blob: 8a697c673b58161140763aa686a7d589c052d35c [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Definitions for the 'struct sk_buff' memory handlers.
3 *
4 * Authors:
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14#ifndef _LINUX_SKBUFF_H
15#define _LINUX_SKBUFF_H
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/kernel.h>
Vegard Nossumfe55f6d2008-08-30 12:16:35 +020018#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/compiler.h>
20#include <linux/time.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050021#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/cache.h>
Eric Dumazet56b17422014-11-03 08:19:53 -080023#include <linux/rbtree.h>
David S. Miller51f3d022014-11-05 16:46:40 -050024#include <linux/socket.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
Arun Sharma600634972011-07-26 16:09:06 -070026#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <asm/types.h>
28#include <linux/spinlock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/net.h>
Thomas Graf3fc7e8a2005-06-23 21:00:17 -070030#include <linux/textsearch.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <net/checksum.h>
Al Viroa80958f2006-12-04 20:41:19 +000032#include <linux/rcupdate.h>
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -070033#include <linux/hrtimer.h>
Ian Campbell131ea662011-08-19 06:25:00 +000034#include <linux/dma-mapping.h>
Michał Mirosławc8f44af2011-11-15 15:29:55 +000035#include <linux/netdev_features.h>
Eric Dumazet363ec39232014-02-26 14:02:11 -080036#include <linux/sched.h>
Jiri Pirko1bd758e2015-05-12 14:56:07 +020037#include <net/flow_dissector.h>
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +020038#include <linux/splice.h>
Bernhard Thaler72b31f72015-05-30 15:27:40 +020039#include <linux/in6.h>
Tom Herbertf70ea012015-07-31 16:52:10 -070040#include <net/flow.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010042/* A. Checksumming of received packets by device.
43 *
44 * CHECKSUM_NONE:
45 *
46 * Device failed to checksum this packet e.g. due to lack of capabilities.
47 * The packet contains full (though not verified) checksum in packet but
48 * not in skb->csum. Thus, skb->csum is undefined in this case.
49 *
50 * CHECKSUM_UNNECESSARY:
51 *
52 * The hardware you're dealing with doesn't calculate the full checksum
53 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
Tom Herbert77cffe22014-08-27 21:26:46 -070054 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
55 * if their checksums are okay. skb->csum is still undefined in this case
56 * though. It is a bad option, but, unfortunately, nowadays most vendors do
57 * this. Apparently with the secret goal to sell you new devices, when you
58 * will add new protocol to your host, f.e. IPv6 8)
59 *
60 * CHECKSUM_UNNECESSARY is applicable to following protocols:
61 * TCP: IPv6 and IPv4.
62 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
63 * zero UDP checksum for either IPv4 or IPv6, the networking stack
64 * may perform further validation in this case.
65 * GRE: only if the checksum is present in the header.
66 * SCTP: indicates the CRC in SCTP header has been validated.
67 *
68 * skb->csum_level indicates the number of consecutive checksums found in
69 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
70 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
71 * and a device is able to verify the checksums for UDP (possibly zero),
72 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
73 * two. If the device were only able to verify the UDP checksum and not
74 * GRE, either because it doesn't support GRE checksum of because GRE
75 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
76 * not considered in this case).
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010077 *
78 * CHECKSUM_COMPLETE:
79 *
80 * This is the most generic way. The device supplied checksum of the _whole_
81 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
82 * hardware doesn't need to parse L3/L4 headers to implement this.
83 *
84 * Note: Even if device supports only some protocols, but is able to produce
85 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
86 *
87 * CHECKSUM_PARTIAL:
88 *
Tom Herbert6edec0e2015-02-10 16:30:28 -080089 * A checksum is set up to be offloaded to a device as described in the
90 * output description for CHECKSUM_PARTIAL. This may occur on a packet
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010091 * received directly from another Linux OS, e.g., a virtualized Linux kernel
Tom Herbert6edec0e2015-02-10 16:30:28 -080092 * on the same host, or it may be set in the input path in GRO or remote
93 * checksum offload. For the purposes of checksum verification, the checksum
94 * referred to by skb->csum_start + skb->csum_offset and any preceding
95 * checksums in the packet are considered verified. Any checksums in the
96 * packet that are after the checksum being offloaded are not considered to
97 * be verified.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010098 *
99 * B. Checksumming on output.
100 *
101 * CHECKSUM_NONE:
102 *
103 * The skb was already checksummed by the protocol, or a checksum is not
104 * required.
105 *
106 * CHECKSUM_PARTIAL:
107 *
108 * The device is required to checksum the packet as seen by hard_start_xmit()
109 * from skb->csum_start up to the end, and to record/write the checksum at
110 * offset skb->csum_start + skb->csum_offset.
111 *
112 * The device must show its capabilities in dev->features, set up at device
113 * setup time, e.g. netdev_features.h:
114 *
115 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything.
116 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over
117 * IPv4. Sigh. Vendors like this way for an unknown reason.
118 * Though, see comment above about CHECKSUM_UNNECESSARY. 8)
119 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead.
120 * NETIF_F_... - Well, you get the picture.
121 *
122 * CHECKSUM_UNNECESSARY:
123 *
124 * Normally, the device will do per protocol specific checksumming. Protocol
125 * implementations that do not want the NIC to perform the checksum
126 * calculation should use this flag in their outgoing skbs.
127 *
128 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC
129 * offload. Correspondingly, the FCoE protocol driver
130 * stack should use CHECKSUM_UNNECESSARY.
131 *
132 * Any questions? No questions, good. --ANK
133 */
134
Herbert Xu60476372007-04-09 11:59:39 -0700135/* Don't change this without changing skb_csum_unnecessary! */
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100136#define CHECKSUM_NONE 0
137#define CHECKSUM_UNNECESSARY 1
138#define CHECKSUM_COMPLETE 2
139#define CHECKSUM_PARTIAL 3
Linus Torvalds1da177e2005-04-16 15:20:36 -0700140
Tom Herbert77cffe22014-08-27 21:26:46 -0700141/* Maximum value in skb->csum_level */
142#define SKB_MAX_CSUM_LEVEL 3
143
Tobias Klauser0bec8c82014-07-22 12:06:23 +0200144#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
David S. Millerfc910a22007-03-25 20:27:59 -0700145#define SKB_WITH_OVERHEAD(X) \
Herbert Xudeea84b2007-10-21 16:27:46 -0700146 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
David S. Millerfc910a22007-03-25 20:27:59 -0700147#define SKB_MAX_ORDER(X, ORDER) \
148 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700149#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
150#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
151
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000152/* return minimum truesize of one skb containing X bytes of data */
153#define SKB_TRUESIZE(X) ((X) + \
154 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
155 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
156
Linus Torvalds1da177e2005-04-16 15:20:36 -0700157struct net_device;
David Howells716ea3a2007-04-02 20:19:53 -0700158struct scatterlist;
Jens Axboe9c55e012007-11-06 23:30:13 -0800159struct pipe_inode_info;
Herbert Xua8f820aa2014-11-07 21:22:22 +0800160struct iov_iter;
Alexander Duyckfd11a832014-12-09 19:40:49 -0800161struct napi_struct;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700162
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700163#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700164struct nf_conntrack {
165 atomic_t use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700166};
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700167#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700168
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200169#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700170struct nf_bridge_info {
Eric Dumazetbf1ac5c2012-04-18 23:19:25 +0000171 atomic_t use;
Florian Westphal3eaf4022015-04-02 14:31:44 +0200172 enum {
173 BRNF_PROTO_UNCHANGED,
174 BRNF_PROTO_8021Q,
175 BRNF_PROTO_PPPOE
Florian Westphal7fb48c52015-05-03 22:05:28 +0200176 } orig_proto:8;
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200177 u8 pkt_otherhost:1;
178 u8 in_prerouting:1;
179 u8 bridged_dnat:1;
Bernhard Thaler411ffb42015-05-30 15:28:28 +0200180 __u16 frag_max_size;
Eric Dumazetbf1ac5c2012-04-18 23:19:25 +0000181 struct net_device *physindev;
Florian Westphal7fb48c52015-05-03 22:05:28 +0200182 union {
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200183 /* prerouting: detect dnat in orig/reply direction */
Bernhard Thaler72b31f72015-05-30 15:27:40 +0200184 __be32 ipv4_daddr;
185 struct in6_addr ipv6_daddr;
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200186
187 /* after prerouting + nat detected: store original source
188 * mac since neigh resolution overwrites it, only used while
189 * skb is out in neigh layer.
190 */
191 char neigh_header[8];
192
193 /* always valid & non-NULL from FORWARD on, for physdev match */
194 struct net_device *physoutdev;
Bernhard Thaler72b31f72015-05-30 15:27:40 +0200195 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700196};
197#endif
198
Linus Torvalds1da177e2005-04-16 15:20:36 -0700199struct sk_buff_head {
200 /* These two members must be first. */
201 struct sk_buff *next;
202 struct sk_buff *prev;
203
204 __u32 qlen;
205 spinlock_t lock;
206};
207
208struct sk_buff;
209
Ian Campbell9d4dde52011-12-22 23:39:14 +0000210/* To allow 64K frame to be packed as single skb without frag_list we
211 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
212 * buffers which do not start on a page boundary.
213 *
214 * Since GRO uses frags we allocate at least 16 regardless of page
215 * size.
Anton Blancharda715dea2011-03-27 14:57:26 +0000216 */
Ian Campbell9d4dde52011-12-22 23:39:14 +0000217#if (65536/PAGE_SIZE + 1) < 16
David S. Millereec00952011-03-29 23:34:08 -0700218#define MAX_SKB_FRAGS 16UL
Anton Blancharda715dea2011-03-27 14:57:26 +0000219#else
Ian Campbell9d4dde52011-12-22 23:39:14 +0000220#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
Anton Blancharda715dea2011-03-27 14:57:26 +0000221#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700222
223typedef struct skb_frag_struct skb_frag_t;
224
225struct skb_frag_struct {
Ian Campbella8605c62011-10-19 23:01:49 +0000226 struct {
227 struct page *p;
228 } page;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000229#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
David S. Millera309bb02007-07-30 18:47:03 -0700230 __u32 page_offset;
231 __u32 size;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000232#else
233 __u16 page_offset;
234 __u16 size;
235#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700236};
237
Eric Dumazet9e903e02011-10-18 21:00:24 +0000238static inline unsigned int skb_frag_size(const skb_frag_t *frag)
239{
240 return frag->size;
241}
242
243static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
244{
245 frag->size = size;
246}
247
248static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
249{
250 frag->size += delta;
251}
252
253static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
254{
255 frag->size -= delta;
256}
257
Patrick Ohlyac45f602009-02-12 05:03:37 +0000258#define HAVE_HW_TIME_STAMP
259
260/**
Randy Dunlapd3a21be2009-03-02 03:15:58 -0800261 * struct skb_shared_hwtstamps - hardware time stamps
Patrick Ohlyac45f602009-02-12 05:03:37 +0000262 * @hwtstamp: hardware time stamp transformed into duration
263 * since arbitrary point in time
Patrick Ohlyac45f602009-02-12 05:03:37 +0000264 *
265 * Software time stamps generated by ktime_get_real() are stored in
Willem de Bruijn4d276eb2014-07-25 18:01:32 -0400266 * skb->tstamp.
Patrick Ohlyac45f602009-02-12 05:03:37 +0000267 *
268 * hwtstamps can only be compared against other hwtstamps from
269 * the same device.
270 *
271 * This structure is attached to packets as part of the
272 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
273 */
274struct skb_shared_hwtstamps {
275 ktime_t hwtstamp;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000276};
277
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000278/* Definitions for tx_flags in struct skb_shared_info */
279enum {
280 /* generate hardware time stamp */
281 SKBTX_HW_TSTAMP = 1 << 0,
282
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400283 /* generate software time stamp when queueing packet to NIC */
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000284 SKBTX_SW_TSTAMP = 1 << 1,
285
286 /* device driver is going to provide hardware time stamp */
287 SKBTX_IN_PROGRESS = 1 << 2,
288
Shirley Maa6686f22011-07-06 12:22:12 +0000289 /* device driver supports TX zero-copy buffers */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000290 SKBTX_DEV_ZEROCOPY = 1 << 3,
Johannes Berg6e3e9392011-11-09 10:15:42 +0100291
292 /* generate wifi status information (where possible) */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000293 SKBTX_WIFI_STATUS = 1 << 4,
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000294
295 /* This indicates at least one fragment might be overwritten
296 * (as in vmsplice(), sendfile() ...)
297 * If we need to compute a TX checksum, we'll need to copy
298 * all frags to avoid possible bad checksum
299 */
300 SKBTX_SHARED_FRAG = 1 << 5,
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400301
302 /* generate software time stamp when entering packet scheduling */
303 SKBTX_SCHED_TSTAMP = 1 << 6,
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400304
305 /* generate software timestamp on peer data acknowledgment */
306 SKBTX_ACK_TSTAMP = 1 << 7,
Shirley Maa6686f22011-07-06 12:22:12 +0000307};
308
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400309#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
310 SKBTX_SCHED_TSTAMP | \
311 SKBTX_ACK_TSTAMP)
Willem de Bruijnf24b9be2014-08-04 22:11:45 -0400312#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
313
Shirley Maa6686f22011-07-06 12:22:12 +0000314/*
315 * The callback notifies userspace to release buffers when skb DMA is done in
316 * lower device, the skb last reference should be 0 when calling this.
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000317 * The zerocopy_success argument is true if zero copy transmit occurred,
318 * false on data copy or out of memory error caused by data copy attempt.
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000319 * The ctx field is used to track device context.
320 * The desc field is used to track userspace buffer index.
Shirley Maa6686f22011-07-06 12:22:12 +0000321 */
322struct ubuf_info {
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000323 void (*callback)(struct ubuf_info *, bool zerocopy_success);
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000324 void *ctx;
Shirley Maa6686f22011-07-06 12:22:12 +0000325 unsigned long desc;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000326};
327
Linus Torvalds1da177e2005-04-16 15:20:36 -0700328/* This data is invariant across clones and lives at
329 * the end of the header data, ie. at skb->end.
330 */
331struct skb_shared_info {
Ian Campbell9f42f122012-01-05 07:13:39 +0000332 unsigned char nr_frags;
333 __u8 tx_flags;
Herbert Xu79671682006-06-22 02:40:14 -0700334 unsigned short gso_size;
335 /* Warning: this field is not always filled in (UFO)! */
336 unsigned short gso_segs;
337 unsigned short gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700338 struct sk_buff *frag_list;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000339 struct skb_shared_hwtstamps hwtstamps;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400340 u32 tskey;
Ian Campbell9f42f122012-01-05 07:13:39 +0000341 __be32 ip6_frag_id;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700342
343 /*
344 * Warning : all fields before dataref are cleared in __alloc_skb()
345 */
346 atomic_t dataref;
347
Johann Baudy69e3c752009-05-18 22:11:22 -0700348 /* Intermediate layers must ensure that destructor_arg
349 * remains valid until skb destructor */
350 void * destructor_arg;
Shirley Maa6686f22011-07-06 12:22:12 +0000351
Eric Dumazetfed66382010-07-22 19:09:08 +0000352 /* must be last field, see pskb_expand_head() */
353 skb_frag_t frags[MAX_SKB_FRAGS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700354};
355
356/* We divide dataref into two halves. The higher 16 bits hold references
357 * to the payload part of skb->data. The lower 16 bits hold references to
Patrick McHardy334a8132007-06-25 04:35:20 -0700358 * the entire skb->data. A clone of a headerless skb holds the length of
359 * the header in skb->hdr_len.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700360 *
361 * All users must obey the rule that the skb->data reference count must be
362 * greater than or equal to the payload reference count.
363 *
364 * Holding a reference to the payload part means that the user does not
365 * care about modifications to the header part of skb->data.
366 */
367#define SKB_DATAREF_SHIFT 16
368#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
369
David S. Millerd179cd12005-08-17 14:57:30 -0700370
371enum {
Vijay Subramanianc8753d52014-10-02 10:00:43 -0700372 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
373 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
374 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
David S. Millerd179cd12005-08-17 14:57:30 -0700375};
376
Herbert Xu79671682006-06-22 02:40:14 -0700377enum {
378 SKB_GSO_TCPV4 = 1 << 0,
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700379 SKB_GSO_UDP = 1 << 1,
Herbert Xu576a30e2006-06-27 13:22:38 -0700380
381 /* This indicates the skb is from an untrusted source. */
382 SKB_GSO_DODGY = 1 << 2,
Michael Chanb0da85372006-06-29 12:30:00 -0700383
384 /* This indicates the tcp segment has CWR set. */
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700385 SKB_GSO_TCP_ECN = 1 << 3,
386
387 SKB_GSO_TCPV6 = 1 << 4,
Chris Leech01d5b2f2009-02-27 14:06:49 -0800388
389 SKB_GSO_FCOE = 1 << 5,
Pravin B Shelar68c33162013-02-14 14:02:41 +0000390
391 SKB_GSO_GRE = 1 << 6,
Pravin B Shelar73136262013-03-07 13:21:51 +0000392
Tom Herbert4b282522014-06-14 23:23:52 -0700393 SKB_GSO_GRE_CSUM = 1 << 7,
Simon Horman0d89d202013-05-23 21:02:52 +0000394
Tom Herbert4b282522014-06-14 23:23:52 -0700395 SKB_GSO_IPIP = 1 << 8,
Eric Dumazetcb32f512013-10-19 11:42:57 -0700396
Tom Herbert4b282522014-06-14 23:23:52 -0700397 SKB_GSO_SIT = 1 << 9,
Eric Dumazet61c1db72013-10-20 20:47:30 -0700398
Tom Herbert4b282522014-06-14 23:23:52 -0700399 SKB_GSO_UDP_TUNNEL = 1 << 10,
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700400
401 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
Tom Herbert4749c092014-06-04 17:20:23 -0700402
Pravin B Shelar59b93b42014-11-05 15:27:48 -0800403 SKB_GSO_TUNNEL_REMCSUM = 1 << 12,
Herbert Xu79671682006-06-22 02:40:14 -0700404};
405
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700406#if BITS_PER_LONG > 32
407#define NET_SKBUFF_DATA_USES_OFFSET 1
408#endif
409
410#ifdef NET_SKBUFF_DATA_USES_OFFSET
411typedef unsigned int sk_buff_data_t;
412#else
413typedef unsigned char *sk_buff_data_t;
414#endif
415
Eric Dumazet363ec39232014-02-26 14:02:11 -0800416/**
417 * struct skb_mstamp - multi resolution time stamps
418 * @stamp_us: timestamp in us resolution
419 * @stamp_jiffies: timestamp in jiffies
420 */
421struct skb_mstamp {
422 union {
423 u64 v64;
424 struct {
425 u32 stamp_us;
426 u32 stamp_jiffies;
427 };
428 };
429};
430
431/**
432 * skb_mstamp_get - get current timestamp
433 * @cl: place to store timestamps
434 */
435static inline void skb_mstamp_get(struct skb_mstamp *cl)
436{
437 u64 val = local_clock();
438
439 do_div(val, NSEC_PER_USEC);
440 cl->stamp_us = (u32)val;
441 cl->stamp_jiffies = (u32)jiffies;
442}
443
444/**
445 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
446 * @t1: pointer to newest sample
447 * @t0: pointer to oldest sample
448 */
449static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
450 const struct skb_mstamp *t0)
451{
452 s32 delta_us = t1->stamp_us - t0->stamp_us;
453 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
454
455 /* If delta_us is negative, this might be because interval is too big,
456 * or local_clock() drift is too big : fallback using jiffies.
457 */
458 if (delta_us <= 0 ||
459 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
460
461 delta_us = jiffies_to_usecs(delta_jiffies);
462
463 return delta_us;
464}
465
466
Linus Torvalds1da177e2005-04-16 15:20:36 -0700467/**
468 * struct sk_buff - socket buffer
469 * @next: Next buffer in list
470 * @prev: Previous buffer in list
Eric Dumazet363ec39232014-02-26 14:02:11 -0800471 * @tstamp: Time we arrived/left
Eric Dumazet56b17422014-11-03 08:19:53 -0800472 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700473 * @sk: Socket we are owned by
Linus Torvalds1da177e2005-04-16 15:20:36 -0700474 * @dev: Device we arrived on/are leaving by
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700475 * @cb: Control buffer. Free for use by every layer. Put private vars here
Eric Dumazet7fee2262010-05-11 23:19:48 +0000476 * @_skb_refdst: destination entry (with norefcount bit)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700477 * @sp: the security path, used for xfrm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700478 * @len: Length of actual data
479 * @data_len: Data length
480 * @mac_len: Length of link layer header
Patrick McHardy334a8132007-06-25 04:35:20 -0700481 * @hdr_len: writable header length of cloned skb
Herbert Xu663ead32007-04-09 11:59:07 -0700482 * @csum: Checksum (must include start/offset pair)
483 * @csum_start: Offset from skb->head where checksumming should start
484 * @csum_offset: Offset from csum_start where checksum should be stored
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700485 * @priority: Packet queueing priority
WANG Cong60ff7462014-05-04 16:39:18 -0700486 * @ignore_df: allow local fragmentation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 * @cloned: Head may be cloned (check refcnt to be sure)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700488 * @ip_summed: Driver fed us an IP checksum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700489 * @nohdr: Payload reference only, must not modify header
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700490 * @nfctinfo: Relationship of this skb to the connection
Linus Torvalds1da177e2005-04-16 15:20:36 -0700491 * @pkt_type: Packet class
Randy Dunlapc83c2482005-10-18 22:07:41 -0700492 * @fclone: skbuff clone status
Randy Dunlapc83c2482005-10-18 22:07:41 -0700493 * @ipvs_property: skbuff is owned by ipvs
Randy Dunlap31729362008-02-18 20:52:13 -0800494 * @peeked: this packet has been seen already, so stats have been
495 * done for it, don't do them again
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700496 * @nf_trace: netfilter packet trace flag
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700497 * @protocol: Packet protocol from driver
498 * @destructor: Destruct function
499 * @nfct: Associated connection, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
Eric Dumazet8964be42009-11-20 15:35:04 -0800501 * @skb_iif: ifindex of device we arrived on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700502 * @tc_index: Traffic control index
503 * @tc_verd: traffic control verdict
Tom Herbert61b905d2014-03-24 15:34:47 -0700504 * @hash: the packet hash
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700505 * @queue_mapping: Queue mapping for multiqueue devices
David S. Miller0b725a22014-08-25 15:51:53 -0700506 * @xmit_more: More SKBs are pending for this queue
Randy Dunlap553a5672008-04-20 10:51:01 -0700507 * @ndisc_nodetype: router type (from link layer)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700508 * @ooo_okay: allow the mapping of a socket to a queue to be changed
Tom Herbert61b905d2014-03-24 15:34:47 -0700509 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
Changli Gao4ca24622011-08-19 07:26:44 -0700510 * ports.
Tom Herberta3b18dd2014-07-01 21:33:17 -0700511 * @sw_hash: indicates hash was computed in software stack
Johannes Berg6e3e9392011-11-09 10:15:42 +0100512 * @wifi_acked_valid: wifi_acked was set
513 * @wifi_acked: whether frame was acked on wifi or not
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000514 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
Eliezer Tamir06021292013-06-10 11:39:50 +0300515 * @napi_id: id of the NAPI struct this skb came from
James Morris984bc162006-06-09 00:29:17 -0700516 * @secmark: security marking
Scott Feldman0c4f6912015-07-18 18:24:48 -0700517 * @offload_fwd_mark: fwding offload mark
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700518 * @mark: Generic packet mark
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000519 * @vlan_proto: vlan encapsulation protocol
Patrick McHardy6aa895b2008-07-14 22:49:06 -0700520 * @vlan_tci: vlan tag control information
Simon Horman0d89d202013-05-23 21:02:52 +0000521 * @inner_protocol: Protocol (encapsulation)
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000522 * @inner_transport_header: Inner transport layer header (encapsulation)
523 * @inner_network_header: Network layer header (encapsulation)
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000524 * @inner_mac_header: Link layer header (encapsulation)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700525 * @transport_header: Transport layer header
526 * @network_header: Network layer header
527 * @mac_header: Link layer header
528 * @tail: Tail pointer
529 * @end: End pointer
530 * @head: Head of buffer
531 * @data: Data head pointer
532 * @truesize: Buffer size
533 * @users: User count - see {datagram,tcp}.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700534 */
535
536struct sk_buff {
Eric Dumazet363ec39232014-02-26 14:02:11 -0800537 union {
Eric Dumazet56b17422014-11-03 08:19:53 -0800538 struct {
539 /* These two members must be first. */
540 struct sk_buff *next;
541 struct sk_buff *prev;
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000542
Eric Dumazet56b17422014-11-03 08:19:53 -0800543 union {
544 ktime_t tstamp;
545 struct skb_mstamp skb_mstamp;
546 };
547 };
548 struct rb_node rbnode; /* used in netem & tcp stack */
549 };
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000550 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700551 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700552
Linus Torvalds1da177e2005-04-16 15:20:36 -0700553 /*
554 * This is the control buffer. It is free to use for every
555 * layer. Please put your private variables there. If you
556 * want to keep them across layers you have to do a skb_clone()
557 * first. This is owned by whoever has the skb queued ATM.
558 */
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000559 char cb[48] __aligned(8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560
Eric Dumazet7fee2262010-05-11 23:19:48 +0000561 unsigned long _skb_refdst;
Eric Dumazetb1937222014-09-28 22:18:47 -0700562 void (*destructor)(struct sk_buff *skb);
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000563#ifdef CONFIG_XFRM
564 struct sec_path *sp;
565#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700566#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
567 struct nf_conntrack *nfct;
568#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200569#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Eric Dumazetb1937222014-09-28 22:18:47 -0700570 struct nf_bridge_info *nf_bridge;
571#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 unsigned int len,
Patrick McHardy334a8132007-06-25 04:35:20 -0700573 data_len;
574 __u16 mac_len,
575 hdr_len;
Eric Dumazetb1937222014-09-28 22:18:47 -0700576
577 /* Following fields are _not_ copied in __copy_skb_header()
578 * Note that queue_mapping is here mostly to fill a hole.
579 */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200580 kmemcheck_bitfield_begin(flags1);
Eric Dumazetb1937222014-09-28 22:18:47 -0700581 __u16 queue_mapping;
582 __u8 cloned:1,
Harald Welte6869c4d2005-08-09 19:24:19 -0700583 nohdr:1,
Patrick McHardyb84f4cc2005-11-20 21:19:21 -0800584 fclone:2,
Herbert Xua59322b2007-12-05 01:53:40 -0800585 peeked:1,
Eric Dumazetb1937222014-09-28 22:18:47 -0700586 head_frag:1,
587 xmit_more:1;
588 /* one bit hole */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200589 kmemcheck_bitfield_end(flags1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590
Eric Dumazetb1937222014-09-28 22:18:47 -0700591 /* fields enclosed in headers_start/headers_end are copied
592 * using a single memcpy() in __copy_skb_header()
593 */
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700594 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700595 __u32 headers_start[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700596 /* public: */
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200597
598/* if you move pkt_type around you also must adapt those constants */
599#ifdef __BIG_ENDIAN_BITFIELD
600#define PKT_TYPE_MAX (7 << 5)
601#else
602#define PKT_TYPE_MAX 7
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100603#endif
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200604#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
605
606 __u8 __pkt_type_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700607 __u8 pkt_type:3;
608 __u8 pfmemalloc:1;
609 __u8 ignore_df:1;
610 __u8 nfctinfo:3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700611
Eric Dumazetb1937222014-09-28 22:18:47 -0700612 __u8 nf_trace:1;
613 __u8 ip_summed:2;
614 __u8 ooo_okay:1;
615 __u8 l4_hash:1;
616 __u8 sw_hash:1;
617 __u8 wifi_acked_valid:1;
618 __u8 wifi_acked:1;
619
620 __u8 no_fcs:1;
621 /* Indicates the inner headers are valid in the skbuff. */
622 __u8 encapsulation:1;
623 __u8 encap_hdr_csum:1;
624 __u8 csum_valid:1;
625 __u8 csum_complete_sw:1;
626 __u8 csum_level:2;
627 __u8 csum_bad:1;
628
629#ifdef CONFIG_IPV6_NDISC_NODETYPE
630 __u8 ndisc_nodetype:2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700631#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700632 __u8 ipvs_property:1;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700633 __u8 inner_protocol_type:1;
Tom Herberte585f232014-11-04 09:06:54 -0800634 __u8 remcsum_offload:1;
635 /* 3 or 5 bit hole */
Alexander Duyck4031ae62012-01-27 06:22:53 +0000636
Linus Torvalds1da177e2005-04-16 15:20:36 -0700637#ifdef CONFIG_NET_SCHED
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700638 __u16 tc_index; /* traffic control index */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700639#ifdef CONFIG_NET_CLS_ACT
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700640 __u16 tc_verd; /* traffic control verdict */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642#endif
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200643
Eric Dumazetb1937222014-09-28 22:18:47 -0700644 union {
645 __wsum csum;
646 struct {
647 __u16 csum_start;
648 __u16 csum_offset;
649 };
650 };
651 __u32 priority;
652 int skb_iif;
653 __u32 hash;
654 __be16 vlan_proto;
655 __u16 vlan_tci;
Eric Dumazet2bd82482015-02-03 23:48:24 -0800656#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
657 union {
658 unsigned int napi_id;
659 unsigned int sender_cpu;
660 };
Chris Leech97fc2f02006-05-23 17:55:33 -0700661#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700662 union {
James Morris984bc162006-06-09 00:29:17 -0700663#ifdef CONFIG_NETWORK_SECMARK
Scott Feldman0c4f6912015-07-18 18:24:48 -0700664 __u32 secmark;
James Morris984bc162006-06-09 00:29:17 -0700665#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700666#ifdef CONFIG_NET_SWITCHDEV
667 __u32 offload_fwd_mark;
668#endif
669 };
670
Neil Horman3b885782009-10-12 13:26:31 -0700671 union {
672 __u32 mark;
Eric Dumazet16fad692013-03-14 05:40:32 +0000673 __u32 reserved_tailroom;
Neil Horman3b885782009-10-12 13:26:31 -0700674 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700675
Tom Herbert8bce6d72014-09-29 20:22:29 -0700676 union {
677 __be16 inner_protocol;
678 __u8 inner_ipproto;
679 };
680
Simon Horman1a37e412013-05-23 21:02:51 +0000681 __u16 inner_transport_header;
682 __u16 inner_network_header;
683 __u16 inner_mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700684
685 __be16 protocol;
Simon Horman1a37e412013-05-23 21:02:51 +0000686 __u16 transport_header;
687 __u16 network_header;
688 __u16 mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700689
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700690 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700691 __u32 headers_end[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700692 /* public: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700693
Linus Torvalds1da177e2005-04-16 15:20:36 -0700694 /* These elements must be at the end, see alloc_skb() for details. */
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700695 sk_buff_data_t tail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700696 sk_buff_data_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700697 unsigned char *head,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700698 *data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700699 unsigned int truesize;
700 atomic_t users;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701};
702
703#ifdef __KERNEL__
704/*
705 * Handling routines are only of interest to the kernel
706 */
707#include <linux/slab.h>
708
Linus Torvalds1da177e2005-04-16 15:20:36 -0700709
Mel Gormanc93bdd02012-07-31 16:44:19 -0700710#define SKB_ALLOC_FCLONE 0x01
711#define SKB_ALLOC_RX 0x02
Alexander Duyckfd11a832014-12-09 19:40:49 -0800712#define SKB_ALLOC_NAPI 0x04
Mel Gormanc93bdd02012-07-31 16:44:19 -0700713
714/* Returns true if the skb was allocated from PFMEMALLOC reserves */
715static inline bool skb_pfmemalloc(const struct sk_buff *skb)
716{
717 return unlikely(skb->pfmemalloc);
718}
719
Eric Dumazet7fee2262010-05-11 23:19:48 +0000720/*
721 * skb might have a dst pointer attached, refcounted or not.
722 * _skb_refdst low order bit is set if refcount was _not_ taken
723 */
724#define SKB_DST_NOREF 1UL
725#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
726
727/**
728 * skb_dst - returns skb dst_entry
729 * @skb: buffer
730 *
731 * Returns skb dst_entry, regardless of reference taken or not.
732 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000733static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
734{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000735 /* If refdst was not refcounted, check we still are in a
736 * rcu_read_lock section
737 */
738 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
739 !rcu_read_lock_held() &&
740 !rcu_read_lock_bh_held());
741 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
Eric Dumazetadf30902009-06-02 05:19:30 +0000742}
743
Eric Dumazet7fee2262010-05-11 23:19:48 +0000744/**
745 * skb_dst_set - sets skb dst
746 * @skb: buffer
747 * @dst: dst entry
748 *
749 * Sets skb dst, assuming a reference was taken on dst and should
750 * be released by skb_dst_drop()
751 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000752static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
753{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000754 skb->_skb_refdst = (unsigned long)dst;
755}
756
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200757/**
758 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
759 * @skb: buffer
760 * @dst: dst entry
761 *
762 * Sets skb dst, assuming a reference was not taken on dst.
763 * If dst entry is cached, we do not take reference and dst_release
764 * will be avoided by refdst_drop. If dst entry is not cached, we take
765 * reference, so that last dst_release can destroy the dst immediately.
766 */
767static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
768{
Hannes Frederic Sowadbfc4fb2014-12-06 19:19:42 +0100769 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
770 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200771}
Eric Dumazet7fee2262010-05-11 23:19:48 +0000772
773/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300774 * skb_dst_is_noref - Test if skb dst isn't refcounted
Eric Dumazet7fee2262010-05-11 23:19:48 +0000775 * @skb: buffer
776 */
777static inline bool skb_dst_is_noref(const struct sk_buff *skb)
778{
779 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
Eric Dumazetadf30902009-06-02 05:19:30 +0000780}
781
Eric Dumazet511c3f92009-06-02 05:14:27 +0000782static inline struct rtable *skb_rtable(const struct sk_buff *skb)
783{
Eric Dumazetadf30902009-06-02 05:19:30 +0000784 return (struct rtable *)skb_dst(skb);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000785}
786
Joe Perches7965bd42013-09-26 14:48:15 -0700787void kfree_skb(struct sk_buff *skb);
788void kfree_skb_list(struct sk_buff *segs);
789void skb_tx_error(struct sk_buff *skb);
790void consume_skb(struct sk_buff *skb);
791void __kfree_skb(struct sk_buff *skb);
Eric Dumazetd7e88832012-04-30 08:10:34 +0000792extern struct kmem_cache *skbuff_head_cache;
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000793
Joe Perches7965bd42013-09-26 14:48:15 -0700794void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
795bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
796 bool *fragstolen, int *delta_truesize);
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000797
Joe Perches7965bd42013-09-26 14:48:15 -0700798struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
799 int node);
Eric Dumazet2ea2f622015-04-24 16:05:01 -0700800struct sk_buff *__build_skb(void *data, unsigned int frag_size);
Joe Perches7965bd42013-09-26 14:48:15 -0700801struct sk_buff *build_skb(void *data, unsigned int frag_size);
David S. Millerd179cd12005-08-17 14:57:30 -0700802static inline struct sk_buff *alloc_skb(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100803 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700804{
Eric Dumazet564824b2010-10-11 19:05:25 +0000805 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700806}
807
Eric Dumazet2e4e4412014-09-17 04:49:49 -0700808struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
809 unsigned long data_len,
810 int max_page_order,
811 int *errcode,
812 gfp_t gfp_mask);
813
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700814/* Layout of fast clones : [skb1][skb2][fclone_ref] */
815struct sk_buff_fclones {
816 struct sk_buff skb1;
817
818 struct sk_buff skb2;
819
820 atomic_t fclone_ref;
821};
822
823/**
824 * skb_fclone_busy - check if fclone is busy
825 * @skb: buffer
826 *
827 * Returns true is skb is a fast clone, and its clone is not freed.
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700828 * Some drivers call skb_orphan() in their ndo_start_xmit(),
829 * so we also check that this didnt happen.
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700830 */
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700831static inline bool skb_fclone_busy(const struct sock *sk,
832 const struct sk_buff *skb)
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700833{
834 const struct sk_buff_fclones *fclones;
835
836 fclones = container_of(skb, struct sk_buff_fclones, skb1);
837
838 return skb->fclone == SKB_FCLONE_ORIG &&
Eric Dumazet6ffe75eb2014-12-03 17:04:39 -0800839 atomic_read(&fclones->fclone_ref) > 1 &&
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700840 fclones->skb2.sk == sk;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700841}
842
David S. Millerd179cd12005-08-17 14:57:30 -0700843static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100844 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700845{
Mel Gormanc93bdd02012-07-31 16:44:19 -0700846 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700847}
848
Joe Perches7965bd42013-09-26 14:48:15 -0700849struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000850static inline struct sk_buff *alloc_skb_head(gfp_t priority)
851{
852 return __alloc_skb_head(priority, -1);
853}
854
Joe Perches7965bd42013-09-26 14:48:15 -0700855struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
856int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
857struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
858struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
Octavian Purdilabad93e92014-06-12 01:36:26 +0300859struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
860 gfp_t gfp_mask, bool fclone);
861static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
862 gfp_t gfp_mask)
863{
864 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
865}
Eric Dumazet117632e2011-12-03 21:39:53 +0000866
Joe Perches7965bd42013-09-26 14:48:15 -0700867int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
868struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
869 unsigned int headroom);
870struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
871 int newtailroom, gfp_t priority);
Fan Du25a91d82014-01-18 09:54:23 +0800872int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
873 int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -0700874int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
875 int len);
876int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
877int skb_pad(struct sk_buff *skb, int pad);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000878#define dev_kfree_skb(a) consume_skb(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700879
Joe Perches7965bd42013-09-26 14:48:15 -0700880int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
881 int getfrag(void *from, char *to, int offset,
882 int len, int odd, struct sk_buff *skb),
883 void *from, int length);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700884
Hannes Frederic Sowabe12a1f2015-05-21 16:59:58 +0200885int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
886 int offset, size_t size);
887
Eric Dumazetd94d9fe2009-11-04 09:50:58 -0800888struct skb_seq_state {
Thomas Graf677e90e2005-06-23 20:59:51 -0700889 __u32 lower_offset;
890 __u32 upper_offset;
891 __u32 frag_idx;
892 __u32 stepped_offset;
893 struct sk_buff *root_skb;
894 struct sk_buff *cur_skb;
895 __u8 *frag_data;
896};
897
Joe Perches7965bd42013-09-26 14:48:15 -0700898void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
899 unsigned int to, struct skb_seq_state *st);
900unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
901 struct skb_seq_state *st);
902void skb_abort_seq_read(struct skb_seq_state *st);
Thomas Graf677e90e2005-06-23 20:59:51 -0700903
Joe Perches7965bd42013-09-26 14:48:15 -0700904unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
Bojan Prtvar059a2442015-02-22 11:46:35 +0100905 unsigned int to, struct ts_config *config);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -0700906
Tom Herbert09323cc2013-12-15 22:16:19 -0800907/*
908 * Packet hash types specify the type of hash in skb_set_hash.
909 *
910 * Hash types refer to the protocol layer addresses which are used to
911 * construct a packet's hash. The hashes are used to differentiate or identify
912 * flows of the protocol layer for the hash type. Hash types are either
913 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
914 *
915 * Properties of hashes:
916 *
917 * 1) Two packets in different flows have different hash values
918 * 2) Two packets in the same flow should have the same hash value
919 *
920 * A hash at a higher layer is considered to be more specific. A driver should
921 * set the most specific hash possible.
922 *
923 * A driver cannot indicate a more specific hash than the layer at which a hash
924 * was computed. For instance an L3 hash cannot be set as an L4 hash.
925 *
926 * A driver may indicate a hash level which is less specific than the
927 * actual layer the hash was computed on. For instance, a hash computed
928 * at L4 may be considered an L3 hash. This should only be done if the
929 * driver can't unambiguously determine that the HW computed the hash at
930 * the higher layer. Note that the "should" in the second property above
931 * permits this.
932 */
933enum pkt_hash_types {
934 PKT_HASH_TYPE_NONE, /* Undefined type */
935 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
936 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
937 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
938};
939
940static inline void
941skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
942{
Tom Herbert61b905d2014-03-24 15:34:47 -0700943 skb->l4_hash = (type == PKT_HASH_TYPE_L4);
Tom Herberta3b18dd2014-07-01 21:33:17 -0700944 skb->sw_hash = 0;
Tom Herbert61b905d2014-03-24 15:34:47 -0700945 skb->hash = hash;
Tom Herbert09323cc2013-12-15 22:16:19 -0800946}
947
Tom Herberte5276932015-09-01 09:24:23 -0700948void __skb_get_hash(struct sk_buff *skb);
949u32 skb_get_poff(const struct sk_buff *skb);
950u32 __skb_get_poff(const struct sk_buff *skb, void *data,
951 const struct flow_keys *keys, int hlen);
952__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
953 void *data, int hlen_proto);
954
955static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
956 int thoff, u8 ip_proto)
957{
958 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
959}
960
961void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
962 const struct flow_dissector_key *key,
963 unsigned int key_count);
964
965bool __skb_flow_dissect(const struct sk_buff *skb,
966 struct flow_dissector *flow_dissector,
967 void *target_container,
968 void *data, __be16 proto, int nhoff, int hlen);
969
970static inline bool skb_flow_dissect(const struct sk_buff *skb,
971 struct flow_dissector *flow_dissector,
972 void *target_container)
973{
974 return __skb_flow_dissect(skb, flow_dissector, target_container,
975 NULL, 0, 0, 0);
976}
977
978static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
979 struct flow_keys *flow)
980{
981 memset(flow, 0, sizeof(*flow));
982 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
983 NULL, 0, 0, 0);
984}
985
986static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
987 void *data, __be16 proto,
988 int nhoff, int hlen)
989{
990 memset(flow, 0, sizeof(*flow));
991 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
992 data, proto, nhoff, hlen);
993}
994
Tom Herbert3958afa1b2013-12-15 22:12:06 -0800995static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +0000996{
Tom Herberta3b18dd2014-07-01 21:33:17 -0700997 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -0800998 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +0000999
Tom Herbert61b905d2014-03-24 15:34:47 -07001000 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001001}
1002
Tom Herbertf70ea012015-07-31 16:52:10 -07001003__u32 __skb_get_hash_flowi6(struct sk_buff *skb, struct flowi6 *fl6);
1004
1005static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, struct flowi6 *fl6)
1006{
1007 if (!skb->l4_hash && !skb->sw_hash)
1008 __skb_get_hash_flowi6(skb, fl6);
1009
1010 return skb->hash;
1011}
1012
1013__u32 __skb_get_hash_flowi4(struct sk_buff *skb, struct flowi4 *fl);
1014
1015static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, struct flowi4 *fl4)
1016{
1017 if (!skb->l4_hash && !skb->sw_hash)
1018 __skb_get_hash_flowi4(skb, fl4);
1019
1020 return skb->hash;
1021}
1022
Tom Herbert50fb7992015-05-01 11:30:12 -07001023__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1024
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001025static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1026{
Tom Herbert61b905d2014-03-24 15:34:47 -07001027 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001028}
1029
Tom Herbert7539fad2013-12-15 22:12:18 -08001030static inline void skb_clear_hash(struct sk_buff *skb)
1031{
Tom Herbert61b905d2014-03-24 15:34:47 -07001032 skb->hash = 0;
Tom Herberta3b18dd2014-07-01 21:33:17 -07001033 skb->sw_hash = 0;
Tom Herbert61b905d2014-03-24 15:34:47 -07001034 skb->l4_hash = 0;
Tom Herbert7539fad2013-12-15 22:12:18 -08001035}
1036
1037static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1038{
Tom Herbert61b905d2014-03-24 15:34:47 -07001039 if (!skb->l4_hash)
Tom Herbert7539fad2013-12-15 22:12:18 -08001040 skb_clear_hash(skb);
1041}
1042
Tom Herbert3df7a742013-12-15 22:16:29 -08001043static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1044{
Tom Herbert61b905d2014-03-24 15:34:47 -07001045 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -07001046 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -07001047 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -08001048};
1049
Eric Dumazetc29390c2015-03-11 18:42:02 -07001050static inline void skb_sender_cpu_clear(struct sk_buff *skb)
1051{
1052#ifdef CONFIG_XPS
1053 skb->sender_cpu = 0;
1054#endif
1055}
1056
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001057#ifdef NET_SKBUFF_DATA_USES_OFFSET
1058static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1059{
1060 return skb->head + skb->end;
1061}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001062
1063static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1064{
1065 return skb->end;
1066}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001067#else
1068static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1069{
1070 return skb->end;
1071}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001072
1073static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1074{
1075 return skb->end - skb->head;
1076}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001077#endif
1078
Linus Torvalds1da177e2005-04-16 15:20:36 -07001079/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001080#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001081
Patrick Ohlyac45f602009-02-12 05:03:37 +00001082static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1083{
1084 return &skb_shinfo(skb)->hwtstamps;
1085}
1086
Linus Torvalds1da177e2005-04-16 15:20:36 -07001087/**
1088 * skb_queue_empty - check if a queue is empty
1089 * @list: queue head
1090 *
1091 * Returns true if the queue is empty, false otherwise.
1092 */
1093static inline int skb_queue_empty(const struct sk_buff_head *list)
1094{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001095 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001096}
1097
1098/**
David S. Millerfc7ebb22008-09-23 00:34:07 -07001099 * skb_queue_is_last - check if skb is the last entry in the queue
1100 * @list: queue head
1101 * @skb: buffer
1102 *
1103 * Returns true if @skb is the last buffer on the list.
1104 */
1105static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1106 const struct sk_buff *skb)
1107{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001108 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -07001109}
1110
1111/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001112 * skb_queue_is_first - check if skb is the first entry in the queue
1113 * @list: queue head
1114 * @skb: buffer
1115 *
1116 * Returns true if @skb is the first buffer on the list.
1117 */
1118static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1119 const struct sk_buff *skb)
1120{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001121 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001122}
1123
1124/**
David S. Miller249c8b42008-09-23 00:44:42 -07001125 * skb_queue_next - return the next packet in the queue
1126 * @list: queue head
1127 * @skb: current buffer
1128 *
1129 * Return the next packet in @list after @skb. It is only valid to
1130 * call this if skb_queue_is_last() evaluates to false.
1131 */
1132static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1133 const struct sk_buff *skb)
1134{
1135 /* This BUG_ON may seem severe, but if we just return then we
1136 * are going to dereference garbage.
1137 */
1138 BUG_ON(skb_queue_is_last(list, skb));
1139 return skb->next;
1140}
1141
1142/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001143 * skb_queue_prev - return the prev packet in the queue
1144 * @list: queue head
1145 * @skb: current buffer
1146 *
1147 * Return the prev packet in @list before @skb. It is only valid to
1148 * call this if skb_queue_is_first() evaluates to false.
1149 */
1150static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1151 const struct sk_buff *skb)
1152{
1153 /* This BUG_ON may seem severe, but if we just return then we
1154 * are going to dereference garbage.
1155 */
1156 BUG_ON(skb_queue_is_first(list, skb));
1157 return skb->prev;
1158}
1159
1160/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001161 * skb_get - reference buffer
1162 * @skb: buffer to reference
1163 *
1164 * Makes another reference to a socket buffer and returns a pointer
1165 * to the buffer.
1166 */
1167static inline struct sk_buff *skb_get(struct sk_buff *skb)
1168{
1169 atomic_inc(&skb->users);
1170 return skb;
1171}
1172
1173/*
1174 * If users == 1, we are the only owner and are can avoid redundant
1175 * atomic change.
1176 */
1177
1178/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001179 * skb_cloned - is the buffer a clone
1180 * @skb: buffer to check
1181 *
1182 * Returns true if the buffer was generated with skb_clone() and is
1183 * one of multiple shared copies of the buffer. Cloned buffers are
1184 * shared data so must not be written to under normal circumstances.
1185 */
1186static inline int skb_cloned(const struct sk_buff *skb)
1187{
1188 return skb->cloned &&
1189 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1190}
1191
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001192static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1193{
1194 might_sleep_if(pri & __GFP_WAIT);
1195
1196 if (skb_cloned(skb))
1197 return pskb_expand_head(skb, 0, 0, pri);
1198
1199 return 0;
1200}
1201
Linus Torvalds1da177e2005-04-16 15:20:36 -07001202/**
1203 * skb_header_cloned - is the header a clone
1204 * @skb: buffer to check
1205 *
1206 * Returns true if modifying the header part of the buffer requires
1207 * the data to be copied.
1208 */
1209static inline int skb_header_cloned(const struct sk_buff *skb)
1210{
1211 int dataref;
1212
1213 if (!skb->cloned)
1214 return 0;
1215
1216 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1217 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1218 return dataref != 1;
1219}
1220
1221/**
1222 * skb_header_release - release reference to header
1223 * @skb: buffer to operate on
1224 *
1225 * Drop a reference to the header part of the buffer. This is done
1226 * by acquiring a payload reference. You must not read from the header
1227 * part of skb->data after this.
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001228 * Note : Check if you can use __skb_header_release() instead.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001229 */
1230static inline void skb_header_release(struct sk_buff *skb)
1231{
1232 BUG_ON(skb->nohdr);
1233 skb->nohdr = 1;
1234 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1235}
1236
1237/**
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001238 * __skb_header_release - release reference to header
1239 * @skb: buffer to operate on
1240 *
1241 * Variant of skb_header_release() assuming skb is private to caller.
1242 * We can avoid one atomic operation.
1243 */
1244static inline void __skb_header_release(struct sk_buff *skb)
1245{
1246 skb->nohdr = 1;
1247 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1248}
1249
1250
1251/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001252 * skb_shared - is the buffer shared
1253 * @skb: buffer to check
1254 *
1255 * Returns true if more than one person has a reference to this
1256 * buffer.
1257 */
1258static inline int skb_shared(const struct sk_buff *skb)
1259{
1260 return atomic_read(&skb->users) != 1;
1261}
1262
1263/**
1264 * skb_share_check - check if buffer is shared and if so clone it
1265 * @skb: buffer to check
1266 * @pri: priority for memory allocation
1267 *
1268 * If the buffer is shared the buffer is cloned and the old copy
1269 * drops a reference. A new clone with a single reference is returned.
1270 * If the buffer is not shared the original buffer is returned. When
1271 * being called from interrupt status or with spinlocks held pri must
1272 * be GFP_ATOMIC.
1273 *
1274 * NULL is returned on a memory allocation failure.
1275 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001276static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001277{
1278 might_sleep_if(pri & __GFP_WAIT);
1279 if (skb_shared(skb)) {
1280 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001281
1282 if (likely(nskb))
1283 consume_skb(skb);
1284 else
1285 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 skb = nskb;
1287 }
1288 return skb;
1289}
1290
1291/*
1292 * Copy shared buffers into a new sk_buff. We effectively do COW on
1293 * packets to handle cases where we have a local reader and forward
1294 * and a couple of other messy ones. The normal one is tcpdumping
1295 * a packet thats being forwarded.
1296 */
1297
1298/**
1299 * skb_unshare - make a copy of a shared buffer
1300 * @skb: buffer to check
1301 * @pri: priority for memory allocation
1302 *
1303 * If the socket buffer is a clone then this function creates a new
1304 * copy of the data, drops a reference count on the old copy and returns
1305 * the new copy with the reference count at 1. If the buffer is not a clone
1306 * the original buffer is returned. When called with a spinlock held or
1307 * from interrupt state @pri must be %GFP_ATOMIC
1308 *
1309 * %NULL is returned on a memory allocation failure.
1310 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001311static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001312 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313{
1314 might_sleep_if(pri & __GFP_WAIT);
1315 if (skb_cloned(skb)) {
1316 struct sk_buff *nskb = skb_copy(skb, pri);
Alexander Aring31eff812014-10-10 23:10:47 +02001317
1318 /* Free our shared copy */
1319 if (likely(nskb))
1320 consume_skb(skb);
1321 else
1322 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001323 skb = nskb;
1324 }
1325 return skb;
1326}
1327
1328/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001329 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001330 * @list_: list to peek at
1331 *
1332 * Peek an &sk_buff. Unlike most other operations you _MUST_
1333 * be careful with this one. A peek leaves the buffer on the
1334 * list and someone else may run off with it. You must hold
1335 * the appropriate locks or have a private queue to do this.
1336 *
1337 * Returns %NULL for an empty list or a pointer to the head element.
1338 * The reference count is not incremented and the reference is therefore
1339 * volatile. Use with caution.
1340 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001341static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001342{
Eric Dumazet18d07002012-04-30 16:31:46 +00001343 struct sk_buff *skb = list_->next;
1344
1345 if (skb == (struct sk_buff *)list_)
1346 skb = NULL;
1347 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001348}
1349
1350/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001351 * skb_peek_next - peek skb following the given one from a queue
1352 * @skb: skb to start from
1353 * @list_: list to peek at
1354 *
1355 * Returns %NULL when the end of the list is met or a pointer to the
1356 * next element. The reference count is not incremented and the
1357 * reference is therefore volatile. Use with caution.
1358 */
1359static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1360 const struct sk_buff_head *list_)
1361{
1362 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001363
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001364 if (next == (struct sk_buff *)list_)
1365 next = NULL;
1366 return next;
1367}
1368
1369/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001370 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001371 * @list_: list to peek at
1372 *
1373 * Peek an &sk_buff. Unlike most other operations you _MUST_
1374 * be careful with this one. A peek leaves the buffer on the
1375 * list and someone else may run off with it. You must hold
1376 * the appropriate locks or have a private queue to do this.
1377 *
1378 * Returns %NULL for an empty list or a pointer to the tail element.
1379 * The reference count is not incremented and the reference is therefore
1380 * volatile. Use with caution.
1381 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001382static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001383{
Eric Dumazet18d07002012-04-30 16:31:46 +00001384 struct sk_buff *skb = list_->prev;
1385
1386 if (skb == (struct sk_buff *)list_)
1387 skb = NULL;
1388 return skb;
1389
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390}
1391
1392/**
1393 * skb_queue_len - get queue length
1394 * @list_: list to measure
1395 *
1396 * Return the length of an &sk_buff queue.
1397 */
1398static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1399{
1400 return list_->qlen;
1401}
1402
David S. Miller67fed452008-09-21 22:36:24 -07001403/**
1404 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1405 * @list: queue to initialize
1406 *
1407 * This initializes only the list and queue length aspects of
1408 * an sk_buff_head object. This allows to initialize the list
1409 * aspects of an sk_buff_head without reinitializing things like
1410 * the spinlock. It can also be used for on-stack sk_buff_head
1411 * objects where the spinlock is known to not be used.
1412 */
1413static inline void __skb_queue_head_init(struct sk_buff_head *list)
1414{
1415 list->prev = list->next = (struct sk_buff *)list;
1416 list->qlen = 0;
1417}
1418
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001419/*
1420 * This function creates a split out lock class for each invocation;
1421 * this is needed for now since a whole lot of users of the skb-queue
1422 * infrastructure in drivers have different locking usage (in hardirq)
1423 * than the networking core (in softirq only). In the long run either the
1424 * network layer or drivers should need annotation to consolidate the
1425 * main types of usage into 3 classes.
1426 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001427static inline void skb_queue_head_init(struct sk_buff_head *list)
1428{
1429 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001430 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001431}
1432
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001433static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1434 struct lock_class_key *class)
1435{
1436 skb_queue_head_init(list);
1437 lockdep_set_class(&list->lock, class);
1438}
1439
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001441 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001442 *
1443 * The "__skb_xxxx()" functions are the non-atomic ones that
1444 * can only be called with interrupts disabled.
1445 */
Joe Perches7965bd42013-09-26 14:48:15 -07001446void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1447 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001448static inline void __skb_insert(struct sk_buff *newsk,
1449 struct sk_buff *prev, struct sk_buff *next,
1450 struct sk_buff_head *list)
1451{
1452 newsk->next = next;
1453 newsk->prev = prev;
1454 next->prev = prev->next = newsk;
1455 list->qlen++;
1456}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457
David S. Miller67fed452008-09-21 22:36:24 -07001458static inline void __skb_queue_splice(const struct sk_buff_head *list,
1459 struct sk_buff *prev,
1460 struct sk_buff *next)
1461{
1462 struct sk_buff *first = list->next;
1463 struct sk_buff *last = list->prev;
1464
1465 first->prev = prev;
1466 prev->next = first;
1467
1468 last->next = next;
1469 next->prev = last;
1470}
1471
1472/**
1473 * skb_queue_splice - join two skb lists, this is designed for stacks
1474 * @list: the new list to add
1475 * @head: the place to add it in the first list
1476 */
1477static inline void skb_queue_splice(const struct sk_buff_head *list,
1478 struct sk_buff_head *head)
1479{
1480 if (!skb_queue_empty(list)) {
1481 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001482 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001483 }
1484}
1485
1486/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001487 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001488 * @list: the new list to add
1489 * @head: the place to add it in the first list
1490 *
1491 * The list at @list is reinitialised
1492 */
1493static inline void skb_queue_splice_init(struct sk_buff_head *list,
1494 struct sk_buff_head *head)
1495{
1496 if (!skb_queue_empty(list)) {
1497 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001498 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001499 __skb_queue_head_init(list);
1500 }
1501}
1502
1503/**
1504 * skb_queue_splice_tail - join two skb lists, each list being a queue
1505 * @list: the new list to add
1506 * @head: the place to add it in the first list
1507 */
1508static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1509 struct sk_buff_head *head)
1510{
1511 if (!skb_queue_empty(list)) {
1512 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001513 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001514 }
1515}
1516
1517/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001518 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001519 * @list: the new list to add
1520 * @head: the place to add it in the first list
1521 *
1522 * Each of the lists is a queue.
1523 * The list at @list is reinitialised
1524 */
1525static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1526 struct sk_buff_head *head)
1527{
1528 if (!skb_queue_empty(list)) {
1529 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001530 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001531 __skb_queue_head_init(list);
1532 }
1533}
1534
Linus Torvalds1da177e2005-04-16 15:20:36 -07001535/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001536 * __skb_queue_after - queue a buffer at the list head
1537 * @list: list to use
1538 * @prev: place after this buffer
1539 * @newsk: buffer to queue
1540 *
1541 * Queue a buffer int the middle of a list. This function takes no locks
1542 * and you must therefore hold required locks before calling it.
1543 *
1544 * A buffer cannot be placed on two lists at the same time.
1545 */
1546static inline void __skb_queue_after(struct sk_buff_head *list,
1547 struct sk_buff *prev,
1548 struct sk_buff *newsk)
1549{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001550 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001551}
1552
Joe Perches7965bd42013-09-26 14:48:15 -07001553void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1554 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001555
Gerrit Renkerf5572852008-04-14 00:05:28 -07001556static inline void __skb_queue_before(struct sk_buff_head *list,
1557 struct sk_buff *next,
1558 struct sk_buff *newsk)
1559{
1560 __skb_insert(newsk, next->prev, next, list);
1561}
1562
Stephen Hemminger300ce172005-10-30 13:47:34 -08001563/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 * __skb_queue_head - queue a buffer at the list head
1565 * @list: list to use
1566 * @newsk: buffer to queue
1567 *
1568 * Queue a buffer at the start of a list. This function takes no locks
1569 * and you must therefore hold required locks before calling it.
1570 *
1571 * A buffer cannot be placed on two lists at the same time.
1572 */
Joe Perches7965bd42013-09-26 14:48:15 -07001573void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574static inline void __skb_queue_head(struct sk_buff_head *list,
1575 struct sk_buff *newsk)
1576{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001577 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001578}
1579
1580/**
1581 * __skb_queue_tail - queue a buffer at the list tail
1582 * @list: list to use
1583 * @newsk: buffer to queue
1584 *
1585 * Queue a buffer at the end of a list. This function takes no locks
1586 * and you must therefore hold required locks before calling it.
1587 *
1588 * A buffer cannot be placed on two lists at the same time.
1589 */
Joe Perches7965bd42013-09-26 14:48:15 -07001590void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001591static inline void __skb_queue_tail(struct sk_buff_head *list,
1592 struct sk_buff *newsk)
1593{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001594 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595}
1596
Linus Torvalds1da177e2005-04-16 15:20:36 -07001597/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001598 * remove sk_buff from list. _Must_ be called atomically, and with
1599 * the list known..
1600 */
Joe Perches7965bd42013-09-26 14:48:15 -07001601void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001602static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1603{
1604 struct sk_buff *next, *prev;
1605
1606 list->qlen--;
1607 next = skb->next;
1608 prev = skb->prev;
1609 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001610 next->prev = prev;
1611 prev->next = next;
1612}
1613
Gerrit Renkerf525c062008-04-14 00:04:12 -07001614/**
1615 * __skb_dequeue - remove from the head of the queue
1616 * @list: list to dequeue from
1617 *
1618 * Remove the head of the list. This function does not take any locks
1619 * so must be used with appropriate locks held only. The head item is
1620 * returned or %NULL if the list is empty.
1621 */
Joe Perches7965bd42013-09-26 14:48:15 -07001622struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001623static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1624{
1625 struct sk_buff *skb = skb_peek(list);
1626 if (skb)
1627 __skb_unlink(skb, list);
1628 return skb;
1629}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001630
1631/**
1632 * __skb_dequeue_tail - remove from the tail of the queue
1633 * @list: list to dequeue from
1634 *
1635 * Remove the tail of the list. This function does not take any locks
1636 * so must be used with appropriate locks held only. The tail item is
1637 * returned or %NULL if the list is empty.
1638 */
Joe Perches7965bd42013-09-26 14:48:15 -07001639struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1641{
1642 struct sk_buff *skb = skb_peek_tail(list);
1643 if (skb)
1644 __skb_unlink(skb, list);
1645 return skb;
1646}
1647
1648
David S. Millerbdcc0922012-03-07 20:53:36 -05001649static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001650{
1651 return skb->data_len;
1652}
1653
1654static inline unsigned int skb_headlen(const struct sk_buff *skb)
1655{
1656 return skb->len - skb->data_len;
1657}
1658
1659static inline int skb_pagelen(const struct sk_buff *skb)
1660{
1661 int i, len = 0;
1662
1663 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001664 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001665 return len + skb_headlen(skb);
1666}
1667
Ian Campbell131ea662011-08-19 06:25:00 +00001668/**
1669 * __skb_fill_page_desc - initialise a paged fragment in an skb
1670 * @skb: buffer containing fragment to be initialised
1671 * @i: paged fragment index to initialise
1672 * @page: the page to use for this fragment
1673 * @off: the offset to the data with @page
1674 * @size: the length of the data
1675 *
1676 * Initialises the @i'th fragment of @skb to point to &size bytes at
1677 * offset @off within @page.
1678 *
1679 * Does not take any additional reference on the fragment.
1680 */
1681static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1682 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683{
1684 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1685
Mel Gormanc48a11c2012-07-31 16:44:23 -07001686 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001687 * Propagate page pfmemalloc to the skb if we can. The problem is
1688 * that not all callers have unique ownership of the page but rely
1689 * on page_is_pfmemalloc doing the right thing(tm).
Mel Gormanc48a11c2012-07-31 16:44:23 -07001690 */
Ian Campbella8605c62011-10-19 23:01:49 +00001691 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001692 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001693 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001694
1695 page = compound_head(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001696 if (page_is_pfmemalloc(page))
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001697 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001698}
1699
1700/**
1701 * skb_fill_page_desc - initialise a paged fragment in an skb
1702 * @skb: buffer containing fragment to be initialised
1703 * @i: paged fragment index to initialise
1704 * @page: the page to use for this fragment
1705 * @off: the offset to the data with @page
1706 * @size: the length of the data
1707 *
1708 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001709 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001710 * addition updates @skb such that @i is the last fragment.
1711 *
1712 * Does not take any additional reference on the fragment.
1713 */
1714static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1715 struct page *page, int off, int size)
1716{
1717 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001718 skb_shinfo(skb)->nr_frags = i + 1;
1719}
1720
Joe Perches7965bd42013-09-26 14:48:15 -07001721void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1722 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001723
Jason Wangf8e617e2013-11-01 14:07:47 +08001724void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1725 unsigned int truesize);
1726
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001728#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001729#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1730
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001731#ifdef NET_SKBUFF_DATA_USES_OFFSET
1732static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1733{
1734 return skb->head + skb->tail;
1735}
1736
1737static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1738{
1739 skb->tail = skb->data - skb->head;
1740}
1741
1742static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1743{
1744 skb_reset_tail_pointer(skb);
1745 skb->tail += offset;
1746}
Simon Horman7cc46192013-05-28 20:34:29 +00001747
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001748#else /* NET_SKBUFF_DATA_USES_OFFSET */
1749static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1750{
1751 return skb->tail;
1752}
1753
1754static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1755{
1756 skb->tail = skb->data;
1757}
1758
1759static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1760{
1761 skb->tail = skb->data + offset;
1762}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001763
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001764#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1765
Linus Torvalds1da177e2005-04-16 15:20:36 -07001766/*
1767 * Add data to an sk_buff
1768 */
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001769unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001770unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001771static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1772{
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001773 unsigned char *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 SKB_LINEAR_ASSERT(skb);
1775 skb->tail += len;
1776 skb->len += len;
1777 return tmp;
1778}
1779
Joe Perches7965bd42013-09-26 14:48:15 -07001780unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001781static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1782{
1783 skb->data -= len;
1784 skb->len += len;
1785 return skb->data;
1786}
1787
Joe Perches7965bd42013-09-26 14:48:15 -07001788unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001789static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1790{
1791 skb->len -= len;
1792 BUG_ON(skb->len < skb->data_len);
1793 return skb->data += len;
1794}
1795
David S. Miller47d29642010-05-02 02:21:44 -07001796static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1797{
1798 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1799}
1800
Joe Perches7965bd42013-09-26 14:48:15 -07001801unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802
1803static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1804{
1805 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07001806 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001807 return NULL;
1808 skb->len -= len;
1809 return skb->data += len;
1810}
1811
1812static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1813{
1814 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1815}
1816
1817static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1818{
1819 if (likely(len <= skb_headlen(skb)))
1820 return 1;
1821 if (unlikely(len > skb->len))
1822 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07001823 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001824}
1825
1826/**
1827 * skb_headroom - bytes at buffer head
1828 * @skb: buffer to check
1829 *
1830 * Return the number of bytes of free space at the head of an &sk_buff.
1831 */
Chuck Leverc2636b42007-10-23 21:07:32 -07001832static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001833{
1834 return skb->data - skb->head;
1835}
1836
1837/**
1838 * skb_tailroom - bytes at buffer end
1839 * @skb: buffer to check
1840 *
1841 * Return the number of bytes of free space at the tail of an sk_buff
1842 */
1843static inline int skb_tailroom(const struct sk_buff *skb)
1844{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001845 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846}
1847
1848/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00001849 * skb_availroom - bytes at buffer end
1850 * @skb: buffer to check
1851 *
1852 * Return the number of bytes of free space at the tail of an sk_buff
1853 * allocated by sk_stream_alloc()
1854 */
1855static inline int skb_availroom(const struct sk_buff *skb)
1856{
Eric Dumazet16fad692013-03-14 05:40:32 +00001857 if (skb_is_nonlinear(skb))
1858 return 0;
1859
1860 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00001861}
1862
1863/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001864 * skb_reserve - adjust headroom
1865 * @skb: buffer to alter
1866 * @len: bytes to move
1867 *
1868 * Increase the headroom of an empty &sk_buff by reducing the tail
1869 * room. This is only allowed for an empty buffer.
1870 */
David S. Miller8243126c2006-01-17 02:54:21 -08001871static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001872{
1873 skb->data += len;
1874 skb->tail += len;
1875}
1876
Tom Herbert8bce6d72014-09-29 20:22:29 -07001877#define ENCAP_TYPE_ETHER 0
1878#define ENCAP_TYPE_IPPROTO 1
1879
1880static inline void skb_set_inner_protocol(struct sk_buff *skb,
1881 __be16 protocol)
1882{
1883 skb->inner_protocol = protocol;
1884 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1885}
1886
1887static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1888 __u8 ipproto)
1889{
1890 skb->inner_ipproto = ipproto;
1891 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
1892}
1893
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001894static inline void skb_reset_inner_headers(struct sk_buff *skb)
1895{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00001896 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001897 skb->inner_network_header = skb->network_header;
1898 skb->inner_transport_header = skb->transport_header;
1899}
1900
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00001901static inline void skb_reset_mac_len(struct sk_buff *skb)
1902{
1903 skb->mac_len = skb->network_header - skb->mac_header;
1904}
1905
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001906static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1907 *skb)
1908{
1909 return skb->head + skb->inner_transport_header;
1910}
1911
1912static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1913{
1914 skb->inner_transport_header = skb->data - skb->head;
1915}
1916
1917static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1918 const int offset)
1919{
1920 skb_reset_inner_transport_header(skb);
1921 skb->inner_transport_header += offset;
1922}
1923
1924static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1925{
1926 return skb->head + skb->inner_network_header;
1927}
1928
1929static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1930{
1931 skb->inner_network_header = skb->data - skb->head;
1932}
1933
1934static inline void skb_set_inner_network_header(struct sk_buff *skb,
1935 const int offset)
1936{
1937 skb_reset_inner_network_header(skb);
1938 skb->inner_network_header += offset;
1939}
1940
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00001941static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1942{
1943 return skb->head + skb->inner_mac_header;
1944}
1945
1946static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1947{
1948 skb->inner_mac_header = skb->data - skb->head;
1949}
1950
1951static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1952 const int offset)
1953{
1954 skb_reset_inner_mac_header(skb);
1955 skb->inner_mac_header += offset;
1956}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00001957static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1958{
Cong Wang35d04612013-05-29 15:16:05 +08001959 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00001960}
1961
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07001962static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1963{
1964 return skb->head + skb->transport_header;
1965}
1966
1967static inline void skb_reset_transport_header(struct sk_buff *skb)
1968{
1969 skb->transport_header = skb->data - skb->head;
1970}
1971
1972static inline void skb_set_transport_header(struct sk_buff *skb,
1973 const int offset)
1974{
1975 skb_reset_transport_header(skb);
1976 skb->transport_header += offset;
1977}
1978
1979static inline unsigned char *skb_network_header(const struct sk_buff *skb)
1980{
1981 return skb->head + skb->network_header;
1982}
1983
1984static inline void skb_reset_network_header(struct sk_buff *skb)
1985{
1986 skb->network_header = skb->data - skb->head;
1987}
1988
1989static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
1990{
1991 skb_reset_network_header(skb);
1992 skb->network_header += offset;
1993}
1994
1995static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
1996{
1997 return skb->head + skb->mac_header;
1998}
1999
2000static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2001{
Cong Wang35d04612013-05-29 15:16:05 +08002002 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002003}
2004
2005static inline void skb_reset_mac_header(struct sk_buff *skb)
2006{
2007 skb->mac_header = skb->data - skb->head;
2008}
2009
2010static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2011{
2012 skb_reset_mac_header(skb);
2013 skb->mac_header += offset;
2014}
2015
Timo Teräs0e3da5b2013-12-16 11:02:09 +02002016static inline void skb_pop_mac_header(struct sk_buff *skb)
2017{
2018 skb->mac_header = skb->network_header;
2019}
2020
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002021static inline void skb_probe_transport_header(struct sk_buff *skb,
2022 const int offset_hint)
2023{
2024 struct flow_keys keys;
2025
2026 if (skb_transport_header_was_set(skb))
2027 return;
Jiri Pirko06635a32015-05-12 14:56:16 +02002028 else if (skb_flow_dissect_flow_keys(skb, &keys))
Tom Herbert42aecaa2015-06-04 09:16:39 -07002029 skb_set_transport_header(skb, keys.control.thoff);
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002030 else
2031 skb_set_transport_header(skb, offset_hint);
2032}
2033
Eric Dumazet03606892012-02-23 10:55:02 +00002034static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2035{
2036 if (skb_mac_header_was_set(skb)) {
2037 const unsigned char *old_mac = skb_mac_header(skb);
2038
2039 skb_set_mac_header(skb, -skb->mac_len);
2040 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2041 }
2042}
2043
Michał Mirosław04fb4512010-12-14 15:24:08 +00002044static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2045{
2046 return skb->csum_start - skb_headroom(skb);
2047}
2048
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002049static inline int skb_transport_offset(const struct sk_buff *skb)
2050{
2051 return skb_transport_header(skb) - skb->data;
2052}
2053
2054static inline u32 skb_network_header_len(const struct sk_buff *skb)
2055{
2056 return skb->transport_header - skb->network_header;
2057}
2058
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002059static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2060{
2061 return skb->inner_transport_header - skb->inner_network_header;
2062}
2063
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002064static inline int skb_network_offset(const struct sk_buff *skb)
2065{
2066 return skb_network_header(skb) - skb->data;
2067}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03002068
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002069static inline int skb_inner_network_offset(const struct sk_buff *skb)
2070{
2071 return skb_inner_network_header(skb) - skb->data;
2072}
2073
Changli Gaof9599ce2010-08-04 04:43:44 +00002074static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2075{
2076 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2077}
2078
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079/*
2080 * CPUs often take a performance hit when accessing unaligned memory
2081 * locations. The actual performance hit varies, it can be small if the
2082 * hardware handles it or large if we have to take an exception and fix it
2083 * in software.
2084 *
2085 * Since an ethernet header is 14 bytes network drivers often end up with
2086 * the IP header at an unaligned offset. The IP header can be aligned by
2087 * shifting the start of the packet by 2 bytes. Drivers should do this
2088 * with:
2089 *
Tobias Klauser8660c122009-07-13 22:48:16 +00002090 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002091 *
2092 * The downside to this alignment of the IP header is that the DMA is now
2093 * unaligned. On some architectures the cost of an unaligned DMA is high
2094 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00002095 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2097 * to be overridden.
2098 */
2099#ifndef NET_IP_ALIGN
2100#define NET_IP_ALIGN 2
2101#endif
2102
Anton Blanchard025be812006-03-31 02:27:06 -08002103/*
2104 * The networking layer reserves some headroom in skb data (via
2105 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2106 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08002107 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08002108 *
2109 * Unfortunately this headroom changes the DMA alignment of the resulting
2110 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2111 * on some architectures. An architecture can override this value,
2112 * perhaps setting it to a cacheline in size (since that will maintain
2113 * cacheline alignment of the DMA). It must be a power of 2.
2114 *
David S. Millerd6301d32009-02-08 19:24:13 -08002115 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08002116 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07002117 *
2118 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2119 * to reduce average number of cache lines per packet.
2120 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07002121 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08002122 */
2123#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07002124#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08002125#endif
2126
Joe Perches7965bd42013-09-26 14:48:15 -07002127int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002128
2129static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2130{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00002131 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07002132 WARN_ON(1);
2133 return;
2134 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002135 skb->len = len;
2136 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002137}
2138
Joe Perches7965bd42013-09-26 14:48:15 -07002139void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002140
2141static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2142{
Herbert Xu3cc0e872006-06-09 16:13:38 -07002143 if (skb->data_len)
2144 return ___pskb_trim(skb, len);
2145 __skb_trim(skb, len);
2146 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002147}
2148
2149static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2150{
2151 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2152}
2153
2154/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07002155 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2156 * @skb: buffer to alter
2157 * @len: new length
2158 *
2159 * This is identical to pskb_trim except that the caller knows that
2160 * the skb is not cloned so we should never get an error due to out-
2161 * of-memory.
2162 */
2163static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2164{
2165 int err = pskb_trim(skb, len);
2166 BUG_ON(err);
2167}
2168
2169/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170 * skb_orphan - orphan a buffer
2171 * @skb: buffer to orphan
2172 *
2173 * If a buffer currently has an owner then we call the owner's
2174 * destructor function and make the @skb unowned. The buffer continues
2175 * to exist but is no longer charged to its former owner.
2176 */
2177static inline void skb_orphan(struct sk_buff *skb)
2178{
Eric Dumazetc34a7612013-07-30 16:11:15 -07002179 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002180 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002181 skb->destructor = NULL;
2182 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07002183 } else {
2184 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002185 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002186}
2187
2188/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002189 * skb_orphan_frags - orphan the frags contained in a buffer
2190 * @skb: buffer to orphan frags from
2191 * @gfp_mask: allocation mask for replacement pages
2192 *
2193 * For each frag in the SKB which needs a destructor (i.e. has an
2194 * owner) create a copy of that frag and release the original
2195 * page by calling the destructor.
2196 */
2197static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2198{
2199 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2200 return 0;
2201 return skb_copy_ubufs(skb, gfp_mask);
2202}
2203
2204/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002205 * __skb_queue_purge - empty a list
2206 * @list: list to empty
2207 *
2208 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2209 * the list and one reference dropped. This function does not take the
2210 * list lock and the caller must hold the relevant locks to use it.
2211 */
Joe Perches7965bd42013-09-26 14:48:15 -07002212void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002213static inline void __skb_queue_purge(struct sk_buff_head *list)
2214{
2215 struct sk_buff *skb;
2216 while ((skb = __skb_dequeue(list)) != NULL)
2217 kfree_skb(skb);
2218}
2219
Joe Perches7965bd42013-09-26 14:48:15 -07002220void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002221
Joe Perches7965bd42013-09-26 14:48:15 -07002222struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2223 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002224
2225/**
2226 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2227 * @dev: network device to receive on
2228 * @length: length to allocate
2229 *
2230 * Allocate a new &sk_buff and assign it a usage count of one. The
2231 * buffer has unspecified headroom built in. Users should allocate
2232 * the headroom they think they need without accounting for the
2233 * built in space. The built in space is used for optimisations.
2234 *
2235 * %NULL is returned if there is no free memory. Although this function
2236 * allocates memory it can be called from an interrupt.
2237 */
2238static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002239 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002240{
2241 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2242}
2243
Eric Dumazet6f532612012-05-18 05:12:12 +00002244/* legacy helper around __netdev_alloc_skb() */
2245static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2246 gfp_t gfp_mask)
2247{
2248 return __netdev_alloc_skb(NULL, length, gfp_mask);
2249}
2250
2251/* legacy helper around netdev_alloc_skb() */
2252static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2253{
2254 return netdev_alloc_skb(NULL, length);
2255}
2256
2257
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002258static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2259 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002260{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002261 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002262
2263 if (NET_IP_ALIGN && skb)
2264 skb_reserve(skb, NET_IP_ALIGN);
2265 return skb;
2266}
2267
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002268static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2269 unsigned int length)
2270{
2271 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2272}
2273
Alexander Duyck181edb22015-05-06 21:12:03 -07002274static inline void skb_free_frag(void *addr)
2275{
2276 __free_page_frag(addr);
2277}
2278
Alexander Duyckffde7322014-12-09 19:40:42 -08002279void *napi_alloc_frag(unsigned int fragsz);
Alexander Duyckfd11a832014-12-09 19:40:49 -08002280struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2281 unsigned int length, gfp_t gfp_mask);
2282static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2283 unsigned int length)
2284{
2285 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2286}
Alexander Duyckffde7322014-12-09 19:40:42 -08002287
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002288/**
Alexander Duyck71dfda52014-11-11 09:26:34 -08002289 * __dev_alloc_pages - allocate page for network Rx
2290 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2291 * @order: size of the allocation
2292 *
2293 * Allocate a new page.
2294 *
2295 * %NULL is returned if there is no free memory.
2296*/
2297static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2298 unsigned int order)
2299{
2300 /* This piece of code contains several assumptions.
2301 * 1. This is for device Rx, therefor a cold page is preferred.
2302 * 2. The expectation is the user wants a compound page.
2303 * 3. If requesting a order 0 page it will not be compound
2304 * due to the check to see if order has a value in prep_new_page
2305 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2306 * code in gfp_to_alloc_flags that should be enforcing this.
2307 */
2308 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2309
2310 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2311}
2312
2313static inline struct page *dev_alloc_pages(unsigned int order)
2314{
2315 return __dev_alloc_pages(GFP_ATOMIC, order);
2316}
2317
2318/**
2319 * __dev_alloc_page - allocate a page for network Rx
2320 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2321 *
2322 * Allocate a new page.
2323 *
2324 * %NULL is returned if there is no free memory.
2325 */
2326static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2327{
2328 return __dev_alloc_pages(gfp_mask, 0);
2329}
2330
2331static inline struct page *dev_alloc_page(void)
2332{
2333 return __dev_alloc_page(GFP_ATOMIC);
2334}
2335
2336/**
Mel Gorman06140022012-07-31 16:44:24 -07002337 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2338 * @page: The page that was allocated from skb_alloc_page
2339 * @skb: The skb that may need pfmemalloc set
2340 */
2341static inline void skb_propagate_pfmemalloc(struct page *page,
2342 struct sk_buff *skb)
2343{
Michal Hocko2f064f32015-08-21 14:11:51 -07002344 if (page_is_pfmemalloc(page))
Mel Gorman06140022012-07-31 16:44:24 -07002345 skb->pfmemalloc = true;
2346}
2347
Eric Dumazet564824b2010-10-11 19:05:25 +00002348/**
Masanari Iidae2278672014-02-18 22:54:36 +09002349 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002350 * @frag: the paged fragment
2351 *
2352 * Returns the &struct page associated with @frag.
2353 */
2354static inline struct page *skb_frag_page(const skb_frag_t *frag)
2355{
Ian Campbella8605c62011-10-19 23:01:49 +00002356 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002357}
2358
2359/**
2360 * __skb_frag_ref - take an addition reference on a paged fragment.
2361 * @frag: the paged fragment
2362 *
2363 * Takes an additional reference on the paged fragment @frag.
2364 */
2365static inline void __skb_frag_ref(skb_frag_t *frag)
2366{
2367 get_page(skb_frag_page(frag));
2368}
2369
2370/**
2371 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2372 * @skb: the buffer
2373 * @f: the fragment offset.
2374 *
2375 * Takes an additional reference on the @f'th paged fragment of @skb.
2376 */
2377static inline void skb_frag_ref(struct sk_buff *skb, int f)
2378{
2379 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2380}
2381
2382/**
2383 * __skb_frag_unref - release a reference on a paged fragment.
2384 * @frag: the paged fragment
2385 *
2386 * Releases a reference on the paged fragment @frag.
2387 */
2388static inline void __skb_frag_unref(skb_frag_t *frag)
2389{
2390 put_page(skb_frag_page(frag));
2391}
2392
2393/**
2394 * skb_frag_unref - release a reference on a paged fragment of an skb.
2395 * @skb: the buffer
2396 * @f: the fragment offset
2397 *
2398 * Releases a reference on the @f'th paged fragment of @skb.
2399 */
2400static inline void skb_frag_unref(struct sk_buff *skb, int f)
2401{
2402 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2403}
2404
2405/**
2406 * skb_frag_address - gets the address of the data contained in a paged fragment
2407 * @frag: the paged fragment buffer
2408 *
2409 * Returns the address of the data within @frag. The page must already
2410 * be mapped.
2411 */
2412static inline void *skb_frag_address(const skb_frag_t *frag)
2413{
2414 return page_address(skb_frag_page(frag)) + frag->page_offset;
2415}
2416
2417/**
2418 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2419 * @frag: the paged fragment buffer
2420 *
2421 * Returns the address of the data within @frag. Checks that the page
2422 * is mapped and returns %NULL otherwise.
2423 */
2424static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2425{
2426 void *ptr = page_address(skb_frag_page(frag));
2427 if (unlikely(!ptr))
2428 return NULL;
2429
2430 return ptr + frag->page_offset;
2431}
2432
2433/**
2434 * __skb_frag_set_page - sets the page contained in a paged fragment
2435 * @frag: the paged fragment
2436 * @page: the page to set
2437 *
2438 * Sets the fragment @frag to contain @page.
2439 */
2440static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2441{
Ian Campbella8605c62011-10-19 23:01:49 +00002442 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002443}
2444
2445/**
2446 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2447 * @skb: the buffer
2448 * @f: the fragment offset
2449 * @page: the page to set
2450 *
2451 * Sets the @f'th fragment of @skb to contain @page.
2452 */
2453static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2454 struct page *page)
2455{
2456 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2457}
2458
Eric Dumazet400dfd32013-10-17 16:27:07 -07002459bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2460
Ian Campbell131ea662011-08-19 06:25:00 +00002461/**
2462 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002463 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002464 * @frag: the paged fragment to map
2465 * @offset: the offset within the fragment (starting at the
2466 * fragment's own offset)
2467 * @size: the number of bytes to map
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002468 * @dir: the direction of the mapping (%PCI_DMA_*)
Ian Campbell131ea662011-08-19 06:25:00 +00002469 *
2470 * Maps the page associated with @frag to @device.
2471 */
2472static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2473 const skb_frag_t *frag,
2474 size_t offset, size_t size,
2475 enum dma_data_direction dir)
2476{
2477 return dma_map_page(dev, skb_frag_page(frag),
2478 frag->page_offset + offset, size, dir);
2479}
2480
Eric Dumazet117632e2011-12-03 21:39:53 +00002481static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2482 gfp_t gfp_mask)
2483{
2484 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2485}
2486
Octavian Purdilabad93e92014-06-12 01:36:26 +03002487
2488static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2489 gfp_t gfp_mask)
2490{
2491 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2492}
2493
2494
Ian Campbell131ea662011-08-19 06:25:00 +00002495/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002496 * skb_clone_writable - is the header of a clone writable
2497 * @skb: buffer to check
2498 * @len: length up to which to write
2499 *
2500 * Returns true if modifying the header part of the cloned buffer
2501 * does not requires the data to be copied.
2502 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002503static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002504{
2505 return !skb_header_cloned(skb) &&
2506 skb_headroom(skb) + len <= skb->hdr_len;
2507}
2508
Herbert Xud9cc2042007-09-16 16:21:16 -07002509static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2510 int cloned)
2511{
2512 int delta = 0;
2513
Herbert Xud9cc2042007-09-16 16:21:16 -07002514 if (headroom > skb_headroom(skb))
2515 delta = headroom - skb_headroom(skb);
2516
2517 if (delta || cloned)
2518 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2519 GFP_ATOMIC);
2520 return 0;
2521}
2522
Patrick McHardy334a8132007-06-25 04:35:20 -07002523/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 * skb_cow - copy header of skb when it is required
2525 * @skb: buffer to cow
2526 * @headroom: needed headroom
2527 *
2528 * If the skb passed lacks sufficient headroom or its data part
2529 * is shared, data is reallocated. If reallocation fails, an error
2530 * is returned and original skb is not changed.
2531 *
2532 * The result is skb with writable area skb->head...skb->tail
2533 * and at least @headroom of space at head.
2534 */
2535static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2536{
Herbert Xud9cc2042007-09-16 16:21:16 -07002537 return __skb_cow(skb, headroom, skb_cloned(skb));
2538}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002539
Herbert Xud9cc2042007-09-16 16:21:16 -07002540/**
2541 * skb_cow_head - skb_cow but only making the head writable
2542 * @skb: buffer to cow
2543 * @headroom: needed headroom
2544 *
2545 * This function is identical to skb_cow except that we replace the
2546 * skb_cloned check by skb_header_cloned. It should be used when
2547 * you only need to push on some header and do not need to modify
2548 * the data.
2549 */
2550static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2551{
2552 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002553}
2554
2555/**
2556 * skb_padto - pad an skbuff up to a minimal size
2557 * @skb: buffer to pad
2558 * @len: minimal length
2559 *
2560 * Pads up a buffer to ensure the trailing bytes exist and are
2561 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002562 * is untouched. Otherwise it is extended. Returns zero on
2563 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002564 */
Herbert Xu5b057c62006-06-23 02:06:41 -07002565static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002566{
2567 unsigned int size = skb->len;
2568 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002569 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002570 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002571}
2572
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002573/**
2574 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2575 * @skb: buffer to pad
2576 * @len: minimal length
2577 *
2578 * Pads up a buffer to ensure the trailing bytes exist and are
2579 * blanked. If the buffer already contains sufficient data it
2580 * is untouched. Otherwise it is extended. Returns zero on
2581 * success. The skb is freed on error.
2582 */
2583static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2584{
2585 unsigned int size = skb->len;
2586
2587 if (unlikely(size < len)) {
2588 len -= size;
2589 if (skb_pad(skb, len))
2590 return -ENOMEM;
2591 __skb_put(skb, len);
2592 }
2593 return 0;
2594}
2595
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596static inline int skb_add_data(struct sk_buff *skb,
Al Viroaf2b0402014-11-27 21:44:24 -05002597 struct iov_iter *from, int copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002598{
2599 const int off = skb->len;
2600
2601 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viroaf2b0402014-11-27 21:44:24 -05002602 __wsum csum = 0;
2603 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2604 &csum, from) == copy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002605 skb->csum = csum_block_add(skb->csum, csum, off);
2606 return 0;
2607 }
Al Viroaf2b0402014-11-27 21:44:24 -05002608 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002609 return 0;
2610
2611 __skb_trim(skb, off);
2612 return -EFAULT;
2613}
2614
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002615static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2616 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617{
2618 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002619 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002620
Ian Campbellea2ab692011-08-22 23:44:58 +00002621 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00002622 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002623 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002624 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002625}
2626
Herbert Xu364c6ba2006-06-09 16:10:40 -07002627static inline int __skb_linearize(struct sk_buff *skb)
2628{
2629 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2630}
2631
Linus Torvalds1da177e2005-04-16 15:20:36 -07002632/**
2633 * skb_linearize - convert paged skb to linear one
2634 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 *
2636 * If there is no free memory -ENOMEM is returned, otherwise zero
2637 * is returned and the old skb data released.
2638 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07002639static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002640{
Herbert Xu364c6ba2006-06-09 16:10:40 -07002641 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2642}
2643
2644/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00002645 * skb_has_shared_frag - can any frag be overwritten
2646 * @skb: buffer to test
2647 *
2648 * Return true if the skb has at least one frag that might be modified
2649 * by an external entity (as in vmsplice()/sendfile())
2650 */
2651static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2652{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00002653 return skb_is_nonlinear(skb) &&
2654 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002655}
2656
2657/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07002658 * skb_linearize_cow - make sure skb is linear and writable
2659 * @skb: buffer to process
2660 *
2661 * If there is no free memory -ENOMEM is returned, otherwise zero
2662 * is returned and the old skb data released.
2663 */
2664static inline int skb_linearize_cow(struct sk_buff *skb)
2665{
2666 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2667 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002668}
2669
2670/**
2671 * skb_postpull_rcsum - update checksum for received skb after pull
2672 * @skb: buffer to update
2673 * @start: start of data before pull
2674 * @len: length of data pulled
2675 *
2676 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07002677 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2678 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002679 */
2680
2681static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f2006-03-20 22:43:56 -08002682 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002683{
Patrick McHardy84fa7932006-08-29 16:44:56 -07002684 if (skb->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002685 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2686}
2687
Herbert Xucbb042f2006-03-20 22:43:56 -08002688unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2689
David S. Miller7ce5a272013-12-02 17:26:05 -05002690/**
2691 * pskb_trim_rcsum - trim received skb and update checksum
2692 * @skb: buffer to trim
2693 * @len: new length
2694 *
2695 * This is exactly the same as pskb_trim except that it ensures the
2696 * checksum of received packets are still valid after the operation.
2697 */
2698
2699static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2700{
2701 if (likely(len >= skb->len))
2702 return 0;
2703 if (skb->ip_summed == CHECKSUM_COMPLETE)
2704 skb->ip_summed = CHECKSUM_NONE;
2705 return __pskb_trim(skb, len);
2706}
2707
Linus Torvalds1da177e2005-04-16 15:20:36 -07002708#define skb_queue_walk(queue, skb) \
2709 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002710 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002711 skb = skb->next)
2712
James Chapman46f89142007-04-30 00:07:31 -07002713#define skb_queue_walk_safe(queue, skb, tmp) \
2714 for (skb = (queue)->next, tmp = skb->next; \
2715 skb != (struct sk_buff *)(queue); \
2716 skb = tmp, tmp = skb->next)
2717
David S. Miller1164f522008-09-23 00:49:44 -07002718#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002719 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07002720 skb = skb->next)
2721
2722#define skb_queue_walk_from_safe(queue, skb, tmp) \
2723 for (tmp = skb->next; \
2724 skb != (struct sk_buff *)(queue); \
2725 skb = tmp, tmp = skb->next)
2726
Stephen Hemminger300ce172005-10-30 13:47:34 -08002727#define skb_queue_reverse_walk(queue, skb) \
2728 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002729 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08002730 skb = skb->prev)
2731
David S. Miller686a2952011-01-20 22:47:32 -08002732#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2733 for (skb = (queue)->prev, tmp = skb->prev; \
2734 skb != (struct sk_buff *)(queue); \
2735 skb = tmp, tmp = skb->prev)
2736
2737#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2738 for (tmp = skb->prev; \
2739 skb != (struct sk_buff *)(queue); \
2740 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741
David S. Miller21dc3302010-08-23 00:13:46 -07002742static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07002743{
2744 return skb_shinfo(skb)->frag_list != NULL;
2745}
2746
2747static inline void skb_frag_list_init(struct sk_buff *skb)
2748{
2749 skb_shinfo(skb)->frag_list = NULL;
2750}
2751
David S. Milleree039872009-06-09 00:17:13 -07002752#define skb_walk_frags(skb, iter) \
2753 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2754
Joe Perches7965bd42013-09-26 14:48:15 -07002755struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2756 int *peeked, int *off, int *err);
2757struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2758 int *err);
2759unsigned int datagram_poll(struct file *file, struct socket *sock,
2760 struct poll_table_struct *wait);
Al Viroc0371da2014-11-24 10:42:55 -05002761int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2762 struct iov_iter *to, int size);
David S. Miller51f3d022014-11-05 16:46:40 -05002763static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2764 struct msghdr *msg, int size)
2765{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002766 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
David S. Miller51f3d022014-11-05 16:46:40 -05002767}
Al Viroe5a4b0b2014-11-24 18:17:55 -05002768int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2769 struct msghdr *msg);
Al Viro3a654f92014-06-19 14:15:22 -04002770int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2771 struct iov_iter *from, int len);
Al Viro3a654f92014-06-19 14:15:22 -04002772int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
Joe Perches7965bd42013-09-26 14:48:15 -07002773void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2774void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2775int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07002776int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2777int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2778__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2779 int len, __wsum csum);
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002780ssize_t skb_socket_splice(struct sock *sk,
2781 struct pipe_inode_info *pipe,
2782 struct splice_pipe_desc *spd);
2783int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
Joe Perches7965bd42013-09-26 14:48:15 -07002784 struct pipe_inode_info *pipe, unsigned int len,
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002785 unsigned int flags,
2786 ssize_t (*splice_cb)(struct sock *,
2787 struct pipe_inode_info *,
2788 struct splice_pipe_desc *));
Joe Perches7965bd42013-09-26 14:48:15 -07002789void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01002790unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002791int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2792 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07002793void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2794int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2795void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01002796unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -07002797struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Vlad Yasevich0d5501c2014-08-08 14:42:13 -04002798struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
Jiri Pirkoe2195122014-11-19 14:05:01 +01002799int skb_ensure_writable(struct sk_buff *skb, int write_len);
Jiri Pirko93515d52014-11-19 14:05:02 +01002800int skb_vlan_pop(struct sk_buff *skb);
2801int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002802
Al Viro6ce8e9c2014-04-06 21:25:44 -04002803static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2804{
Al Viro21226ab2014-11-28 15:48:29 -05002805 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro6ce8e9c2014-04-06 21:25:44 -04002806}
2807
Al Viro7eab8d92014-04-06 21:51:23 -04002808static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2809{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002810 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro7eab8d92014-04-06 21:51:23 -04002811}
2812
Daniel Borkmann2817a332013-10-30 11:50:51 +01002813struct skb_checksum_ops {
2814 __wsum (*update)(const void *mem, int len, __wsum wsum);
2815 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2816};
2817
2818__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2819 __wsum csum, const struct skb_checksum_ops *ops);
2820__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2821 __wsum csum);
2822
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07002823static inline void * __must_check
2824__skb_header_pointer(const struct sk_buff *skb, int offset,
2825 int len, void *data, int hlen, void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002826{
Patrick McHardy55820ee2005-07-05 14:08:10 -07002827 if (hlen - offset >= len)
David S. Miller690e36e2014-08-23 12:13:41 -07002828 return data + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002829
David S. Miller690e36e2014-08-23 12:13:41 -07002830 if (!skb ||
2831 skb_copy_bits(skb, offset, buffer, len) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002832 return NULL;
2833
2834 return buffer;
2835}
2836
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07002837static inline void * __must_check
2838skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
David S. Miller690e36e2014-08-23 12:13:41 -07002839{
2840 return __skb_header_pointer(skb, offset, len, skb->data,
2841 skb_headlen(skb), buffer);
2842}
2843
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01002844/**
2845 * skb_needs_linearize - check if we need to linearize a given skb
2846 * depending on the given device features.
2847 * @skb: socket buffer to check
2848 * @features: net device features
2849 *
2850 * Returns true if either:
2851 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2852 * 2. skb is fragmented and the device does not support SG.
2853 */
2854static inline bool skb_needs_linearize(struct sk_buff *skb,
2855 netdev_features_t features)
2856{
2857 return skb_is_nonlinear(skb) &&
2858 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2859 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2860}
2861
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002862static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2863 void *to,
2864 const unsigned int len)
2865{
2866 memcpy(to, skb->data, len);
2867}
2868
2869static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2870 const int offset, void *to,
2871 const unsigned int len)
2872{
2873 memcpy(to, skb->data + offset, len);
2874}
2875
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03002876static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2877 const void *from,
2878 const unsigned int len)
2879{
2880 memcpy(skb->data, from, len);
2881}
2882
2883static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2884 const int offset,
2885 const void *from,
2886 const unsigned int len)
2887{
2888 memcpy(skb->data + offset, from, len);
2889}
2890
Joe Perches7965bd42013-09-26 14:48:15 -07002891void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002892
Patrick Ohlyac45f602009-02-12 05:03:37 +00002893static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2894{
2895 return skb->tstamp;
2896}
2897
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002898/**
2899 * skb_get_timestamp - get timestamp from a skb
2900 * @skb: skb to get stamp from
2901 * @stamp: pointer to struct timeval to store stamp in
2902 *
2903 * Timestamps are stored in the skb as offsets to a base timestamp.
2904 * This function converts the offset back to a struct timeval and stores
2905 * it in stamp.
2906 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00002907static inline void skb_get_timestamp(const struct sk_buff *skb,
2908 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002909{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002910 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002911}
2912
Patrick Ohlyac45f602009-02-12 05:03:37 +00002913static inline void skb_get_timestampns(const struct sk_buff *skb,
2914 struct timespec *stamp)
2915{
2916 *stamp = ktime_to_timespec(skb->tstamp);
2917}
2918
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002919static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002920{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002921 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002922}
2923
Stephen Hemminger164891a2007-04-23 22:26:16 -07002924static inline ktime_t net_timedelta(ktime_t t)
2925{
2926 return ktime_sub(ktime_get_real(), t);
2927}
2928
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07002929static inline ktime_t net_invalid_timestamp(void)
2930{
2931 return ktime_set(0, 0);
2932}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002933
Alexander Duyck62bccb82014-09-04 13:31:35 -04002934struct sk_buff *skb_clone_sk(struct sk_buff *skb);
2935
Richard Cochranc1f19b52010-07-17 08:49:36 +00002936#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2937
Joe Perches7965bd42013-09-26 14:48:15 -07002938void skb_clone_tx_timestamp(struct sk_buff *skb);
2939bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00002940
2941#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2942
2943static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2944{
2945}
2946
2947static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2948{
2949 return false;
2950}
2951
2952#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2953
2954/**
2955 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2956 *
Richard Cochranda92b192011-10-21 00:49:15 +00002957 * PHY drivers may accept clones of transmitted packets for
2958 * timestamping via their phy_driver.txtstamp method. These drivers
Benjamin Poirier7a76a022015-08-07 09:32:21 -07002959 * must call this function to return the skb back to the stack with a
2960 * timestamp.
Richard Cochranda92b192011-10-21 00:49:15 +00002961 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00002962 * @skb: clone of the the original outgoing packet
Benjamin Poirier7a76a022015-08-07 09:32:21 -07002963 * @hwtstamps: hardware time stamps
Richard Cochranc1f19b52010-07-17 08:49:36 +00002964 *
2965 */
2966void skb_complete_tx_timestamp(struct sk_buff *skb,
2967 struct skb_shared_hwtstamps *hwtstamps);
2968
Willem de Bruijne7fd2882014-08-04 22:11:48 -04002969void __skb_tstamp_tx(struct sk_buff *orig_skb,
2970 struct skb_shared_hwtstamps *hwtstamps,
2971 struct sock *sk, int tstype);
2972
Patrick Ohlyac45f602009-02-12 05:03:37 +00002973/**
2974 * skb_tstamp_tx - queue clone of skb with send time stamps
2975 * @orig_skb: the original outgoing packet
2976 * @hwtstamps: hardware time stamps, may be NULL if not available
2977 *
2978 * If the skb has a socket associated, then this function clones the
2979 * skb (thus sharing the actual data and optional structures), stores
2980 * the optional hardware time stamping information (if non NULL) or
2981 * generates a software time stamp (otherwise), then queues the clone
2982 * to the error queue of the socket. Errors are silently ignored.
2983 */
Joe Perches7965bd42013-09-26 14:48:15 -07002984void skb_tstamp_tx(struct sk_buff *orig_skb,
2985 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00002986
Richard Cochran4507a712010-07-17 08:48:28 +00002987static inline void sw_tx_timestamp(struct sk_buff *skb)
2988{
Oliver Hartkopp2244d072010-08-17 08:59:14 +00002989 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
2990 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
Richard Cochran4507a712010-07-17 08:48:28 +00002991 skb_tstamp_tx(skb, NULL);
2992}
2993
2994/**
2995 * skb_tx_timestamp() - Driver hook for transmit timestamping
2996 *
2997 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00002998 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00002999 *
David S. Miller73409f32013-12-27 13:04:33 -05003000 * Specifically, one should make absolutely sure that this function is
3001 * called before TX completion of this packet can trigger. Otherwise
3002 * the packet could potentially already be freed.
3003 *
Richard Cochran4507a712010-07-17 08:48:28 +00003004 * @skb: A socket buffer.
3005 */
3006static inline void skb_tx_timestamp(struct sk_buff *skb)
3007{
Richard Cochranc1f19b52010-07-17 08:49:36 +00003008 skb_clone_tx_timestamp(skb);
Richard Cochran4507a712010-07-17 08:48:28 +00003009 sw_tx_timestamp(skb);
3010}
3011
Johannes Berg6e3e9392011-11-09 10:15:42 +01003012/**
3013 * skb_complete_wifi_ack - deliver skb with wifi status
3014 *
3015 * @skb: the original outgoing packet
3016 * @acked: ack status
3017 *
3018 */
3019void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3020
Joe Perches7965bd42013-09-26 14:48:15 -07003021__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3022__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003023
Herbert Xu60476372007-04-09 11:59:39 -07003024static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3025{
Tom Herbert6edec0e2015-02-10 16:30:28 -08003026 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3027 skb->csum_valid ||
3028 (skb->ip_summed == CHECKSUM_PARTIAL &&
3029 skb_checksum_start_offset(skb) >= 0));
Herbert Xu60476372007-04-09 11:59:39 -07003030}
3031
Herbert Xufb286bb2005-11-10 13:01:24 -08003032/**
3033 * skb_checksum_complete - Calculate checksum of an entire packet
3034 * @skb: packet to process
3035 *
3036 * This function calculates the checksum over the entire packet plus
3037 * the value of skb->csum. The latter can be used to supply the
3038 * checksum of a pseudo header as used by TCP/UDP. It returns the
3039 * checksum.
3040 *
3041 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3042 * this function can be used to verify that checksum on received
3043 * packets. In that case the function should return zero if the
3044 * checksum is correct. In particular, this function will return zero
3045 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3046 * hardware has already verified the correctness of the checksum.
3047 */
Al Viro4381ca32007-07-15 21:00:11 +01003048static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08003049{
Herbert Xu60476372007-04-09 11:59:39 -07003050 return skb_csum_unnecessary(skb) ?
3051 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003052}
3053
Tom Herbert77cffe22014-08-27 21:26:46 -07003054static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3055{
3056 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3057 if (skb->csum_level == 0)
3058 skb->ip_summed = CHECKSUM_NONE;
3059 else
3060 skb->csum_level--;
3061 }
3062}
3063
3064static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3065{
3066 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3067 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3068 skb->csum_level++;
3069 } else if (skb->ip_summed == CHECKSUM_NONE) {
3070 skb->ip_summed = CHECKSUM_UNNECESSARY;
3071 skb->csum_level = 0;
3072 }
3073}
3074
Tom Herbert5a212322014-08-31 15:12:41 -07003075static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3076{
3077 /* Mark current checksum as bad (typically called from GRO
3078 * path). In the case that ip_summed is CHECKSUM_NONE
3079 * this must be the first checksum encountered in the packet.
3080 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3081 * checksum after the last one validated. For UDP, a zero
3082 * checksum can not be marked as bad.
3083 */
3084
3085 if (skb->ip_summed == CHECKSUM_NONE ||
3086 skb->ip_summed == CHECKSUM_UNNECESSARY)
3087 skb->csum_bad = 1;
3088}
3089
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003090/* Check if we need to perform checksum complete validation.
3091 *
3092 * Returns true if checksum complete is needed, false otherwise
3093 * (either checksum is unnecessary or zero checksum is allowed).
3094 */
3095static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3096 bool zero_okay,
3097 __sum16 check)
3098{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003099 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3100 skb->csum_valid = 1;
Tom Herbert77cffe22014-08-27 21:26:46 -07003101 __skb_decr_checksum_unnecessary(skb);
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003102 return false;
3103 }
3104
3105 return true;
3106}
3107
3108/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3109 * in checksum_init.
3110 */
3111#define CHECKSUM_BREAK 76
3112
Tom Herbert4e18b9a2015-04-20 14:10:04 -07003113/* Unset checksum-complete
3114 *
3115 * Unset checksum complete can be done when packet is being modified
3116 * (uncompressed for instance) and checksum-complete value is
3117 * invalidated.
3118 */
3119static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3120{
3121 if (skb->ip_summed == CHECKSUM_COMPLETE)
3122 skb->ip_summed = CHECKSUM_NONE;
3123}
3124
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003125/* Validate (init) checksum based on checksum complete.
3126 *
3127 * Return values:
3128 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3129 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3130 * checksum is stored in skb->csum for use in __skb_checksum_complete
3131 * non-zero: value of invalid checksum
3132 *
3133 */
3134static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3135 bool complete,
3136 __wsum psum)
3137{
3138 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3139 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003140 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003141 return 0;
3142 }
Tom Herbert5a212322014-08-31 15:12:41 -07003143 } else if (skb->csum_bad) {
3144 /* ip_summed == CHECKSUM_NONE in this case */
Eric Dumazetc91d4602015-05-15 05:48:07 -07003145 return (__force __sum16)1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003146 }
3147
3148 skb->csum = psum;
3149
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003150 if (complete || skb->len <= CHECKSUM_BREAK) {
3151 __sum16 csum;
3152
3153 csum = __skb_checksum_complete(skb);
3154 skb->csum_valid = !csum;
3155 return csum;
3156 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003157
3158 return 0;
3159}
3160
3161static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3162{
3163 return 0;
3164}
3165
3166/* Perform checksum validate (init). Note that this is a macro since we only
3167 * want to calculate the pseudo header which is an input function if necessary.
3168 * First we try to validate without any computation (checksum unnecessary) and
3169 * then calculate based on checksum complete calling the function to compute
3170 * pseudo header.
3171 *
3172 * Return values:
3173 * 0: checksum is validated or try to in skb_checksum_complete
3174 * non-zero: value of invalid checksum
3175 */
3176#define __skb_checksum_validate(skb, proto, complete, \
3177 zero_okay, check, compute_pseudo) \
3178({ \
3179 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003180 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003181 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3182 __ret = __skb_checksum_validate_complete(skb, \
3183 complete, compute_pseudo(skb, proto)); \
3184 __ret; \
3185})
3186
3187#define skb_checksum_init(skb, proto, compute_pseudo) \
3188 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3189
3190#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3191 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3192
3193#define skb_checksum_validate(skb, proto, compute_pseudo) \
3194 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3195
3196#define skb_checksum_validate_zero_check(skb, proto, check, \
3197 compute_pseudo) \
Sabrina Dubroca096a4cf2015-02-06 18:54:19 +01003198 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003199
3200#define skb_checksum_simple_validate(skb) \
3201 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3202
Tom Herbertd96535a2014-08-31 15:12:42 -07003203static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3204{
3205 return (skb->ip_summed == CHECKSUM_NONE &&
3206 skb->csum_valid && !skb->csum_bad);
3207}
3208
3209static inline void __skb_checksum_convert(struct sk_buff *skb,
3210 __sum16 check, __wsum pseudo)
3211{
3212 skb->csum = ~pseudo;
3213 skb->ip_summed = CHECKSUM_COMPLETE;
3214}
3215
3216#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3217do { \
3218 if (__skb_checksum_convert_check(skb)) \
3219 __skb_checksum_convert(skb, check, \
3220 compute_pseudo(skb, proto)); \
3221} while (0)
3222
Tom Herbert15e23962015-02-10 16:30:31 -08003223static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3224 u16 start, u16 offset)
3225{
3226 skb->ip_summed = CHECKSUM_PARTIAL;
3227 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3228 skb->csum_offset = offset - start;
3229}
3230
Tom Herbertdcdc8992015-02-02 16:07:34 -08003231/* Update skbuf and packet to reflect the remote checksum offload operation.
3232 * When called, ptr indicates the starting point for skb->csum when
3233 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3234 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3235 */
3236static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
Tom Herbert15e23962015-02-10 16:30:31 -08003237 int start, int offset, bool nopartial)
Tom Herbertdcdc8992015-02-02 16:07:34 -08003238{
3239 __wsum delta;
3240
Tom Herbert15e23962015-02-10 16:30:31 -08003241 if (!nopartial) {
3242 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3243 return;
3244 }
3245
Tom Herbertdcdc8992015-02-02 16:07:34 -08003246 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3247 __skb_checksum_complete(skb);
3248 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3249 }
3250
3251 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3252
3253 /* Adjust skb->csum since we changed the packet */
3254 skb->csum = csum_add(skb->csum, delta);
3255}
3256
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003257#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07003258void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003259static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3260{
3261 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07003262 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003263}
3264static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3265{
3266 if (nfct)
3267 atomic_inc(&nfct->use);
3268}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003269#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003270#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003271static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3272{
3273 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3274 kfree(nf_bridge);
3275}
3276static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3277{
3278 if (nf_bridge)
3279 atomic_inc(&nf_bridge->use);
3280}
3281#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08003282static inline void nf_reset(struct sk_buff *skb)
3283{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003284#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003285 nf_conntrack_put(skb->nfct);
3286 skb->nfct = NULL;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003287#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003288#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003289 nf_bridge_put(skb->nf_bridge);
3290 skb->nf_bridge = NULL;
3291#endif
3292}
3293
Patrick McHardy124dff02013-04-05 20:42:05 +02003294static inline void nf_reset_trace(struct sk_buff *skb)
3295{
Florian Westphal478b3602014-02-15 23:48:45 +01003296#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00003297 skb->nf_trace = 0;
3298#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003299}
3300
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003301/* Note: This doesn't put any conntrack and bridge info in dst. */
Eric Dumazetb1937222014-09-28 22:18:47 -07003302static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3303 bool copy)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003304{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003305#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003306 dst->nfct = src->nfct;
3307 nf_conntrack_get(src->nfct);
Eric Dumazetb1937222014-09-28 22:18:47 -07003308 if (copy)
3309 dst->nfctinfo = src->nfctinfo;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003310#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003311#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003312 dst->nf_bridge = src->nf_bridge;
3313 nf_bridge_get(src->nf_bridge);
3314#endif
Florian Westphal478b3602014-02-15 23:48:45 +01003315#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Eric Dumazetb1937222014-09-28 22:18:47 -07003316 if (copy)
3317 dst->nf_trace = src->nf_trace;
Florian Westphal478b3602014-02-15 23:48:45 +01003318#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003319}
3320
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003321static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3322{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003323#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003324 nf_conntrack_put(dst->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003325#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003326#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003327 nf_bridge_put(dst->nf_bridge);
3328#endif
Eric Dumazetb1937222014-09-28 22:18:47 -07003329 __nf_copy(dst, src, true);
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003330}
3331
James Morris984bc162006-06-09 00:29:17 -07003332#ifdef CONFIG_NETWORK_SECMARK
3333static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3334{
3335 to->secmark = from->secmark;
3336}
3337
3338static inline void skb_init_secmark(struct sk_buff *skb)
3339{
3340 skb->secmark = 0;
3341}
3342#else
3343static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3344{ }
3345
3346static inline void skb_init_secmark(struct sk_buff *skb)
3347{ }
3348#endif
3349
Eric W. Biederman574f7192014-04-01 12:20:24 -07003350static inline bool skb_irq_freeable(const struct sk_buff *skb)
3351{
3352 return !skb->destructor &&
3353#if IS_ENABLED(CONFIG_XFRM)
3354 !skb->sp &&
3355#endif
3356#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3357 !skb->nfct &&
3358#endif
3359 !skb->_skb_refdst &&
3360 !skb_has_frag_list(skb);
3361}
3362
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003363static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3364{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003365 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003366}
3367
Stephen Hemminger92477442009-03-21 13:39:26 -07003368static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003369{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003370 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003371}
3372
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003373static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3374{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003375 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003376}
3377
David S. Millerd5a9e242009-01-27 16:22:11 -08003378static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3379{
3380 skb->queue_mapping = rx_queue + 1;
3381}
3382
Stephen Hemminger92477442009-03-21 13:39:26 -07003383static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003384{
3385 return skb->queue_mapping - 1;
3386}
3387
Stephen Hemminger92477442009-03-21 13:39:26 -07003388static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003389{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003390 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003391}
3392
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003393static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3394{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003395#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003396 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003397#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003398 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003399#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003400}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003401
Pravin B Shelar68c33162013-02-14 14:02:41 +00003402/* Keeps track of mac header offset relative to skb->head.
3403 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3404 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003405 * tunnel skb it points to outer mac header.
3406 * Keeps track of level of encapsulation of network headers.
3407 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003408struct skb_gso_cb {
Eric Dumazet3347c962013-10-19 11:42:56 -07003409 int mac_offset;
3410 int encap_level;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003411 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003412};
3413#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3414
3415static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3416{
3417 return (skb_mac_header(inner_skb) - inner_skb->head) -
3418 SKB_GSO_CB(inner_skb)->mac_offset;
3419}
3420
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003421static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3422{
3423 int new_headroom, headroom;
3424 int ret;
3425
3426 headroom = skb_headroom(skb);
3427 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3428 if (ret)
3429 return ret;
3430
3431 new_headroom = skb_headroom(skb);
3432 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3433 return 0;
3434}
3435
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003436/* Compute the checksum for a gso segment. First compute the checksum value
3437 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3438 * then add in skb->csum (checksum from csum_start to end of packet).
3439 * skb->csum and csum_start are then updated to reflect the checksum of the
3440 * resultant packet starting from the transport header-- the resultant checksum
3441 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3442 * header.
3443 */
3444static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3445{
3446 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
Eric Dumazetc91d4602015-05-15 05:48:07 -07003447 skb_transport_offset(skb);
3448 __wsum partial;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003449
Eric Dumazetc91d4602015-05-15 05:48:07 -07003450 partial = csum_partial(skb_transport_header(skb), plen, skb->csum);
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003451 skb->csum = res;
3452 SKB_GSO_CB(skb)->csum_start -= plen;
3453
Eric Dumazetc91d4602015-05-15 05:48:07 -07003454 return csum_fold(partial);
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003455}
3456
David S. Millerbdcc0922012-03-07 20:53:36 -05003457static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07003458{
3459 return skb_shinfo(skb)->gso_size;
3460}
3461
Eric Dumazet36a8f392013-09-29 01:21:32 -07003462/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05003463static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02003464{
3465 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3466}
3467
Joe Perches7965bd42013-09-26 14:48:15 -07003468void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07003469
3470static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3471{
3472 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3473 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003474 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3475
Alexander Duyckb78462e2010-06-02 12:24:37 +00003476 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3477 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07003478 __skb_warn_lro_forwarding(skb);
3479 return true;
3480 }
3481 return false;
3482}
3483
Herbert Xu35fc92a2007-03-26 23:22:20 -07003484static inline void skb_forward_csum(struct sk_buff *skb)
3485{
3486 /* Unfortunately we don't support this one. Any brave souls? */
3487 if (skb->ip_summed == CHECKSUM_COMPLETE)
3488 skb->ip_summed = CHECKSUM_NONE;
3489}
3490
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003491/**
3492 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3493 * @skb: skb to check
3494 *
3495 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3496 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3497 * use this helper, to document places where we make this assertion.
3498 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003499static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003500{
3501#ifdef DEBUG
3502 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3503#endif
3504}
3505
Rusty Russellf35d9d82008-02-04 23:49:54 -05003506bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00003507
Paul Durranted1f50c2014-01-09 10:02:46 +00003508int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
Linus Lüssing9afd85c2015-05-02 14:01:07 +02003509struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3510 unsigned int transport_len,
3511 __sum16(*skb_chkf)(struct sk_buff *skb));
Paul Durranted1f50c2014-01-09 10:02:46 +00003512
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00003513/**
3514 * skb_head_is_locked - Determine if the skb->head is locked down
3515 * @skb: skb to check
3516 *
3517 * The head on skbs build around a head frag can be removed if they are
3518 * not cloned. This function returns true if the skb head is locked down
3519 * due to either being allocated via kmalloc, or by being a clone with
3520 * multiple references to the head.
3521 */
3522static inline bool skb_head_is_locked(const struct sk_buff *skb)
3523{
3524 return !skb->head_frag || skb_cloned(skb);
3525}
Florian Westphalfe6cc552014-02-13 23:09:12 +01003526
3527/**
3528 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3529 *
3530 * @skb: GSO skb
3531 *
3532 * skb_gso_network_seglen is used to determine the real size of the
3533 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3534 *
3535 * The MAC/L2 header is not accounted for.
3536 */
3537static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3538{
3539 unsigned int hdr_len = skb_transport_header(skb) -
3540 skb_network_header(skb);
3541 return hdr_len + skb_gso_transport_seglen(skb);
3542}
Thomas Grafee122c72015-07-21 10:43:58 +02003543
Linus Torvalds1da177e2005-04-16 15:20:36 -07003544#endif /* __KERNEL__ */
3545#endif /* _LINUX_SKBUFF_H */