blob: 477f0f60db45a84295a394a9c7b714f7f19eeb03 [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>
23
Arun Sharma600634972011-07-26 16:09:06 -070024#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025#include <asm/types.h>
26#include <linux/spinlock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <linux/net.h>
Thomas Graf3fc7e8a2005-06-23 21:00:17 -070028#include <linux/textsearch.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <net/checksum.h>
Al Viroa80958f2006-12-04 20:41:19 +000030#include <linux/rcupdate.h>
Chris Leech97fc2f02006-05-23 17:55:33 -070031#include <linux/dmaengine.h>
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -070032#include <linux/hrtimer.h>
Ian Campbell131ea662011-08-19 06:25:00 +000033#include <linux/dma-mapping.h>
Michał Mirosławc8f44af2011-11-15 15:29:55 +000034#include <linux/netdev_features.h>
Eric Dumazet363ec39232014-02-26 14:02:11 -080035#include <linux/sched.h>
Jason Wang5203cd22013-03-26 23:11:21 +000036#include <net/flow_keys.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070037
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010038/* A. Checksumming of received packets by device.
39 *
40 * CHECKSUM_NONE:
41 *
42 * Device failed to checksum this packet e.g. due to lack of capabilities.
43 * The packet contains full (though not verified) checksum in packet but
44 * not in skb->csum. Thus, skb->csum is undefined in this case.
45 *
46 * CHECKSUM_UNNECESSARY:
47 *
48 * The hardware you're dealing with doesn't calculate the full checksum
49 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
50 * for specific protocols e.g. TCP/UDP/SCTP, then, for such packets it will
51 * set CHECKSUM_UNNECESSARY if their checksums are okay. skb->csum is still
52 * undefined in this case though. It is a bad option, but, unfortunately,
53 * nowadays most vendors do this. Apparently with the secret goal to sell
54 * you new devices, when you will add new protocol to your host, f.e. IPv6 8)
55 *
56 * CHECKSUM_COMPLETE:
57 *
58 * This is the most generic way. The device supplied checksum of the _whole_
59 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
60 * hardware doesn't need to parse L3/L4 headers to implement this.
61 *
62 * Note: Even if device supports only some protocols, but is able to produce
63 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
64 *
65 * CHECKSUM_PARTIAL:
66 *
67 * This is identical to the case for output below. This may occur on a packet
68 * received directly from another Linux OS, e.g., a virtualized Linux kernel
69 * on the same host. The packet can be treated in the same way as
70 * CHECKSUM_UNNECESSARY, except that on output (i.e., forwarding) the
71 * checksum must be filled in by the OS or the hardware.
72 *
73 * B. Checksumming on output.
74 *
75 * CHECKSUM_NONE:
76 *
77 * The skb was already checksummed by the protocol, or a checksum is not
78 * required.
79 *
80 * CHECKSUM_PARTIAL:
81 *
82 * The device is required to checksum the packet as seen by hard_start_xmit()
83 * from skb->csum_start up to the end, and to record/write the checksum at
84 * offset skb->csum_start + skb->csum_offset.
85 *
86 * The device must show its capabilities in dev->features, set up at device
87 * setup time, e.g. netdev_features.h:
88 *
89 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
90 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
91 * IPv4. Sigh. Vendors like this way for an unknown reason.
92 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
93 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
94 * NETIF_F_... - Well, you get the picture.
95 *
96 * CHECKSUM_UNNECESSARY:
97 *
98 * Normally, the device will do per protocol specific checksumming. Protocol
99 * implementations that do not want the NIC to perform the checksum
100 * calculation should use this flag in their outgoing skbs.
101 *
102 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
103 * offload. Correspondingly, the FCoE protocol driver
104 * stack should use CHECKSUM_UNNECESSARY.
105 *
106 * Any questions? No questions, good. --ANK
107 */
108
Herbert Xu60476372007-04-09 11:59:39 -0700109/* Don't change this without changing skb_csum_unnecessary! */
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100110#define CHECKSUM_NONE 0
111#define CHECKSUM_UNNECESSARY 1
112#define CHECKSUM_COMPLETE 2
113#define CHECKSUM_PARTIAL 3
Linus Torvalds1da177e2005-04-16 15:20:36 -0700114
Tobias Klauser0bec8c82014-07-22 12:06:23 +0200115#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
David S. Millerfc910a22007-03-25 20:27:59 -0700116#define SKB_WITH_OVERHEAD(X) \
Herbert Xudeea84b2007-10-21 16:27:46 -0700117 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
David S. Millerfc910a22007-03-25 20:27:59 -0700118#define SKB_MAX_ORDER(X, ORDER) \
119 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700120#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
121#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
122
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000123/* return minimum truesize of one skb containing X bytes of data */
124#define SKB_TRUESIZE(X) ((X) + \
125 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
126 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
127
Linus Torvalds1da177e2005-04-16 15:20:36 -0700128struct net_device;
David Howells716ea3a2007-04-02 20:19:53 -0700129struct scatterlist;
Jens Axboe9c55e012007-11-06 23:30:13 -0800130struct pipe_inode_info;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700131
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700132#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700133struct nf_conntrack {
134 atomic_t use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700135};
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700136#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700137
138#ifdef CONFIG_BRIDGE_NETFILTER
139struct nf_bridge_info {
Eric Dumazetbf1ac5c2012-04-18 23:19:25 +0000140 atomic_t use;
141 unsigned int mask;
142 struct net_device *physindev;
143 struct net_device *physoutdev;
144 unsigned long data[32 / sizeof(unsigned long)];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700145};
146#endif
147
Linus Torvalds1da177e2005-04-16 15:20:36 -0700148struct sk_buff_head {
149 /* These two members must be first. */
150 struct sk_buff *next;
151 struct sk_buff *prev;
152
153 __u32 qlen;
154 spinlock_t lock;
155};
156
157struct sk_buff;
158
Ian Campbell9d4dde52011-12-22 23:39:14 +0000159/* To allow 64K frame to be packed as single skb without frag_list we
160 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
161 * buffers which do not start on a page boundary.
162 *
163 * Since GRO uses frags we allocate at least 16 regardless of page
164 * size.
Anton Blancharda715dea2011-03-27 14:57:26 +0000165 */
Ian Campbell9d4dde52011-12-22 23:39:14 +0000166#if (65536/PAGE_SIZE + 1) < 16
David S. Millereec00952011-03-29 23:34:08 -0700167#define MAX_SKB_FRAGS 16UL
Anton Blancharda715dea2011-03-27 14:57:26 +0000168#else
Ian Campbell9d4dde52011-12-22 23:39:14 +0000169#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
Anton Blancharda715dea2011-03-27 14:57:26 +0000170#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700171
172typedef struct skb_frag_struct skb_frag_t;
173
174struct skb_frag_struct {
Ian Campbella8605c62011-10-19 23:01:49 +0000175 struct {
176 struct page *p;
177 } page;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000178#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
David S. Millera309bb02007-07-30 18:47:03 -0700179 __u32 page_offset;
180 __u32 size;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000181#else
182 __u16 page_offset;
183 __u16 size;
184#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700185};
186
Eric Dumazet9e903e02011-10-18 21:00:24 +0000187static inline unsigned int skb_frag_size(const skb_frag_t *frag)
188{
189 return frag->size;
190}
191
192static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
193{
194 frag->size = size;
195}
196
197static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
198{
199 frag->size += delta;
200}
201
202static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
203{
204 frag->size -= delta;
205}
206
Patrick Ohlyac45f602009-02-12 05:03:37 +0000207#define HAVE_HW_TIME_STAMP
208
209/**
Randy Dunlapd3a21be2009-03-02 03:15:58 -0800210 * struct skb_shared_hwtstamps - hardware time stamps
Patrick Ohlyac45f602009-02-12 05:03:37 +0000211 * @hwtstamp: hardware time stamp transformed into duration
212 * since arbitrary point in time
Patrick Ohlyac45f602009-02-12 05:03:37 +0000213 *
214 * Software time stamps generated by ktime_get_real() are stored in
Willem de Bruijn4d276eb2014-07-25 18:01:32 -0400215 * skb->tstamp.
Patrick Ohlyac45f602009-02-12 05:03:37 +0000216 *
217 * hwtstamps can only be compared against other hwtstamps from
218 * the same device.
219 *
220 * This structure is attached to packets as part of the
221 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
222 */
223struct skb_shared_hwtstamps {
224 ktime_t hwtstamp;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000225};
226
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000227/* Definitions for tx_flags in struct skb_shared_info */
228enum {
229 /* generate hardware time stamp */
230 SKBTX_HW_TSTAMP = 1 << 0,
231
232 /* generate software time stamp */
233 SKBTX_SW_TSTAMP = 1 << 1,
234
235 /* device driver is going to provide hardware time stamp */
236 SKBTX_IN_PROGRESS = 1 << 2,
237
Shirley Maa6686f22011-07-06 12:22:12 +0000238 /* device driver supports TX zero-copy buffers */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000239 SKBTX_DEV_ZEROCOPY = 1 << 3,
Johannes Berg6e3e9392011-11-09 10:15:42 +0100240
241 /* generate wifi status information (where possible) */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000242 SKBTX_WIFI_STATUS = 1 << 4,
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000243
244 /* This indicates at least one fragment might be overwritten
245 * (as in vmsplice(), sendfile() ...)
246 * If we need to compute a TX checksum, we'll need to copy
247 * all frags to avoid possible bad checksum
248 */
249 SKBTX_SHARED_FRAG = 1 << 5,
Shirley Maa6686f22011-07-06 12:22:12 +0000250};
251
Willem de Bruijnf24b9be2014-08-04 22:11:45 -0400252#define SKBTX_ANY_SW_TSTAMP SKBTX_SW_TSTAMP
253#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
254
Shirley Maa6686f22011-07-06 12:22:12 +0000255/*
256 * The callback notifies userspace to release buffers when skb DMA is done in
257 * lower device, the skb last reference should be 0 when calling this.
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000258 * The zerocopy_success argument is true if zero copy transmit occurred,
259 * false on data copy or out of memory error caused by data copy attempt.
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000260 * The ctx field is used to track device context.
261 * The desc field is used to track userspace buffer index.
Shirley Maa6686f22011-07-06 12:22:12 +0000262 */
263struct ubuf_info {
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000264 void (*callback)(struct ubuf_info *, bool zerocopy_success);
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000265 void *ctx;
Shirley Maa6686f22011-07-06 12:22:12 +0000266 unsigned long desc;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000267};
268
Linus Torvalds1da177e2005-04-16 15:20:36 -0700269/* This data is invariant across clones and lives at
270 * the end of the header data, ie. at skb->end.
271 */
272struct skb_shared_info {
Ian Campbell9f42f122012-01-05 07:13:39 +0000273 unsigned char nr_frags;
274 __u8 tx_flags;
Herbert Xu79671682006-06-22 02:40:14 -0700275 unsigned short gso_size;
276 /* Warning: this field is not always filled in (UFO)! */
277 unsigned short gso_segs;
278 unsigned short gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279 struct sk_buff *frag_list;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000280 struct skb_shared_hwtstamps hwtstamps;
Ian Campbell9f42f122012-01-05 07:13:39 +0000281 __be32 ip6_frag_id;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700282
283 /*
284 * Warning : all fields before dataref are cleared in __alloc_skb()
285 */
286 atomic_t dataref;
287
Johann Baudy69e3c752009-05-18 22:11:22 -0700288 /* Intermediate layers must ensure that destructor_arg
289 * remains valid until skb destructor */
290 void * destructor_arg;
Shirley Maa6686f22011-07-06 12:22:12 +0000291
Eric Dumazetfed66382010-07-22 19:09:08 +0000292 /* must be last field, see pskb_expand_head() */
293 skb_frag_t frags[MAX_SKB_FRAGS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700294};
295
296/* We divide dataref into two halves. The higher 16 bits hold references
297 * to the payload part of skb->data. The lower 16 bits hold references to
Patrick McHardy334a8132007-06-25 04:35:20 -0700298 * the entire skb->data. A clone of a headerless skb holds the length of
299 * the header in skb->hdr_len.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700300 *
301 * All users must obey the rule that the skb->data reference count must be
302 * greater than or equal to the payload reference count.
303 *
304 * Holding a reference to the payload part means that the user does not
305 * care about modifications to the header part of skb->data.
306 */
307#define SKB_DATAREF_SHIFT 16
308#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
309
David S. Millerd179cd12005-08-17 14:57:30 -0700310
311enum {
312 SKB_FCLONE_UNAVAILABLE,
313 SKB_FCLONE_ORIG,
314 SKB_FCLONE_CLONE,
315};
316
Herbert Xu79671682006-06-22 02:40:14 -0700317enum {
318 SKB_GSO_TCPV4 = 1 << 0,
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700319 SKB_GSO_UDP = 1 << 1,
Herbert Xu576a30e2006-06-27 13:22:38 -0700320
321 /* This indicates the skb is from an untrusted source. */
322 SKB_GSO_DODGY = 1 << 2,
Michael Chanb0da85372006-06-29 12:30:00 -0700323
324 /* This indicates the tcp segment has CWR set. */
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700325 SKB_GSO_TCP_ECN = 1 << 3,
326
327 SKB_GSO_TCPV6 = 1 << 4,
Chris Leech01d5b2f2009-02-27 14:06:49 -0800328
329 SKB_GSO_FCOE = 1 << 5,
Pravin B Shelar68c33162013-02-14 14:02:41 +0000330
331 SKB_GSO_GRE = 1 << 6,
Pravin B Shelar73136262013-03-07 13:21:51 +0000332
Tom Herbert4b282522014-06-14 23:23:52 -0700333 SKB_GSO_GRE_CSUM = 1 << 7,
Simon Horman0d89d202013-05-23 21:02:52 +0000334
Tom Herbert4b282522014-06-14 23:23:52 -0700335 SKB_GSO_IPIP = 1 << 8,
Eric Dumazetcb32f512013-10-19 11:42:57 -0700336
Tom Herbert4b282522014-06-14 23:23:52 -0700337 SKB_GSO_SIT = 1 << 9,
Eric Dumazet61c1db72013-10-20 20:47:30 -0700338
Tom Herbert4b282522014-06-14 23:23:52 -0700339 SKB_GSO_UDP_TUNNEL = 1 << 10,
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700340
341 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
Tom Herbert4749c092014-06-04 17:20:23 -0700342
Tom Herbert4b282522014-06-14 23:23:52 -0700343 SKB_GSO_MPLS = 1 << 12,
344
Herbert Xu79671682006-06-22 02:40:14 -0700345};
346
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700347#if BITS_PER_LONG > 32
348#define NET_SKBUFF_DATA_USES_OFFSET 1
349#endif
350
351#ifdef NET_SKBUFF_DATA_USES_OFFSET
352typedef unsigned int sk_buff_data_t;
353#else
354typedef unsigned char *sk_buff_data_t;
355#endif
356
Eric Dumazet363ec39232014-02-26 14:02:11 -0800357/**
358 * struct skb_mstamp - multi resolution time stamps
359 * @stamp_us: timestamp in us resolution
360 * @stamp_jiffies: timestamp in jiffies
361 */
362struct skb_mstamp {
363 union {
364 u64 v64;
365 struct {
366 u32 stamp_us;
367 u32 stamp_jiffies;
368 };
369 };
370};
371
372/**
373 * skb_mstamp_get - get current timestamp
374 * @cl: place to store timestamps
375 */
376static inline void skb_mstamp_get(struct skb_mstamp *cl)
377{
378 u64 val = local_clock();
379
380 do_div(val, NSEC_PER_USEC);
381 cl->stamp_us = (u32)val;
382 cl->stamp_jiffies = (u32)jiffies;
383}
384
385/**
386 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
387 * @t1: pointer to newest sample
388 * @t0: pointer to oldest sample
389 */
390static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
391 const struct skb_mstamp *t0)
392{
393 s32 delta_us = t1->stamp_us - t0->stamp_us;
394 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
395
396 /* If delta_us is negative, this might be because interval is too big,
397 * or local_clock() drift is too big : fallback using jiffies.
398 */
399 if (delta_us <= 0 ||
400 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
401
402 delta_us = jiffies_to_usecs(delta_jiffies);
403
404 return delta_us;
405}
406
407
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408/**
409 * struct sk_buff - socket buffer
410 * @next: Next buffer in list
411 * @prev: Previous buffer in list
Eric Dumazet363ec39232014-02-26 14:02:11 -0800412 * @tstamp: Time we arrived/left
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700413 * @sk: Socket we are owned by
Linus Torvalds1da177e2005-04-16 15:20:36 -0700414 * @dev: Device we arrived on/are leaving by
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700415 * @cb: Control buffer. Free for use by every layer. Put private vars here
Eric Dumazet7fee2262010-05-11 23:19:48 +0000416 * @_skb_refdst: destination entry (with norefcount bit)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700417 * @sp: the security path, used for xfrm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 * @len: Length of actual data
419 * @data_len: Data length
420 * @mac_len: Length of link layer header
Patrick McHardy334a8132007-06-25 04:35:20 -0700421 * @hdr_len: writable header length of cloned skb
Herbert Xu663ead32007-04-09 11:59:07 -0700422 * @csum: Checksum (must include start/offset pair)
423 * @csum_start: Offset from skb->head where checksumming should start
424 * @csum_offset: Offset from csum_start where checksum should be stored
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700425 * @priority: Packet queueing priority
WANG Cong60ff7462014-05-04 16:39:18 -0700426 * @ignore_df: allow local fragmentation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700427 * @cloned: Head may be cloned (check refcnt to be sure)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700428 * @ip_summed: Driver fed us an IP checksum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700429 * @nohdr: Payload reference only, must not modify header
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700430 * @nfctinfo: Relationship of this skb to the connection
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431 * @pkt_type: Packet class
Randy Dunlapc83c2482005-10-18 22:07:41 -0700432 * @fclone: skbuff clone status
Randy Dunlapc83c2482005-10-18 22:07:41 -0700433 * @ipvs_property: skbuff is owned by ipvs
Randy Dunlap31729362008-02-18 20:52:13 -0800434 * @peeked: this packet has been seen already, so stats have been
435 * done for it, don't do them again
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700436 * @nf_trace: netfilter packet trace flag
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700437 * @protocol: Packet protocol from driver
438 * @destructor: Destruct function
439 * @nfct: Associated connection, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
Eric Dumazet8964be42009-11-20 15:35:04 -0800441 * @skb_iif: ifindex of device we arrived on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 * @tc_index: Traffic control index
443 * @tc_verd: traffic control verdict
Tom Herbert61b905d2014-03-24 15:34:47 -0700444 * @hash: the packet hash
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700445 * @queue_mapping: Queue mapping for multiqueue devices
Randy Dunlap553a5672008-04-20 10:51:01 -0700446 * @ndisc_nodetype: router type (from link layer)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700447 * @ooo_okay: allow the mapping of a socket to a queue to be changed
Tom Herbert61b905d2014-03-24 15:34:47 -0700448 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
Changli Gao4ca24622011-08-19 07:26:44 -0700449 * ports.
Tom Herberta3b18dd2014-07-01 21:33:17 -0700450 * @sw_hash: indicates hash was computed in software stack
Johannes Berg6e3e9392011-11-09 10:15:42 +0100451 * @wifi_acked_valid: wifi_acked was set
452 * @wifi_acked: whether frame was acked on wifi or not
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000453 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
Randy Dunlapf4b8ea72006-06-22 16:00:11 -0700454 * @dma_cookie: a cookie to one of several possible DMA operations
455 * done by skb DMA functions
Eliezer Tamir06021292013-06-10 11:39:50 +0300456 * @napi_id: id of the NAPI struct this skb came from
James Morris984bc162006-06-09 00:29:17 -0700457 * @secmark: security marking
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700458 * @mark: Generic packet mark
459 * @dropcount: total number of sk_receive_queue overflows
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000460 * @vlan_proto: vlan encapsulation protocol
Patrick McHardy6aa895b02008-07-14 22:49:06 -0700461 * @vlan_tci: vlan tag control information
Simon Horman0d89d202013-05-23 21:02:52 +0000462 * @inner_protocol: Protocol (encapsulation)
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000463 * @inner_transport_header: Inner transport layer header (encapsulation)
464 * @inner_network_header: Network layer header (encapsulation)
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000465 * @inner_mac_header: Link layer header (encapsulation)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700466 * @transport_header: Transport layer header
467 * @network_header: Network layer header
468 * @mac_header: Link layer header
469 * @tail: Tail pointer
470 * @end: End pointer
471 * @head: Head of buffer
472 * @data: Data head pointer
473 * @truesize: Buffer size
474 * @users: User count - see {datagram,tcp}.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700475 */
476
477struct sk_buff {
478 /* These two members must be first. */
479 struct sk_buff *next;
480 struct sk_buff *prev;
481
Eric Dumazet363ec39232014-02-26 14:02:11 -0800482 union {
483 ktime_t tstamp;
484 struct skb_mstamp skb_mstamp;
485 };
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000486
487 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489
Linus Torvalds1da177e2005-04-16 15:20:36 -0700490 /*
491 * This is the control buffer. It is free to use for every
492 * layer. Please put your private variables there. If you
493 * want to keep them across layers you have to do a skb_clone()
494 * first. This is owned by whoever has the skb queued ATM.
495 */
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000496 char cb[48] __aligned(8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700497
Eric Dumazet7fee2262010-05-11 23:19:48 +0000498 unsigned long _skb_refdst;
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000499#ifdef CONFIG_XFRM
500 struct sec_path *sp;
501#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 unsigned int len,
Patrick McHardy334a8132007-06-25 04:35:20 -0700503 data_len;
504 __u16 mac_len,
505 hdr_len;
Al Viroff1dcad2006-11-20 18:07:29 -0800506 union {
507 __wsum csum;
Herbert Xu663ead32007-04-09 11:59:07 -0700508 struct {
509 __u16 csum_start;
510 __u16 csum_offset;
511 };
Al Viroff1dcad2006-11-20 18:07:29 -0800512 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700513 __u32 priority;
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200514 kmemcheck_bitfield_begin(flags1);
WANG Cong60ff7462014-05-04 16:39:18 -0700515 __u8 ignore_df:1,
Thomas Graf1cbb3382005-07-05 14:13:41 -0700516 cloned:1,
517 ip_summed:2,
Harald Welte6869c4d2005-08-09 19:24:19 -0700518 nohdr:1,
519 nfctinfo:3;
David S. Millerd179cd12005-08-17 14:57:30 -0700520 __u8 pkt_type:3,
Patrick McHardyb84f4cc2005-11-20 21:19:21 -0800521 fclone:2,
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700522 ipvs_property:1,
Herbert Xua59322b2007-12-05 01:53:40 -0800523 peeked:1,
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700524 nf_trace:1;
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200525 kmemcheck_bitfield_end(flags1);
Eric Dumazet4ab408d2010-03-01 03:09:26 +0000526 __be16 protocol;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700527
528 void (*destructor)(struct sk_buff *skb);
Yasuyuki Kozakai9fb9cbb2005-11-09 16:38:16 -0800529#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700530 struct nf_conntrack *nfct;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100531#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700532#ifdef CONFIG_BRIDGE_NETFILTER
533 struct nf_bridge_info *nf_bridge;
534#endif
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -0700535
Eric Dumazet8964be42009-11-20 15:35:04 -0800536 int skb_iif;
Alexander Duyck4031ae62012-01-27 06:22:53 +0000537
Tom Herbert61b905d2014-03-24 15:34:47 -0700538 __u32 hash;
Alexander Duyck4031ae62012-01-27 06:22:53 +0000539
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000540 __be16 vlan_proto;
Alexander Duyck4031ae62012-01-27 06:22:53 +0000541 __u16 vlan_tci;
542
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543#ifdef CONFIG_NET_SCHED
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700544 __u16 tc_index; /* traffic control index */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700545#ifdef CONFIG_NET_CLS_ACT
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700546 __u16 tc_verd; /* traffic control verdict */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700547#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700548#endif
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200549
Eric Dumazet0a148422011-04-20 09:27:32 +0000550 __u16 queue_mapping;
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200551 kmemcheck_bitfield_begin(flags2);
YOSHIFUJI Hideakide357cc2008-03-15 23:59:18 -0400552#ifdef CONFIG_IPV6_NDISC_NODETYPE
Jiri Pirko8a4eb572011-03-12 03:14:39 +0000553 __u8 ndisc_nodetype:2;
YOSHIFUJI Hideakide357cc2008-03-15 23:59:18 -0400554#endif
Mel Gormanc93bdd02012-07-31 16:44:19 -0700555 __u8 pfmemalloc:1;
Tom Herbert3853b582010-11-21 13:17:29 +0000556 __u8 ooo_okay:1;
Tom Herbert61b905d2014-03-24 15:34:47 -0700557 __u8 l4_hash:1;
Tom Herberta3b18dd2014-07-01 21:33:17 -0700558 __u8 sw_hash:1;
Johannes Berg6e3e9392011-11-09 10:15:42 +0100559 __u8 wifi_acked_valid:1;
560 __u8 wifi_acked:1;
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000561 __u8 no_fcs:1;
Eric Dumazetd3836f22012-04-27 00:33:38 +0000562 __u8 head_frag:1;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000563 /* Encapsulation protocol and NIC drivers should use
564 * this flag to indicate to each other if the skb contains
565 * encapsulated packet or not and maybe use the inner packet
566 * headers if needed
567 */
568 __u8 encapsulation:1;
Tom Herbert7e2b10c2014-06-04 17:20:02 -0700569 __u8 encap_hdr_csum:1;
Tom Herbert5d0c2b92014-06-10 18:54:13 -0700570 __u8 csum_valid:1;
Tom Herbert7e3cead2014-06-10 18:54:19 -0700571 __u8 csum_complete_sw:1;
Tom Herberta3b18dd2014-07-01 21:33:17 -0700572 /* 2/4 bit hole (depending on ndisc_nodetype presence) */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200573 kmemcheck_bitfield_end(flags2);
574
Cong Wange0d10952013-08-01 11:10:25 +0800575#if defined CONFIG_NET_DMA || defined CONFIG_NET_RX_BUSY_POLL
Eliezer Tamir06021292013-06-10 11:39:50 +0300576 union {
577 unsigned int napi_id;
578 dma_cookie_t dma_cookie;
579 };
Chris Leech97fc2f02006-05-23 17:55:33 -0700580#endif
James Morris984bc162006-06-09 00:29:17 -0700581#ifdef CONFIG_NETWORK_SECMARK
582 __u32 secmark;
583#endif
Neil Horman3b885782009-10-12 13:26:31 -0700584 union {
585 __u32 mark;
586 __u32 dropcount;
Eric Dumazet16fad692013-03-14 05:40:32 +0000587 __u32 reserved_tailroom;
Neil Horman3b885782009-10-12 13:26:31 -0700588 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589
Simon Horman0d89d202013-05-23 21:02:52 +0000590 __be16 inner_protocol;
Simon Horman1a37e412013-05-23 21:02:51 +0000591 __u16 inner_transport_header;
592 __u16 inner_network_header;
593 __u16 inner_mac_header;
594 __u16 transport_header;
595 __u16 network_header;
596 __u16 mac_header;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 /* These elements must be at the end, see alloc_skb() for details. */
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700598 sk_buff_data_t tail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700599 sk_buff_data_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700600 unsigned char *head,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700601 *data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700602 unsigned int truesize;
603 atomic_t users;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700604};
605
606#ifdef __KERNEL__
607/*
608 * Handling routines are only of interest to the kernel
609 */
610#include <linux/slab.h>
611
Linus Torvalds1da177e2005-04-16 15:20:36 -0700612
Mel Gormanc93bdd02012-07-31 16:44:19 -0700613#define SKB_ALLOC_FCLONE 0x01
614#define SKB_ALLOC_RX 0x02
615
616/* Returns true if the skb was allocated from PFMEMALLOC reserves */
617static inline bool skb_pfmemalloc(const struct sk_buff *skb)
618{
619 return unlikely(skb->pfmemalloc);
620}
621
Eric Dumazet7fee2262010-05-11 23:19:48 +0000622/*
623 * skb might have a dst pointer attached, refcounted or not.
624 * _skb_refdst low order bit is set if refcount was _not_ taken
625 */
626#define SKB_DST_NOREF 1UL
627#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
628
629/**
630 * skb_dst - returns skb dst_entry
631 * @skb: buffer
632 *
633 * Returns skb dst_entry, regardless of reference taken or not.
634 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000635static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
636{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000637 /* If refdst was not refcounted, check we still are in a
638 * rcu_read_lock section
639 */
640 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
641 !rcu_read_lock_held() &&
642 !rcu_read_lock_bh_held());
643 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
Eric Dumazetadf30902009-06-02 05:19:30 +0000644}
645
Eric Dumazet7fee2262010-05-11 23:19:48 +0000646/**
647 * skb_dst_set - sets skb dst
648 * @skb: buffer
649 * @dst: dst entry
650 *
651 * Sets skb dst, assuming a reference was taken on dst and should
652 * be released by skb_dst_drop()
653 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000654static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
655{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000656 skb->_skb_refdst = (unsigned long)dst;
657}
658
Joe Perches7965bd42013-09-26 14:48:15 -0700659void __skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst,
660 bool force);
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200661
662/**
663 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
664 * @skb: buffer
665 * @dst: dst entry
666 *
667 * Sets skb dst, assuming a reference was not taken on dst.
668 * If dst entry is cached, we do not take reference and dst_release
669 * will be avoided by refdst_drop. If dst entry is not cached, we take
670 * reference, so that last dst_release can destroy the dst immediately.
671 */
672static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
673{
674 __skb_dst_set_noref(skb, dst, false);
675}
676
677/**
678 * skb_dst_set_noref_force - sets skb dst, without taking reference
679 * @skb: buffer
680 * @dst: dst entry
681 *
682 * Sets skb dst, assuming a reference was not taken on dst.
683 * No reference is taken and no dst_release will be called. While for
684 * cached dsts deferred reclaim is a basic feature, for entries that are
685 * not cached it is caller's job to guarantee that last dst_release for
686 * provided dst happens when nobody uses it, eg. after a RCU grace period.
687 */
688static inline void skb_dst_set_noref_force(struct sk_buff *skb,
689 struct dst_entry *dst)
690{
691 __skb_dst_set_noref(skb, dst, true);
692}
Eric Dumazet7fee2262010-05-11 23:19:48 +0000693
694/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300695 * skb_dst_is_noref - Test if skb dst isn't refcounted
Eric Dumazet7fee2262010-05-11 23:19:48 +0000696 * @skb: buffer
697 */
698static inline bool skb_dst_is_noref(const struct sk_buff *skb)
699{
700 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
Eric Dumazetadf30902009-06-02 05:19:30 +0000701}
702
Eric Dumazet511c3f92009-06-02 05:14:27 +0000703static inline struct rtable *skb_rtable(const struct sk_buff *skb)
704{
Eric Dumazetadf30902009-06-02 05:19:30 +0000705 return (struct rtable *)skb_dst(skb);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000706}
707
Joe Perches7965bd42013-09-26 14:48:15 -0700708void kfree_skb(struct sk_buff *skb);
709void kfree_skb_list(struct sk_buff *segs);
710void skb_tx_error(struct sk_buff *skb);
711void consume_skb(struct sk_buff *skb);
712void __kfree_skb(struct sk_buff *skb);
Eric Dumazetd7e88832012-04-30 08:10:34 +0000713extern struct kmem_cache *skbuff_head_cache;
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000714
Joe Perches7965bd42013-09-26 14:48:15 -0700715void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
716bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
717 bool *fragstolen, int *delta_truesize);
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000718
Joe Perches7965bd42013-09-26 14:48:15 -0700719struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
720 int node);
721struct sk_buff *build_skb(void *data, unsigned int frag_size);
David S. Millerd179cd12005-08-17 14:57:30 -0700722static inline struct sk_buff *alloc_skb(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100723 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700724{
Eric Dumazet564824b2010-10-11 19:05:25 +0000725 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700726}
727
728static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100729 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700730{
Mel Gormanc93bdd02012-07-31 16:44:19 -0700731 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700732}
733
Joe Perches7965bd42013-09-26 14:48:15 -0700734struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000735static inline struct sk_buff *alloc_skb_head(gfp_t priority)
736{
737 return __alloc_skb_head(priority, -1);
738}
739
Joe Perches7965bd42013-09-26 14:48:15 -0700740struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
741int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
742struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
743struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
Octavian Purdilabad93e92014-06-12 01:36:26 +0300744struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
745 gfp_t gfp_mask, bool fclone);
746static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
747 gfp_t gfp_mask)
748{
749 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
750}
Eric Dumazet117632e2011-12-03 21:39:53 +0000751
Joe Perches7965bd42013-09-26 14:48:15 -0700752int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
753struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
754 unsigned int headroom);
755struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
756 int newtailroom, gfp_t priority);
Fan Du25a91d82014-01-18 09:54:23 +0800757int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
758 int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -0700759int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
760 int len);
761int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
762int skb_pad(struct sk_buff *skb, int pad);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000763#define dev_kfree_skb(a) consume_skb(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Joe Perches7965bd42013-09-26 14:48:15 -0700765int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
766 int getfrag(void *from, char *to, int offset,
767 int len, int odd, struct sk_buff *skb),
768 void *from, int length);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700769
Eric Dumazetd94d9fe2009-11-04 09:50:58 -0800770struct skb_seq_state {
Thomas Graf677e90e2005-06-23 20:59:51 -0700771 __u32 lower_offset;
772 __u32 upper_offset;
773 __u32 frag_idx;
774 __u32 stepped_offset;
775 struct sk_buff *root_skb;
776 struct sk_buff *cur_skb;
777 __u8 *frag_data;
778};
779
Joe Perches7965bd42013-09-26 14:48:15 -0700780void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
781 unsigned int to, struct skb_seq_state *st);
782unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
783 struct skb_seq_state *st);
784void skb_abort_seq_read(struct skb_seq_state *st);
Thomas Graf677e90e2005-06-23 20:59:51 -0700785
Joe Perches7965bd42013-09-26 14:48:15 -0700786unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
787 unsigned int to, struct ts_config *config,
788 struct ts_state *state);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -0700789
Tom Herbert09323cc2013-12-15 22:16:19 -0800790/*
791 * Packet hash types specify the type of hash in skb_set_hash.
792 *
793 * Hash types refer to the protocol layer addresses which are used to
794 * construct a packet's hash. The hashes are used to differentiate or identify
795 * flows of the protocol layer for the hash type. Hash types are either
796 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
797 *
798 * Properties of hashes:
799 *
800 * 1) Two packets in different flows have different hash values
801 * 2) Two packets in the same flow should have the same hash value
802 *
803 * A hash at a higher layer is considered to be more specific. A driver should
804 * set the most specific hash possible.
805 *
806 * A driver cannot indicate a more specific hash than the layer at which a hash
807 * was computed. For instance an L3 hash cannot be set as an L4 hash.
808 *
809 * A driver may indicate a hash level which is less specific than the
810 * actual layer the hash was computed on. For instance, a hash computed
811 * at L4 may be considered an L3 hash. This should only be done if the
812 * driver can't unambiguously determine that the HW computed the hash at
813 * the higher layer. Note that the "should" in the second property above
814 * permits this.
815 */
816enum pkt_hash_types {
817 PKT_HASH_TYPE_NONE, /* Undefined type */
818 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
819 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
820 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
821};
822
823static inline void
824skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
825{
Tom Herbert61b905d2014-03-24 15:34:47 -0700826 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
Tom Herberta3b18dd2014-07-01 21:33:17 -0700827 skb->sw_hash = 0;
Tom Herbert61b905d2014-03-24 15:34:47 -0700828 skb->hash = hash;
Tom Herbert09323cc2013-12-15 22:16:19 -0800829}
830
Tom Herbert3958afa1b2013-12-15 22:12:06 -0800831void __skb_get_hash(struct sk_buff *skb);
832static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +0000833{
Tom Herberta3b18dd2014-07-01 21:33:17 -0700834 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -0800835 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +0000836
Tom Herbert61b905d2014-03-24 15:34:47 -0700837 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +0000838}
839
Tom Herbert57bdf7f42014-01-15 08:57:54 -0800840static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
841{
Tom Herbert61b905d2014-03-24 15:34:47 -0700842 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -0800843}
844
Tom Herbert7539fad2013-12-15 22:12:18 -0800845static inline void skb_clear_hash(struct sk_buff *skb)
846{
Tom Herbert61b905d2014-03-24 15:34:47 -0700847 skb->hash = 0;
Tom Herberta3b18dd2014-07-01 21:33:17 -0700848 skb->sw_hash = 0;
Tom Herbert61b905d2014-03-24 15:34:47 -0700849 skb->l4_hash = 0;
Tom Herbert7539fad2013-12-15 22:12:18 -0800850}
851
852static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
853{
Tom Herbert61b905d2014-03-24 15:34:47 -0700854 if (!skb->l4_hash)
Tom Herbert7539fad2013-12-15 22:12:18 -0800855 skb_clear_hash(skb);
856}
857
Tom Herbert3df7a742013-12-15 22:16:29 -0800858static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
859{
Tom Herbert61b905d2014-03-24 15:34:47 -0700860 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -0700861 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -0700862 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -0800863};
864
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700865#ifdef NET_SKBUFF_DATA_USES_OFFSET
866static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
867{
868 return skb->head + skb->end;
869}
Alexander Duyckec47ea82012-05-04 14:26:56 +0000870
871static inline unsigned int skb_end_offset(const struct sk_buff *skb)
872{
873 return skb->end;
874}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700875#else
876static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
877{
878 return skb->end;
879}
Alexander Duyckec47ea82012-05-04 14:26:56 +0000880
881static inline unsigned int skb_end_offset(const struct sk_buff *skb)
882{
883 return skb->end - skb->head;
884}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700885#endif
886
Linus Torvalds1da177e2005-04-16 15:20:36 -0700887/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700888#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700889
Patrick Ohlyac45f602009-02-12 05:03:37 +0000890static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
891{
892 return &skb_shinfo(skb)->hwtstamps;
893}
894
Linus Torvalds1da177e2005-04-16 15:20:36 -0700895/**
896 * skb_queue_empty - check if a queue is empty
897 * @list: queue head
898 *
899 * Returns true if the queue is empty, false otherwise.
900 */
901static inline int skb_queue_empty(const struct sk_buff_head *list)
902{
Daniel Borkmannfd44b932014-01-07 23:23:44 +0100903 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700904}
905
906/**
David S. Millerfc7ebb22008-09-23 00:34:07 -0700907 * skb_queue_is_last - check if skb is the last entry in the queue
908 * @list: queue head
909 * @skb: buffer
910 *
911 * Returns true if @skb is the last buffer on the list.
912 */
913static inline bool skb_queue_is_last(const struct sk_buff_head *list,
914 const struct sk_buff *skb)
915{
Daniel Borkmannfd44b932014-01-07 23:23:44 +0100916 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -0700917}
918
919/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -0800920 * skb_queue_is_first - check if skb is the first entry in the queue
921 * @list: queue head
922 * @skb: buffer
923 *
924 * Returns true if @skb is the first buffer on the list.
925 */
926static inline bool skb_queue_is_first(const struct sk_buff_head *list,
927 const struct sk_buff *skb)
928{
Daniel Borkmannfd44b932014-01-07 23:23:44 +0100929 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -0800930}
931
932/**
David S. Miller249c8b42008-09-23 00:44:42 -0700933 * skb_queue_next - return the next packet in the queue
934 * @list: queue head
935 * @skb: current buffer
936 *
937 * Return the next packet in @list after @skb. It is only valid to
938 * call this if skb_queue_is_last() evaluates to false.
939 */
940static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
941 const struct sk_buff *skb)
942{
943 /* This BUG_ON may seem severe, but if we just return then we
944 * are going to dereference garbage.
945 */
946 BUG_ON(skb_queue_is_last(list, skb));
947 return skb->next;
948}
949
950/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -0800951 * skb_queue_prev - return the prev packet in the queue
952 * @list: queue head
953 * @skb: current buffer
954 *
955 * Return the prev packet in @list before @skb. It is only valid to
956 * call this if skb_queue_is_first() evaluates to false.
957 */
958static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
959 const struct sk_buff *skb)
960{
961 /* This BUG_ON may seem severe, but if we just return then we
962 * are going to dereference garbage.
963 */
964 BUG_ON(skb_queue_is_first(list, skb));
965 return skb->prev;
966}
967
968/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969 * skb_get - reference buffer
970 * @skb: buffer to reference
971 *
972 * Makes another reference to a socket buffer and returns a pointer
973 * to the buffer.
974 */
975static inline struct sk_buff *skb_get(struct sk_buff *skb)
976{
977 atomic_inc(&skb->users);
978 return skb;
979}
980
981/*
982 * If users == 1, we are the only owner and are can avoid redundant
983 * atomic change.
984 */
985
986/**
Linus Torvalds1da177e2005-04-16 15:20:36 -0700987 * skb_cloned - is the buffer a clone
988 * @skb: buffer to check
989 *
990 * Returns true if the buffer was generated with skb_clone() and is
991 * one of multiple shared copies of the buffer. Cloned buffers are
992 * shared data so must not be written to under normal circumstances.
993 */
994static inline int skb_cloned(const struct sk_buff *skb)
995{
996 return skb->cloned &&
997 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
998}
999
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001000static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1001{
1002 might_sleep_if(pri & __GFP_WAIT);
1003
1004 if (skb_cloned(skb))
1005 return pskb_expand_head(skb, 0, 0, pri);
1006
1007 return 0;
1008}
1009
Linus Torvalds1da177e2005-04-16 15:20:36 -07001010/**
1011 * skb_header_cloned - is the header a clone
1012 * @skb: buffer to check
1013 *
1014 * Returns true if modifying the header part of the buffer requires
1015 * the data to be copied.
1016 */
1017static inline int skb_header_cloned(const struct sk_buff *skb)
1018{
1019 int dataref;
1020
1021 if (!skb->cloned)
1022 return 0;
1023
1024 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1025 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1026 return dataref != 1;
1027}
1028
1029/**
1030 * skb_header_release - release reference to header
1031 * @skb: buffer to operate on
1032 *
1033 * Drop a reference to the header part of the buffer. This is done
1034 * by acquiring a payload reference. You must not read from the header
1035 * part of skb->data after this.
1036 */
1037static inline void skb_header_release(struct sk_buff *skb)
1038{
1039 BUG_ON(skb->nohdr);
1040 skb->nohdr = 1;
1041 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1042}
1043
1044/**
1045 * skb_shared - is the buffer shared
1046 * @skb: buffer to check
1047 *
1048 * Returns true if more than one person has a reference to this
1049 * buffer.
1050 */
1051static inline int skb_shared(const struct sk_buff *skb)
1052{
1053 return atomic_read(&skb->users) != 1;
1054}
1055
1056/**
1057 * skb_share_check - check if buffer is shared and if so clone it
1058 * @skb: buffer to check
1059 * @pri: priority for memory allocation
1060 *
1061 * If the buffer is shared the buffer is cloned and the old copy
1062 * drops a reference. A new clone with a single reference is returned.
1063 * If the buffer is not shared the original buffer is returned. When
1064 * being called from interrupt status or with spinlocks held pri must
1065 * be GFP_ATOMIC.
1066 *
1067 * NULL is returned on a memory allocation failure.
1068 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001069static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070{
1071 might_sleep_if(pri & __GFP_WAIT);
1072 if (skb_shared(skb)) {
1073 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001074
1075 if (likely(nskb))
1076 consume_skb(skb);
1077 else
1078 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079 skb = nskb;
1080 }
1081 return skb;
1082}
1083
1084/*
1085 * Copy shared buffers into a new sk_buff. We effectively do COW on
1086 * packets to handle cases where we have a local reader and forward
1087 * and a couple of other messy ones. The normal one is tcpdumping
1088 * a packet thats being forwarded.
1089 */
1090
1091/**
1092 * skb_unshare - make a copy of a shared buffer
1093 * @skb: buffer to check
1094 * @pri: priority for memory allocation
1095 *
1096 * If the socket buffer is a clone then this function creates a new
1097 * copy of the data, drops a reference count on the old copy and returns
1098 * the new copy with the reference count at 1. If the buffer is not a clone
1099 * the original buffer is returned. When called with a spinlock held or
1100 * from interrupt state @pri must be %GFP_ATOMIC
1101 *
1102 * %NULL is returned on a memory allocation failure.
1103 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001104static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001105 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001106{
1107 might_sleep_if(pri & __GFP_WAIT);
1108 if (skb_cloned(skb)) {
1109 struct sk_buff *nskb = skb_copy(skb, pri);
1110 kfree_skb(skb); /* Free our shared copy */
1111 skb = nskb;
1112 }
1113 return skb;
1114}
1115
1116/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001117 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001118 * @list_: list to peek at
1119 *
1120 * Peek an &sk_buff. Unlike most other operations you _MUST_
1121 * be careful with this one. A peek leaves the buffer on the
1122 * list and someone else may run off with it. You must hold
1123 * the appropriate locks or have a private queue to do this.
1124 *
1125 * Returns %NULL for an empty list or a pointer to the head element.
1126 * The reference count is not incremented and the reference is therefore
1127 * volatile. Use with caution.
1128 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001129static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001130{
Eric Dumazet18d07002012-04-30 16:31:46 +00001131 struct sk_buff *skb = list_->next;
1132
1133 if (skb == (struct sk_buff *)list_)
1134 skb = NULL;
1135 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001136}
1137
1138/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001139 * skb_peek_next - peek skb following the given one from a queue
1140 * @skb: skb to start from
1141 * @list_: list to peek at
1142 *
1143 * Returns %NULL when the end of the list is met or a pointer to the
1144 * next element. The reference count is not incremented and the
1145 * reference is therefore volatile. Use with caution.
1146 */
1147static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1148 const struct sk_buff_head *list_)
1149{
1150 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001151
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001152 if (next == (struct sk_buff *)list_)
1153 next = NULL;
1154 return next;
1155}
1156
1157/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001158 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001159 * @list_: list to peek at
1160 *
1161 * Peek an &sk_buff. Unlike most other operations you _MUST_
1162 * be careful with this one. A peek leaves the buffer on the
1163 * list and someone else may run off with it. You must hold
1164 * the appropriate locks or have a private queue to do this.
1165 *
1166 * Returns %NULL for an empty list or a pointer to the tail element.
1167 * The reference count is not incremented and the reference is therefore
1168 * volatile. Use with caution.
1169 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001170static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001171{
Eric Dumazet18d07002012-04-30 16:31:46 +00001172 struct sk_buff *skb = list_->prev;
1173
1174 if (skb == (struct sk_buff *)list_)
1175 skb = NULL;
1176 return skb;
1177
Linus Torvalds1da177e2005-04-16 15:20:36 -07001178}
1179
1180/**
1181 * skb_queue_len - get queue length
1182 * @list_: list to measure
1183 *
1184 * Return the length of an &sk_buff queue.
1185 */
1186static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1187{
1188 return list_->qlen;
1189}
1190
David S. Miller67fed452008-09-21 22:36:24 -07001191/**
1192 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1193 * @list: queue to initialize
1194 *
1195 * This initializes only the list and queue length aspects of
1196 * an sk_buff_head object. This allows to initialize the list
1197 * aspects of an sk_buff_head without reinitializing things like
1198 * the spinlock. It can also be used for on-stack sk_buff_head
1199 * objects where the spinlock is known to not be used.
1200 */
1201static inline void __skb_queue_head_init(struct sk_buff_head *list)
1202{
1203 list->prev = list->next = (struct sk_buff *)list;
1204 list->qlen = 0;
1205}
1206
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001207/*
1208 * This function creates a split out lock class for each invocation;
1209 * this is needed for now since a whole lot of users of the skb-queue
1210 * infrastructure in drivers have different locking usage (in hardirq)
1211 * than the networking core (in softirq only). In the long run either the
1212 * network layer or drivers should need annotation to consolidate the
1213 * main types of usage into 3 classes.
1214 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001215static inline void skb_queue_head_init(struct sk_buff_head *list)
1216{
1217 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001218 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001219}
1220
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001221static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1222 struct lock_class_key *class)
1223{
1224 skb_queue_head_init(list);
1225 lockdep_set_class(&list->lock, class);
1226}
1227
Linus Torvalds1da177e2005-04-16 15:20:36 -07001228/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001229 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001230 *
1231 * The "__skb_xxxx()" functions are the non-atomic ones that
1232 * can only be called with interrupts disabled.
1233 */
Joe Perches7965bd42013-09-26 14:48:15 -07001234void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1235 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001236static inline void __skb_insert(struct sk_buff *newsk,
1237 struct sk_buff *prev, struct sk_buff *next,
1238 struct sk_buff_head *list)
1239{
1240 newsk->next = next;
1241 newsk->prev = prev;
1242 next->prev = prev->next = newsk;
1243 list->qlen++;
1244}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001245
David S. Miller67fed452008-09-21 22:36:24 -07001246static inline void __skb_queue_splice(const struct sk_buff_head *list,
1247 struct sk_buff *prev,
1248 struct sk_buff *next)
1249{
1250 struct sk_buff *first = list->next;
1251 struct sk_buff *last = list->prev;
1252
1253 first->prev = prev;
1254 prev->next = first;
1255
1256 last->next = next;
1257 next->prev = last;
1258}
1259
1260/**
1261 * skb_queue_splice - join two skb lists, this is designed for stacks
1262 * @list: the new list to add
1263 * @head: the place to add it in the first list
1264 */
1265static inline void skb_queue_splice(const struct sk_buff_head *list,
1266 struct sk_buff_head *head)
1267{
1268 if (!skb_queue_empty(list)) {
1269 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001270 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001271 }
1272}
1273
1274/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001275 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001276 * @list: the new list to add
1277 * @head: the place to add it in the first list
1278 *
1279 * The list at @list is reinitialised
1280 */
1281static inline void skb_queue_splice_init(struct sk_buff_head *list,
1282 struct sk_buff_head *head)
1283{
1284 if (!skb_queue_empty(list)) {
1285 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001286 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001287 __skb_queue_head_init(list);
1288 }
1289}
1290
1291/**
1292 * skb_queue_splice_tail - join two skb lists, each list being a queue
1293 * @list: the new list to add
1294 * @head: the place to add it in the first list
1295 */
1296static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1297 struct sk_buff_head *head)
1298{
1299 if (!skb_queue_empty(list)) {
1300 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001301 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001302 }
1303}
1304
1305/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001306 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001307 * @list: the new list to add
1308 * @head: the place to add it in the first list
1309 *
1310 * Each of the lists is a queue.
1311 * The list at @list is reinitialised
1312 */
1313static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1314 struct sk_buff_head *head)
1315{
1316 if (!skb_queue_empty(list)) {
1317 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001318 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001319 __skb_queue_head_init(list);
1320 }
1321}
1322
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001324 * __skb_queue_after - queue a buffer at the list head
1325 * @list: list to use
1326 * @prev: place after this buffer
1327 * @newsk: buffer to queue
1328 *
1329 * Queue a buffer int the middle of a list. This function takes no locks
1330 * and you must therefore hold required locks before calling it.
1331 *
1332 * A buffer cannot be placed on two lists at the same time.
1333 */
1334static inline void __skb_queue_after(struct sk_buff_head *list,
1335 struct sk_buff *prev,
1336 struct sk_buff *newsk)
1337{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001338 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001339}
1340
Joe Perches7965bd42013-09-26 14:48:15 -07001341void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1342 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001343
Gerrit Renkerf5572852008-04-14 00:05:28 -07001344static inline void __skb_queue_before(struct sk_buff_head *list,
1345 struct sk_buff *next,
1346 struct sk_buff *newsk)
1347{
1348 __skb_insert(newsk, next->prev, next, list);
1349}
1350
Stephen Hemminger300ce172005-10-30 13:47:34 -08001351/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001352 * __skb_queue_head - queue a buffer at the list head
1353 * @list: list to use
1354 * @newsk: buffer to queue
1355 *
1356 * Queue a buffer at the start of a list. This function takes no locks
1357 * and you must therefore hold required locks before calling it.
1358 *
1359 * A buffer cannot be placed on two lists at the same time.
1360 */
Joe Perches7965bd42013-09-26 14:48:15 -07001361void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362static inline void __skb_queue_head(struct sk_buff_head *list,
1363 struct sk_buff *newsk)
1364{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001365 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001366}
1367
1368/**
1369 * __skb_queue_tail - queue a buffer at the list tail
1370 * @list: list to use
1371 * @newsk: buffer to queue
1372 *
1373 * Queue a buffer at the end of a list. This function takes no locks
1374 * and you must therefore hold required locks before calling it.
1375 *
1376 * A buffer cannot be placed on two lists at the same time.
1377 */
Joe Perches7965bd42013-09-26 14:48:15 -07001378void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001379static inline void __skb_queue_tail(struct sk_buff_head *list,
1380 struct sk_buff *newsk)
1381{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001382 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383}
1384
Linus Torvalds1da177e2005-04-16 15:20:36 -07001385/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386 * remove sk_buff from list. _Must_ be called atomically, and with
1387 * the list known..
1388 */
Joe Perches7965bd42013-09-26 14:48:15 -07001389void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1391{
1392 struct sk_buff *next, *prev;
1393
1394 list->qlen--;
1395 next = skb->next;
1396 prev = skb->prev;
1397 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 next->prev = prev;
1399 prev->next = next;
1400}
1401
Gerrit Renkerf525c062008-04-14 00:04:12 -07001402/**
1403 * __skb_dequeue - remove from the head of the queue
1404 * @list: list to dequeue from
1405 *
1406 * Remove the head of the list. This function does not take any locks
1407 * so must be used with appropriate locks held only. The head item is
1408 * returned or %NULL if the list is empty.
1409 */
Joe Perches7965bd42013-09-26 14:48:15 -07001410struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001411static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1412{
1413 struct sk_buff *skb = skb_peek(list);
1414 if (skb)
1415 __skb_unlink(skb, list);
1416 return skb;
1417}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001418
1419/**
1420 * __skb_dequeue_tail - remove from the tail of the queue
1421 * @list: list to dequeue from
1422 *
1423 * Remove the tail of the list. This function does not take any locks
1424 * so must be used with appropriate locks held only. The tail item is
1425 * returned or %NULL if the list is empty.
1426 */
Joe Perches7965bd42013-09-26 14:48:15 -07001427struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001428static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1429{
1430 struct sk_buff *skb = skb_peek_tail(list);
1431 if (skb)
1432 __skb_unlink(skb, list);
1433 return skb;
1434}
1435
1436
David S. Millerbdcc0922012-03-07 20:53:36 -05001437static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001438{
1439 return skb->data_len;
1440}
1441
1442static inline unsigned int skb_headlen(const struct sk_buff *skb)
1443{
1444 return skb->len - skb->data_len;
1445}
1446
1447static inline int skb_pagelen(const struct sk_buff *skb)
1448{
1449 int i, len = 0;
1450
1451 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001452 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453 return len + skb_headlen(skb);
1454}
1455
Ian Campbell131ea662011-08-19 06:25:00 +00001456/**
1457 * __skb_fill_page_desc - initialise a paged fragment in an skb
1458 * @skb: buffer containing fragment to be initialised
1459 * @i: paged fragment index to initialise
1460 * @page: the page to use for this fragment
1461 * @off: the offset to the data with @page
1462 * @size: the length of the data
1463 *
1464 * Initialises the @i'th fragment of @skb to point to &size bytes at
1465 * offset @off within @page.
1466 *
1467 * Does not take any additional reference on the fragment.
1468 */
1469static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1470 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001471{
1472 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1473
Mel Gormanc48a11c2012-07-31 16:44:23 -07001474 /*
1475 * Propagate page->pfmemalloc to the skb if we can. The problem is
1476 * that not all callers have unique ownership of the page. If
1477 * pfmemalloc is set, we check the mapping as a mapping implies
1478 * page->index is set (index and pfmemalloc share space).
1479 * If it's a valid mapping, we cannot use page->pfmemalloc but we
1480 * do not lose pfmemalloc information as the pages would not be
1481 * allocated using __GFP_MEMALLOC.
1482 */
Ian Campbella8605c62011-10-19 23:01:49 +00001483 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001484 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001485 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001486
1487 page = compound_head(page);
1488 if (page->pfmemalloc && !page->mapping)
1489 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001490}
1491
1492/**
1493 * skb_fill_page_desc - initialise a paged fragment in an skb
1494 * @skb: buffer containing fragment to be initialised
1495 * @i: paged fragment index to initialise
1496 * @page: the page to use for this fragment
1497 * @off: the offset to the data with @page
1498 * @size: the length of the data
1499 *
1500 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001501 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001502 * addition updates @skb such that @i is the last fragment.
1503 *
1504 * Does not take any additional reference on the fragment.
1505 */
1506static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1507 struct page *page, int off, int size)
1508{
1509 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001510 skb_shinfo(skb)->nr_frags = i + 1;
1511}
1512
Joe Perches7965bd42013-09-26 14:48:15 -07001513void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1514 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001515
Jason Wangf8e617e2013-11-01 14:07:47 +08001516void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1517 unsigned int truesize);
1518
Linus Torvalds1da177e2005-04-16 15:20:36 -07001519#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001520#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001521#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1522
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001523#ifdef NET_SKBUFF_DATA_USES_OFFSET
1524static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1525{
1526 return skb->head + skb->tail;
1527}
1528
1529static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1530{
1531 skb->tail = skb->data - skb->head;
1532}
1533
1534static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1535{
1536 skb_reset_tail_pointer(skb);
1537 skb->tail += offset;
1538}
Simon Horman7cc46192013-05-28 20:34:29 +00001539
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001540#else /* NET_SKBUFF_DATA_USES_OFFSET */
1541static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1542{
1543 return skb->tail;
1544}
1545
1546static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1547{
1548 skb->tail = skb->data;
1549}
1550
1551static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1552{
1553 skb->tail = skb->data + offset;
1554}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001555
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001556#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1557
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558/*
1559 * Add data to an sk_buff
1560 */
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001561unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001562unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1564{
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001565 unsigned char *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566 SKB_LINEAR_ASSERT(skb);
1567 skb->tail += len;
1568 skb->len += len;
1569 return tmp;
1570}
1571
Joe Perches7965bd42013-09-26 14:48:15 -07001572unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001573static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1574{
1575 skb->data -= len;
1576 skb->len += len;
1577 return skb->data;
1578}
1579
Joe Perches7965bd42013-09-26 14:48:15 -07001580unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001581static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1582{
1583 skb->len -= len;
1584 BUG_ON(skb->len < skb->data_len);
1585 return skb->data += len;
1586}
1587
David S. Miller47d29642010-05-02 02:21:44 -07001588static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1589{
1590 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1591}
1592
Joe Perches7965bd42013-09-26 14:48:15 -07001593unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594
1595static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1596{
1597 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07001598 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 return NULL;
1600 skb->len -= len;
1601 return skb->data += len;
1602}
1603
1604static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1605{
1606 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1607}
1608
1609static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1610{
1611 if (likely(len <= skb_headlen(skb)))
1612 return 1;
1613 if (unlikely(len > skb->len))
1614 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07001615 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616}
1617
1618/**
1619 * skb_headroom - bytes at buffer head
1620 * @skb: buffer to check
1621 *
1622 * Return the number of bytes of free space at the head of an &sk_buff.
1623 */
Chuck Leverc2636b42007-10-23 21:07:32 -07001624static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625{
1626 return skb->data - skb->head;
1627}
1628
1629/**
1630 * skb_tailroom - bytes at buffer end
1631 * @skb: buffer to check
1632 *
1633 * Return the number of bytes of free space at the tail of an sk_buff
1634 */
1635static inline int skb_tailroom(const struct sk_buff *skb)
1636{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001637 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001638}
1639
1640/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00001641 * skb_availroom - bytes at buffer end
1642 * @skb: buffer to check
1643 *
1644 * Return the number of bytes of free space at the tail of an sk_buff
1645 * allocated by sk_stream_alloc()
1646 */
1647static inline int skb_availroom(const struct sk_buff *skb)
1648{
Eric Dumazet16fad692013-03-14 05:40:32 +00001649 if (skb_is_nonlinear(skb))
1650 return 0;
1651
1652 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00001653}
1654
1655/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001656 * skb_reserve - adjust headroom
1657 * @skb: buffer to alter
1658 * @len: bytes to move
1659 *
1660 * Increase the headroom of an empty &sk_buff by reducing the tail
1661 * room. This is only allowed for an empty buffer.
1662 */
David S. Miller8243126c2006-01-17 02:54:21 -08001663static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001664{
1665 skb->data += len;
1666 skb->tail += len;
1667}
1668
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001669static inline void skb_reset_inner_headers(struct sk_buff *skb)
1670{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00001671 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001672 skb->inner_network_header = skb->network_header;
1673 skb->inner_transport_header = skb->transport_header;
1674}
1675
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00001676static inline void skb_reset_mac_len(struct sk_buff *skb)
1677{
1678 skb->mac_len = skb->network_header - skb->mac_header;
1679}
1680
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001681static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1682 *skb)
1683{
1684 return skb->head + skb->inner_transport_header;
1685}
1686
1687static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1688{
1689 skb->inner_transport_header = skb->data - skb->head;
1690}
1691
1692static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1693 const int offset)
1694{
1695 skb_reset_inner_transport_header(skb);
1696 skb->inner_transport_header += offset;
1697}
1698
1699static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1700{
1701 return skb->head + skb->inner_network_header;
1702}
1703
1704static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1705{
1706 skb->inner_network_header = skb->data - skb->head;
1707}
1708
1709static inline void skb_set_inner_network_header(struct sk_buff *skb,
1710 const int offset)
1711{
1712 skb_reset_inner_network_header(skb);
1713 skb->inner_network_header += offset;
1714}
1715
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00001716static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1717{
1718 return skb->head + skb->inner_mac_header;
1719}
1720
1721static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1722{
1723 skb->inner_mac_header = skb->data - skb->head;
1724}
1725
1726static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1727 const int offset)
1728{
1729 skb_reset_inner_mac_header(skb);
1730 skb->inner_mac_header += offset;
1731}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00001732static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1733{
Cong Wang35d04612013-05-29 15:16:05 +08001734 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00001735}
1736
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07001737static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1738{
1739 return skb->head + skb->transport_header;
1740}
1741
1742static inline void skb_reset_transport_header(struct sk_buff *skb)
1743{
1744 skb->transport_header = skb->data - skb->head;
1745}
1746
1747static inline void skb_set_transport_header(struct sk_buff *skb,
1748 const int offset)
1749{
1750 skb_reset_transport_header(skb);
1751 skb->transport_header += offset;
1752}
1753
1754static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1755{
1756 return skb->head + skb->network_header;
1757}
1758
1759static inline void skb_reset_network_header(struct sk_buff *skb)
1760{
1761 skb->network_header = skb->data - skb->head;
1762}
1763
1764static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1765{
1766 skb_reset_network_header(skb);
1767 skb->network_header += offset;
1768}
1769
1770static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1771{
1772 return skb->head + skb->mac_header;
1773}
1774
1775static inline int skb_mac_header_was_set(const struct sk_buff *skb)
1776{
Cong Wang35d04612013-05-29 15:16:05 +08001777 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07001778}
1779
1780static inline void skb_reset_mac_header(struct sk_buff *skb)
1781{
1782 skb->mac_header = skb->data - skb->head;
1783}
1784
1785static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
1786{
1787 skb_reset_mac_header(skb);
1788 skb->mac_header += offset;
1789}
1790
Timo Teräs0e3da5b2013-12-16 11:02:09 +02001791static inline void skb_pop_mac_header(struct sk_buff *skb)
1792{
1793 skb->mac_header = skb->network_header;
1794}
1795
Ying Xuefbbdb8f2013-03-27 16:46:06 +00001796static inline void skb_probe_transport_header(struct sk_buff *skb,
1797 const int offset_hint)
1798{
1799 struct flow_keys keys;
1800
1801 if (skb_transport_header_was_set(skb))
1802 return;
1803 else if (skb_flow_dissect(skb, &keys))
1804 skb_set_transport_header(skb, keys.thoff);
1805 else
1806 skb_set_transport_header(skb, offset_hint);
1807}
1808
Eric Dumazet03606892012-02-23 10:55:02 +00001809static inline void skb_mac_header_rebuild(struct sk_buff *skb)
1810{
1811 if (skb_mac_header_was_set(skb)) {
1812 const unsigned char *old_mac = skb_mac_header(skb);
1813
1814 skb_set_mac_header(skb, -skb->mac_len);
1815 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
1816 }
1817}
1818
Michał Mirosław04fb4512010-12-14 15:24:08 +00001819static inline int skb_checksum_start_offset(const struct sk_buff *skb)
1820{
1821 return skb->csum_start - skb_headroom(skb);
1822}
1823
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07001824static inline int skb_transport_offset(const struct sk_buff *skb)
1825{
1826 return skb_transport_header(skb) - skb->data;
1827}
1828
1829static inline u32 skb_network_header_len(const struct sk_buff *skb)
1830{
1831 return skb->transport_header - skb->network_header;
1832}
1833
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001834static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
1835{
1836 return skb->inner_transport_header - skb->inner_network_header;
1837}
1838
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07001839static inline int skb_network_offset(const struct sk_buff *skb)
1840{
1841 return skb_network_header(skb) - skb->data;
1842}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03001843
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001844static inline int skb_inner_network_offset(const struct sk_buff *skb)
1845{
1846 return skb_inner_network_header(skb) - skb->data;
1847}
1848
Changli Gaof9599ce2010-08-04 04:43:44 +00001849static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
1850{
1851 return pskb_may_pull(skb, skb_network_offset(skb) + len);
1852}
1853
Tom Herberte5eb4e32014-06-14 23:24:28 -07001854static inline void skb_pop_rcv_encapsulation(struct sk_buff *skb)
1855{
1856 /* Only continue with checksum unnecessary if device indicated
1857 * it is valid across encapsulation (skb->encapsulation was set).
1858 */
1859 if (skb->ip_summed == CHECKSUM_UNNECESSARY && !skb->encapsulation)
1860 skb->ip_summed = CHECKSUM_NONE;
1861
1862 skb->encapsulation = 0;
1863 skb->csum_valid = 0;
1864}
1865
Linus Torvalds1da177e2005-04-16 15:20:36 -07001866/*
1867 * CPUs often take a performance hit when accessing unaligned memory
1868 * locations. The actual performance hit varies, it can be small if the
1869 * hardware handles it or large if we have to take an exception and fix it
1870 * in software.
1871 *
1872 * Since an ethernet header is 14 bytes network drivers often end up with
1873 * the IP header at an unaligned offset. The IP header can be aligned by
1874 * shifting the start of the packet by 2 bytes. Drivers should do this
1875 * with:
1876 *
Tobias Klauser8660c122009-07-13 22:48:16 +00001877 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878 *
1879 * The downside to this alignment of the IP header is that the DMA is now
1880 * unaligned. On some architectures the cost of an unaligned DMA is high
1881 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00001882 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
1884 * to be overridden.
1885 */
1886#ifndef NET_IP_ALIGN
1887#define NET_IP_ALIGN 2
1888#endif
1889
Anton Blanchard025be812006-03-31 02:27:06 -08001890/*
1891 * The networking layer reserves some headroom in skb data (via
1892 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
1893 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08001894 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08001895 *
1896 * Unfortunately this headroom changes the DMA alignment of the resulting
1897 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
1898 * on some architectures. An architecture can override this value,
1899 * perhaps setting it to a cacheline in size (since that will maintain
1900 * cacheline alignment of the DMA). It must be a power of 2.
1901 *
David S. Millerd6301d32009-02-08 19:24:13 -08001902 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08001903 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07001904 *
1905 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
1906 * to reduce average number of cache lines per packet.
1907 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07001908 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08001909 */
1910#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07001911#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08001912#endif
1913
Joe Perches7965bd42013-09-26 14:48:15 -07001914int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001915
1916static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
1917{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00001918 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07001919 WARN_ON(1);
1920 return;
1921 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001922 skb->len = len;
1923 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924}
1925
Joe Perches7965bd42013-09-26 14:48:15 -07001926void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001927
1928static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
1929{
Herbert Xu3cc0e872006-06-09 16:13:38 -07001930 if (skb->data_len)
1931 return ___pskb_trim(skb, len);
1932 __skb_trim(skb, len);
1933 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934}
1935
1936static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
1937{
1938 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
1939}
1940
1941/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07001942 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
1943 * @skb: buffer to alter
1944 * @len: new length
1945 *
1946 * This is identical to pskb_trim except that the caller knows that
1947 * the skb is not cloned so we should never get an error due to out-
1948 * of-memory.
1949 */
1950static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
1951{
1952 int err = pskb_trim(skb, len);
1953 BUG_ON(err);
1954}
1955
1956/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957 * skb_orphan - orphan a buffer
1958 * @skb: buffer to orphan
1959 *
1960 * If a buffer currently has an owner then we call the owner's
1961 * destructor function and make the @skb unowned. The buffer continues
1962 * to exist but is no longer charged to its former owner.
1963 */
1964static inline void skb_orphan(struct sk_buff *skb)
1965{
Eric Dumazetc34a7612013-07-30 16:11:15 -07001966 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07001967 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07001968 skb->destructor = NULL;
1969 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07001970 } else {
1971 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07001972 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973}
1974
1975/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00001976 * skb_orphan_frags - orphan the frags contained in a buffer
1977 * @skb: buffer to orphan frags from
1978 * @gfp_mask: allocation mask for replacement pages
1979 *
1980 * For each frag in the SKB which needs a destructor (i.e. has an
1981 * owner) create a copy of that frag and release the original
1982 * page by calling the destructor.
1983 */
1984static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
1985{
1986 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
1987 return 0;
1988 return skb_copy_ubufs(skb, gfp_mask);
1989}
1990
1991/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992 * __skb_queue_purge - empty a list
1993 * @list: list to empty
1994 *
1995 * Delete all buffers on an &sk_buff list. Each buffer is removed from
1996 * the list and one reference dropped. This function does not take the
1997 * list lock and the caller must hold the relevant locks to use it.
1998 */
Joe Perches7965bd42013-09-26 14:48:15 -07001999void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000static inline void __skb_queue_purge(struct sk_buff_head *list)
2001{
2002 struct sk_buff *skb;
2003 while ((skb = __skb_dequeue(list)) != NULL)
2004 kfree_skb(skb);
2005}
2006
Alexander Duycke5e67302013-02-08 10:17:15 +00002007#define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768)
2008#define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER)
2009#define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE
2010
Joe Perches7965bd42013-09-26 14:48:15 -07002011void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002012
Joe Perches7965bd42013-09-26 14:48:15 -07002013struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2014 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002015
2016/**
2017 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2018 * @dev: network device to receive on
2019 * @length: length to allocate
2020 *
2021 * Allocate a new &sk_buff and assign it a usage count of one. The
2022 * buffer has unspecified headroom built in. Users should allocate
2023 * the headroom they think they need without accounting for the
2024 * built in space. The built in space is used for optimisations.
2025 *
2026 * %NULL is returned if there is no free memory. Although this function
2027 * allocates memory it can be called from an interrupt.
2028 */
2029static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002030 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002031{
2032 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2033}
2034
Eric Dumazet6f532612012-05-18 05:12:12 +00002035/* legacy helper around __netdev_alloc_skb() */
2036static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2037 gfp_t gfp_mask)
2038{
2039 return __netdev_alloc_skb(NULL, length, gfp_mask);
2040}
2041
2042/* legacy helper around netdev_alloc_skb() */
2043static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2044{
2045 return netdev_alloc_skb(NULL, length);
2046}
2047
2048
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002049static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2050 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002051{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002052 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002053
2054 if (NET_IP_ALIGN && skb)
2055 skb_reserve(skb, NET_IP_ALIGN);
2056 return skb;
2057}
2058
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002059static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2060 unsigned int length)
2061{
2062 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2063}
2064
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002065/**
2066 * __skb_alloc_pages - allocate pages for ps-rx on a skb and preserve pfmemalloc data
Mel Gorman06140022012-07-31 16:44:24 -07002067 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2068 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2069 * @order: size of the allocation
2070 *
2071 * Allocate a new page.
2072 *
2073 * %NULL is returned if there is no free memory.
2074*/
2075static inline struct page *__skb_alloc_pages(gfp_t gfp_mask,
2076 struct sk_buff *skb,
2077 unsigned int order)
2078{
2079 struct page *page;
2080
2081 gfp_mask |= __GFP_COLD;
2082
2083 if (!(gfp_mask & __GFP_NOMEMALLOC))
2084 gfp_mask |= __GFP_MEMALLOC;
2085
2086 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2087 if (skb && page && page->pfmemalloc)
2088 skb->pfmemalloc = true;
2089
2090 return page;
2091}
2092
2093/**
2094 * __skb_alloc_page - allocate a page for ps-rx for a given skb and preserve pfmemalloc data
2095 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX
2096 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used
2097 *
2098 * Allocate a new page.
2099 *
2100 * %NULL is returned if there is no free memory.
2101 */
2102static inline struct page *__skb_alloc_page(gfp_t gfp_mask,
2103 struct sk_buff *skb)
2104{
2105 return __skb_alloc_pages(gfp_mask, skb, 0);
2106}
2107
2108/**
2109 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2110 * @page: The page that was allocated from skb_alloc_page
2111 * @skb: The skb that may need pfmemalloc set
2112 */
2113static inline void skb_propagate_pfmemalloc(struct page *page,
2114 struct sk_buff *skb)
2115{
2116 if (page && page->pfmemalloc)
2117 skb->pfmemalloc = true;
2118}
2119
Eric Dumazet564824b2010-10-11 19:05:25 +00002120/**
Masanari Iidae2278672014-02-18 22:54:36 +09002121 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002122 * @frag: the paged fragment
2123 *
2124 * Returns the &struct page associated with @frag.
2125 */
2126static inline struct page *skb_frag_page(const skb_frag_t *frag)
2127{
Ian Campbella8605c62011-10-19 23:01:49 +00002128 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002129}
2130
2131/**
2132 * __skb_frag_ref - take an addition reference on a paged fragment.
2133 * @frag: the paged fragment
2134 *
2135 * Takes an additional reference on the paged fragment @frag.
2136 */
2137static inline void __skb_frag_ref(skb_frag_t *frag)
2138{
2139 get_page(skb_frag_page(frag));
2140}
2141
2142/**
2143 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2144 * @skb: the buffer
2145 * @f: the fragment offset.
2146 *
2147 * Takes an additional reference on the @f'th paged fragment of @skb.
2148 */
2149static inline void skb_frag_ref(struct sk_buff *skb, int f)
2150{
2151 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2152}
2153
2154/**
2155 * __skb_frag_unref - release a reference on a paged fragment.
2156 * @frag: the paged fragment
2157 *
2158 * Releases a reference on the paged fragment @frag.
2159 */
2160static inline void __skb_frag_unref(skb_frag_t *frag)
2161{
2162 put_page(skb_frag_page(frag));
2163}
2164
2165/**
2166 * skb_frag_unref - release a reference on a paged fragment of an skb.
2167 * @skb: the buffer
2168 * @f: the fragment offset
2169 *
2170 * Releases a reference on the @f'th paged fragment of @skb.
2171 */
2172static inline void skb_frag_unref(struct sk_buff *skb, int f)
2173{
2174 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2175}
2176
2177/**
2178 * skb_frag_address - gets the address of the data contained in a paged fragment
2179 * @frag: the paged fragment buffer
2180 *
2181 * Returns the address of the data within @frag. The page must already
2182 * be mapped.
2183 */
2184static inline void *skb_frag_address(const skb_frag_t *frag)
2185{
2186 return page_address(skb_frag_page(frag)) + frag->page_offset;
2187}
2188
2189/**
2190 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2191 * @frag: the paged fragment buffer
2192 *
2193 * Returns the address of the data within @frag. Checks that the page
2194 * is mapped and returns %NULL otherwise.
2195 */
2196static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2197{
2198 void *ptr = page_address(skb_frag_page(frag));
2199 if (unlikely(!ptr))
2200 return NULL;
2201
2202 return ptr + frag->page_offset;
2203}
2204
2205/**
2206 * __skb_frag_set_page - sets the page contained in a paged fragment
2207 * @frag: the paged fragment
2208 * @page: the page to set
2209 *
2210 * Sets the fragment @frag to contain @page.
2211 */
2212static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2213{
Ian Campbella8605c62011-10-19 23:01:49 +00002214 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002215}
2216
2217/**
2218 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2219 * @skb: the buffer
2220 * @f: the fragment offset
2221 * @page: the page to set
2222 *
2223 * Sets the @f'th fragment of @skb to contain @page.
2224 */
2225static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2226 struct page *page)
2227{
2228 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2229}
2230
Eric Dumazet400dfd32013-10-17 16:27:07 -07002231bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2232
Ian Campbell131ea662011-08-19 06:25:00 +00002233/**
2234 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002235 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002236 * @frag: the paged fragment to map
2237 * @offset: the offset within the fragment (starting at the
2238 * fragment's own offset)
2239 * @size: the number of bytes to map
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002240 * @dir: the direction of the mapping (%PCI_DMA_*)
Ian Campbell131ea662011-08-19 06:25:00 +00002241 *
2242 * Maps the page associated with @frag to @device.
2243 */
2244static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2245 const skb_frag_t *frag,
2246 size_t offset, size_t size,
2247 enum dma_data_direction dir)
2248{
2249 return dma_map_page(dev, skb_frag_page(frag),
2250 frag->page_offset + offset, size, dir);
2251}
2252
Eric Dumazet117632e2011-12-03 21:39:53 +00002253static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2254 gfp_t gfp_mask)
2255{
2256 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2257}
2258
Octavian Purdilabad93e92014-06-12 01:36:26 +03002259
2260static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2261 gfp_t gfp_mask)
2262{
2263 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2264}
2265
2266
Ian Campbell131ea662011-08-19 06:25:00 +00002267/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002268 * skb_clone_writable - is the header of a clone writable
2269 * @skb: buffer to check
2270 * @len: length up to which to write
2271 *
2272 * Returns true if modifying the header part of the cloned buffer
2273 * does not requires the data to be copied.
2274 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002275static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002276{
2277 return !skb_header_cloned(skb) &&
2278 skb_headroom(skb) + len <= skb->hdr_len;
2279}
2280
Herbert Xud9cc2042007-09-16 16:21:16 -07002281static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2282 int cloned)
2283{
2284 int delta = 0;
2285
Herbert Xud9cc2042007-09-16 16:21:16 -07002286 if (headroom > skb_headroom(skb))
2287 delta = headroom - skb_headroom(skb);
2288
2289 if (delta || cloned)
2290 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2291 GFP_ATOMIC);
2292 return 0;
2293}
2294
Patrick McHardy334a8132007-06-25 04:35:20 -07002295/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296 * skb_cow - copy header of skb when it is required
2297 * @skb: buffer to cow
2298 * @headroom: needed headroom
2299 *
2300 * If the skb passed lacks sufficient headroom or its data part
2301 * is shared, data is reallocated. If reallocation fails, an error
2302 * is returned and original skb is not changed.
2303 *
2304 * The result is skb with writable area skb->head...skb->tail
2305 * and at least @headroom of space at head.
2306 */
2307static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2308{
Herbert Xud9cc2042007-09-16 16:21:16 -07002309 return __skb_cow(skb, headroom, skb_cloned(skb));
2310}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002311
Herbert Xud9cc2042007-09-16 16:21:16 -07002312/**
2313 * skb_cow_head - skb_cow but only making the head writable
2314 * @skb: buffer to cow
2315 * @headroom: needed headroom
2316 *
2317 * This function is identical to skb_cow except that we replace the
2318 * skb_cloned check by skb_header_cloned. It should be used when
2319 * you only need to push on some header and do not need to modify
2320 * the data.
2321 */
2322static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2323{
2324 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325}
2326
2327/**
2328 * skb_padto - pad an skbuff up to a minimal size
2329 * @skb: buffer to pad
2330 * @len: minimal length
2331 *
2332 * Pads up a buffer to ensure the trailing bytes exist and are
2333 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002334 * is untouched. Otherwise it is extended. Returns zero on
2335 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 */
2337
Herbert Xu5b057c62006-06-23 02:06:41 -07002338static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002339{
2340 unsigned int size = skb->len;
2341 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002342 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002343 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002344}
2345
2346static inline int skb_add_data(struct sk_buff *skb,
2347 char __user *from, int copy)
2348{
2349 const int off = skb->len;
2350
2351 if (skb->ip_summed == CHECKSUM_NONE) {
2352 int err = 0;
Al Viro50842052006-11-14 21:36:34 -08002353 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy),
Linus Torvalds1da177e2005-04-16 15:20:36 -07002354 copy, 0, &err);
2355 if (!err) {
2356 skb->csum = csum_block_add(skb->csum, csum, off);
2357 return 0;
2358 }
2359 } else if (!copy_from_user(skb_put(skb, copy), from, copy))
2360 return 0;
2361
2362 __skb_trim(skb, off);
2363 return -EFAULT;
2364}
2365
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002366static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2367 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002368{
2369 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002370 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002371
Ian Campbellea2ab692011-08-22 23:44:58 +00002372 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00002373 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002374 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002375 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002376}
2377
Herbert Xu364c6ba2006-06-09 16:10:40 -07002378static inline int __skb_linearize(struct sk_buff *skb)
2379{
2380 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2381}
2382
Linus Torvalds1da177e2005-04-16 15:20:36 -07002383/**
2384 * skb_linearize - convert paged skb to linear one
2385 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07002386 *
2387 * If there is no free memory -ENOMEM is returned, otherwise zero
2388 * is returned and the old skb data released.
2389 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07002390static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002391{
Herbert Xu364c6ba2006-06-09 16:10:40 -07002392 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2393}
2394
2395/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00002396 * skb_has_shared_frag - can any frag be overwritten
2397 * @skb: buffer to test
2398 *
2399 * Return true if the skb has at least one frag that might be modified
2400 * by an external entity (as in vmsplice()/sendfile())
2401 */
2402static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2403{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00002404 return skb_is_nonlinear(skb) &&
2405 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002406}
2407
2408/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07002409 * skb_linearize_cow - make sure skb is linear and writable
2410 * @skb: buffer to process
2411 *
2412 * If there is no free memory -ENOMEM is returned, otherwise zero
2413 * is returned and the old skb data released.
2414 */
2415static inline int skb_linearize_cow(struct sk_buff *skb)
2416{
2417 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2418 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002419}
2420
2421/**
2422 * skb_postpull_rcsum - update checksum for received skb after pull
2423 * @skb: buffer to update
2424 * @start: start of data before pull
2425 * @len: length of data pulled
2426 *
2427 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07002428 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2429 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002430 */
2431
2432static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f2006-03-20 22:43:56 -08002433 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002434{
Patrick McHardy84fa7932006-08-29 16:44:56 -07002435 if (skb->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002436 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2437}
2438
Herbert Xucbb042f2006-03-20 22:43:56 -08002439unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2440
David S. Miller7ce5a272013-12-02 17:26:05 -05002441/**
2442 * pskb_trim_rcsum - trim received skb and update checksum
2443 * @skb: buffer to trim
2444 * @len: new length
2445 *
2446 * This is exactly the same as pskb_trim except that it ensures the
2447 * checksum of received packets are still valid after the operation.
2448 */
2449
2450static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2451{
2452 if (likely(len >= skb->len))
2453 return 0;
2454 if (skb->ip_summed == CHECKSUM_COMPLETE)
2455 skb->ip_summed = CHECKSUM_NONE;
2456 return __pskb_trim(skb, len);
2457}
2458
Linus Torvalds1da177e2005-04-16 15:20:36 -07002459#define skb_queue_walk(queue, skb) \
2460 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002461 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462 skb = skb->next)
2463
James Chapman46f89142007-04-30 00:07:31 -07002464#define skb_queue_walk_safe(queue, skb, tmp) \
2465 for (skb = (queue)->next, tmp = skb->next; \
2466 skb != (struct sk_buff *)(queue); \
2467 skb = tmp, tmp = skb->next)
2468
David S. Miller1164f522008-09-23 00:49:44 -07002469#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002470 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07002471 skb = skb->next)
2472
2473#define skb_queue_walk_from_safe(queue, skb, tmp) \
2474 for (tmp = skb->next; \
2475 skb != (struct sk_buff *)(queue); \
2476 skb = tmp, tmp = skb->next)
2477
Stephen Hemminger300ce172005-10-30 13:47:34 -08002478#define skb_queue_reverse_walk(queue, skb) \
2479 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002480 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08002481 skb = skb->prev)
2482
David S. Miller686a2952011-01-20 22:47:32 -08002483#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2484 for (skb = (queue)->prev, tmp = skb->prev; \
2485 skb != (struct sk_buff *)(queue); \
2486 skb = tmp, tmp = skb->prev)
2487
2488#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2489 for (tmp = skb->prev; \
2490 skb != (struct sk_buff *)(queue); \
2491 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002492
David S. Miller21dc3302010-08-23 00:13:46 -07002493static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07002494{
2495 return skb_shinfo(skb)->frag_list != NULL;
2496}
2497
2498static inline void skb_frag_list_init(struct sk_buff *skb)
2499{
2500 skb_shinfo(skb)->frag_list = NULL;
2501}
2502
2503static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag)
2504{
2505 frag->next = skb_shinfo(skb)->frag_list;
2506 skb_shinfo(skb)->frag_list = frag;
2507}
2508
2509#define skb_walk_frags(skb, iter) \
2510 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2511
Joe Perches7965bd42013-09-26 14:48:15 -07002512struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2513 int *peeked, int *off, int *err);
2514struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2515 int *err);
2516unsigned int datagram_poll(struct file *file, struct socket *sock,
2517 struct poll_table_struct *wait);
2518int skb_copy_datagram_iovec(const struct sk_buff *from, int offset,
2519 struct iovec *to, int size);
2520int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, int hlen,
2521 struct iovec *iov);
2522int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset,
2523 const struct iovec *from, int from_offset,
2524 int len);
2525int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *frm,
2526 int offset, size_t count);
2527int skb_copy_datagram_const_iovec(const struct sk_buff *from, int offset,
2528 const struct iovec *to, int to_offset,
2529 int size);
2530void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2531void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2532int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07002533int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2534int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2535__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2536 int len, __wsum csum);
2537int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
2538 struct pipe_inode_info *pipe, unsigned int len,
2539 unsigned int flags);
2540void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01002541unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002542int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2543 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07002544void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2545int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2546void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01002547unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -07002548struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002549
Daniel Borkmann2817a332013-10-30 11:50:51 +01002550struct skb_checksum_ops {
2551 __wsum (*update)(const void *mem, int len, __wsum wsum);
2552 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2553};
2554
2555__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2556 __wsum csum, const struct skb_checksum_ops *ops);
2557__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2558 __wsum csum);
2559
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560static inline void *skb_header_pointer(const struct sk_buff *skb, int offset,
2561 int len, void *buffer)
2562{
2563 int hlen = skb_headlen(skb);
2564
Patrick McHardy55820ee2005-07-05 14:08:10 -07002565 if (hlen - offset >= len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566 return skb->data + offset;
2567
2568 if (skb_copy_bits(skb, offset, buffer, len) < 0)
2569 return NULL;
2570
2571 return buffer;
2572}
2573
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01002574/**
2575 * skb_needs_linearize - check if we need to linearize a given skb
2576 * depending on the given device features.
2577 * @skb: socket buffer to check
2578 * @features: net device features
2579 *
2580 * Returns true if either:
2581 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2582 * 2. skb is fragmented and the device does not support SG.
2583 */
2584static inline bool skb_needs_linearize(struct sk_buff *skb,
2585 netdev_features_t features)
2586{
2587 return skb_is_nonlinear(skb) &&
2588 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2589 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2590}
2591
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002592static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2593 void *to,
2594 const unsigned int len)
2595{
2596 memcpy(to, skb->data, len);
2597}
2598
2599static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2600 const int offset, void *to,
2601 const unsigned int len)
2602{
2603 memcpy(to, skb->data + offset, len);
2604}
2605
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03002606static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2607 const void *from,
2608 const unsigned int len)
2609{
2610 memcpy(skb->data, from, len);
2611}
2612
2613static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2614 const int offset,
2615 const void *from,
2616 const unsigned int len)
2617{
2618 memcpy(skb->data + offset, from, len);
2619}
2620
Joe Perches7965bd42013-09-26 14:48:15 -07002621void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002622
Patrick Ohlyac45f602009-02-12 05:03:37 +00002623static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2624{
2625 return skb->tstamp;
2626}
2627
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002628/**
2629 * skb_get_timestamp - get timestamp from a skb
2630 * @skb: skb to get stamp from
2631 * @stamp: pointer to struct timeval to store stamp in
2632 *
2633 * Timestamps are stored in the skb as offsets to a base timestamp.
2634 * This function converts the offset back to a struct timeval and stores
2635 * it in stamp.
2636 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00002637static inline void skb_get_timestamp(const struct sk_buff *skb,
2638 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002639{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002640 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002641}
2642
Patrick Ohlyac45f602009-02-12 05:03:37 +00002643static inline void skb_get_timestampns(const struct sk_buff *skb,
2644 struct timespec *stamp)
2645{
2646 *stamp = ktime_to_timespec(skb->tstamp);
2647}
2648
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002649static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002650{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002651 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002652}
2653
Stephen Hemminger164891a2007-04-23 22:26:16 -07002654static inline ktime_t net_timedelta(ktime_t t)
2655{
2656 return ktime_sub(ktime_get_real(), t);
2657}
2658
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07002659static inline ktime_t net_invalid_timestamp(void)
2660{
2661 return ktime_set(0, 0);
2662}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002663
Richard Cochranc1f19b52010-07-17 08:49:36 +00002664#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2665
Joe Perches7965bd42013-09-26 14:48:15 -07002666void skb_clone_tx_timestamp(struct sk_buff *skb);
2667bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00002668
2669#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2670
2671static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2672{
2673}
2674
2675static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2676{
2677 return false;
2678}
2679
2680#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2681
2682/**
2683 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2684 *
Richard Cochranda92b192011-10-21 00:49:15 +00002685 * PHY drivers may accept clones of transmitted packets for
2686 * timestamping via their phy_driver.txtstamp method. These drivers
2687 * must call this function to return the skb back to the stack, with
2688 * or without a timestamp.
2689 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00002690 * @skb: clone of the the original outgoing packet
Richard Cochranda92b192011-10-21 00:49:15 +00002691 * @hwtstamps: hardware time stamps, may be NULL if not available
Richard Cochranc1f19b52010-07-17 08:49:36 +00002692 *
2693 */
2694void skb_complete_tx_timestamp(struct sk_buff *skb,
2695 struct skb_shared_hwtstamps *hwtstamps);
2696
Patrick Ohlyac45f602009-02-12 05:03:37 +00002697/**
2698 * skb_tstamp_tx - queue clone of skb with send time stamps
2699 * @orig_skb: the original outgoing packet
2700 * @hwtstamps: hardware time stamps, may be NULL if not available
2701 *
2702 * If the skb has a socket associated, then this function clones the
2703 * skb (thus sharing the actual data and optional structures), stores
2704 * the optional hardware time stamping information (if non NULL) or
2705 * generates a software time stamp (otherwise), then queues the clone
2706 * to the error queue of the socket. Errors are silently ignored.
2707 */
Joe Perches7965bd42013-09-26 14:48:15 -07002708void skb_tstamp_tx(struct sk_buff *orig_skb,
2709 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00002710
Richard Cochran4507a712010-07-17 08:48:28 +00002711static inline void sw_tx_timestamp(struct sk_buff *skb)
2712{
Oliver Hartkopp2244d072010-08-17 08:59:14 +00002713 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2714 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
Richard Cochran4507a712010-07-17 08:48:28 +00002715 skb_tstamp_tx(skb, NULL);
2716}
2717
2718/**
2719 * skb_tx_timestamp() - Driver hook for transmit timestamping
2720 *
2721 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00002722 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00002723 *
David S. Miller73409f32013-12-27 13:04:33 -05002724 * Specifically, one should make absolutely sure that this function is
2725 * called before TX completion of this packet can trigger. Otherwise
2726 * the packet could potentially already be freed.
2727 *
Richard Cochran4507a712010-07-17 08:48:28 +00002728 * @skb: A socket buffer.
2729 */
2730static inline void skb_tx_timestamp(struct sk_buff *skb)
2731{
Richard Cochranc1f19b52010-07-17 08:49:36 +00002732 skb_clone_tx_timestamp(skb);
Richard Cochran4507a712010-07-17 08:48:28 +00002733 sw_tx_timestamp(skb);
2734}
2735
Johannes Berg6e3e9392011-11-09 10:15:42 +01002736/**
2737 * skb_complete_wifi_ack - deliver skb with wifi status
2738 *
2739 * @skb: the original outgoing packet
2740 * @acked: ack status
2741 *
2742 */
2743void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
2744
Joe Perches7965bd42013-09-26 14:48:15 -07002745__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
2746__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08002747
Herbert Xu60476372007-04-09 11:59:39 -07002748static inline int skb_csum_unnecessary(const struct sk_buff *skb)
2749{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07002750 return ((skb->ip_summed & CHECKSUM_UNNECESSARY) || skb->csum_valid);
Herbert Xu60476372007-04-09 11:59:39 -07002751}
2752
Herbert Xufb286bb2005-11-10 13:01:24 -08002753/**
2754 * skb_checksum_complete - Calculate checksum of an entire packet
2755 * @skb: packet to process
2756 *
2757 * This function calculates the checksum over the entire packet plus
2758 * the value of skb->csum. The latter can be used to supply the
2759 * checksum of a pseudo header as used by TCP/UDP. It returns the
2760 * checksum.
2761 *
2762 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
2763 * this function can be used to verify that checksum on received
2764 * packets. In that case the function should return zero if the
2765 * checksum is correct. In particular, this function will return zero
2766 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
2767 * hardware has already verified the correctness of the checksum.
2768 */
Al Viro4381ca32007-07-15 21:00:11 +01002769static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08002770{
Herbert Xu60476372007-04-09 11:59:39 -07002771 return skb_csum_unnecessary(skb) ?
2772 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08002773}
2774
Tom Herbert76ba0aa2014-05-02 16:29:18 -07002775/* Check if we need to perform checksum complete validation.
2776 *
2777 * Returns true if checksum complete is needed, false otherwise
2778 * (either checksum is unnecessary or zero checksum is allowed).
2779 */
2780static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
2781 bool zero_okay,
2782 __sum16 check)
2783{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07002784 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
2785 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07002786 return false;
2787 }
2788
2789 return true;
2790}
2791
2792/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
2793 * in checksum_init.
2794 */
2795#define CHECKSUM_BREAK 76
2796
2797/* Validate (init) checksum based on checksum complete.
2798 *
2799 * Return values:
2800 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
2801 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
2802 * checksum is stored in skb->csum for use in __skb_checksum_complete
2803 * non-zero: value of invalid checksum
2804 *
2805 */
2806static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
2807 bool complete,
2808 __wsum psum)
2809{
2810 if (skb->ip_summed == CHECKSUM_COMPLETE) {
2811 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07002812 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07002813 return 0;
2814 }
2815 }
2816
2817 skb->csum = psum;
2818
Tom Herbert5d0c2b92014-06-10 18:54:13 -07002819 if (complete || skb->len <= CHECKSUM_BREAK) {
2820 __sum16 csum;
2821
2822 csum = __skb_checksum_complete(skb);
2823 skb->csum_valid = !csum;
2824 return csum;
2825 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07002826
2827 return 0;
2828}
2829
2830static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
2831{
2832 return 0;
2833}
2834
2835/* Perform checksum validate (init). Note that this is a macro since we only
2836 * want to calculate the pseudo header which is an input function if necessary.
2837 * First we try to validate without any computation (checksum unnecessary) and
2838 * then calculate based on checksum complete calling the function to compute
2839 * pseudo header.
2840 *
2841 * Return values:
2842 * 0: checksum is validated or try to in skb_checksum_complete
2843 * non-zero: value of invalid checksum
2844 */
2845#define __skb_checksum_validate(skb, proto, complete, \
2846 zero_okay, check, compute_pseudo) \
2847({ \
2848 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07002849 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07002850 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
2851 __ret = __skb_checksum_validate_complete(skb, \
2852 complete, compute_pseudo(skb, proto)); \
2853 __ret; \
2854})
2855
2856#define skb_checksum_init(skb, proto, compute_pseudo) \
2857 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
2858
2859#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
2860 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
2861
2862#define skb_checksum_validate(skb, proto, compute_pseudo) \
2863 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
2864
2865#define skb_checksum_validate_zero_check(skb, proto, check, \
2866 compute_pseudo) \
2867 __skb_checksum_validate_(skb, proto, true, true, check, compute_pseudo)
2868
2869#define skb_checksum_simple_validate(skb) \
2870 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
2871
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07002872#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07002873void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002874static inline void nf_conntrack_put(struct nf_conntrack *nfct)
2875{
2876 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07002877 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002878}
2879static inline void nf_conntrack_get(struct nf_conntrack *nfct)
2880{
2881 if (nfct)
2882 atomic_inc(&nfct->use);
2883}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01002884#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885#ifdef CONFIG_BRIDGE_NETFILTER
2886static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
2887{
2888 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
2889 kfree(nf_bridge);
2890}
2891static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
2892{
2893 if (nf_bridge)
2894 atomic_inc(&nf_bridge->use);
2895}
2896#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08002897static inline void nf_reset(struct sk_buff *skb)
2898{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07002899#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Patrick McHardya193a4a2006-03-20 19:23:05 -08002900 nf_conntrack_put(skb->nfct);
2901 skb->nfct = NULL;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01002902#endif
Patrick McHardya193a4a2006-03-20 19:23:05 -08002903#ifdef CONFIG_BRIDGE_NETFILTER
2904 nf_bridge_put(skb->nf_bridge);
2905 skb->nf_bridge = NULL;
2906#endif
2907}
2908
Patrick McHardy124dff02013-04-05 20:42:05 +02002909static inline void nf_reset_trace(struct sk_buff *skb)
2910{
Florian Westphal478b3602014-02-15 23:48:45 +01002911#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00002912 skb->nf_trace = 0;
2913#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07002914}
2915
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07002916/* Note: This doesn't put any conntrack and bridge info in dst. */
2917static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2918{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07002919#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07002920 dst->nfct = src->nfct;
2921 nf_conntrack_get(src->nfct);
2922 dst->nfctinfo = src->nfctinfo;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01002923#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07002924#ifdef CONFIG_BRIDGE_NETFILTER
2925 dst->nf_bridge = src->nf_bridge;
2926 nf_bridge_get(src->nf_bridge);
2927#endif
Florian Westphal478b3602014-02-15 23:48:45 +01002928#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
2929 dst->nf_trace = src->nf_trace;
2930#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07002931}
2932
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07002933static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
2934{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07002935#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07002936 nf_conntrack_put(dst->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01002937#endif
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07002938#ifdef CONFIG_BRIDGE_NETFILTER
2939 nf_bridge_put(dst->nf_bridge);
2940#endif
2941 __nf_copy(dst, src);
2942}
2943
James Morris984bc162006-06-09 00:29:17 -07002944#ifdef CONFIG_NETWORK_SECMARK
2945static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2946{
2947 to->secmark = from->secmark;
2948}
2949
2950static inline void skb_init_secmark(struct sk_buff *skb)
2951{
2952 skb->secmark = 0;
2953}
2954#else
2955static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
2956{ }
2957
2958static inline void skb_init_secmark(struct sk_buff *skb)
2959{ }
2960#endif
2961
Eric W. Biederman574f7192014-04-01 12:20:24 -07002962static inline bool skb_irq_freeable(const struct sk_buff *skb)
2963{
2964 return !skb->destructor &&
2965#if IS_ENABLED(CONFIG_XFRM)
2966 !skb->sp &&
2967#endif
2968#if IS_ENABLED(CONFIG_NF_CONNTRACK)
2969 !skb->nfct &&
2970#endif
2971 !skb->_skb_refdst &&
2972 !skb_has_frag_list(skb);
2973}
2974
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07002975static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
2976{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07002977 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07002978}
2979
Stephen Hemminger92477442009-03-21 13:39:26 -07002980static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07002981{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07002982 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07002983}
2984
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07002985static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
2986{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07002987 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07002988}
2989
David S. Millerd5a9e242009-01-27 16:22:11 -08002990static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
2991{
2992 skb->queue_mapping = rx_queue + 1;
2993}
2994
Stephen Hemminger92477442009-03-21 13:39:26 -07002995static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08002996{
2997 return skb->queue_mapping - 1;
2998}
2999
Stephen Hemminger92477442009-03-21 13:39:26 -07003000static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003001{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003002 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003003}
3004
Tom Herbert0e001612014-07-01 21:32:27 -07003005u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb,
Joe Perches7965bd42013-09-26 14:48:15 -07003006 unsigned int num_tx_queues);
Stephen Hemminger92477442009-03-21 13:39:26 -07003007
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003008static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3009{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003010#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003011 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003012#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003013 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003014#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003015}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003016
Pravin B Shelar68c33162013-02-14 14:02:41 +00003017/* Keeps track of mac header offset relative to skb->head.
3018 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3019 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003020 * tunnel skb it points to outer mac header.
3021 * Keeps track of level of encapsulation of network headers.
3022 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003023struct skb_gso_cb {
Eric Dumazet3347c962013-10-19 11:42:56 -07003024 int mac_offset;
3025 int encap_level;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003026 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003027};
3028#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3029
3030static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3031{
3032 return (skb_mac_header(inner_skb) - inner_skb->head) -
3033 SKB_GSO_CB(inner_skb)->mac_offset;
3034}
3035
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003036static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3037{
3038 int new_headroom, headroom;
3039 int ret;
3040
3041 headroom = skb_headroom(skb);
3042 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3043 if (ret)
3044 return ret;
3045
3046 new_headroom = skb_headroom(skb);
3047 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3048 return 0;
3049}
3050
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003051/* Compute the checksum for a gso segment. First compute the checksum value
3052 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3053 * then add in skb->csum (checksum from csum_start to end of packet).
3054 * skb->csum and csum_start are then updated to reflect the checksum of the
3055 * resultant packet starting from the transport header-- the resultant checksum
3056 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3057 * header.
3058 */
3059static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3060{
3061 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
3062 skb_transport_offset(skb);
3063 __u16 csum;
3064
3065 csum = csum_fold(csum_partial(skb_transport_header(skb),
3066 plen, skb->csum));
3067 skb->csum = res;
3068 SKB_GSO_CB(skb)->csum_start -= plen;
3069
3070 return csum;
3071}
3072
David S. Millerbdcc0922012-03-07 20:53:36 -05003073static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07003074{
3075 return skb_shinfo(skb)->gso_size;
3076}
3077
Eric Dumazet36a8f392013-09-29 01:21:32 -07003078/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05003079static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02003080{
3081 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3082}
3083
Joe Perches7965bd42013-09-26 14:48:15 -07003084void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07003085
3086static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3087{
3088 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3089 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003090 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3091
Alexander Duyckb78462e2010-06-02 12:24:37 +00003092 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3093 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07003094 __skb_warn_lro_forwarding(skb);
3095 return true;
3096 }
3097 return false;
3098}
3099
Herbert Xu35fc92a2007-03-26 23:22:20 -07003100static inline void skb_forward_csum(struct sk_buff *skb)
3101{
3102 /* Unfortunately we don't support this one. Any brave souls? */
3103 if (skb->ip_summed == CHECKSUM_COMPLETE)
3104 skb->ip_summed = CHECKSUM_NONE;
3105}
3106
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003107/**
3108 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3109 * @skb: skb to check
3110 *
3111 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3112 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3113 * use this helper, to document places where we make this assertion.
3114 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003115static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003116{
3117#ifdef DEBUG
3118 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3119#endif
3120}
3121
Rusty Russellf35d9d82008-02-04 23:49:54 -05003122bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00003123
Paul Durranted1f50c2014-01-09 10:02:46 +00003124int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
3125
Daniel Borkmannf77668d2013-03-19 06:39:30 +00003126u32 __skb_get_poff(const struct sk_buff *skb);
3127
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00003128/**
3129 * skb_head_is_locked - Determine if the skb->head is locked down
3130 * @skb: skb to check
3131 *
3132 * The head on skbs build around a head frag can be removed if they are
3133 * not cloned. This function returns true if the skb head is locked down
3134 * due to either being allocated via kmalloc, or by being a clone with
3135 * multiple references to the head.
3136 */
3137static inline bool skb_head_is_locked(const struct sk_buff *skb)
3138{
3139 return !skb->head_frag || skb_cloned(skb);
3140}
Florian Westphalfe6cc552014-02-13 23:09:12 +01003141
3142/**
3143 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3144 *
3145 * @skb: GSO skb
3146 *
3147 * skb_gso_network_seglen is used to determine the real size of the
3148 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3149 *
3150 * The MAC/L2 header is not accounted for.
3151 */
3152static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3153{
3154 unsigned int hdr_len = skb_transport_header(skb) -
3155 skb_network_header(skb);
3156 return hdr_len + skb_gso_transport_seglen(skb);
3157}
Linus Torvalds1da177e2005-04-16 15:20:36 -07003158#endif /* __KERNEL__ */
3159#endif /* _LINUX_SKBUFF_H */