xref: /linux-6.15/include/linux/skbuff.h (revision bbb03029)
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 
467 	/* This indicates the skb is from an untrusted source. */
468 	SKB_GSO_DODGY = 1 << 1,
469 
470 	/* This indicates the tcp segment has CWR set. */
471 	SKB_GSO_TCP_ECN = 1 << 2,
472 
473 	SKB_GSO_TCP_FIXEDID = 1 << 3,
474 
475 	SKB_GSO_TCPV6 = 1 << 4,
476 
477 	SKB_GSO_FCOE = 1 << 5,
478 
479 	SKB_GSO_GRE = 1 << 6,
480 
481 	SKB_GSO_GRE_CSUM = 1 << 7,
482 
483 	SKB_GSO_IPXIP4 = 1 << 8,
484 
485 	SKB_GSO_IPXIP6 = 1 << 9,
486 
487 	SKB_GSO_UDP_TUNNEL = 1 << 10,
488 
489 	SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
490 
491 	SKB_GSO_PARTIAL = 1 << 12,
492 
493 	SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
494 
495 	SKB_GSO_SCTP = 1 << 14,
496 
497 	SKB_GSO_ESP = 1 << 15,
498 };
499 
500 #if BITS_PER_LONG > 32
501 #define NET_SKBUFF_DATA_USES_OFFSET 1
502 #endif
503 
504 #ifdef NET_SKBUFF_DATA_USES_OFFSET
505 typedef unsigned int sk_buff_data_t;
506 #else
507 typedef unsigned char *sk_buff_data_t;
508 #endif
509 
510 /**
511  *	struct sk_buff - socket buffer
512  *	@next: Next buffer in list
513  *	@prev: Previous buffer in list
514  *	@tstamp: Time we arrived/left
515  *	@rbnode: RB tree node, alternative to next/prev for netem/tcp
516  *	@sk: Socket we are owned by
517  *	@dev: Device we arrived on/are leaving by
518  *	@cb: Control buffer. Free for use by every layer. Put private vars here
519  *	@_skb_refdst: destination entry (with norefcount bit)
520  *	@sp: the security path, used for xfrm
521  *	@len: Length of actual data
522  *	@data_len: Data length
523  *	@mac_len: Length of link layer header
524  *	@hdr_len: writable header length of cloned skb
525  *	@csum: Checksum (must include start/offset pair)
526  *	@csum_start: Offset from skb->head where checksumming should start
527  *	@csum_offset: Offset from csum_start where checksum should be stored
528  *	@priority: Packet queueing priority
529  *	@ignore_df: allow local fragmentation
530  *	@cloned: Head may be cloned (check refcnt to be sure)
531  *	@ip_summed: Driver fed us an IP checksum
532  *	@nohdr: Payload reference only, must not modify header
533  *	@pkt_type: Packet class
534  *	@fclone: skbuff clone status
535  *	@ipvs_property: skbuff is owned by ipvs
536  *	@tc_skip_classify: do not classify packet. set by IFB device
537  *	@tc_at_ingress: used within tc_classify to distinguish in/egress
538  *	@tc_redirected: packet was redirected by a tc action
539  *	@tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
540  *	@peeked: this packet has been seen already, so stats have been
541  *		done for it, don't do them again
542  *	@nf_trace: netfilter packet trace flag
543  *	@protocol: Packet protocol from driver
544  *	@destructor: Destruct function
545  *	@_nfct: Associated connection, if any (with nfctinfo bits)
546  *	@nf_bridge: Saved data about a bridged frame - see br_netfilter.c
547  *	@skb_iif: ifindex of device we arrived on
548  *	@tc_index: Traffic control index
549  *	@hash: the packet hash
550  *	@queue_mapping: Queue mapping for multiqueue devices
551  *	@xmit_more: More SKBs are pending for this queue
552  *	@ndisc_nodetype: router type (from link layer)
553  *	@ooo_okay: allow the mapping of a socket to a queue to be changed
554  *	@l4_hash: indicate hash is a canonical 4-tuple hash over transport
555  *		ports.
556  *	@sw_hash: indicates hash was computed in software stack
557  *	@wifi_acked_valid: wifi_acked was set
558  *	@wifi_acked: whether frame was acked on wifi or not
559  *	@no_fcs:  Request NIC to treat last 4 bytes as Ethernet FCS
560  *	@csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
561  *	@dst_pending_confirm: need to confirm neighbour
562   *	@napi_id: id of the NAPI struct this skb came from
563  *	@secmark: security marking
564  *	@mark: Generic packet mark
565  *	@vlan_proto: vlan encapsulation protocol
566  *	@vlan_tci: vlan tag control information
567  *	@inner_protocol: Protocol (encapsulation)
568  *	@inner_transport_header: Inner transport layer header (encapsulation)
569  *	@inner_network_header: Network layer header (encapsulation)
570  *	@inner_mac_header: Link layer header (encapsulation)
571  *	@transport_header: Transport layer header
572  *	@network_header: Network layer header
573  *	@mac_header: Link layer header
574  *	@tail: Tail pointer
575  *	@end: End pointer
576  *	@head: Head of buffer
577  *	@data: Data head pointer
578  *	@truesize: Buffer size
579  *	@users: User count - see {datagram,tcp}.c
580  */
581 
582 struct sk_buff {
583 	union {
584 		struct {
585 			/* These two members must be first. */
586 			struct sk_buff		*next;
587 			struct sk_buff		*prev;
588 
589 			union {
590 				ktime_t		tstamp;
591 				u64		skb_mstamp;
592 			};
593 		};
594 		struct rb_node	rbnode; /* used in netem & tcp stack */
595 	};
596 	struct sock		*sk;
597 
598 	union {
599 		struct net_device	*dev;
600 		/* Some protocols might use this space to store information,
601 		 * while device pointer would be NULL.
602 		 * UDP receive path is one user.
603 		 */
604 		unsigned long		dev_scratch;
605 	};
606 	/*
607 	 * This is the control buffer. It is free to use for every
608 	 * layer. Please put your private variables there. If you
609 	 * want to keep them across layers you have to do a skb_clone()
610 	 * first. This is owned by whoever has the skb queued ATM.
611 	 */
612 	char			cb[48] __aligned(8);
613 
614 	unsigned long		_skb_refdst;
615 	void			(*destructor)(struct sk_buff *skb);
616 #ifdef CONFIG_XFRM
617 	struct	sec_path	*sp;
618 #endif
619 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
620 	unsigned long		 _nfct;
621 #endif
622 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
623 	struct nf_bridge_info	*nf_bridge;
624 #endif
625 	unsigned int		len,
626 				data_len;
627 	__u16			mac_len,
628 				hdr_len;
629 
630 	/* Following fields are _not_ copied in __copy_skb_header()
631 	 * Note that queue_mapping is here mostly to fill a hole.
632 	 */
633 	kmemcheck_bitfield_begin(flags1);
634 	__u16			queue_mapping;
635 
636 /* if you move cloned around you also must adapt those constants */
637 #ifdef __BIG_ENDIAN_BITFIELD
638 #define CLONED_MASK	(1 << 7)
639 #else
640 #define CLONED_MASK	1
641 #endif
642 #define CLONED_OFFSET()		offsetof(struct sk_buff, __cloned_offset)
643 
644 	__u8			__cloned_offset[0];
645 	__u8			cloned:1,
646 				nohdr:1,
647 				fclone:2,
648 				peeked:1,
649 				head_frag:1,
650 				xmit_more:1,
651 				__unused:1; /* one bit hole */
652 	kmemcheck_bitfield_end(flags1);
653 
654 	/* fields enclosed in headers_start/headers_end are copied
655 	 * using a single memcpy() in __copy_skb_header()
656 	 */
657 	/* private: */
658 	__u32			headers_start[0];
659 	/* public: */
660 
661 /* if you move pkt_type around you also must adapt those constants */
662 #ifdef __BIG_ENDIAN_BITFIELD
663 #define PKT_TYPE_MAX	(7 << 5)
664 #else
665 #define PKT_TYPE_MAX	7
666 #endif
667 #define PKT_TYPE_OFFSET()	offsetof(struct sk_buff, __pkt_type_offset)
668 
669 	__u8			__pkt_type_offset[0];
670 	__u8			pkt_type:3;
671 	__u8			pfmemalloc:1;
672 	__u8			ignore_df:1;
673 
674 	__u8			nf_trace:1;
675 	__u8			ip_summed:2;
676 	__u8			ooo_okay:1;
677 	__u8			l4_hash:1;
678 	__u8			sw_hash:1;
679 	__u8			wifi_acked_valid:1;
680 	__u8			wifi_acked:1;
681 
682 	__u8			no_fcs:1;
683 	/* Indicates the inner headers are valid in the skbuff. */
684 	__u8			encapsulation:1;
685 	__u8			encap_hdr_csum:1;
686 	__u8			csum_valid:1;
687 	__u8			csum_complete_sw:1;
688 	__u8			csum_level:2;
689 	__u8			csum_not_inet:1;
690 
691 	__u8			dst_pending_confirm:1;
692 #ifdef CONFIG_IPV6_NDISC_NODETYPE
693 	__u8			ndisc_nodetype:2;
694 #endif
695 	__u8			ipvs_property:1;
696 	__u8			inner_protocol_type:1;
697 	__u8			remcsum_offload:1;
698 #ifdef CONFIG_NET_SWITCHDEV
699 	__u8			offload_fwd_mark:1;
700 #endif
701 #ifdef CONFIG_NET_CLS_ACT
702 	__u8			tc_skip_classify:1;
703 	__u8			tc_at_ingress:1;
704 	__u8			tc_redirected:1;
705 	__u8			tc_from_ingress:1;
706 #endif
707 
708 #ifdef CONFIG_NET_SCHED
709 	__u16			tc_index;	/* traffic control index */
710 #endif
711 
712 	union {
713 		__wsum		csum;
714 		struct {
715 			__u16	csum_start;
716 			__u16	csum_offset;
717 		};
718 	};
719 	__u32			priority;
720 	int			skb_iif;
721 	__u32			hash;
722 	__be16			vlan_proto;
723 	__u16			vlan_tci;
724 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
725 	union {
726 		unsigned int	napi_id;
727 		unsigned int	sender_cpu;
728 	};
729 #endif
730 #ifdef CONFIG_NETWORK_SECMARK
731 	__u32		secmark;
732 #endif
733 
734 	union {
735 		__u32		mark;
736 		__u32		reserved_tailroom;
737 	};
738 
739 	union {
740 		__be16		inner_protocol;
741 		__u8		inner_ipproto;
742 	};
743 
744 	__u16			inner_transport_header;
745 	__u16			inner_network_header;
746 	__u16			inner_mac_header;
747 
748 	__be16			protocol;
749 	__u16			transport_header;
750 	__u16			network_header;
751 	__u16			mac_header;
752 
753 	/* private: */
754 	__u32			headers_end[0];
755 	/* public: */
756 
757 	/* These elements must be at the end, see alloc_skb() for details.  */
758 	sk_buff_data_t		tail;
759 	sk_buff_data_t		end;
760 	unsigned char		*head,
761 				*data;
762 	unsigned int		truesize;
763 	refcount_t		users;
764 };
765 
766 #ifdef __KERNEL__
767 /*
768  *	Handling routines are only of interest to the kernel
769  */
770 #include <linux/slab.h>
771 
772 
773 #define SKB_ALLOC_FCLONE	0x01
774 #define SKB_ALLOC_RX		0x02
775 #define SKB_ALLOC_NAPI		0x04
776 
777 /* Returns true if the skb was allocated from PFMEMALLOC reserves */
778 static inline bool skb_pfmemalloc(const struct sk_buff *skb)
779 {
780 	return unlikely(skb->pfmemalloc);
781 }
782 
783 /*
784  * skb might have a dst pointer attached, refcounted or not.
785  * _skb_refdst low order bit is set if refcount was _not_ taken
786  */
787 #define SKB_DST_NOREF	1UL
788 #define SKB_DST_PTRMASK	~(SKB_DST_NOREF)
789 
790 #define SKB_NFCT_PTRMASK	~(7UL)
791 /**
792  * skb_dst - returns skb dst_entry
793  * @skb: buffer
794  *
795  * Returns skb dst_entry, regardless of reference taken or not.
796  */
797 static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
798 {
799 	/* If refdst was not refcounted, check we still are in a
800 	 * rcu_read_lock section
801 	 */
802 	WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
803 		!rcu_read_lock_held() &&
804 		!rcu_read_lock_bh_held());
805 	return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
806 }
807 
808 /**
809  * skb_dst_set - sets skb dst
810  * @skb: buffer
811  * @dst: dst entry
812  *
813  * Sets skb dst, assuming a reference was taken on dst and should
814  * be released by skb_dst_drop()
815  */
816 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
817 {
818 	skb->_skb_refdst = (unsigned long)dst;
819 }
820 
821 /**
822  * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
823  * @skb: buffer
824  * @dst: dst entry
825  *
826  * Sets skb dst, assuming a reference was not taken on dst.
827  * If dst entry is cached, we do not take reference and dst_release
828  * will be avoided by refdst_drop. If dst entry is not cached, we take
829  * reference, so that last dst_release can destroy the dst immediately.
830  */
831 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
832 {
833 	WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
834 	skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
835 }
836 
837 /**
838  * skb_dst_is_noref - Test if skb dst isn't refcounted
839  * @skb: buffer
840  */
841 static inline bool skb_dst_is_noref(const struct sk_buff *skb)
842 {
843 	return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
844 }
845 
846 static inline struct rtable *skb_rtable(const struct sk_buff *skb)
847 {
848 	return (struct rtable *)skb_dst(skb);
849 }
850 
851 /* For mangling skb->pkt_type from user space side from applications
852  * such as nft, tc, etc, we only allow a conservative subset of
853  * possible pkt_types to be set.
854 */
855 static inline bool skb_pkt_type_ok(u32 ptype)
856 {
857 	return ptype <= PACKET_OTHERHOST;
858 }
859 
860 static inline unsigned int skb_napi_id(const struct sk_buff *skb)
861 {
862 #ifdef CONFIG_NET_RX_BUSY_POLL
863 	return skb->napi_id;
864 #else
865 	return 0;
866 #endif
867 }
868 
869 /* decrement the reference count and return true if we can free the skb */
870 static inline bool skb_unref(struct sk_buff *skb)
871 {
872 	if (unlikely(!skb))
873 		return false;
874 	if (likely(refcount_read(&skb->users) == 1))
875 		smp_rmb();
876 	else if (likely(!refcount_dec_and_test(&skb->users)))
877 		return false;
878 
879 	return true;
880 }
881 
882 void skb_release_head_state(struct sk_buff *skb);
883 void kfree_skb(struct sk_buff *skb);
884 void kfree_skb_list(struct sk_buff *segs);
885 void skb_tx_error(struct sk_buff *skb);
886 void consume_skb(struct sk_buff *skb);
887 void consume_stateless_skb(struct sk_buff *skb);
888 void  __kfree_skb(struct sk_buff *skb);
889 extern struct kmem_cache *skbuff_head_cache;
890 
891 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
892 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
893 		      bool *fragstolen, int *delta_truesize);
894 
895 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
896 			    int node);
897 struct sk_buff *__build_skb(void *data, unsigned int frag_size);
898 struct sk_buff *build_skb(void *data, unsigned int frag_size);
899 static inline struct sk_buff *alloc_skb(unsigned int size,
900 					gfp_t priority)
901 {
902 	return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
903 }
904 
905 struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
906 				     unsigned long data_len,
907 				     int max_page_order,
908 				     int *errcode,
909 				     gfp_t gfp_mask);
910 
911 /* Layout of fast clones : [skb1][skb2][fclone_ref] */
912 struct sk_buff_fclones {
913 	struct sk_buff	skb1;
914 
915 	struct sk_buff	skb2;
916 
917 	refcount_t	fclone_ref;
918 };
919 
920 /**
921  *	skb_fclone_busy - check if fclone is busy
922  *	@sk: socket
923  *	@skb: buffer
924  *
925  * Returns true if skb is a fast clone, and its clone is not freed.
926  * Some drivers call skb_orphan() in their ndo_start_xmit(),
927  * so we also check that this didnt happen.
928  */
929 static inline bool skb_fclone_busy(const struct sock *sk,
930 				   const struct sk_buff *skb)
931 {
932 	const struct sk_buff_fclones *fclones;
933 
934 	fclones = container_of(skb, struct sk_buff_fclones, skb1);
935 
936 	return skb->fclone == SKB_FCLONE_ORIG &&
937 	       refcount_read(&fclones->fclone_ref) > 1 &&
938 	       fclones->skb2.sk == sk;
939 }
940 
941 static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
942 					       gfp_t priority)
943 {
944 	return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
945 }
946 
947 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
948 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
949 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
950 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
951 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
952 				   gfp_t gfp_mask, bool fclone);
953 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
954 					  gfp_t gfp_mask)
955 {
956 	return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
957 }
958 
959 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
960 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
961 				     unsigned int headroom);
962 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
963 				int newtailroom, gfp_t priority);
964 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
965 				     int offset, int len);
966 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
967 			      int offset, int len);
968 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
969 int skb_pad(struct sk_buff *skb, int pad);
970 #define dev_kfree_skb(a)	consume_skb(a)
971 
972 int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
973 			    int getfrag(void *from, char *to, int offset,
974 					int len, int odd, struct sk_buff *skb),
975 			    void *from, int length);
976 
977 int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
978 			 int offset, size_t size);
979 
980 struct skb_seq_state {
981 	__u32		lower_offset;
982 	__u32		upper_offset;
983 	__u32		frag_idx;
984 	__u32		stepped_offset;
985 	struct sk_buff	*root_skb;
986 	struct sk_buff	*cur_skb;
987 	__u8		*frag_data;
988 };
989 
990 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
991 			  unsigned int to, struct skb_seq_state *st);
992 unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
993 			  struct skb_seq_state *st);
994 void skb_abort_seq_read(struct skb_seq_state *st);
995 
996 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
997 			   unsigned int to, struct ts_config *config);
998 
999 /*
1000  * Packet hash types specify the type of hash in skb_set_hash.
1001  *
1002  * Hash types refer to the protocol layer addresses which are used to
1003  * construct a packet's hash. The hashes are used to differentiate or identify
1004  * flows of the protocol layer for the hash type. Hash types are either
1005  * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1006  *
1007  * Properties of hashes:
1008  *
1009  * 1) Two packets in different flows have different hash values
1010  * 2) Two packets in the same flow should have the same hash value
1011  *
1012  * A hash at a higher layer is considered to be more specific. A driver should
1013  * set the most specific hash possible.
1014  *
1015  * A driver cannot indicate a more specific hash than the layer at which a hash
1016  * was computed. For instance an L3 hash cannot be set as an L4 hash.
1017  *
1018  * A driver may indicate a hash level which is less specific than the
1019  * actual layer the hash was computed on. For instance, a hash computed
1020  * at L4 may be considered an L3 hash. This should only be done if the
1021  * driver can't unambiguously determine that the HW computed the hash at
1022  * the higher layer. Note that the "should" in the second property above
1023  * permits this.
1024  */
1025 enum pkt_hash_types {
1026 	PKT_HASH_TYPE_NONE,	/* Undefined type */
1027 	PKT_HASH_TYPE_L2,	/* Input: src_MAC, dest_MAC */
1028 	PKT_HASH_TYPE_L3,	/* Input: src_IP, dst_IP */
1029 	PKT_HASH_TYPE_L4,	/* Input: src_IP, dst_IP, src_port, dst_port */
1030 };
1031 
1032 static inline void skb_clear_hash(struct sk_buff *skb)
1033 {
1034 	skb->hash = 0;
1035 	skb->sw_hash = 0;
1036 	skb->l4_hash = 0;
1037 }
1038 
1039 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1040 {
1041 	if (!skb->l4_hash)
1042 		skb_clear_hash(skb);
1043 }
1044 
1045 static inline void
1046 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1047 {
1048 	skb->l4_hash = is_l4;
1049 	skb->sw_hash = is_sw;
1050 	skb->hash = hash;
1051 }
1052 
1053 static inline void
1054 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1055 {
1056 	/* Used by drivers to set hash from HW */
1057 	__skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1058 }
1059 
1060 static inline void
1061 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1062 {
1063 	__skb_set_hash(skb, hash, true, is_l4);
1064 }
1065 
1066 void __skb_get_hash(struct sk_buff *skb);
1067 u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1068 u32 skb_get_poff(const struct sk_buff *skb);
1069 u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1070 		   const struct flow_keys *keys, int hlen);
1071 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1072 			    void *data, int hlen_proto);
1073 
1074 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1075 					int thoff, u8 ip_proto)
1076 {
1077 	return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1078 }
1079 
1080 void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1081 			     const struct flow_dissector_key *key,
1082 			     unsigned int key_count);
1083 
1084 bool __skb_flow_dissect(const struct sk_buff *skb,
1085 			struct flow_dissector *flow_dissector,
1086 			void *target_container,
1087 			void *data, __be16 proto, int nhoff, int hlen,
1088 			unsigned int flags);
1089 
1090 static inline bool skb_flow_dissect(const struct sk_buff *skb,
1091 				    struct flow_dissector *flow_dissector,
1092 				    void *target_container, unsigned int flags)
1093 {
1094 	return __skb_flow_dissect(skb, flow_dissector, target_container,
1095 				  NULL, 0, 0, 0, flags);
1096 }
1097 
1098 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1099 					      struct flow_keys *flow,
1100 					      unsigned int flags)
1101 {
1102 	memset(flow, 0, sizeof(*flow));
1103 	return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1104 				  NULL, 0, 0, 0, flags);
1105 }
1106 
1107 static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1108 						  void *data, __be16 proto,
1109 						  int nhoff, int hlen,
1110 						  unsigned int flags)
1111 {
1112 	memset(flow, 0, sizeof(*flow));
1113 	return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1114 				  data, proto, nhoff, hlen, flags);
1115 }
1116 
1117 static inline __u32 skb_get_hash(struct sk_buff *skb)
1118 {
1119 	if (!skb->l4_hash && !skb->sw_hash)
1120 		__skb_get_hash(skb);
1121 
1122 	return skb->hash;
1123 }
1124 
1125 __u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
1126 
1127 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1128 {
1129 	if (!skb->l4_hash && !skb->sw_hash) {
1130 		struct flow_keys keys;
1131 		__u32 hash = __get_hash_from_flowi6(fl6, &keys);
1132 
1133 		__skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1134 	}
1135 
1136 	return skb->hash;
1137 }
1138 
1139 __u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
1140 
1141 static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
1142 {
1143 	if (!skb->l4_hash && !skb->sw_hash) {
1144 		struct flow_keys keys;
1145 		__u32 hash = __get_hash_from_flowi4(fl4, &keys);
1146 
1147 		__skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1148 	}
1149 
1150 	return skb->hash;
1151 }
1152 
1153 __u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1154 
1155 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1156 {
1157 	return skb->hash;
1158 }
1159 
1160 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1161 {
1162 	to->hash = from->hash;
1163 	to->sw_hash = from->sw_hash;
1164 	to->l4_hash = from->l4_hash;
1165 };
1166 
1167 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1168 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1169 {
1170 	return skb->head + skb->end;
1171 }
1172 
1173 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1174 {
1175 	return skb->end;
1176 }
1177 #else
1178 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1179 {
1180 	return skb->end;
1181 }
1182 
1183 static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1184 {
1185 	return skb->end - skb->head;
1186 }
1187 #endif
1188 
1189 /* Internal */
1190 #define skb_shinfo(SKB)	((struct skb_shared_info *)(skb_end_pointer(SKB)))
1191 
1192 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1193 {
1194 	return &skb_shinfo(skb)->hwtstamps;
1195 }
1196 
1197 /**
1198  *	skb_queue_empty - check if a queue is empty
1199  *	@list: queue head
1200  *
1201  *	Returns true if the queue is empty, false otherwise.
1202  */
1203 static inline int skb_queue_empty(const struct sk_buff_head *list)
1204 {
1205 	return list->next == (const struct sk_buff *) list;
1206 }
1207 
1208 /**
1209  *	skb_queue_is_last - check if skb is the last entry in the queue
1210  *	@list: queue head
1211  *	@skb: buffer
1212  *
1213  *	Returns true if @skb is the last buffer on the list.
1214  */
1215 static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1216 				     const struct sk_buff *skb)
1217 {
1218 	return skb->next == (const struct sk_buff *) list;
1219 }
1220 
1221 /**
1222  *	skb_queue_is_first - check if skb is the first entry in the queue
1223  *	@list: queue head
1224  *	@skb: buffer
1225  *
1226  *	Returns true if @skb is the first buffer on the list.
1227  */
1228 static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1229 				      const struct sk_buff *skb)
1230 {
1231 	return skb->prev == (const struct sk_buff *) list;
1232 }
1233 
1234 /**
1235  *	skb_queue_next - return the next packet in the queue
1236  *	@list: queue head
1237  *	@skb: current buffer
1238  *
1239  *	Return the next packet in @list after @skb.  It is only valid to
1240  *	call this if skb_queue_is_last() evaluates to false.
1241  */
1242 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1243 					     const struct sk_buff *skb)
1244 {
1245 	/* This BUG_ON may seem severe, but if we just return then we
1246 	 * are going to dereference garbage.
1247 	 */
1248 	BUG_ON(skb_queue_is_last(list, skb));
1249 	return skb->next;
1250 }
1251 
1252 /**
1253  *	skb_queue_prev - return the prev packet in the queue
1254  *	@list: queue head
1255  *	@skb: current buffer
1256  *
1257  *	Return the prev packet in @list before @skb.  It is only valid to
1258  *	call this if skb_queue_is_first() evaluates to false.
1259  */
1260 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1261 					     const struct sk_buff *skb)
1262 {
1263 	/* This BUG_ON may seem severe, but if we just return then we
1264 	 * are going to dereference garbage.
1265 	 */
1266 	BUG_ON(skb_queue_is_first(list, skb));
1267 	return skb->prev;
1268 }
1269 
1270 /**
1271  *	skb_get - reference buffer
1272  *	@skb: buffer to reference
1273  *
1274  *	Makes another reference to a socket buffer and returns a pointer
1275  *	to the buffer.
1276  */
1277 static inline struct sk_buff *skb_get(struct sk_buff *skb)
1278 {
1279 	refcount_inc(&skb->users);
1280 	return skb;
1281 }
1282 
1283 /*
1284  * If users == 1, we are the only owner and are can avoid redundant
1285  * atomic change.
1286  */
1287 
1288 /**
1289  *	skb_cloned - is the buffer a clone
1290  *	@skb: buffer to check
1291  *
1292  *	Returns true if the buffer was generated with skb_clone() and is
1293  *	one of multiple shared copies of the buffer. Cloned buffers are
1294  *	shared data so must not be written to under normal circumstances.
1295  */
1296 static inline int skb_cloned(const struct sk_buff *skb)
1297 {
1298 	return skb->cloned &&
1299 	       (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1300 }
1301 
1302 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1303 {
1304 	might_sleep_if(gfpflags_allow_blocking(pri));
1305 
1306 	if (skb_cloned(skb))
1307 		return pskb_expand_head(skb, 0, 0, pri);
1308 
1309 	return 0;
1310 }
1311 
1312 /**
1313  *	skb_header_cloned - is the header a clone
1314  *	@skb: buffer to check
1315  *
1316  *	Returns true if modifying the header part of the buffer requires
1317  *	the data to be copied.
1318  */
1319 static inline int skb_header_cloned(const struct sk_buff *skb)
1320 {
1321 	int dataref;
1322 
1323 	if (!skb->cloned)
1324 		return 0;
1325 
1326 	dataref = atomic_read(&skb_shinfo(skb)->dataref);
1327 	dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1328 	return dataref != 1;
1329 }
1330 
1331 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1332 {
1333 	might_sleep_if(gfpflags_allow_blocking(pri));
1334 
1335 	if (skb_header_cloned(skb))
1336 		return pskb_expand_head(skb, 0, 0, pri);
1337 
1338 	return 0;
1339 }
1340 
1341 /**
1342  *	skb_header_release - release reference to header
1343  *	@skb: buffer to operate on
1344  *
1345  *	Drop a reference to the header part of the buffer.  This is done
1346  *	by acquiring a payload reference.  You must not read from the header
1347  *	part of skb->data after this.
1348  *	Note : Check if you can use __skb_header_release() instead.
1349  */
1350 static inline void skb_header_release(struct sk_buff *skb)
1351 {
1352 	BUG_ON(skb->nohdr);
1353 	skb->nohdr = 1;
1354 	atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1355 }
1356 
1357 /**
1358  *	__skb_header_release - release reference to header
1359  *	@skb: buffer to operate on
1360  *
1361  *	Variant of skb_header_release() assuming skb is private to caller.
1362  *	We can avoid one atomic operation.
1363  */
1364 static inline void __skb_header_release(struct sk_buff *skb)
1365 {
1366 	skb->nohdr = 1;
1367 	atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1368 }
1369 
1370 
1371 /**
1372  *	skb_shared - is the buffer shared
1373  *	@skb: buffer to check
1374  *
1375  *	Returns true if more than one person has a reference to this
1376  *	buffer.
1377  */
1378 static inline int skb_shared(const struct sk_buff *skb)
1379 {
1380 	return refcount_read(&skb->users) != 1;
1381 }
1382 
1383 /**
1384  *	skb_share_check - check if buffer is shared and if so clone it
1385  *	@skb: buffer to check
1386  *	@pri: priority for memory allocation
1387  *
1388  *	If the buffer is shared the buffer is cloned and the old copy
1389  *	drops a reference. A new clone with a single reference is returned.
1390  *	If the buffer is not shared the original buffer is returned. When
1391  *	being called from interrupt status or with spinlocks held pri must
1392  *	be GFP_ATOMIC.
1393  *
1394  *	NULL is returned on a memory allocation failure.
1395  */
1396 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1397 {
1398 	might_sleep_if(gfpflags_allow_blocking(pri));
1399 	if (skb_shared(skb)) {
1400 		struct sk_buff *nskb = skb_clone(skb, pri);
1401 
1402 		if (likely(nskb))
1403 			consume_skb(skb);
1404 		else
1405 			kfree_skb(skb);
1406 		skb = nskb;
1407 	}
1408 	return skb;
1409 }
1410 
1411 /*
1412  *	Copy shared buffers into a new sk_buff. We effectively do COW on
1413  *	packets to handle cases where we have a local reader and forward
1414  *	and a couple of other messy ones. The normal one is tcpdumping
1415  *	a packet thats being forwarded.
1416  */
1417 
1418 /**
1419  *	skb_unshare - make a copy of a shared buffer
1420  *	@skb: buffer to check
1421  *	@pri: priority for memory allocation
1422  *
1423  *	If the socket buffer is a clone then this function creates a new
1424  *	copy of the data, drops a reference count on the old copy and returns
1425  *	the new copy with the reference count at 1. If the buffer is not a clone
1426  *	the original buffer is returned. When called with a spinlock held or
1427  *	from interrupt state @pri must be %GFP_ATOMIC
1428  *
1429  *	%NULL is returned on a memory allocation failure.
1430  */
1431 static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1432 					  gfp_t pri)
1433 {
1434 	might_sleep_if(gfpflags_allow_blocking(pri));
1435 	if (skb_cloned(skb)) {
1436 		struct sk_buff *nskb = skb_copy(skb, pri);
1437 
1438 		/* Free our shared copy */
1439 		if (likely(nskb))
1440 			consume_skb(skb);
1441 		else
1442 			kfree_skb(skb);
1443 		skb = nskb;
1444 	}
1445 	return skb;
1446 }
1447 
1448 /**
1449  *	skb_peek - peek at the head of an &sk_buff_head
1450  *	@list_: list to peek at
1451  *
1452  *	Peek an &sk_buff. Unlike most other operations you _MUST_
1453  *	be careful with this one. A peek leaves the buffer on the
1454  *	list and someone else may run off with it. You must hold
1455  *	the appropriate locks or have a private queue to do this.
1456  *
1457  *	Returns %NULL for an empty list or a pointer to the head element.
1458  *	The reference count is not incremented and the reference is therefore
1459  *	volatile. Use with caution.
1460  */
1461 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1462 {
1463 	struct sk_buff *skb = list_->next;
1464 
1465 	if (skb == (struct sk_buff *)list_)
1466 		skb = NULL;
1467 	return skb;
1468 }
1469 
1470 /**
1471  *	skb_peek_next - peek skb following the given one from a queue
1472  *	@skb: skb to start from
1473  *	@list_: list to peek at
1474  *
1475  *	Returns %NULL when the end of the list is met or a pointer to the
1476  *	next element. The reference count is not incremented and the
1477  *	reference is therefore volatile. Use with caution.
1478  */
1479 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1480 		const struct sk_buff_head *list_)
1481 {
1482 	struct sk_buff *next = skb->next;
1483 
1484 	if (next == (struct sk_buff *)list_)
1485 		next = NULL;
1486 	return next;
1487 }
1488 
1489 /**
1490  *	skb_peek_tail - peek at the tail of an &sk_buff_head
1491  *	@list_: list to peek at
1492  *
1493  *	Peek an &sk_buff. Unlike most other operations you _MUST_
1494  *	be careful with this one. A peek leaves the buffer on the
1495  *	list and someone else may run off with it. You must hold
1496  *	the appropriate locks or have a private queue to do this.
1497  *
1498  *	Returns %NULL for an empty list or a pointer to the tail element.
1499  *	The reference count is not incremented and the reference is therefore
1500  *	volatile. Use with caution.
1501  */
1502 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1503 {
1504 	struct sk_buff *skb = list_->prev;
1505 
1506 	if (skb == (struct sk_buff *)list_)
1507 		skb = NULL;
1508 	return skb;
1509 
1510 }
1511 
1512 /**
1513  *	skb_queue_len	- get queue length
1514  *	@list_: list to measure
1515  *
1516  *	Return the length of an &sk_buff queue.
1517  */
1518 static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1519 {
1520 	return list_->qlen;
1521 }
1522 
1523 /**
1524  *	__skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1525  *	@list: queue to initialize
1526  *
1527  *	This initializes only the list and queue length aspects of
1528  *	an sk_buff_head object.  This allows to initialize the list
1529  *	aspects of an sk_buff_head without reinitializing things like
1530  *	the spinlock.  It can also be used for on-stack sk_buff_head
1531  *	objects where the spinlock is known to not be used.
1532  */
1533 static inline void __skb_queue_head_init(struct sk_buff_head *list)
1534 {
1535 	list->prev = list->next = (struct sk_buff *)list;
1536 	list->qlen = 0;
1537 }
1538 
1539 /*
1540  * This function creates a split out lock class for each invocation;
1541  * this is needed for now since a whole lot of users of the skb-queue
1542  * infrastructure in drivers have different locking usage (in hardirq)
1543  * than the networking core (in softirq only). In the long run either the
1544  * network layer or drivers should need annotation to consolidate the
1545  * main types of usage into 3 classes.
1546  */
1547 static inline void skb_queue_head_init(struct sk_buff_head *list)
1548 {
1549 	spin_lock_init(&list->lock);
1550 	__skb_queue_head_init(list);
1551 }
1552 
1553 static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1554 		struct lock_class_key *class)
1555 {
1556 	skb_queue_head_init(list);
1557 	lockdep_set_class(&list->lock, class);
1558 }
1559 
1560 /*
1561  *	Insert an sk_buff on a list.
1562  *
1563  *	The "__skb_xxxx()" functions are the non-atomic ones that
1564  *	can only be called with interrupts disabled.
1565  */
1566 void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1567 		struct sk_buff_head *list);
1568 static inline void __skb_insert(struct sk_buff *newsk,
1569 				struct sk_buff *prev, struct sk_buff *next,
1570 				struct sk_buff_head *list)
1571 {
1572 	newsk->next = next;
1573 	newsk->prev = prev;
1574 	next->prev  = prev->next = newsk;
1575 	list->qlen++;
1576 }
1577 
1578 static inline void __skb_queue_splice(const struct sk_buff_head *list,
1579 				      struct sk_buff *prev,
1580 				      struct sk_buff *next)
1581 {
1582 	struct sk_buff *first = list->next;
1583 	struct sk_buff *last = list->prev;
1584 
1585 	first->prev = prev;
1586 	prev->next = first;
1587 
1588 	last->next = next;
1589 	next->prev = last;
1590 }
1591 
1592 /**
1593  *	skb_queue_splice - join two skb lists, this is designed for stacks
1594  *	@list: the new list to add
1595  *	@head: the place to add it in the first list
1596  */
1597 static inline void skb_queue_splice(const struct sk_buff_head *list,
1598 				    struct sk_buff_head *head)
1599 {
1600 	if (!skb_queue_empty(list)) {
1601 		__skb_queue_splice(list, (struct sk_buff *) head, head->next);
1602 		head->qlen += list->qlen;
1603 	}
1604 }
1605 
1606 /**
1607  *	skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1608  *	@list: the new list to add
1609  *	@head: the place to add it in the first list
1610  *
1611  *	The list at @list is reinitialised
1612  */
1613 static inline void skb_queue_splice_init(struct sk_buff_head *list,
1614 					 struct sk_buff_head *head)
1615 {
1616 	if (!skb_queue_empty(list)) {
1617 		__skb_queue_splice(list, (struct sk_buff *) head, head->next);
1618 		head->qlen += list->qlen;
1619 		__skb_queue_head_init(list);
1620 	}
1621 }
1622 
1623 /**
1624  *	skb_queue_splice_tail - join two skb lists, each list being a queue
1625  *	@list: the new list to add
1626  *	@head: the place to add it in the first list
1627  */
1628 static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1629 					 struct sk_buff_head *head)
1630 {
1631 	if (!skb_queue_empty(list)) {
1632 		__skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1633 		head->qlen += list->qlen;
1634 	}
1635 }
1636 
1637 /**
1638  *	skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1639  *	@list: the new list to add
1640  *	@head: the place to add it in the first list
1641  *
1642  *	Each of the lists is a queue.
1643  *	The list at @list is reinitialised
1644  */
1645 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1646 					      struct sk_buff_head *head)
1647 {
1648 	if (!skb_queue_empty(list)) {
1649 		__skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1650 		head->qlen += list->qlen;
1651 		__skb_queue_head_init(list);
1652 	}
1653 }
1654 
1655 /**
1656  *	__skb_queue_after - queue a buffer at the list head
1657  *	@list: list to use
1658  *	@prev: place after this buffer
1659  *	@newsk: buffer to queue
1660  *
1661  *	Queue a buffer int the middle of a list. This function takes no locks
1662  *	and you must therefore hold required locks before calling it.
1663  *
1664  *	A buffer cannot be placed on two lists at the same time.
1665  */
1666 static inline void __skb_queue_after(struct sk_buff_head *list,
1667 				     struct sk_buff *prev,
1668 				     struct sk_buff *newsk)
1669 {
1670 	__skb_insert(newsk, prev, prev->next, list);
1671 }
1672 
1673 void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1674 		struct sk_buff_head *list);
1675 
1676 static inline void __skb_queue_before(struct sk_buff_head *list,
1677 				      struct sk_buff *next,
1678 				      struct sk_buff *newsk)
1679 {
1680 	__skb_insert(newsk, next->prev, next, list);
1681 }
1682 
1683 /**
1684  *	__skb_queue_head - queue a buffer at the list head
1685  *	@list: list to use
1686  *	@newsk: buffer to queue
1687  *
1688  *	Queue a buffer at the start of a list. This function takes no locks
1689  *	and you must therefore hold required locks before calling it.
1690  *
1691  *	A buffer cannot be placed on two lists at the same time.
1692  */
1693 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1694 static inline void __skb_queue_head(struct sk_buff_head *list,
1695 				    struct sk_buff *newsk)
1696 {
1697 	__skb_queue_after(list, (struct sk_buff *)list, newsk);
1698 }
1699 
1700 /**
1701  *	__skb_queue_tail - queue a buffer at the list tail
1702  *	@list: list to use
1703  *	@newsk: buffer to queue
1704  *
1705  *	Queue a buffer at the end of a list. This function takes no locks
1706  *	and you must therefore hold required locks before calling it.
1707  *
1708  *	A buffer cannot be placed on two lists at the same time.
1709  */
1710 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1711 static inline void __skb_queue_tail(struct sk_buff_head *list,
1712 				   struct sk_buff *newsk)
1713 {
1714 	__skb_queue_before(list, (struct sk_buff *)list, newsk);
1715 }
1716 
1717 /*
1718  * remove sk_buff from list. _Must_ be called atomically, and with
1719  * the list known..
1720  */
1721 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1722 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1723 {
1724 	struct sk_buff *next, *prev;
1725 
1726 	list->qlen--;
1727 	next	   = skb->next;
1728 	prev	   = skb->prev;
1729 	skb->next  = skb->prev = NULL;
1730 	next->prev = prev;
1731 	prev->next = next;
1732 }
1733 
1734 /**
1735  *	__skb_dequeue - remove from the head of the queue
1736  *	@list: list to dequeue from
1737  *
1738  *	Remove the head of the list. This function does not take any locks
1739  *	so must be used with appropriate locks held only. The head item is
1740  *	returned or %NULL if the list is empty.
1741  */
1742 struct sk_buff *skb_dequeue(struct sk_buff_head *list);
1743 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1744 {
1745 	struct sk_buff *skb = skb_peek(list);
1746 	if (skb)
1747 		__skb_unlink(skb, list);
1748 	return skb;
1749 }
1750 
1751 /**
1752  *	__skb_dequeue_tail - remove from the tail of the queue
1753  *	@list: list to dequeue from
1754  *
1755  *	Remove the tail of the list. This function does not take any locks
1756  *	so must be used with appropriate locks held only. The tail item is
1757  *	returned or %NULL if the list is empty.
1758  */
1759 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1760 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1761 {
1762 	struct sk_buff *skb = skb_peek_tail(list);
1763 	if (skb)
1764 		__skb_unlink(skb, list);
1765 	return skb;
1766 }
1767 
1768 
1769 static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1770 {
1771 	return skb->data_len;
1772 }
1773 
1774 static inline unsigned int skb_headlen(const struct sk_buff *skb)
1775 {
1776 	return skb->len - skb->data_len;
1777 }
1778 
1779 static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1780 {
1781 	unsigned int i, len = 0;
1782 
1783 	for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
1784 		len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1785 	return len + skb_headlen(skb);
1786 }
1787 
1788 /**
1789  * __skb_fill_page_desc - initialise a paged fragment in an skb
1790  * @skb: buffer containing fragment to be initialised
1791  * @i: paged fragment index to initialise
1792  * @page: the page to use for this fragment
1793  * @off: the offset to the data with @page
1794  * @size: the length of the data
1795  *
1796  * Initialises the @i'th fragment of @skb to point to &size bytes at
1797  * offset @off within @page.
1798  *
1799  * Does not take any additional reference on the fragment.
1800  */
1801 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1802 					struct page *page, int off, int size)
1803 {
1804 	skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1805 
1806 	/*
1807 	 * Propagate page pfmemalloc to the skb if we can. The problem is
1808 	 * that not all callers have unique ownership of the page but rely
1809 	 * on page_is_pfmemalloc doing the right thing(tm).
1810 	 */
1811 	frag->page.p		  = page;
1812 	frag->page_offset	  = off;
1813 	skb_frag_size_set(frag, size);
1814 
1815 	page = compound_head(page);
1816 	if (page_is_pfmemalloc(page))
1817 		skb->pfmemalloc	= true;
1818 }
1819 
1820 /**
1821  * skb_fill_page_desc - initialise a paged fragment in an skb
1822  * @skb: buffer containing fragment to be initialised
1823  * @i: paged fragment index to initialise
1824  * @page: the page to use for this fragment
1825  * @off: the offset to the data with @page
1826  * @size: the length of the data
1827  *
1828  * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
1829  * @skb to point to @size bytes at offset @off within @page. In
1830  * addition updates @skb such that @i is the last fragment.
1831  *
1832  * Does not take any additional reference on the fragment.
1833  */
1834 static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1835 				      struct page *page, int off, int size)
1836 {
1837 	__skb_fill_page_desc(skb, i, page, off, size);
1838 	skb_shinfo(skb)->nr_frags = i + 1;
1839 }
1840 
1841 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1842 		     int size, unsigned int truesize);
1843 
1844 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1845 			  unsigned int truesize);
1846 
1847 #define SKB_PAGE_ASSERT(skb) 	BUG_ON(skb_shinfo(skb)->nr_frags)
1848 #define SKB_FRAG_ASSERT(skb) 	BUG_ON(skb_has_frag_list(skb))
1849 #define SKB_LINEAR_ASSERT(skb)  BUG_ON(skb_is_nonlinear(skb))
1850 
1851 #ifdef NET_SKBUFF_DATA_USES_OFFSET
1852 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1853 {
1854 	return skb->head + skb->tail;
1855 }
1856 
1857 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1858 {
1859 	skb->tail = skb->data - skb->head;
1860 }
1861 
1862 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1863 {
1864 	skb_reset_tail_pointer(skb);
1865 	skb->tail += offset;
1866 }
1867 
1868 #else /* NET_SKBUFF_DATA_USES_OFFSET */
1869 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1870 {
1871 	return skb->tail;
1872 }
1873 
1874 static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1875 {
1876 	skb->tail = skb->data;
1877 }
1878 
1879 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1880 {
1881 	skb->tail = skb->data + offset;
1882 }
1883 
1884 #endif /* NET_SKBUFF_DATA_USES_OFFSET */
1885 
1886 /*
1887  *	Add data to an sk_buff
1888  */
1889 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1890 void *skb_put(struct sk_buff *skb, unsigned int len);
1891 static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
1892 {
1893 	void *tmp = skb_tail_pointer(skb);
1894 	SKB_LINEAR_ASSERT(skb);
1895 	skb->tail += len;
1896 	skb->len  += len;
1897 	return tmp;
1898 }
1899 
1900 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
1901 {
1902 	void *tmp = __skb_put(skb, len);
1903 
1904 	memset(tmp, 0, len);
1905 	return tmp;
1906 }
1907 
1908 static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
1909 				   unsigned int len)
1910 {
1911 	void *tmp = __skb_put(skb, len);
1912 
1913 	memcpy(tmp, data, len);
1914 	return tmp;
1915 }
1916 
1917 static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
1918 {
1919 	*(u8 *)__skb_put(skb, 1) = val;
1920 }
1921 
1922 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
1923 {
1924 	void *tmp = skb_put(skb, len);
1925 
1926 	memset(tmp, 0, len);
1927 
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 
1938 	return tmp;
1939 }
1940 
1941 static inline void skb_put_u8(struct sk_buff *skb, u8 val)
1942 {
1943 	*(u8 *)skb_put(skb, 1) = val;
1944 }
1945 
1946 void *skb_push(struct sk_buff *skb, unsigned int len);
1947 static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
1948 {
1949 	skb->data -= len;
1950 	skb->len  += len;
1951 	return skb->data;
1952 }
1953 
1954 void *skb_pull(struct sk_buff *skb, unsigned int len);
1955 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
1956 {
1957 	skb->len -= len;
1958 	BUG_ON(skb->len < skb->data_len);
1959 	return skb->data += len;
1960 }
1961 
1962 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1963 {
1964 	return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1965 }
1966 
1967 void *__pskb_pull_tail(struct sk_buff *skb, int delta);
1968 
1969 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
1970 {
1971 	if (len > skb_headlen(skb) &&
1972 	    !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1973 		return NULL;
1974 	skb->len -= len;
1975 	return skb->data += len;
1976 }
1977 
1978 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
1979 {
1980 	return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1981 }
1982 
1983 static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1984 {
1985 	if (likely(len <= skb_headlen(skb)))
1986 		return 1;
1987 	if (unlikely(len > skb->len))
1988 		return 0;
1989 	return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1990 }
1991 
1992 void skb_condense(struct sk_buff *skb);
1993 
1994 /**
1995  *	skb_headroom - bytes at buffer head
1996  *	@skb: buffer to check
1997  *
1998  *	Return the number of bytes of free space at the head of an &sk_buff.
1999  */
2000 static inline unsigned int skb_headroom(const struct sk_buff *skb)
2001 {
2002 	return skb->data - skb->head;
2003 }
2004 
2005 /**
2006  *	skb_tailroom - bytes at buffer end
2007  *	@skb: buffer to check
2008  *
2009  *	Return the number of bytes of free space at the tail of an sk_buff
2010  */
2011 static inline int skb_tailroom(const struct sk_buff *skb)
2012 {
2013 	return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2014 }
2015 
2016 /**
2017  *	skb_availroom - bytes at buffer end
2018  *	@skb: buffer to check
2019  *
2020  *	Return the number of bytes of free space at the tail of an sk_buff
2021  *	allocated by sk_stream_alloc()
2022  */
2023 static inline int skb_availroom(const struct sk_buff *skb)
2024 {
2025 	if (skb_is_nonlinear(skb))
2026 		return 0;
2027 
2028 	return skb->end - skb->tail - skb->reserved_tailroom;
2029 }
2030 
2031 /**
2032  *	skb_reserve - adjust headroom
2033  *	@skb: buffer to alter
2034  *	@len: bytes to move
2035  *
2036  *	Increase the headroom of an empty &sk_buff by reducing the tail
2037  *	room. This is only allowed for an empty buffer.
2038  */
2039 static inline void skb_reserve(struct sk_buff *skb, int len)
2040 {
2041 	skb->data += len;
2042 	skb->tail += len;
2043 }
2044 
2045 /**
2046  *	skb_tailroom_reserve - adjust reserved_tailroom
2047  *	@skb: buffer to alter
2048  *	@mtu: maximum amount of headlen permitted
2049  *	@needed_tailroom: minimum amount of reserved_tailroom
2050  *
2051  *	Set reserved_tailroom so that headlen can be as large as possible but
2052  *	not larger than mtu and tailroom cannot be smaller than
2053  *	needed_tailroom.
2054  *	The required headroom should already have been reserved before using
2055  *	this function.
2056  */
2057 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2058 					unsigned int needed_tailroom)
2059 {
2060 	SKB_LINEAR_ASSERT(skb);
2061 	if (mtu < skb_tailroom(skb) - needed_tailroom)
2062 		/* use at most mtu */
2063 		skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2064 	else
2065 		/* use up to all available space */
2066 		skb->reserved_tailroom = needed_tailroom;
2067 }
2068 
2069 #define ENCAP_TYPE_ETHER	0
2070 #define ENCAP_TYPE_IPPROTO	1
2071 
2072 static inline void skb_set_inner_protocol(struct sk_buff *skb,
2073 					  __be16 protocol)
2074 {
2075 	skb->inner_protocol = protocol;
2076 	skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2077 }
2078 
2079 static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2080 					 __u8 ipproto)
2081 {
2082 	skb->inner_ipproto = ipproto;
2083 	skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2084 }
2085 
2086 static inline void skb_reset_inner_headers(struct sk_buff *skb)
2087 {
2088 	skb->inner_mac_header = skb->mac_header;
2089 	skb->inner_network_header = skb->network_header;
2090 	skb->inner_transport_header = skb->transport_header;
2091 }
2092 
2093 static inline void skb_reset_mac_len(struct sk_buff *skb)
2094 {
2095 	skb->mac_len = skb->network_header - skb->mac_header;
2096 }
2097 
2098 static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2099 							*skb)
2100 {
2101 	return skb->head + skb->inner_transport_header;
2102 }
2103 
2104 static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2105 {
2106 	return skb_inner_transport_header(skb) - skb->data;
2107 }
2108 
2109 static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2110 {
2111 	skb->inner_transport_header = skb->data - skb->head;
2112 }
2113 
2114 static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2115 						   const int offset)
2116 {
2117 	skb_reset_inner_transport_header(skb);
2118 	skb->inner_transport_header += offset;
2119 }
2120 
2121 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2122 {
2123 	return skb->head + skb->inner_network_header;
2124 }
2125 
2126 static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2127 {
2128 	skb->inner_network_header = skb->data - skb->head;
2129 }
2130 
2131 static inline void skb_set_inner_network_header(struct sk_buff *skb,
2132 						const int offset)
2133 {
2134 	skb_reset_inner_network_header(skb);
2135 	skb->inner_network_header += offset;
2136 }
2137 
2138 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2139 {
2140 	return skb->head + skb->inner_mac_header;
2141 }
2142 
2143 static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2144 {
2145 	skb->inner_mac_header = skb->data - skb->head;
2146 }
2147 
2148 static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2149 					    const int offset)
2150 {
2151 	skb_reset_inner_mac_header(skb);
2152 	skb->inner_mac_header += offset;
2153 }
2154 static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2155 {
2156 	return skb->transport_header != (typeof(skb->transport_header))~0U;
2157 }
2158 
2159 static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2160 {
2161 	return skb->head + skb->transport_header;
2162 }
2163 
2164 static inline void skb_reset_transport_header(struct sk_buff *skb)
2165 {
2166 	skb->transport_header = skb->data - skb->head;
2167 }
2168 
2169 static inline void skb_set_transport_header(struct sk_buff *skb,
2170 					    const int offset)
2171 {
2172 	skb_reset_transport_header(skb);
2173 	skb->transport_header += offset;
2174 }
2175 
2176 static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2177 {
2178 	return skb->head + skb->network_header;
2179 }
2180 
2181 static inline void skb_reset_network_header(struct sk_buff *skb)
2182 {
2183 	skb->network_header = skb->data - skb->head;
2184 }
2185 
2186 static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2187 {
2188 	skb_reset_network_header(skb);
2189 	skb->network_header += offset;
2190 }
2191 
2192 static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2193 {
2194 	return skb->head + skb->mac_header;
2195 }
2196 
2197 static inline int skb_mac_offset(const struct sk_buff *skb)
2198 {
2199 	return skb_mac_header(skb) - skb->data;
2200 }
2201 
2202 static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2203 {
2204 	return skb->network_header - skb->mac_header;
2205 }
2206 
2207 static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2208 {
2209 	return skb->mac_header != (typeof(skb->mac_header))~0U;
2210 }
2211 
2212 static inline void skb_reset_mac_header(struct sk_buff *skb)
2213 {
2214 	skb->mac_header = skb->data - skb->head;
2215 }
2216 
2217 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2218 {
2219 	skb_reset_mac_header(skb);
2220 	skb->mac_header += offset;
2221 }
2222 
2223 static inline void skb_pop_mac_header(struct sk_buff *skb)
2224 {
2225 	skb->mac_header = skb->network_header;
2226 }
2227 
2228 static inline void skb_probe_transport_header(struct sk_buff *skb,
2229 					      const int offset_hint)
2230 {
2231 	struct flow_keys keys;
2232 
2233 	if (skb_transport_header_was_set(skb))
2234 		return;
2235 	else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
2236 		skb_set_transport_header(skb, keys.control.thoff);
2237 	else
2238 		skb_set_transport_header(skb, offset_hint);
2239 }
2240 
2241 static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2242 {
2243 	if (skb_mac_header_was_set(skb)) {
2244 		const unsigned char *old_mac = skb_mac_header(skb);
2245 
2246 		skb_set_mac_header(skb, -skb->mac_len);
2247 		memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2248 	}
2249 }
2250 
2251 static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2252 {
2253 	return skb->csum_start - skb_headroom(skb);
2254 }
2255 
2256 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2257 {
2258 	return skb->head + skb->csum_start;
2259 }
2260 
2261 static inline int skb_transport_offset(const struct sk_buff *skb)
2262 {
2263 	return skb_transport_header(skb) - skb->data;
2264 }
2265 
2266 static inline u32 skb_network_header_len(const struct sk_buff *skb)
2267 {
2268 	return skb->transport_header - skb->network_header;
2269 }
2270 
2271 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2272 {
2273 	return skb->inner_transport_header - skb->inner_network_header;
2274 }
2275 
2276 static inline int skb_network_offset(const struct sk_buff *skb)
2277 {
2278 	return skb_network_header(skb) - skb->data;
2279 }
2280 
2281 static inline int skb_inner_network_offset(const struct sk_buff *skb)
2282 {
2283 	return skb_inner_network_header(skb) - skb->data;
2284 }
2285 
2286 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2287 {
2288 	return pskb_may_pull(skb, skb_network_offset(skb) + len);
2289 }
2290 
2291 /*
2292  * CPUs often take a performance hit when accessing unaligned memory
2293  * locations. The actual performance hit varies, it can be small if the
2294  * hardware handles it or large if we have to take an exception and fix it
2295  * in software.
2296  *
2297  * Since an ethernet header is 14 bytes network drivers often end up with
2298  * the IP header at an unaligned offset. The IP header can be aligned by
2299  * shifting the start of the packet by 2 bytes. Drivers should do this
2300  * with:
2301  *
2302  * skb_reserve(skb, NET_IP_ALIGN);
2303  *
2304  * The downside to this alignment of the IP header is that the DMA is now
2305  * unaligned. On some architectures the cost of an unaligned DMA is high
2306  * and this cost outweighs the gains made by aligning the IP header.
2307  *
2308  * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2309  * to be overridden.
2310  */
2311 #ifndef NET_IP_ALIGN
2312 #define NET_IP_ALIGN	2
2313 #endif
2314 
2315 /*
2316  * The networking layer reserves some headroom in skb data (via
2317  * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2318  * the header has to grow. In the default case, if the header has to grow
2319  * 32 bytes or less we avoid the reallocation.
2320  *
2321  * Unfortunately this headroom changes the DMA alignment of the resulting
2322  * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2323  * on some architectures. An architecture can override this value,
2324  * perhaps setting it to a cacheline in size (since that will maintain
2325  * cacheline alignment of the DMA). It must be a power of 2.
2326  *
2327  * Various parts of the networking layer expect at least 32 bytes of
2328  * headroom, you should not reduce this.
2329  *
2330  * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2331  * to reduce average number of cache lines per packet.
2332  * get_rps_cpus() for example only access one 64 bytes aligned block :
2333  * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2334  */
2335 #ifndef NET_SKB_PAD
2336 #define NET_SKB_PAD	max(32, L1_CACHE_BYTES)
2337 #endif
2338 
2339 int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2340 
2341 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2342 {
2343 	if (unlikely(skb_is_nonlinear(skb))) {
2344 		WARN_ON(1);
2345 		return;
2346 	}
2347 	skb->len = len;
2348 	skb_set_tail_pointer(skb, len);
2349 }
2350 
2351 static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2352 {
2353 	__skb_set_length(skb, len);
2354 }
2355 
2356 void skb_trim(struct sk_buff *skb, unsigned int len);
2357 
2358 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2359 {
2360 	if (skb->data_len)
2361 		return ___pskb_trim(skb, len);
2362 	__skb_trim(skb, len);
2363 	return 0;
2364 }
2365 
2366 static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2367 {
2368 	return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2369 }
2370 
2371 /**
2372  *	pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2373  *	@skb: buffer to alter
2374  *	@len: new length
2375  *
2376  *	This is identical to pskb_trim except that the caller knows that
2377  *	the skb is not cloned so we should never get an error due to out-
2378  *	of-memory.
2379  */
2380 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2381 {
2382 	int err = pskb_trim(skb, len);
2383 	BUG_ON(err);
2384 }
2385 
2386 static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2387 {
2388 	unsigned int diff = len - skb->len;
2389 
2390 	if (skb_tailroom(skb) < diff) {
2391 		int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2392 					   GFP_ATOMIC);
2393 		if (ret)
2394 			return ret;
2395 	}
2396 	__skb_set_length(skb, len);
2397 	return 0;
2398 }
2399 
2400 /**
2401  *	skb_orphan - orphan a buffer
2402  *	@skb: buffer to orphan
2403  *
2404  *	If a buffer currently has an owner then we call the owner's
2405  *	destructor function and make the @skb unowned. The buffer continues
2406  *	to exist but is no longer charged to its former owner.
2407  */
2408 static inline void skb_orphan(struct sk_buff *skb)
2409 {
2410 	if (skb->destructor) {
2411 		skb->destructor(skb);
2412 		skb->destructor = NULL;
2413 		skb->sk		= NULL;
2414 	} else {
2415 		BUG_ON(skb->sk);
2416 	}
2417 }
2418 
2419 /**
2420  *	skb_orphan_frags - orphan the frags contained in a buffer
2421  *	@skb: buffer to orphan frags from
2422  *	@gfp_mask: allocation mask for replacement pages
2423  *
2424  *	For each frag in the SKB which needs a destructor (i.e. has an
2425  *	owner) create a copy of that frag and release the original
2426  *	page by calling the destructor.
2427  */
2428 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2429 {
2430 	if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2431 		return 0;
2432 	return skb_copy_ubufs(skb, gfp_mask);
2433 }
2434 
2435 /**
2436  *	__skb_queue_purge - empty a list
2437  *	@list: list to empty
2438  *
2439  *	Delete all buffers on an &sk_buff list. Each buffer is removed from
2440  *	the list and one reference dropped. This function does not take the
2441  *	list lock and the caller must hold the relevant locks to use it.
2442  */
2443 void skb_queue_purge(struct sk_buff_head *list);
2444 static inline void __skb_queue_purge(struct sk_buff_head *list)
2445 {
2446 	struct sk_buff *skb;
2447 	while ((skb = __skb_dequeue(list)) != NULL)
2448 		kfree_skb(skb);
2449 }
2450 
2451 void skb_rbtree_purge(struct rb_root *root);
2452 
2453 void *netdev_alloc_frag(unsigned int fragsz);
2454 
2455 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2456 				   gfp_t gfp_mask);
2457 
2458 /**
2459  *	netdev_alloc_skb - allocate an skbuff for rx on a specific device
2460  *	@dev: network device to receive on
2461  *	@length: length to allocate
2462  *
2463  *	Allocate a new &sk_buff and assign it a usage count of one. The
2464  *	buffer has unspecified headroom built in. Users should allocate
2465  *	the headroom they think they need without accounting for the
2466  *	built in space. The built in space is used for optimisations.
2467  *
2468  *	%NULL is returned if there is no free memory. Although this function
2469  *	allocates memory it can be called from an interrupt.
2470  */
2471 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2472 					       unsigned int length)
2473 {
2474 	return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2475 }
2476 
2477 /* legacy helper around __netdev_alloc_skb() */
2478 static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2479 					      gfp_t gfp_mask)
2480 {
2481 	return __netdev_alloc_skb(NULL, length, gfp_mask);
2482 }
2483 
2484 /* legacy helper around netdev_alloc_skb() */
2485 static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2486 {
2487 	return netdev_alloc_skb(NULL, length);
2488 }
2489 
2490 
2491 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2492 		unsigned int length, gfp_t gfp)
2493 {
2494 	struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2495 
2496 	if (NET_IP_ALIGN && skb)
2497 		skb_reserve(skb, NET_IP_ALIGN);
2498 	return skb;
2499 }
2500 
2501 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2502 		unsigned int length)
2503 {
2504 	return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2505 }
2506 
2507 static inline void skb_free_frag(void *addr)
2508 {
2509 	page_frag_free(addr);
2510 }
2511 
2512 void *napi_alloc_frag(unsigned int fragsz);
2513 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2514 				 unsigned int length, gfp_t gfp_mask);
2515 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2516 					     unsigned int length)
2517 {
2518 	return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2519 }
2520 void napi_consume_skb(struct sk_buff *skb, int budget);
2521 
2522 void __kfree_skb_flush(void);
2523 void __kfree_skb_defer(struct sk_buff *skb);
2524 
2525 /**
2526  * __dev_alloc_pages - allocate page for network Rx
2527  * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2528  * @order: size of the allocation
2529  *
2530  * Allocate a new page.
2531  *
2532  * %NULL is returned if there is no free memory.
2533 */
2534 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2535 					     unsigned int order)
2536 {
2537 	/* This piece of code contains several assumptions.
2538 	 * 1.  This is for device Rx, therefor a cold page is preferred.
2539 	 * 2.  The expectation is the user wants a compound page.
2540 	 * 3.  If requesting a order 0 page it will not be compound
2541 	 *     due to the check to see if order has a value in prep_new_page
2542 	 * 4.  __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2543 	 *     code in gfp_to_alloc_flags that should be enforcing this.
2544 	 */
2545 	gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2546 
2547 	return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2548 }
2549 
2550 static inline struct page *dev_alloc_pages(unsigned int order)
2551 {
2552 	return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2553 }
2554 
2555 /**
2556  * __dev_alloc_page - allocate a page for network Rx
2557  * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2558  *
2559  * Allocate a new page.
2560  *
2561  * %NULL is returned if there is no free memory.
2562  */
2563 static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2564 {
2565 	return __dev_alloc_pages(gfp_mask, 0);
2566 }
2567 
2568 static inline struct page *dev_alloc_page(void)
2569 {
2570 	return dev_alloc_pages(0);
2571 }
2572 
2573 /**
2574  *	skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2575  *	@page: The page that was allocated from skb_alloc_page
2576  *	@skb: The skb that may need pfmemalloc set
2577  */
2578 static inline void skb_propagate_pfmemalloc(struct page *page,
2579 					     struct sk_buff *skb)
2580 {
2581 	if (page_is_pfmemalloc(page))
2582 		skb->pfmemalloc = true;
2583 }
2584 
2585 /**
2586  * skb_frag_page - retrieve the page referred to by a paged fragment
2587  * @frag: the paged fragment
2588  *
2589  * Returns the &struct page associated with @frag.
2590  */
2591 static inline struct page *skb_frag_page(const skb_frag_t *frag)
2592 {
2593 	return frag->page.p;
2594 }
2595 
2596 /**
2597  * __skb_frag_ref - take an addition reference on a paged fragment.
2598  * @frag: the paged fragment
2599  *
2600  * Takes an additional reference on the paged fragment @frag.
2601  */
2602 static inline void __skb_frag_ref(skb_frag_t *frag)
2603 {
2604 	get_page(skb_frag_page(frag));
2605 }
2606 
2607 /**
2608  * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2609  * @skb: the buffer
2610  * @f: the fragment offset.
2611  *
2612  * Takes an additional reference on the @f'th paged fragment of @skb.
2613  */
2614 static inline void skb_frag_ref(struct sk_buff *skb, int f)
2615 {
2616 	__skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2617 }
2618 
2619 /**
2620  * __skb_frag_unref - release a reference on a paged fragment.
2621  * @frag: the paged fragment
2622  *
2623  * Releases a reference on the paged fragment @frag.
2624  */
2625 static inline void __skb_frag_unref(skb_frag_t *frag)
2626 {
2627 	put_page(skb_frag_page(frag));
2628 }
2629 
2630 /**
2631  * skb_frag_unref - release a reference on a paged fragment of an skb.
2632  * @skb: the buffer
2633  * @f: the fragment offset
2634  *
2635  * Releases a reference on the @f'th paged fragment of @skb.
2636  */
2637 static inline void skb_frag_unref(struct sk_buff *skb, int f)
2638 {
2639 	__skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2640 }
2641 
2642 /**
2643  * skb_frag_address - gets the address of the data contained in a paged fragment
2644  * @frag: the paged fragment buffer
2645  *
2646  * Returns the address of the data within @frag. The page must already
2647  * be mapped.
2648  */
2649 static inline void *skb_frag_address(const skb_frag_t *frag)
2650 {
2651 	return page_address(skb_frag_page(frag)) + frag->page_offset;
2652 }
2653 
2654 /**
2655  * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2656  * @frag: the paged fragment buffer
2657  *
2658  * Returns the address of the data within @frag. Checks that the page
2659  * is mapped and returns %NULL otherwise.
2660  */
2661 static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2662 {
2663 	void *ptr = page_address(skb_frag_page(frag));
2664 	if (unlikely(!ptr))
2665 		return NULL;
2666 
2667 	return ptr + frag->page_offset;
2668 }
2669 
2670 /**
2671  * __skb_frag_set_page - sets the page contained in a paged fragment
2672  * @frag: the paged fragment
2673  * @page: the page to set
2674  *
2675  * Sets the fragment @frag to contain @page.
2676  */
2677 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2678 {
2679 	frag->page.p = page;
2680 }
2681 
2682 /**
2683  * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2684  * @skb: the buffer
2685  * @f: the fragment offset
2686  * @page: the page to set
2687  *
2688  * Sets the @f'th fragment of @skb to contain @page.
2689  */
2690 static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2691 				     struct page *page)
2692 {
2693 	__skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2694 }
2695 
2696 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2697 
2698 /**
2699  * skb_frag_dma_map - maps a paged fragment via the DMA API
2700  * @dev: the device to map the fragment to
2701  * @frag: the paged fragment to map
2702  * @offset: the offset within the fragment (starting at the
2703  *          fragment's own offset)
2704  * @size: the number of bytes to map
2705  * @dir: the direction of the mapping (``PCI_DMA_*``)
2706  *
2707  * Maps the page associated with @frag to @device.
2708  */
2709 static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2710 					  const skb_frag_t *frag,
2711 					  size_t offset, size_t size,
2712 					  enum dma_data_direction dir)
2713 {
2714 	return dma_map_page(dev, skb_frag_page(frag),
2715 			    frag->page_offset + offset, size, dir);
2716 }
2717 
2718 static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2719 					gfp_t gfp_mask)
2720 {
2721 	return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2722 }
2723 
2724 
2725 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2726 						  gfp_t gfp_mask)
2727 {
2728 	return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2729 }
2730 
2731 
2732 /**
2733  *	skb_clone_writable - is the header of a clone writable
2734  *	@skb: buffer to check
2735  *	@len: length up to which to write
2736  *
2737  *	Returns true if modifying the header part of the cloned buffer
2738  *	does not requires the data to be copied.
2739  */
2740 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
2741 {
2742 	return !skb_header_cloned(skb) &&
2743 	       skb_headroom(skb) + len <= skb->hdr_len;
2744 }
2745 
2746 static inline int skb_try_make_writable(struct sk_buff *skb,
2747 					unsigned int write_len)
2748 {
2749 	return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2750 	       pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2751 }
2752 
2753 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2754 			    int cloned)
2755 {
2756 	int delta = 0;
2757 
2758 	if (headroom > skb_headroom(skb))
2759 		delta = headroom - skb_headroom(skb);
2760 
2761 	if (delta || cloned)
2762 		return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2763 					GFP_ATOMIC);
2764 	return 0;
2765 }
2766 
2767 /**
2768  *	skb_cow - copy header of skb when it is required
2769  *	@skb: buffer to cow
2770  *	@headroom: needed headroom
2771  *
2772  *	If the skb passed lacks sufficient headroom or its data part
2773  *	is shared, data is reallocated. If reallocation fails, an error
2774  *	is returned and original skb is not changed.
2775  *
2776  *	The result is skb with writable area skb->head...skb->tail
2777  *	and at least @headroom of space at head.
2778  */
2779 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2780 {
2781 	return __skb_cow(skb, headroom, skb_cloned(skb));
2782 }
2783 
2784 /**
2785  *	skb_cow_head - skb_cow but only making the head writable
2786  *	@skb: buffer to cow
2787  *	@headroom: needed headroom
2788  *
2789  *	This function is identical to skb_cow except that we replace the
2790  *	skb_cloned check by skb_header_cloned.  It should be used when
2791  *	you only need to push on some header and do not need to modify
2792  *	the data.
2793  */
2794 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2795 {
2796 	return __skb_cow(skb, headroom, skb_header_cloned(skb));
2797 }
2798 
2799 /**
2800  *	skb_padto	- pad an skbuff up to a minimal size
2801  *	@skb: buffer to pad
2802  *	@len: minimal length
2803  *
2804  *	Pads up a buffer to ensure the trailing bytes exist and are
2805  *	blanked. If the buffer already contains sufficient data it
2806  *	is untouched. Otherwise it is extended. Returns zero on
2807  *	success. The skb is freed on error.
2808  */
2809 static inline int skb_padto(struct sk_buff *skb, unsigned int len)
2810 {
2811 	unsigned int size = skb->len;
2812 	if (likely(size >= len))
2813 		return 0;
2814 	return skb_pad(skb, len - size);
2815 }
2816 
2817 /**
2818  *	skb_put_padto - increase size and pad an skbuff up to a minimal size
2819  *	@skb: buffer to pad
2820  *	@len: minimal length
2821  *
2822  *	Pads up a buffer to ensure the trailing bytes exist and are
2823  *	blanked. If the buffer already contains sufficient data it
2824  *	is untouched. Otherwise it is extended. Returns zero on
2825  *	success. The skb is freed on error.
2826  */
2827 static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2828 {
2829 	unsigned int size = skb->len;
2830 
2831 	if (unlikely(size < len)) {
2832 		len -= size;
2833 		if (skb_pad(skb, len))
2834 			return -ENOMEM;
2835 		__skb_put(skb, len);
2836 	}
2837 	return 0;
2838 }
2839 
2840 static inline int skb_add_data(struct sk_buff *skb,
2841 			       struct iov_iter *from, int copy)
2842 {
2843 	const int off = skb->len;
2844 
2845 	if (skb->ip_summed == CHECKSUM_NONE) {
2846 		__wsum csum = 0;
2847 		if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2848 					         &csum, from)) {
2849 			skb->csum = csum_block_add(skb->csum, csum, off);
2850 			return 0;
2851 		}
2852 	} else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
2853 		return 0;
2854 
2855 	__skb_trim(skb, off);
2856 	return -EFAULT;
2857 }
2858 
2859 static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2860 				    const struct page *page, int off)
2861 {
2862 	if (i) {
2863 		const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
2864 
2865 		return page == skb_frag_page(frag) &&
2866 		       off == frag->page_offset + skb_frag_size(frag);
2867 	}
2868 	return false;
2869 }
2870 
2871 static inline int __skb_linearize(struct sk_buff *skb)
2872 {
2873 	return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2874 }
2875 
2876 /**
2877  *	skb_linearize - convert paged skb to linear one
2878  *	@skb: buffer to linarize
2879  *
2880  *	If there is no free memory -ENOMEM is returned, otherwise zero
2881  *	is returned and the old skb data released.
2882  */
2883 static inline int skb_linearize(struct sk_buff *skb)
2884 {
2885 	return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2886 }
2887 
2888 /**
2889  * skb_has_shared_frag - can any frag be overwritten
2890  * @skb: buffer to test
2891  *
2892  * Return true if the skb has at least one frag that might be modified
2893  * by an external entity (as in vmsplice()/sendfile())
2894  */
2895 static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2896 {
2897 	return skb_is_nonlinear(skb) &&
2898 	       skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
2899 }
2900 
2901 /**
2902  *	skb_linearize_cow - make sure skb is linear and writable
2903  *	@skb: buffer to process
2904  *
2905  *	If there is no free memory -ENOMEM is returned, otherwise zero
2906  *	is returned and the old skb data released.
2907  */
2908 static inline int skb_linearize_cow(struct sk_buff *skb)
2909 {
2910 	return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2911 	       __skb_linearize(skb) : 0;
2912 }
2913 
2914 static __always_inline void
2915 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2916 		     unsigned int off)
2917 {
2918 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2919 		skb->csum = csum_block_sub(skb->csum,
2920 					   csum_partial(start, len, 0), off);
2921 	else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2922 		 skb_checksum_start_offset(skb) < 0)
2923 		skb->ip_summed = CHECKSUM_NONE;
2924 }
2925 
2926 /**
2927  *	skb_postpull_rcsum - update checksum for received skb after pull
2928  *	@skb: buffer to update
2929  *	@start: start of data before pull
2930  *	@len: length of data pulled
2931  *
2932  *	After doing a pull on a received packet, you need to call this to
2933  *	update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2934  *	CHECKSUM_NONE so that it can be recomputed from scratch.
2935  */
2936 static inline void skb_postpull_rcsum(struct sk_buff *skb,
2937 				      const void *start, unsigned int len)
2938 {
2939 	__skb_postpull_rcsum(skb, start, len, 0);
2940 }
2941 
2942 static __always_inline void
2943 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2944 		     unsigned int off)
2945 {
2946 	if (skb->ip_summed == CHECKSUM_COMPLETE)
2947 		skb->csum = csum_block_add(skb->csum,
2948 					   csum_partial(start, len, 0), off);
2949 }
2950 
2951 /**
2952  *	skb_postpush_rcsum - update checksum for received skb after push
2953  *	@skb: buffer to update
2954  *	@start: start of data after push
2955  *	@len: length of data pushed
2956  *
2957  *	After doing a push on a received packet, you need to call this to
2958  *	update the CHECKSUM_COMPLETE checksum.
2959  */
2960 static inline void skb_postpush_rcsum(struct sk_buff *skb,
2961 				      const void *start, unsigned int len)
2962 {
2963 	__skb_postpush_rcsum(skb, start, len, 0);
2964 }
2965 
2966 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2967 
2968 /**
2969  *	skb_push_rcsum - push skb and update receive checksum
2970  *	@skb: buffer to update
2971  *	@len: length of data pulled
2972  *
2973  *	This function performs an skb_push on the packet and updates
2974  *	the CHECKSUM_COMPLETE checksum.  It should be used on
2975  *	receive path processing instead of skb_push unless you know
2976  *	that the checksum difference is zero (e.g., a valid IP header)
2977  *	or you are setting ip_summed to CHECKSUM_NONE.
2978  */
2979 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
2980 {
2981 	skb_push(skb, len);
2982 	skb_postpush_rcsum(skb, skb->data, len);
2983 	return skb->data;
2984 }
2985 
2986 /**
2987  *	pskb_trim_rcsum - trim received skb and update checksum
2988  *	@skb: buffer to trim
2989  *	@len: new length
2990  *
2991  *	This is exactly the same as pskb_trim except that it ensures the
2992  *	checksum of received packets are still valid after the operation.
2993  */
2994 
2995 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2996 {
2997 	if (likely(len >= skb->len))
2998 		return 0;
2999 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3000 		skb->ip_summed = CHECKSUM_NONE;
3001 	return __pskb_trim(skb, len);
3002 }
3003 
3004 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3005 {
3006 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3007 		skb->ip_summed = CHECKSUM_NONE;
3008 	__skb_trim(skb, len);
3009 	return 0;
3010 }
3011 
3012 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3013 {
3014 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3015 		skb->ip_summed = CHECKSUM_NONE;
3016 	return __skb_grow(skb, len);
3017 }
3018 
3019 #define skb_queue_walk(queue, skb) \
3020 		for (skb = (queue)->next;					\
3021 		     skb != (struct sk_buff *)(queue);				\
3022 		     skb = skb->next)
3023 
3024 #define skb_queue_walk_safe(queue, skb, tmp)					\
3025 		for (skb = (queue)->next, tmp = skb->next;			\
3026 		     skb != (struct sk_buff *)(queue);				\
3027 		     skb = tmp, tmp = skb->next)
3028 
3029 #define skb_queue_walk_from(queue, skb)						\
3030 		for (; skb != (struct sk_buff *)(queue);			\
3031 		     skb = skb->next)
3032 
3033 #define skb_queue_walk_from_safe(queue, skb, tmp)				\
3034 		for (tmp = skb->next;						\
3035 		     skb != (struct sk_buff *)(queue);				\
3036 		     skb = tmp, tmp = skb->next)
3037 
3038 #define skb_queue_reverse_walk(queue, skb) \
3039 		for (skb = (queue)->prev;					\
3040 		     skb != (struct sk_buff *)(queue);				\
3041 		     skb = skb->prev)
3042 
3043 #define skb_queue_reverse_walk_safe(queue, skb, tmp)				\
3044 		for (skb = (queue)->prev, tmp = skb->prev;			\
3045 		     skb != (struct sk_buff *)(queue);				\
3046 		     skb = tmp, tmp = skb->prev)
3047 
3048 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp)			\
3049 		for (tmp = skb->prev;						\
3050 		     skb != (struct sk_buff *)(queue);				\
3051 		     skb = tmp, tmp = skb->prev)
3052 
3053 static inline bool skb_has_frag_list(const struct sk_buff *skb)
3054 {
3055 	return skb_shinfo(skb)->frag_list != NULL;
3056 }
3057 
3058 static inline void skb_frag_list_init(struct sk_buff *skb)
3059 {
3060 	skb_shinfo(skb)->frag_list = NULL;
3061 }
3062 
3063 #define skb_walk_frags(skb, iter)	\
3064 	for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3065 
3066 
3067 int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3068 				const struct sk_buff *skb);
3069 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3070 					  struct sk_buff_head *queue,
3071 					  unsigned int flags,
3072 					  void (*destructor)(struct sock *sk,
3073 							   struct sk_buff *skb),
3074 					  int *peeked, int *off, int *err,
3075 					  struct sk_buff **last);
3076 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3077 					void (*destructor)(struct sock *sk,
3078 							   struct sk_buff *skb),
3079 					int *peeked, int *off, int *err,
3080 					struct sk_buff **last);
3081 struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3082 				    void (*destructor)(struct sock *sk,
3083 						       struct sk_buff *skb),
3084 				    int *peeked, int *off, int *err);
3085 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3086 				  int *err);
3087 unsigned int datagram_poll(struct file *file, struct socket *sock,
3088 			   struct poll_table_struct *wait);
3089 int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3090 			   struct iov_iter *to, int size);
3091 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3092 					struct msghdr *msg, int size)
3093 {
3094 	return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3095 }
3096 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3097 				   struct msghdr *msg);
3098 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3099 				 struct iov_iter *from, int len);
3100 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3101 void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3102 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3103 static inline void skb_free_datagram_locked(struct sock *sk,
3104 					    struct sk_buff *skb)
3105 {
3106 	__skb_free_datagram_locked(sk, skb, 0);
3107 }
3108 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3109 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3110 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3111 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3112 			      int len, __wsum csum);
3113 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3114 		    struct pipe_inode_info *pipe, unsigned int len,
3115 		    unsigned int flags);
3116 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3117 			 int len);
3118 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
3119 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3120 unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3121 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3122 		 int len, int hlen);
3123 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3124 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3125 void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3126 unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
3127 bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
3128 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3129 struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3130 int skb_ensure_writable(struct sk_buff *skb, int write_len);
3131 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3132 int skb_vlan_pop(struct sk_buff *skb);
3133 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3134 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3135 			     gfp_t gfp);
3136 
3137 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3138 {
3139 	return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3140 }
3141 
3142 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3143 {
3144 	return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3145 }
3146 
3147 struct skb_checksum_ops {
3148 	__wsum (*update)(const void *mem, int len, __wsum wsum);
3149 	__wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3150 };
3151 
3152 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3153 
3154 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3155 		      __wsum csum, const struct skb_checksum_ops *ops);
3156 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3157 		    __wsum csum);
3158 
3159 static inline void * __must_check
3160 __skb_header_pointer(const struct sk_buff *skb, int offset,
3161 		     int len, void *data, int hlen, void *buffer)
3162 {
3163 	if (hlen - offset >= len)
3164 		return data + offset;
3165 
3166 	if (!skb ||
3167 	    skb_copy_bits(skb, offset, buffer, len) < 0)
3168 		return NULL;
3169 
3170 	return buffer;
3171 }
3172 
3173 static inline void * __must_check
3174 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3175 {
3176 	return __skb_header_pointer(skb, offset, len, skb->data,
3177 				    skb_headlen(skb), buffer);
3178 }
3179 
3180 /**
3181  *	skb_needs_linearize - check if we need to linearize a given skb
3182  *			      depending on the given device features.
3183  *	@skb: socket buffer to check
3184  *	@features: net device features
3185  *
3186  *	Returns true if either:
3187  *	1. skb has frag_list and the device doesn't support FRAGLIST, or
3188  *	2. skb is fragmented and the device does not support SG.
3189  */
3190 static inline bool skb_needs_linearize(struct sk_buff *skb,
3191 				       netdev_features_t features)
3192 {
3193 	return skb_is_nonlinear(skb) &&
3194 	       ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3195 		(skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3196 }
3197 
3198 static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3199 					     void *to,
3200 					     const unsigned int len)
3201 {
3202 	memcpy(to, skb->data, len);
3203 }
3204 
3205 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3206 						    const int offset, void *to,
3207 						    const unsigned int len)
3208 {
3209 	memcpy(to, skb->data + offset, len);
3210 }
3211 
3212 static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3213 					   const void *from,
3214 					   const unsigned int len)
3215 {
3216 	memcpy(skb->data, from, len);
3217 }
3218 
3219 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3220 						  const int offset,
3221 						  const void *from,
3222 						  const unsigned int len)
3223 {
3224 	memcpy(skb->data + offset, from, len);
3225 }
3226 
3227 void skb_init(void);
3228 
3229 static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3230 {
3231 	return skb->tstamp;
3232 }
3233 
3234 /**
3235  *	skb_get_timestamp - get timestamp from a skb
3236  *	@skb: skb to get stamp from
3237  *	@stamp: pointer to struct timeval to store stamp in
3238  *
3239  *	Timestamps are stored in the skb as offsets to a base timestamp.
3240  *	This function converts the offset back to a struct timeval and stores
3241  *	it in stamp.
3242  */
3243 static inline void skb_get_timestamp(const struct sk_buff *skb,
3244 				     struct timeval *stamp)
3245 {
3246 	*stamp = ktime_to_timeval(skb->tstamp);
3247 }
3248 
3249 static inline void skb_get_timestampns(const struct sk_buff *skb,
3250 				       struct timespec *stamp)
3251 {
3252 	*stamp = ktime_to_timespec(skb->tstamp);
3253 }
3254 
3255 static inline void __net_timestamp(struct sk_buff *skb)
3256 {
3257 	skb->tstamp = ktime_get_real();
3258 }
3259 
3260 static inline ktime_t net_timedelta(ktime_t t)
3261 {
3262 	return ktime_sub(ktime_get_real(), t);
3263 }
3264 
3265 static inline ktime_t net_invalid_timestamp(void)
3266 {
3267 	return 0;
3268 }
3269 
3270 struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3271 
3272 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3273 
3274 void skb_clone_tx_timestamp(struct sk_buff *skb);
3275 bool skb_defer_rx_timestamp(struct sk_buff *skb);
3276 
3277 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3278 
3279 static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3280 {
3281 }
3282 
3283 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3284 {
3285 	return false;
3286 }
3287 
3288 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3289 
3290 /**
3291  * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3292  *
3293  * PHY drivers may accept clones of transmitted packets for
3294  * timestamping via their phy_driver.txtstamp method. These drivers
3295  * must call this function to return the skb back to the stack with a
3296  * timestamp.
3297  *
3298  * @skb: clone of the the original outgoing packet
3299  * @hwtstamps: hardware time stamps
3300  *
3301  */
3302 void skb_complete_tx_timestamp(struct sk_buff *skb,
3303 			       struct skb_shared_hwtstamps *hwtstamps);
3304 
3305 void __skb_tstamp_tx(struct sk_buff *orig_skb,
3306 		     struct skb_shared_hwtstamps *hwtstamps,
3307 		     struct sock *sk, int tstype);
3308 
3309 /**
3310  * skb_tstamp_tx - queue clone of skb with send time stamps
3311  * @orig_skb:	the original outgoing packet
3312  * @hwtstamps:	hardware time stamps, may be NULL if not available
3313  *
3314  * If the skb has a socket associated, then this function clones the
3315  * skb (thus sharing the actual data and optional structures), stores
3316  * the optional hardware time stamping information (if non NULL) or
3317  * generates a software time stamp (otherwise), then queues the clone
3318  * to the error queue of the socket.  Errors are silently ignored.
3319  */
3320 void skb_tstamp_tx(struct sk_buff *orig_skb,
3321 		   struct skb_shared_hwtstamps *hwtstamps);
3322 
3323 /**
3324  * skb_tx_timestamp() - Driver hook for transmit timestamping
3325  *
3326  * Ethernet MAC Drivers should call this function in their hard_xmit()
3327  * function immediately before giving the sk_buff to the MAC hardware.
3328  *
3329  * Specifically, one should make absolutely sure that this function is
3330  * called before TX completion of this packet can trigger.  Otherwise
3331  * the packet could potentially already be freed.
3332  *
3333  * @skb: A socket buffer.
3334  */
3335 static inline void skb_tx_timestamp(struct sk_buff *skb)
3336 {
3337 	skb_clone_tx_timestamp(skb);
3338 	if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3339 		skb_tstamp_tx(skb, NULL);
3340 }
3341 
3342 /**
3343  * skb_complete_wifi_ack - deliver skb with wifi status
3344  *
3345  * @skb: the original outgoing packet
3346  * @acked: ack status
3347  *
3348  */
3349 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3350 
3351 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3352 __sum16 __skb_checksum_complete(struct sk_buff *skb);
3353 
3354 static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3355 {
3356 	return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3357 		skb->csum_valid ||
3358 		(skb->ip_summed == CHECKSUM_PARTIAL &&
3359 		 skb_checksum_start_offset(skb) >= 0));
3360 }
3361 
3362 /**
3363  *	skb_checksum_complete - Calculate checksum of an entire packet
3364  *	@skb: packet to process
3365  *
3366  *	This function calculates the checksum over the entire packet plus
3367  *	the value of skb->csum.  The latter can be used to supply the
3368  *	checksum of a pseudo header as used by TCP/UDP.  It returns the
3369  *	checksum.
3370  *
3371  *	For protocols that contain complete checksums such as ICMP/TCP/UDP,
3372  *	this function can be used to verify that checksum on received
3373  *	packets.  In that case the function should return zero if the
3374  *	checksum is correct.  In particular, this function will return zero
3375  *	if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3376  *	hardware has already verified the correctness of the checksum.
3377  */
3378 static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3379 {
3380 	return skb_csum_unnecessary(skb) ?
3381 	       0 : __skb_checksum_complete(skb);
3382 }
3383 
3384 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3385 {
3386 	if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3387 		if (skb->csum_level == 0)
3388 			skb->ip_summed = CHECKSUM_NONE;
3389 		else
3390 			skb->csum_level--;
3391 	}
3392 }
3393 
3394 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3395 {
3396 	if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3397 		if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3398 			skb->csum_level++;
3399 	} else if (skb->ip_summed == CHECKSUM_NONE) {
3400 		skb->ip_summed = CHECKSUM_UNNECESSARY;
3401 		skb->csum_level = 0;
3402 	}
3403 }
3404 
3405 /* Check if we need to perform checksum complete validation.
3406  *
3407  * Returns true if checksum complete is needed, false otherwise
3408  * (either checksum is unnecessary or zero checksum is allowed).
3409  */
3410 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3411 						  bool zero_okay,
3412 						  __sum16 check)
3413 {
3414 	if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3415 		skb->csum_valid = 1;
3416 		__skb_decr_checksum_unnecessary(skb);
3417 		return false;
3418 	}
3419 
3420 	return true;
3421 }
3422 
3423 /* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3424  * in checksum_init.
3425  */
3426 #define CHECKSUM_BREAK 76
3427 
3428 /* Unset checksum-complete
3429  *
3430  * Unset checksum complete can be done when packet is being modified
3431  * (uncompressed for instance) and checksum-complete value is
3432  * invalidated.
3433  */
3434 static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3435 {
3436 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3437 		skb->ip_summed = CHECKSUM_NONE;
3438 }
3439 
3440 /* Validate (init) checksum based on checksum complete.
3441  *
3442  * Return values:
3443  *   0: checksum is validated or try to in skb_checksum_complete. In the latter
3444  *	case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3445  *	checksum is stored in skb->csum for use in __skb_checksum_complete
3446  *   non-zero: value of invalid checksum
3447  *
3448  */
3449 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3450 						       bool complete,
3451 						       __wsum psum)
3452 {
3453 	if (skb->ip_summed == CHECKSUM_COMPLETE) {
3454 		if (!csum_fold(csum_add(psum, skb->csum))) {
3455 			skb->csum_valid = 1;
3456 			return 0;
3457 		}
3458 	}
3459 
3460 	skb->csum = psum;
3461 
3462 	if (complete || skb->len <= CHECKSUM_BREAK) {
3463 		__sum16 csum;
3464 
3465 		csum = __skb_checksum_complete(skb);
3466 		skb->csum_valid = !csum;
3467 		return csum;
3468 	}
3469 
3470 	return 0;
3471 }
3472 
3473 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3474 {
3475 	return 0;
3476 }
3477 
3478 /* Perform checksum validate (init). Note that this is a macro since we only
3479  * want to calculate the pseudo header which is an input function if necessary.
3480  * First we try to validate without any computation (checksum unnecessary) and
3481  * then calculate based on checksum complete calling the function to compute
3482  * pseudo header.
3483  *
3484  * Return values:
3485  *   0: checksum is validated or try to in skb_checksum_complete
3486  *   non-zero: value of invalid checksum
3487  */
3488 #define __skb_checksum_validate(skb, proto, complete,			\
3489 				zero_okay, check, compute_pseudo)	\
3490 ({									\
3491 	__sum16 __ret = 0;						\
3492 	skb->csum_valid = 0;						\
3493 	if (__skb_checksum_validate_needed(skb, zero_okay, check))	\
3494 		__ret = __skb_checksum_validate_complete(skb,		\
3495 				complete, compute_pseudo(skb, proto));	\
3496 	__ret;								\
3497 })
3498 
3499 #define skb_checksum_init(skb, proto, compute_pseudo)			\
3500 	__skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3501 
3502 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo)	\
3503 	__skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3504 
3505 #define skb_checksum_validate(skb, proto, compute_pseudo)		\
3506 	__skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3507 
3508 #define skb_checksum_validate_zero_check(skb, proto, check,		\
3509 					 compute_pseudo)		\
3510 	__skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3511 
3512 #define skb_checksum_simple_validate(skb)				\
3513 	__skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3514 
3515 static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3516 {
3517 	return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3518 }
3519 
3520 static inline void __skb_checksum_convert(struct sk_buff *skb,
3521 					  __sum16 check, __wsum pseudo)
3522 {
3523 	skb->csum = ~pseudo;
3524 	skb->ip_summed = CHECKSUM_COMPLETE;
3525 }
3526 
3527 #define skb_checksum_try_convert(skb, proto, check, compute_pseudo)	\
3528 do {									\
3529 	if (__skb_checksum_convert_check(skb))				\
3530 		__skb_checksum_convert(skb, check,			\
3531 				       compute_pseudo(skb, proto));	\
3532 } while (0)
3533 
3534 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3535 					      u16 start, u16 offset)
3536 {
3537 	skb->ip_summed = CHECKSUM_PARTIAL;
3538 	skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3539 	skb->csum_offset = offset - start;
3540 }
3541 
3542 /* Update skbuf and packet to reflect the remote checksum offload operation.
3543  * When called, ptr indicates the starting point for skb->csum when
3544  * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3545  * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3546  */
3547 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3548 				       int start, int offset, bool nopartial)
3549 {
3550 	__wsum delta;
3551 
3552 	if (!nopartial) {
3553 		skb_remcsum_adjust_partial(skb, ptr, start, offset);
3554 		return;
3555 	}
3556 
3557 	 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3558 		__skb_checksum_complete(skb);
3559 		skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3560 	}
3561 
3562 	delta = remcsum_adjust(ptr, skb->csum, start, offset);
3563 
3564 	/* Adjust skb->csum since we changed the packet */
3565 	skb->csum = csum_add(skb->csum, delta);
3566 }
3567 
3568 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3569 {
3570 #if IS_ENABLED(CONFIG_NF_CONNTRACK)
3571 	return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3572 #else
3573 	return NULL;
3574 #endif
3575 }
3576 
3577 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3578 void nf_conntrack_destroy(struct nf_conntrack *nfct);
3579 static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3580 {
3581 	if (nfct && atomic_dec_and_test(&nfct->use))
3582 		nf_conntrack_destroy(nfct);
3583 }
3584 static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3585 {
3586 	if (nfct)
3587 		atomic_inc(&nfct->use);
3588 }
3589 #endif
3590 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3591 static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3592 {
3593 	if (nf_bridge && refcount_dec_and_test(&nf_bridge->use))
3594 		kfree(nf_bridge);
3595 }
3596 static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3597 {
3598 	if (nf_bridge)
3599 		refcount_inc(&nf_bridge->use);
3600 }
3601 #endif /* CONFIG_BRIDGE_NETFILTER */
3602 static inline void nf_reset(struct sk_buff *skb)
3603 {
3604 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3605 	nf_conntrack_put(skb_nfct(skb));
3606 	skb->_nfct = 0;
3607 #endif
3608 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3609 	nf_bridge_put(skb->nf_bridge);
3610 	skb->nf_bridge = NULL;
3611 #endif
3612 }
3613 
3614 static inline void nf_reset_trace(struct sk_buff *skb)
3615 {
3616 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3617 	skb->nf_trace = 0;
3618 #endif
3619 }
3620 
3621 /* Note: This doesn't put any conntrack and bridge info in dst. */
3622 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3623 			     bool copy)
3624 {
3625 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3626 	dst->_nfct = src->_nfct;
3627 	nf_conntrack_get(skb_nfct(src));
3628 #endif
3629 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3630 	dst->nf_bridge  = src->nf_bridge;
3631 	nf_bridge_get(src->nf_bridge);
3632 #endif
3633 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
3634 	if (copy)
3635 		dst->nf_trace = src->nf_trace;
3636 #endif
3637 }
3638 
3639 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3640 {
3641 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3642 	nf_conntrack_put(skb_nfct(dst));
3643 #endif
3644 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3645 	nf_bridge_put(dst->nf_bridge);
3646 #endif
3647 	__nf_copy(dst, src, true);
3648 }
3649 
3650 #ifdef CONFIG_NETWORK_SECMARK
3651 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3652 {
3653 	to->secmark = from->secmark;
3654 }
3655 
3656 static inline void skb_init_secmark(struct sk_buff *skb)
3657 {
3658 	skb->secmark = 0;
3659 }
3660 #else
3661 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3662 { }
3663 
3664 static inline void skb_init_secmark(struct sk_buff *skb)
3665 { }
3666 #endif
3667 
3668 static inline bool skb_irq_freeable(const struct sk_buff *skb)
3669 {
3670 	return !skb->destructor &&
3671 #if IS_ENABLED(CONFIG_XFRM)
3672 		!skb->sp &&
3673 #endif
3674 		!skb_nfct(skb) &&
3675 		!skb->_skb_refdst &&
3676 		!skb_has_frag_list(skb);
3677 }
3678 
3679 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3680 {
3681 	skb->queue_mapping = queue_mapping;
3682 }
3683 
3684 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
3685 {
3686 	return skb->queue_mapping;
3687 }
3688 
3689 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3690 {
3691 	to->queue_mapping = from->queue_mapping;
3692 }
3693 
3694 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3695 {
3696 	skb->queue_mapping = rx_queue + 1;
3697 }
3698 
3699 static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
3700 {
3701 	return skb->queue_mapping - 1;
3702 }
3703 
3704 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
3705 {
3706 	return skb->queue_mapping != 0;
3707 }
3708 
3709 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3710 {
3711 	skb->dst_pending_confirm = val;
3712 }
3713 
3714 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3715 {
3716 	return skb->dst_pending_confirm != 0;
3717 }
3718 
3719 static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3720 {
3721 #ifdef CONFIG_XFRM
3722 	return skb->sp;
3723 #else
3724 	return NULL;
3725 #endif
3726 }
3727 
3728 /* Keeps track of mac header offset relative to skb->head.
3729  * It is useful for TSO of Tunneling protocol. e.g. GRE.
3730  * For non-tunnel skb it points to skb_mac_header() and for
3731  * tunnel skb it points to outer mac header.
3732  * Keeps track of level of encapsulation of network headers.
3733  */
3734 struct skb_gso_cb {
3735 	union {
3736 		int	mac_offset;
3737 		int	data_offset;
3738 	};
3739 	int	encap_level;
3740 	__wsum	csum;
3741 	__u16	csum_start;
3742 };
3743 #define SKB_SGO_CB_OFFSET	32
3744 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
3745 
3746 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3747 {
3748 	return (skb_mac_header(inner_skb) - inner_skb->head) -
3749 		SKB_GSO_CB(inner_skb)->mac_offset;
3750 }
3751 
3752 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3753 {
3754 	int new_headroom, headroom;
3755 	int ret;
3756 
3757 	headroom = skb_headroom(skb);
3758 	ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3759 	if (ret)
3760 		return ret;
3761 
3762 	new_headroom = skb_headroom(skb);
3763 	SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3764 	return 0;
3765 }
3766 
3767 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3768 {
3769 	/* Do not update partial checksums if remote checksum is enabled. */
3770 	if (skb->remcsum_offload)
3771 		return;
3772 
3773 	SKB_GSO_CB(skb)->csum = res;
3774 	SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3775 }
3776 
3777 /* Compute the checksum for a gso segment. First compute the checksum value
3778  * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3779  * then add in skb->csum (checksum from csum_start to end of packet).
3780  * skb->csum and csum_start are then updated to reflect the checksum of the
3781  * resultant packet starting from the transport header-- the resultant checksum
3782  * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3783  * header.
3784  */
3785 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3786 {
3787 	unsigned char *csum_start = skb_transport_header(skb);
3788 	int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3789 	__wsum partial = SKB_GSO_CB(skb)->csum;
3790 
3791 	SKB_GSO_CB(skb)->csum = res;
3792 	SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
3793 
3794 	return csum_fold(csum_partial(csum_start, plen, partial));
3795 }
3796 
3797 static inline bool skb_is_gso(const struct sk_buff *skb)
3798 {
3799 	return skb_shinfo(skb)->gso_size;
3800 }
3801 
3802 /* Note: Should be called only if skb_is_gso(skb) is true */
3803 static inline bool skb_is_gso_v6(const struct sk_buff *skb)
3804 {
3805 	return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3806 }
3807 
3808 static inline void skb_gso_reset(struct sk_buff *skb)
3809 {
3810 	skb_shinfo(skb)->gso_size = 0;
3811 	skb_shinfo(skb)->gso_segs = 0;
3812 	skb_shinfo(skb)->gso_type = 0;
3813 }
3814 
3815 void __skb_warn_lro_forwarding(const struct sk_buff *skb);
3816 
3817 static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3818 {
3819 	/* LRO sets gso_size but not gso_type, whereas if GSO is really
3820 	 * wanted then gso_type will be set. */
3821 	const struct skb_shared_info *shinfo = skb_shinfo(skb);
3822 
3823 	if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3824 	    unlikely(shinfo->gso_type == 0)) {
3825 		__skb_warn_lro_forwarding(skb);
3826 		return true;
3827 	}
3828 	return false;
3829 }
3830 
3831 static inline void skb_forward_csum(struct sk_buff *skb)
3832 {
3833 	/* Unfortunately we don't support this one.  Any brave souls? */
3834 	if (skb->ip_summed == CHECKSUM_COMPLETE)
3835 		skb->ip_summed = CHECKSUM_NONE;
3836 }
3837 
3838 /**
3839  * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3840  * @skb: skb to check
3841  *
3842  * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3843  * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3844  * use this helper, to document places where we make this assertion.
3845  */
3846 static inline void skb_checksum_none_assert(const struct sk_buff *skb)
3847 {
3848 #ifdef DEBUG
3849 	BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3850 #endif
3851 }
3852 
3853 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
3854 
3855 int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3856 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3857 				     unsigned int transport_len,
3858 				     __sum16(*skb_chkf)(struct sk_buff *skb));
3859 
3860 /**
3861  * skb_head_is_locked - Determine if the skb->head is locked down
3862  * @skb: skb to check
3863  *
3864  * The head on skbs build around a head frag can be removed if they are
3865  * not cloned.  This function returns true if the skb head is locked down
3866  * due to either being allocated via kmalloc, or by being a clone with
3867  * multiple references to the head.
3868  */
3869 static inline bool skb_head_is_locked(const struct sk_buff *skb)
3870 {
3871 	return !skb->head_frag || skb_cloned(skb);
3872 }
3873 
3874 /**
3875  * skb_gso_network_seglen - Return length of individual segments of a gso packet
3876  *
3877  * @skb: GSO skb
3878  *
3879  * skb_gso_network_seglen is used to determine the real size of the
3880  * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3881  *
3882  * The MAC/L2 header is not accounted for.
3883  */
3884 static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3885 {
3886 	unsigned int hdr_len = skb_transport_header(skb) -
3887 			       skb_network_header(skb);
3888 	return hdr_len + skb_gso_transport_seglen(skb);
3889 }
3890 
3891 /* Local Checksum Offload.
3892  * Compute outer checksum based on the assumption that the
3893  * inner checksum will be offloaded later.
3894  * See Documentation/networking/checksum-offloads.txt for
3895  * explanation of how this works.
3896  * Fill in outer checksum adjustment (e.g. with sum of outer
3897  * pseudo-header) before calling.
3898  * Also ensure that inner checksum is in linear data area.
3899  */
3900 static inline __wsum lco_csum(struct sk_buff *skb)
3901 {
3902 	unsigned char *csum_start = skb_checksum_start(skb);
3903 	unsigned char *l4_hdr = skb_transport_header(skb);
3904 	__wsum partial;
3905 
3906 	/* Start with complement of inner checksum adjustment */
3907 	partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3908 						    skb->csum_offset));
3909 
3910 	/* Add in checksum of our headers (incl. outer checksum
3911 	 * adjustment filled in by caller) and return result.
3912 	 */
3913 	return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
3914 }
3915 
3916 #endif	/* __KERNEL__ */
3917 #endif	/* _LINUX_SKBUFF_H */
3918