skb as offsets to a base timestamp.
3277  *	This function converts the offset back to a struct timeval and stores
3278  *	it in stamp.
3279  */
3280 static inline void skb_get_timestamp(const struct sk_buff *skb,
3281 				     struct timeval *stamp)
3282 {
3283 	*stamp = ktime_to_timeval(skb->tstamp);
3284 }
3285 
3286 static inline void skb_get_timestampns(const struct sk_buff *skb,
3287 				       struct timespec *stamp)
3288 {
3289 	*stamp = ktime_to_timespec(skb->tstamp);
3290 }
3291 
3292 static inline void __net_timestamp(struct sk_buff *skb)
3293 {
3294 	skb->tstamp = ktime_get_real();
3295 }
3296 
3297 static inline ktime_t net_timedelta(ktime_t t)
3298 {
3299 	return ktime_sub(ktime_get_real(), t);
3300 }
3301 
3302 static inline ktime_t net_invalid_timestamp(void)
3303 {
3304 	return 0;
3305 }
3306 
3307 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3308 
3309 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3310 
3311 void skb_clone_tx_timestamp(struct sk_buff *skb);
3312 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3313 
3314 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3315 
3316 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3317 {
3318 }
3319 
3320 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3321 {
3322 	return false;
3323 }
3324 
3325 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3326 
3327 /**
3328  * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3329  *
3330  * PHY drivers may accept clones of transmitted packets for
3331  * timestamping via their phy_driver.txtstamp method. These drivers
3332  * must call this function to return the skb back to the stack with a
3333  * timestamp.
3334  *
3335  * @skb: clone of the the original outgoing packet
3336  * @hwtstamps: hardware time stamps
3337  *
3338  */
3339 void skb_complete_tx_timestamp(struct sk_buff *skb,
3340 			       struct skb_shared_hwtstamps *hwtstamps);
3341 
3342 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3343 		     struct skb_shared_hwtstamps *hwtstamps,
3344 		     struct sock *sk, int tstype);
3345 
3346 /**
3347  * skb_tstamp_tx - queue clone of skb with send time stamps
3348  * @orig_skb:	the original outgoing packet
3349  * @hwtstamps:	hardware time stamps, may be NULL if not available
3350  *
3351  * If the skb has a socket associated, then this function clones the
3352  * skb (thus sharing the actual data and optional structures), stores
3353  * the optional hardware time stamping information (if non NULL) or
3354  * generates a software time stamp (otherwise), then queues the clone
3355  * to the error queue of the socket.  Errors are silently ignored.
3356  */
3357 void skb_tstamp_tx(struct sk_buff *orig_skb,
3358 		   struct skb_shared_hwtstamps *hwtstamps);
3359 
3360 /**
3361  * skb_tx_timestamp() - Driver hook for transmit timestamping
3362  *
3363  * Ethernet MAC Drivers should call this function in their hard_xmit()
3364  * function immediately before giving the sk_buff to the MAC hardware.
3365  *
3366  * Specifically, one should make absolutely sure that this function is
3367  * called before TX completion of this packet can trigger.  Otherwise
3368  * the packet could potentially already be freed.
3369  *
3370  * @skb: A socket buffer.
3371  */
3372 static inline void skb_tx_timestamp(struct sk_buff *skb)
3373 {
3374 	skb_clone_tx_timestamp(skb);
3375 	if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3376 		skb_tstamp_tx(skb, NULL);
3377 }
3378 
3379 /**
3380  * skb_complete_wifi_ack - deliver skb with wifi status
3381  *
3382  * @skb: the original outgoing packet
3383  * @acked: ack status
3384  *
3385  */
3386 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3387 
3388 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3389 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3390 
3391 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3392 {
3393 	return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3394 		skb->csum_valid ||
3395 		(skb->ip_summed == CHECKSUM_PARTIAL &&
3396 		 skb_checksum_start_offset(skb) >= 0));
3397 }
3398 
3399 /**
3400  *	skb_checksum_complete - Calculate checksum of an entire packet
3401  *	@skb: packet to process
3402  *
3403  *	This function calculates the checksum over the entire packet plus
3404  *	the value of skb->csum.  The latter can be used to supply the
3405  *	checksum of a pseudo header as used by TCP/UDP.  It returns the
3406  *	checksum.
3407  *
3408  *	For protocols that contain complete checksums such as ICMP/TCP/UDP,
3409  *	this function can be used to verify that checksum on received
3410  *	packets.  In that case the function should return zero if the
3411  *	checksum is correct.  In particular, this function will return zero
3412  *	if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3413  *	hardware has already verified the correctness of the checksum.
3414  */
3415 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3416 {
3417 	return skb_csum_unnecessary(skb) ?
3418 	       0 : __skb_checksum_complete(skb);
3419 }
3420 
3421 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3422 {
3423 	if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3424 		if (skb->csum_level == 0)
3425 			skb->ip_summed = CHECKSUM_NONE;
3426 		else
3427 			skb->csum_level--;
3428 	}
3429 }
3430 
3431 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3432 {
3433 	if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3434 		if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3435 			skb->csum_level++;
3436 	} else if (skb->ip_summed == CHECKSUM_NONE) {
3437 		skb->ip_summed = CHECKSUM_UNNECESSARY;
3438 		skb->csum_level = 0;
3439 	}
3440 }
3441 
3442 /* Check if we need to perform checksum complete validation.
3443  *
3444  * Returns true if checksum complete is needed, false otherwise
3445  * (either checksum is unnecessary or zero checksum is allowed).
3446  */
3447 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3448 						  bool zero_okay,
3449 						  __sum16 check)
3450 {
3451 	if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3452 		skb->csum_valid = 1;
3453 		__skb_decr_checksum_unnecessary(skb);
3454 		return false;
3455 	}
3456 
3457 	return true;
3458 }
3459 
3460 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3461  * in checksum_init.
3462  */
3463 #define CHECKSUM_BREAK 76
3464 
3465 /* Unset checksum-complete
3466  *
3467  * Unset checksum complete can be done when packet is being modified
3468  * (uncompressed for instance) and checksum-complete value is
3469  * invalidated.
3470  */
3471 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3472 {
3473 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3474 		skb->ip_summed = CHECKSUM_NONE;
3475 }
3476 
3477 /* Validate (init) checksum based on checksum complete.
3478  *
3479  * Return values:
3480  *   0: checksum is validated or try to in skb_checksum_complete. In the latter
3481  *	case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3482  *	checksum is stored in skb->csum for use in __skb_checksum_complete
3483  *   non-zero: value of invalid checksum
3484  *
3485  */
3486 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3487 						       bool complete,
3488 						       __wsum psum)
3489 {
3490 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
3491 		if (!csum_fold(csum_add(psum, skb->csum))) {
3492 			skb->csum_valid = 1;
3493 			return 0;
3494 		}
3495 	}
3496 
3497 	skb->csum = psum;
3498 
3499 	if (complete || skb->len <= CHECKSUM_BREAK) {
3500 		__sum16 csum;
3501 
3502 		csum = __skb_checksum_complete(skb);
3503 		skb->csum_valid = !csum;
3504 		return csum;
3505 	}
3506 
3507 	return 0;
3508 }
3509 
3510 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3511 {
3512 	return 0;
3513 }
3514 
3515 /* Perform checksum validate (init). Note that this is a macro since we only
3516  * want to calculate the pseudo header which is an input function if necessary.
3517  * First we try to validate without any computation (checksum unnecessary) and
3518  * then calculate based on checksum complete calling the function to compute
3519  * pseudo header.
3520  *
3521  * Return values:
3522  *   0: checksum is validated or try to in skb_checksum_complete
3523  *   non-zero: value of invalid checksum
3524  */
3525 #define __skb_checksum_validate(skb, proto, complete,			\
3526 				zero_okay, check, compute_pseudo)	\
3527 ({									\
3528 	__sum16 __ret = 0;						\
3529 	skb->csum_valid = 0;						\
3530 	if (__skb_checksum_validate_needed(skb, zero_okay, check))	\
3531 		__ret = __skb_checksum_validate_complete(skb,		\
3532 				complete, compute_pseudo(skb, proto));	\
3533 	__ret;								\
3534 })
3535 
3536 #define skb_checksum_init(skb, proto, compute_pseudo)			\
3537 	__skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3538 
3539 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo)	\
3540 	__skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3541 
3542 #define skb_checksum_validate(skb, proto, compute_pseudo)		\
3543 	__skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3544 
3545 #define skb_checksum_validate_zero_check(skb, proto, check,		\
3546 					 compute_pseudo)		\
3547 	__skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3548 
3549 #define skb_checksum_simple_validate(skb)				\
3550 	__skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3551 
3552 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3553 {
3554 	return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3555 }
3556 
3557 static inline void __skb_checksum_convert(struct sk_buff *skb,
3558 					  __sum16 check, __wsum pseudo)
3559 {
3560 	skb->csum = ~pseudo;
3561 	skb->ip_summed = CHECKSUM_COMPLETE;
3562 }
3563 
3564 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo)	\
3565 do {									\
3566 	if (__skb_checksum_convert_check(skb))				\
3567 		__skb_checksum_convert(skb, check,			\
3568 				       compute_pseudo(skb, proto));	\
3569 } while (0)
3570 
3571 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3572 					      u16 start, u16 offset)
3573 {
3574 	skb->ip_summed = CHECKSUM_PARTIAL;
3575 	skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3576 	skb->csum_offset = offset - start;
3577 }
3578 
3579 /* Update skbuf and packet to reflect the remote checksum offload operation.
3580  * When called, ptr indicates the starting point for skb->csum when
3581  * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3582  * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3583  */
3584 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3585 				       int start, int offset, bool nopartial)
3586 {
3587 	__wsum delta;
3588 
3589 	if (!nopartial) {
3590 		skb_remcsum_adjust_partial(skb, ptr, start, offset);
3591 		return;
3592 	}
3593 
3594 	 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3595 		__skb_checksum_complete(skb);
3596 		skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3597 	}
3598 
3599 	delta = remcsum_adjust(ptr, skb->csum, start, offset);
3600 
3601 	/* Adjust skb->csum since we changed the packet */
3602 	skb->csum = csum_add(skb->csum, delta);
3603 }
3604 
3605 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3606 {
3607 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3608 	return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3609 #else
3610 	return NULL;
3611 #endif
3612 }
3613 
3614 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3615 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3616 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3617 {
3618 	if (nfct && atomic_dec_and_test(&nfct->use))
3619 		nf_conntrack_destroy(nfct);
3620 }
3621 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3622 {
3623 	if (nfct)
3624 		atomic_inc(&nfct->use);
3625 }
3626 #endif
3627 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3628 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3629 {
3630 	if (nf_bridge && refcount_dec_and_test(&nf_bridge->use))
3631 		kfree(nf_bridge);
3632 }
3633 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3634 {
3635 	if (nf_bridge)
3636 		refcount_inc(&nf_bridge->use);
3637 }
3638 #endif /* CONFIG_BRIDGE_NETFILTER */
3639 static inline void nf_reset(struct sk_buff *skb)
3640 {
3641 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3642 	nf_conntrack_put(skb_nfct(skb));
3643 	skb->_nfct = 0;
3644 #endif
3645 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3646 	nf_bridge_put(skb->nf_bridge);
3647 	skb->nf_bridge = NULL;
3648 #endif
3649 }
3650 
3651 static inline void nf_reset_trace(struct sk_buff *skb)
3652 {
3653 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3654 	skb->nf_trace = 0;
3655 #endif
3656 }
3657 
3658 /* Note: This doesn't put any conntrack and bridge info in dst. */
3659 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3660 			     bool copy)
3661 {
3662 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3663 	dst->_nfct = src->_nfct;
3664 	nf_conntrack_get(skb_nfct(src));
3665 #endif
3666 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3667 	dst->nf_bridge  = src->nf_bridge;
3668 	nf_bridge_get(src->nf_bridge);
3669 #endif
3670 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3671 	if (copy)
3672 		dst->nf_trace = src->nf_trace;
3673 #endif
3674 }
3675 
3676 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3677 {
3678 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3679 	nf_conntrack_put(skb_nfct(dst));
3680 #endif
3681 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3682 	nf_bridge_put(dst->nf_bridge);
3683 #endif
3684 	__nf_copy(dst, src, true);
3685 }
3686 
3687 #ifdef CONFIG_NETWORK_SECMARK
3688 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3689 {
3690 	to->secmark = from->secmark;
3691 }
3692 
3693 static inline void skb_init_secmark(struct sk_buff *skb)
3694 {
3695 	skb->secmark = 0;
3696 }
3697 #else
3698 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3699 { }
3700 
3701 static inline void skb_init_secmark(struct sk_buff *skb)
3702 { }
3703 #endif
3704 
3705 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3706 {
3707 	return !skb->destructor &&
3708 #if IS_ENABLED(CONFIG_XFRM)
3709 		!skb->sp &&
3710 #endif
3711 		!skb_nfct(skb) &&
3712 		!skb->_skb_refdst &&
3713 		!skb_has_frag_list(skb);
3714 }
3715 
3716 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3717 {
3718 	skb->queue_mapping = queue_mapping;
3719 }
3720 
3721 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3722 {
3723 	return skb->queue_mapping;
3724 }
3725 
3726 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3727 {
3728 	to->queue_mapping = from->queue_mapping;
3729 }
3730 
3731 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3732 {
3733 	skb->queue_mapping = rx_queue + 1;
3734 }
3735 
3736 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3737 {
3738 	return skb->queue_mapping - 1;
3739 }
3740 
3741 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3742 {
3743 	return skb->queue_mapping != 0;
3744 }
3745 
3746 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3747 {
3748 	skb->dst_pending_confirm = val;
3749 }
3750 
3751 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3752 {
3753 	return skb->dst_pending_confirm != 0;
3754 }
3755 
3756 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3757 {
3758 #ifdef CONFIG_XFRM
3759 	return skb->sp;
3760 #else
3761 	return NULL;
3762 #endif
3763 }
3764 
3765 /* Keeps track of mac header offset relative to skb->head.
3766  * It is useful for TSO of Tunneling protocol. e.g. GRE.
3767  * For non-tunnel skb it points to skb_mac_header() and for
3768  * tunnel skb it points to outer mac header.
3769  * Keeps track of level of encapsulation of network headers.
3770  */
3771 struct skb_gso_cb {
3772 	union {
3773 		int	mac_offset;
3774 		int	data_offset;
3775 	};
3776 	int	encap_level;
3777 	__wsum	csum;
3778 	__u16	csum_start;
3779 };
3780 #define SKB_SGO_CB_OFFSET	32
3781 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3782 
3783 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3784 {
3785 	return (skb_mac_header(inner_skb) - inner_skb->head) -
3786 		SKB_GSO_CB(inner_skb)->mac_offset;
3787 }
3788 
3789 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3790 {
3791 	int new_headroom, headroom;
3792 	int ret;
3793 
3794 	headroom = skb_headroom(skb);
3795 	ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3796 	if (ret)
3797 		return ret;
3798 
3799 	new_headroom = skb_headroom(skb);
3800 	SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3801 	return 0;
3802 }
3803 
3804 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3805 {
3806 	/* Do not update partial checksums if remote checksum is enabled. */
3807 	if (skb->remcsum_offload)
3808 		return;
3809 
3810 	SKB_GSO_CB(skb)->csum = res;
3811 	SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3812 }
3813 
3814 /* Compute the checksum for a gso segment. First compute the checksum value
3815  * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3816  * then add in skb->csum (checksum from csum_start to end of packet).
3817  * skb->csum and csum_start are then updated to reflect the checksum of the
3818  * resultant packet starting from the transport header-- the resultant checksum
3819  * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3820  * header.
3821  */
3822 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3823 {
3824 	unsigned char *csum_start = skb_transport_header(skb);
3825 	int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3826 	__wsum partial = SKB_GSO_CB(skb)->csum;
3827 
3828 	SKB_GSO_CB(skb)->csum = res;
3829 	SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3830 
3831 	return csum_fold(csum_partial(csum_start, plen, partial));
3832 }
3833 
3834 static inline bool skb_is_gso(const struct sk_buff *skb)
3835 {
3836 	return skb_shinfo(skb)->gso_size;
3837 }
3838 
3839 /* Note: Should be called only if skb_is_gso(skb) is true */
3840 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3841 {
3842 	return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3843 }
3844 
3845 static inline void skb_gso_reset(struct sk_buff *skb)
3846 {
3847 	skb_shinfo(skb)->gso_size = 0;
3848 	skb_shinfo(skb)->gso_segs = 0;
3849 	skb_shinfo(skb)->gso_type = 0;
3850 }
3851 
3852 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3853 
3854 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3855 {
3856 	/* LRO sets gso_size but not gso_type, whereas if GSO is really
3857 	 * wanted then gso_type will be set. */
3858 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
3859 
3860 	if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3861 	    unlikely(shinfo->gso_type == 0)) {
3862 		__skb_warn_lro_forwarding(skb);
3863 		return true;
3864 	}
3865 	return false;
3866 }
3867 
3868 static inline void skb_forward_csum(struct sk_buff *skb)
3869 {
3870 	/* Unfortunately we don't support this one.  Any brave souls? */
3871 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3872 		skb->ip_summed = CHECKSUM_NONE;
3873 }
3874 
3875 /**
3876  * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3877  * @skb: skb to check
3878  *
3879  * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3880  * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3881  * use this helper, to document places where we make this assertion.
3882  */
3883 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3884 {
3885 #ifdef DEBUG
3886 	BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3887 #endif
3888 }
3889 
3890 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3891 
3892 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3893 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3894 				     unsigned int transport_len,
3895 				     __sum16(*skb_chkf)(struct sk_buff *skb));
3896 
3897 /**
3898  * skb_head_is_locked - Determine if the skb->head is locked down
3899  * @skb: skb to check
3900  *
3901  * The head on skbs build around a head frag can be removed if they are
3902  * not cloned.  This function returns true if the skb head is locked down
3903  * due to either being allocated via kmalloc, or by being a clone with
3904  * multiple references to the head.
3905  */
3906 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3907 {
3908 	return !skb->head_frag || skb_cloned(skb);
3909 }
3910 
3911 /**
3912  * skb_gso_network_seglen - Return length of individual segments of a gso packet
3913  *
3914  * @skb: GSO skb
3915  *
3916  * skb_gso_network_seglen is used to determine the real size of the
3917  * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3918  *
3919  * The MAC/L2 header is not accounted for.
3920  */
3921 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3922 {
3923 	unsigned int hdr_len = skb_transport_header(skb) -
3924 			       skb_network_header(skb);
3925 	return hdr_len + skb_gso_transport_seglen(skb);
3926 }
3927 
3928 /* Local Checksum Offload.
3929  * Compute outer checksum based on the assumption that the
3930  * inner checksum will be offloaded later.
3931  * See Documentation/networking/checksum-offloads.txt for
3932  * explanation of how this works.
3933  * Fill in outer checksum adjustment (e.g. with sum of outer
3934  * pseudo-header) before calling.
3935  * Also ensure that inner checksum is in linear data area.
3936  */
3937 static inline __wsum lco_csum(struct sk_buff *skb)
3938 {
3939 	unsigned char *csum_start = skb_checksum_start(skb);
3940 	unsigned char *l4_hdr = skb_transport_header(skb);
3941 	__wsum partial;
3942 
3943 	/* Start with complement of inner checksum adjustment */
3944 	partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3945 						    skb->csum_offset));
3946 
3947 	/* Add in checksum of our headers (incl. outer checksum
3948 	 * adjustment filled in by caller) and return result.
3949 	 */
3950 	return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3951 }
3952 
3953 #endif	/* __KERNEL__ */
3954 #endif	/* _LINUX_SKBUFF_H */
3955