net: convert sk_buff.users from atomic_t to refcount_t
[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 22#include <linux/cache.h>
56b17425 23#include <linux/rbtree.h>
51f3d02b 24#include <linux/socket.h>
1da177e4 25
60063497 26#include <linux/atomic.h>
1da177e4
LT
27#include <asm/types.h>
28#include <linux/spinlock.h>
1da177e4 29#include <linux/net.h>
3fc7e8a6 30#include <linux/textsearch.h>
1da177e4 31#include <net/checksum.h>
a80958f4 32#include <linux/rcupdate.h>
b7aa0bf7 33#include <linux/hrtimer.h>
131ea667 34#include <linux/dma-mapping.h>
c8f44aff 35#include <linux/netdev_features.h>
363ec392 36#include <linux/sched.h>
e6017571 37#include <linux/sched/clock.h>
1bd758eb 38#include <net/flow_dissector.h>
a60e3cc7 39#include <linux/splice.h>
72b31f72 40#include <linux/in6.h>
8b10cab6 41#include <linux/if_packet.h>
f70ea018 42#include <net/flow.h>
1da177e4 43
7a6ae71b
TH
44/* The interface for checksum offload between the stack and networking drivers
45 * is as follows...
46 *
47 * A. IP checksum related features
48 *
49 * Drivers advertise checksum offload capabilities in the features of a device.
50 * From the stack's point of view these are capabilities offered by the driver,
51 * a driver typically only advertises features that it is capable of offloading
52 * to its device.
53 *
54 * The checksum related features are:
55 *
56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
57 * IP (one's complement) checksum for any combination
58 * of protocols or protocol layering. The checksum is
59 * computed and set in a packet per the CHECKSUM_PARTIAL
60 * interface (see below).
61 *
62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
63 * TCP or UDP packets over IPv4. These are specifically
64 * unencapsulated packets of the form IPv4|TCP or
65 * IPv4|UDP where the Protocol field in the IPv4 header
66 * is TCP or UDP. The IPv4 header may contain IP options
67 * This feature cannot be set in features for a device
68 * with NETIF_F_HW_CSUM also set. This feature is being
69 * DEPRECATED (see below).
70 *
71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
72 * TCP or UDP packets over IPv6. These are specifically
73 * unencapsulated packets of the form IPv6|TCP or
74 * IPv4|UDP where the Next Header field in the IPv6
75 * header is either TCP or UDP. IPv6 extension headers
76 * are not supported with this feature. This feature
77 * cannot be set in features for a device with
78 * NETIF_F_HW_CSUM also set. This feature is being
79 * DEPRECATED (see below).
80 *
81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
82 * This flag is used only used to disable the RX checksum
83 * feature for a device. The stack will accept receive
84 * checksum indication in packets received on a device
85 * regardless of whether NETIF_F_RXCSUM is set.
86 *
87 * B. Checksumming of received packets by device. Indication of checksum
88 * verification is in set skb->ip_summed. Possible values are:
78ea85f1
DB
89 *
90 * CHECKSUM_NONE:
91 *
7a6ae71b 92 * Device did not checksum this packet e.g. due to lack of capabilities.
78ea85f1
DB
93 * The packet contains full (though not verified) checksum in packet but
94 * not in skb->csum. Thus, skb->csum is undefined in this case.
95 *
96 * CHECKSUM_UNNECESSARY:
97 *
98 * The hardware you're dealing with doesn't calculate the full checksum
99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
77cffe23
TH
100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
101 * if their checksums are okay. skb->csum is still undefined in this case
7a6ae71b
TH
102 * though. A driver or device must never modify the checksum field in the
103 * packet even if checksum is verified.
77cffe23
TH
104 *
105 * CHECKSUM_UNNECESSARY is applicable to following protocols:
106 * TCP: IPv6 and IPv4.
107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
108 * zero UDP checksum for either IPv4 or IPv6, the networking stack
109 * may perform further validation in this case.
110 * GRE: only if the checksum is present in the header.
111 * SCTP: indicates the CRC in SCTP header has been validated.
b4759dcd 112 * FCOE: indicates the CRC in FC frame has been validated.
77cffe23
TH
113 *
114 * skb->csum_level indicates the number of consecutive checksums found in
115 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
116 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
117 * and a device is able to verify the checksums for UDP (possibly zero),
118 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
119 * two. If the device were only able to verify the UDP checksum and not
120 * GRE, either because it doesn't support GRE checksum of because GRE
121 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
122 * not considered in this case).
78ea85f1
DB
123 *
124 * CHECKSUM_COMPLETE:
125 *
126 * This is the most generic way. The device supplied checksum of the _whole_
127 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
128 * hardware doesn't need to parse L3/L4 headers to implement this.
129 *
b4759dcd
DC
130 * Notes:
131 * - Even if device supports only some protocols, but is able to produce
132 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
133 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
78ea85f1
DB
134 *
135 * CHECKSUM_PARTIAL:
136 *
6edec0e6
TH
137 * A checksum is set up to be offloaded to a device as described in the
138 * output description for CHECKSUM_PARTIAL. This may occur on a packet
78ea85f1 139 * received directly from another Linux OS, e.g., a virtualized Linux kernel
6edec0e6
TH
140 * on the same host, or it may be set in the input path in GRO or remote
141 * checksum offload. For the purposes of checksum verification, the checksum
142 * referred to by skb->csum_start + skb->csum_offset and any preceding
143 * checksums in the packet are considered verified. Any checksums in the
144 * packet that are after the checksum being offloaded are not considered to
145 * be verified.
78ea85f1 146 *
7a6ae71b
TH
147 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
148 * in the skb->ip_summed for a packet. Values are:
78ea85f1
DB
149 *
150 * CHECKSUM_PARTIAL:
151 *
7a6ae71b 152 * The driver is required to checksum the packet as seen by hard_start_xmit()
78ea85f1 153 * from skb->csum_start up to the end, and to record/write the checksum at
7a6ae71b
TH
154 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
155 * csum_start and csum_offset values are valid values given the length and
156 * offset of the packet, however they should not attempt to validate that the
157 * checksum refers to a legitimate transport layer checksum-- it is the
158 * purview of the stack to validate that csum_start and csum_offset are set
159 * correctly.
160 *
161 * When the stack requests checksum offload for a packet, the driver MUST
162 * ensure that the checksum is set correctly. A driver can either offload the
163 * checksum calculation to the device, or call skb_checksum_help (in the case
164 * that the device does not support offload for a particular checksum).
165 *
166 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
167 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
43c26a1a
DC
168 * checksum offload capability.
169 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
170 * on network device checksumming capabilities: if a packet does not match
171 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
172 * csum_not_inet, see item D.) is called to resolve the checksum.
78ea85f1 173 *
7a6ae71b 174 * CHECKSUM_NONE:
78ea85f1 175 *
7a6ae71b
TH
176 * The skb was already checksummed by the protocol, or a checksum is not
177 * required.
78ea85f1
DB
178 *
179 * CHECKSUM_UNNECESSARY:
180 *
7a6ae71b
TH
181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
182 * output.
78ea85f1 183 *
7a6ae71b
TH
184 * CHECKSUM_COMPLETE:
185 * Not used in checksum output. If a driver observes a packet with this value
186 * set in skbuff, if should treat as CHECKSUM_NONE being set.
187 *
188 * D. Non-IP checksum (CRC) offloads
189 *
190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
191 * offloading the SCTP CRC in a packet. To perform this offload the stack
dba00306
DC
192 * will set set csum_start and csum_offset accordingly, set ip_summed to
193 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
194 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
195 * A driver that supports both IP checksum offload and SCTP CRC32c offload
196 * must verify which offload is configured for a packet by testing the
197 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
198 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
7a6ae71b
TH
199 *
200 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
201 * offloading the FCOE CRC in a packet. To perform this offload the stack
202 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
203 * accordingly. Note the there is no indication in the skbuff that the
204 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
205 * both IP checksum offload and FCOE CRC offload must verify which offload
206 * is configured for a packet presumably by inspecting packet headers.
207 *
208 * E. Checksumming on output with GSO.
209 *
210 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
211 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
212 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
213 * part of the GSO operation is implied. If a checksum is being offloaded
214 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
215 * are set to refer to the outermost checksum being offload (two offloaded
216 * checksums are possible with UDP encapsulation).
78ea85f1
DB
217 */
218
60476372 219/* Don't change this without changing skb_csum_unnecessary! */
78ea85f1
DB
220#define CHECKSUM_NONE 0
221#define CHECKSUM_UNNECESSARY 1
222#define CHECKSUM_COMPLETE 2
223#define CHECKSUM_PARTIAL 3
1da177e4 224
77cffe23
TH
225/* Maximum value in skb->csum_level */
226#define SKB_MAX_CSUM_LEVEL 3
227
0bec8c88 228#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
fc910a27 229#define SKB_WITH_OVERHEAD(X) \
deea84b0 230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
fc910a27
DM
231#define SKB_MAX_ORDER(X, ORDER) \
232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
1da177e4
LT
233#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
234#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
235
87fb4b7b
ED
236/* return minimum truesize of one skb containing X bytes of data */
237#define SKB_TRUESIZE(X) ((X) + \
238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
240
1da177e4 241struct net_device;
716ea3a7 242struct scatterlist;
9c55e01c 243struct pipe_inode_info;
a8f820aa 244struct iov_iter;
fd11a83d 245struct napi_struct;
1da177e4 246
5f79e0f9 247#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
1da177e4
LT
248struct nf_conntrack {
249 atomic_t use;
1da177e4 250};
5f79e0f9 251#endif
1da177e4 252
34666d46 253#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1da177e4 254struct nf_bridge_info {
53869ceb 255 refcount_t use;
3eaf4025
FW
256 enum {
257 BRNF_PROTO_UNCHANGED,
258 BRNF_PROTO_8021Q,
259 BRNF_PROTO_PPPOE
7fb48c5b 260 } orig_proto:8;
72b1e5e4
FW
261 u8 pkt_otherhost:1;
262 u8 in_prerouting:1;
263 u8 bridged_dnat:1;
411ffb4f 264 __u16 frag_max_size;
bf1ac5ca 265 struct net_device *physindev;
63cdbc06
FW
266
267 /* always valid & non-NULL from FORWARD on, for physdev match */
268 struct net_device *physoutdev;
7fb48c5b 269 union {
72b1e5e4 270 /* prerouting: detect dnat in orig/reply direction */
72b31f72
BT
271 __be32 ipv4_daddr;
272 struct in6_addr ipv6_daddr;
72b1e5e4
FW
273
274 /* after prerouting + nat detected: store original source
275 * mac since neigh resolution overwrites it, only used while
276 * skb is out in neigh layer.
277 */
278 char neigh_header[8];
72b31f72 279 };
1da177e4
LT
280};
281#endif
282
1da177e4
LT
283struct sk_buff_head {
284 /* These two members must be first. */
285 struct sk_buff *next;
286 struct sk_buff *prev;
287
288 __u32 qlen;
289 spinlock_t lock;
290};
291
292struct sk_buff;
293
9d4dde52
IC
294/* To allow 64K frame to be packed as single skb without frag_list we
295 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
296 * buffers which do not start on a page boundary.
297 *
298 * Since GRO uses frags we allocate at least 16 regardless of page
299 * size.
a715dea3 300 */
9d4dde52 301#if (65536/PAGE_SIZE + 1) < 16
eec00954 302#define MAX_SKB_FRAGS 16UL
a715dea3 303#else
9d4dde52 304#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
a715dea3 305#endif
5f74f82e 306extern int sysctl_max_skb_frags;
1da177e4 307
3953c46c
MRL
308/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
309 * segment using its current segmentation instead.
310 */
311#define GSO_BY_FRAGS 0xFFFF
312
1da177e4
LT
313typedef struct skb_frag_struct skb_frag_t;
314
315struct skb_frag_struct {
a8605c60
IC
316 struct {
317 struct page *p;
318 } page;
cb4dfe56 319#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
a309bb07
DM
320 __u32 page_offset;
321 __u32 size;
cb4dfe56
ED
322#else
323 __u16 page_offset;
324 __u16 size;
325#endif
1da177e4
LT
326};
327
9e903e08
ED
328static inline unsigned int skb_frag_size(const skb_frag_t *frag)
329{
330 return frag->size;
331}
332
333static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
334{
335 frag->size = size;
336}
337
338static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
339{
340 frag->size += delta;
341}
342
343static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
344{
345 frag->size -= delta;
346}
347
ac45f602
PO
348#define HAVE_HW_TIME_STAMP
349
350/**
d3a21be8 351 * struct skb_shared_hwtstamps - hardware time stamps
ac45f602
PO
352 * @hwtstamp: hardware time stamp transformed into duration
353 * since arbitrary point in time
ac45f602
PO
354 *
355 * Software time stamps generated by ktime_get_real() are stored in
4d276eb6 356 * skb->tstamp.
ac45f602
PO
357 *
358 * hwtstamps can only be compared against other hwtstamps from
359 * the same device.
360 *
361 * This structure is attached to packets as part of the
362 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
363 */
364struct skb_shared_hwtstamps {
365 ktime_t hwtstamp;
ac45f602
PO
366};
367
2244d07b
OH
368/* Definitions for tx_flags in struct skb_shared_info */
369enum {
370 /* generate hardware time stamp */
371 SKBTX_HW_TSTAMP = 1 << 0,
372
e7fd2885 373 /* generate software time stamp when queueing packet to NIC */
2244d07b
OH
374 SKBTX_SW_TSTAMP = 1 << 1,
375
376 /* device driver is going to provide hardware time stamp */
377 SKBTX_IN_PROGRESS = 1 << 2,
378
a6686f2f 379 /* device driver supports TX zero-copy buffers */
62b1a8ab 380 SKBTX_DEV_ZEROCOPY = 1 << 3,
6e3e939f
JB
381
382 /* generate wifi status information (where possible) */
62b1a8ab 383 SKBTX_WIFI_STATUS = 1 << 4,
c9af6db4
PS
384
385 /* This indicates at least one fragment might be overwritten
386 * (as in vmsplice(), sendfile() ...)
387 * If we need to compute a TX checksum, we'll need to copy
388 * all frags to avoid possible bad checksum
389 */
390 SKBTX_SHARED_FRAG = 1 << 5,
e7fd2885
WB
391
392 /* generate software time stamp when entering packet scheduling */
393 SKBTX_SCHED_TSTAMP = 1 << 6,
a6686f2f
SM
394};
395
e1c8a607 396#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
0a2cf20c 397 SKBTX_SCHED_TSTAMP)
f24b9be5
WB
398#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
399
a6686f2f
SM
400/*
401 * The callback notifies userspace to release buffers when skb DMA is done in
402 * lower device, the skb last reference should be 0 when calling this.
e19d6763
MT
403 * The zerocopy_success argument is true if zero copy transmit occurred,
404 * false on data copy or out of memory error caused by data copy attempt.
ca8f4fb2
MT
405 * The ctx field is used to track device context.
406 * The desc field is used to track userspace buffer index.
a6686f2f
SM
407 */
408struct ubuf_info {
e19d6763 409 void (*callback)(struct ubuf_info *, bool zerocopy_success);
ca8f4fb2 410 void *ctx;
a6686f2f 411 unsigned long desc;
ac45f602
PO
412};
413
1da177e4
LT
414/* This data is invariant across clones and lives at
415 * the end of the header data, ie. at skb->end.
416 */
417struct skb_shared_info {
7f564528 418 unsigned short _unused;
9f42f126
IC
419 unsigned char nr_frags;
420 __u8 tx_flags;
7967168c
HX
421 unsigned short gso_size;
422 /* Warning: this field is not always filled in (UFO)! */
423 unsigned short gso_segs;
1da177e4 424 struct sk_buff *frag_list;
ac45f602 425 struct skb_shared_hwtstamps hwtstamps;
7f564528 426 unsigned int gso_type;
09c2d251 427 u32 tskey;
9f42f126 428 __be32 ip6_frag_id;
ec7d2f2c
ED
429
430 /*
431 * Warning : all fields before dataref are cleared in __alloc_skb()
432 */
433 atomic_t dataref;
434
69e3c75f
JB
435 /* Intermediate layers must ensure that destructor_arg
436 * remains valid until skb destructor */
437 void * destructor_arg;
a6686f2f 438
fed66381
ED
439 /* must be last field, see pskb_expand_head() */
440 skb_frag_t frags[MAX_SKB_FRAGS];
1da177e4
LT
441};
442
443/* We divide dataref into two halves. The higher 16 bits hold references
444 * to the payload part of skb->data. The lower 16 bits hold references to
334a8132
PM
445 * the entire skb->data. A clone of a headerless skb holds the length of
446 * the header in skb->hdr_len.
1da177e4
LT
447 *
448 * All users must obey the rule that the skb->data reference count must be
449 * greater than or equal to the payload reference count.
450 *
451 * Holding a reference to the payload part means that the user does not
452 * care about modifications to the header part of skb->data.
453 */
454#define SKB_DATAREF_SHIFT 16
455#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
456
d179cd12
DM
457
458enum {
c8753d55
VS
459 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
460 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
461 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
d179cd12
DM
462};
463
7967168c
HX
464enum {
465 SKB_GSO_TCPV4 = 1 << 0,
f83ef8c0 466 SKB_GSO_UDP = 1 << 1,
576a30eb
HX
467
468 /* This indicates the skb is from an untrusted source. */
469 SKB_GSO_DODGY = 1 << 2,
b0da8537
MC
470
471 /* This indicates the tcp segment has CWR set. */
f83ef8c0
HX
472 SKB_GSO_TCP_ECN = 1 << 3,
473
cbc53e08 474 SKB_GSO_TCP_FIXEDID = 1 << 4,
01d5b2fc 475
cbc53e08 476 SKB_GSO_TCPV6 = 1 << 5,
68c33163 477
cbc53e08 478 SKB_GSO_FCOE = 1 << 6,
73136267 479
cbc53e08 480 SKB_GSO_GRE = 1 << 7,
0d89d203 481
cbc53e08 482 SKB_GSO_GRE_CSUM = 1 << 8,
cb32f511 483
7e13318d 484 SKB_GSO_IPXIP4 = 1 << 9,
61c1db7f 485
7e13318d 486 SKB_GSO_IPXIP6 = 1 << 10,
0f4f4ffa 487
cbc53e08 488 SKB_GSO_UDP_TUNNEL = 1 << 11,
4749c09c 489
cbc53e08
AD
490 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
491
802ab55a
AD
492 SKB_GSO_PARTIAL = 1 << 13,
493
494 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
90017acc
MRL
495
496 SKB_GSO_SCTP = 1 << 15,
c7ef8f0c
SK
497
498 SKB_GSO_ESP = 1 << 16,
7967168c
HX
499};
500
2e07fa9c
ACM
501#if BITS_PER_LONG > 32
502#define NET_SKBUFF_DATA_USES_OFFSET 1
503#endif
504
505#ifdef NET_SKBUFF_DATA_USES_OFFSET
506typedef unsigned int sk_buff_data_t;
507#else
508typedef unsigned char *sk_buff_data_t;
509#endif
510
1da177e4
LT
511/**
512 * struct sk_buff - socket buffer
513 * @next: Next buffer in list
514 * @prev: Previous buffer in list
363ec392 515 * @tstamp: Time we arrived/left
56b17425 516 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
d84e0bd7 517 * @sk: Socket we are owned by
1da177e4 518 * @dev: Device we arrived on/are leaving by
d84e0bd7 519 * @cb: Control buffer. Free for use by every layer. Put private vars here
7fee226a 520 * @_skb_refdst: destination entry (with norefcount bit)
67be2dd1 521 * @sp: the security path, used for xfrm
1da177e4
LT
522 * @len: Length of actual data
523 * @data_len: Data length
524 * @mac_len: Length of link layer header
334a8132 525 * @hdr_len: writable header length of cloned skb
663ead3b
HX
526 * @csum: Checksum (must include start/offset pair)
527 * @csum_start: Offset from skb->head where checksumming should start
528 * @csum_offset: Offset from csum_start where checksum should be stored
d84e0bd7 529 * @priority: Packet queueing priority
60ff7467 530 * @ignore_df: allow local fragmentation
1da177e4 531 * @cloned: Head may be cloned (check refcnt to be sure)
d84e0bd7 532 * @ip_summed: Driver fed us an IP checksum
1da177e4
LT
533 * @nohdr: Payload reference only, must not modify header
534 * @pkt_type: Packet class
c83c2486 535 * @fclone: skbuff clone status
c83c2486 536 * @ipvs_property: skbuff is owned by ipvs
e7246e12 537 * @tc_skip_classify: do not classify packet. set by IFB device
8dc07fdb 538 * @tc_at_ingress: used within tc_classify to distinguish in/egress
bc31c905
WB
539 * @tc_redirected: packet was redirected by a tc action
540 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
31729363
RD
541 * @peeked: this packet has been seen already, so stats have been
542 * done for it, don't do them again
ba9dda3a 543 * @nf_trace: netfilter packet trace flag
d84e0bd7
DB
544 * @protocol: Packet protocol from driver
545 * @destructor: Destruct function
a9e419dc 546 * @_nfct: Associated connection, if any (with nfctinfo bits)
1da177e4 547 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
8964be4a 548 * @skb_iif: ifindex of device we arrived on
1da177e4 549 * @tc_index: Traffic control index
61b905da 550 * @hash: the packet hash
d84e0bd7 551 * @queue_mapping: Queue mapping for multiqueue devices
0b725a2c 552 * @xmit_more: More SKBs are pending for this queue
553a5672 553 * @ndisc_nodetype: router type (from link layer)
d84e0bd7 554 * @ooo_okay: allow the mapping of a socket to a queue to be changed
61b905da 555 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
4ca2462e 556 * ports.
a3b18ddb 557 * @sw_hash: indicates hash was computed in software stack
6e3e939f
JB
558 * @wifi_acked_valid: wifi_acked was set
559 * @wifi_acked: whether frame was acked on wifi or not
3bdc0eba 560 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
dba00306 561 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
4ff06203 562 * @dst_pending_confirm: need to confirm neighbour
06021292 563 * @napi_id: id of the NAPI struct this skb came from
984bc16c 564 * @secmark: security marking
d84e0bd7 565 * @mark: Generic packet mark
86a9bad3 566 * @vlan_proto: vlan encapsulation protocol
6aa895b0 567 * @vlan_tci: vlan tag control information
0d89d203 568 * @inner_protocol: Protocol (encapsulation)
6a674e9c
JG
569 * @inner_transport_header: Inner transport layer header (encapsulation)
570 * @inner_network_header: Network layer header (encapsulation)
aefbd2b3 571 * @inner_mac_header: Link layer header (encapsulation)
d84e0bd7
DB
572 * @transport_header: Transport layer header
573 * @network_header: Network layer header
574 * @mac_header: Link layer header
575 * @tail: Tail pointer
576 * @end: End pointer
577 * @head: Head of buffer
578 * @data: Data head pointer
579 * @truesize: Buffer size
580 * @users: User count - see {datagram,tcp}.c
1da177e4
LT
581 */
582
583struct sk_buff {
363ec392 584 union {
56b17425
ED
585 struct {
586 /* These two members must be first. */
587 struct sk_buff *next;
588 struct sk_buff *prev;
589
590 union {
591 ktime_t tstamp;
9a568de4 592 u64 skb_mstamp;
56b17425
ED
593 };
594 };
595 struct rb_node rbnode; /* used in netem & tcp stack */
363ec392 596 };
da3f5cf1 597 struct sock *sk;
1da177e4 598
c84d9490
ED
599 union {
600 struct net_device *dev;
601 /* Some protocols might use this space to store information,
602 * while device pointer would be NULL.
603 * UDP receive path is one user.
604 */
605 unsigned long dev_scratch;
606 };
1da177e4
LT
607 /*
608 * This is the control buffer. It is free to use for every
609 * layer. Please put your private variables there. If you
610 * want to keep them across layers you have to do a skb_clone()
611 * first. This is owned by whoever has the skb queued ATM.
612 */
da3f5cf1 613 char cb[48] __aligned(8);
1da177e4 614
7fee226a 615 unsigned long _skb_refdst;
b1937227 616 void (*destructor)(struct sk_buff *skb);
da3f5cf1
FF
617#ifdef CONFIG_XFRM
618 struct sec_path *sp;
b1937227
ED
619#endif
620#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
a9e419dc 621 unsigned long _nfct;
b1937227 622#endif
85224844 623#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
b1937227 624 struct nf_bridge_info *nf_bridge;
da3f5cf1 625#endif
1da177e4 626 unsigned int len,
334a8132
PM
627 data_len;
628 __u16 mac_len,
629 hdr_len;
b1937227
ED
630
631 /* Following fields are _not_ copied in __copy_skb_header()
632 * Note that queue_mapping is here mostly to fill a hole.
633 */
fe55f6d5 634 kmemcheck_bitfield_begin(flags1);
b1937227 635 __u16 queue_mapping;
36bbef52
DB
636
637/* if you move cloned around you also must adapt those constants */
638#ifdef __BIG_ENDIAN_BITFIELD
639#define CLONED_MASK (1 << 7)
640#else
641#define CLONED_MASK 1
642#endif
643#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
644
645 __u8 __cloned_offset[0];
b1937227 646 __u8 cloned:1,
6869c4d8 647 nohdr:1,
b84f4cc9 648 fclone:2,
a59322be 649 peeked:1,
b1937227 650 head_frag:1,
36bbef52
DB
651 xmit_more:1,
652 __unused:1; /* one bit hole */
fe55f6d5 653 kmemcheck_bitfield_end(flags1);
4031ae6e 654
b1937227
ED
655 /* fields enclosed in headers_start/headers_end are copied
656 * using a single memcpy() in __copy_skb_header()
657 */
ebcf34f3 658 /* private: */
b1937227 659 __u32 headers_start[0];
ebcf34f3 660 /* public: */
4031ae6e 661
233577a2
HFS
662/* if you move pkt_type around you also must adapt those constants */
663#ifdef __BIG_ENDIAN_BITFIELD
664#define PKT_TYPE_MAX (7 << 5)
665#else
666#define PKT_TYPE_MAX 7
1da177e4 667#endif
233577a2 668#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
fe55f6d5 669
233577a2 670 __u8 __pkt_type_offset[0];
b1937227 671 __u8 pkt_type:3;
c93bdd0e 672 __u8 pfmemalloc:1;
b1937227 673 __u8 ignore_df:1;
b1937227
ED
674
675 __u8 nf_trace:1;
676 __u8 ip_summed:2;
3853b584 677 __u8 ooo_okay:1;
61b905da 678 __u8 l4_hash:1;
a3b18ddb 679 __u8 sw_hash:1;
6e3e939f
JB
680 __u8 wifi_acked_valid:1;
681 __u8 wifi_acked:1;
b1937227 682
3bdc0eba 683 __u8 no_fcs:1;
77cffe23 684 /* Indicates the inner headers are valid in the skbuff. */
6a674e9c 685 __u8 encapsulation:1;
7e2b10c1 686 __u8 encap_hdr_csum:1;
5d0c2b95 687 __u8 csum_valid:1;
7e3cead5 688 __u8 csum_complete_sw:1;
b1937227 689 __u8 csum_level:2;
dba00306 690 __u8 csum_not_inet:1;
fe55f6d5 691
4ff06203 692 __u8 dst_pending_confirm:1;
b1937227
ED
693#ifdef CONFIG_IPV6_NDISC_NODETYPE
694 __u8 ndisc_nodetype:2;
695#endif
696 __u8 ipvs_property:1;
8bce6d7d 697 __u8 inner_protocol_type:1;
e585f236 698 __u8 remcsum_offload:1;
6bc506b4
IS
699#ifdef CONFIG_NET_SWITCHDEV
700 __u8 offload_fwd_mark:1;
701#endif
e7246e12
WB
702#ifdef CONFIG_NET_CLS_ACT
703 __u8 tc_skip_classify:1;
8dc07fdb 704 __u8 tc_at_ingress:1;
bc31c905
WB
705 __u8 tc_redirected:1;
706 __u8 tc_from_ingress:1;
e7246e12 707#endif
b1937227
ED
708
709#ifdef CONFIG_NET_SCHED
710 __u16 tc_index; /* traffic control index */
b1937227 711#endif
fe55f6d5 712
b1937227
ED
713 union {
714 __wsum csum;
715 struct {
716 __u16 csum_start;
717 __u16 csum_offset;
718 };
719 };
720 __u32 priority;
721 int skb_iif;
722 __u32 hash;
723 __be16 vlan_proto;
724 __u16 vlan_tci;
2bd82484
ED
725#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
726 union {
727 unsigned int napi_id;
728 unsigned int sender_cpu;
729 };
97fc2f08 730#endif
984bc16c 731#ifdef CONFIG_NETWORK_SECMARK
6bc506b4 732 __u32 secmark;
0c4f691f 733#endif
0c4f691f 734
3b885787
NH
735 union {
736 __u32 mark;
16fad69c 737 __u32 reserved_tailroom;
3b885787 738 };
1da177e4 739
8bce6d7d
TH
740 union {
741 __be16 inner_protocol;
742 __u8 inner_ipproto;
743 };
744
1a37e412
SH
745 __u16 inner_transport_header;
746 __u16 inner_network_header;
747 __u16 inner_mac_header;
b1937227
ED
748
749 __be16 protocol;
1a37e412
SH
750 __u16 transport_header;
751 __u16 network_header;
752 __u16 mac_header;
b1937227 753
ebcf34f3 754 /* private: */
b1937227 755 __u32 headers_end[0];
ebcf34f3 756 /* public: */
b1937227 757
1da177e4 758 /* These elements must be at the end, see alloc_skb() for details. */
27a884dc 759 sk_buff_data_t tail;
4305b541 760 sk_buff_data_t end;
1da177e4 761 unsigned char *head,
4305b541 762 *data;
27a884dc 763 unsigned int truesize;
63354797 764 refcount_t users;
1da177e4
LT
765};
766
767#ifdef __KERNEL__
768/*
769 * Handling routines are only of interest to the kernel
770 */
771#include <linux/slab.h>
772
1da177e4 773
c93bdd0e
MG
774#define SKB_ALLOC_FCLONE 0x01
775#define SKB_ALLOC_RX 0x02
fd11a83d 776#define SKB_ALLOC_NAPI 0x04
c93bdd0e
MG
777
778/* Returns true if the skb was allocated from PFMEMALLOC reserves */
779static inline bool skb_pfmemalloc(const struct sk_buff *skb)
780{
781 return unlikely(skb->pfmemalloc);
782}
783
7fee226a
ED
784/*
785 * skb might have a dst pointer attached, refcounted or not.
786 * _skb_refdst low order bit is set if refcount was _not_ taken
787 */
788#define SKB_DST_NOREF 1UL
789#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
790
a9e419dc 791#define SKB_NFCT_PTRMASK ~(7UL)
7fee226a
ED
792/**
793 * skb_dst - returns skb dst_entry
794 * @skb: buffer
795 *
796 * Returns skb dst_entry, regardless of reference taken or not.
797 */
adf30907
ED
798static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
799{
7fee226a
ED
800 /* If refdst was not refcounted, check we still are in a
801 * rcu_read_lock section
802 */
803 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
804 !rcu_read_lock_held() &&
805 !rcu_read_lock_bh_held());
806 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
adf30907
ED
807}
808
7fee226a
ED
809/**
810 * skb_dst_set - sets skb dst
811 * @skb: buffer
812 * @dst: dst entry
813 *
814 * Sets skb dst, assuming a reference was taken on dst and should
815 * be released by skb_dst_drop()
816 */
adf30907
ED
817static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
818{
7fee226a
ED
819 skb->_skb_refdst = (unsigned long)dst;
820}
821
932bc4d7
JA
822/**
823 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
824 * @skb: buffer
825 * @dst: dst entry
826 *
827 * Sets skb dst, assuming a reference was not taken on dst.
828 * If dst entry is cached, we do not take reference and dst_release
829 * will be avoided by refdst_drop. If dst entry is not cached, we take
830 * reference, so that last dst_release can destroy the dst immediately.
831 */
832static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
833{
dbfc4fb7
HFS
834 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
835 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
932bc4d7 836}
7fee226a
ED
837
838/**
25985edc 839 * skb_dst_is_noref - Test if skb dst isn't refcounted
7fee226a
ED
840 * @skb: buffer
841 */
842static inline bool skb_dst_is_noref(const struct sk_buff *skb)
843{
844 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
adf30907
ED
845}
846
511c3f92
ED
847static inline struct rtable *skb_rtable(const struct sk_buff *skb)
848{
adf30907 849 return (struct rtable *)skb_dst(skb);
511c3f92
ED
850}
851
8b10cab6
JHS
852/* For mangling skb->pkt_type from user space side from applications
853 * such as nft, tc, etc, we only allow a conservative subset of
854 * possible pkt_types to be set.
855*/
856static inline bool skb_pkt_type_ok(u32 ptype)
857{
858 return ptype <= PACKET_OTHERHOST;
859}
860
90b602f8
ML
861static inline unsigned int skb_napi_id(const struct sk_buff *skb)
862{
863#ifdef CONFIG_NET_RX_BUSY_POLL
864 return skb->napi_id;
865#else
866 return 0;
867#endif
868}
869
3889a803
PA
870/* decrement the reference count and return true if we can free the skb */
871static inline bool skb_unref(struct sk_buff *skb)
872{
873 if (unlikely(!skb))
874 return false;
63354797 875 if (likely(refcount_read(&skb->users) == 1))
3889a803 876 smp_rmb();
63354797 877 else if (likely(!refcount_dec_and_test(&skb->users)))
3889a803
PA
878 return false;
879
880 return true;
881}
882
0a463c78 883void skb_release_head_state(struct sk_buff *skb);
7965bd4d
JP
884void kfree_skb(struct sk_buff *skb);
885void kfree_skb_list(struct sk_buff *segs);
886void skb_tx_error(struct sk_buff *skb);
887void consume_skb(struct sk_buff *skb);
0a463c78 888void consume_stateless_skb(struct sk_buff *skb);
7965bd4d 889void __kfree_skb(struct sk_buff *skb);
d7e8883c 890extern struct kmem_cache *skbuff_head_cache;
bad43ca8 891
7965bd4d
JP
892void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
893bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
894 bool *fragstolen, int *delta_truesize);
bad43ca8 895
7965bd4d
JP
896struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
897 int node);
2ea2f62c 898struct sk_buff *__build_skb(void *data, unsigned int frag_size);
7965bd4d 899struct sk_buff *build_skb(void *data, unsigned int frag_size);
d179cd12 900static inline struct sk_buff *alloc_skb(unsigned int size,
dd0fc66f 901 gfp_t priority)
d179cd12 902{
564824b0 903 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
d179cd12
DM
904}
905
2e4e4410
ED
906struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
907 unsigned long data_len,
908 int max_page_order,
909 int *errcode,
910 gfp_t gfp_mask);
911
d0bf4a9e
ED
912/* Layout of fast clones : [skb1][skb2][fclone_ref] */
913struct sk_buff_fclones {
914 struct sk_buff skb1;
915
916 struct sk_buff skb2;
917
918 atomic_t fclone_ref;
919};
920
921/**
922 * skb_fclone_busy - check if fclone is busy
293de7de 923 * @sk: socket
d0bf4a9e
ED
924 * @skb: buffer
925 *
bda13fed 926 * Returns true if skb is a fast clone, and its clone is not freed.
39bb5e62
ED
927 * Some drivers call skb_orphan() in their ndo_start_xmit(),
928 * so we also check that this didnt happen.
d0bf4a9e 929 */
39bb5e62
ED
930static inline bool skb_fclone_busy(const struct sock *sk,
931 const struct sk_buff *skb)
d0bf4a9e
ED
932{
933 const struct sk_buff_fclones *fclones;
934
935 fclones = container_of(skb, struct sk_buff_fclones, skb1);
936
937 return skb->fclone == SKB_FCLONE_ORIG &&
6ffe75eb 938 atomic_read(&fclones->fclone_ref) > 1 &&
39bb5e62 939 fclones->skb2.sk == sk;
d0bf4a9e
ED
940}
941
d179cd12 942static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
dd0fc66f 943 gfp_t priority)
d179cd12 944{
c93bdd0e 945 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
d179cd12
DM
946}
947
7965bd4d 948struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
0ebd0ac5
PM
949static inline struct sk_buff *alloc_skb_head(gfp_t priority)
950{
951 return __alloc_skb_head(priority, -1);
952}
953
7965bd4d
JP
954struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
955int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
956struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
957struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
bad93e9d
OP
958struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
959 gfp_t gfp_mask, bool fclone);
960static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
961 gfp_t gfp_mask)
962{
963 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
964}
7965bd4d
JP
965
966int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
967struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
968 unsigned int headroom);
969struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
970 int newtailroom, gfp_t priority);
48a1df65
JD
971int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
972 int offset, int len);
973int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
974 int offset, int len);
7965bd4d
JP
975int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
976int skb_pad(struct sk_buff *skb, int pad);
ead2ceb0 977#define dev_kfree_skb(a) consume_skb(a)
1da177e4 978
7965bd4d
JP
979int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
980 int getfrag(void *from, char *to, int offset,
981 int len, int odd, struct sk_buff *skb),
982 void *from, int length);
e89e9cf5 983
be12a1fe
HFS
984int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
985 int offset, size_t size);
986
d94d9fee 987struct skb_seq_state {
677e90ed
TG
988 __u32 lower_offset;
989 __u32 upper_offset;
990 __u32 frag_idx;
991 __u32 stepped_offset;
992 struct sk_buff *root_skb;
993 struct sk_buff *cur_skb;
994 __u8 *frag_data;
995};
996
7965bd4d
JP
997void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
998 unsigned int to, struct skb_seq_state *st);
999unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1000 struct skb_seq_state *st);
1001void skb_abort_seq_read(struct skb_seq_state *st);
677e90ed 1002
7965bd4d 1003unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
059a2440 1004 unsigned int to, struct ts_config *config);
3fc7e8a6 1005
09323cc4
TH
1006/*
1007 * Packet hash types specify the type of hash in skb_set_hash.
1008 *
1009 * Hash types refer to the protocol layer addresses which are used to
1010 * construct a packet's hash. The hashes are used to differentiate or identify
1011 * flows of the protocol layer for the hash type. Hash types are either
1012 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1013 *
1014 * Properties of hashes:
1015 *
1016 * 1) Two packets in different flows have different hash values
1017 * 2) Two packets in the same flow should have the same hash value
1018 *
1019 * A hash at a higher layer is considered to be more specific. A driver should
1020 * set the most specific hash possible.
1021 *
1022 * A driver cannot indicate a more specific hash than the layer at which a hash
1023 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1024 *
1025 * A driver may indicate a hash level which is less specific than the
1026 * actual layer the hash was computed on. For instance, a hash computed
1027 * at L4 may be considered an L3 hash. This should only be done if the
1028 * driver can't unambiguously determine that the HW computed the hash at
1029 * the higher layer. Note that the "should" in the second property above
1030 * permits this.
1031 */
1032enum pkt_hash_types {
1033 PKT_HASH_TYPE_NONE, /* Undefined type */
1034 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1035 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1036 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1037};
1038
bcc83839 1039static inline void skb_clear_hash(struct sk_buff *skb)
09323cc4 1040{
bcc83839 1041 skb->hash = 0;
a3b18ddb 1042 skb->sw_hash = 0;
bcc83839
TH
1043 skb->l4_hash = 0;
1044}
1045
1046static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1047{
1048 if (!skb->l4_hash)
1049 skb_clear_hash(skb);
1050}
1051
1052static inline void
1053__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1054{
1055 skb->l4_hash = is_l4;
1056 skb->sw_hash = is_sw;
61b905da 1057 skb->hash = hash;
09323cc4
TH
1058}
1059
bcc83839
TH
1060static inline void
1061skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1062{
1063 /* Used by drivers to set hash from HW */
1064 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1065}
1066
1067static inline void
1068__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1069{
1070 __skb_set_hash(skb, hash, true, is_l4);
1071}
1072
e5276937 1073void __skb_get_hash(struct sk_buff *skb);
b917783c 1074u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
e5276937
TH
1075u32 skb_get_poff(const struct sk_buff *skb);
1076u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1077 const struct flow_keys *keys, int hlen);
1078__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1079 void *data, int hlen_proto);
1080
1081static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1082 int thoff, u8 ip_proto)
1083{
1084 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1085}
1086
1087void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1088 const struct flow_dissector_key *key,
1089 unsigned int key_count);
1090
1091bool __skb_flow_dissect(const struct sk_buff *skb,
1092 struct flow_dissector *flow_dissector,
1093 void *target_container,
cd79a238
TH
1094 void *data, __be16 proto, int nhoff, int hlen,
1095 unsigned int flags);
e5276937
TH
1096
1097static inline bool skb_flow_dissect(const struct sk_buff *skb,
1098 struct flow_dissector *flow_dissector,
cd79a238 1099 void *target_container, unsigned int flags)
e5276937
TH
1100{
1101 return __skb_flow_dissect(skb, flow_dissector, target_container,
cd79a238 1102 NULL, 0, 0, 0, flags);
e5276937
TH
1103}
1104
1105static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
cd79a238
TH
1106 struct flow_keys *flow,
1107 unsigned int flags)
e5276937
TH
1108{
1109 memset(flow, 0, sizeof(*flow));
1110 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
cd79a238 1111 NULL, 0, 0, 0, flags);
e5276937
TH
1112}
1113
1114static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1115 void *data, __be16 proto,
cd79a238
TH
1116 int nhoff, int hlen,
1117 unsigned int flags)
e5276937
TH
1118{
1119 memset(flow, 0, sizeof(*flow));
1120 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
cd79a238 1121 data, proto, nhoff, hlen, flags);
e5276937
TH
1122}
1123
3958afa1 1124static inline __u32 skb_get_hash(struct sk_buff *skb)
bfb564e7 1125{
a3b18ddb 1126 if (!skb->l4_hash && !skb->sw_hash)
3958afa1 1127 __skb_get_hash(skb);
bfb564e7 1128
61b905da 1129 return skb->hash;
bfb564e7
KK
1130}
1131
20a17bf6 1132__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
f70ea018 1133
20a17bf6 1134static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
f70ea018 1135{
c6cc1ca7
TH
1136 if (!skb->l4_hash && !skb->sw_hash) {
1137 struct flow_keys keys;
de4c1f8b 1138 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
c6cc1ca7 1139
de4c1f8b 1140 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
c6cc1ca7 1141 }
f70ea018
TH
1142
1143 return skb->hash;
1144}
1145
20a17bf6 1146__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
f70ea018 1147
20a17bf6 1148static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
f70ea018 1149{
c6cc1ca7
TH
1150 if (!skb->l4_hash && !skb->sw_hash) {
1151 struct flow_keys keys;
de4c1f8b 1152 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
c6cc1ca7 1153
de4c1f8b 1154 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
c6cc1ca7 1155 }
f70ea018
TH
1156
1157 return skb->hash;
1158}
1159
50fb7992
TH
1160__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1161
57bdf7f4
TH
1162static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1163{
61b905da 1164 return skb->hash;
57bdf7f4
TH
1165}
1166
3df7a74e
TH
1167static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1168{
61b905da 1169 to->hash = from->hash;
a3b18ddb 1170 to->sw_hash = from->sw_hash;
61b905da 1171 to->l4_hash = from->l4_hash;
3df7a74e
TH
1172};
1173
4305b541
ACM
1174#ifdef NET_SKBUFF_DATA_USES_OFFSET
1175static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1176{
1177 return skb->head + skb->end;
1178}
ec47ea82
AD
1179
1180static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1181{
1182 return skb->end;
1183}
4305b541
ACM
1184#else
1185static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1186{
1187 return skb->end;
1188}
ec47ea82
AD
1189
1190static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1191{
1192 return skb->end - skb->head;
1193}
4305b541
ACM
1194#endif
1195
1da177e4 1196/* Internal */
4305b541 1197#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1da177e4 1198
ac45f602
PO
1199static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1200{
1201 return &skb_shinfo(skb)->hwtstamps;
1202}
1203
1da177e4
LT
1204/**
1205 * skb_queue_empty - check if a queue is empty
1206 * @list: queue head
1207 *
1208 * Returns true if the queue is empty, false otherwise.
1209 */
1210static inline int skb_queue_empty(const struct sk_buff_head *list)
1211{
fd44b93c 1212 return list->next == (const struct sk_buff *) list;
1da177e4
LT
1213}
1214
fc7ebb21
DM
1215/**
1216 * skb_queue_is_last - check if skb is the last entry in the queue
1217 * @list: queue head
1218 * @skb: buffer
1219 *
1220 * Returns true if @skb is the last buffer on the list.
1221 */
1222static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1223 const struct sk_buff *skb)
1224{
fd44b93c 1225 return skb->next == (const struct sk_buff *) list;
fc7ebb21
DM
1226}
1227
832d11c5
IJ
1228/**
1229 * skb_queue_is_first - check if skb is the first entry in the queue
1230 * @list: queue head
1231 * @skb: buffer
1232 *
1233 * Returns true if @skb is the first buffer on the list.
1234 */
1235static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1236 const struct sk_buff *skb)
1237{
fd44b93c 1238 return skb->prev == (const struct sk_buff *) list;
832d11c5
IJ
1239}
1240
249c8b42
DM
1241/**
1242 * skb_queue_next - return the next packet in the queue
1243 * @list: queue head
1244 * @skb: current buffer
1245 *
1246 * Return the next packet in @list after @skb. It is only valid to
1247 * call this if skb_queue_is_last() evaluates to false.
1248 */
1249static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1250 const struct sk_buff *skb)
1251{
1252 /* This BUG_ON may seem severe, but if we just return then we
1253 * are going to dereference garbage.
1254 */
1255 BUG_ON(skb_queue_is_last(list, skb));
1256 return skb->next;
1257}
1258
832d11c5
IJ
1259/**
1260 * skb_queue_prev - return the prev packet in the queue
1261 * @list: queue head
1262 * @skb: current buffer
1263 *
1264 * Return the prev packet in @list before @skb. It is only valid to
1265 * call this if skb_queue_is_first() evaluates to false.
1266 */
1267static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1268 const struct sk_buff *skb)
1269{
1270 /* This BUG_ON may seem severe, but if we just return then we
1271 * are going to dereference garbage.
1272 */
1273 BUG_ON(skb_queue_is_first(list, skb));
1274 return skb->prev;
1275}
1276
1da177e4
LT
1277/**
1278 * skb_get - reference buffer
1279 * @skb: buffer to reference
1280 *
1281 * Makes another reference to a socket buffer and returns a pointer
1282 * to the buffer.
1283 */
1284static inline struct sk_buff *skb_get(struct sk_buff *skb)
1285{
63354797 1286 refcount_inc(&skb->users);
1da177e4
LT
1287 return skb;
1288}
1289
1290/*
1291 * If users == 1, we are the only owner and are can avoid redundant
1292 * atomic change.
1293 */
1294
1da177e4
LT
1295/**
1296 * skb_cloned - is the buffer a clone
1297 * @skb: buffer to check
1298 *
1299 * Returns true if the buffer was generated with skb_clone() and is
1300 * one of multiple shared copies of the buffer. Cloned buffers are
1301 * shared data so must not be written to under normal circumstances.
1302 */
1303static inline int skb_cloned(const struct sk_buff *skb)
1304{
1305 return skb->cloned &&
1306 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1307}
1308
14bbd6a5
PS
1309static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1310{
d0164adc 1311 might_sleep_if(gfpflags_allow_blocking(pri));
14bbd6a5
PS
1312
1313 if (skb_cloned(skb))
1314 return pskb_expand_head(skb, 0, 0, pri);
1315
1316 return 0;
1317}
1318
1da177e4
LT
1319/**
1320 * skb_header_cloned - is the header a clone
1321 * @skb: buffer to check
1322 *
1323 * Returns true if modifying the header part of the buffer requires
1324 * the data to be copied.
1325 */
1326static inline int skb_header_cloned(const struct sk_buff *skb)
1327{
1328 int dataref;
1329
1330 if (!skb->cloned)
1331 return 0;
1332
1333 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1334 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1335 return dataref != 1;
1336}
1337
9580bf2e
ED
1338static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1339{
1340 might_sleep_if(gfpflags_allow_blocking(pri));
1341
1342 if (skb_header_cloned(skb))
1343 return pskb_expand_head(skb, 0, 0, pri);
1344
1345 return 0;
1346}
1347
1da177e4
LT
1348/**
1349 * skb_header_release - release reference to header
1350 * @skb: buffer to operate on
1351 *
1352 * Drop a reference to the header part of the buffer. This is done
1353 * by acquiring a payload reference. You must not read from the header
1354 * part of skb->data after this.
f4a775d1 1355 * Note : Check if you can use __skb_header_release() instead.
1da177e4
LT
1356 */
1357static inline void skb_header_release(struct sk_buff *skb)
1358{
1359 BUG_ON(skb->nohdr);
1360 skb->nohdr = 1;
1361 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1362}
1363
f4a775d1
ED
1364/**
1365 * __skb_header_release - release reference to header
1366 * @skb: buffer to operate on
1367 *
1368 * Variant of skb_header_release() assuming skb is private to caller.
1369 * We can avoid one atomic operation.
1370 */
1371static inline void __skb_header_release(struct sk_buff *skb)
1372{
1373 skb->nohdr = 1;
1374 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1375}
1376
1377
1da177e4
LT
1378/**
1379 * skb_shared - is the buffer shared
1380 * @skb: buffer to check
1381 *
1382 * Returns true if more than one person has a reference to this
1383 * buffer.
1384 */
1385static inline int skb_shared(const struct sk_buff *skb)
1386{
63354797 1387 return refcount_read(&skb->users) != 1;
1da177e4
LT
1388}
1389
1390/**
1391 * skb_share_check - check if buffer is shared and if so clone it
1392 * @skb: buffer to check
1393 * @pri: priority for memory allocation
1394 *
1395 * If the buffer is shared the buffer is cloned and the old copy
1396 * drops a reference. A new clone with a single reference is returned.
1397 * If the buffer is not shared the original buffer is returned. When
1398 * being called from interrupt status or with spinlocks held pri must
1399 * be GFP_ATOMIC.
1400 *
1401 * NULL is returned on a memory allocation failure.
1402 */
47061bc4 1403static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1da177e4 1404{
d0164adc 1405 might_sleep_if(gfpflags_allow_blocking(pri));
1da177e4
LT
1406 if (skb_shared(skb)) {
1407 struct sk_buff *nskb = skb_clone(skb, pri);
47061bc4
ED
1408
1409 if (likely(nskb))
1410 consume_skb(skb);
1411 else
1412 kfree_skb(skb);
1da177e4
LT
1413 skb = nskb;
1414 }
1415 return skb;
1416}
1417
1418/*
1419 * Copy shared buffers into a new sk_buff. We effectively do COW on
1420 * packets to handle cases where we have a local reader and forward
1421 * and a couple of other messy ones. The normal one is tcpdumping
1422 * a packet thats being forwarded.
1423 */
1424
1425/**
1426 * skb_unshare - make a copy of a shared buffer
1427 * @skb: buffer to check
1428 * @pri: priority for memory allocation
1429 *
1430 * If the socket buffer is a clone then this function creates a new
1431 * copy of the data, drops a reference count on the old copy and returns
1432 * the new copy with the reference count at 1. If the buffer is not a clone
1433 * the original buffer is returned. When called with a spinlock held or
1434 * from interrupt state @pri must be %GFP_ATOMIC
1435 *
1436 * %NULL is returned on a memory allocation failure.
1437 */
e2bf521d 1438static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
dd0fc66f 1439 gfp_t pri)
1da177e4 1440{
d0164adc 1441 might_sleep_if(gfpflags_allow_blocking(pri));
1da177e4
LT
1442 if (skb_cloned(skb)) {
1443 struct sk_buff *nskb = skb_copy(skb, pri);
31eff81e
AA
1444
1445 /* Free our shared copy */
1446 if (likely(nskb))
1447 consume_skb(skb);
1448 else
1449 kfree_skb(skb);
1da177e4
LT
1450 skb = nskb;
1451 }
1452 return skb;
1453}
1454
1455/**
1a5778aa 1456 * skb_peek - peek at the head of an &sk_buff_head
1da177e4
LT
1457 * @list_: list to peek at
1458 *
1459 * Peek an &sk_buff. Unlike most other operations you _MUST_
1460 * be careful with this one. A peek leaves the buffer on the
1461 * list and someone else may run off with it. You must hold
1462 * the appropriate locks or have a private queue to do this.
1463 *
1464 * Returns %NULL for an empty list or a pointer to the head element.
1465 * The reference count is not incremented and the reference is therefore
1466 * volatile. Use with caution.
1467 */
05bdd2f1 1468static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1da177e4 1469{
18d07000
ED
1470 struct sk_buff *skb = list_->next;
1471
1472 if (skb == (struct sk_buff *)list_)
1473 skb = NULL;
1474 return skb;
1da177e4
LT
1475}
1476
da5ef6e5
PE
1477/**
1478 * skb_peek_next - peek skb following the given one from a queue
1479 * @skb: skb to start from
1480 * @list_: list to peek at
1481 *
1482 * Returns %NULL when the end of the list is met or a pointer to the
1483 * next element. The reference count is not incremented and the
1484 * reference is therefore volatile. Use with caution.
1485 */
1486static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1487 const struct sk_buff_head *list_)
1488{
1489 struct sk_buff *next = skb->next;
18d07000 1490
da5ef6e5
PE
1491 if (next == (struct sk_buff *)list_)
1492 next = NULL;
1493 return next;
1494}
1495
1da177e4 1496/**
1a5778aa 1497 * skb_peek_tail - peek at the tail of an &sk_buff_head
1da177e4
LT
1498 * @list_: list to peek at
1499 *
1500 * Peek an &sk_buff. Unlike most other operations you _MUST_
1501 * be careful with this one. A peek leaves the buffer on the
1502 * list and someone else may run off with it. You must hold
1503 * the appropriate locks or have a private queue to do this.
1504 *
1505 * Returns %NULL for an empty list or a pointer to the tail element.
1506 * The reference count is not incremented and the reference is therefore
1507 * volatile. Use with caution.
1508 */
05bdd2f1 1509static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1da177e4 1510{
18d07000
ED
1511 struct sk_buff *skb = list_->prev;
1512
1513 if (skb == (struct sk_buff *)list_)
1514 skb = NULL;
1515 return skb;
1516
1da177e4
LT
1517}
1518
1519/**
1520 * skb_queue_len - get queue length
1521 * @list_: list to measure
1522 *
1523 * Return the length of an &sk_buff queue.
1524 */
1525static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1526{
1527 return list_->qlen;
1528}
1529
67fed459
DM
1530/**
1531 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1532 * @list: queue to initialize
1533 *
1534 * This initializes only the list and queue length aspects of
1535 * an sk_buff_head object. This allows to initialize the list
1536 * aspects of an sk_buff_head without reinitializing things like
1537 * the spinlock. It can also be used for on-stack sk_buff_head
1538 * objects where the spinlock is known to not be used.
1539 */
1540static inline void __skb_queue_head_init(struct sk_buff_head *list)
1541{
1542 list->prev = list->next = (struct sk_buff *)list;
1543 list->qlen = 0;
1544}
1545
76f10ad0
AV
1546/*
1547 * This function creates a split out lock class for each invocation;
1548 * this is needed for now since a whole lot of users of the skb-queue
1549 * infrastructure in drivers have different locking usage (in hardirq)
1550 * than the networking core (in softirq only). In the long run either the
1551 * network layer or drivers should need annotation to consolidate the
1552 * main types of usage into 3 classes.
1553 */
1da177e4
LT
1554static inline void skb_queue_head_init(struct sk_buff_head *list)
1555{
1556 spin_lock_init(&list->lock);
67fed459 1557 __skb_queue_head_init(list);
1da177e4
LT
1558}
1559
c2ecba71
PE
1560static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1561 struct lock_class_key *class)
1562{
1563 skb_queue_head_init(list);
1564 lockdep_set_class(&list->lock, class);
1565}
1566
1da177e4 1567/*
bf299275 1568 * Insert an sk_buff on a list.
1da177e4
LT
1569 *
1570 * The "__skb_xxxx()" functions are the non-atomic ones that
1571 * can only be called with interrupts disabled.
1572 */
7965bd4d
JP
1573void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1574 struct sk_buff_head *list);
bf299275
GR
1575static inline void __skb_insert(struct sk_buff *newsk,
1576 struct sk_buff *prev, struct sk_buff *next,
1577 struct sk_buff_head *list)
1578{
1579 newsk->next = next;
1580 newsk->prev = prev;
1581 next->prev = prev->next = newsk;
1582 list->qlen++;
1583}
1da177e4 1584
67fed459
DM
1585static inline void __skb_queue_splice(const struct sk_buff_head *list,
1586 struct sk_buff *prev,
1587 struct sk_buff *next)
1588{
1589 struct sk_buff *first = list->next;
1590 struct sk_buff *last = list->prev;
1591
1592 first->prev = prev;
1593 prev->next = first;
1594
1595 last->next = next;
1596 next->prev = last;
1597}
1598
1599/**
1600 * skb_queue_splice - join two skb lists, this is designed for stacks
1601 * @list: the new list to add
1602 * @head: the place to add it in the first list
1603 */
1604static inline void skb_queue_splice(const struct sk_buff_head *list,
1605 struct sk_buff_head *head)
1606{
1607 if (!skb_queue_empty(list)) {
1608 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1d4a31dd 1609 head->qlen += list->qlen;
67fed459
DM
1610 }
1611}
1612
1613/**
d9619496 1614 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
67fed459
DM
1615 * @list: the new list to add
1616 * @head: the place to add it in the first list
1617 *
1618 * The list at @list is reinitialised
1619 */
1620static inline void skb_queue_splice_init(struct sk_buff_head *list,
1621 struct sk_buff_head *head)
1622{
1623 if (!skb_queue_empty(list)) {
1624 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1d4a31dd 1625 head->qlen += list->qlen;
67fed459
DM
1626 __skb_queue_head_init(list);
1627 }
1628}
1629
1630/**
1631 * skb_queue_splice_tail - join two skb lists, each list being a queue
1632 * @list: the new list to add
1633 * @head: the place to add it in the first list
1634 */
1635static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1636 struct sk_buff_head *head)
1637{
1638 if (!skb_queue_empty(list)) {
1639 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1d4a31dd 1640 head->qlen += list->qlen;
67fed459
DM
1641 }
1642}
1643
1644/**
d9619496 1645 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
67fed459
DM
1646 * @list: the new list to add
1647 * @head: the place to add it in the first list
1648 *
1649 * Each of the lists is a queue.
1650 * The list at @list is reinitialised
1651 */
1652static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1653 struct sk_buff_head *head)
1654{
1655 if (!skb_queue_empty(list)) {
1656 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1d4a31dd 1657 head->qlen += list->qlen;
67fed459
DM
1658 __skb_queue_head_init(list);
1659 }
1660}
1661
1da177e4 1662/**
300ce174 1663 * __skb_queue_after - queue a buffer at the list head
1da177e4 1664 * @list: list to use
300ce174 1665 * @prev: place after this buffer
1da177e4
LT
1666 * @newsk: buffer to queue
1667 *
300ce174 1668 * Queue a buffer int the middle of a list. This function takes no locks
1da177e4
LT
1669 * and you must therefore hold required locks before calling it.
1670 *
1671 * A buffer cannot be placed on two lists at the same time.
1672 */
300ce174
SH
1673static inline void __skb_queue_after(struct sk_buff_head *list,
1674 struct sk_buff *prev,
1675 struct sk_buff *newsk)
1da177e4 1676{
bf299275 1677 __skb_insert(newsk, prev, prev->next, list);
1da177e4
LT
1678}
1679
7965bd4d
JP
1680void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1681 struct sk_buff_head *list);
7de6c033 1682
f5572855
GR
1683static inline void __skb_queue_before(struct sk_buff_head *list,
1684 struct sk_buff *next,
1685 struct sk_buff *newsk)
1686{
1687 __skb_insert(newsk, next->prev, next, list);
1688}
1689
300ce174
SH
1690/**
1691 * __skb_queue_head - queue a buffer at the list head
1692 * @list: list to use
1693 * @newsk: buffer to queue
1694 *
1695 * Queue a buffer at the start of a list. This function takes no locks
1696 * and you must therefore hold required locks before calling it.
1697 *
1698 * A buffer cannot be placed on two lists at the same time.
1699 */
7965bd4d 1700void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
300ce174
SH
1701static inline void __skb_queue_head(struct sk_buff_head *list,
1702 struct sk_buff *newsk)
1703{
1704 __skb_queue_after(list, (struct sk_buff *)list, newsk);
1705}
1706
1da177e4
LT
1707/**
1708 * __skb_queue_tail - queue a buffer at the list tail
1709 * @list: list to use
1710 * @newsk: buffer to queue
1711 *
1712 * Queue a buffer at the end of a list. This function takes no locks
1713 * and you must therefore hold required locks before calling it.
1714 *
1715 * A buffer cannot be placed on two lists at the same time.
1716 */
7965bd4d 1717void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1da177e4
LT
1718static inline void __skb_queue_tail(struct sk_buff_head *list,
1719 struct sk_buff *newsk)
1720{
f5572855 1721 __skb_queue_before(list, (struct sk_buff *)list, newsk);
1da177e4
LT
1722}
1723
1da177e4
LT
1724/*
1725 * remove sk_buff from list. _Must_ be called atomically, and with
1726 * the list known..
1727 */
7965bd4d 1728void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1da177e4
LT
1729static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1730{
1731 struct sk_buff *next, *prev;
1732
1733 list->qlen--;
1734 next = skb->next;
1735 prev = skb->prev;
1736 skb->next = skb->prev = NULL;
1da177e4
LT
1737 next->prev = prev;
1738 prev->next = next;
1739}
1740
f525c06d
GR
1741/**
1742 * __skb_dequeue - remove from the head of the queue
1743 * @list: list to dequeue from
1744 *
1745 * Remove the head of the list. This function does not take any locks
1746 * so must be used with appropriate locks held only. The head item is
1747 * returned or %NULL if the list is empty.
1748 */
7965bd4d 1749struct sk_buff *skb_dequeue(struct sk_buff_head *list);
f525c06d
GR
1750static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1751{
1752 struct sk_buff *skb = skb_peek(list);
1753 if (skb)
1754 __skb_unlink(skb, list);
1755 return skb;
1756}
1da177e4
LT
1757
1758/**
1759 * __skb_dequeue_tail - remove from the tail of the queue
1760 * @list: list to dequeue from
1761 *
1762 * Remove the tail of the list. This function does not take any locks
1763 * so must be used with appropriate locks held only. The tail item is
1764 * returned or %NULL if the list is empty.
1765 */
7965bd4d 1766struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
1da177e4
LT
1767static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1768{
1769 struct sk_buff *skb = skb_peek_tail(list);
1770 if (skb)
1771 __skb_unlink(skb, list);
1772 return skb;
1773}
1774
1775
bdcc0924 1776static inline bool skb_is_nonlinear(const struct sk_buff *skb)
1da177e4
LT
1777{
1778 return skb->data_len;
1779}
1780
1781static inline unsigned int skb_headlen(const struct sk_buff *skb)
1782{
1783 return skb->len - skb->data_len;
1784}
1785
c72d8cda 1786static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1da177e4 1787{
c72d8cda 1788 unsigned int i, len = 0;
1da177e4 1789
c72d8cda 1790 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
9e903e08 1791 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
1da177e4
LT
1792 return len + skb_headlen(skb);
1793}
1794
131ea667
IC
1795/**
1796 * __skb_fill_page_desc - initialise a paged fragment in an skb
1797 * @skb: buffer containing fragment to be initialised
1798 * @i: paged fragment index to initialise
1799 * @page: the page to use for this fragment
1800 * @off: the offset to the data with @page
1801 * @size: the length of the data
1802 *
1803 * Initialises the @i'th fragment of @skb to point to &size bytes at
1804 * offset @off within @page.
1805 *
1806 * Does not take any additional reference on the fragment.
1807 */
1808static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1809 struct page *page, int off, int size)
1da177e4
LT
1810{
1811 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1812
c48a11c7 1813 /*
2f064f34
MH
1814 * Propagate page pfmemalloc to the skb if we can. The problem is
1815 * that not all callers have unique ownership of the page but rely
1816 * on page_is_pfmemalloc doing the right thing(tm).
c48a11c7 1817 */
a8605c60 1818 frag->page.p = page;
1da177e4 1819 frag->page_offset = off;
9e903e08 1820 skb_frag_size_set(frag, size);
cca7af38
PE
1821
1822 page = compound_head(page);
2f064f34 1823 if (page_is_pfmemalloc(page))
cca7af38 1824 skb->pfmemalloc = true;
131ea667
IC
1825}
1826
1827/**
1828 * skb_fill_page_desc - initialise a paged fragment in an skb
1829 * @skb: buffer containing fragment to be initialised
1830 * @i: paged fragment index to initialise
1831 * @page: the page to use for this fragment
1832 * @off: the offset to the data with @page
1833 * @size: the length of the data
1834 *
1835 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
bc32383c 1836 * @skb to point to @size bytes at offset @off within @page. In
131ea667
IC
1837 * addition updates @skb such that @i is the last fragment.
1838 *
1839 * Does not take any additional reference on the fragment.
1840 */
1841static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1842 struct page *page, int off, int size)
1843{
1844 __skb_fill_page_desc(skb, i, page, off, size);
1da177e4
LT
1845 skb_shinfo(skb)->nr_frags = i + 1;
1846}
1847
7965bd4d
JP
1848void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1849 int size, unsigned int truesize);
654bed16 1850
f8e617e1
JW
1851void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1852 unsigned int truesize);
1853
1da177e4 1854#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
21dc3301 1855#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
1da177e4
LT
1856#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1857
27a884dc
ACM
1858#ifdef NET_SKBUFF_DATA_USES_OFFSET
1859static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1860{
1861 return skb->head + skb->tail;
1862}
1863
1864static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1865{
1866 skb->tail = skb->data - skb->head;
1867}
1868
1869static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1870{
1871 skb_reset_tail_pointer(skb);
1872 skb->tail += offset;
1873}
7cc46190 1874
27a884dc
ACM
1875#else /* NET_SKBUFF_DATA_USES_OFFSET */
1876static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1877{
1878 return skb->tail;
1879}
1880
1881static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1882{
1883 skb->tail = skb->data;
1884}
1885
1886static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1887{
1888 skb->tail = skb->data + offset;
1889}
4305b541 1890
27a884dc
ACM
1891#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1892
1da177e4
LT
1893/*
1894 * Add data to an sk_buff
1895 */
4df864c1
JB
1896void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1897void *skb_put(struct sk_buff *skb, unsigned int len);
1898static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
1da177e4 1899{
4df864c1 1900 void *tmp = skb_tail_pointer(skb);
1da177e4
LT
1901 SKB_LINEAR_ASSERT(skb);
1902 skb->tail += len;
1903 skb->len += len;
1904 return tmp;
1905}
1906
de77b966 1907static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
1908{
1909 void *tmp = __skb_put(skb, len);
1910
1911 memset(tmp, 0, len);
1912 return tmp;
1913}
1914
1915static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
1916 unsigned int len)
1917{
1918 void *tmp = __skb_put(skb, len);
1919
1920 memcpy(tmp, data, len);
1921 return tmp;
1922}
1923
1924static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
1925{
1926 *(u8 *)__skb_put(skb, 1) = val;
1927}
1928
83ad357d 1929static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
e45a79da 1930{
83ad357d 1931 void *tmp = skb_put(skb, len);
e45a79da
JB
1932
1933 memset(tmp, 0, len);
1934
1935 return tmp;
1936}
1937
59ae1d12
JB
1938static inline void *skb_put_data(struct sk_buff *skb, const void *data,
1939 unsigned int len)
1940{
1941 void *tmp = skb_put(skb, len);
1942
1943 memcpy(tmp, data, len);
1944
1945 return tmp;
1946}
1947
634fef61
JB
1948static inline void skb_put_u8(struct sk_buff *skb, u8 val)
1949{
1950 *(u8 *)skb_put(skb, 1) = val;
1951}
1952
d58ff351
JB
1953void *skb_push(struct sk_buff *skb, unsigned int len);
1954static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
1da177e4
LT
1955{
1956 skb->data -= len;
1957 skb->len += len;
1958 return skb->data;
1959}
1960
af72868b
JB
1961void *skb_pull(struct sk_buff *skb, unsigned int len);
1962static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
1da177e4
LT
1963{
1964 skb->len -= len;
1965 BUG_ON(skb->len < skb->data_len);
1966 return skb->data += len;
1967}
1968
af72868b 1969static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
47d29646
DM
1970{
1971 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1972}
1973
af72868b 1974void *__pskb_pull_tail(struct sk_buff *skb, int delta);
1da177e4 1975
af72868b 1976static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
1da177e4
LT
1977{
1978 if (len > skb_headlen(skb) &&
987c402a 1979 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
1da177e4
LT
1980 return NULL;
1981 skb->len -= len;
1982 return skb->data += len;
1983}
1984
af72868b 1985static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
1da177e4
LT
1986{
1987 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1988}
1989
1990static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1991{
1992 if (likely(len <= skb_headlen(skb)))
1993 return 1;
1994 if (unlikely(len > skb->len))
1995 return 0;
987c402a 1996 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
1da177e4
LT
1997}
1998
c8c8b127
ED
1999void skb_condense(struct sk_buff *skb);
2000
1da177e4
LT
2001/**
2002 * skb_headroom - bytes at buffer head
2003 * @skb: buffer to check
2004 *
2005 * Return the number of bytes of free space at the head of an &sk_buff.
2006 */
c2636b4d 2007static inline unsigned int skb_headroom(const struct sk_buff *skb)
1da177e4
LT
2008{
2009 return skb->data - skb->head;
2010}
2011
2012/**
2013 * skb_tailroom - bytes at buffer end
2014 * @skb: buffer to check
2015 *
2016 * Return the number of bytes of free space at the tail of an sk_buff
2017 */
2018static inline int skb_tailroom(const struct sk_buff *skb)
2019{
4305b541 2020 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
1da177e4
LT
2021}
2022
a21d4572
ED
2023/**
2024 * skb_availroom - bytes at buffer end
2025 * @skb: buffer to check
2026 *
2027 * Return the number of bytes of free space at the tail of an sk_buff
2028 * allocated by sk_stream_alloc()
2029 */
2030static inline int skb_availroom(const struct sk_buff *skb)
2031{
16fad69c
ED
2032 if (skb_is_nonlinear(skb))
2033 return 0;
2034
2035 return skb->end - skb->tail - skb->reserved_tailroom;
a21d4572
ED
2036}
2037
1da177e4
LT
2038/**
2039 * skb_reserve - adjust headroom
2040 * @skb: buffer to alter
2041 * @len: bytes to move
2042 *
2043 * Increase the headroom of an empty &sk_buff by reducing the tail
2044 * room. This is only allowed for an empty buffer.
2045 */
8243126c 2046static inline void skb_reserve(struct sk_buff *skb, int len)
1da177e4
LT
2047{
2048 skb->data += len;
2049 skb->tail += len;
2050}
2051
1837b2e2
BP
2052/**
2053 * skb_tailroom_reserve - adjust reserved_tailroom
2054 * @skb: buffer to alter
2055 * @mtu: maximum amount of headlen permitted
2056 * @needed_tailroom: minimum amount of reserved_tailroom
2057 *
2058 * Set reserved_tailroom so that headlen can be as large as possible but
2059 * not larger than mtu and tailroom cannot be smaller than
2060 * needed_tailroom.
2061 * The required headroom should already have been reserved before using
2062 * this function.
2063 */
2064static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2065 unsigned int needed_tailroom)
2066{
2067 SKB_LINEAR_ASSERT(skb);
2068 if (mtu < skb_tailroom(skb) - needed_tailroom)
2069 /* use at most mtu */
2070 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2071 else
2072 /* use up to all available space */
2073 skb->reserved_tailroom = needed_tailroom;
2074}
2075
8bce6d7d
TH
2076#define ENCAP_TYPE_ETHER 0
2077#define ENCAP_TYPE_IPPROTO 1
2078
2079static inline void skb_set_inner_protocol(struct sk_buff *skb,
2080 __be16 protocol)
2081{
2082 skb->inner_protocol = protocol;
2083 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2084}
2085
2086static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2087 __u8 ipproto)
2088{
2089 skb->inner_ipproto = ipproto;
2090 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2091}
2092
6a674e9c
JG
2093static inline void skb_reset_inner_headers(struct sk_buff *skb)
2094{
aefbd2b3 2095 skb->inner_mac_header = skb->mac_header;
6a674e9c
JG
2096 skb->inner_network_header = skb->network_header;
2097 skb->inner_transport_header = skb->transport_header;
2098}
2099
0b5c9db1
JP
2100static inline void skb_reset_mac_len(struct sk_buff *skb)
2101{
2102 skb->mac_len = skb->network_header - skb->mac_header;
2103}
2104
6a674e9c
JG
2105static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2106 *skb)
2107{
2108 return skb->head + skb->inner_transport_header;
2109}
2110
55dc5a9f
TH
2111static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2112{
2113 return skb_inner_transport_header(skb) - skb->data;
2114}
2115
6a674e9c
JG
2116static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2117{
2118 skb->inner_transport_header = skb->data - skb->head;
2119}
2120
2121static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2122 const int offset)
2123{
2124 skb_reset_inner_transport_header(skb);
2125 skb->inner_transport_header += offset;
2126}
2127
2128static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2129{
2130 return skb->head + skb->inner_network_header;
2131}
2132
2133static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2134{
2135 skb->inner_network_header = skb->data - skb->head;
2136}
2137
2138static inline void skb_set_inner_network_header(struct sk_buff *skb,
2139 const int offset)
2140{
2141 skb_reset_inner_network_header(skb);
2142 skb->inner_network_header += offset;
2143}
2144
aefbd2b3
PS
2145static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2146{
2147 return skb->head + skb->inner_mac_header;
2148}
2149
2150static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2151{
2152 skb->inner_mac_header = skb->data - skb->head;
2153}
2154
2155static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2156 const int offset)
2157{
2158 skb_reset_inner_mac_header(skb);
2159 skb->inner_mac_header += offset;
2160}
fda55eca
ED
2161static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2162{
35d04610 2163 return skb->transport_header != (typeof(skb->transport_header))~0U;
fda55eca
ED
2164}
2165
9c70220b
ACM
2166static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2167{
2e07fa9c 2168 return skb->head + skb->transport_header;
9c70220b
ACM
2169}
2170
badff6d0
ACM
2171static inline void skb_reset_transport_header(struct sk_buff *skb)
2172{
2e07fa9c 2173 skb->transport_header = skb->data - skb->head;
badff6d0
ACM
2174}
2175
967b05f6
ACM
2176static inline void skb_set_transport_header(struct sk_buff *skb,
2177 const int offset)
2178{
2e07fa9c
ACM
2179 skb_reset_transport_header(skb);
2180 skb->transport_header += offset;
ea2ae17d
ACM
2181}
2182
d56f90a7
ACM
2183static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2184{
2e07fa9c 2185 return skb->head + skb->network_header;
d56f90a7
ACM
2186}
2187
c1d2bbe1
ACM
2188static inline void skb_reset_network_header(struct sk_buff *skb)
2189{
2e07fa9c 2190 skb->network_header = skb->data - skb->head;
c1d2bbe1
ACM
2191}
2192
c14d2450
ACM
2193static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2194{
2e07fa9c
ACM
2195 skb_reset_network_header(skb);
2196 skb->network_header += offset;
c14d2450
ACM
2197}
2198
2e07fa9c 2199static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
bbe735e4 2200{
2e07fa9c 2201 return skb->head + skb->mac_header;
bbe735e4
ACM
2202}
2203
ea6da4fd
AV
2204static inline int skb_mac_offset(const struct sk_buff *skb)
2205{
2206 return skb_mac_header(skb) - skb->data;
2207}
2208
2e07fa9c 2209static inline int skb_mac_header_was_set(const struct sk_buff *skb)
cfe1fc77 2210{
35d04610 2211 return skb->mac_header != (typeof(skb->mac_header))~0U;
2e07fa9c
ACM
2212}
2213
2214static inline void skb_reset_mac_header(struct sk_buff *skb)
2215{
2216 skb->mac_header = skb->data - skb->head;
2217}
2218
2219static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2220{
2221 skb_reset_mac_header(skb);
2222 skb->mac_header += offset;
2223}
2224
0e3da5bb
TT
2225static inline void skb_pop_mac_header(struct sk_buff *skb)
2226{
2227 skb->mac_header = skb->network_header;
2228}
2229
fbbdb8f0
YX
2230static inline void skb_probe_transport_header(struct sk_buff *skb,
2231 const int offset_hint)
2232{
2233 struct flow_keys keys;
2234
2235 if (skb_transport_header_was_set(skb))
2236 return;
cd79a238 2237 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
42aecaa9 2238 skb_set_transport_header(skb, keys.control.thoff);
fbbdb8f0
YX
2239 else
2240 skb_set_transport_header(skb, offset_hint);
2241}
2242
03606895
ED
2243static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2244{
2245 if (skb_mac_header_was_set(skb)) {
2246 const unsigned char *old_mac = skb_mac_header(skb);
2247
2248 skb_set_mac_header(skb, -skb->mac_len);
2249 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2250 }
2251}
2252
04fb451e
MM
2253static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2254{
2255 return skb->csum_start - skb_headroom(skb);
2256}
2257
08b64fcc
AD
2258static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2259{
2260 return skb->head + skb->csum_start;
2261}
2262
2e07fa9c
ACM
2263static inline int skb_transport_offset(const struct sk_buff *skb)
2264{
2265 return skb_transport_header(skb) - skb->data;
2266}
2267
2268static inline u32 skb_network_header_len(const struct sk_buff *skb)
2269{
2270 return skb->transport_header - skb->network_header;
2271}
2272
6a674e9c
JG
2273static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2274{
2275 return skb->inner_transport_header - skb->inner_network_header;
2276}
2277
2e07fa9c
ACM
2278static inline int skb_network_offset(const struct sk_buff *skb)
2279{
2280 return skb_network_header(skb) - skb->data;
2281}
48d49d0c 2282
6a674e9c
JG
2283static inline int skb_inner_network_offset(const struct sk_buff *skb)
2284{
2285 return skb_inner_network_header(skb) - skb->data;
2286}
2287
f9599ce1
CG
2288static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2289{
2290 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2291}
2292
1da177e4
LT
2293/*
2294 * CPUs often take a performance hit when accessing unaligned memory
2295 * locations. The actual performance hit varies, it can be small if the
2296 * hardware handles it or large if we have to take an exception and fix it
2297 * in software.
2298 *
2299 * Since an ethernet header is 14 bytes network drivers often end up with
2300 * the IP header at an unaligned offset. The IP header can be aligned by
2301 * shifting the start of the packet by 2 bytes. Drivers should do this
2302 * with:
2303 *
8660c124 2304 * skb_reserve(skb, NET_IP_ALIGN);
1da177e4
LT
2305 *
2306 * The downside to this alignment of the IP header is that the DMA is now
2307 * unaligned. On some architectures the cost of an unaligned DMA is high
2308 * and this cost outweighs the gains made by aligning the IP header.
8660c124 2309 *
1da177e4
LT
2310 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2311 * to be overridden.
2312 */
2313#ifndef NET_IP_ALIGN
2314#define NET_IP_ALIGN 2
2315#endif
2316
025be81e
AB
2317/*
2318 * The networking layer reserves some headroom in skb data (via
2319 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2320 * the header has to grow. In the default case, if the header has to grow
d6301d3d 2321 * 32 bytes or less we avoid the reallocation.
025be81e
AB
2322 *
2323 * Unfortunately this headroom changes the DMA alignment of the resulting
2324 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2325 * on some architectures. An architecture can override this value,
2326 * perhaps setting it to a cacheline in size (since that will maintain
2327 * cacheline alignment of the DMA). It must be a power of 2.
2328 *
d6301d3d 2329 * Various parts of the networking layer expect at least 32 bytes of
025be81e 2330 * headroom, you should not reduce this.
5933dd2f
ED
2331 *
2332 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2333 * to reduce average number of cache lines per packet.
2334 * get_rps_cpus() for example only access one 64 bytes aligned block :
18e8c134 2335 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
025be81e
AB
2336 */
2337#ifndef NET_SKB_PAD
5933dd2f 2338#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
025be81e
AB
2339#endif
2340
7965bd4d 2341int ___pskb_trim(struct sk_buff *skb, unsigned int len);
1da177e4 2342
5293efe6 2343static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
1da177e4 2344{
c4264f27 2345 if (unlikely(skb_is_nonlinear(skb))) {
3cc0e873
HX
2346 WARN_ON(1);
2347 return;
2348 }
27a884dc
ACM
2349 skb->len = len;
2350 skb_set_tail_pointer(skb, len);
1da177e4
LT
2351}
2352
5293efe6
DB
2353static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2354{
2355 __skb_set_length(skb, len);
2356}
2357
7965bd4d 2358void skb_trim(struct sk_buff *skb, unsigned int len);
1da177e4
LT
2359
2360static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2361{
3cc0e873
HX
2362 if (skb->data_len)
2363 return ___pskb_trim(skb, len);
2364 __skb_trim(skb, len);
2365 return 0;
1da177e4
LT
2366}
2367
2368static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2369{
2370 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2371}
2372
e9fa4f7b
HX
2373/**
2374 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2375 * @skb: buffer to alter
2376 * @len: new length
2377 *
2378 * This is identical to pskb_trim except that the caller knows that
2379 * the skb is not cloned so we should never get an error due to out-
2380 * of-memory.
2381 */
2382static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2383{
2384 int err = pskb_trim(skb, len);
2385 BUG_ON(err);
2386}
2387
5293efe6
DB
2388static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2389{
2390 unsigned int diff = len - skb->len;
2391
2392 if (skb_tailroom(skb) < diff) {
2393 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2394 GFP_ATOMIC);
2395 if (ret)
2396 return ret;
2397 }
2398 __skb_set_length(skb, len);
2399 return 0;
2400}
2401
1da177e4
LT
2402/**
2403 * skb_orphan - orphan a buffer
2404 * @skb: buffer to orphan
2405 *
2406 * If a buffer currently has an owner then we call the owner's
2407 * destructor function and make the @skb unowned. The buffer continues
2408 * to exist but is no longer charged to its former owner.
2409 */
2410static inline void skb_orphan(struct sk_buff *skb)
2411{
c34a7612 2412 if (skb->destructor) {
1da177e4 2413 skb->destructor(skb);
c34a7612
ED
2414 skb->destructor = NULL;
2415 skb->sk = NULL;
376c7311
ED
2416 } else {
2417 BUG_ON(skb->sk);
c34a7612 2418 }
1da177e4
LT
2419}
2420
a353e0ce
MT
2421/**
2422 * skb_orphan_frags - orphan the frags contained in a buffer
2423 * @skb: buffer to orphan frags from
2424 * @gfp_mask: allocation mask for replacement pages
2425 *
2426 * For each frag in the SKB which needs a destructor (i.e. has an
2427 * owner) create a copy of that frag and release the original
2428 * page by calling the destructor.
2429 */
2430static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2431{
2432 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2433 return 0;
2434 return skb_copy_ubufs(skb, gfp_mask);
2435}
2436
1da177e4
LT
2437/**
2438 * __skb_queue_purge - empty a list
2439 * @list: list to empty
2440 *
2441 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2442 * the list and one reference dropped. This function does not take the
2443 * list lock and the caller must hold the relevant locks to use it.
2444 */
7965bd4d 2445void skb_queue_purge(struct sk_buff_head *list);
1da177e4
LT
2446static inline void __skb_queue_purge(struct sk_buff_head *list)
2447{
2448 struct sk_buff *skb;
2449 while ((skb = __skb_dequeue(list)) != NULL)
2450 kfree_skb(skb);
2451}
2452
9f5afeae
YW
2453void skb_rbtree_purge(struct rb_root *root);
2454
7965bd4d 2455void *netdev_alloc_frag(unsigned int fragsz);
1da177e4 2456
7965bd4d
JP
2457struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2458 gfp_t gfp_mask);
8af27456
CH
2459
2460/**
2461 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2462 * @dev: network device to receive on
2463 * @length: length to allocate
2464 *
2465 * Allocate a new &sk_buff and assign it a usage count of one. The
2466 * buffer has unspecified headroom built in. Users should allocate
2467 * the headroom they think they need without accounting for the
2468 * built in space. The built in space is used for optimisations.
2469 *
2470 * %NULL is returned if there is no free memory. Although this function
2471 * allocates memory it can be called from an interrupt.
2472 */
2473static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
6f532612 2474 unsigned int length)
8af27456
CH
2475{
2476 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2477}
2478
6f532612
ED
2479/* legacy helper around __netdev_alloc_skb() */
2480static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2481 gfp_t gfp_mask)
2482{
2483 return __netdev_alloc_skb(NULL, length, gfp_mask);
2484}
2485
2486/* legacy helper around netdev_alloc_skb() */
2487static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2488{
2489 return netdev_alloc_skb(NULL, length);
2490}
2491
2492
4915a0de
ED
2493static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2494 unsigned int length, gfp_t gfp)
61321bbd 2495{
4915a0de 2496 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
61321bbd
ED
2497
2498 if (NET_IP_ALIGN && skb)
2499 skb_reserve(skb, NET_IP_ALIGN);
2500 return skb;
2501}
2502
4915a0de
ED
2503static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2504 unsigned int length)
2505{
2506 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2507}
2508
181edb2b
AD
2509static inline void skb_free_frag(void *addr)
2510{
8c2dd3e4 2511 page_frag_free(addr);
181edb2b
AD
2512}
2513
ffde7328 2514void *napi_alloc_frag(unsigned int fragsz);
fd11a83d
AD
2515struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2516 unsigned int length, gfp_t gfp_mask);
2517static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2518 unsigned int length)
2519{
2520 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2521}
795bb1c0
JDB
2522void napi_consume_skb(struct sk_buff *skb, int budget);
2523
2524void __kfree_skb_flush(void);
15fad714 2525void __kfree_skb_defer(struct sk_buff *skb);
ffde7328 2526
71dfda58
AD
2527/**
2528 * __dev_alloc_pages - allocate page for network Rx
2529 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2530 * @order: size of the allocation
2531 *
2532 * Allocate a new page.
2533 *
2534 * %NULL is returned if there is no free memory.
2535*/
2536static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2537 unsigned int order)
2538{
2539 /* This piece of code contains several assumptions.
2540 * 1. This is for device Rx, therefor a cold page is preferred.
2541 * 2. The expectation is the user wants a compound page.
2542 * 3. If requesting a order 0 page it will not be compound
2543 * due to the check to see if order has a value in prep_new_page
2544 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2545 * code in gfp_to_alloc_flags that should be enforcing this.
2546 */
2547 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2548
2549 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2550}
2551
2552static inline struct page *dev_alloc_pages(unsigned int order)
2553{
95829b3a 2554 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
71dfda58
AD
2555}
2556
2557/**
2558 * __dev_alloc_page - allocate a page for network Rx
2559 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2560 *
2561 * Allocate a new page.
2562 *
2563 * %NULL is returned if there is no free memory.
2564 */
2565static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2566{
2567 return __dev_alloc_pages(gfp_mask, 0);
2568}
2569
2570static inline struct page *dev_alloc_page(void)
2571{
95829b3a 2572 return dev_alloc_pages(0);
71dfda58
AD
2573}
2574
0614002b
MG
2575/**
2576 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2577 * @page: The page that was allocated from skb_alloc_page
2578 * @skb: The skb that may need pfmemalloc set
2579 */
2580static inline void skb_propagate_pfmemalloc(struct page *page,
2581 struct sk_buff *skb)
2582{
2f064f34 2583 if (page_is_pfmemalloc(page))
0614002b
MG
2584 skb->pfmemalloc = true;
2585}
2586
131ea667 2587/**
e227867f 2588 * skb_frag_page - retrieve the page referred to by a paged fragment
131ea667
IC
2589 * @frag: the paged fragment
2590 *
2591 * Returns the &struct page associated with @frag.
2592 */
2593static inline struct page *skb_frag_page(const skb_frag_t *frag)
2594{
a8605c60 2595 return frag->page.p;
131ea667
IC
2596}
2597
2598/**
2599 * __skb_frag_ref - take an addition reference on a paged fragment.
2600 * @frag: the paged fragment
2601 *
2602 * Takes an additional reference on the paged fragment @frag.
2603 */
2604static inline void __skb_frag_ref(skb_frag_t *frag)
2605{
2606 get_page(skb_frag_page(frag));
2607}
2608
2609/**
2610 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2611 * @skb: the buffer
2612 * @f: the fragment offset.
2613 *
2614 * Takes an additional reference on the @f'th paged fragment of @skb.
2615 */
2616static inline void skb_frag_ref(struct sk_buff *skb, int f)
2617{
2618 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2619}
2620
2621/**
2622 * __skb_frag_unref - release a reference on a paged fragment.
2623 * @frag: the paged fragment
2624 *
2625 * Releases a reference on the paged fragment @frag.
2626 */
2627static inline void __skb_frag_unref(skb_frag_t *frag)
2628{
2629 put_page(skb_frag_page(frag));
2630}
2631
2632/**
2633 * skb_frag_unref - release a reference on a paged fragment of an skb.
2634 * @skb: the buffer
2635 * @f: the fragment offset
2636 *
2637 * Releases a reference on the @f'th paged fragment of @skb.
2638 */
2639static inline void skb_frag_unref(struct sk_buff *skb, int f)
2640{
2641 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2642}
2643
2644/**
2645 * skb_frag_address - gets the address of the data contained in a paged fragment
2646 * @frag: the paged fragment buffer
2647 *
2648 * Returns the address of the data within @frag. The page must already
2649 * be mapped.
2650 */
2651static inline void *skb_frag_address(const skb_frag_t *frag)
2652{
2653 return page_address(skb_frag_page(frag)) + frag->page_offset;
2654}
2655
2656/**
2657 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2658 * @frag: the paged fragment buffer
2659 *
2660 * Returns the address of the data within @frag. Checks that the page
2661 * is mapped and returns %NULL otherwise.
2662 */
2663static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2664{
2665 void *ptr = page_address(skb_frag_page(frag));
2666 if (unlikely(!ptr))
2667 return NULL;
2668
2669 return ptr + frag->page_offset;
2670}
2671
2672/**
2673 * __skb_frag_set_page - sets the page contained in a paged fragment
2674 * @frag: the paged fragment
2675 * @page: the page to set
2676 *
2677 * Sets the fragment @frag to contain @page.
2678 */
2679static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2680{
a8605c60 2681 frag->page.p = page;
131ea667
IC
2682}
2683
2684/**
2685 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2686 * @skb: the buffer
2687 * @f: the fragment offset
2688 * @page: the page to set
2689 *
2690 * Sets the @f'th fragment of @skb to contain @page.
2691 */
2692static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2693 struct page *page)
2694{
2695 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2696}
2697
400dfd3a
ED
2698bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2699
131ea667
IC
2700/**
2701 * skb_frag_dma_map - maps a paged fragment via the DMA API
f83347df 2702 * @dev: the device to map the fragment to
131ea667
IC
2703 * @frag: the paged fragment to map
2704 * @offset: the offset within the fragment (starting at the
2705 * fragment's own offset)
2706 * @size: the number of bytes to map
f83347df 2707 * @dir: the direction of the mapping (%PCI_DMA_*)
131ea667
IC
2708 *
2709 * Maps the page associated with @frag to @device.
2710 */
2711static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2712 const skb_frag_t *frag,
2713 size_t offset, size_t size,
2714 enum dma_data_direction dir)
2715{
2716 return dma_map_page(dev, skb_frag_page(frag),
2717 frag->page_offset + offset, size, dir);
2718}
2719
117632e6
ED
2720static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2721 gfp_t gfp_mask)
2722{
2723 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2724}
2725
bad93e9d
OP
2726
2727static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2728 gfp_t gfp_mask)
2729{
2730 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2731}
2732
2733
334a8132
PM
2734/**
2735 * skb_clone_writable - is the header of a clone writable
2736 * @skb: buffer to check
2737 * @len: length up to which to write
2738 *
2739 * Returns true if modifying the header part of the cloned buffer
2740 * does not requires the data to be copied.
2741 */
05bdd2f1 2742static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
334a8132
PM
2743{
2744 return !skb_header_cloned(skb) &&
2745 skb_headroom(skb) + len <= skb->hdr_len;
2746}
2747
3697649f
DB
2748static inline int skb_try_make_writable(struct sk_buff *skb,
2749 unsigned int write_len)
2750{
2751 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2752 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2753}
2754
d9cc2048
HX
2755static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2756 int cloned)
2757{
2758 int delta = 0;
2759
d9cc2048
HX
2760 if (headroom > skb_headroom(skb))
2761 delta = headroom - skb_headroom(skb);
2762
2763 if (delta || cloned)
2764 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2765 GFP_ATOMIC);
2766 return 0;
2767}
2768
1da177e4
LT
2769/**
2770 * skb_cow - copy header of skb when it is required
2771 * @skb: buffer to cow
2772 * @headroom: needed headroom
2773 *
2774 * If the skb passed lacks sufficient headroom or its data part
2775 * is shared, data is reallocated. If reallocation fails, an error
2776 * is returned and original skb is not changed.
2777 *
2778 * The result is skb with writable area skb->head...skb->tail
2779 * and at least @headroom of space at head.
2780 */
2781static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2782{
d9cc2048
HX
2783 return __skb_cow(skb, headroom, skb_cloned(skb));
2784}
1da177e4 2785
d9cc2048
HX
2786/**
2787 * skb_cow_head - skb_cow but only making the head writable
2788 * @skb: buffer to cow
2789 * @headroom: needed headroom
2790 *
2791 * This function is identical to skb_cow except that we replace the
2792 * skb_cloned check by skb_header_cloned. It should be used when
2793 * you only need to push on some header and do not need to modify
2794 * the data.
2795 */
2796static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2797{
2798 return __skb_cow(skb, headroom, skb_header_cloned(skb));
1da177e4
LT
2799}
2800
2801/**
2802 * skb_padto - pad an skbuff up to a minimal size
2803 * @skb: buffer to pad
2804 * @len: minimal length
2805 *
2806 * Pads up a buffer to ensure the trailing bytes exist and are
2807 * blanked. If the buffer already contains sufficient data it
5b057c6b
HX
2808 * is untouched. Otherwise it is extended. Returns zero on
2809 * success. The skb is freed on error.
1da177e4 2810 */
5b057c6b 2811static inline int skb_padto(struct sk_buff *skb, unsigned int len)
1da177e4
LT
2812{
2813 unsigned int size = skb->len;
2814 if (likely(size >= len))
5b057c6b 2815 return 0;
987c402a 2816 return skb_pad(skb, len - size);
1da177e4
LT
2817}
2818
9c0c1124
AD
2819/**
2820 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2821 * @skb: buffer to pad
2822 * @len: minimal length
2823 *
2824 * Pads up a buffer to ensure the trailing bytes exist and are
2825 * blanked. If the buffer already contains sufficient data it
2826 * is untouched. Otherwise it is extended. Returns zero on
2827 * success. The skb is freed on error.
2828 */
2829static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2830{
2831 unsigned int size = skb->len;
2832
2833 if (unlikely(size < len)) {
2834 len -= size;
2835 if (skb_pad(skb, len))
2836 return -ENOMEM;
2837 __skb_put(skb, len);
2838 }
2839 return 0;
2840}
2841
1da177e4 2842static inline int skb_add_data(struct sk_buff *skb,
af2b040e 2843 struct iov_iter *from, int copy)
1da177e4
LT
2844{
2845 const int off = skb->len;
2846
2847 if (skb->ip_summed == CHECKSUM_NONE) {
af2b040e 2848 __wsum csum = 0;
15e6cb46
AV
2849 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2850 &csum, from)) {
1da177e4
LT
2851 skb->csum = csum_block_add(skb->csum, csum, off);
2852 return 0;
2853 }
15e6cb46 2854 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
1da177e4
LT
2855 return 0;
2856
2857 __skb_trim(skb, off);
2858 return -EFAULT;
2859}
2860
38ba0a65
ED
2861static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2862 const struct page *page, int off)
1da177e4
LT
2863{
2864 if (i) {
9e903e08 2865 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
1da177e4 2866
ea2ab693 2867 return page == skb_frag_page(frag) &&
9e903e08 2868 off == frag->page_offset + skb_frag_size(frag);
1da177e4 2869 }
38ba0a65 2870 return false;
1da177e4
LT
2871}
2872
364c6bad
HX
2873static inline int __skb_linearize(struct sk_buff *skb)
2874{
2875 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2876}
2877
1da177e4
LT
2878/**
2879 * skb_linearize - convert paged skb to linear one
2880 * @skb: buffer to linarize
1da177e4
LT
2881 *
2882 * If there is no free memory -ENOMEM is returned, otherwise zero
2883 * is returned and the old skb data released.
2884 */
364c6bad
HX
2885static inline int skb_linearize(struct sk_buff *skb)
2886{
2887 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2888}
2889
cef401de
ED
2890/**
2891 * skb_has_shared_frag - can any frag be overwritten
2892 * @skb: buffer to test
2893 *
2894 * Return true if the skb has at least one frag that might be modified
2895 * by an external entity (as in vmsplice()/sendfile())
2896 */
2897static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2898{
c9af6db4
PS
2899 return skb_is_nonlinear(skb) &&
2900 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
cef401de
ED
2901}
2902
364c6bad
HX
2903/**
2904 * skb_linearize_cow - make sure skb is linear and writable
2905 * @skb: buffer to process
2906 *
2907 * If there is no free memory -ENOMEM is returned, otherwise zero
2908 * is returned and the old skb data released.
2909 */
2910static inline int skb_linearize_cow(struct sk_buff *skb)
1da177e4 2911{
364c6bad
HX
2912 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2913 __skb_linearize(skb) : 0;
1da177e4
LT
2914}
2915
479ffccc
DB
2916static __always_inline void
2917__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2918 unsigned int off)
2919{
2920 if (skb->ip_summed == CHECKSUM_COMPLETE)
2921 skb->csum = csum_block_sub(skb->csum,
2922 csum_partial(start, len, 0), off);
2923 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
2924 skb_checksum_start_offset(skb) < 0)
2925 skb->ip_summed = CHECKSUM_NONE;
2926}
2927
1da177e4
LT
2928/**
2929 * skb_postpull_rcsum - update checksum for received skb after pull
2930 * @skb: buffer to update
2931 * @start: start of data before pull
2932 * @len: length of data pulled
2933 *
2934 * After doing a pull on a received packet, you need to call this to
84fa7933
PM
2935 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2936 * CHECKSUM_NONE so that it can be recomputed from scratch.
1da177e4 2937 */
1da177e4 2938static inline void skb_postpull_rcsum(struct sk_buff *skb,
cbb042f9 2939 const void *start, unsigned int len)
1da177e4 2940{
479ffccc 2941 __skb_postpull_rcsum(skb, start, len, 0);
1da177e4
LT
2942}
2943
479ffccc
DB
2944static __always_inline void
2945__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
2946 unsigned int off)
2947{
2948 if (skb->ip_summed == CHECKSUM_COMPLETE)
2949 skb->csum = csum_block_add(skb->csum,
2950 csum_partial(start, len, 0), off);
2951}
cbb042f9 2952
479ffccc
DB
2953/**
2954 * skb_postpush_rcsum - update checksum for received skb after push
2955 * @skb: buffer to update
2956 * @start: start of data after push
2957 * @len: length of data pushed
2958 *
2959 * After doing a push on a received packet, you need to call this to
2960 * update the CHECKSUM_COMPLETE checksum.
2961 */
f8ffad69
DB
2962static inline void skb_postpush_rcsum(struct sk_buff *skb,
2963 const void *start, unsigned int len)
2964{
479ffccc 2965 __skb_postpush_rcsum(skb, start, len, 0);
f8ffad69
DB
2966}
2967
af72868b 2968void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
479ffccc 2969
82a31b92
WC
2970/**
2971 * skb_push_rcsum - push skb and update receive checksum
2972 * @skb: buffer to update
2973 * @len: length of data pulled
2974 *
2975 * This function performs an skb_push on the packet and updates
2976 * the CHECKSUM_COMPLETE checksum. It should be used on
2977 * receive path processing instead of skb_push unless you know
2978 * that the checksum difference is zero (e.g., a valid IP header)
2979 * or you are setting ip_summed to CHECKSUM_NONE.
2980 */
d58ff351 2981static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
82a31b92
WC
2982{
2983 skb_push(skb, len);
2984 skb_postpush_rcsum(skb, skb->data, len);
2985 return skb->data;
2986}
2987
7ce5a27f
DM
2988/**
2989 * pskb_trim_rcsum - trim received skb and update checksum
2990 * @skb: buffer to trim
2991 * @len: new length
2992 *
2993 * This is exactly the same as pskb_trim except that it ensures the
2994 * checksum of received packets are still valid after the operation.
2995 */
2996
2997static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2998{
2999 if (likely(len >= skb->len))
3000 return 0;
3001 if (skb->ip_summed == CHECKSUM_COMPLETE)
3002 skb->ip_summed = CHECKSUM_NONE;
3003 return __pskb_trim(skb, len);
3004}
3005
5293efe6
DB
3006static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3007{
3008 if (skb->ip_summed == CHECKSUM_COMPLETE)
3009 skb->ip_summed = CHECKSUM_NONE;
3010 __skb_trim(skb, len);
3011 return 0;
3012}
3013
3014static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3015{
3016 if (skb->ip_summed == CHECKSUM_COMPLETE)
3017 skb->ip_summed = CHECKSUM_NONE;
3018 return __skb_grow(skb, len);
3019}
3020
1da177e4
LT
3021#define skb_queue_walk(queue, skb) \
3022 for (skb = (queue)->next; \
a1e4891f 3023 skb != (struct sk_buff *)(queue); \
1da177e4
LT
3024 skb = skb->next)
3025
46f8914e
JC
3026#define skb_queue_walk_safe(queue, skb, tmp) \
3027 for (skb = (queue)->next, tmp = skb->next; \
3028 skb != (struct sk_buff *)(queue); \
3029 skb = tmp, tmp = skb->next)
3030
1164f52a 3031#define skb_queue_walk_from(queue, skb) \
a1e4891f 3032 for (; skb != (struct sk_buff *)(queue); \
1164f52a
DM
3033 skb = skb->next)
3034
3035#define skb_queue_walk_from_safe(queue, skb, tmp) \
3036 for (tmp = skb->next; \
3037 skb != (struct sk_buff *)(queue); \
3038 skb = tmp, tmp = skb->next)
3039
300ce174
SH
3040#define skb_queue_reverse_walk(queue, skb) \
3041 for (skb = (queue)->prev; \
a1e4891f 3042 skb != (struct sk_buff *)(queue); \
300ce174
SH
3043 skb = skb->prev)
3044
686a2955
DM
3045#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3046 for (skb = (queue)->prev, tmp = skb->prev; \
3047 skb != (struct sk_buff *)(queue); \
3048 skb = tmp, tmp = skb->prev)
3049
3050#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3051 for (tmp = skb->prev; \
3052 skb != (struct sk_buff *)(queue); \
3053 skb = tmp, tmp = skb->prev)
1da177e4 3054
21dc3301 3055static inline bool skb_has_frag_list(const struct sk_buff *skb)
ee039871
DM
3056{
3057 return skb_shinfo(skb)->frag_list != NULL;
3058}
3059
3060static inline void skb_frag_list_init(struct sk_buff *skb)
3061{
3062 skb_shinfo(skb)->frag_list = NULL;
3063}
3064
ee039871
DM
3065#define skb_walk_frags(skb, iter) \
3066 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3067
ea3793ee
RW
3068
3069int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3070 const struct sk_buff *skb);
65101aec
PA
3071struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3072 struct sk_buff_head *queue,
3073 unsigned int flags,
3074 void (*destructor)(struct sock *sk,
3075 struct sk_buff *skb),
3076 int *peeked, int *off, int *err,
3077 struct sk_buff **last);
ea3793ee 3078struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
7c13f97f
PA
3079 void (*destructor)(struct sock *sk,
3080 struct sk_buff *skb),
ea3793ee
RW
3081 int *peeked, int *off, int *err,
3082 struct sk_buff **last);
7965bd4d 3083struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
7c13f97f
PA
3084 void (*destructor)(struct sock *sk,
3085 struct sk_buff *skb),
7965bd4d
JP
3086 int *peeked, int *off, int *err);
3087struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3088 int *err);
3089unsigned int datagram_poll(struct file *file, struct socket *sock,
3090 struct poll_table_struct *wait);
c0371da6
AV
3091int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3092 struct iov_iter *to, int size);
51f3d02b
DM
3093static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3094 struct msghdr *msg, int size)
3095{
e5a4b0bb 3096 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
227158db 3097}
e5a4b0bb
AV
3098int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3099 struct msghdr *msg);
3a654f97
AV
3100int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3101 struct iov_iter *from, int len);
3a654f97 3102int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
7965bd4d 3103void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
627d2d6b 3104void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3105static inline void skb_free_datagram_locked(struct sock *sk,
3106 struct sk_buff *skb)
3107{
3108 __skb_free_datagram_locked(sk, skb, 0);
3109}
7965bd4d 3110int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
7965bd4d
JP
3111int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3112int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3113__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3114 int len, __wsum csum);
a60e3cc7 3115int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
7965bd4d 3116 struct pipe_inode_info *pipe, unsigned int len,
25869262 3117 unsigned int flags);
7965bd4d 3118void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
af2806f8 3119unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
36d5fe6a
ZK
3120int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3121 int len, int hlen);
7965bd4d
JP
3122void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3123int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3124void skb_scrub_packet(struct sk_buff *skb, bool xnet);
de960aa9 3125unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
ae7ef81e 3126bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
7965bd4d 3127struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
0d5501c1 3128struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
e2195121 3129int skb_ensure_writable(struct sk_buff *skb, int write_len);
bfca4c52 3130int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
93515d53
JP
3131int skb_vlan_pop(struct sk_buff *skb);
3132int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
6fa01ccd
SV
3133struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3134 gfp_t gfp);
20380731 3135
6ce8e9ce
AV
3136static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3137{
3073f070 3138 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
6ce8e9ce
AV
3139}
3140
7eab8d9e
AV
3141static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3142{
e5a4b0bb 3143 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
7eab8d9e
AV
3144}
3145
2817a336
DB
3146struct skb_checksum_ops {
3147 __wsum (*update)(const void *mem, int len, __wsum wsum);
3148 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3149};
3150
9617813d
DC
3151extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3152
2817a336
DB
3153__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3154 __wsum csum, const struct skb_checksum_ops *ops);
3155__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3156 __wsum csum);
3157
1e98a0f0
ED
3158static inline void * __must_check
3159__skb_header_pointer(const struct sk_buff *skb, int offset,
3160 int len, void *data, int hlen, void *buffer)
1da177e4 3161{
55820ee2 3162 if (hlen - offset >= len)
690e36e7 3163 return data + offset;
1da177e4 3164
690e36e7
DM
3165 if (!skb ||
3166 skb_copy_bits(skb, offset, buffer, len) < 0)
1da177e4
LT
3167 return NULL;
3168
3169 return buffer;
3170}
3171
1e98a0f0
ED
3172static inline void * __must_check
3173skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
690e36e7
DM
3174{
3175 return __skb_header_pointer(skb, offset, len, skb->data,
3176 skb_headlen(skb), buffer);
3177}
3178
4262e5cc
DB
3179/**
3180 * skb_needs_linearize - check if we need to linearize a given skb
3181 * depending on the given device features.
3182 * @skb: socket buffer to check
3183 * @features: net device features
3184 *
3185 * Returns true if either:
3186 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3187 * 2. skb is fragmented and the device does not support SG.
3188 */
3189static inline bool skb_needs_linearize(struct sk_buff *skb,
3190 netdev_features_t features)
3191{
3192 return skb_is_nonlinear(skb) &&
3193 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3194 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3195}
3196
d626f62b
ACM
3197static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3198 void *to,
3199 const unsigned int len)
3200{
3201 memcpy(to, skb->data, len);
3202}
3203
3204static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3205 const int offset, void *to,
3206 const unsigned int len)
3207{
3208 memcpy(to, skb->data + offset, len);
3209}
3210
27d7ff46
ACM
3211static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3212 const void *from,
3213 const unsigned int len)
3214{
3215 memcpy(skb->data, from, len);
3216}
3217
3218static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3219 const int offset,
3220 const void *from,
3221 const unsigned int len)
3222{
3223 memcpy(skb->data + offset, from, len);
3224}
3225
7965bd4d 3226void skb_init(void);
1da177e4 3227
ac45f602
PO
3228static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3229{
3230 return skb->tstamp;
3231}
3232
a61bbcf2
PM
3233/**
3234 * skb_get_timestamp - get timestamp from a skb
3235 * @skb: skb to get stamp from
3236 * @stamp: pointer to struct timeval to store stamp in
3237 *
3238 * Timestamps are stored in the skb as offsets to a base timestamp.
3239 * This function converts the offset back to a struct timeval and stores
3240 * it in stamp.
3241 */
ac45f602
PO
3242static inline void skb_get_timestamp(const struct sk_buff *skb,
3243 struct timeval *stamp)
a61bbcf2 3244{
b7aa0bf7 3245 *stamp = ktime_to_timeval(skb->tstamp);
a61bbcf2
PM
3246}
3247
ac45f602
PO
3248static inline void skb_get_timestampns(const struct sk_buff *skb,
3249 struct timespec *stamp)
3250{
3251 *stamp = ktime_to_timespec(skb->tstamp);
3252}
3253
b7aa0bf7 3254static inline void __net_timestamp(struct sk_buff *skb)
a61bbcf2 3255{
b7aa0bf7 3256 skb->tstamp = ktime_get_real();
a61bbcf2
PM
3257}
3258
164891aa
SH
3259static inline ktime_t net_timedelta(ktime_t t)
3260{
3261 return ktime_sub(ktime_get_real(), t);
3262}
3263
b9ce204f
IJ
3264static inline ktime_t net_invalid_timestamp(void)
3265{
8b0e1953 3266 return 0;
b9ce204f 3267}
a61bbcf2 3268
62bccb8c
AD
3269struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3270
c1f19b51
RC
3271#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3272
7965bd4d
JP
3273void skb_clone_tx_timestamp(struct sk_buff *skb);
3274bool skb_defer_rx_timestamp(struct sk_buff *skb);
c1f19b51
RC
3275
3276#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3277
3278static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3279{
3280}
3281
3282static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3283{
3284 return false;
3285}
3286
3287#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3288
3289/**
3290 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3291 *
da92b194
RC
3292 * PHY drivers may accept clones of transmitted packets for
3293 * timestamping via their phy_driver.txtstamp method. These drivers
7a76a021
BP
3294 * must call this function to return the skb back to the stack with a
3295 * timestamp.
da92b194 3296 *
c1f19b51 3297 * @skb: clone of the the original outgoing packet
7a76a021 3298 * @hwtstamps: hardware time stamps
c1f19b51
RC
3299 *
3300 */
3301void skb_complete_tx_timestamp(struct sk_buff *skb,
3302 struct skb_shared_hwtstamps *hwtstamps);
3303
e7fd2885
WB
3304void __skb_tstamp_tx(struct sk_buff *orig_skb,
3305 struct skb_shared_hwtstamps *hwtstamps,
3306 struct sock *sk, int tstype);
3307
ac45f602
PO
3308/**
3309 * skb_tstamp_tx - queue clone of skb with send time stamps
3310 * @orig_skb: the original outgoing packet
3311 * @hwtstamps: hardware time stamps, may be NULL if not available
3312 *
3313 * If the skb has a socket associated, then this function clones the
3314 * skb (thus sharing the actual data and optional structures), stores
3315 * the optional hardware time stamping information (if non NULL) or
3316 * generates a software time stamp (otherwise), then queues the clone
3317 * to the error queue of the socket. Errors are silently ignored.
3318 */
7965bd4d
JP
3319void skb_tstamp_tx(struct sk_buff *orig_skb,
3320 struct skb_shared_hwtstamps *hwtstamps);
ac45f602 3321
4507a715
RC
3322/**
3323 * skb_tx_timestamp() - Driver hook for transmit timestamping
3324 *
3325 * Ethernet MAC Drivers should call this function in their hard_xmit()
4ff75b7c 3326 * function immediately before giving the sk_buff to the MAC hardware.
4507a715 3327 *
73409f3b
DM
3328 * Specifically, one should make absolutely sure that this function is
3329 * called before TX completion of this packet can trigger. Otherwise
3330 * the packet could potentially already be freed.
3331 *
4507a715
RC
3332 * @skb: A socket buffer.
3333 */
3334static inline void skb_tx_timestamp(struct sk_buff *skb)
3335{
c1f19b51 3336 skb_clone_tx_timestamp(skb);
b50a5c70
ML
3337 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3338 skb_tstamp_tx(skb, NULL);
4507a715
RC
3339}
3340
6e3e939f
JB
3341/**
3342 * skb_complete_wifi_ack - deliver skb with wifi status
3343 *
3344 * @skb: the original outgoing packet
3345 * @acked: ack status
3346 *
3347 */
3348void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3349
7965bd4d
JP
3350__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3351__sum16 __skb_checksum_complete(struct sk_buff *skb);
fb286bb2 3352
60476372
HX
3353static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3354{
6edec0e6
TH
3355 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3356 skb->csum_valid ||
3357 (skb->ip_summed == CHECKSUM_PARTIAL &&
3358 skb_checksum_start_offset(skb) >= 0));
60476372
HX
3359}
3360
fb286bb2
HX
3361/**
3362 * skb_checksum_complete - Calculate checksum of an entire packet
3363 * @skb: packet to process
3364 *
3365 * This function calculates the checksum over the entire packet plus
3366 * the value of skb->csum. The latter can be used to supply the
3367 * checksum of a pseudo header as used by TCP/UDP. It returns the
3368 * checksum.
3369 *
3370 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3371 * this function can be used to verify that checksum on received
3372 * packets. In that case the function should return zero if the
3373 * checksum is correct. In particular, this function will return zero
3374 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3375 * hardware has already verified the correctness of the checksum.
3376 */
4381ca3c 3377static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
fb286bb2 3378{
60476372
HX
3379 return skb_csum_unnecessary(skb) ?
3380 0 : __skb_checksum_complete(skb);
fb286bb2
HX
3381}
3382
77cffe23
TH
3383static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3384{
3385 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3386 if (skb->csum_level == 0)
3387 skb->ip_summed = CHECKSUM_NONE;
3388 else
3389 skb->csum_level--;
3390 }
3391}
3392
3393static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3394{
3395 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3396 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3397 skb->csum_level++;
3398 } else if (skb->ip_summed == CHECKSUM_NONE) {
3399 skb->ip_summed = CHECKSUM_UNNECESSARY;
3400 skb->csum_level = 0;
3401 }
3402}
3403
76ba0aae
TH
3404/* Check if we need to perform checksum complete validation.
3405 *
3406 * Returns true if checksum complete is needed, false otherwise
3407 * (either checksum is unnecessary or zero checksum is allowed).
3408 */
3409static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3410 bool zero_okay,
3411 __sum16 check)
3412{
5d0c2b95
TH
3413 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3414 skb->csum_valid = 1;
77cffe23 3415 __skb_decr_checksum_unnecessary(skb);
76ba0aae
TH
3416 return false;
3417 }
3418
3419 return true;
3420}
3421
3422/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3423 * in checksum_init.
3424 */
3425#define CHECKSUM_BREAK 76
3426
4e18b9ad
TH
3427/* Unset checksum-complete
3428 *
3429 * Unset checksum complete can be done when packet is being modified
3430 * (uncompressed for instance) and checksum-complete value is
3431 * invalidated.
3432 */
3433static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3434{
3435 if (skb->ip_summed == CHECKSUM_COMPLETE)
3436 skb->ip_summed = CHECKSUM_NONE;
3437}
3438
76ba0aae
TH
3439/* Validate (init) checksum based on checksum complete.
3440 *
3441 * Return values:
3442 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3443 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3444 * checksum is stored in skb->csum for use in __skb_checksum_complete
3445 * non-zero: value of invalid checksum
3446 *
3447 */
3448static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3449 bool complete,
3450 __wsum psum)
3451{
3452 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3453 if (!csum_fold(csum_add(psum, skb->csum))) {
5d0c2b95 3454 skb->csum_valid = 1;
76ba0aae
TH
3455 return 0;
3456 }
3457 }
3458
3459 skb->csum = psum;
3460
5d0c2b95
TH
3461 if (complete || skb->len <= CHECKSUM_BREAK) {
3462 __sum16 csum;
3463
3464 csum = __skb_checksum_complete(skb);
3465 skb->csum_valid = !csum;
3466 return csum;
3467 }
76ba0aae
TH
3468
3469 return 0;
3470}
3471
3472static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3473{
3474 return 0;
3475}
3476
3477/* Perform checksum validate (init). Note that this is a macro since we only
3478 * want to calculate the pseudo header which is an input function if necessary.
3479 * First we try to validate without any computation (checksum unnecessary) and
3480 * then calculate based on checksum complete calling the function to compute
3481 * pseudo header.
3482 *
3483 * Return values:
3484 * 0: checksum is validated or try to in skb_checksum_complete
3485 * non-zero: value of invalid checksum
3486 */
3487#define __skb_checksum_validate(skb, proto, complete, \
3488 zero_okay, check, compute_pseudo) \
3489({ \
3490 __sum16 __ret = 0; \
5d0c2b95 3491 skb->csum_valid = 0; \
76ba0aae
TH
3492 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3493 __ret = __skb_checksum_validate_complete(skb, \
3494 complete, compute_pseudo(skb, proto)); \
3495 __ret; \
3496})
3497
3498#define skb_checksum_init(skb, proto, compute_pseudo) \
3499 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3500
3501#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3502 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3503
3504#define skb_checksum_validate(skb, proto, compute_pseudo) \
3505 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3506
3507#define skb_checksum_validate_zero_check(skb, proto, check, \
3508 compute_pseudo) \
096a4cfa 3509 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
76ba0aae
TH
3510
3511#define skb_checksum_simple_validate(skb) \
3512 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3513
d96535a1
TH
3514static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3515{
219f1d79 3516 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
d96535a1
TH
3517}
3518
3519static inline void __skb_checksum_convert(struct sk_buff *skb,
3520 __sum16 check, __wsum pseudo)
3521{
3522 skb->csum = ~pseudo;
3523 skb->ip_summed = CHECKSUM_COMPLETE;
3524}
3525
3526#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3527do { \
3528 if (__skb_checksum_convert_check(skb)) \
3529 __skb_checksum_convert(skb, check, \
3530 compute_pseudo(skb, proto)); \
3531} while (0)
3532
15e2396d
TH
3533static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3534 u16 start, u16 offset)
3535{
3536 skb->ip_summed = CHECKSUM_PARTIAL;
3537 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3538 skb->csum_offset = offset - start;
3539}
3540
dcdc8994
TH
3541/* Update skbuf and packet to reflect the remote checksum offload operation.
3542 * When called, ptr indicates the starting point for skb->csum when
3543 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3544 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3545 */
3546static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
15e2396d 3547 int start, int offset, bool nopartial)
dcdc8994
TH
3548{
3549 __wsum delta;
3550
15e2396d
TH
3551 if (!nopartial) {
3552 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3553 return;
3554 }
3555
dcdc8994
TH
3556 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3557 __skb_checksum_complete(skb);
3558 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3559 }
3560
3561 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3562
3563 /* Adjust skb->csum since we changed the packet */
3564 skb->csum = csum_add(skb->csum, delta);
3565}
3566
cb9c6836
FW
3567static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3568{
3569#if IS_ENABLED(CONFIG_NF_CONNTRACK)
a9e419dc 3570 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
cb9c6836
FW
3571#else
3572 return NULL;
3573#endif
3574}
3575
5f79e0f9 3576#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
7965bd4d 3577void nf_conntrack_destroy(struct nf_conntrack *nfct);
1da177e4
LT
3578static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3579{
3580 if (nfct && atomic_dec_and_test(&nfct->use))
de6e05c4 3581 nf_conntrack_destroy(nfct);
1da177e4
LT
3582}
3583static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3584{
3585 if (nfct)
3586 atomic_inc(&nfct->use);
3587}
2fc72c7b 3588#endif
34666d46 3589#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
1da177e4
LT
3590static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3591{
53869ceb 3592 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use))
1da177e4
LT
3593 kfree(nf_bridge);
3594}
3595static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3596{
3597 if (nf_bridge)
53869ceb 3598 refcount_inc(&nf_bridge->use);
1da177e4
LT
3599}
3600#endif /* CONFIG_BRIDGE_NETFILTER */
a193a4ab
PM
3601static inline void nf_reset(struct sk_buff *skb)
3602{
5f79e0f9 3603#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
a9e419dc
FW
3604 nf_conntrack_put(skb_nfct(skb));
3605 skb->_nfct = 0;
2fc72c7b 3606#endif
34666d46 3607#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
a193a4ab
PM
3608 nf_bridge_put(skb->nf_bridge);
3609 skb->nf_bridge = NULL;
3610#endif
3611}
3612
124dff01
PM
3613static inline void nf_reset_trace(struct sk_buff *skb)
3614{
478b360a 3615#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
130549fe
G
3616 skb->nf_trace = 0;
3617#endif
a193a4ab
PM
3618}
3619
edda553c 3620/* Note: This doesn't put any conntrack and bridge info in dst. */
b1937227
ED
3621static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3622 bool copy)
edda553c 3623{
5f79e0f9 3624#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
a9e419dc
FW
3625 dst->_nfct = src->_nfct;
3626 nf_conntrack_get(skb_nfct(src));
2fc72c7b 3627#endif
34666d46 3628#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
edda553c
YK
3629 dst->nf_bridge = src->nf_bridge;
3630 nf_bridge_get(src->nf_bridge);
3631#endif
478b360a 3632#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
b1937227
ED
3633 if (copy)
3634 dst->nf_trace = src->nf_trace;
478b360a 3635#endif
edda553c
YK
3636}
3637
e7ac05f3
YK
3638static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3639{
e7ac05f3 3640#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
a9e419dc 3641 nf_conntrack_put(skb_nfct(dst));
2fc72c7b 3642#endif
34666d46 3643#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
e7ac05f3
YK
3644 nf_bridge_put(dst->nf_bridge);
3645#endif
b1937227 3646 __nf_copy(dst, src, true);
e7ac05f3
YK
3647}
3648
984bc16c
JM
3649#ifdef CONFIG_NETWORK_SECMARK
3650static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3651{
3652 to->secmark = from->secmark;
3653}
3654
3655static inline void skb_init_secmark(struct sk_buff *skb)
3656{
3657 skb->secmark = 0;
3658}
3659#else
3660static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3661{ }
3662
3663static inline void skb_init_secmark(struct sk_buff *skb)
3664{ }
3665#endif
3666
574f7194
EB
3667static inline bool skb_irq_freeable(const struct sk_buff *skb)
3668{
3669 return !skb->destructor &&
3670#if IS_ENABLED(CONFIG_XFRM)
3671 !skb->sp &&
3672#endif
cb9c6836 3673 !skb_nfct(skb) &&
574f7194
EB
3674 !skb->_skb_refdst &&
3675 !skb_has_frag_list(skb);
3676}
3677
f25f4e44
PWJ
3678static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3679{
f25f4e44 3680 skb->queue_mapping = queue_mapping;
f25f4e44
PWJ
3681}
3682
9247744e 3683static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4e3ab47a 3684{
4e3ab47a 3685 return skb->queue_mapping;
4e3ab47a
PE
3686}
3687
f25f4e44
PWJ
3688static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3689{
f25f4e44 3690 to->queue_mapping = from->queue_mapping;
f25f4e44
PWJ
3691}
3692
d5a9e24a
DM
3693static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3694{
3695 skb->queue_mapping = rx_queue + 1;
3696}
3697
9247744e 3698static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
d5a9e24a
DM
3699{
3700 return skb->queue_mapping - 1;
3701}
3702
9247744e 3703static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
d5a9e24a 3704{
a02cec21 3705 return skb->queue_mapping != 0;
d5a9e24a
DM
3706}
3707
4ff06203
JA
3708static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3709{
3710 skb->dst_pending_confirm = val;
3711}
3712
3713static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3714{
3715 return skb->dst_pending_confirm != 0;
3716}
3717
def8b4fa
AD
3718static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3719{
0b3d8e08 3720#ifdef CONFIG_XFRM
def8b4fa 3721 return skb->sp;
def8b4fa 3722#else
def8b4fa 3723 return NULL;
def8b4fa 3724#endif
0b3d8e08 3725}
def8b4fa 3726
68c33163
PS
3727/* Keeps track of mac header offset relative to skb->head.
3728 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3729 * For non-tunnel skb it points to skb_mac_header() and for
3347c960
ED
3730 * tunnel skb it points to outer mac header.
3731 * Keeps track of level of encapsulation of network headers.
3732 */
68c33163 3733struct skb_gso_cb {
802ab55a
AD
3734 union {
3735 int mac_offset;
3736 int data_offset;
3737 };
3347c960 3738 int encap_level;
76443456 3739 __wsum csum;
7e2b10c1 3740 __u16 csum_start;
68c33163 3741};
9207f9d4
KK
3742#define SKB_SGO_CB_OFFSET 32
3743#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
68c33163
PS
3744
3745static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3746{
3747 return (skb_mac_header(inner_skb) - inner_skb->head) -
3748 SKB_GSO_CB(inner_skb)->mac_offset;
3749}
3750
1e2bd517
PS
3751static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3752{
3753 int new_headroom, headroom;
3754 int ret;
3755
3756 headroom = skb_headroom(skb);
3757 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3758 if (ret)
3759 return ret;
3760
3761 new_headroom = skb_headroom(skb);
3762 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3763 return 0;
3764}
3765
08b64fcc
AD
3766static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3767{
3768 /* Do not update partial checksums if remote checksum is enabled. */
3769 if (skb->remcsum_offload)
3770 return;
3771
3772 SKB_GSO_CB(skb)->csum = res;
3773 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3774}
3775
7e2b10c1
TH
3776/* Compute the checksum for a gso segment. First compute the checksum value
3777 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3778 * then add in skb->csum (checksum from csum_start to end of packet).
3779 * skb->csum and csum_start are then updated to reflect the checksum of the
3780 * resultant packet starting from the transport header-- the resultant checksum
3781 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3782 * header.
3783 */
3784static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3785{
76443456
AD
3786 unsigned char *csum_start = skb_transport_header(skb);
3787 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3788 __wsum partial = SKB_GSO_CB(skb)->csum;
7e2b10c1 3789
76443456
AD
3790 SKB_GSO_CB(skb)->csum = res;
3791 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
7e2b10c1 3792
76443456 3793 return csum_fold(csum_partial(csum_start, plen, partial));
7e2b10c1
TH
3794}
3795
bdcc0924 3796static inline bool skb_is_gso(const struct sk_buff *skb)
89114afd
HX
3797{
3798 return skb_shinfo(skb)->gso_size;
3799}
3800
36a8f39e 3801/* Note: Should be called only if skb_is_gso(skb) is true */
bdcc0924 3802static inline bool skb_is_gso_v6(const struct sk_buff *skb)
eabd7e35
BG
3803{
3804 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3805}
3806
5293efe6
DB
3807static inline void skb_gso_reset(struct sk_buff *skb)
3808{
3809 skb_shinfo(skb)->gso_size = 0;
3810 skb_shinfo(skb)->gso_segs = 0;
3811 skb_shinfo(skb)->gso_type = 0;
3812}
3813
7965bd4d 3814void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4497b076
BH
3815
3816static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3817{
3818 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3819 * wanted then gso_type will be set. */
05bdd2f1
ED
3820 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3821
b78462eb
AD
3822 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3823 unlikely(shinfo->gso_type == 0)) {
4497b076
BH
3824 __skb_warn_lro_forwarding(skb);
3825 return true;
3826 }
3827 return false;
3828}
3829
35fc92a9
HX
3830static inline void skb_forward_csum(struct sk_buff *skb)
3831{
3832 /* Unfortunately we don't support this one. Any brave souls? */
3833 if (skb->ip_summed == CHECKSUM_COMPLETE)
3834 skb->ip_summed = CHECKSUM_NONE;
3835}
3836
bc8acf2c
ED
3837/**
3838 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3839 * @skb: skb to check
3840 *
3841 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3842 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3843 * use this helper, to document places where we make this assertion.
3844 */
05bdd2f1 3845static inline void skb_checksum_none_assert(const struct sk_buff *skb)
bc8acf2c
ED
3846{
3847#ifdef DEBUG
3848 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3849#endif
3850}
3851
f35d9d8a 3852bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
a6686f2f 3853
ed1f50c3 3854int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
9afd85c9
LL
3855struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3856 unsigned int transport_len,
3857 __sum16(*skb_chkf)(struct sk_buff *skb));
ed1f50c3 3858
3a7c1ee4
AD
3859/**
3860 * skb_head_is_locked - Determine if the skb->head is locked down
3861 * @skb: skb to check
3862 *
3863 * The head on skbs build around a head frag can be removed if they are
3864 * not cloned. This function returns true if the skb head is locked down
3865 * due to either being allocated via kmalloc, or by being a clone with
3866 * multiple references to the head.
3867 */
3868static inline bool skb_head_is_locked(const struct sk_buff *skb)
3869{
3870 return !skb->head_frag || skb_cloned(skb);
3871}
fe6cc55f
FW
3872
3873/**
3874 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3875 *
3876 * @skb: GSO skb
3877 *
3878 * skb_gso_network_seglen is used to determine the real size of the
3879 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3880 *
3881 * The MAC/L2 header is not accounted for.
3882 */
3883static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3884{
3885 unsigned int hdr_len = skb_transport_header(skb) -
3886 skb_network_header(skb);
3887 return hdr_len + skb_gso_transport_seglen(skb);
3888}
ee122c79 3889
179bc67f
EC
3890/* Local Checksum Offload.
3891 * Compute outer checksum based on the assumption that the
3892 * inner checksum will be offloaded later.
e8ae7b00
EC
3893 * See Documentation/networking/checksum-offloads.txt for
3894 * explanation of how this works.
179bc67f
EC
3895 * Fill in outer checksum adjustment (e.g. with sum of outer
3896 * pseudo-header) before calling.
3897 * Also ensure that inner checksum is in linear data area.
3898 */
3899static inline __wsum lco_csum(struct sk_buff *skb)
3900{
9e74a6da
AD
3901 unsigned char *csum_start = skb_checksum_start(skb);
3902 unsigned char *l4_hdr = skb_transport_header(skb);
3903 __wsum partial;
179bc67f
EC
3904
3905 /* Start with complement of inner checksum adjustment */
9e74a6da
AD
3906 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3907 skb->csum_offset));
3908
179bc67f 3909 /* Add in checksum of our headers (incl. outer checksum
9e74a6da 3910 * adjustment filled in by caller) and return result.
179bc67f 3911 */
9e74a6da 3912 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
179bc67f
EC
3913}
3914
1da177e4
LT
3915#endif /* __KERNEL__ */
3916#endif /* _LINUX_SKBUFF_H */