blob: 97dc5f8123b326cc7ed1c102a765ada789592176 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
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
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/kernel.h>
Vegard Nossumfe55f6d2008-08-30 12:16:35 +020018#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/compiler.h>
20#include <linux/time.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050021#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/cache.h>
Eric Dumazet56b17422014-11-03 08:19:53 -080023#include <linux/rbtree.h>
David S. Miller51f3d022014-11-05 16:46:40 -050024#include <linux/socket.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
Arun Sharma600634972011-07-26 16:09:06 -070026#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <asm/types.h>
28#include <linux/spinlock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/net.h>
Thomas Graf3fc7e8a2005-06-23 21:00:17 -070030#include <linux/textsearch.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <net/checksum.h>
Al Viroa80958f2006-12-04 20:41:19 +000032#include <linux/rcupdate.h>
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -070033#include <linux/hrtimer.h>
Ian Campbell131ea662011-08-19 06:25:00 +000034#include <linux/dma-mapping.h>
Michał Mirosławc8f44af2011-11-15 15:29:55 +000035#include <linux/netdev_features.h>
Eric Dumazet363ec39232014-02-26 14:02:11 -080036#include <linux/sched.h>
Jason Wang5203cd22013-03-26 23:11:21 +000037#include <net/flow_keys.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070038
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010039/* A. Checksumming of received packets by device.
40 *
41 * CHECKSUM_NONE:
42 *
43 * Device failed to checksum this packet e.g. due to lack of capabilities.
44 * The packet contains full (though not verified) checksum in packet but
45 * not in skb->csum. Thus, skb->csum is undefined in this case.
46 *
47 * CHECKSUM_UNNECESSARY:
48 *
49 * The hardware you're dealing with doesn't calculate the full checksum
50 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
Tom Herbert77cffe22014-08-27 21:26:46 -070051 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
52 * if their checksums are okay. skb->csum is still undefined in this case
53 * though. It is a bad option, but, unfortunately, nowadays most vendors do
54 * this. Apparently with the secret goal to sell you new devices, when you
55 * will add new protocol to your host, f.e. IPv6 8)
56 *
57 * CHECKSUM_UNNECESSARY is applicable to following protocols:
58 * TCP: IPv6 and IPv4.
59 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
60 * zero UDP checksum for either IPv4 or IPv6, the networking stack
61 * may perform further validation in this case.
62 * GRE: only if the checksum is present in the header.
63 * SCTP: indicates the CRC in SCTP header has been validated.
64 *
65 * skb->csum_level indicates the number of consecutive checksums found in
66 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
67 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
68 * and a device is able to verify the checksums for UDP (possibly zero),
69 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
70 * two. If the device were only able to verify the UDP checksum and not
71 * GRE, either because it doesn't support GRE checksum of because GRE
72 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
73 * not considered in this case).
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010074 *
75 * CHECKSUM_COMPLETE:
76 *
77 * This is the most generic way. The device supplied checksum of the _whole_
78 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
79 * hardware doesn't need to parse L3/L4 headers to implement this.
80 *
81 * Note: Even if device supports only some protocols, but is able to produce
82 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
83 *
84 * CHECKSUM_PARTIAL:
85 *
86 * This is identical to the case for output below. This may occur on a packet
87 * received directly from another Linux OS, e.g., a virtualized Linux kernel
88 * on the same host. The packet can be treated in the same way as
89 * CHECKSUM_UNNECESSARY, except that on output (i.e., forwarding) the
90 * checksum must be filled in by the OS or the hardware.
91 *
92 * B. Checksumming on output.
93 *
94 * CHECKSUM_NONE:
95 *
96 * The skb was already checksummed by the protocol, or a checksum is not
97 * required.
98 *
99 * CHECKSUM_PARTIAL:
100 *
101 * The device is required to checksum the packet as seen by hard_start_xmit()
102 * from skb->csum_start up to the end, and to record/write the checksum at
103 * offset skb->csum_start + skb->csum_offset.
104 *
105 * The device must show its capabilities in dev->features, set up at device
106 * setup time, e.g. netdev_features.h:
107 *
108 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
109 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
110 * IPv4. Sigh. Vendors like this way for an unknown reason.
111 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
112 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
113 * NETIF_F_... - Well, you get the picture.
114 *
115 * CHECKSUM_UNNECESSARY:
116 *
117 * Normally, the device will do per protocol specific checksumming. Protocol
118 * implementations that do not want the NIC to perform the checksum
119 * calculation should use this flag in their outgoing skbs.
120 *
121 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
122 * offload. Correspondingly, the FCoE protocol driver
123 * stack should use CHECKSUM_UNNECESSARY.
124 *
125 * Any questions? No questions, good. --ANK
126 */
127
Herbert Xu60476372007-04-09 11:59:39 -0700128/* Don't change this without changing skb_csum_unnecessary! */
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100129#define CHECKSUM_NONE 0
130#define CHECKSUM_UNNECESSARY 1
131#define CHECKSUM_COMPLETE 2
132#define CHECKSUM_PARTIAL 3
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133
Tom Herbert77cffe22014-08-27 21:26:46 -0700134/* Maximum value in skb->csum_level */
135#define SKB_MAX_CSUM_LEVEL 3
136
Tobias Klauser0bec8c82014-07-22 12:06:23 +0200137#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
David S. Millerfc910a22007-03-25 20:27:59 -0700138#define SKB_WITH_OVERHEAD(X) \
Herbert Xudeea84b2007-10-21 16:27:46 -0700139 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
David S. Millerfc910a22007-03-25 20:27:59 -0700140#define SKB_MAX_ORDER(X, ORDER) \
141 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700142#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
143#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
144
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000145/* return minimum truesize of one skb containing X bytes of data */
146#define SKB_TRUESIZE(X) ((X) + \
147 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
148 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
149
Linus Torvalds1da177e2005-04-16 15:20:36 -0700150struct net_device;
David Howells716ea3a2007-04-02 20:19:53 -0700151struct scatterlist;
Jens Axboe9c55e012007-11-06 23:30:13 -0800152struct pipe_inode_info;
Herbert Xua8f820aa2014-11-07 21:22:22 +0800153struct iov_iter;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700154
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700155#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700156struct nf_conntrack {
157 atomic_t use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700158};
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700159#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700160
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200161#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162struct nf_bridge_info {
Eric Dumazetbf1ac5c2012-04-18 23:19:25 +0000163 atomic_t use;
164 unsigned int mask;
165 struct net_device *physindev;
166 struct net_device *physoutdev;
167 unsigned long data[32 / sizeof(unsigned long)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168};
169#endif
170
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171struct sk_buff_head {
172 /* These two members must be first. */
173 struct sk_buff *next;
174 struct sk_buff *prev;
175
176 __u32 qlen;
177 spinlock_t lock;
178};
179
180struct sk_buff;
181
Ian Campbell9d4dde52011-12-22 23:39:14 +0000182/* To allow 64K frame to be packed as single skb without frag_list we
183 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
184 * buffers which do not start on a page boundary.
185 *
186 * Since GRO uses frags we allocate at least 16 regardless of page
187 * size.
Anton Blancharda715dea2011-03-27 14:57:26 +0000188 */
Ian Campbell9d4dde52011-12-22 23:39:14 +0000189#if (65536/PAGE_SIZE + 1) < 16
David S. Millereec00952011-03-29 23:34:08 -0700190#define MAX_SKB_FRAGS 16UL
Anton Blancharda715dea2011-03-27 14:57:26 +0000191#else
Ian Campbell9d4dde52011-12-22 23:39:14 +0000192#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
Anton Blancharda715dea2011-03-27 14:57:26 +0000193#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700194
195typedef struct skb_frag_struct skb_frag_t;
196
197struct skb_frag_struct {
Ian Campbella8605c62011-10-19 23:01:49 +0000198 struct {
199 struct page *p;
200 } page;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000201#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
David S. Millera309bb02007-07-30 18:47:03 -0700202 __u32 page_offset;
203 __u32 size;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000204#else
205 __u16 page_offset;
206 __u16 size;
207#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700208};
209
Eric Dumazet9e903e02011-10-18 21:00:24 +0000210static inline unsigned int skb_frag_size(const skb_frag_t *frag)
211{
212 return frag->size;
213}
214
215static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
216{
217 frag->size = size;
218}
219
220static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
221{
222 frag->size += delta;
223}
224
225static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
226{
227 frag->size -= delta;
228}
229
Patrick Ohlyac45f602009-02-12 05:03:37 +0000230#define HAVE_HW_TIME_STAMP
231
232/**
Randy Dunlapd3a21be2009-03-02 03:15:58 -0800233 * struct skb_shared_hwtstamps - hardware time stamps
Patrick Ohlyac45f602009-02-12 05:03:37 +0000234 * @hwtstamp: hardware time stamp transformed into duration
235 * since arbitrary point in time
Patrick Ohlyac45f602009-02-12 05:03:37 +0000236 *
237 * Software time stamps generated by ktime_get_real() are stored in
Willem de Bruijn4d276eb2014-07-25 18:01:32 -0400238 * skb->tstamp.
Patrick Ohlyac45f602009-02-12 05:03:37 +0000239 *
240 * hwtstamps can only be compared against other hwtstamps from
241 * the same device.
242 *
243 * This structure is attached to packets as part of the
244 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
245 */
246struct skb_shared_hwtstamps {
247 ktime_t hwtstamp;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000248};
249
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000250/* Definitions for tx_flags in struct skb_shared_info */
251enum {
252 /* generate hardware time stamp */
253 SKBTX_HW_TSTAMP = 1 << 0,
254
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400255 /* generate software time stamp when queueing packet to NIC */
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000256 SKBTX_SW_TSTAMP = 1 << 1,
257
258 /* device driver is going to provide hardware time stamp */
259 SKBTX_IN_PROGRESS = 1 << 2,
260
Shirley Maa6686f22011-07-06 12:22:12 +0000261 /* device driver supports TX zero-copy buffers */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000262 SKBTX_DEV_ZEROCOPY = 1 << 3,
Johannes Berg6e3e9392011-11-09 10:15:42 +0100263
264 /* generate wifi status information (where possible) */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000265 SKBTX_WIFI_STATUS = 1 << 4,
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000266
267 /* This indicates at least one fragment might be overwritten
268 * (as in vmsplice(), sendfile() ...)
269 * If we need to compute a TX checksum, we'll need to copy
270 * all frags to avoid possible bad checksum
271 */
272 SKBTX_SHARED_FRAG = 1 << 5,
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400273
274 /* generate software time stamp when entering packet scheduling */
275 SKBTX_SCHED_TSTAMP = 1 << 6,
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400276
277 /* generate software timestamp on peer data acknowledgment */
278 SKBTX_ACK_TSTAMP = 1 << 7,
Shirley Maa6686f22011-07-06 12:22:12 +0000279};
280
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400281#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
282 SKBTX_SCHED_TSTAMP | \
283 SKBTX_ACK_TSTAMP)
Willem de Bruijnf24b9be2014-08-04 22:11:45 -0400284#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
285
Shirley Maa6686f22011-07-06 12:22:12 +0000286/*
287 * The callback notifies userspace to release buffers when skb DMA is done in
288 * lower device, the skb last reference should be 0 when calling this.
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000289 * The zerocopy_success argument is true if zero copy transmit occurred,
290 * false on data copy or out of memory error caused by data copy attempt.
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000291 * The ctx field is used to track device context.
292 * The desc field is used to track userspace buffer index.
Shirley Maa6686f22011-07-06 12:22:12 +0000293 */
294struct ubuf_info {
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000295 void (*callback)(struct ubuf_info *, bool zerocopy_success);
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000296 void *ctx;
Shirley Maa6686f22011-07-06 12:22:12 +0000297 unsigned long desc;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000298};
299
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300/* This data is invariant across clones and lives at
301 * the end of the header data, ie. at skb->end.
302 */
303struct skb_shared_info {
Ian Campbell9f42f122012-01-05 07:13:39 +0000304 unsigned char nr_frags;
305 __u8 tx_flags;
Herbert Xu79671682006-06-22 02:40:14 -0700306 unsigned short gso_size;
307 /* Warning: this field is not always filled in (UFO)! */
308 unsigned short gso_segs;
309 unsigned short gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700310 struct sk_buff *frag_list;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000311 struct skb_shared_hwtstamps hwtstamps;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400312 u32 tskey;
Ian Campbell9f42f122012-01-05 07:13:39 +0000313 __be32 ip6_frag_id;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700314
315 /*
316 * Warning : all fields before dataref are cleared in __alloc_skb()
317 */
318 atomic_t dataref;
319
Johann Baudy69e3c752009-05-18 22:11:22 -0700320 /* Intermediate layers must ensure that destructor_arg
321 * remains valid until skb destructor */
322 void * destructor_arg;
Shirley Maa6686f22011-07-06 12:22:12 +0000323
Eric Dumazetfed66382010-07-22 19:09:08 +0000324 /* must be last field, see pskb_expand_head() */
325 skb_frag_t frags[MAX_SKB_FRAGS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700326};
327
328/* We divide dataref into two halves. The higher 16 bits hold references
329 * to the payload part of skb->data. The lower 16 bits hold references to
Patrick McHardy334a8132007-06-25 04:35:20 -0700330 * the entire skb->data. A clone of a headerless skb holds the length of
331 * the header in skb->hdr_len.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700332 *
333 * All users must obey the rule that the skb->data reference count must be
334 * greater than or equal to the payload reference count.
335 *
336 * Holding a reference to the payload part means that the user does not
337 * care about modifications to the header part of skb->data.
338 */
339#define SKB_DATAREF_SHIFT 16
340#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
341
David S. Millerd179cd12005-08-17 14:57:30 -0700342
343enum {
Vijay Subramanianc8753d52014-10-02 10:00:43 -0700344 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
345 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
346 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
347 SKB_FCLONE_FREE, /* this companion fclone skb is available */
David S. Millerd179cd12005-08-17 14:57:30 -0700348};
349
Herbert Xu79671682006-06-22 02:40:14 -0700350enum {
351 SKB_GSO_TCPV4 = 1 << 0,
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700352 SKB_GSO_UDP = 1 << 1,
Herbert Xu576a30e2006-06-27 13:22:38 -0700353
354 /* This indicates the skb is from an untrusted source. */
355 SKB_GSO_DODGY = 1 << 2,
Michael Chanb0da85372006-06-29 12:30:00 -0700356
357 /* This indicates the tcp segment has CWR set. */
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700358 SKB_GSO_TCP_ECN = 1 << 3,
359
360 SKB_GSO_TCPV6 = 1 << 4,
Chris Leech01d5b2f2009-02-27 14:06:49 -0800361
362 SKB_GSO_FCOE = 1 << 5,
Pravin B Shelar68c33162013-02-14 14:02:41 +0000363
364 SKB_GSO_GRE = 1 << 6,
Pravin B Shelar73136262013-03-07 13:21:51 +0000365
Tom Herbert4b282522014-06-14 23:23:52 -0700366 SKB_GSO_GRE_CSUM = 1 << 7,
Simon Horman0d89d202013-05-23 21:02:52 +0000367
Tom Herbert4b282522014-06-14 23:23:52 -0700368 SKB_GSO_IPIP = 1 << 8,
Eric Dumazetcb32f512013-10-19 11:42:57 -0700369
Tom Herbert4b282522014-06-14 23:23:52 -0700370 SKB_GSO_SIT = 1 << 9,
Eric Dumazet61c1db72013-10-20 20:47:30 -0700371
Tom Herbert4b282522014-06-14 23:23:52 -0700372 SKB_GSO_UDP_TUNNEL = 1 << 10,
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700373
374 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
Tom Herbert4749c092014-06-04 17:20:23 -0700375
Pravin B Shelar59b93b42014-11-05 15:27:48 -0800376 SKB_GSO_TUNNEL_REMCSUM = 1 << 12,
Herbert Xu79671682006-06-22 02:40:14 -0700377};
378
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700379#if BITS_PER_LONG > 32
380#define NET_SKBUFF_DATA_USES_OFFSET 1
381#endif
382
383#ifdef NET_SKBUFF_DATA_USES_OFFSET
384typedef unsigned int sk_buff_data_t;
385#else
386typedef unsigned char *sk_buff_data_t;
387#endif
388
Eric Dumazet363ec39232014-02-26 14:02:11 -0800389/**
390 * struct skb_mstamp - multi resolution time stamps
391 * @stamp_us: timestamp in us resolution
392 * @stamp_jiffies: timestamp in jiffies
393 */
394struct skb_mstamp {
395 union {
396 u64 v64;
397 struct {
398 u32 stamp_us;
399 u32 stamp_jiffies;
400 };
401 };
402};
403
404/**
405 * skb_mstamp_get - get current timestamp
406 * @cl: place to store timestamps
407 */
408static inline void skb_mstamp_get(struct skb_mstamp *cl)
409{
410 u64 val = local_clock();
411
412 do_div(val, NSEC_PER_USEC);
413 cl->stamp_us = (u32)val;
414 cl->stamp_jiffies = (u32)jiffies;
415}
416
417/**
418 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
419 * @t1: pointer to newest sample
420 * @t0: pointer to oldest sample
421 */
422static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
423 const struct skb_mstamp *t0)
424{
425 s32 delta_us = t1->stamp_us - t0->stamp_us;
426 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
427
428 /* If delta_us is negative, this might be because interval is too big,
429 * or local_clock() drift is too big : fallback using jiffies.
430 */
431 if (delta_us <= 0 ||
432 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
433
434 delta_us = jiffies_to_usecs(delta_jiffies);
435
436 return delta_us;
437}
438
439
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440/**
441 * struct sk_buff - socket buffer
442 * @next: Next buffer in list
443 * @prev: Previous buffer in list
Eric Dumazet363ec39232014-02-26 14:02:11 -0800444 * @tstamp: Time we arrived/left
Eric Dumazet56b17422014-11-03 08:19:53 -0800445 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700446 * @sk: Socket we are owned by
Linus Torvalds1da177e2005-04-16 15:20:36 -0700447 * @dev: Device we arrived on/are leaving by
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700448 * @cb: Control buffer. Free for use by every layer. Put private vars here
Eric Dumazet7fee2262010-05-11 23:19:48 +0000449 * @_skb_refdst: destination entry (with norefcount bit)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700450 * @sp: the security path, used for xfrm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700451 * @len: Length of actual data
452 * @data_len: Data length
453 * @mac_len: Length of link layer header
Patrick McHardy334a8132007-06-25 04:35:20 -0700454 * @hdr_len: writable header length of cloned skb
Herbert Xu663ead32007-04-09 11:59:07 -0700455 * @csum: Checksum (must include start/offset pair)
456 * @csum_start: Offset from skb->head where checksumming should start
457 * @csum_offset: Offset from csum_start where checksum should be stored
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700458 * @priority: Packet queueing priority
WANG Cong60ff7462014-05-04 16:39:18 -0700459 * @ignore_df: allow local fragmentation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700460 * @cloned: Head may be cloned (check refcnt to be sure)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700461 * @ip_summed: Driver fed us an IP checksum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700462 * @nohdr: Payload reference only, must not modify header
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700463 * @nfctinfo: Relationship of this skb to the connection
Linus Torvalds1da177e2005-04-16 15:20:36 -0700464 * @pkt_type: Packet class
Randy Dunlapc83c2482005-10-18 22:07:41 -0700465 * @fclone: skbuff clone status
Randy Dunlapc83c2482005-10-18 22:07:41 -0700466 * @ipvs_property: skbuff is owned by ipvs
Randy Dunlap31729362008-02-18 20:52:13 -0800467 * @peeked: this packet has been seen already, so stats have been
468 * done for it, don't do them again
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700469 * @nf_trace: netfilter packet trace flag
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700470 * @protocol: Packet protocol from driver
471 * @destructor: Destruct function
472 * @nfct: Associated connection, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700473 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
Eric Dumazet8964be42009-11-20 15:35:04 -0800474 * @skb_iif: ifindex of device we arrived on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 * @tc_index: Traffic control index
476 * @tc_verd: traffic control verdict
Tom Herbert61b905d2014-03-24 15:34:47 -0700477 * @hash: the packet hash
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700478 * @queue_mapping: Queue mapping for multiqueue devices
David S. Miller0b725a22014-08-25 15:51:53 -0700479 * @xmit_more: More SKBs are pending for this queue
Randy Dunlap553a5672008-04-20 10:51:01 -0700480 * @ndisc_nodetype: router type (from link layer)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700481 * @ooo_okay: allow the mapping of a socket to a queue to be changed
Tom Herbert61b905d2014-03-24 15:34:47 -0700482 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
Changli Gao4ca24622011-08-19 07:26:44 -0700483 * ports.
Tom Herberta3b18dd2014-07-01 21:33:17 -0700484 * @sw_hash: indicates hash was computed in software stack
Johannes Berg6e3e9392011-11-09 10:15:42 +0100485 * @wifi_acked_valid: wifi_acked was set
486 * @wifi_acked: whether frame was acked on wifi or not
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000487 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
Eliezer Tamir06021292013-06-10 11:39:50 +0300488 * @napi_id: id of the NAPI struct this skb came from
James Morris984bc162006-06-09 00:29:17 -0700489 * @secmark: security marking
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700490 * @mark: Generic packet mark
491 * @dropcount: total number of sk_receive_queue overflows
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000492 * @vlan_proto: vlan encapsulation protocol
Patrick McHardy6aa895b02008-07-14 22:49:06 -0700493 * @vlan_tci: vlan tag control information
Simon Horman0d89d202013-05-23 21:02:52 +0000494 * @inner_protocol: Protocol (encapsulation)
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000495 * @inner_transport_header: Inner transport layer header (encapsulation)
496 * @inner_network_header: Network layer header (encapsulation)
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000497 * @inner_mac_header: Link layer header (encapsulation)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700498 * @transport_header: Transport layer header
499 * @network_header: Network layer header
500 * @mac_header: Link layer header
501 * @tail: Tail pointer
502 * @end: End pointer
503 * @head: Head of buffer
504 * @data: Data head pointer
505 * @truesize: Buffer size
506 * @users: User count - see {datagram,tcp}.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700507 */
508
509struct sk_buff {
Eric Dumazet363ec39232014-02-26 14:02:11 -0800510 union {
Eric Dumazet56b17422014-11-03 08:19:53 -0800511 struct {
512 /* These two members must be first. */
513 struct sk_buff *next;
514 struct sk_buff *prev;
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000515
Eric Dumazet56b17422014-11-03 08:19:53 -0800516 union {
517 ktime_t tstamp;
518 struct skb_mstamp skb_mstamp;
519 };
520 };
521 struct rb_node rbnode; /* used in netem & tcp stack */
522 };
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000523 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700524 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700525
Linus Torvalds1da177e2005-04-16 15:20:36 -0700526 /*
527 * This is the control buffer. It is free to use for every
528 * layer. Please put your private variables there. If you
529 * want to keep them across layers you have to do a skb_clone()
530 * first. This is owned by whoever has the skb queued ATM.
531 */
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000532 char cb[48] __aligned(8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700533
Eric Dumazet7fee2262010-05-11 23:19:48 +0000534 unsigned long _skb_refdst;
Eric Dumazetb1937222014-09-28 22:18:47 -0700535 void (*destructor)(struct sk_buff *skb);
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000536#ifdef CONFIG_XFRM
537 struct sec_path *sp;
538#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700539#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
540 struct nf_conntrack *nfct;
541#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200542#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Eric Dumazetb1937222014-09-28 22:18:47 -0700543 struct nf_bridge_info *nf_bridge;
544#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545 unsigned int len,
Patrick McHardy334a8132007-06-25 04:35:20 -0700546 data_len;
547 __u16 mac_len,
548 hdr_len;
Eric Dumazetb1937222014-09-28 22:18:47 -0700549
550 /* Following fields are _not_ copied in __copy_skb_header()
551 * Note that queue_mapping is here mostly to fill a hole.
552 */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200553 kmemcheck_bitfield_begin(flags1);
Eric Dumazetb1937222014-09-28 22:18:47 -0700554 __u16 queue_mapping;
555 __u8 cloned:1,
Harald Welte6869c4d2005-08-09 19:24:19 -0700556 nohdr:1,
Patrick McHardyb84f4cc2005-11-20 21:19:21 -0800557 fclone:2,
Herbert Xua59322b2007-12-05 01:53:40 -0800558 peeked:1,
Eric Dumazetb1937222014-09-28 22:18:47 -0700559 head_frag:1,
560 xmit_more:1;
561 /* one bit hole */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200562 kmemcheck_bitfield_end(flags1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563
Eric Dumazetb1937222014-09-28 22:18:47 -0700564 /* fields enclosed in headers_start/headers_end are copied
565 * using a single memcpy() in __copy_skb_header()
566 */
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700567 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700568 __u32 headers_start[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700569 /* public: */
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200570
571/* if you move pkt_type around you also must adapt those constants */
572#ifdef __BIG_ENDIAN_BITFIELD
573#define PKT_TYPE_MAX (7 << 5)
574#else
575#define PKT_TYPE_MAX 7
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100576#endif
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200577#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
578
579 __u8 __pkt_type_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700580 __u8 pkt_type:3;
581 __u8 pfmemalloc:1;
582 __u8 ignore_df:1;
583 __u8 nfctinfo:3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584
Eric Dumazetb1937222014-09-28 22:18:47 -0700585 __u8 nf_trace:1;
586 __u8 ip_summed:2;
587 __u8 ooo_okay:1;
588 __u8 l4_hash:1;
589 __u8 sw_hash:1;
590 __u8 wifi_acked_valid:1;
591 __u8 wifi_acked:1;
592
593 __u8 no_fcs:1;
594 /* Indicates the inner headers are valid in the skbuff. */
595 __u8 encapsulation:1;
596 __u8 encap_hdr_csum:1;
597 __u8 csum_valid:1;
598 __u8 csum_complete_sw:1;
599 __u8 csum_level:2;
600 __u8 csum_bad:1;
601
602#ifdef CONFIG_IPV6_NDISC_NODETYPE
603 __u8 ndisc_nodetype:2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700605 __u8 ipvs_property:1;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700606 __u8 inner_protocol_type:1;
Tom Herberte585f232014-11-04 09:06:54 -0800607 __u8 remcsum_offload:1;
608 /* 3 or 5 bit hole */
Alexander Duyck4031ae62012-01-27 06:22:53 +0000609
Linus Torvalds1da177e2005-04-16 15:20:36 -0700610#ifdef CONFIG_NET_SCHED
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700611 __u16 tc_index; /* traffic control index */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612#ifdef CONFIG_NET_CLS_ACT
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700613 __u16 tc_verd; /* traffic control verdict */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700614#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700615#endif
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200616
Eric Dumazetb1937222014-09-28 22:18:47 -0700617 union {
618 __wsum csum;
619 struct {
620 __u16 csum_start;
621 __u16 csum_offset;
622 };
623 };
624 __u32 priority;
625 int skb_iif;
626 __u32 hash;
627 __be16 vlan_proto;
628 __u16 vlan_tci;
Dan Williams7bced392013-12-30 12:37:29 -0800629#ifdef CONFIG_NET_RX_BUSY_POLL
630 unsigned int napi_id;
Chris Leech97fc2f02006-05-23 17:55:33 -0700631#endif
James Morris984bc162006-06-09 00:29:17 -0700632#ifdef CONFIG_NETWORK_SECMARK
633 __u32 secmark;
634#endif
Neil Horman3b885782009-10-12 13:26:31 -0700635 union {
636 __u32 mark;
637 __u32 dropcount;
Eric Dumazet16fad692013-03-14 05:40:32 +0000638 __u32 reserved_tailroom;
Neil Horman3b885782009-10-12 13:26:31 -0700639 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640
Tom Herbert8bce6d72014-09-29 20:22:29 -0700641 union {
642 __be16 inner_protocol;
643 __u8 inner_ipproto;
644 };
645
Simon Horman1a37e412013-05-23 21:02:51 +0000646 __u16 inner_transport_header;
647 __u16 inner_network_header;
648 __u16 inner_mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700649
650 __be16 protocol;
Simon Horman1a37e412013-05-23 21:02:51 +0000651 __u16 transport_header;
652 __u16 network_header;
653 __u16 mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700654
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700655 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700656 __u32 headers_end[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700657 /* public: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700658
Linus Torvalds1da177e2005-04-16 15:20:36 -0700659 /* These elements must be at the end, see alloc_skb() for details. */
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700660 sk_buff_data_t tail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700661 sk_buff_data_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 unsigned char *head,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700663 *data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700664 unsigned int truesize;
665 atomic_t users;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666};
667
668#ifdef __KERNEL__
669/*
670 * Handling routines are only of interest to the kernel
671 */
672#include <linux/slab.h>
673
Linus Torvalds1da177e2005-04-16 15:20:36 -0700674
Mel Gormanc93bdd02012-07-31 16:44:19 -0700675#define SKB_ALLOC_FCLONE 0x01
676#define SKB_ALLOC_RX 0x02
677
678/* Returns true if the skb was allocated from PFMEMALLOC reserves */
679static inline bool skb_pfmemalloc(const struct sk_buff *skb)
680{
681 return unlikely(skb->pfmemalloc);
682}
683
Eric Dumazet7fee2262010-05-11 23:19:48 +0000684/*
685 * skb might have a dst pointer attached, refcounted or not.
686 * _skb_refdst low order bit is set if refcount was _not_ taken
687 */
688#define SKB_DST_NOREF 1UL
689#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
690
691/**
692 * skb_dst - returns skb dst_entry
693 * @skb: buffer
694 *
695 * Returns skb dst_entry, regardless of reference taken or not.
696 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000697static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
698{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000699 /* If refdst was not refcounted, check we still are in a
700 * rcu_read_lock section
701 */
702 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
703 !rcu_read_lock_held() &&
704 !rcu_read_lock_bh_held());
705 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
Eric Dumazetadf30902009-06-02 05:19:30 +0000706}
707
Eric Dumazet7fee2262010-05-11 23:19:48 +0000708/**
709 * skb_dst_set - sets skb dst
710 * @skb: buffer
711 * @dst: dst entry
712 *
713 * Sets skb dst, assuming a reference was taken on dst and should
714 * be released by skb_dst_drop()
715 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000716static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
717{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000718 skb->_skb_refdst = (unsigned long)dst;
719}
720
Joe Perches7965bd42013-09-26 14:48:15 -0700721void __skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst,
722 bool force);
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200723
724/**
725 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
726 * @skb: buffer
727 * @dst: dst entry
728 *
729 * Sets skb dst, assuming a reference was not taken on dst.
730 * If dst entry is cached, we do not take reference and dst_release
731 * will be avoided by refdst_drop. If dst entry is not cached, we take
732 * reference, so that last dst_release can destroy the dst immediately.
733 */
734static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
735{
736 __skb_dst_set_noref(skb, dst, false);
737}
738
739/**
740 * skb_dst_set_noref_force - sets skb dst, without taking reference
741 * @skb: buffer
742 * @dst: dst entry
743 *
744 * Sets skb dst, assuming a reference was not taken on dst.
745 * No reference is taken and no dst_release will be called. While for
746 * cached dsts deferred reclaim is a basic feature, for entries that are
747 * not cached it is caller's job to guarantee that last dst_release for
748 * provided dst happens when nobody uses it, eg. after a RCU grace period.
749 */
750static inline void skb_dst_set_noref_force(struct sk_buff *skb,
751 struct dst_entry *dst)
752{
753 __skb_dst_set_noref(skb, dst, true);
754}
Eric Dumazet7fee2262010-05-11 23:19:48 +0000755
756/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300757 * skb_dst_is_noref - Test if skb dst isn't refcounted
Eric Dumazet7fee2262010-05-11 23:19:48 +0000758 * @skb: buffer
759 */
760static inline bool skb_dst_is_noref(const struct sk_buff *skb)
761{
762 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
Eric Dumazetadf30902009-06-02 05:19:30 +0000763}
764
Eric Dumazet511c3f92009-06-02 05:14:27 +0000765static inline struct rtable *skb_rtable(const struct sk_buff *skb)
766{
Eric Dumazetadf30902009-06-02 05:19:30 +0000767 return (struct rtable *)skb_dst(skb);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000768}
769
Joe Perches7965bd42013-09-26 14:48:15 -0700770void kfree_skb(struct sk_buff *skb);
771void kfree_skb_list(struct sk_buff *segs);
772void skb_tx_error(struct sk_buff *skb);
773void consume_skb(struct sk_buff *skb);
774void __kfree_skb(struct sk_buff *skb);
Eric Dumazetd7e88832012-04-30 08:10:34 +0000775extern struct kmem_cache *skbuff_head_cache;
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000776
Joe Perches7965bd42013-09-26 14:48:15 -0700777void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
778bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
779 bool *fragstolen, int *delta_truesize);
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000780
Joe Perches7965bd42013-09-26 14:48:15 -0700781struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
782 int node);
783struct sk_buff *build_skb(void *data, unsigned int frag_size);
David S. Millerd179cd12005-08-17 14:57:30 -0700784static inline struct sk_buff *alloc_skb(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100785 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700786{
Eric Dumazet564824b2010-10-11 19:05:25 +0000787 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700788}
789
Eric Dumazet2e4e4412014-09-17 04:49:49 -0700790struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
791 unsigned long data_len,
792 int max_page_order,
793 int *errcode,
794 gfp_t gfp_mask);
795
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700796/* Layout of fast clones : [skb1][skb2][fclone_ref] */
797struct sk_buff_fclones {
798 struct sk_buff skb1;
799
800 struct sk_buff skb2;
801
802 atomic_t fclone_ref;
803};
804
805/**
806 * skb_fclone_busy - check if fclone is busy
807 * @skb: buffer
808 *
809 * Returns true is skb is a fast clone, and its clone is not freed.
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700810 * Some drivers call skb_orphan() in their ndo_start_xmit(),
811 * so we also check that this didnt happen.
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700812 */
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700813static inline bool skb_fclone_busy(const struct sock *sk,
814 const struct sk_buff *skb)
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700815{
816 const struct sk_buff_fclones *fclones;
817
818 fclones = container_of(skb, struct sk_buff_fclones, skb1);
819
820 return skb->fclone == SKB_FCLONE_ORIG &&
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700821 fclones->skb2.fclone == SKB_FCLONE_CLONE &&
822 fclones->skb2.sk == sk;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700823}
824
David S. Millerd179cd12005-08-17 14:57:30 -0700825static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100826 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700827{
Mel Gormanc93bdd02012-07-31 16:44:19 -0700828 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700829}
830
Joe Perches7965bd42013-09-26 14:48:15 -0700831struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000832static inline struct sk_buff *alloc_skb_head(gfp_t priority)
833{
834 return __alloc_skb_head(priority, -1);
835}
836
Joe Perches7965bd42013-09-26 14:48:15 -0700837struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
838int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
839struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
840struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
Octavian Purdilabad93e92014-06-12 01:36:26 +0300841struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
842 gfp_t gfp_mask, bool fclone);
843static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
844 gfp_t gfp_mask)
845{
846 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
847}
Eric Dumazet117632e2011-12-03 21:39:53 +0000848
Joe Perches7965bd42013-09-26 14:48:15 -0700849int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
850struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
851 unsigned int headroom);
852struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
853 int newtailroom, gfp_t priority);
Fan Du25a91d82014-01-18 09:54:23 +0800854int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
855 int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -0700856int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
857 int len);
858int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
859int skb_pad(struct sk_buff *skb, int pad);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000860#define dev_kfree_skb(a) consume_skb(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700861
Joe Perches7965bd42013-09-26 14:48:15 -0700862int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
863 int getfrag(void *from, char *to, int offset,
864 int len, int odd, struct sk_buff *skb),
865 void *from, int length);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700866
Eric Dumazetd94d9fe2009-11-04 09:50:58 -0800867struct skb_seq_state {
Thomas Graf677e90e2005-06-23 20:59:51 -0700868 __u32 lower_offset;
869 __u32 upper_offset;
870 __u32 frag_idx;
871 __u32 stepped_offset;
872 struct sk_buff *root_skb;
873 struct sk_buff *cur_skb;
874 __u8 *frag_data;
875};
876
Joe Perches7965bd42013-09-26 14:48:15 -0700877void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
878 unsigned int to, struct skb_seq_state *st);
879unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
880 struct skb_seq_state *st);
881void skb_abort_seq_read(struct skb_seq_state *st);
Thomas Graf677e90e2005-06-23 20:59:51 -0700882
Joe Perches7965bd42013-09-26 14:48:15 -0700883unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
884 unsigned int to, struct ts_config *config,
885 struct ts_state *state);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -0700886
Tom Herbert09323cc2013-12-15 22:16:19 -0800887/*
888 * Packet hash types specify the type of hash in skb_set_hash.
889 *
890 * Hash types refer to the protocol layer addresses which are used to
891 * construct a packet's hash. The hashes are used to differentiate or identify
892 * flows of the protocol layer for the hash type. Hash types are either
893 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
894 *
895 * Properties of hashes:
896 *
897 * 1) Two packets in different flows have different hash values
898 * 2) Two packets in the same flow should have the same hash value
899 *
900 * A hash at a higher layer is considered to be more specific. A driver should
901 * set the most specific hash possible.
902 *
903 * A driver cannot indicate a more specific hash than the layer at which a hash
904 * was computed. For instance an L3 hash cannot be set as an L4 hash.
905 *
906 * A driver may indicate a hash level which is less specific than the
907 * actual layer the hash was computed on. For instance, a hash computed
908 * at L4 may be considered an L3 hash. This should only be done if the
909 * driver can't unambiguously determine that the HW computed the hash at
910 * the higher layer. Note that the "should" in the second property above
911 * permits this.
912 */
913enum pkt_hash_types {
914 PKT_HASH_TYPE_NONE, /* Undefined type */
915 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
916 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
917 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
918};
919
920static inline void
921skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
922{
Tom Herbert61b905d2014-03-24 15:34:47 -0700923 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
Tom Herberta3b18dd2014-07-01 21:33:17 -0700924 skb->sw_hash = 0;
Tom Herbert61b905d2014-03-24 15:34:47 -0700925 skb->hash = hash;
Tom Herbert09323cc2013-12-15 22:16:19 -0800926}
927
Tom Herbert3958afa1b2013-12-15 22:12:06 -0800928void __skb_get_hash(struct sk_buff *skb);
929static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +0000930{
Tom Herberta3b18dd2014-07-01 21:33:17 -0700931 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -0800932 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +0000933
Tom Herbert61b905d2014-03-24 15:34:47 -0700934 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +0000935}
936
Tom Herbert57bdf7f42014-01-15 08:57:54 -0800937static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
938{
Tom Herbert61b905d2014-03-24 15:34:47 -0700939 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -0800940}
941
Tom Herbert7539fad2013-12-15 22:12:18 -0800942static inline void skb_clear_hash(struct sk_buff *skb)
943{
Tom Herbert61b905d2014-03-24 15:34:47 -0700944 skb->hash = 0;
Tom Herberta3b18dd2014-07-01 21:33:17 -0700945 skb->sw_hash = 0;
Tom Herbert61b905d2014-03-24 15:34:47 -0700946 skb->l4_hash = 0;
Tom Herbert7539fad2013-12-15 22:12:18 -0800947}
948
949static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
950{
Tom Herbert61b905d2014-03-24 15:34:47 -0700951 if (!skb->l4_hash)
Tom Herbert7539fad2013-12-15 22:12:18 -0800952 skb_clear_hash(skb);
953}
954
Tom Herbert3df7a742013-12-15 22:16:29 -0800955static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
956{
Tom Herbert61b905d2014-03-24 15:34:47 -0700957 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -0700958 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -0700959 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -0800960};
961
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700962#ifdef NET_SKBUFF_DATA_USES_OFFSET
963static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
964{
965 return skb->head + skb->end;
966}
Alexander Duyckec47ea82012-05-04 14:26:56 +0000967
968static inline unsigned int skb_end_offset(const struct sk_buff *skb)
969{
970 return skb->end;
971}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700972#else
973static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
974{
975 return skb->end;
976}
Alexander Duyckec47ea82012-05-04 14:26:56 +0000977
978static inline unsigned int skb_end_offset(const struct sk_buff *skb)
979{
980 return skb->end - skb->head;
981}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700982#endif
983
Linus Torvalds1da177e2005-04-16 15:20:36 -0700984/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700985#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700986
Patrick Ohlyac45f602009-02-12 05:03:37 +0000987static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
988{
989 return &skb_shinfo(skb)->hwtstamps;
990}
991
Linus Torvalds1da177e2005-04-16 15:20:36 -0700992/**
993 * skb_queue_empty - check if a queue is empty
994 * @list: queue head
995 *
996 * Returns true if the queue is empty, false otherwise.
997 */
998static inline int skb_queue_empty(const struct sk_buff_head *list)
999{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001000 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001001}
1002
1003/**
David S. Millerfc7ebb22008-09-23 00:34:07 -07001004 * skb_queue_is_last - check if skb is the last entry in the queue
1005 * @list: queue head
1006 * @skb: buffer
1007 *
1008 * Returns true if @skb is the last buffer on the list.
1009 */
1010static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1011 const struct sk_buff *skb)
1012{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001013 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -07001014}
1015
1016/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001017 * skb_queue_is_first - check if skb is the first entry in the queue
1018 * @list: queue head
1019 * @skb: buffer
1020 *
1021 * Returns true if @skb is the first buffer on the list.
1022 */
1023static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1024 const struct sk_buff *skb)
1025{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001026 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001027}
1028
1029/**
David S. Miller249c8b42008-09-23 00:44:42 -07001030 * skb_queue_next - return the next packet in the queue
1031 * @list: queue head
1032 * @skb: current buffer
1033 *
1034 * Return the next packet in @list after @skb. It is only valid to
1035 * call this if skb_queue_is_last() evaluates to false.
1036 */
1037static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1038 const struct sk_buff *skb)
1039{
1040 /* This BUG_ON may seem severe, but if we just return then we
1041 * are going to dereference garbage.
1042 */
1043 BUG_ON(skb_queue_is_last(list, skb));
1044 return skb->next;
1045}
1046
1047/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001048 * skb_queue_prev - return the prev packet in the queue
1049 * @list: queue head
1050 * @skb: current buffer
1051 *
1052 * Return the prev packet in @list before @skb. It is only valid to
1053 * call this if skb_queue_is_first() evaluates to false.
1054 */
1055static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1056 const struct sk_buff *skb)
1057{
1058 /* This BUG_ON may seem severe, but if we just return then we
1059 * are going to dereference garbage.
1060 */
1061 BUG_ON(skb_queue_is_first(list, skb));
1062 return skb->prev;
1063}
1064
1065/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001066 * skb_get - reference buffer
1067 * @skb: buffer to reference
1068 *
1069 * Makes another reference to a socket buffer and returns a pointer
1070 * to the buffer.
1071 */
1072static inline struct sk_buff *skb_get(struct sk_buff *skb)
1073{
1074 atomic_inc(&skb->users);
1075 return skb;
1076}
1077
1078/*
1079 * If users == 1, we are the only owner and are can avoid redundant
1080 * atomic change.
1081 */
1082
1083/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001084 * skb_cloned - is the buffer a clone
1085 * @skb: buffer to check
1086 *
1087 * Returns true if the buffer was generated with skb_clone() and is
1088 * one of multiple shared copies of the buffer. Cloned buffers are
1089 * shared data so must not be written to under normal circumstances.
1090 */
1091static inline int skb_cloned(const struct sk_buff *skb)
1092{
1093 return skb->cloned &&
1094 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1095}
1096
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001097static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1098{
1099 might_sleep_if(pri & __GFP_WAIT);
1100
1101 if (skb_cloned(skb))
1102 return pskb_expand_head(skb, 0, 0, pri);
1103
1104 return 0;
1105}
1106
Linus Torvalds1da177e2005-04-16 15:20:36 -07001107/**
1108 * skb_header_cloned - is the header a clone
1109 * @skb: buffer to check
1110 *
1111 * Returns true if modifying the header part of the buffer requires
1112 * the data to be copied.
1113 */
1114static inline int skb_header_cloned(const struct sk_buff *skb)
1115{
1116 int dataref;
1117
1118 if (!skb->cloned)
1119 return 0;
1120
1121 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1122 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1123 return dataref != 1;
1124}
1125
1126/**
1127 * skb_header_release - release reference to header
1128 * @skb: buffer to operate on
1129 *
1130 * Drop a reference to the header part of the buffer. This is done
1131 * by acquiring a payload reference. You must not read from the header
1132 * part of skb->data after this.
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001133 * Note : Check if you can use __skb_header_release() instead.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001134 */
1135static inline void skb_header_release(struct sk_buff *skb)
1136{
1137 BUG_ON(skb->nohdr);
1138 skb->nohdr = 1;
1139 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1140}
1141
1142/**
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001143 * __skb_header_release - release reference to header
1144 * @skb: buffer to operate on
1145 *
1146 * Variant of skb_header_release() assuming skb is private to caller.
1147 * We can avoid one atomic operation.
1148 */
1149static inline void __skb_header_release(struct sk_buff *skb)
1150{
1151 skb->nohdr = 1;
1152 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1153}
1154
1155
1156/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001157 * skb_shared - is the buffer shared
1158 * @skb: buffer to check
1159 *
1160 * Returns true if more than one person has a reference to this
1161 * buffer.
1162 */
1163static inline int skb_shared(const struct sk_buff *skb)
1164{
1165 return atomic_read(&skb->users) != 1;
1166}
1167
1168/**
1169 * skb_share_check - check if buffer is shared and if so clone it
1170 * @skb: buffer to check
1171 * @pri: priority for memory allocation
1172 *
1173 * If the buffer is shared the buffer is cloned and the old copy
1174 * drops a reference. A new clone with a single reference is returned.
1175 * If the buffer is not shared the original buffer is returned. When
1176 * being called from interrupt status or with spinlocks held pri must
1177 * be GFP_ATOMIC.
1178 *
1179 * NULL is returned on a memory allocation failure.
1180 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001181static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182{
1183 might_sleep_if(pri & __GFP_WAIT);
1184 if (skb_shared(skb)) {
1185 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001186
1187 if (likely(nskb))
1188 consume_skb(skb);
1189 else
1190 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 skb = nskb;
1192 }
1193 return skb;
1194}
1195
1196/*
1197 * Copy shared buffers into a new sk_buff. We effectively do COW on
1198 * packets to handle cases where we have a local reader and forward
1199 * and a couple of other messy ones. The normal one is tcpdumping
1200 * a packet thats being forwarded.
1201 */
1202
1203/**
1204 * skb_unshare - make a copy of a shared buffer
1205 * @skb: buffer to check
1206 * @pri: priority for memory allocation
1207 *
1208 * If the socket buffer is a clone then this function creates a new
1209 * copy of the data, drops a reference count on the old copy and returns
1210 * the new copy with the reference count at 1. If the buffer is not a clone
1211 * the original buffer is returned. When called with a spinlock held or
1212 * from interrupt state @pri must be %GFP_ATOMIC
1213 *
1214 * %NULL is returned on a memory allocation failure.
1215 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001216static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001217 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001218{
1219 might_sleep_if(pri & __GFP_WAIT);
1220 if (skb_cloned(skb)) {
1221 struct sk_buff *nskb = skb_copy(skb, pri);
Alexander Aring31eff812014-10-10 23:10:47 +02001222
1223 /* Free our shared copy */
1224 if (likely(nskb))
1225 consume_skb(skb);
1226 else
1227 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228 skb = nskb;
1229 }
1230 return skb;
1231}
1232
1233/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001234 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001235 * @list_: list to peek at
1236 *
1237 * Peek an &sk_buff. Unlike most other operations you _MUST_
1238 * be careful with this one. A peek leaves the buffer on the
1239 * list and someone else may run off with it. You must hold
1240 * the appropriate locks or have a private queue to do this.
1241 *
1242 * Returns %NULL for an empty list or a pointer to the head element.
1243 * The reference count is not incremented and the reference is therefore
1244 * volatile. Use with caution.
1245 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001246static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001247{
Eric Dumazet18d07002012-04-30 16:31:46 +00001248 struct sk_buff *skb = list_->next;
1249
1250 if (skb == (struct sk_buff *)list_)
1251 skb = NULL;
1252 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001253}
1254
1255/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001256 * skb_peek_next - peek skb following the given one from a queue
1257 * @skb: skb to start from
1258 * @list_: list to peek at
1259 *
1260 * Returns %NULL when the end of the list is met or a pointer to the
1261 * next element. The reference count is not incremented and the
1262 * reference is therefore volatile. Use with caution.
1263 */
1264static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1265 const struct sk_buff_head *list_)
1266{
1267 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001268
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001269 if (next == (struct sk_buff *)list_)
1270 next = NULL;
1271 return next;
1272}
1273
1274/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001275 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001276 * @list_: list to peek at
1277 *
1278 * Peek an &sk_buff. Unlike most other operations you _MUST_
1279 * be careful with this one. A peek leaves the buffer on the
1280 * list and someone else may run off with it. You must hold
1281 * the appropriate locks or have a private queue to do this.
1282 *
1283 * Returns %NULL for an empty list or a pointer to the tail element.
1284 * The reference count is not incremented and the reference is therefore
1285 * volatile. Use with caution.
1286 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001287static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288{
Eric Dumazet18d07002012-04-30 16:31:46 +00001289 struct sk_buff *skb = list_->prev;
1290
1291 if (skb == (struct sk_buff *)list_)
1292 skb = NULL;
1293 return skb;
1294
Linus Torvalds1da177e2005-04-16 15:20:36 -07001295}
1296
1297/**
1298 * skb_queue_len - get queue length
1299 * @list_: list to measure
1300 *
1301 * Return the length of an &sk_buff queue.
1302 */
1303static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1304{
1305 return list_->qlen;
1306}
1307
David S. Miller67fed452008-09-21 22:36:24 -07001308/**
1309 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1310 * @list: queue to initialize
1311 *
1312 * This initializes only the list and queue length aspects of
1313 * an sk_buff_head object. This allows to initialize the list
1314 * aspects of an sk_buff_head without reinitializing things like
1315 * the spinlock. It can also be used for on-stack sk_buff_head
1316 * objects where the spinlock is known to not be used.
1317 */
1318static inline void __skb_queue_head_init(struct sk_buff_head *list)
1319{
1320 list->prev = list->next = (struct sk_buff *)list;
1321 list->qlen = 0;
1322}
1323
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001324/*
1325 * This function creates a split out lock class for each invocation;
1326 * this is needed for now since a whole lot of users of the skb-queue
1327 * infrastructure in drivers have different locking usage (in hardirq)
1328 * than the networking core (in softirq only). In the long run either the
1329 * network layer or drivers should need annotation to consolidate the
1330 * main types of usage into 3 classes.
1331 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001332static inline void skb_queue_head_init(struct sk_buff_head *list)
1333{
1334 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001335 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001336}
1337
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001338static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1339 struct lock_class_key *class)
1340{
1341 skb_queue_head_init(list);
1342 lockdep_set_class(&list->lock, class);
1343}
1344
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001346 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001347 *
1348 * The "__skb_xxxx()" functions are the non-atomic ones that
1349 * can only be called with interrupts disabled.
1350 */
Joe Perches7965bd42013-09-26 14:48:15 -07001351void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1352 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001353static inline void __skb_insert(struct sk_buff *newsk,
1354 struct sk_buff *prev, struct sk_buff *next,
1355 struct sk_buff_head *list)
1356{
1357 newsk->next = next;
1358 newsk->prev = prev;
1359 next->prev = prev->next = newsk;
1360 list->qlen++;
1361}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362
David S. Miller67fed452008-09-21 22:36:24 -07001363static inline void __skb_queue_splice(const struct sk_buff_head *list,
1364 struct sk_buff *prev,
1365 struct sk_buff *next)
1366{
1367 struct sk_buff *first = list->next;
1368 struct sk_buff *last = list->prev;
1369
1370 first->prev = prev;
1371 prev->next = first;
1372
1373 last->next = next;
1374 next->prev = last;
1375}
1376
1377/**
1378 * skb_queue_splice - join two skb lists, this is designed for stacks
1379 * @list: the new list to add
1380 * @head: the place to add it in the first list
1381 */
1382static inline void skb_queue_splice(const struct sk_buff_head *list,
1383 struct sk_buff_head *head)
1384{
1385 if (!skb_queue_empty(list)) {
1386 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001387 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001388 }
1389}
1390
1391/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001392 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001393 * @list: the new list to add
1394 * @head: the place to add it in the first list
1395 *
1396 * The list at @list is reinitialised
1397 */
1398static inline void skb_queue_splice_init(struct sk_buff_head *list,
1399 struct sk_buff_head *head)
1400{
1401 if (!skb_queue_empty(list)) {
1402 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001403 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001404 __skb_queue_head_init(list);
1405 }
1406}
1407
1408/**
1409 * skb_queue_splice_tail - join two skb lists, each list being a queue
1410 * @list: the new list to add
1411 * @head: the place to add it in the first list
1412 */
1413static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1414 struct sk_buff_head *head)
1415{
1416 if (!skb_queue_empty(list)) {
1417 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001418 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001419 }
1420}
1421
1422/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001423 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001424 * @list: the new list to add
1425 * @head: the place to add it in the first list
1426 *
1427 * Each of the lists is a queue.
1428 * The list at @list is reinitialised
1429 */
1430static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1431 struct sk_buff_head *head)
1432{
1433 if (!skb_queue_empty(list)) {
1434 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001435 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001436 __skb_queue_head_init(list);
1437 }
1438}
1439
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001441 * __skb_queue_after - queue a buffer at the list head
1442 * @list: list to use
1443 * @prev: place after this buffer
1444 * @newsk: buffer to queue
1445 *
1446 * Queue a buffer int the middle of a list. This function takes no locks
1447 * and you must therefore hold required locks before calling it.
1448 *
1449 * A buffer cannot be placed on two lists at the same time.
1450 */
1451static inline void __skb_queue_after(struct sk_buff_head *list,
1452 struct sk_buff *prev,
1453 struct sk_buff *newsk)
1454{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001455 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001456}
1457
Joe Perches7965bd42013-09-26 14:48:15 -07001458void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1459 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001460
Gerrit Renkerf5572852008-04-14 00:05:28 -07001461static inline void __skb_queue_before(struct sk_buff_head *list,
1462 struct sk_buff *next,
1463 struct sk_buff *newsk)
1464{
1465 __skb_insert(newsk, next->prev, next, list);
1466}
1467
Stephen Hemminger300ce172005-10-30 13:47:34 -08001468/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001469 * __skb_queue_head - queue a buffer at the list head
1470 * @list: list to use
1471 * @newsk: buffer to queue
1472 *
1473 * Queue a buffer at the start of a list. This function takes no locks
1474 * and you must therefore hold required locks before calling it.
1475 *
1476 * A buffer cannot be placed on two lists at the same time.
1477 */
Joe Perches7965bd42013-09-26 14:48:15 -07001478void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001479static inline void __skb_queue_head(struct sk_buff_head *list,
1480 struct sk_buff *newsk)
1481{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001482 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001483}
1484
1485/**
1486 * __skb_queue_tail - queue a buffer at the list tail
1487 * @list: list to use
1488 * @newsk: buffer to queue
1489 *
1490 * Queue a buffer at the end of a list. This function takes no locks
1491 * and you must therefore hold required locks before calling it.
1492 *
1493 * A buffer cannot be placed on two lists at the same time.
1494 */
Joe Perches7965bd42013-09-26 14:48:15 -07001495void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001496static inline void __skb_queue_tail(struct sk_buff_head *list,
1497 struct sk_buff *newsk)
1498{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001499 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500}
1501
Linus Torvalds1da177e2005-04-16 15:20:36 -07001502/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001503 * remove sk_buff from list. _Must_ be called atomically, and with
1504 * the list known..
1505 */
Joe Perches7965bd42013-09-26 14:48:15 -07001506void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1508{
1509 struct sk_buff *next, *prev;
1510
1511 list->qlen--;
1512 next = skb->next;
1513 prev = skb->prev;
1514 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001515 next->prev = prev;
1516 prev->next = next;
1517}
1518
Gerrit Renkerf525c062008-04-14 00:04:12 -07001519/**
1520 * __skb_dequeue - remove from the head of the queue
1521 * @list: list to dequeue from
1522 *
1523 * Remove the head of the list. This function does not take any locks
1524 * so must be used with appropriate locks held only. The head item is
1525 * returned or %NULL if the list is empty.
1526 */
Joe Perches7965bd42013-09-26 14:48:15 -07001527struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001528static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1529{
1530 struct sk_buff *skb = skb_peek(list);
1531 if (skb)
1532 __skb_unlink(skb, list);
1533 return skb;
1534}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535
1536/**
1537 * __skb_dequeue_tail - remove from the tail of the queue
1538 * @list: list to dequeue from
1539 *
1540 * Remove the tail of the list. This function does not take any locks
1541 * so must be used with appropriate locks held only. The tail item is
1542 * returned or %NULL if the list is empty.
1543 */
Joe Perches7965bd42013-09-26 14:48:15 -07001544struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001545static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1546{
1547 struct sk_buff *skb = skb_peek_tail(list);
1548 if (skb)
1549 __skb_unlink(skb, list);
1550 return skb;
1551}
1552
1553
David S. Millerbdcc0922012-03-07 20:53:36 -05001554static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001555{
1556 return skb->data_len;
1557}
1558
1559static inline unsigned int skb_headlen(const struct sk_buff *skb)
1560{
1561 return skb->len - skb->data_len;
1562}
1563
1564static inline int skb_pagelen(const struct sk_buff *skb)
1565{
1566 int i, len = 0;
1567
1568 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001569 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570 return len + skb_headlen(skb);
1571}
1572
Ian Campbell131ea662011-08-19 06:25:00 +00001573/**
1574 * __skb_fill_page_desc - initialise a paged fragment in an skb
1575 * @skb: buffer containing fragment to be initialised
1576 * @i: paged fragment index to initialise
1577 * @page: the page to use for this fragment
1578 * @off: the offset to the data with @page
1579 * @size: the length of the data
1580 *
1581 * Initialises the @i'th fragment of @skb to point to &size bytes at
1582 * offset @off within @page.
1583 *
1584 * Does not take any additional reference on the fragment.
1585 */
1586static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1587 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001588{
1589 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1590
Mel Gormanc48a11c2012-07-31 16:44:23 -07001591 /*
1592 * Propagate page->pfmemalloc to the skb if we can. The problem is
1593 * that not all callers have unique ownership of the page. If
1594 * pfmemalloc is set, we check the mapping as a mapping implies
1595 * page->index is set (index and pfmemalloc share space).
1596 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1597 * do not lose pfmemalloc information as the pages would not be
1598 * allocated using __GFP_MEMALLOC.
1599 */
Ian Campbella8605c62011-10-19 23:01:49 +00001600 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001601 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001602 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001603
1604 page = compound_head(page);
1605 if (page->pfmemalloc && !page->mapping)
1606 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001607}
1608
1609/**
1610 * skb_fill_page_desc - initialise a paged fragment in an skb
1611 * @skb: buffer containing fragment to be initialised
1612 * @i: paged fragment index to initialise
1613 * @page: the page to use for this fragment
1614 * @off: the offset to the data with @page
1615 * @size: the length of the data
1616 *
1617 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001618 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001619 * addition updates @skb such that @i is the last fragment.
1620 *
1621 * Does not take any additional reference on the fragment.
1622 */
1623static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1624 struct page *page, int off, int size)
1625{
1626 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627 skb_shinfo(skb)->nr_frags = i + 1;
1628}
1629
Joe Perches7965bd42013-09-26 14:48:15 -07001630void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1631 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001632
Jason Wangf8e617e2013-11-01 14:07:47 +08001633void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1634 unsigned int truesize);
1635
Linus Torvalds1da177e2005-04-16 15:20:36 -07001636#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001637#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1639
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001640#ifdef NET_SKBUFF_DATA_USES_OFFSET
1641static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1642{
1643 return skb->head + skb->tail;
1644}
1645
1646static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1647{
1648 skb->tail = skb->data - skb->head;
1649}
1650
1651static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1652{
1653 skb_reset_tail_pointer(skb);
1654 skb->tail += offset;
1655}
Simon Horman7cc46192013-05-28 20:34:29 +00001656
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001657#else /* NET_SKBUFF_DATA_USES_OFFSET */
1658static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1659{
1660 return skb->tail;
1661}
1662
1663static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1664{
1665 skb->tail = skb->data;
1666}
1667
1668static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1669{
1670 skb->tail = skb->data + offset;
1671}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001672
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001673#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1674
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675/*
1676 * Add data to an sk_buff
1677 */
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001678unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001679unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1681{
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001682 unsigned char *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683 SKB_LINEAR_ASSERT(skb);
1684 skb->tail += len;
1685 skb->len += len;
1686 return tmp;
1687}
1688
Joe Perches7965bd42013-09-26 14:48:15 -07001689unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001690static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1691{
1692 skb->data -= len;
1693 skb->len += len;
1694 return skb->data;
1695}
1696
Joe Perches7965bd42013-09-26 14:48:15 -07001697unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001698static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1699{
1700 skb->len -= len;
1701 BUG_ON(skb->len < skb->data_len);
1702 return skb->data += len;
1703}
1704
David S. Miller47d29642010-05-02 02:21:44 -07001705static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1706{
1707 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1708}
1709
Joe Perches7965bd42013-09-26 14:48:15 -07001710unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711
1712static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1713{
1714 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07001715 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001716 return NULL;
1717 skb->len -= len;
1718 return skb->data += len;
1719}
1720
1721static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1722{
1723 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1724}
1725
1726static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1727{
1728 if (likely(len <= skb_headlen(skb)))
1729 return 1;
1730 if (unlikely(len > skb->len))
1731 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07001732 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001733}
1734
1735/**
1736 * skb_headroom - bytes at buffer head
1737 * @skb: buffer to check
1738 *
1739 * Return the number of bytes of free space at the head of an &sk_buff.
1740 */
Chuck Leverc2636b42007-10-23 21:07:32 -07001741static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001742{
1743 return skb->data - skb->head;
1744}
1745
1746/**
1747 * skb_tailroom - bytes at buffer end
1748 * @skb: buffer to check
1749 *
1750 * Return the number of bytes of free space at the tail of an sk_buff
1751 */
1752static inline int skb_tailroom(const struct sk_buff *skb)
1753{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001754 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001755}
1756
1757/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00001758 * skb_availroom - bytes at buffer end
1759 * @skb: buffer to check
1760 *
1761 * Return the number of bytes of free space at the tail of an sk_buff
1762 * allocated by sk_stream_alloc()
1763 */
1764static inline int skb_availroom(const struct sk_buff *skb)
1765{
Eric Dumazet16fad692013-03-14 05:40:32 +00001766 if (skb_is_nonlinear(skb))
1767 return 0;
1768
1769 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00001770}
1771
1772/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001773 * skb_reserve - adjust headroom
1774 * @skb: buffer to alter
1775 * @len: bytes to move
1776 *
1777 * Increase the headroom of an empty &sk_buff by reducing the tail
1778 * room. This is only allowed for an empty buffer.
1779 */
David S. Miller8243126c2006-01-17 02:54:21 -08001780static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781{
1782 skb->data += len;
1783 skb->tail += len;
1784}
1785
Tom Herbert8bce6d72014-09-29 20:22:29 -07001786#define ENCAP_TYPE_ETHER 0
1787#define ENCAP_TYPE_IPPROTO 1
1788
1789static inline void skb_set_inner_protocol(struct sk_buff *skb,
1790 __be16 protocol)
1791{
1792 skb->inner_protocol = protocol;
1793 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1794}
1795
1796static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1797 __u8 ipproto)
1798{
1799 skb->inner_ipproto = ipproto;
1800 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
1801}
1802
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001803static inline void skb_reset_inner_headers(struct sk_buff *skb)
1804{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00001805 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001806 skb->inner_network_header = skb->network_header;
1807 skb->inner_transport_header = skb->transport_header;
1808}
1809
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00001810static inline void skb_reset_mac_len(struct sk_buff *skb)
1811{
1812 skb->mac_len = skb->network_header - skb->mac_header;
1813}
1814
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001815static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1816 *skb)
1817{
1818 return skb->head + skb->inner_transport_header;
1819}
1820
1821static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1822{
1823 skb->inner_transport_header = skb->data - skb->head;
1824}
1825
1826static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1827 const int offset)
1828{
1829 skb_reset_inner_transport_header(skb);
1830 skb->inner_transport_header += offset;
1831}
1832
1833static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1834{
1835 return skb->head + skb->inner_network_header;
1836}
1837
1838static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1839{
1840 skb->inner_network_header = skb->data - skb->head;
1841}
1842
1843static inline void skb_set_inner_network_header(struct sk_buff *skb,
1844 const int offset)
1845{
1846 skb_reset_inner_network_header(skb);
1847 skb->inner_network_header += offset;
1848}
1849
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00001850static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1851{
1852 return skb->head + skb->inner_mac_header;
1853}
1854
1855static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1856{
1857 skb->inner_mac_header = skb->data - skb->head;
1858}
1859
1860static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1861 const int offset)
1862{
1863 skb_reset_inner_mac_header(skb);
1864 skb->inner_mac_header += offset;
1865}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00001866static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1867{
Cong Wang35d04612013-05-29 15:16:05 +08001868 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00001869}
1870
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07001871static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1872{
1873 return skb->head + skb->transport_header;
1874}
1875
1876static inline void skb_reset_transport_header(struct sk_buff *skb)
1877{
1878 skb->transport_header = skb->data - skb->head;
1879}
1880
1881static inline void skb_set_transport_header(struct sk_buff *skb,
1882 const int offset)
1883{
1884 skb_reset_transport_header(skb);
1885 skb->transport_header += offset;
1886}
1887
1888static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1889{
1890 return skb->head + skb->network_header;
1891}
1892
1893static inline void skb_reset_network_header(struct sk_buff *skb)
1894{
1895 skb->network_header = skb->data - skb->head;
1896}
1897
1898static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1899{
1900 skb_reset_network_header(skb);
1901 skb->network_header += offset;
1902}
1903
1904static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1905{
1906 return skb->head + skb->mac_header;
1907}
1908
1909static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1910{
Cong Wang35d04612013-05-29 15:16:05 +08001911 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07001912}
1913
1914static inline void skb_reset_mac_header(struct sk_buff *skb)
1915{
1916 skb->mac_header = skb->data - skb->head;
1917}
1918
1919static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1920{
1921 skb_reset_mac_header(skb);
1922 skb->mac_header += offset;
1923}
1924
Timo Teräs0e3da5b2013-12-16 11:02:09 +02001925static inline void skb_pop_mac_header(struct sk_buff *skb)
1926{
1927 skb->mac_header = skb->network_header;
1928}
1929
Ying Xuefbbdb8f2013-03-27 16:46:06 +00001930static inline void skb_probe_transport_header(struct sk_buff *skb,
1931 const int offset_hint)
1932{
1933 struct flow_keys keys;
1934
1935 if (skb_transport_header_was_set(skb))
1936 return;
1937 else if (skb_flow_dissect(skb, &keys))
1938 skb_set_transport_header(skb, keys.thoff);
1939 else
1940 skb_set_transport_header(skb, offset_hint);
1941}
1942
Eric Dumazet03606892012-02-23 10:55:02 +00001943static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1944{
1945 if (skb_mac_header_was_set(skb)) {
1946 const unsigned char *old_mac = skb_mac_header(skb);
1947
1948 skb_set_mac_header(skb, -skb->mac_len);
1949 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1950 }
1951}
1952
Michał Mirosław04fb4512010-12-14 15:24:08 +00001953static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1954{
1955 return skb->csum_start - skb_headroom(skb);
1956}
1957
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07001958static inline int skb_transport_offset(const struct sk_buff *skb)
1959{
1960 return skb_transport_header(skb) - skb->data;
1961}
1962
1963static inline u32 skb_network_header_len(const struct sk_buff *skb)
1964{
1965 return skb->transport_header - skb->network_header;
1966}
1967
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001968static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1969{
1970 return skb->inner_transport_header - skb->inner_network_header;
1971}
1972
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07001973static inline int skb_network_offset(const struct sk_buff *skb)
1974{
1975 return skb_network_header(skb) - skb->data;
1976}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03001977
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001978static inline int skb_inner_network_offset(const struct sk_buff *skb)
1979{
1980 return skb_inner_network_header(skb) - skb->data;
1981}
1982
Changli Gaof9599ce2010-08-04 04:43:44 +00001983static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1984{
1985 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1986}
1987
Linus Torvalds1da177e2005-04-16 15:20:36 -07001988/*
1989 * CPUs often take a performance hit when accessing unaligned memory
1990 * locations. The actual performance hit varies, it can be small if the
1991 * hardware handles it or large if we have to take an exception and fix it
1992 * in software.
1993 *
1994 * Since an ethernet header is 14 bytes network drivers often end up with
1995 * the IP header at an unaligned offset. The IP header can be aligned by
1996 * shifting the start of the packet by 2 bytes. Drivers should do this
1997 * with:
1998 *
Tobias Klauser8660c122009-07-13 22:48:16 +00001999 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 *
2001 * The downside to this alignment of the IP header is that the DMA is now
2002 * unaligned. On some architectures the cost of an unaligned DMA is high
2003 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00002004 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2006 * to be overridden.
2007 */
2008#ifndef NET_IP_ALIGN
2009#define NET_IP_ALIGN 2
2010#endif
2011
Anton Blanchard025be812006-03-31 02:27:06 -08002012/*
2013 * The networking layer reserves some headroom in skb data (via
2014 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2015 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08002016 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08002017 *
2018 * Unfortunately this headroom changes the DMA alignment of the resulting
2019 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2020 * on some architectures. An architecture can override this value,
2021 * perhaps setting it to a cacheline in size (since that will maintain
2022 * cacheline alignment of the DMA). It must be a power of 2.
2023 *
David S. Millerd6301d32009-02-08 19:24:13 -08002024 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08002025 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07002026 *
2027 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2028 * to reduce average number of cache lines per packet.
2029 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07002030 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08002031 */
2032#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07002033#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08002034#endif
2035
Joe Perches7965bd42013-09-26 14:48:15 -07002036int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002037
2038static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2039{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00002040 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07002041 WARN_ON(1);
2042 return;
2043 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002044 skb->len = len;
2045 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002046}
2047
Joe Perches7965bd42013-09-26 14:48:15 -07002048void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002049
2050static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2051{
Herbert Xu3cc0e872006-06-09 16:13:38 -07002052 if (skb->data_len)
2053 return ___pskb_trim(skb, len);
2054 __skb_trim(skb, len);
2055 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002056}
2057
2058static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2059{
2060 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2061}
2062
2063/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07002064 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2065 * @skb: buffer to alter
2066 * @len: new length
2067 *
2068 * This is identical to pskb_trim except that the caller knows that
2069 * the skb is not cloned so we should never get an error due to out-
2070 * of-memory.
2071 */
2072static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2073{
2074 int err = pskb_trim(skb, len);
2075 BUG_ON(err);
2076}
2077
2078/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 * skb_orphan - orphan a buffer
2080 * @skb: buffer to orphan
2081 *
2082 * If a buffer currently has an owner then we call the owner's
2083 * destructor function and make the @skb unowned. The buffer continues
2084 * to exist but is no longer charged to its former owner.
2085 */
2086static inline void skb_orphan(struct sk_buff *skb)
2087{
Eric Dumazetc34a7612013-07-30 16:11:15 -07002088 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002089 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002090 skb->destructor = NULL;
2091 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07002092 } else {
2093 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002094 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002095}
2096
2097/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002098 * skb_orphan_frags - orphan the frags contained in a buffer
2099 * @skb: buffer to orphan frags from
2100 * @gfp_mask: allocation mask for replacement pages
2101 *
2102 * For each frag in the SKB which needs a destructor (i.e. has an
2103 * owner) create a copy of that frag and release the original
2104 * page by calling the destructor.
2105 */
2106static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2107{
2108 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2109 return 0;
2110 return skb_copy_ubufs(skb, gfp_mask);
2111}
2112
2113/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114 * __skb_queue_purge - empty a list
2115 * @list: list to empty
2116 *
2117 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2118 * the list and one reference dropped. This function does not take the
2119 * list lock and the caller must hold the relevant locks to use it.
2120 */
Joe Perches7965bd42013-09-26 14:48:15 -07002121void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002122static inline void __skb_queue_purge(struct sk_buff_head *list)
2123{
2124 struct sk_buff *skb;
2125 while ((skb = __skb_dequeue(list)) != NULL)
2126 kfree_skb(skb);
2127}
2128
Alexander Duycke5e67302013-02-08 10:17:15 +00002129#define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
2130#define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
2131#define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
2132
Joe Perches7965bd42013-09-26 14:48:15 -07002133void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002134
Joe Perches7965bd42013-09-26 14:48:15 -07002135struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2136 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002137
2138/**
2139 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2140 * @dev: network device to receive on
2141 * @length: length to allocate
2142 *
2143 * Allocate a new &sk_buff and assign it a usage count of one. The
2144 * buffer has unspecified headroom built in. Users should allocate
2145 * the headroom they think they need without accounting for the
2146 * built in space. The built in space is used for optimisations.
2147 *
2148 * %NULL is returned if there is no free memory. Although this function
2149 * allocates memory it can be called from an interrupt.
2150 */
2151static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002152 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002153{
2154 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2155}
2156
Eric Dumazet6f532612012-05-18 05:12:12 +00002157/* legacy helper around __netdev_alloc_skb() */
2158static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2159 gfp_t gfp_mask)
2160{
2161 return __netdev_alloc_skb(NULL, length, gfp_mask);
2162}
2163
2164/* legacy helper around netdev_alloc_skb() */
2165static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2166{
2167 return netdev_alloc_skb(NULL, length);
2168}
2169
2170
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002171static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2172 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002173{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002174 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002175
2176 if (NET_IP_ALIGN && skb)
2177 skb_reserve(skb, NET_IP_ALIGN);
2178 return skb;
2179}
2180
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002181static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2182 unsigned int length)
2183{
2184 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2185}
2186
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002187/**
Alexander Duyck71dfda52014-11-11 09:26:34 -08002188 * __dev_alloc_pages - allocate page for network Rx
2189 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2190 * @order: size of the allocation
2191 *
2192 * Allocate a new page.
2193 *
2194 * %NULL is returned if there is no free memory.
2195*/
2196static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2197 unsigned int order)
2198{
2199 /* This piece of code contains several assumptions.
2200 * 1. This is for device Rx, therefor a cold page is preferred.
2201 * 2. The expectation is the user wants a compound page.
2202 * 3. If requesting a order 0 page it will not be compound
2203 * due to the check to see if order has a value in prep_new_page
2204 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2205 * code in gfp_to_alloc_flags that should be enforcing this.
2206 */
2207 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2208
2209 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2210}
2211
2212static inline struct page *dev_alloc_pages(unsigned int order)
2213{
2214 return __dev_alloc_pages(GFP_ATOMIC, order);
2215}
2216
2217/**
2218 * __dev_alloc_page - allocate a page for network Rx
2219 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2220 *
2221 * Allocate a new page.
2222 *
2223 * %NULL is returned if there is no free memory.
2224 */
2225static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2226{
2227 return __dev_alloc_pages(gfp_mask, 0);
2228}
2229
2230static inline struct page *dev_alloc_page(void)
2231{
2232 return __dev_alloc_page(GFP_ATOMIC);
2233}
2234
2235/**
Mel Gorman06140022012-07-31 16:44:24 -07002236 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2237 * @page: The page that was allocated from skb_alloc_page
2238 * @skb: The skb that may need pfmemalloc set
2239 */
2240static inline void skb_propagate_pfmemalloc(struct page *page,
2241 struct sk_buff *skb)
2242{
2243 if (page && page->pfmemalloc)
2244 skb->pfmemalloc = true;
2245}
2246
Eric Dumazet564824b2010-10-11 19:05:25 +00002247/**
Masanari Iidae2278672014-02-18 22:54:36 +09002248 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002249 * @frag: the paged fragment
2250 *
2251 * Returns the &struct page associated with @frag.
2252 */
2253static inline struct page *skb_frag_page(const skb_frag_t *frag)
2254{
Ian Campbella8605c62011-10-19 23:01:49 +00002255 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002256}
2257
2258/**
2259 * __skb_frag_ref - take an addition reference on a paged fragment.
2260 * @frag: the paged fragment
2261 *
2262 * Takes an additional reference on the paged fragment @frag.
2263 */
2264static inline void __skb_frag_ref(skb_frag_t *frag)
2265{
2266 get_page(skb_frag_page(frag));
2267}
2268
2269/**
2270 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2271 * @skb: the buffer
2272 * @f: the fragment offset.
2273 *
2274 * Takes an additional reference on the @f'th paged fragment of @skb.
2275 */
2276static inline void skb_frag_ref(struct sk_buff *skb, int f)
2277{
2278 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2279}
2280
2281/**
2282 * __skb_frag_unref - release a reference on a paged fragment.
2283 * @frag: the paged fragment
2284 *
2285 * Releases a reference on the paged fragment @frag.
2286 */
2287static inline void __skb_frag_unref(skb_frag_t *frag)
2288{
2289 put_page(skb_frag_page(frag));
2290}
2291
2292/**
2293 * skb_frag_unref - release a reference on a paged fragment of an skb.
2294 * @skb: the buffer
2295 * @f: the fragment offset
2296 *
2297 * Releases a reference on the @f'th paged fragment of @skb.
2298 */
2299static inline void skb_frag_unref(struct sk_buff *skb, int f)
2300{
2301 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2302}
2303
2304/**
2305 * skb_frag_address - gets the address of the data contained in a paged fragment
2306 * @frag: the paged fragment buffer
2307 *
2308 * Returns the address of the data within @frag. The page must already
2309 * be mapped.
2310 */
2311static inline void *skb_frag_address(const skb_frag_t *frag)
2312{
2313 return page_address(skb_frag_page(frag)) + frag->page_offset;
2314}
2315
2316/**
2317 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2318 * @frag: the paged fragment buffer
2319 *
2320 * Returns the address of the data within @frag. Checks that the page
2321 * is mapped and returns %NULL otherwise.
2322 */
2323static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2324{
2325 void *ptr = page_address(skb_frag_page(frag));
2326 if (unlikely(!ptr))
2327 return NULL;
2328
2329 return ptr + frag->page_offset;
2330}
2331
2332/**
2333 * __skb_frag_set_page - sets the page contained in a paged fragment
2334 * @frag: the paged fragment
2335 * @page: the page to set
2336 *
2337 * Sets the fragment @frag to contain @page.
2338 */
2339static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2340{
Ian Campbella8605c62011-10-19 23:01:49 +00002341 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002342}
2343
2344/**
2345 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2346 * @skb: the buffer
2347 * @f: the fragment offset
2348 * @page: the page to set
2349 *
2350 * Sets the @f'th fragment of @skb to contain @page.
2351 */
2352static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2353 struct page *page)
2354{
2355 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2356}
2357
Eric Dumazet400dfd32013-10-17 16:27:07 -07002358bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2359
Ian Campbell131ea662011-08-19 06:25:00 +00002360/**
2361 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002362 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002363 * @frag: the paged fragment to map
2364 * @offset: the offset within the fragment (starting at the
2365 * fragment's own offset)
2366 * @size: the number of bytes to map
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002367 * @dir: the direction of the mapping (%PCI_DMA_*)
Ian Campbell131ea662011-08-19 06:25:00 +00002368 *
2369 * Maps the page associated with @frag to @device.
2370 */
2371static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2372 const skb_frag_t *frag,
2373 size_t offset, size_t size,
2374 enum dma_data_direction dir)
2375{
2376 return dma_map_page(dev, skb_frag_page(frag),
2377 frag->page_offset + offset, size, dir);
2378}
2379
Eric Dumazet117632e2011-12-03 21:39:53 +00002380static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2381 gfp_t gfp_mask)
2382{
2383 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2384}
2385
Octavian Purdilabad93e92014-06-12 01:36:26 +03002386
2387static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2388 gfp_t gfp_mask)
2389{
2390 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2391}
2392
2393
Ian Campbell131ea662011-08-19 06:25:00 +00002394/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002395 * skb_clone_writable - is the header of a clone writable
2396 * @skb: buffer to check
2397 * @len: length up to which to write
2398 *
2399 * Returns true if modifying the header part of the cloned buffer
2400 * does not requires the data to be copied.
2401 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002402static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002403{
2404 return !skb_header_cloned(skb) &&
2405 skb_headroom(skb) + len <= skb->hdr_len;
2406}
2407
Herbert Xud9cc2042007-09-16 16:21:16 -07002408static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2409 int cloned)
2410{
2411 int delta = 0;
2412
Herbert Xud9cc2042007-09-16 16:21:16 -07002413 if (headroom > skb_headroom(skb))
2414 delta = headroom - skb_headroom(skb);
2415
2416 if (delta || cloned)
2417 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2418 GFP_ATOMIC);
2419 return 0;
2420}
2421
Patrick McHardy334a8132007-06-25 04:35:20 -07002422/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002423 * skb_cow - copy header of skb when it is required
2424 * @skb: buffer to cow
2425 * @headroom: needed headroom
2426 *
2427 * If the skb passed lacks sufficient headroom or its data part
2428 * is shared, data is reallocated. If reallocation fails, an error
2429 * is returned and original skb is not changed.
2430 *
2431 * The result is skb with writable area skb->head...skb->tail
2432 * and at least @headroom of space at head.
2433 */
2434static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2435{
Herbert Xud9cc2042007-09-16 16:21:16 -07002436 return __skb_cow(skb, headroom, skb_cloned(skb));
2437}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002438
Herbert Xud9cc2042007-09-16 16:21:16 -07002439/**
2440 * skb_cow_head - skb_cow but only making the head writable
2441 * @skb: buffer to cow
2442 * @headroom: needed headroom
2443 *
2444 * This function is identical to skb_cow except that we replace the
2445 * skb_cloned check by skb_header_cloned. It should be used when
2446 * you only need to push on some header and do not need to modify
2447 * the data.
2448 */
2449static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2450{
2451 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002452}
2453
2454/**
2455 * skb_padto - pad an skbuff up to a minimal size
2456 * @skb: buffer to pad
2457 * @len: minimal length
2458 *
2459 * Pads up a buffer to ensure the trailing bytes exist and are
2460 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002461 * is untouched. Otherwise it is extended. Returns zero on
2462 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463 */
2464
Herbert Xu5b057c62006-06-23 02:06:41 -07002465static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002466{
2467 unsigned int size = skb->len;
2468 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002469 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002470 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002471}
2472
2473static inline int skb_add_data(struct sk_buff *skb,
2474 char __user *from, int copy)
2475{
2476 const int off = skb->len;
2477
2478 if (skb->ip_summed == CHECKSUM_NONE) {
2479 int err = 0;
Al Viro50842052006-11-14 21:36:34 -08002480 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002481 copy, 0, &err);
2482 if (!err) {
2483 skb->csum = csum_block_add(skb->csum, csum, off);
2484 return 0;
2485 }
2486 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
2487 return 0;
2488
2489 __skb_trim(skb, off);
2490 return -EFAULT;
2491}
2492
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002493static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2494 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002495{
2496 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002497 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002498
Ian Campbellea2ab692011-08-22 23:44:58 +00002499 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00002500 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002501 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002502 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002503}
2504
Herbert Xu364c6ba2006-06-09 16:10:40 -07002505static inline int __skb_linearize(struct sk_buff *skb)
2506{
2507 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2508}
2509
Linus Torvalds1da177e2005-04-16 15:20:36 -07002510/**
2511 * skb_linearize - convert paged skb to linear one
2512 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07002513 *
2514 * If there is no free memory -ENOMEM is returned, otherwise zero
2515 * is returned and the old skb data released.
2516 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07002517static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002518{
Herbert Xu364c6ba2006-06-09 16:10:40 -07002519 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2520}
2521
2522/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00002523 * skb_has_shared_frag - can any frag be overwritten
2524 * @skb: buffer to test
2525 *
2526 * Return true if the skb has at least one frag that might be modified
2527 * by an external entity (as in vmsplice()/sendfile())
2528 */
2529static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2530{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00002531 return skb_is_nonlinear(skb) &&
2532 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002533}
2534
2535/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07002536 * skb_linearize_cow - make sure skb is linear and writable
2537 * @skb: buffer to process
2538 *
2539 * If there is no free memory -ENOMEM is returned, otherwise zero
2540 * is returned and the old skb data released.
2541 */
2542static inline int skb_linearize_cow(struct sk_buff *skb)
2543{
2544 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2545 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002546}
2547
2548/**
2549 * skb_postpull_rcsum - update checksum for received skb after pull
2550 * @skb: buffer to update
2551 * @start: start of data before pull
2552 * @len: length of data pulled
2553 *
2554 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07002555 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2556 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002557 */
2558
2559static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f2006-03-20 22:43:56 -08002560 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002561{
Patrick McHardy84fa7932006-08-29 16:44:56 -07002562 if (skb->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002563 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2564}
2565
Herbert Xucbb042f2006-03-20 22:43:56 -08002566unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2567
David S. Miller7ce5a272013-12-02 17:26:05 -05002568/**
2569 * pskb_trim_rcsum - trim received skb and update checksum
2570 * @skb: buffer to trim
2571 * @len: new length
2572 *
2573 * This is exactly the same as pskb_trim except that it ensures the
2574 * checksum of received packets are still valid after the operation.
2575 */
2576
2577static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2578{
2579 if (likely(len >= skb->len))
2580 return 0;
2581 if (skb->ip_summed == CHECKSUM_COMPLETE)
2582 skb->ip_summed = CHECKSUM_NONE;
2583 return __pskb_trim(skb, len);
2584}
2585
Linus Torvalds1da177e2005-04-16 15:20:36 -07002586#define skb_queue_walk(queue, skb) \
2587 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002588 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002589 skb = skb->next)
2590
James Chapman46f89142007-04-30 00:07:31 -07002591#define skb_queue_walk_safe(queue, skb, tmp) \
2592 for (skb = (queue)->next, tmp = skb->next; \
2593 skb != (struct sk_buff *)(queue); \
2594 skb = tmp, tmp = skb->next)
2595
David S. Miller1164f522008-09-23 00:49:44 -07002596#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002597 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07002598 skb = skb->next)
2599
2600#define skb_queue_walk_from_safe(queue, skb, tmp) \
2601 for (tmp = skb->next; \
2602 skb != (struct sk_buff *)(queue); \
2603 skb = tmp, tmp = skb->next)
2604
Stephen Hemminger300ce172005-10-30 13:47:34 -08002605#define skb_queue_reverse_walk(queue, skb) \
2606 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002607 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08002608 skb = skb->prev)
2609
David S. Miller686a2952011-01-20 22:47:32 -08002610#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2611 for (skb = (queue)->prev, tmp = skb->prev; \
2612 skb != (struct sk_buff *)(queue); \
2613 skb = tmp, tmp = skb->prev)
2614
2615#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2616 for (tmp = skb->prev; \
2617 skb != (struct sk_buff *)(queue); \
2618 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619
David S. Miller21dc3302010-08-23 00:13:46 -07002620static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07002621{
2622 return skb_shinfo(skb)->frag_list != NULL;
2623}
2624
2625static inline void skb_frag_list_init(struct sk_buff *skb)
2626{
2627 skb_shinfo(skb)->frag_list = NULL;
2628}
2629
2630static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2631{
2632 frag->next = skb_shinfo(skb)->frag_list;
2633 skb_shinfo(skb)->frag_list = frag;
2634}
2635
2636#define skb_walk_frags(skb, iter) \
2637 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2638
Joe Perches7965bd42013-09-26 14:48:15 -07002639struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2640 int *peeked, int *off, int *err);
2641struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2642 int *err);
2643unsigned int datagram_poll(struct file *file, struct socket *sock,
2644 struct poll_table_struct *wait);
2645int skb_copy_datagram_iovec(const struct sk_buff *from, int offset,
2646 struct iovec *to, int size);
David S. Miller51f3d022014-11-05 16:46:40 -05002647static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2648 struct msghdr *msg, int size)
2649{
2650 return skb_copy_datagram_iovec(from, offset, msg->msg_iov, size);
2651}
Joe Perches7965bd42013-09-26 14:48:15 -07002652int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, int hlen,
2653 struct iovec *iov);
Al Viro227158d2014-04-06 18:47:38 -04002654static inline int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2655 struct msghdr *msg)
2656{
2657 return skb_copy_and_csum_datagram_iovec(skb, hlen, msg->msg_iov);
2658}
Joe Perches7965bd42013-09-26 14:48:15 -07002659int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
2660 const struct iovec *from, int from_offset,
2661 int len);
2662int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *frm,
2663 int offset, size_t count);
Herbert Xua8f820aa2014-11-07 21:22:22 +08002664int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2665 struct iov_iter *to, int size);
Joe Perches7965bd42013-09-26 14:48:15 -07002666void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2667void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2668int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07002669int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2670int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2671__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2672 int len, __wsum csum);
2673int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2674 struct pipe_inode_info *pipe, unsigned int len,
2675 unsigned int flags);
2676void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01002677unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002678int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2679 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07002680void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2681int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2682void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01002683unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -07002684struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Vlad Yasevich0d5501c2014-08-08 14:42:13 -04002685struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
Jiri Pirkoe2195122014-11-19 14:05:01 +01002686int skb_ensure_writable(struct sk_buff *skb, int write_len);
Jiri Pirko93515d52014-11-19 14:05:02 +01002687int skb_vlan_pop(struct sk_buff *skb);
2688int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002689
Al Viro6ce8e9c2014-04-06 21:25:44 -04002690static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2691{
2692 return memcpy_fromiovec(data, msg->msg_iov, len);
2693}
2694
Daniel Borkmann2817a332013-10-30 11:50:51 +01002695struct skb_checksum_ops {
2696 __wsum (*update)(const void *mem, int len, __wsum wsum);
2697 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2698};
2699
2700__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2701 __wsum csum, const struct skb_checksum_ops *ops);
2702__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2703 __wsum csum);
2704
David S. Miller690e36e2014-08-23 12:13:41 -07002705static inline void *__skb_header_pointer(const struct sk_buff *skb, int offset,
2706 int len, void *data, int hlen, void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707{
Patrick McHardy55820ee2005-07-05 14:08:10 -07002708 if (hlen - offset >= len)
David S. Miller690e36e2014-08-23 12:13:41 -07002709 return data + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002710
David S. Miller690e36e2014-08-23 12:13:41 -07002711 if (!skb ||
2712 skb_copy_bits(skb, offset, buffer, len) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713 return NULL;
2714
2715 return buffer;
2716}
2717
David S. Miller690e36e2014-08-23 12:13:41 -07002718static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2719 int len, void *buffer)
2720{
2721 return __skb_header_pointer(skb, offset, len, skb->data,
2722 skb_headlen(skb), buffer);
2723}
2724
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01002725/**
2726 * skb_needs_linearize - check if we need to linearize a given skb
2727 * depending on the given device features.
2728 * @skb: socket buffer to check
2729 * @features: net device features
2730 *
2731 * Returns true if either:
2732 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2733 * 2. skb is fragmented and the device does not support SG.
2734 */
2735static inline bool skb_needs_linearize(struct sk_buff *skb,
2736 netdev_features_t features)
2737{
2738 return skb_is_nonlinear(skb) &&
2739 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2740 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2741}
2742
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002743static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2744 void *to,
2745 const unsigned int len)
2746{
2747 memcpy(to, skb->data, len);
2748}
2749
2750static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2751 const int offset, void *to,
2752 const unsigned int len)
2753{
2754 memcpy(to, skb->data + offset, len);
2755}
2756
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03002757static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2758 const void *from,
2759 const unsigned int len)
2760{
2761 memcpy(skb->data, from, len);
2762}
2763
2764static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2765 const int offset,
2766 const void *from,
2767 const unsigned int len)
2768{
2769 memcpy(skb->data + offset, from, len);
2770}
2771
Joe Perches7965bd42013-09-26 14:48:15 -07002772void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773
Patrick Ohlyac45f602009-02-12 05:03:37 +00002774static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2775{
2776 return skb->tstamp;
2777}
2778
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002779/**
2780 * skb_get_timestamp - get timestamp from a skb
2781 * @skb: skb to get stamp from
2782 * @stamp: pointer to struct timeval to store stamp in
2783 *
2784 * Timestamps are stored in the skb as offsets to a base timestamp.
2785 * This function converts the offset back to a struct timeval and stores
2786 * it in stamp.
2787 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00002788static inline void skb_get_timestamp(const struct sk_buff *skb,
2789 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002790{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002791 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002792}
2793
Patrick Ohlyac45f602009-02-12 05:03:37 +00002794static inline void skb_get_timestampns(const struct sk_buff *skb,
2795 struct timespec *stamp)
2796{
2797 *stamp = ktime_to_timespec(skb->tstamp);
2798}
2799
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002800static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002801{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002802 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002803}
2804
Stephen Hemminger164891a2007-04-23 22:26:16 -07002805static inline ktime_t net_timedelta(ktime_t t)
2806{
2807 return ktime_sub(ktime_get_real(), t);
2808}
2809
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07002810static inline ktime_t net_invalid_timestamp(void)
2811{
2812 return ktime_set(0, 0);
2813}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002814
Alexander Duyck62bccb82014-09-04 13:31:35 -04002815struct sk_buff *skb_clone_sk(struct sk_buff *skb);
2816
Richard Cochranc1f19b52010-07-17 08:49:36 +00002817#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2818
Joe Perches7965bd42013-09-26 14:48:15 -07002819void skb_clone_tx_timestamp(struct sk_buff *skb);
2820bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00002821
2822#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2823
2824static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2825{
2826}
2827
2828static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2829{
2830 return false;
2831}
2832
2833#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2834
2835/**
2836 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2837 *
Richard Cochranda92b192011-10-21 00:49:15 +00002838 * PHY drivers may accept clones of transmitted packets for
2839 * timestamping via their phy_driver.txtstamp method. These drivers
2840 * must call this function to return the skb back to the stack, with
2841 * or without a timestamp.
2842 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00002843 * @skb: clone of the the original outgoing packet
Richard Cochranda92b192011-10-21 00:49:15 +00002844 * @hwtstamps: hardware time stamps, may be NULL if not available
Richard Cochranc1f19b52010-07-17 08:49:36 +00002845 *
2846 */
2847void skb_complete_tx_timestamp(struct sk_buff *skb,
2848 struct skb_shared_hwtstamps *hwtstamps);
2849
Willem de Bruijne7fd2882014-08-04 22:11:48 -04002850void __skb_tstamp_tx(struct sk_buff *orig_skb,
2851 struct skb_shared_hwtstamps *hwtstamps,
2852 struct sock *sk, int tstype);
2853
Patrick Ohlyac45f602009-02-12 05:03:37 +00002854/**
2855 * skb_tstamp_tx - queue clone of skb with send time stamps
2856 * @orig_skb: the original outgoing packet
2857 * @hwtstamps: hardware time stamps, may be NULL if not available
2858 *
2859 * If the skb has a socket associated, then this function clones the
2860 * skb (thus sharing the actual data and optional structures), stores
2861 * the optional hardware time stamping information (if non NULL) or
2862 * generates a software time stamp (otherwise), then queues the clone
2863 * to the error queue of the socket. Errors are silently ignored.
2864 */
Joe Perches7965bd42013-09-26 14:48:15 -07002865void skb_tstamp_tx(struct sk_buff *orig_skb,
2866 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00002867
Richard Cochran4507a712010-07-17 08:48:28 +00002868static inline void sw_tx_timestamp(struct sk_buff *skb)
2869{
Oliver Hartkopp2244d072010-08-17 08:59:14 +00002870 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2871 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
Richard Cochran4507a712010-07-17 08:48:28 +00002872 skb_tstamp_tx(skb, NULL);
2873}
2874
2875/**
2876 * skb_tx_timestamp() - Driver hook for transmit timestamping
2877 *
2878 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00002879 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00002880 *
David S. Miller73409f32013-12-27 13:04:33 -05002881 * Specifically, one should make absolutely sure that this function is
2882 * called before TX completion of this packet can trigger. Otherwise
2883 * the packet could potentially already be freed.
2884 *
Richard Cochran4507a712010-07-17 08:48:28 +00002885 * @skb: A socket buffer.
2886 */
2887static inline void skb_tx_timestamp(struct sk_buff *skb)
2888{
Richard Cochranc1f19b52010-07-17 08:49:36 +00002889 skb_clone_tx_timestamp(skb);
Richard Cochran4507a712010-07-17 08:48:28 +00002890 sw_tx_timestamp(skb);
2891}
2892
Johannes Berg6e3e9392011-11-09 10:15:42 +01002893/**
2894 * skb_complete_wifi_ack - deliver skb with wifi status
2895 *
2896 * @skb: the original outgoing packet
2897 * @acked: ack status
2898 *
2899 */
2900void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2901
Joe Perches7965bd42013-09-26 14:48:15 -07002902__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2903__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08002904
Herbert Xu60476372007-04-09 11:59:39 -07002905static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2906{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07002907 return ((skb->ip_summed & CHECKSUM_UNNECESSARY) || skb->csum_valid);
Herbert Xu60476372007-04-09 11:59:39 -07002908}
2909
Herbert Xufb286bb2005-11-10 13:01:24 -08002910/**
2911 * skb_checksum_complete - Calculate checksum of an entire packet
2912 * @skb: packet to process
2913 *
2914 * This function calculates the checksum over the entire packet plus
2915 * the value of skb->csum. The latter can be used to supply the
2916 * checksum of a pseudo header as used by TCP/UDP. It returns the
2917 * checksum.
2918 *
2919 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2920 * this function can be used to verify that checksum on received
2921 * packets. In that case the function should return zero if the
2922 * checksum is correct. In particular, this function will return zero
2923 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2924 * hardware has already verified the correctness of the checksum.
2925 */
Al Viro4381ca32007-07-15 21:00:11 +01002926static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08002927{
Herbert Xu60476372007-04-09 11:59:39 -07002928 return skb_csum_unnecessary(skb) ?
2929 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08002930}
2931
Tom Herbert77cffe22014-08-27 21:26:46 -07002932static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
2933{
2934 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2935 if (skb->csum_level == 0)
2936 skb->ip_summed = CHECKSUM_NONE;
2937 else
2938 skb->csum_level--;
2939 }
2940}
2941
2942static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
2943{
2944 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
2945 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
2946 skb->csum_level++;
2947 } else if (skb->ip_summed == CHECKSUM_NONE) {
2948 skb->ip_summed = CHECKSUM_UNNECESSARY;
2949 skb->csum_level = 0;
2950 }
2951}
2952
Tom Herbert5a212322014-08-31 15:12:41 -07002953static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
2954{
2955 /* Mark current checksum as bad (typically called from GRO
2956 * path). In the case that ip_summed is CHECKSUM_NONE
2957 * this must be the first checksum encountered in the packet.
2958 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
2959 * checksum after the last one validated. For UDP, a zero
2960 * checksum can not be marked as bad.
2961 */
2962
2963 if (skb->ip_summed == CHECKSUM_NONE ||
2964 skb->ip_summed == CHECKSUM_UNNECESSARY)
2965 skb->csum_bad = 1;
2966}
2967
Tom Herbert76ba0aa2014-05-02 16:29:18 -07002968/* Check if we need to perform checksum complete validation.
2969 *
2970 * Returns true if checksum complete is needed, false otherwise
2971 * (either checksum is unnecessary or zero checksum is allowed).
2972 */
2973static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
2974 bool zero_okay,
2975 __sum16 check)
2976{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07002977 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
2978 skb->csum_valid = 1;
Tom Herbert77cffe22014-08-27 21:26:46 -07002979 __skb_decr_checksum_unnecessary(skb);
Tom Herbert76ba0aa2014-05-02 16:29:18 -07002980 return false;
2981 }
2982
2983 return true;
2984}
2985
2986/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
2987 * in checksum_init.
2988 */
2989#define CHECKSUM_BREAK 76
2990
2991/* Validate (init) checksum based on checksum complete.
2992 *
2993 * Return values:
2994 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
2995 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
2996 * checksum is stored in skb->csum for use in __skb_checksum_complete
2997 * non-zero: value of invalid checksum
2998 *
2999 */
3000static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3001 bool complete,
3002 __wsum psum)
3003{
3004 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3005 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003006 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003007 return 0;
3008 }
Tom Herbert5a212322014-08-31 15:12:41 -07003009 } else if (skb->csum_bad) {
3010 /* ip_summed == CHECKSUM_NONE in this case */
3011 return 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003012 }
3013
3014 skb->csum = psum;
3015
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003016 if (complete || skb->len <= CHECKSUM_BREAK) {
3017 __sum16 csum;
3018
3019 csum = __skb_checksum_complete(skb);
3020 skb->csum_valid = !csum;
3021 return csum;
3022 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003023
3024 return 0;
3025}
3026
3027static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3028{
3029 return 0;
3030}
3031
3032/* Perform checksum validate (init). Note that this is a macro since we only
3033 * want to calculate the pseudo header which is an input function if necessary.
3034 * First we try to validate without any computation (checksum unnecessary) and
3035 * then calculate based on checksum complete calling the function to compute
3036 * pseudo header.
3037 *
3038 * Return values:
3039 * 0: checksum is validated or try to in skb_checksum_complete
3040 * non-zero: value of invalid checksum
3041 */
3042#define __skb_checksum_validate(skb, proto, complete, \
3043 zero_okay, check, compute_pseudo) \
3044({ \
3045 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003046 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003047 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3048 __ret = __skb_checksum_validate_complete(skb, \
3049 complete, compute_pseudo(skb, proto)); \
3050 __ret; \
3051})
3052
3053#define skb_checksum_init(skb, proto, compute_pseudo) \
3054 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3055
3056#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3057 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3058
3059#define skb_checksum_validate(skb, proto, compute_pseudo) \
3060 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3061
3062#define skb_checksum_validate_zero_check(skb, proto, check, \
3063 compute_pseudo) \
3064 __skb_checksum_validate_(skb, proto, true, true, check, compute_pseudo)
3065
3066#define skb_checksum_simple_validate(skb) \
3067 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3068
Tom Herbertd96535a2014-08-31 15:12:42 -07003069static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3070{
3071 return (skb->ip_summed == CHECKSUM_NONE &&
3072 skb->csum_valid && !skb->csum_bad);
3073}
3074
3075static inline void __skb_checksum_convert(struct sk_buff *skb,
3076 __sum16 check, __wsum pseudo)
3077{
3078 skb->csum = ~pseudo;
3079 skb->ip_summed = CHECKSUM_COMPLETE;
3080}
3081
3082#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3083do { \
3084 if (__skb_checksum_convert_check(skb)) \
3085 __skb_checksum_convert(skb, check, \
3086 compute_pseudo(skb, proto)); \
3087} while (0)
3088
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003089#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07003090void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003091static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3092{
3093 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07003094 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003095}
3096static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3097{
3098 if (nfct)
3099 atomic_inc(&nfct->use);
3100}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003101#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003102#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003103static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3104{
3105 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3106 kfree(nf_bridge);
3107}
3108static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3109{
3110 if (nf_bridge)
3111 atomic_inc(&nf_bridge->use);
3112}
3113#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08003114static inline void nf_reset(struct sk_buff *skb)
3115{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003116#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003117 nf_conntrack_put(skb->nfct);
3118 skb->nfct = NULL;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003119#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003120#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003121 nf_bridge_put(skb->nf_bridge);
3122 skb->nf_bridge = NULL;
3123#endif
3124}
3125
Patrick McHardy124dff02013-04-05 20:42:05 +02003126static inline void nf_reset_trace(struct sk_buff *skb)
3127{
Florian Westphal478b3602014-02-15 23:48:45 +01003128#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00003129 skb->nf_trace = 0;
3130#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003131}
3132
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003133/* Note: This doesn't put any conntrack and bridge info in dst. */
Eric Dumazetb1937222014-09-28 22:18:47 -07003134static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3135 bool copy)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003136{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003137#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003138 dst->nfct = src->nfct;
3139 nf_conntrack_get(src->nfct);
Eric Dumazetb1937222014-09-28 22:18:47 -07003140 if (copy)
3141 dst->nfctinfo = src->nfctinfo;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003142#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003143#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003144 dst->nf_bridge = src->nf_bridge;
3145 nf_bridge_get(src->nf_bridge);
3146#endif
Florian Westphal478b3602014-02-15 23:48:45 +01003147#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Eric Dumazetb1937222014-09-28 22:18:47 -07003148 if (copy)
3149 dst->nf_trace = src->nf_trace;
Florian Westphal478b3602014-02-15 23:48:45 +01003150#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003151}
3152
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003153static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3154{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003155#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003156 nf_conntrack_put(dst->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003157#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003158#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003159 nf_bridge_put(dst->nf_bridge);
3160#endif
Eric Dumazetb1937222014-09-28 22:18:47 -07003161 __nf_copy(dst, src, true);
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003162}
3163
James Morris984bc162006-06-09 00:29:17 -07003164#ifdef CONFIG_NETWORK_SECMARK
3165static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3166{
3167 to->secmark = from->secmark;
3168}
3169
3170static inline void skb_init_secmark(struct sk_buff *skb)
3171{
3172 skb->secmark = 0;
3173}
3174#else
3175static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3176{ }
3177
3178static inline void skb_init_secmark(struct sk_buff *skb)
3179{ }
3180#endif
3181
Eric W. Biederman574f7192014-04-01 12:20:24 -07003182static inline bool skb_irq_freeable(const struct sk_buff *skb)
3183{
3184 return !skb->destructor &&
3185#if IS_ENABLED(CONFIG_XFRM)
3186 !skb->sp &&
3187#endif
3188#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3189 !skb->nfct &&
3190#endif
3191 !skb->_skb_refdst &&
3192 !skb_has_frag_list(skb);
3193}
3194
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003195static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3196{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003197 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003198}
3199
Stephen Hemminger92477442009-03-21 13:39:26 -07003200static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003201{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003202 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003203}
3204
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003205static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3206{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003207 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003208}
3209
David S. Millerd5a9e242009-01-27 16:22:11 -08003210static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3211{
3212 skb->queue_mapping = rx_queue + 1;
3213}
3214
Stephen Hemminger92477442009-03-21 13:39:26 -07003215static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003216{
3217 return skb->queue_mapping - 1;
3218}
3219
Stephen Hemminger92477442009-03-21 13:39:26 -07003220static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003221{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003222 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003223}
3224
Tom Herbert0e001612014-07-01 21:32:27 -07003225u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
Joe Perches7965bd42013-09-26 14:48:15 -07003226 unsigned int num_tx_queues);
Stephen Hemminger92477442009-03-21 13:39:26 -07003227
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003228static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3229{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003230#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003231 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003232#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003233 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003234#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003235}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003236
Pravin B Shelar68c33162013-02-14 14:02:41 +00003237/* Keeps track of mac header offset relative to skb->head.
3238 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3239 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003240 * tunnel skb it points to outer mac header.
3241 * Keeps track of level of encapsulation of network headers.
3242 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003243struct skb_gso_cb {
Eric Dumazet3347c962013-10-19 11:42:56 -07003244 int mac_offset;
3245 int encap_level;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003246 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003247};
3248#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3249
3250static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3251{
3252 return (skb_mac_header(inner_skb) - inner_skb->head) -
3253 SKB_GSO_CB(inner_skb)->mac_offset;
3254}
3255
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003256static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3257{
3258 int new_headroom, headroom;
3259 int ret;
3260
3261 headroom = skb_headroom(skb);
3262 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3263 if (ret)
3264 return ret;
3265
3266 new_headroom = skb_headroom(skb);
3267 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3268 return 0;
3269}
3270
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003271/* Compute the checksum for a gso segment. First compute the checksum value
3272 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3273 * then add in skb->csum (checksum from csum_start to end of packet).
3274 * skb->csum and csum_start are then updated to reflect the checksum of the
3275 * resultant packet starting from the transport header-- the resultant checksum
3276 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3277 * header.
3278 */
3279static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3280{
3281 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3282 skb_transport_offset(skb);
3283 __u16 csum;
3284
3285 csum = csum_fold(csum_partial(skb_transport_header(skb),
3286 plen, skb->csum));
3287 skb->csum = res;
3288 SKB_GSO_CB(skb)->csum_start -= plen;
3289
3290 return csum;
3291}
3292
David S. Millerbdcc0922012-03-07 20:53:36 -05003293static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07003294{
3295 return skb_shinfo(skb)->gso_size;
3296}
3297
Eric Dumazet36a8f392013-09-29 01:21:32 -07003298/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05003299static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02003300{
3301 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3302}
3303
Joe Perches7965bd42013-09-26 14:48:15 -07003304void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07003305
3306static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3307{
3308 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3309 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003310 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3311
Alexander Duyckb78462e2010-06-02 12:24:37 +00003312 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3313 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07003314 __skb_warn_lro_forwarding(skb);
3315 return true;
3316 }
3317 return false;
3318}
3319
Herbert Xu35fc92a2007-03-26 23:22:20 -07003320static inline void skb_forward_csum(struct sk_buff *skb)
3321{
3322 /* Unfortunately we don't support this one. Any brave souls? */
3323 if (skb->ip_summed == CHECKSUM_COMPLETE)
3324 skb->ip_summed = CHECKSUM_NONE;
3325}
3326
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003327/**
3328 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3329 * @skb: skb to check
3330 *
3331 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3332 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3333 * use this helper, to document places where we make this assertion.
3334 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003335static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003336{
3337#ifdef DEBUG
3338 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3339#endif
3340}
3341
Rusty Russellf35d9d82008-02-04 23:49:54 -05003342bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00003343
Paul Durranted1f50c2014-01-09 10:02:46 +00003344int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3345
Alexander Duyck56193d12014-09-05 19:20:26 -04003346u32 skb_get_poff(const struct sk_buff *skb);
3347u32 __skb_get_poff(const struct sk_buff *skb, void *data,
3348 const struct flow_keys *keys, int hlen);
Daniel Borkmannf77668d2013-03-19 06:39:30 +00003349
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00003350/**
3351 * skb_head_is_locked - Determine if the skb->head is locked down
3352 * @skb: skb to check
3353 *
3354 * The head on skbs build around a head frag can be removed if they are
3355 * not cloned. This function returns true if the skb head is locked down
3356 * due to either being allocated via kmalloc, or by being a clone with
3357 * multiple references to the head.
3358 */
3359static inline bool skb_head_is_locked(const struct sk_buff *skb)
3360{
3361 return !skb->head_frag || skb_cloned(skb);
3362}
Florian Westphalfe6cc552014-02-13 23:09:12 +01003363
3364/**
3365 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3366 *
3367 * @skb: GSO skb
3368 *
3369 * skb_gso_network_seglen is used to determine the real size of the
3370 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3371 *
3372 * The MAC/L2 header is not accounted for.
3373 */
3374static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3375{
3376 unsigned int hdr_len = skb_transport_header(skb) -
3377 skb_network_header(skb);
3378 return hdr_len + skb_gso_transport_seglen(skb);
3379}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003380#endif /* __KERNEL__ */
3381#endif /* _LINUX_SKBUFF_H */