net: Add GSO support for UDP tunnels with checksum
[linux-2.6-block.git] / include / linux / skbuff.h
CommitLineData
1da177e4
LT
1/*
2 * Definitions for the 'struct sk_buff' memory handlers.
3 *
4 * Authors:
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
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
1da177e4 17#include <linux/kernel.h>
fe55f6d5 18#include <linux/kmemcheck.h>
1da177e4
LT
19#include <linux/compiler.h>
20#include <linux/time.h>
187f1882 21#include <linux/bug.h>
1da177e4
LT
22#include <linux/cache.h>
23
60063497 24#include <linux/atomic.h>
1da177e4
LT
25#include <asm/types.h>
26#include <linux/spinlock.h>
1da177e4 27#include <linux/net.h>
3fc7e8a6 28#include <linux/textsearch.h>
1da177e4 29#include <net/checksum.h>
a80958f4 30#include <linux/rcupdate.h>
97fc2f08 31#include <linux/dmaengine.h>
b7aa0bf7 32#include <linux/hrtimer.h>
131ea667 33#include <linux/dma-mapping.h>
c8f44aff 34#include <linux/netdev_features.h>
363ec392 35#include <linux/sched.h>
5203cd28 36#include <net/flow_keys.h>
1da177e4 37
78ea85f1
DB
38/* A. Checksumming of received packets by device.
39 *
40 * CHECKSUM_NONE:
41 *
42 * Device failed to checksum this packet e.g. due to lack of capabilities.
43 * The packet contains full (though not verified) checksum in packet but
44 * not in skb->csum. Thus, skb->csum is undefined in this case.
45 *
46 * CHECKSUM_UNNECESSARY:
47 *
48 * The hardware you're dealing with doesn't calculate the full checksum
49 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
50 * for specific protocols e.g. TCP/UDP/SCTP, then, for such packets it will
51 * set CHECKSUM_UNNECESSARY if their checksums are okay. skb->csum is still
52 * undefined in this case though. It is a bad option, but, unfortunately,
53 * nowadays most vendors do this. Apparently with the secret goal to sell
54 * you new devices, when you will add new protocol to your host, f.e. IPv6 8)
55 *
56 * CHECKSUM_COMPLETE:
57 *
58 * This is the most generic way. The device supplied checksum of the _whole_
59 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
60 * hardware doesn't need to parse L3/L4 headers to implement this.
61 *
62 * Note: Even if device supports only some protocols, but is able to produce
63 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
64 *
65 * CHECKSUM_PARTIAL:
66 *
67 * This is identical to the case for output below. This may occur on a packet
68 * received directly from another Linux OS, e.g., a virtualized Linux kernel
69 * on the same host. The packet can be treated in the same way as
70 * CHECKSUM_UNNECESSARY, except that on output (i.e., forwarding) the
71 * checksum must be filled in by the OS or the hardware.
72 *
73 * B. Checksumming on output.
74 *
75 * CHECKSUM_NONE:
76 *
77 * The skb was already checksummed by the protocol, or a checksum is not
78 * required.
79 *
80 * CHECKSUM_PARTIAL:
81 *
82 * The device is required to checksum the packet as seen by hard_start_xmit()
83 * from skb->csum_start up to the end, and to record/write the checksum at
84 * offset skb->csum_start + skb->csum_offset.
85 *
86 * The device must show its capabilities in dev->features, set up at device
87 * setup time, e.g. netdev_features.h:
88 *
89 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
90 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
91 * IPv4. Sigh. Vendors like this way for an unknown reason.
92 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
93 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
94 * NETIF_F_... - Well, you get the picture.
95 *
96 * CHECKSUM_UNNECESSARY:
97 *
98 * Normally, the device will do per protocol specific checksumming. Protocol
99 * implementations that do not want the NIC to perform the checksum
100 * calculation should use this flag in their outgoing skbs.
101 *
102 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
103 * offload. Correspondingly, the FCoE protocol driver
104 * stack should use CHECKSUM_UNNECESSARY.
105 *
106 * Any questions? No questions, good. --ANK
107 */
108
60476372 109/* Don't change this without changing skb_csum_unnecessary! */
78ea85f1
DB
110#define CHECKSUM_NONE 0
111#define CHECKSUM_UNNECESSARY 1
112#define CHECKSUM_COMPLETE 2
113#define CHECKSUM_PARTIAL 3
1da177e4
LT
114
115#define SKB_DATA_ALIGN(X) (((X) + (SMP_CACHE_BYTES - 1)) & \
116 ~(SMP_CACHE_BYTES - 1))
fc910a27 117#define SKB_WITH_OVERHEAD(X) \
deea84b0 118 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
fc910a27
DM
119#define SKB_MAX_ORDER(X, ORDER) \
120 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
1da177e4
LT
121#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
122#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
123
87fb4b7b
ED
124/* return minimum truesize of one skb containing X bytes of data */
125#define SKB_TRUESIZE(X) ((X) + \
126 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
127 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
128
1da177e4 129struct net_device;
716ea3a7 130struct scatterlist;
9c55e01c 131struct pipe_inode_info;
1da177e4 132
5f79e0f9 133#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1da177e4
LT
134struct nf_conntrack {
135 atomic_t use;
1da177e4 136};
5f79e0f9 137#endif
1da177e4
LT
138
139#ifdef CONFIG_BRIDGE_NETFILTER
140struct nf_bridge_info {
bf1ac5ca
ED
141 atomic_t use;
142 unsigned int mask;
143 struct net_device *physindev;
144 struct net_device *physoutdev;
145 unsigned long data[32 / sizeof(unsigned long)];
1da177e4
LT
146};
147#endif
148
1da177e4
LT
149struct sk_buff_head {
150 /* These two members must be first. */
151 struct sk_buff *next;
152 struct sk_buff *prev;
153
154 __u32 qlen;
155 spinlock_t lock;
156};
157
158struct sk_buff;
159
9d4dde52
IC
160/* To allow 64K frame to be packed as single skb without frag_list we
161 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
162 * buffers which do not start on a page boundary.
163 *
164 * Since GRO uses frags we allocate at least 16 regardless of page
165 * size.
a715dea3 166 */
9d4dde52 167#if (65536/PAGE_SIZE + 1) < 16
eec00954 168#define MAX_SKB_FRAGS 16UL
a715dea3 169#else
9d4dde52 170#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
a715dea3 171#endif
1da177e4
LT
172
173typedef struct skb_frag_struct skb_frag_t;
174
175struct skb_frag_struct {
a8605c60
IC
176 struct {
177 struct page *p;
178 } page;
cb4dfe56 179#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
a309bb07
DM
180 __u32 page_offset;
181 __u32 size;
cb4dfe56
ED
182#else
183 __u16 page_offset;
184 __u16 size;
185#endif
1da177e4
LT
186};
187
9e903e08
ED
188static inline unsigned int skb_frag_size(const skb_frag_t *frag)
189{
190 return frag->size;
191}
192
193static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
194{
195 frag->size = size;
196}
197
198static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
199{
200 frag->size += delta;
201}
202
203static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
204{
205 frag->size -= delta;
206}
207
ac45f602
PO
208#define HAVE_HW_TIME_STAMP
209
210/**
d3a21be8 211 * struct skb_shared_hwtstamps - hardware time stamps
ac45f602
PO
212 * @hwtstamp: hardware time stamp transformed into duration
213 * since arbitrary point in time
214 * @syststamp: hwtstamp transformed to system time base
215 *
216 * Software time stamps generated by ktime_get_real() are stored in
217 * skb->tstamp. The relation between the different kinds of time
218 * stamps is as follows:
219 *
220 * syststamp and tstamp can be compared against each other in
221 * arbitrary combinations. The accuracy of a
222 * syststamp/tstamp/"syststamp from other device" comparison is
223 * limited by the accuracy of the transformation into system time
224 * base. This depends on the device driver and its underlying
225 * hardware.
226 *
227 * hwtstamps can only be compared against other hwtstamps from
228 * the same device.
229 *
230 * This structure is attached to packets as part of the
231 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
232 */
233struct skb_shared_hwtstamps {
234 ktime_t hwtstamp;
235 ktime_t syststamp;
236};
237
2244d07b
OH
238/* Definitions for tx_flags in struct skb_shared_info */
239enum {
240 /* generate hardware time stamp */
241 SKBTX_HW_TSTAMP = 1 << 0,
242
243 /* generate software time stamp */
244 SKBTX_SW_TSTAMP = 1 << 1,
245
246 /* device driver is going to provide hardware time stamp */
247 SKBTX_IN_PROGRESS = 1 << 2,
248
a6686f2f 249 /* device driver supports TX zero-copy buffers */
62b1a8ab 250 SKBTX_DEV_ZEROCOPY = 1 << 3,
6e3e939f
JB
251
252 /* generate wifi status information (where possible) */
62b1a8ab 253 SKBTX_WIFI_STATUS = 1 << 4,
c9af6db4
PS
254
255 /* This indicates at least one fragment might be overwritten
256 * (as in vmsplice(), sendfile() ...)
257 * If we need to compute a TX checksum, we'll need to copy
258 * all frags to avoid possible bad checksum
259 */
260 SKBTX_SHARED_FRAG = 1 << 5,
a6686f2f
SM
261};
262
263/*
264 * The callback notifies userspace to release buffers when skb DMA is done in
265 * lower device, the skb last reference should be 0 when calling this.
e19d6763
MT
266 * The zerocopy_success argument is true if zero copy transmit occurred,
267 * false on data copy or out of memory error caused by data copy attempt.
ca8f4fb2
MT
268 * The ctx field is used to track device context.
269 * The desc field is used to track userspace buffer index.
a6686f2f
SM
270 */
271struct ubuf_info {
e19d6763 272 void (*callback)(struct ubuf_info *, bool zerocopy_success);
ca8f4fb2 273 void *ctx;
a6686f2f 274 unsigned long desc;
ac45f602
PO
275};
276
1da177e4
LT
277/* This data is invariant across clones and lives at
278 * the end of the header data, ie. at skb->end.
279 */
280struct skb_shared_info {
9f42f126
IC
281 unsigned char nr_frags;
282 __u8 tx_flags;
7967168c
HX
283 unsigned short gso_size;
284 /* Warning: this field is not always filled in (UFO)! */
285 unsigned short gso_segs;
286 unsigned short gso_type;
1da177e4 287 struct sk_buff *frag_list;
ac45f602 288 struct skb_shared_hwtstamps hwtstamps;
9f42f126 289 __be32 ip6_frag_id;
ec7d2f2c
ED
290
291 /*
292 * Warning : all fields before dataref are cleared in __alloc_skb()
293 */
294 atomic_t dataref;
295
69e3c75f
JB
296 /* Intermediate layers must ensure that destructor_arg
297 * remains valid until skb destructor */
298 void * destructor_arg;
a6686f2f 299
fed66381
ED
300 /* must be last field, see pskb_expand_head() */
301 skb_frag_t frags[MAX_SKB_FRAGS];
1da177e4
LT
302};
303
304/* We divide dataref into two halves. The higher 16 bits hold references
305 * to the payload part of skb->data. The lower 16 bits hold references to
334a8132
PM
306 * the entire skb->data. A clone of a headerless skb holds the length of
307 * the header in skb->hdr_len.
1da177e4
LT
308 *
309 * All users must obey the rule that the skb->data reference count must be
310 * greater than or equal to the payload reference count.
311 *
312 * Holding a reference to the payload part means that the user does not
313 * care about modifications to the header part of skb->data.
314 */
315#define SKB_DATAREF_SHIFT 16
316#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
317
d179cd12
DM
318
319enum {
320 SKB_FCLONE_UNAVAILABLE,
321 SKB_FCLONE_ORIG,
322 SKB_FCLONE_CLONE,
323};
324
7967168c
HX
325enum {
326 SKB_GSO_TCPV4 = 1 << 0,
f83ef8c0 327 SKB_GSO_UDP = 1 << 1,
576a30eb
HX
328
329 /* This indicates the skb is from an untrusted source. */
330 SKB_GSO_DODGY = 1 << 2,
b0da8537
MC
331
332 /* This indicates the tcp segment has CWR set. */
f83ef8c0
HX
333 SKB_GSO_TCP_ECN = 1 << 3,
334
335 SKB_GSO_TCPV6 = 1 << 4,
01d5b2fc
CL
336
337 SKB_GSO_FCOE = 1 << 5,
68c33163
PS
338
339 SKB_GSO_GRE = 1 << 6,
73136267 340
cb32f511 341 SKB_GSO_IPIP = 1 << 7,
0d89d203 342
61c1db7f 343 SKB_GSO_SIT = 1 << 8,
cb32f511 344
61c1db7f
ED
345 SKB_GSO_UDP_TUNNEL = 1 << 9,
346
347 SKB_GSO_MPLS = 1 << 10,
0f4f4ffa
TH
348
349 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
7967168c
HX
350};
351
2e07fa9c
ACM
352#if BITS_PER_LONG > 32
353#define NET_SKBUFF_DATA_USES_OFFSET 1
354#endif
355
356#ifdef NET_SKBUFF_DATA_USES_OFFSET
357typedef unsigned int sk_buff_data_t;
358#else
359typedef unsigned char *sk_buff_data_t;
360#endif
361
363ec392
ED
362/**
363 * struct skb_mstamp - multi resolution time stamps
364 * @stamp_us: timestamp in us resolution
365 * @stamp_jiffies: timestamp in jiffies
366 */
367struct skb_mstamp {
368 union {
369 u64 v64;
370 struct {
371 u32 stamp_us;
372 u32 stamp_jiffies;
373 };
374 };
375};
376
377/**
378 * skb_mstamp_get - get current timestamp
379 * @cl: place to store timestamps
380 */
381static inline void skb_mstamp_get(struct skb_mstamp *cl)
382{
383 u64 val = local_clock();
384
385 do_div(val, NSEC_PER_USEC);
386 cl->stamp_us = (u32)val;
387 cl->stamp_jiffies = (u32)jiffies;
388}
389
390/**
391 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
392 * @t1: pointer to newest sample
393 * @t0: pointer to oldest sample
394 */
395static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
396 const struct skb_mstamp *t0)
397{
398 s32 delta_us = t1->stamp_us - t0->stamp_us;
399 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
400
401 /* If delta_us is negative, this might be because interval is too big,
402 * or local_clock() drift is too big : fallback using jiffies.
403 */
404 if (delta_us <= 0 ||
405 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
406
407 delta_us = jiffies_to_usecs(delta_jiffies);
408
409 return delta_us;
410}
411
412
1da177e4
LT
413/**
414 * struct sk_buff - socket buffer
415 * @next: Next buffer in list
416 * @prev: Previous buffer in list
363ec392 417 * @tstamp: Time we arrived/left
d84e0bd7 418 * @sk: Socket we are owned by
1da177e4 419 * @dev: Device we arrived on/are leaving by
d84e0bd7 420 * @cb: Control buffer. Free for use by every layer. Put private vars here
7fee226a 421 * @_skb_refdst: destination entry (with norefcount bit)
67be2dd1 422 * @sp: the security path, used for xfrm
1da177e4
LT
423 * @len: Length of actual data
424 * @data_len: Data length
425 * @mac_len: Length of link layer header
334a8132 426 * @hdr_len: writable header length of cloned skb
663ead3b
HX
427 * @csum: Checksum (must include start/offset pair)
428 * @csum_start: Offset from skb->head where checksumming should start
429 * @csum_offset: Offset from csum_start where checksum should be stored
d84e0bd7 430 * @priority: Packet queueing priority
60ff7467 431 * @ignore_df: allow local fragmentation
1da177e4 432 * @cloned: Head may be cloned (check refcnt to be sure)
d84e0bd7 433 * @ip_summed: Driver fed us an IP checksum
1da177e4 434 * @nohdr: Payload reference only, must not modify header
d84e0bd7 435 * @nfctinfo: Relationship of this skb to the connection
1da177e4 436 * @pkt_type: Packet class
c83c2486 437 * @fclone: skbuff clone status
c83c2486 438 * @ipvs_property: skbuff is owned by ipvs
31729363
RD
439 * @peeked: this packet has been seen already, so stats have been
440 * done for it, don't do them again
ba9dda3a 441 * @nf_trace: netfilter packet trace flag
d84e0bd7
DB
442 * @protocol: Packet protocol from driver
443 * @destructor: Destruct function
444 * @nfct: Associated connection, if any
1da177e4 445 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
8964be4a 446 * @skb_iif: ifindex of device we arrived on
1da177e4
LT
447 * @tc_index: Traffic control index
448 * @tc_verd: traffic control verdict
61b905da 449 * @hash: the packet hash
d84e0bd7 450 * @queue_mapping: Queue mapping for multiqueue devices
553a5672 451 * @ndisc_nodetype: router type (from link layer)
d84e0bd7 452 * @ooo_okay: allow the mapping of a socket to a queue to be changed
61b905da 453 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
4ca2462e 454 * ports.
6e3e939f
JB
455 * @wifi_acked_valid: wifi_acked was set
456 * @wifi_acked: whether frame was acked on wifi or not
3bdc0eba 457 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
f4b8ea78
RD
458 * @dma_cookie: a cookie to one of several possible DMA operations
459 * done by skb DMA functions
06021292 460 * @napi_id: id of the NAPI struct this skb came from
984bc16c 461 * @secmark: security marking
d84e0bd7
DB
462 * @mark: Generic packet mark
463 * @dropcount: total number of sk_receive_queue overflows
86a9bad3 464 * @vlan_proto: vlan encapsulation protocol
6aa895b0 465 * @vlan_tci: vlan tag control information
0d89d203 466 * @inner_protocol: Protocol (encapsulation)
6a674e9c
JG
467 * @inner_transport_header: Inner transport layer header (encapsulation)
468 * @inner_network_header: Network layer header (encapsulation)
aefbd2b3 469 * @inner_mac_header: Link layer header (encapsulation)
d84e0bd7
DB
470 * @transport_header: Transport layer header
471 * @network_header: Network layer header
472 * @mac_header: Link layer header
473 * @tail: Tail pointer
474 * @end: End pointer
475 * @head: Head of buffer
476 * @data: Data head pointer
477 * @truesize: Buffer size
478 * @users: User count - see {datagram,tcp}.c
1da177e4
LT
479 */
480
481struct sk_buff {
482 /* These two members must be first. */
483 struct sk_buff *next;
484 struct sk_buff *prev;
485
363ec392
ED
486 union {
487 ktime_t tstamp;
488 struct skb_mstamp skb_mstamp;
489 };
da3f5cf1
FF
490
491 struct sock *sk;
1da177e4 492 struct net_device *dev;
1da177e4 493
1da177e4
LT
494 /*
495 * This is the control buffer. It is free to use for every
496 * layer. Please put your private variables there. If you
497 * want to keep them across layers you have to do a skb_clone()
498 * first. This is owned by whoever has the skb queued ATM.
499 */
da3f5cf1 500 char cb[48] __aligned(8);
1da177e4 501
7fee226a 502 unsigned long _skb_refdst;
da3f5cf1
FF
503#ifdef CONFIG_XFRM
504 struct sec_path *sp;
505#endif
1da177e4 506 unsigned int len,
334a8132
PM
507 data_len;
508 __u16 mac_len,
509 hdr_len;
ff1dcadb
AV
510 union {
511 __wsum csum;
663ead3b
HX
512 struct {
513 __u16 csum_start;
514 __u16 csum_offset;
515 };
ff1dcadb 516 };
1da177e4 517 __u32 priority;
fe55f6d5 518 kmemcheck_bitfield_begin(flags1);
60ff7467 519 __u8 ignore_df:1,
1cbb3380
TG
520 cloned:1,
521 ip_summed:2,
6869c4d8
HW
522 nohdr:1,
523 nfctinfo:3;
d179cd12 524 __u8 pkt_type:3,
b84f4cc9 525 fclone:2,
ba9dda3a 526 ipvs_property:1,
a59322be 527 peeked:1,
ba9dda3a 528 nf_trace:1;
fe55f6d5 529 kmemcheck_bitfield_end(flags1);
4ab408de 530 __be16 protocol;
1da177e4
LT
531
532 void (*destructor)(struct sk_buff *skb);
9fb9cbb1 533#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
5f79e0f9 534 struct nf_conntrack *nfct;
2fc72c7b 535#endif
1da177e4
LT
536#ifdef CONFIG_BRIDGE_NETFILTER
537 struct nf_bridge_info *nf_bridge;
538#endif
f25f4e44 539
8964be4a 540 int skb_iif;
4031ae6e 541
61b905da 542 __u32 hash;
4031ae6e 543
86a9bad3 544 __be16 vlan_proto;
4031ae6e
AD
545 __u16 vlan_tci;
546
1da177e4 547#ifdef CONFIG_NET_SCHED
b6b99eb5 548 __u16 tc_index; /* traffic control index */
1da177e4 549#ifdef CONFIG_NET_CLS_ACT
b6b99eb5 550 __u16 tc_verd; /* traffic control verdict */
1da177e4 551#endif
1da177e4 552#endif
fe55f6d5 553
0a14842f 554 __u16 queue_mapping;
fe55f6d5 555 kmemcheck_bitfield_begin(flags2);
de357cc0 556#ifdef CONFIG_IPV6_NDISC_NODETYPE
8a4eb573 557 __u8 ndisc_nodetype:2;
d0f09804 558#endif
c93bdd0e 559 __u8 pfmemalloc:1;
3853b584 560 __u8 ooo_okay:1;
61b905da 561 __u8 l4_hash:1;
6e3e939f
JB
562 __u8 wifi_acked_valid:1;
563 __u8 wifi_acked:1;
3bdc0eba 564 __u8 no_fcs:1;
d3836f21 565 __u8 head_frag:1;
6a674e9c
JG
566 /* Encapsulation protocol and NIC drivers should use
567 * this flag to indicate to each other if the skb contains
568 * encapsulated packet or not and maybe use the inner packet
569 * headers if needed
570 */
571 __u8 encapsulation:1;
7e2b10c1
TH
572 __u8 encap_hdr_csum:1;
573 /* 5/7 bit hole (depending on ndisc_nodetype presence) */
fe55f6d5
VN
574 kmemcheck_bitfield_end(flags2);
575
e0d1095a 576#if defined CONFIG_NET_DMA || defined CONFIG_NET_RX_BUSY_POLL
06021292
ET
577 union {
578 unsigned int napi_id;
579 dma_cookie_t dma_cookie;
580 };
97fc2f08 581#endif
984bc16c
JM
582#ifdef CONFIG_NETWORK_SECMARK
583 __u32 secmark;
584#endif
3b885787
NH
585 union {
586 __u32 mark;
587 __u32 dropcount;
16fad69c 588 __u32 reserved_tailroom;
3b885787 589 };
1da177e4 590
0d89d203 591 __be16 inner_protocol;
1a37e412
SH
592 __u16 inner_transport_header;
593 __u16 inner_network_header;
594 __u16 inner_mac_header;
595 __u16 transport_header;
596 __u16 network_header;
597 __u16 mac_header;
1da177e4 598 /* These elements must be at the end, see alloc_skb() for details. */
27a884dc 599 sk_buff_data_t tail;
4305b541 600 sk_buff_data_t end;
1da177e4 601 unsigned char *head,
4305b541 602 *data;
27a884dc
ACM
603 unsigned int truesize;
604 atomic_t users;
1da177e4
LT
605};
606
607#ifdef __KERNEL__
608/*
609 * Handling routines are only of interest to the kernel
610 */
611#include <linux/slab.h>
612
1da177e4 613
c93bdd0e
MG
614#define SKB_ALLOC_FCLONE 0x01
615#define SKB_ALLOC_RX 0x02
616
617/* Returns true if the skb was allocated from PFMEMALLOC reserves */
618static inline bool skb_pfmemalloc(const struct sk_buff *skb)
619{
620 return unlikely(skb->pfmemalloc);
621}
622
7fee226a
ED
623/*
624 * skb might have a dst pointer attached, refcounted or not.
625 * _skb_refdst low order bit is set if refcount was _not_ taken
626 */
627#define SKB_DST_NOREF 1UL
628#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
629
630/**
631 * skb_dst - returns skb dst_entry
632 * @skb: buffer
633 *
634 * Returns skb dst_entry, regardless of reference taken or not.
635 */
adf30907
ED
636static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
637{
7fee226a
ED
638 /* If refdst was not refcounted, check we still are in a
639 * rcu_read_lock section
640 */
641 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
642 !rcu_read_lock_held() &&
643 !rcu_read_lock_bh_held());
644 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
adf30907
ED
645}
646
7fee226a
ED
647/**
648 * skb_dst_set - sets skb dst
649 * @skb: buffer
650 * @dst: dst entry
651 *
652 * Sets skb dst, assuming a reference was taken on dst and should
653 * be released by skb_dst_drop()
654 */
adf30907
ED
655static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
656{
7fee226a
ED
657 skb->_skb_refdst = (unsigned long)dst;
658}
659
7965bd4d
JP
660void __skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst,
661 bool force);
932bc4d7
JA
662
663/**
664 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
665 * @skb: buffer
666 * @dst: dst entry
667 *
668 * Sets skb dst, assuming a reference was not taken on dst.
669 * If dst entry is cached, we do not take reference and dst_release
670 * will be avoided by refdst_drop. If dst entry is not cached, we take
671 * reference, so that last dst_release can destroy the dst immediately.
672 */
673static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
674{
675 __skb_dst_set_noref(skb, dst, false);
676}
677
678/**
679 * skb_dst_set_noref_force - sets skb dst, without taking reference
680 * @skb: buffer
681 * @dst: dst entry
682 *
683 * Sets skb dst, assuming a reference was not taken on dst.
684 * No reference is taken and no dst_release will be called. While for
685 * cached dsts deferred reclaim is a basic feature, for entries that are
686 * not cached it is caller's job to guarantee that last dst_release for
687 * provided dst happens when nobody uses it, eg. after a RCU grace period.
688 */
689static inline void skb_dst_set_noref_force(struct sk_buff *skb,
690 struct dst_entry *dst)
691{
692 __skb_dst_set_noref(skb, dst, true);
693}
7fee226a
ED
694
695/**
25985edc 696 * skb_dst_is_noref - Test if skb dst isn't refcounted
7fee226a
ED
697 * @skb: buffer
698 */
699static inline bool skb_dst_is_noref(const struct sk_buff *skb)
700{
701 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
adf30907
ED
702}
703
511c3f92
ED
704static inline struct rtable *skb_rtable(const struct sk_buff *skb)
705{
adf30907 706 return (struct rtable *)skb_dst(skb);
511c3f92
ED
707}
708
7965bd4d
JP
709void kfree_skb(struct sk_buff *skb);
710void kfree_skb_list(struct sk_buff *segs);
711void skb_tx_error(struct sk_buff *skb);
712void consume_skb(struct sk_buff *skb);
713void __kfree_skb(struct sk_buff *skb);
d7e8883c 714extern struct kmem_cache *skbuff_head_cache;
bad43ca8 715
7965bd4d
JP
716void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
717bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
718 bool *fragstolen, int *delta_truesize);
bad43ca8 719
7965bd4d
JP
720struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
721 int node);
722struct sk_buff *build_skb(void *data, unsigned int frag_size);
d179cd12 723static inline struct sk_buff *alloc_skb(unsigned int size,
dd0fc66f 724 gfp_t priority)
d179cd12 725{
564824b0 726 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
d179cd12
DM
727}
728
729static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
dd0fc66f 730 gfp_t priority)
d179cd12 731{
c93bdd0e 732 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
d179cd12
DM
733}
734
7965bd4d 735struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
0ebd0ac5
PM
736static inline struct sk_buff *alloc_skb_head(gfp_t priority)
737{
738 return __alloc_skb_head(priority, -1);
739}
740
7965bd4d
JP
741struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
742int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
743struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
744struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
745struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask);
746
747int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
748struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
749 unsigned int headroom);
750struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
751 int newtailroom, gfp_t priority);
25a91d8d
FD
752int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
753 int offset, int len);
7965bd4d
JP
754int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
755 int len);
756int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
757int skb_pad(struct sk_buff *skb, int pad);
ead2ceb0 758#define dev_kfree_skb(a) consume_skb(a)
1da177e4 759
7965bd4d
JP
760int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
761 int getfrag(void *from, char *to, int offset,
762 int len, int odd, struct sk_buff *skb),
763 void *from, int length);
e89e9cf5 764
d94d9fee 765struct skb_seq_state {
677e90ed
TG
766 __u32 lower_offset;
767 __u32 upper_offset;
768 __u32 frag_idx;
769 __u32 stepped_offset;
770 struct sk_buff *root_skb;
771 struct sk_buff *cur_skb;
772 __u8 *frag_data;
773};
774
7965bd4d
JP
775void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
776 unsigned int to, struct skb_seq_state *st);
777unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
778 struct skb_seq_state *st);
779void skb_abort_seq_read(struct skb_seq_state *st);
677e90ed 780
7965bd4d
JP
781unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
782 unsigned int to, struct ts_config *config,
783 struct ts_state *state);
3fc7e8a6 784
09323cc4
TH
785/*
786 * Packet hash types specify the type of hash in skb_set_hash.
787 *
788 * Hash types refer to the protocol layer addresses which are used to
789 * construct a packet's hash. The hashes are used to differentiate or identify
790 * flows of the protocol layer for the hash type. Hash types are either
791 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
792 *
793 * Properties of hashes:
794 *
795 * 1) Two packets in different flows have different hash values
796 * 2) Two packets in the same flow should have the same hash value
797 *
798 * A hash at a higher layer is considered to be more specific. A driver should
799 * set the most specific hash possible.
800 *
801 * A driver cannot indicate a more specific hash than the layer at which a hash
802 * was computed. For instance an L3 hash cannot be set as an L4 hash.
803 *
804 * A driver may indicate a hash level which is less specific than the
805 * actual layer the hash was computed on. For instance, a hash computed
806 * at L4 may be considered an L3 hash. This should only be done if the
807 * driver can't unambiguously determine that the HW computed the hash at
808 * the higher layer. Note that the "should" in the second property above
809 * permits this.
810 */
811enum pkt_hash_types {
812 PKT_HASH_TYPE_NONE, /* Undefined type */
813 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
814 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
815 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
816};
817
818static inline void
819skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
820{
61b905da
TH
821 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
822 skb->hash = hash;
09323cc4
TH
823}
824
3958afa1
TH
825void __skb_get_hash(struct sk_buff *skb);
826static inline __u32 skb_get_hash(struct sk_buff *skb)
bfb564e7 827{
61b905da 828 if (!skb->l4_hash)
3958afa1 829 __skb_get_hash(skb);
bfb564e7 830
61b905da 831 return skb->hash;
bfb564e7
KK
832}
833
57bdf7f4
TH
834static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
835{
61b905da 836 return skb->hash;
57bdf7f4
TH
837}
838
7539fadc
TH
839static inline void skb_clear_hash(struct sk_buff *skb)
840{
61b905da
TH
841 skb->hash = 0;
842 skb->l4_hash = 0;
7539fadc
TH
843}
844
845static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
846{
61b905da 847 if (!skb->l4_hash)
7539fadc
TH
848 skb_clear_hash(skb);
849}
850
3df7a74e
TH
851static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
852{
61b905da
TH
853 to->hash = from->hash;
854 to->l4_hash = from->l4_hash;
3df7a74e
TH
855};
856
4305b541
ACM
857#ifdef NET_SKBUFF_DATA_USES_OFFSET
858static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
859{
860 return skb->head + skb->end;
861}
ec47ea82
AD
862
863static inline unsigned int skb_end_offset(const struct sk_buff *skb)
864{
865 return skb->end;
866}
4305b541
ACM
867#else
868static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
869{
870 return skb->end;
871}
ec47ea82
AD
872
873static inline unsigned int skb_end_offset(const struct sk_buff *skb)
874{
875 return skb->end - skb->head;
876}
4305b541
ACM
877#endif
878
1da177e4 879/* Internal */
4305b541 880#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1da177e4 881
ac45f602
PO
882static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
883{
884 return &skb_shinfo(skb)->hwtstamps;
885}
886
1da177e4
LT
887/**
888 * skb_queue_empty - check if a queue is empty
889 * @list: queue head
890 *
891 * Returns true if the queue is empty, false otherwise.
892 */
893static inline int skb_queue_empty(const struct sk_buff_head *list)
894{
fd44b93c 895 return list->next == (const struct sk_buff *) list;
1da177e4
LT
896}
897
fc7ebb21
DM
898/**
899 * skb_queue_is_last - check if skb is the last entry in the queue
900 * @list: queue head
901 * @skb: buffer
902 *
903 * Returns true if @skb is the last buffer on the list.
904 */
905static inline bool skb_queue_is_last(const struct sk_buff_head *list,
906 const struct sk_buff *skb)
907{
fd44b93c 908 return skb->next == (const struct sk_buff *) list;
fc7ebb21
DM
909}
910
832d11c5
IJ
911/**
912 * skb_queue_is_first - check if skb is the first entry in the queue
913 * @list: queue head
914 * @skb: buffer
915 *
916 * Returns true if @skb is the first buffer on the list.
917 */
918static inline bool skb_queue_is_first(const struct sk_buff_head *list,
919 const struct sk_buff *skb)
920{
fd44b93c 921 return skb->prev == (const struct sk_buff *) list;
832d11c5
IJ
922}
923
249c8b42
DM
924/**
925 * skb_queue_next - return the next packet in the queue
926 * @list: queue head
927 * @skb: current buffer
928 *
929 * Return the next packet in @list after @skb. It is only valid to
930 * call this if skb_queue_is_last() evaluates to false.
931 */
932static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
933 const struct sk_buff *skb)
934{
935 /* This BUG_ON may seem severe, but if we just return then we
936 * are going to dereference garbage.
937 */
938 BUG_ON(skb_queue_is_last(list, skb));
939 return skb->next;
940}
941
832d11c5
IJ
942/**
943 * skb_queue_prev - return the prev packet in the queue
944 * @list: queue head
945 * @skb: current buffer
946 *
947 * Return the prev packet in @list before @skb. It is only valid to
948 * call this if skb_queue_is_first() evaluates to false.
949 */
950static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
951 const struct sk_buff *skb)
952{
953 /* This BUG_ON may seem severe, but if we just return then we
954 * are going to dereference garbage.
955 */
956 BUG_ON(skb_queue_is_first(list, skb));
957 return skb->prev;
958}
959
1da177e4
LT
960/**
961 * skb_get - reference buffer
962 * @skb: buffer to reference
963 *
964 * Makes another reference to a socket buffer and returns a pointer
965 * to the buffer.
966 */
967static inline struct sk_buff *skb_get(struct sk_buff *skb)
968{
969 atomic_inc(&skb->users);
970 return skb;
971}
972
973/*
974 * If users == 1, we are the only owner and are can avoid redundant
975 * atomic change.
976 */
977
1da177e4
LT
978/**
979 * skb_cloned - is the buffer a clone
980 * @skb: buffer to check
981 *
982 * Returns true if the buffer was generated with skb_clone() and is
983 * one of multiple shared copies of the buffer. Cloned buffers are
984 * shared data so must not be written to under normal circumstances.
985 */
986static inline int skb_cloned(const struct sk_buff *skb)
987{
988 return skb->cloned &&
989 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
990}
991
14bbd6a5
PS
992static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
993{
994 might_sleep_if(pri & __GFP_WAIT);
995
996 if (skb_cloned(skb))
997 return pskb_expand_head(skb, 0, 0, pri);
998
999 return 0;
1000}
1001
1da177e4
LT
1002/**
1003 * skb_header_cloned - is the header a clone
1004 * @skb: buffer to check
1005 *
1006 * Returns true if modifying the header part of the buffer requires
1007 * the data to be copied.
1008 */
1009static inline int skb_header_cloned(const struct sk_buff *skb)
1010{
1011 int dataref;
1012
1013 if (!skb->cloned)
1014 return 0;
1015
1016 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1017 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1018 return dataref != 1;
1019}
1020
1021/**
1022 * skb_header_release - release reference to header
1023 * @skb: buffer to operate on
1024 *
1025 * Drop a reference to the header part of the buffer. This is done
1026 * by acquiring a payload reference. You must not read from the header
1027 * part of skb->data after this.
1028 */
1029static inline void skb_header_release(struct sk_buff *skb)
1030{
1031 BUG_ON(skb->nohdr);
1032 skb->nohdr = 1;
1033 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1034}
1035
1036/**
1037 * skb_shared - is the buffer shared
1038 * @skb: buffer to check
1039 *
1040 * Returns true if more than one person has a reference to this
1041 * buffer.
1042 */
1043static inline int skb_shared(const struct sk_buff *skb)
1044{
1045 return atomic_read(&skb->users) != 1;
1046}
1047
1048/**
1049 * skb_share_check - check if buffer is shared and if so clone it
1050 * @skb: buffer to check
1051 * @pri: priority for memory allocation
1052 *
1053 * If the buffer is shared the buffer is cloned and the old copy
1054 * drops a reference. A new clone with a single reference is returned.
1055 * If the buffer is not shared the original buffer is returned. When
1056 * being called from interrupt status or with spinlocks held pri must
1057 * be GFP_ATOMIC.
1058 *
1059 * NULL is returned on a memory allocation failure.
1060 */
47061bc4 1061static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1da177e4
LT
1062{
1063 might_sleep_if(pri & __GFP_WAIT);
1064 if (skb_shared(skb)) {
1065 struct sk_buff *nskb = skb_clone(skb, pri);
47061bc4
ED
1066
1067 if (likely(nskb))
1068 consume_skb(skb);
1069 else
1070 kfree_skb(skb);
1da177e4
LT
1071 skb = nskb;
1072 }
1073 return skb;
1074}
1075
1076/*
1077 * Copy shared buffers into a new sk_buff. We effectively do COW on
1078 * packets to handle cases where we have a local reader and forward
1079 * and a couple of other messy ones. The normal one is tcpdumping
1080 * a packet thats being forwarded.
1081 */
1082
1083/**
1084 * skb_unshare - make a copy of a shared buffer
1085 * @skb: buffer to check
1086 * @pri: priority for memory allocation
1087 *
1088 * If the socket buffer is a clone then this function creates a new
1089 * copy of the data, drops a reference count on the old copy and returns
1090 * the new copy with the reference count at 1. If the buffer is not a clone
1091 * the original buffer is returned. When called with a spinlock held or
1092 * from interrupt state @pri must be %GFP_ATOMIC
1093 *
1094 * %NULL is returned on a memory allocation failure.
1095 */
e2bf521d 1096static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
dd0fc66f 1097 gfp_t pri)
1da177e4
LT
1098{
1099 might_sleep_if(pri & __GFP_WAIT);
1100 if (skb_cloned(skb)) {
1101 struct sk_buff *nskb = skb_copy(skb, pri);
1102 kfree_skb(skb); /* Free our shared copy */
1103 skb = nskb;
1104 }
1105 return skb;
1106}
1107
1108/**
1a5778aa 1109 * skb_peek - peek at the head of an &sk_buff_head
1da177e4
LT
1110 * @list_: list to peek at
1111 *
1112 * Peek an &sk_buff. Unlike most other operations you _MUST_
1113 * be careful with this one. A peek leaves the buffer on the
1114 * list and someone else may run off with it. You must hold
1115 * the appropriate locks or have a private queue to do this.
1116 *
1117 * Returns %NULL for an empty list or a pointer to the head element.
1118 * The reference count is not incremented and the reference is therefore
1119 * volatile. Use with caution.
1120 */
05bdd2f1 1121static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1da177e4 1122{
18d07000
ED
1123 struct sk_buff *skb = list_->next;
1124
1125 if (skb == (struct sk_buff *)list_)
1126 skb = NULL;
1127 return skb;
1da177e4
LT
1128}
1129
da5ef6e5
PE
1130/**
1131 * skb_peek_next - peek skb following the given one from a queue
1132 * @skb: skb to start from
1133 * @list_: list to peek at
1134 *
1135 * Returns %NULL when the end of the list is met or a pointer to the
1136 * next element. The reference count is not incremented and the
1137 * reference is therefore volatile. Use with caution.
1138 */
1139static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1140 const struct sk_buff_head *list_)
1141{
1142 struct sk_buff *next = skb->next;
18d07000 1143
da5ef6e5
PE
1144 if (next == (struct sk_buff *)list_)
1145 next = NULL;
1146 return next;
1147}
1148
1da177e4 1149/**
1a5778aa 1150 * skb_peek_tail - peek at the tail of an &sk_buff_head
1da177e4
LT
1151 * @list_: list to peek at
1152 *
1153 * Peek an &sk_buff. Unlike most other operations you _MUST_
1154 * be careful with this one. A peek leaves the buffer on the
1155 * list and someone else may run off with it. You must hold
1156 * the appropriate locks or have a private queue to do this.
1157 *
1158 * Returns %NULL for an empty list or a pointer to the tail element.
1159 * The reference count is not incremented and the reference is therefore
1160 * volatile. Use with caution.
1161 */
05bdd2f1 1162static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1da177e4 1163{
18d07000
ED
1164 struct sk_buff *skb = list_->prev;
1165
1166 if (skb == (struct sk_buff *)list_)
1167 skb = NULL;
1168 return skb;
1169
1da177e4
LT
1170}
1171
1172/**
1173 * skb_queue_len - get queue length
1174 * @list_: list to measure
1175 *
1176 * Return the length of an &sk_buff queue.
1177 */
1178static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1179{
1180 return list_->qlen;
1181}
1182
67fed459
DM
1183/**
1184 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1185 * @list: queue to initialize
1186 *
1187 * This initializes only the list and queue length aspects of
1188 * an sk_buff_head object. This allows to initialize the list
1189 * aspects of an sk_buff_head without reinitializing things like
1190 * the spinlock. It can also be used for on-stack sk_buff_head
1191 * objects where the spinlock is known to not be used.
1192 */
1193static inline void __skb_queue_head_init(struct sk_buff_head *list)
1194{
1195 list->prev = list->next = (struct sk_buff *)list;
1196 list->qlen = 0;
1197}
1198
76f10ad0
AV
1199/*
1200 * This function creates a split out lock class for each invocation;
1201 * this is needed for now since a whole lot of users of the skb-queue
1202 * infrastructure in drivers have different locking usage (in hardirq)
1203 * than the networking core (in softirq only). In the long run either the
1204 * network layer or drivers should need annotation to consolidate the
1205 * main types of usage into 3 classes.
1206 */
1da177e4
LT
1207static inline void skb_queue_head_init(struct sk_buff_head *list)
1208{
1209 spin_lock_init(&list->lock);
67fed459 1210 __skb_queue_head_init(list);
1da177e4
LT
1211}
1212
c2ecba71
PE
1213static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1214 struct lock_class_key *class)
1215{
1216 skb_queue_head_init(list);
1217 lockdep_set_class(&list->lock, class);
1218}
1219
1da177e4 1220/*
bf299275 1221 * Insert an sk_buff on a list.
1da177e4
LT
1222 *
1223 * The "__skb_xxxx()" functions are the non-atomic ones that
1224 * can only be called with interrupts disabled.
1225 */
7965bd4d
JP
1226void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1227 struct sk_buff_head *list);
bf299275
GR
1228static inline void __skb_insert(struct sk_buff *newsk,
1229 struct sk_buff *prev, struct sk_buff *next,
1230 struct sk_buff_head *list)
1231{
1232 newsk->next = next;
1233 newsk->prev = prev;
1234 next->prev = prev->next = newsk;
1235 list->qlen++;
1236}
1da177e4 1237
67fed459
DM
1238static inline void __skb_queue_splice(const struct sk_buff_head *list,
1239 struct sk_buff *prev,
1240 struct sk_buff *next)
1241{
1242 struct sk_buff *first = list->next;
1243 struct sk_buff *last = list->prev;
1244
1245 first->prev = prev;
1246 prev->next = first;
1247
1248 last->next = next;
1249 next->prev = last;
1250}
1251
1252/**
1253 * skb_queue_splice - join two skb lists, this is designed for stacks
1254 * @list: the new list to add
1255 * @head: the place to add it in the first list
1256 */
1257static inline void skb_queue_splice(const struct sk_buff_head *list,
1258 struct sk_buff_head *head)
1259{
1260 if (!skb_queue_empty(list)) {
1261 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1d4a31dd 1262 head->qlen += list->qlen;
67fed459
DM
1263 }
1264}
1265
1266/**
d9619496 1267 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
67fed459
DM
1268 * @list: the new list to add
1269 * @head: the place to add it in the first list
1270 *
1271 * The list at @list is reinitialised
1272 */
1273static inline void skb_queue_splice_init(struct sk_buff_head *list,
1274 struct sk_buff_head *head)
1275{
1276 if (!skb_queue_empty(list)) {
1277 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1d4a31dd 1278 head->qlen += list->qlen;
67fed459
DM
1279 __skb_queue_head_init(list);
1280 }
1281}
1282
1283/**
1284 * skb_queue_splice_tail - join two skb lists, each list being a queue
1285 * @list: the new list to add
1286 * @head: the place to add it in the first list
1287 */
1288static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1289 struct sk_buff_head *head)
1290{
1291 if (!skb_queue_empty(list)) {
1292 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1d4a31dd 1293 head->qlen += list->qlen;
67fed459
DM
1294 }
1295}
1296
1297/**
d9619496 1298 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
67fed459
DM
1299 * @list: the new list to add
1300 * @head: the place to add it in the first list
1301 *
1302 * Each of the lists is a queue.
1303 * The list at @list is reinitialised
1304 */
1305static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1306 struct sk_buff_head *head)
1307{
1308 if (!skb_queue_empty(list)) {
1309 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1d4a31dd 1310 head->qlen += list->qlen;
67fed459
DM
1311 __skb_queue_head_init(list);
1312 }
1313}
1314
1da177e4 1315/**
300ce174 1316 * __skb_queue_after - queue a buffer at the list head
1da177e4 1317 * @list: list to use
300ce174 1318 * @prev: place after this buffer
1da177e4
LT
1319 * @newsk: buffer to queue
1320 *
300ce174 1321 * Queue a buffer int the middle of a list. This function takes no locks
1da177e4
LT
1322 * and you must therefore hold required locks before calling it.
1323 *
1324 * A buffer cannot be placed on two lists at the same time.
1325 */
300ce174
SH
1326static inline void __skb_queue_after(struct sk_buff_head *list,
1327 struct sk_buff *prev,
1328 struct sk_buff *newsk)
1da177e4 1329{
bf299275 1330 __skb_insert(newsk, prev, prev->next, list);
1da177e4
LT
1331}
1332
7965bd4d
JP
1333void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1334 struct sk_buff_head *list);
7de6c033 1335
f5572855
GR
1336static inline void __skb_queue_before(struct sk_buff_head *list,
1337 struct sk_buff *next,
1338 struct sk_buff *newsk)
1339{
1340 __skb_insert(newsk, next->prev, next, list);
1341}
1342
300ce174
SH
1343/**
1344 * __skb_queue_head - queue a buffer at the list head
1345 * @list: list to use
1346 * @newsk: buffer to queue
1347 *
1348 * Queue a buffer at the start of a list. This function takes no locks
1349 * and you must therefore hold required locks before calling it.
1350 *
1351 * A buffer cannot be placed on two lists at the same time.
1352 */
7965bd4d 1353void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
300ce174
SH
1354static inline void __skb_queue_head(struct sk_buff_head *list,
1355 struct sk_buff *newsk)
1356{
1357 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1358}
1359
1da177e4
LT
1360/**
1361 * __skb_queue_tail - queue a buffer at the list tail
1362 * @list: list to use
1363 * @newsk: buffer to queue
1364 *
1365 * Queue a buffer at the end of a list. This function takes no locks
1366 * and you must therefore hold required locks before calling it.
1367 *
1368 * A buffer cannot be placed on two lists at the same time.
1369 */
7965bd4d 1370void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1da177e4
LT
1371static inline void __skb_queue_tail(struct sk_buff_head *list,
1372 struct sk_buff *newsk)
1373{
f5572855 1374 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1da177e4
LT
1375}
1376
1da177e4
LT
1377/*
1378 * remove sk_buff from list. _Must_ be called atomically, and with
1379 * the list known..
1380 */
7965bd4d 1381void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1da177e4
LT
1382static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1383{
1384 struct sk_buff *next, *prev;
1385
1386 list->qlen--;
1387 next = skb->next;
1388 prev = skb->prev;
1389 skb->next = skb->prev = NULL;
1da177e4
LT
1390 next->prev = prev;
1391 prev->next = next;
1392}
1393
f525c06d
GR
1394/**
1395 * __skb_dequeue - remove from the head of the queue
1396 * @list: list to dequeue from
1397 *
1398 * Remove the head of the list. This function does not take any locks
1399 * so must be used with appropriate locks held only. The head item is
1400 * returned or %NULL if the list is empty.
1401 */
7965bd4d 1402struct sk_buff *skb_dequeue(struct sk_buff_head *list);
f525c06d
GR
1403static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1404{
1405 struct sk_buff *skb = skb_peek(list);
1406 if (skb)
1407 __skb_unlink(skb, list);
1408 return skb;
1409}
1da177e4
LT
1410
1411/**
1412 * __skb_dequeue_tail - remove from the tail of the queue
1413 * @list: list to dequeue from
1414 *
1415 * Remove the tail of the list. This function does not take any locks
1416 * so must be used with appropriate locks held only. The tail item is
1417 * returned or %NULL if the list is empty.
1418 */
7965bd4d 1419struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1da177e4
LT
1420static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1421{
1422 struct sk_buff *skb = skb_peek_tail(list);
1423 if (skb)
1424 __skb_unlink(skb, list);
1425 return skb;
1426}
1427
1428
bdcc0924 1429static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1da177e4
LT
1430{
1431 return skb->data_len;
1432}
1433
1434static inline unsigned int skb_headlen(const struct sk_buff *skb)
1435{
1436 return skb->len - skb->data_len;
1437}
1438
1439static inline int skb_pagelen(const struct sk_buff *skb)
1440{
1441 int i, len = 0;
1442
1443 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
9e903e08 1444 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1da177e4
LT
1445 return len + skb_headlen(skb);
1446}
1447
131ea667
IC
1448/**
1449 * __skb_fill_page_desc - initialise a paged fragment in an skb
1450 * @skb: buffer containing fragment to be initialised
1451 * @i: paged fragment index to initialise
1452 * @page: the page to use for this fragment
1453 * @off: the offset to the data with @page
1454 * @size: the length of the data
1455 *
1456 * Initialises the @i'th fragment of @skb to point to &size bytes at
1457 * offset @off within @page.
1458 *
1459 * Does not take any additional reference on the fragment.
1460 */
1461static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1462 struct page *page, int off, int size)
1da177e4
LT
1463{
1464 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1465
c48a11c7
MG
1466 /*
1467 * Propagate page->pfmemalloc to the skb if we can. The problem is
1468 * that not all callers have unique ownership of the page. If
1469 * pfmemalloc is set, we check the mapping as a mapping implies
1470 * page->index is set (index and pfmemalloc share space).
1471 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1472 * do not lose pfmemalloc information as the pages would not be
1473 * allocated using __GFP_MEMALLOC.
1474 */
a8605c60 1475 frag->page.p = page;
1da177e4 1476 frag->page_offset = off;
9e903e08 1477 skb_frag_size_set(frag, size);
cca7af38
PE
1478
1479 page = compound_head(page);
1480 if (page->pfmemalloc && !page->mapping)
1481 skb->pfmemalloc = true;
131ea667
IC
1482}
1483
1484/**
1485 * skb_fill_page_desc - initialise a paged fragment in an skb
1486 * @skb: buffer containing fragment to be initialised
1487 * @i: paged fragment index to initialise
1488 * @page: the page to use for this fragment
1489 * @off: the offset to the data with @page
1490 * @size: the length of the data
1491 *
1492 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
bc32383c 1493 * @skb to point to @size bytes at offset @off within @page. In
131ea667
IC
1494 * addition updates @skb such that @i is the last fragment.
1495 *
1496 * Does not take any additional reference on the fragment.
1497 */
1498static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1499 struct page *page, int off, int size)
1500{
1501 __skb_fill_page_desc(skb, i, page, off, size);
1da177e4
LT
1502 skb_shinfo(skb)->nr_frags = i + 1;
1503}
1504
7965bd4d
JP
1505void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1506 int size, unsigned int truesize);
654bed16 1507
f8e617e1
JW
1508void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1509 unsigned int truesize);
1510
1da177e4 1511#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
21dc3301 1512#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1da177e4
LT
1513#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1514
27a884dc
ACM
1515#ifdef NET_SKBUFF_DATA_USES_OFFSET
1516static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1517{
1518 return skb->head + skb->tail;
1519}
1520
1521static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1522{
1523 skb->tail = skb->data - skb->head;
1524}
1525
1526static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1527{
1528 skb_reset_tail_pointer(skb);
1529 skb->tail += offset;
1530}
7cc46190 1531
27a884dc
ACM
1532#else /* NET_SKBUFF_DATA_USES_OFFSET */
1533static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1534{
1535 return skb->tail;
1536}
1537
1538static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1539{
1540 skb->tail = skb->data;
1541}
1542
1543static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1544{
1545 skb->tail = skb->data + offset;
1546}
4305b541 1547
27a884dc
ACM
1548#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1549
1da177e4
LT
1550/*
1551 * Add data to an sk_buff
1552 */
0c7ddf36 1553unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
7965bd4d 1554unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
1da177e4
LT
1555static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1556{
27a884dc 1557 unsigned char *tmp = skb_tail_pointer(skb);
1da177e4
LT
1558 SKB_LINEAR_ASSERT(skb);
1559 skb->tail += len;
1560 skb->len += len;
1561 return tmp;
1562}
1563
7965bd4d 1564unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
1da177e4
LT
1565static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1566{
1567 skb->data -= len;
1568 skb->len += len;
1569 return skb->data;
1570}
1571
7965bd4d 1572unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
1da177e4
LT
1573static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1574{
1575 skb->len -= len;
1576 BUG_ON(skb->len < skb->data_len);
1577 return skb->data += len;
1578}
1579
47d29646
DM
1580static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1581{
1582 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1583}
1584
7965bd4d 1585unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
1da177e4
LT
1586
1587static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1588{
1589 if (len > skb_headlen(skb) &&
987c402a 1590 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1da177e4
LT
1591 return NULL;
1592 skb->len -= len;
1593 return skb->data += len;
1594}
1595
1596static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1597{
1598 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1599}
1600
1601static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1602{
1603 if (likely(len <= skb_headlen(skb)))
1604 return 1;
1605 if (unlikely(len > skb->len))
1606 return 0;
987c402a 1607 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1da177e4
LT
1608}
1609
1610/**
1611 * skb_headroom - bytes at buffer head
1612 * @skb: buffer to check
1613 *
1614 * Return the number of bytes of free space at the head of an &sk_buff.
1615 */
c2636b4d 1616static inline unsigned int skb_headroom(const struct sk_buff *skb)
1da177e4
LT
1617{
1618 return skb->data - skb->head;
1619}
1620
1621/**
1622 * skb_tailroom - bytes at buffer end
1623 * @skb: buffer to check
1624 *
1625 * Return the number of bytes of free space at the tail of an sk_buff
1626 */
1627static inline int skb_tailroom(const struct sk_buff *skb)
1628{
4305b541 1629 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1da177e4
LT
1630}
1631
a21d4572
ED
1632/**
1633 * skb_availroom - bytes at buffer end
1634 * @skb: buffer to check
1635 *
1636 * Return the number of bytes of free space at the tail of an sk_buff
1637 * allocated by sk_stream_alloc()
1638 */
1639static inline int skb_availroom(const struct sk_buff *skb)
1640{
16fad69c
ED
1641 if (skb_is_nonlinear(skb))
1642 return 0;
1643
1644 return skb->end - skb->tail - skb->reserved_tailroom;
a21d4572
ED
1645}
1646
1da177e4
LT
1647/**
1648 * skb_reserve - adjust headroom
1649 * @skb: buffer to alter
1650 * @len: bytes to move
1651 *
1652 * Increase the headroom of an empty &sk_buff by reducing the tail
1653 * room. This is only allowed for an empty buffer.
1654 */
8243126c 1655static inline void skb_reserve(struct sk_buff *skb, int len)
1da177e4
LT
1656{
1657 skb->data += len;
1658 skb->tail += len;
1659}
1660
6a674e9c
JG
1661static inline void skb_reset_inner_headers(struct sk_buff *skb)
1662{
aefbd2b3 1663 skb->inner_mac_header = skb->mac_header;
6a674e9c
JG
1664 skb->inner_network_header = skb->network_header;
1665 skb->inner_transport_header = skb->transport_header;
1666}
1667
0b5c9db1
JP
1668static inline void skb_reset_mac_len(struct sk_buff *skb)
1669{
1670 skb->mac_len = skb->network_header - skb->mac_header;
1671}
1672
6a674e9c
JG
1673static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1674 *skb)
1675{
1676 return skb->head + skb->inner_transport_header;
1677}
1678
1679static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1680{
1681 skb->inner_transport_header = skb->data - skb->head;
1682}
1683
1684static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1685 const int offset)
1686{
1687 skb_reset_inner_transport_header(skb);
1688 skb->inner_transport_header += offset;
1689}
1690
1691static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1692{
1693 return skb->head + skb->inner_network_header;
1694}
1695
1696static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1697{
1698 skb->inner_network_header = skb->data - skb->head;
1699}
1700
1701static inline void skb_set_inner_network_header(struct sk_buff *skb,
1702 const int offset)
1703{
1704 skb_reset_inner_network_header(skb);
1705 skb->inner_network_header += offset;
1706}
1707
aefbd2b3
PS
1708static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1709{
1710 return skb->head + skb->inner_mac_header;
1711}
1712
1713static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1714{
1715 skb->inner_mac_header = skb->data - skb->head;
1716}
1717
1718static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1719 const int offset)
1720{
1721 skb_reset_inner_mac_header(skb);
1722 skb->inner_mac_header += offset;
1723}
fda55eca
ED
1724static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1725{
35d04610 1726 return skb->transport_header != (typeof(skb->transport_header))~0U;
fda55eca
ED
1727}
1728
9c70220b
ACM
1729static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1730{
2e07fa9c 1731 return skb->head + skb->transport_header;
9c70220b
ACM
1732}
1733
badff6d0
ACM
1734static inline void skb_reset_transport_header(struct sk_buff *skb)
1735{
2e07fa9c 1736 skb->transport_header = skb->data - skb->head;
badff6d0
ACM
1737}
1738
967b05f6
ACM
1739static inline void skb_set_transport_header(struct sk_buff *skb,
1740 const int offset)
1741{
2e07fa9c
ACM
1742 skb_reset_transport_header(skb);
1743 skb->transport_header += offset;
ea2ae17d
ACM
1744}
1745
d56f90a7
ACM
1746static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1747{
2e07fa9c 1748 return skb->head + skb->network_header;
d56f90a7
ACM
1749}
1750
c1d2bbe1
ACM
1751static inline void skb_reset_network_header(struct sk_buff *skb)
1752{
2e07fa9c 1753 skb->network_header = skb->data - skb->head;
c1d2bbe1
ACM
1754}
1755
c14d2450
ACM
1756static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1757{
2e07fa9c
ACM
1758 skb_reset_network_header(skb);
1759 skb->network_header += offset;
c14d2450
ACM
1760}
1761
2e07fa9c 1762static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
bbe735e4 1763{
2e07fa9c 1764 return skb->head + skb->mac_header;
bbe735e4
ACM
1765}
1766
2e07fa9c 1767static inline int skb_mac_header_was_set(const struct sk_buff *skb)
cfe1fc77 1768{
35d04610 1769 return skb->mac_header != (typeof(skb->mac_header))~0U;
2e07fa9c
ACM
1770}
1771
1772static inline void skb_reset_mac_header(struct sk_buff *skb)
1773{
1774 skb->mac_header = skb->data - skb->head;
1775}
1776
1777static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1778{
1779 skb_reset_mac_header(skb);
1780 skb->mac_header += offset;
1781}
1782
0e3da5bb
TT
1783static inline void skb_pop_mac_header(struct sk_buff *skb)
1784{
1785 skb->mac_header = skb->network_header;
1786}
1787
fbbdb8f0
YX
1788static inline void skb_probe_transport_header(struct sk_buff *skb,
1789 const int offset_hint)
1790{
1791 struct flow_keys keys;
1792
1793 if (skb_transport_header_was_set(skb))
1794 return;
1795 else if (skb_flow_dissect(skb, &keys))
1796 skb_set_transport_header(skb, keys.thoff);
1797 else
1798 skb_set_transport_header(skb, offset_hint);
1799}
1800
03606895
ED
1801static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1802{
1803 if (skb_mac_header_was_set(skb)) {
1804 const unsigned char *old_mac = skb_mac_header(skb);
1805
1806 skb_set_mac_header(skb, -skb->mac_len);
1807 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1808 }
1809}
1810
04fb451e
MM
1811static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1812{
1813 return skb->csum_start - skb_headroom(skb);
1814}
1815
2e07fa9c
ACM
1816static inline int skb_transport_offset(const struct sk_buff *skb)
1817{
1818 return skb_transport_header(skb) - skb->data;
1819}
1820
1821static inline u32 skb_network_header_len(const struct sk_buff *skb)
1822{
1823 return skb->transport_header - skb->network_header;
1824}
1825
6a674e9c
JG
1826static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1827{
1828 return skb->inner_transport_header - skb->inner_network_header;
1829}
1830
2e07fa9c
ACM
1831static inline int skb_network_offset(const struct sk_buff *skb)
1832{
1833 return skb_network_header(skb) - skb->data;
1834}
48d49d0c 1835
6a674e9c
JG
1836static inline int skb_inner_network_offset(const struct sk_buff *skb)
1837{
1838 return skb_inner_network_header(skb) - skb->data;
1839}
1840
f9599ce1
CG
1841static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1842{
1843 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1844}
1845
1da177e4
LT
1846/*
1847 * CPUs often take a performance hit when accessing unaligned memory
1848 * locations. The actual performance hit varies, it can be small if the
1849 * hardware handles it or large if we have to take an exception and fix it
1850 * in software.
1851 *
1852 * Since an ethernet header is 14 bytes network drivers often end up with
1853 * the IP header at an unaligned offset. The IP header can be aligned by
1854 * shifting the start of the packet by 2 bytes. Drivers should do this
1855 * with:
1856 *
8660c124 1857 * skb_reserve(skb, NET_IP_ALIGN);
1da177e4
LT
1858 *
1859 * The downside to this alignment of the IP header is that the DMA is now
1860 * unaligned. On some architectures the cost of an unaligned DMA is high
1861 * and this cost outweighs the gains made by aligning the IP header.
8660c124 1862 *
1da177e4
LT
1863 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1864 * to be overridden.
1865 */
1866#ifndef NET_IP_ALIGN
1867#define NET_IP_ALIGN 2
1868#endif
1869
025be81e
AB
1870/*
1871 * The networking layer reserves some headroom in skb data (via
1872 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1873 * the header has to grow. In the default case, if the header has to grow
d6301d3d 1874 * 32 bytes or less we avoid the reallocation.
025be81e
AB
1875 *
1876 * Unfortunately this headroom changes the DMA alignment of the resulting
1877 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1878 * on some architectures. An architecture can override this value,
1879 * perhaps setting it to a cacheline in size (since that will maintain
1880 * cacheline alignment of the DMA). It must be a power of 2.
1881 *
d6301d3d 1882 * Various parts of the networking layer expect at least 32 bytes of
025be81e 1883 * headroom, you should not reduce this.
5933dd2f
ED
1884 *
1885 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
1886 * to reduce average number of cache lines per packet.
1887 * get_rps_cpus() for example only access one 64 bytes aligned block :
18e8c134 1888 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
025be81e
AB
1889 */
1890#ifndef NET_SKB_PAD
5933dd2f 1891#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
025be81e
AB
1892#endif
1893
7965bd4d 1894int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1da177e4
LT
1895
1896static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1897{
c4264f27 1898 if (unlikely(skb_is_nonlinear(skb))) {
3cc0e873
HX
1899 WARN_ON(1);
1900 return;
1901 }
27a884dc
ACM
1902 skb->len = len;
1903 skb_set_tail_pointer(skb, len);
1da177e4
LT
1904}
1905
7965bd4d 1906void skb_trim(struct sk_buff *skb, unsigned int len);
1da177e4
LT
1907
1908static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1909{
3cc0e873
HX
1910 if (skb->data_len)
1911 return ___pskb_trim(skb, len);
1912 __skb_trim(skb, len);
1913 return 0;
1da177e4
LT
1914}
1915
1916static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1917{
1918 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1919}
1920
e9fa4f7b
HX
1921/**
1922 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1923 * @skb: buffer to alter
1924 * @len: new length
1925 *
1926 * This is identical to pskb_trim except that the caller knows that
1927 * the skb is not cloned so we should never get an error due to out-
1928 * of-memory.
1929 */
1930static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1931{
1932 int err = pskb_trim(skb, len);
1933 BUG_ON(err);
1934}
1935
1da177e4
LT
1936/**
1937 * skb_orphan - orphan a buffer
1938 * @skb: buffer to orphan
1939 *
1940 * If a buffer currently has an owner then we call the owner's
1941 * destructor function and make the @skb unowned. The buffer continues
1942 * to exist but is no longer charged to its former owner.
1943 */
1944static inline void skb_orphan(struct sk_buff *skb)
1945{
c34a7612 1946 if (skb->destructor) {
1da177e4 1947 skb->destructor(skb);
c34a7612
ED
1948 skb->destructor = NULL;
1949 skb->sk = NULL;
376c7311
ED
1950 } else {
1951 BUG_ON(skb->sk);
c34a7612 1952 }
1da177e4
LT
1953}
1954
a353e0ce
MT
1955/**
1956 * skb_orphan_frags - orphan the frags contained in a buffer
1957 * @skb: buffer to orphan frags from
1958 * @gfp_mask: allocation mask for replacement pages
1959 *
1960 * For each frag in the SKB which needs a destructor (i.e. has an
1961 * owner) create a copy of that frag and release the original
1962 * page by calling the destructor.
1963 */
1964static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
1965{
1966 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
1967 return 0;
1968 return skb_copy_ubufs(skb, gfp_mask);
1969}
1970
1da177e4
LT
1971/**
1972 * __skb_queue_purge - empty a list
1973 * @list: list to empty
1974 *
1975 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1976 * the list and one reference dropped. This function does not take the
1977 * list lock and the caller must hold the relevant locks to use it.
1978 */
7965bd4d 1979void skb_queue_purge(struct sk_buff_head *list);
1da177e4
LT
1980static inline void __skb_queue_purge(struct sk_buff_head *list)
1981{
1982 struct sk_buff *skb;
1983 while ((skb = __skb_dequeue(list)) != NULL)
1984 kfree_skb(skb);
1985}
1986
e5e67305
AD
1987#define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
1988#define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
1989#define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
1990
7965bd4d 1991void *netdev_alloc_frag(unsigned int fragsz);
1da177e4 1992
7965bd4d
JP
1993struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
1994 gfp_t gfp_mask);
8af27456
CH
1995
1996/**
1997 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
1998 * @dev: network device to receive on
1999 * @length: length to allocate
2000 *
2001 * Allocate a new &sk_buff and assign it a usage count of one. The
2002 * buffer has unspecified headroom built in. Users should allocate
2003 * the headroom they think they need without accounting for the
2004 * built in space. The built in space is used for optimisations.
2005 *
2006 * %NULL is returned if there is no free memory. Although this function
2007 * allocates memory it can be called from an interrupt.
2008 */
2009static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
6f532612 2010 unsigned int length)
8af27456
CH
2011{
2012 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2013}
2014
6f532612
ED
2015/* legacy helper around __netdev_alloc_skb() */
2016static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2017 gfp_t gfp_mask)
2018{
2019 return __netdev_alloc_skb(NULL, length, gfp_mask);
2020}
2021
2022/* legacy helper around netdev_alloc_skb() */
2023static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2024{
2025 return netdev_alloc_skb(NULL, length);
2026}
2027
2028
4915a0de
ED
2029static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2030 unsigned int length, gfp_t gfp)
61321bbd 2031{
4915a0de 2032 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
61321bbd
ED
2033
2034 if (NET_IP_ALIGN && skb)
2035 skb_reserve(skb, NET_IP_ALIGN);
2036 return skb;
2037}
2038
4915a0de
ED
2039static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2040 unsigned int length)
2041{
2042 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2043}
2044
bc6fc9fa
FF
2045/**
2046 * __skb_alloc_pages - allocate pages for ps-rx on a skb and preserve pfmemalloc data
0614002b
MG
2047 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2048 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2049 * @order: size of the allocation
2050 *
2051 * Allocate a new page.
2052 *
2053 * %NULL is returned if there is no free memory.
2054*/
2055static inline struct page *__skb_alloc_pages(gfp_t gfp_mask,
2056 struct sk_buff *skb,
2057 unsigned int order)
2058{
2059 struct page *page;
2060
2061 gfp_mask |= __GFP_COLD;
2062
2063 if (!(gfp_mask & __GFP_NOMEMALLOC))
2064 gfp_mask |= __GFP_MEMALLOC;
2065
2066 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2067 if (skb && page && page->pfmemalloc)
2068 skb->pfmemalloc = true;
2069
2070 return page;
2071}
2072
2073/**
2074 * __skb_alloc_page - allocate a page for ps-rx for a given skb and preserve pfmemalloc data
2075 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2076 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2077 *
2078 * Allocate a new page.
2079 *
2080 * %NULL is returned if there is no free memory.
2081 */
2082static inline struct page *__skb_alloc_page(gfp_t gfp_mask,
2083 struct sk_buff *skb)
2084{
2085 return __skb_alloc_pages(gfp_mask, skb, 0);
2086}
2087
2088/**
2089 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2090 * @page: The page that was allocated from skb_alloc_page
2091 * @skb: The skb that may need pfmemalloc set
2092 */
2093static inline void skb_propagate_pfmemalloc(struct page *page,
2094 struct sk_buff *skb)
2095{
2096 if (page && page->pfmemalloc)
2097 skb->pfmemalloc = true;
2098}
2099
131ea667 2100/**
e227867f 2101 * skb_frag_page - retrieve the page referred to by a paged fragment
131ea667
IC
2102 * @frag: the paged fragment
2103 *
2104 * Returns the &struct page associated with @frag.
2105 */
2106static inline struct page *skb_frag_page(const skb_frag_t *frag)
2107{
a8605c60 2108 return frag->page.p;
131ea667
IC
2109}
2110
2111/**
2112 * __skb_frag_ref - take an addition reference on a paged fragment.
2113 * @frag: the paged fragment
2114 *
2115 * Takes an additional reference on the paged fragment @frag.
2116 */
2117static inline void __skb_frag_ref(skb_frag_t *frag)
2118{
2119 get_page(skb_frag_page(frag));
2120}
2121
2122/**
2123 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2124 * @skb: the buffer
2125 * @f: the fragment offset.
2126 *
2127 * Takes an additional reference on the @f'th paged fragment of @skb.
2128 */
2129static inline void skb_frag_ref(struct sk_buff *skb, int f)
2130{
2131 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2132}
2133
2134/**
2135 * __skb_frag_unref - release a reference on a paged fragment.
2136 * @frag: the paged fragment
2137 *
2138 * Releases a reference on the paged fragment @frag.
2139 */
2140static inline void __skb_frag_unref(skb_frag_t *frag)
2141{
2142 put_page(skb_frag_page(frag));
2143}
2144
2145/**
2146 * skb_frag_unref - release a reference on a paged fragment of an skb.
2147 * @skb: the buffer
2148 * @f: the fragment offset
2149 *
2150 * Releases a reference on the @f'th paged fragment of @skb.
2151 */
2152static inline void skb_frag_unref(struct sk_buff *skb, int f)
2153{
2154 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2155}
2156
2157/**
2158 * skb_frag_address - gets the address of the data contained in a paged fragment
2159 * @frag: the paged fragment buffer
2160 *
2161 * Returns the address of the data within @frag. The page must already
2162 * be mapped.
2163 */
2164static inline void *skb_frag_address(const skb_frag_t *frag)
2165{
2166 return page_address(skb_frag_page(frag)) + frag->page_offset;
2167}
2168
2169/**
2170 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2171 * @frag: the paged fragment buffer
2172 *
2173 * Returns the address of the data within @frag. Checks that the page
2174 * is mapped and returns %NULL otherwise.
2175 */
2176static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2177{
2178 void *ptr = page_address(skb_frag_page(frag));
2179 if (unlikely(!ptr))
2180 return NULL;
2181
2182 return ptr + frag->page_offset;
2183}
2184
2185/**
2186 * __skb_frag_set_page - sets the page contained in a paged fragment
2187 * @frag: the paged fragment
2188 * @page: the page to set
2189 *
2190 * Sets the fragment @frag to contain @page.
2191 */
2192static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2193{
a8605c60 2194 frag->page.p = page;
131ea667
IC
2195}
2196
2197/**
2198 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2199 * @skb: the buffer
2200 * @f: the fragment offset
2201 * @page: the page to set
2202 *
2203 * Sets the @f'th fragment of @skb to contain @page.
2204 */
2205static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2206 struct page *page)
2207{
2208 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2209}
2210
400dfd3a
ED
2211bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2212
131ea667
IC
2213/**
2214 * skb_frag_dma_map - maps a paged fragment via the DMA API
f83347df 2215 * @dev: the device to map the fragment to
131ea667
IC
2216 * @frag: the paged fragment to map
2217 * @offset: the offset within the fragment (starting at the
2218 * fragment's own offset)
2219 * @size: the number of bytes to map
f83347df 2220 * @dir: the direction of the mapping (%PCI_DMA_*)
131ea667
IC
2221 *
2222 * Maps the page associated with @frag to @device.
2223 */
2224static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2225 const skb_frag_t *frag,
2226 size_t offset, size_t size,
2227 enum dma_data_direction dir)
2228{
2229 return dma_map_page(dev, skb_frag_page(frag),
2230 frag->page_offset + offset, size, dir);
2231}
2232
117632e6
ED
2233static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2234 gfp_t gfp_mask)
2235{
2236 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2237}
2238
334a8132
PM
2239/**
2240 * skb_clone_writable - is the header of a clone writable
2241 * @skb: buffer to check
2242 * @len: length up to which to write
2243 *
2244 * Returns true if modifying the header part of the cloned buffer
2245 * does not requires the data to be copied.
2246 */
05bdd2f1 2247static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
334a8132
PM
2248{
2249 return !skb_header_cloned(skb) &&
2250 skb_headroom(skb) + len <= skb->hdr_len;
2251}
2252
d9cc2048
HX
2253static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2254 int cloned)
2255{
2256 int delta = 0;
2257
d9cc2048
HX
2258 if (headroom > skb_headroom(skb))
2259 delta = headroom - skb_headroom(skb);
2260
2261 if (delta || cloned)
2262 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2263 GFP_ATOMIC);
2264 return 0;
2265}
2266
1da177e4
LT
2267/**
2268 * skb_cow - copy header of skb when it is required
2269 * @skb: buffer to cow
2270 * @headroom: needed headroom
2271 *
2272 * If the skb passed lacks sufficient headroom or its data part
2273 * is shared, data is reallocated. If reallocation fails, an error
2274 * is returned and original skb is not changed.
2275 *
2276 * The result is skb with writable area skb->head...skb->tail
2277 * and at least @headroom of space at head.
2278 */
2279static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2280{
d9cc2048
HX
2281 return __skb_cow(skb, headroom, skb_cloned(skb));
2282}
1da177e4 2283
d9cc2048
HX
2284/**
2285 * skb_cow_head - skb_cow but only making the head writable
2286 * @skb: buffer to cow
2287 * @headroom: needed headroom
2288 *
2289 * This function is identical to skb_cow except that we replace the
2290 * skb_cloned check by skb_header_cloned. It should be used when
2291 * you only need to push on some header and do not need to modify
2292 * the data.
2293 */
2294static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2295{
2296 return __skb_cow(skb, headroom, skb_header_cloned(skb));
1da177e4
LT
2297}
2298
2299/**
2300 * skb_padto - pad an skbuff up to a minimal size
2301 * @skb: buffer to pad
2302 * @len: minimal length
2303 *
2304 * Pads up a buffer to ensure the trailing bytes exist and are
2305 * blanked. If the buffer already contains sufficient data it
5b057c6b
HX
2306 * is untouched. Otherwise it is extended. Returns zero on
2307 * success. The skb is freed on error.
1da177e4
LT
2308 */
2309
5b057c6b 2310static inline int skb_padto(struct sk_buff *skb, unsigned int len)
1da177e4
LT
2311{
2312 unsigned int size = skb->len;
2313 if (likely(size >= len))
5b057c6b 2314 return 0;
987c402a 2315 return skb_pad(skb, len - size);
1da177e4
LT
2316}
2317
2318static inline int skb_add_data(struct sk_buff *skb,
2319 char __user *from, int copy)
2320{
2321 const int off = skb->len;
2322
2323 if (skb->ip_summed == CHECKSUM_NONE) {
2324 int err = 0;
5084205f 2325 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
1da177e4
LT
2326 copy, 0, &err);
2327 if (!err) {
2328 skb->csum = csum_block_add(skb->csum, csum, off);
2329 return 0;
2330 }
2331 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
2332 return 0;
2333
2334 __skb_trim(skb, off);
2335 return -EFAULT;
2336}
2337
38ba0a65
ED
2338static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2339 const struct page *page, int off)
1da177e4
LT
2340{
2341 if (i) {
9e903e08 2342 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1da177e4 2343
ea2ab693 2344 return page == skb_frag_page(frag) &&
9e903e08 2345 off == frag->page_offset + skb_frag_size(frag);
1da177e4 2346 }
38ba0a65 2347 return false;
1da177e4
LT
2348}
2349
364c6bad
HX
2350static inline int __skb_linearize(struct sk_buff *skb)
2351{
2352 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2353}
2354
1da177e4
LT
2355/**
2356 * skb_linearize - convert paged skb to linear one
2357 * @skb: buffer to linarize
1da177e4
LT
2358 *
2359 * If there is no free memory -ENOMEM is returned, otherwise zero
2360 * is returned and the old skb data released.
2361 */
364c6bad
HX
2362static inline int skb_linearize(struct sk_buff *skb)
2363{
2364 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2365}
2366
cef401de
ED
2367/**
2368 * skb_has_shared_frag - can any frag be overwritten
2369 * @skb: buffer to test
2370 *
2371 * Return true if the skb has at least one frag that might be modified
2372 * by an external entity (as in vmsplice()/sendfile())
2373 */
2374static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2375{
c9af6db4
PS
2376 return skb_is_nonlinear(skb) &&
2377 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
cef401de
ED
2378}
2379
364c6bad
HX
2380/**
2381 * skb_linearize_cow - make sure skb is linear and writable
2382 * @skb: buffer to process
2383 *
2384 * If there is no free memory -ENOMEM is returned, otherwise zero
2385 * is returned and the old skb data released.
2386 */
2387static inline int skb_linearize_cow(struct sk_buff *skb)
1da177e4 2388{
364c6bad
HX
2389 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2390 __skb_linearize(skb) : 0;
1da177e4
LT
2391}
2392
2393/**
2394 * skb_postpull_rcsum - update checksum for received skb after pull
2395 * @skb: buffer to update
2396 * @start: start of data before pull
2397 * @len: length of data pulled
2398 *
2399 * After doing a pull on a received packet, you need to call this to
84fa7933
PM
2400 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2401 * CHECKSUM_NONE so that it can be recomputed from scratch.
1da177e4
LT
2402 */
2403
2404static inline void skb_postpull_rcsum(struct sk_buff *skb,
cbb042f9 2405 const void *start, unsigned int len)
1da177e4 2406{
84fa7933 2407 if (skb->ip_summed == CHECKSUM_COMPLETE)
1da177e4
LT
2408 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2409}
2410
cbb042f9
HX
2411unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2412
7ce5a27f
DM
2413/**
2414 * pskb_trim_rcsum - trim received skb and update checksum
2415 * @skb: buffer to trim
2416 * @len: new length
2417 *
2418 * This is exactly the same as pskb_trim except that it ensures the
2419 * checksum of received packets are still valid after the operation.
2420 */
2421
2422static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2423{
2424 if (likely(len >= skb->len))
2425 return 0;
2426 if (skb->ip_summed == CHECKSUM_COMPLETE)
2427 skb->ip_summed = CHECKSUM_NONE;
2428 return __pskb_trim(skb, len);
2429}
2430
1da177e4
LT
2431#define skb_queue_walk(queue, skb) \
2432 for (skb = (queue)->next; \
a1e4891f 2433 skb != (struct sk_buff *)(queue); \
1da177e4
LT
2434 skb = skb->next)
2435
46f8914e
JC
2436#define skb_queue_walk_safe(queue, skb, tmp) \
2437 for (skb = (queue)->next, tmp = skb->next; \
2438 skb != (struct sk_buff *)(queue); \
2439 skb = tmp, tmp = skb->next)
2440
1164f52a 2441#define skb_queue_walk_from(queue, skb) \
a1e4891f 2442 for (; skb != (struct sk_buff *)(queue); \
1164f52a
DM
2443 skb = skb->next)
2444
2445#define skb_queue_walk_from_safe(queue, skb, tmp) \
2446 for (tmp = skb->next; \
2447 skb != (struct sk_buff *)(queue); \
2448 skb = tmp, tmp = skb->next)
2449
300ce174
SH
2450#define skb_queue_reverse_walk(queue, skb) \
2451 for (skb = (queue)->prev; \
a1e4891f 2452 skb != (struct sk_buff *)(queue); \
300ce174
SH
2453 skb = skb->prev)
2454
686a2955
DM
2455#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2456 for (skb = (queue)->prev, tmp = skb->prev; \
2457 skb != (struct sk_buff *)(queue); \
2458 skb = tmp, tmp = skb->prev)
2459
2460#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2461 for (tmp = skb->prev; \
2462 skb != (struct sk_buff *)(queue); \
2463 skb = tmp, tmp = skb->prev)
1da177e4 2464
21dc3301 2465static inline bool skb_has_frag_list(const struct sk_buff *skb)
ee039871
DM
2466{
2467 return skb_shinfo(skb)->frag_list != NULL;
2468}
2469
2470static inline void skb_frag_list_init(struct sk_buff *skb)
2471{
2472 skb_shinfo(skb)->frag_list = NULL;
2473}
2474
2475static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2476{
2477 frag->next = skb_shinfo(skb)->frag_list;
2478 skb_shinfo(skb)->frag_list = frag;
2479}
2480
2481#define skb_walk_frags(skb, iter) \
2482 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2483
7965bd4d
JP
2484struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2485 int *peeked, int *off, int *err);
2486struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2487 int *err);
2488unsigned int datagram_poll(struct file *file, struct socket *sock,
2489 struct poll_table_struct *wait);
2490int skb_copy_datagram_iovec(const struct sk_buff *from, int offset,
2491 struct iovec *to, int size);
2492int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, int hlen,
2493 struct iovec *iov);
2494int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
2495 const struct iovec *from, int from_offset,
2496 int len);
2497int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *frm,
2498 int offset, size_t count);
2499int skb_copy_datagram_const_iovec(const struct sk_buff *from, int offset,
2500 const struct iovec *to, int to_offset,
2501 int size);
2502void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2503void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2504int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
7965bd4d
JP
2505int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2506int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2507__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2508 int len, __wsum csum);
2509int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2510 struct pipe_inode_info *pipe, unsigned int len,
2511 unsigned int flags);
2512void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
af2806f8 2513unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
36d5fe6a
ZK
2514int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2515 int len, int hlen);
7965bd4d
JP
2516void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2517int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2518void skb_scrub_packet(struct sk_buff *skb, bool xnet);
de960aa9 2519unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
7965bd4d 2520struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
20380731 2521
2817a336
DB
2522struct skb_checksum_ops {
2523 __wsum (*update)(const void *mem, int len, __wsum wsum);
2524 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2525};
2526
2527__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2528 __wsum csum, const struct skb_checksum_ops *ops);
2529__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2530 __wsum csum);
2531
1da177e4
LT
2532static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2533 int len, void *buffer)
2534{
2535 int hlen = skb_headlen(skb);
2536
55820ee2 2537 if (hlen - offset >= len)
1da177e4
LT
2538 return skb->data + offset;
2539
2540 if (skb_copy_bits(skb, offset, buffer, len) < 0)
2541 return NULL;
2542
2543 return buffer;
2544}
2545
4262e5cc
DB
2546/**
2547 * skb_needs_linearize - check if we need to linearize a given skb
2548 * depending on the given device features.
2549 * @skb: socket buffer to check
2550 * @features: net device features
2551 *
2552 * Returns true if either:
2553 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2554 * 2. skb is fragmented and the device does not support SG.
2555 */
2556static inline bool skb_needs_linearize(struct sk_buff *skb,
2557 netdev_features_t features)
2558{
2559 return skb_is_nonlinear(skb) &&
2560 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2561 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2562}
2563
d626f62b
ACM
2564static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2565 void *to,
2566 const unsigned int len)
2567{
2568 memcpy(to, skb->data, len);
2569}
2570
2571static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2572 const int offset, void *to,
2573 const unsigned int len)
2574{
2575 memcpy(to, skb->data + offset, len);
2576}
2577
27d7ff46
ACM
2578static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2579 const void *from,
2580 const unsigned int len)
2581{
2582 memcpy(skb->data, from, len);
2583}
2584
2585static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2586 const int offset,
2587 const void *from,
2588 const unsigned int len)
2589{
2590 memcpy(skb->data + offset, from, len);
2591}
2592
7965bd4d 2593void skb_init(void);
1da177e4 2594
ac45f602
PO
2595static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2596{
2597 return skb->tstamp;
2598}
2599
a61bbcf2
PM
2600/**
2601 * skb_get_timestamp - get timestamp from a skb
2602 * @skb: skb to get stamp from
2603 * @stamp: pointer to struct timeval to store stamp in
2604 *
2605 * Timestamps are stored in the skb as offsets to a base timestamp.
2606 * This function converts the offset back to a struct timeval and stores
2607 * it in stamp.
2608 */
ac45f602
PO
2609static inline void skb_get_timestamp(const struct sk_buff *skb,
2610 struct timeval *stamp)
a61bbcf2 2611{
b7aa0bf7 2612 *stamp = ktime_to_timeval(skb->tstamp);
a61bbcf2
PM
2613}
2614
ac45f602
PO
2615static inline void skb_get_timestampns(const struct sk_buff *skb,
2616 struct timespec *stamp)
2617{
2618 *stamp = ktime_to_timespec(skb->tstamp);
2619}
2620
b7aa0bf7 2621static inline void __net_timestamp(struct sk_buff *skb)
a61bbcf2 2622{
b7aa0bf7 2623 skb->tstamp = ktime_get_real();
a61bbcf2
PM
2624}
2625
164891aa
SH
2626static inline ktime_t net_timedelta(ktime_t t)
2627{
2628 return ktime_sub(ktime_get_real(), t);
2629}
2630
b9ce204f
IJ
2631static inline ktime_t net_invalid_timestamp(void)
2632{
2633 return ktime_set(0, 0);
2634}
a61bbcf2 2635
c1f19b51
RC
2636#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2637
7965bd4d
JP
2638void skb_clone_tx_timestamp(struct sk_buff *skb);
2639bool skb_defer_rx_timestamp(struct sk_buff *skb);
c1f19b51
RC
2640
2641#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2642
2643static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2644{
2645}
2646
2647static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2648{
2649 return false;
2650}
2651
2652#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2653
2654/**
2655 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2656 *
da92b194
RC
2657 * PHY drivers may accept clones of transmitted packets for
2658 * timestamping via their phy_driver.txtstamp method. These drivers
2659 * must call this function to return the skb back to the stack, with
2660 * or without a timestamp.
2661 *
c1f19b51 2662 * @skb: clone of the the original outgoing packet
da92b194 2663 * @hwtstamps: hardware time stamps, may be NULL if not available
c1f19b51
RC
2664 *
2665 */
2666void skb_complete_tx_timestamp(struct sk_buff *skb,
2667 struct skb_shared_hwtstamps *hwtstamps);
2668
ac45f602
PO
2669/**
2670 * skb_tstamp_tx - queue clone of skb with send time stamps
2671 * @orig_skb: the original outgoing packet
2672 * @hwtstamps: hardware time stamps, may be NULL if not available
2673 *
2674 * If the skb has a socket associated, then this function clones the
2675 * skb (thus sharing the actual data and optional structures), stores
2676 * the optional hardware time stamping information (if non NULL) or
2677 * generates a software time stamp (otherwise), then queues the clone
2678 * to the error queue of the socket. Errors are silently ignored.
2679 */
7965bd4d
JP
2680void skb_tstamp_tx(struct sk_buff *orig_skb,
2681 struct skb_shared_hwtstamps *hwtstamps);
ac45f602 2682
4507a715
RC
2683static inline void sw_tx_timestamp(struct sk_buff *skb)
2684{
2244d07b
OH
2685 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2686 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
4507a715
RC
2687 skb_tstamp_tx(skb, NULL);
2688}
2689
2690/**
2691 * skb_tx_timestamp() - Driver hook for transmit timestamping
2692 *
2693 * Ethernet MAC Drivers should call this function in their hard_xmit()
4ff75b7c 2694 * function immediately before giving the sk_buff to the MAC hardware.
4507a715 2695 *
73409f3b
DM
2696 * Specifically, one should make absolutely sure that this function is
2697 * called before TX completion of this packet can trigger. Otherwise
2698 * the packet could potentially already be freed.
2699 *
4507a715
RC
2700 * @skb: A socket buffer.
2701 */
2702static inline void skb_tx_timestamp(struct sk_buff *skb)
2703{
c1f19b51 2704 skb_clone_tx_timestamp(skb);
4507a715
RC
2705 sw_tx_timestamp(skb);
2706}
2707
6e3e939f
JB
2708/**
2709 * skb_complete_wifi_ack - deliver skb with wifi status
2710 *
2711 * @skb: the original outgoing packet
2712 * @acked: ack status
2713 *
2714 */
2715void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2716
7965bd4d
JP
2717__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2718__sum16 __skb_checksum_complete(struct sk_buff *skb);
fb286bb2 2719
60476372
HX
2720static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2721{
2722 return skb->ip_summed & CHECKSUM_UNNECESSARY;
2723}
2724
fb286bb2
HX
2725/**
2726 * skb_checksum_complete - Calculate checksum of an entire packet
2727 * @skb: packet to process
2728 *
2729 * This function calculates the checksum over the entire packet plus
2730 * the value of skb->csum. The latter can be used to supply the
2731 * checksum of a pseudo header as used by TCP/UDP. It returns the
2732 * checksum.
2733 *
2734 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2735 * this function can be used to verify that checksum on received
2736 * packets. In that case the function should return zero if the
2737 * checksum is correct. In particular, this function will return zero
2738 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2739 * hardware has already verified the correctness of the checksum.
2740 */
4381ca3c 2741static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
fb286bb2 2742{
60476372
HX
2743 return skb_csum_unnecessary(skb) ?
2744 0 : __skb_checksum_complete(skb);
fb286bb2
HX
2745}
2746
76ba0aae
TH
2747/* Check if we need to perform checksum complete validation.
2748 *
2749 * Returns true if checksum complete is needed, false otherwise
2750 * (either checksum is unnecessary or zero checksum is allowed).
2751 */
2752static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
2753 bool zero_okay,
2754 __sum16 check)
2755{
2756 if (skb_csum_unnecessary(skb)) {
2757 return false;
2758 } else if (zero_okay && !check) {
2759 skb->ip_summed = CHECKSUM_UNNECESSARY;
2760 return false;
2761 }
2762
2763 return true;
2764}
2765
2766/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
2767 * in checksum_init.
2768 */
2769#define CHECKSUM_BREAK 76
2770
2771/* Validate (init) checksum based on checksum complete.
2772 *
2773 * Return values:
2774 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
2775 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
2776 * checksum is stored in skb->csum for use in __skb_checksum_complete
2777 * non-zero: value of invalid checksum
2778 *
2779 */
2780static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
2781 bool complete,
2782 __wsum psum)
2783{
2784 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2785 if (!csum_fold(csum_add(psum, skb->csum))) {
2786 skb->ip_summed = CHECKSUM_UNNECESSARY;
2787 return 0;
2788 }
2789 }
2790
2791 skb->csum = psum;
2792
2793 if (complete || skb->len <= CHECKSUM_BREAK)
2794 return __skb_checksum_complete(skb);
2795
2796 return 0;
2797}
2798
2799static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
2800{
2801 return 0;
2802}
2803
2804/* Perform checksum validate (init). Note that this is a macro since we only
2805 * want to calculate the pseudo header which is an input function if necessary.
2806 * First we try to validate without any computation (checksum unnecessary) and
2807 * then calculate based on checksum complete calling the function to compute
2808 * pseudo header.
2809 *
2810 * Return values:
2811 * 0: checksum is validated or try to in skb_checksum_complete
2812 * non-zero: value of invalid checksum
2813 */
2814#define __skb_checksum_validate(skb, proto, complete, \
2815 zero_okay, check, compute_pseudo) \
2816({ \
2817 __sum16 __ret = 0; \
2818 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
2819 __ret = __skb_checksum_validate_complete(skb, \
2820 complete, compute_pseudo(skb, proto)); \
2821 __ret; \
2822})
2823
2824#define skb_checksum_init(skb, proto, compute_pseudo) \
2825 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
2826
2827#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
2828 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
2829
2830#define skb_checksum_validate(skb, proto, compute_pseudo) \
2831 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
2832
2833#define skb_checksum_validate_zero_check(skb, proto, check, \
2834 compute_pseudo) \
2835 __skb_checksum_validate_(skb, proto, true, true, check, compute_pseudo)
2836
2837#define skb_checksum_simple_validate(skb) \
2838 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
2839
5f79e0f9 2840#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
7965bd4d 2841void nf_conntrack_destroy(struct nf_conntrack *nfct);
1da177e4
LT
2842static inline void nf_conntrack_put(struct nf_conntrack *nfct)
2843{
2844 if (nfct && atomic_dec_and_test(&nfct->use))
de6e05c4 2845 nf_conntrack_destroy(nfct);
1da177e4
LT
2846}
2847static inline void nf_conntrack_get(struct nf_conntrack *nfct)
2848{
2849 if (nfct)
2850 atomic_inc(&nfct->use);
2851}
2fc72c7b 2852#endif
1da177e4
LT
2853#ifdef CONFIG_BRIDGE_NETFILTER
2854static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
2855{
2856 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
2857 kfree(nf_bridge);
2858}
2859static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
2860{
2861 if (nf_bridge)
2862 atomic_inc(&nf_bridge->use);
2863}
2864#endif /* CONFIG_BRIDGE_NETFILTER */
a193a4ab
PM
2865static inline void nf_reset(struct sk_buff *skb)
2866{
5f79e0f9 2867#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
a193a4ab
PM
2868 nf_conntrack_put(skb->nfct);
2869 skb->nfct = NULL;
2fc72c7b 2870#endif
a193a4ab
PM
2871#ifdef CONFIG_BRIDGE_NETFILTER
2872 nf_bridge_put(skb->nf_bridge);
2873 skb->nf_bridge = NULL;
2874#endif
2875}
2876
124dff01
PM
2877static inline void nf_reset_trace(struct sk_buff *skb)
2878{
478b360a 2879#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
130549fe
G
2880 skb->nf_trace = 0;
2881#endif
a193a4ab
PM
2882}
2883
edda553c
YK
2884/* Note: This doesn't put any conntrack and bridge info in dst. */
2885static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2886{
5f79e0f9 2887#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
edda553c
YK
2888 dst->nfct = src->nfct;
2889 nf_conntrack_get(src->nfct);
2890 dst->nfctinfo = src->nfctinfo;
2fc72c7b 2891#endif
edda553c
YK
2892#ifdef CONFIG_BRIDGE_NETFILTER
2893 dst->nf_bridge = src->nf_bridge;
2894 nf_bridge_get(src->nf_bridge);
2895#endif
478b360a
FW
2896#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2897 dst->nf_trace = src->nf_trace;
2898#endif
edda553c
YK
2899}
2900
e7ac05f3
YK
2901static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2902{
e7ac05f3 2903#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
5f79e0f9 2904 nf_conntrack_put(dst->nfct);
2fc72c7b 2905#endif
e7ac05f3
YK
2906#ifdef CONFIG_BRIDGE_NETFILTER
2907 nf_bridge_put(dst->nf_bridge);
2908#endif
2909 __nf_copy(dst, src);
2910}
2911
984bc16c
JM
2912#ifdef CONFIG_NETWORK_SECMARK
2913static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2914{
2915 to->secmark = from->secmark;
2916}
2917
2918static inline void skb_init_secmark(struct sk_buff *skb)
2919{
2920 skb->secmark = 0;
2921}
2922#else
2923static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2924{ }
2925
2926static inline void skb_init_secmark(struct sk_buff *skb)
2927{ }
2928#endif
2929
574f7194
EB
2930static inline bool skb_irq_freeable(const struct sk_buff *skb)
2931{
2932 return !skb->destructor &&
2933#if IS_ENABLED(CONFIG_XFRM)
2934 !skb->sp &&
2935#endif
2936#if IS_ENABLED(CONFIG_NF_CONNTRACK)
2937 !skb->nfct &&
2938#endif
2939 !skb->_skb_refdst &&
2940 !skb_has_frag_list(skb);
2941}
2942
f25f4e44
PWJ
2943static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
2944{
f25f4e44 2945 skb->queue_mapping = queue_mapping;
f25f4e44
PWJ
2946}
2947
9247744e 2948static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4e3ab47a 2949{
4e3ab47a 2950 return skb->queue_mapping;
4e3ab47a
PE
2951}
2952
f25f4e44
PWJ
2953static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
2954{
f25f4e44 2955 to->queue_mapping = from->queue_mapping;
f25f4e44
PWJ
2956}
2957
d5a9e24a
DM
2958static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
2959{
2960 skb->queue_mapping = rx_queue + 1;
2961}
2962
9247744e 2963static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
d5a9e24a
DM
2964{
2965 return skb->queue_mapping - 1;
2966}
2967
9247744e 2968static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
d5a9e24a 2969{
a02cec21 2970 return skb->queue_mapping != 0;
d5a9e24a
DM
2971}
2972
7965bd4d
JP
2973u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb,
2974 unsigned int num_tx_queues);
9247744e 2975
def8b4fa
AD
2976static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
2977{
0b3d8e08 2978#ifdef CONFIG_XFRM
def8b4fa 2979 return skb->sp;
def8b4fa 2980#else
def8b4fa 2981 return NULL;
def8b4fa 2982#endif
0b3d8e08 2983}
def8b4fa 2984
68c33163
PS
2985/* Keeps track of mac header offset relative to skb->head.
2986 * It is useful for TSO of Tunneling protocol. e.g. GRE.
2987 * For non-tunnel skb it points to skb_mac_header() and for
3347c960
ED
2988 * tunnel skb it points to outer mac header.
2989 * Keeps track of level of encapsulation of network headers.
2990 */
68c33163 2991struct skb_gso_cb {
3347c960
ED
2992 int mac_offset;
2993 int encap_level;
7e2b10c1 2994 __u16 csum_start;
68c33163
PS
2995};
2996#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
2997
2998static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
2999{
3000 return (skb_mac_header(inner_skb) - inner_skb->head) -
3001 SKB_GSO_CB(inner_skb)->mac_offset;
3002}
3003
1e2bd517
PS
3004static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3005{
3006 int new_headroom, headroom;
3007 int ret;
3008
3009 headroom = skb_headroom(skb);
3010 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3011 if (ret)
3012 return ret;
3013
3014 new_headroom = skb_headroom(skb);
3015 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3016 return 0;
3017}
3018
7e2b10c1
TH
3019/* Compute the checksum for a gso segment. First compute the checksum value
3020 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3021 * then add in skb->csum (checksum from csum_start to end of packet).
3022 * skb->csum and csum_start are then updated to reflect the checksum of the
3023 * resultant packet starting from the transport header-- the resultant checksum
3024 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3025 * header.
3026 */
3027static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3028{
3029 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3030 skb_transport_offset(skb);
3031 __u16 csum;
3032
3033 csum = csum_fold(csum_partial(skb_transport_header(skb),
3034 plen, skb->csum));
3035 skb->csum = res;
3036 SKB_GSO_CB(skb)->csum_start -= plen;
3037
3038 return csum;
3039}
3040
bdcc0924 3041static inline bool skb_is_gso(const struct sk_buff *skb)
89114afd
HX
3042{
3043 return skb_shinfo(skb)->gso_size;
3044}
3045
36a8f39e 3046/* Note: Should be called only if skb_is_gso(skb) is true */
bdcc0924 3047static inline bool skb_is_gso_v6(const struct sk_buff *skb)
eabd7e35
BG
3048{
3049 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3050}
3051
7965bd4d 3052void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4497b076
BH
3053
3054static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3055{
3056 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3057 * wanted then gso_type will be set. */
05bdd2f1
ED
3058 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3059
b78462eb
AD
3060 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3061 unlikely(shinfo->gso_type == 0)) {
4497b076
BH
3062 __skb_warn_lro_forwarding(skb);
3063 return true;
3064 }
3065 return false;
3066}
3067
35fc92a9
HX
3068static inline void skb_forward_csum(struct sk_buff *skb)
3069{
3070 /* Unfortunately we don't support this one. Any brave souls? */
3071 if (skb->ip_summed == CHECKSUM_COMPLETE)
3072 skb->ip_summed = CHECKSUM_NONE;
3073}
3074
bc8acf2c
ED
3075/**
3076 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3077 * @skb: skb to check
3078 *
3079 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3080 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3081 * use this helper, to document places where we make this assertion.
3082 */
05bdd2f1 3083static inline void skb_checksum_none_assert(const struct sk_buff *skb)
bc8acf2c
ED
3084{
3085#ifdef DEBUG
3086 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3087#endif
3088}
3089
f35d9d8a 3090bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
a6686f2f 3091
ed1f50c3
PD
3092int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3093
f77668dc
DB
3094u32 __skb_get_poff(const struct sk_buff *skb);
3095
3a7c1ee4
AD
3096/**
3097 * skb_head_is_locked - Determine if the skb->head is locked down
3098 * @skb: skb to check
3099 *
3100 * The head on skbs build around a head frag can be removed if they are
3101 * not cloned. This function returns true if the skb head is locked down
3102 * due to either being allocated via kmalloc, or by being a clone with
3103 * multiple references to the head.
3104 */
3105static inline bool skb_head_is_locked(const struct sk_buff *skb)
3106{
3107 return !skb->head_frag || skb_cloned(skb);
3108}
fe6cc55f
FW
3109
3110/**
3111 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3112 *
3113 * @skb: GSO skb
3114 *
3115 * skb_gso_network_seglen is used to determine the real size of the
3116 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3117 *
3118 * The MAC/L2 header is not accounted for.
3119 */
3120static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3121{
3122 unsigned int hdr_len = skb_transport_header(skb) -
3123 skb_network_header(skb);
3124 return hdr_len + skb_gso_transport_seglen(skb);
3125}
1da177e4
LT
3126#endif /* __KERNEL__ */
3127#endif /* _LINUX_SKBUFF_H */