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