blob: bbe41bccfc5f297a037946a7a5055b3ea8e1f4c6 [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
Tom Herbertbcc83832015-09-01 09:24:24 -0700940static inline void skb_clear_hash(struct sk_buff *skb)
941{
942 skb->hash = 0;
943 skb->sw_hash = 0;
944 skb->l4_hash = 0;
945}
946
947static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
948{
949 if (!skb->l4_hash)
950 skb_clear_hash(skb);
951}
952
953static inline void
954__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
955{
956 skb->l4_hash = is_l4;
957 skb->sw_hash = is_sw;
958 skb->hash = hash;
959}
960
Tom Herbert09323cc2013-12-15 22:16:19 -0800961static inline void
962skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
963{
Tom Herbertbcc83832015-09-01 09:24:24 -0700964 /* Used by drivers to set hash from HW */
965 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
966}
967
968static inline void
969__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
970{
971 __skb_set_hash(skb, hash, true, is_l4);
Tom Herbert09323cc2013-12-15 22:16:19 -0800972}
973
Tom Herberte5276932015-09-01 09:24:23 -0700974void __skb_get_hash(struct sk_buff *skb);
975u32 skb_get_poff(const struct sk_buff *skb);
976u32 __skb_get_poff(const struct sk_buff *skb, void *data,
977 const struct flow_keys *keys, int hlen);
978__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
979 void *data, int hlen_proto);
980
981static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
982 int thoff, u8 ip_proto)
983{
984 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
985}
986
987void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
988 const struct flow_dissector_key *key,
989 unsigned int key_count);
990
991bool __skb_flow_dissect(const struct sk_buff *skb,
992 struct flow_dissector *flow_dissector,
993 void *target_container,
994 void *data, __be16 proto, int nhoff, int hlen);
995
996static inline bool skb_flow_dissect(const struct sk_buff *skb,
997 struct flow_dissector *flow_dissector,
998 void *target_container)
999{
1000 return __skb_flow_dissect(skb, flow_dissector, target_container,
1001 NULL, 0, 0, 0);
1002}
1003
1004static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1005 struct flow_keys *flow)
1006{
1007 memset(flow, 0, sizeof(*flow));
1008 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
1009 NULL, 0, 0, 0);
1010}
1011
1012static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1013 void *data, __be16 proto,
1014 int nhoff, int hlen)
1015{
1016 memset(flow, 0, sizeof(*flow));
1017 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
1018 data, proto, nhoff, hlen);
1019}
1020
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001021static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001022{
Tom Herberta3b18dd2014-07-01 21:33:17 -07001023 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001024 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001025
Tom Herbert61b905d2014-03-24 15:34:47 -07001026 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001027}
1028
Tom Herbertf70ea012015-07-31 16:52:10 -07001029__u32 __skb_get_hash_flowi6(struct sk_buff *skb, struct flowi6 *fl6);
1030
1031static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, struct flowi6 *fl6)
1032{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001033 if (!skb->l4_hash && !skb->sw_hash) {
1034 struct flow_keys keys;
1035
1036 __skb_set_sw_hash(skb, __get_hash_from_flowi6(fl6, &keys),
1037 flow_keys_have_l4(&keys));
1038 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001039
1040 return skb->hash;
1041}
1042
1043__u32 __skb_get_hash_flowi4(struct sk_buff *skb, struct flowi4 *fl);
1044
1045static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, struct flowi4 *fl4)
1046{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001047 if (!skb->l4_hash && !skb->sw_hash) {
1048 struct flow_keys keys;
1049
1050 __skb_set_sw_hash(skb, __get_hash_from_flowi4(fl4, &keys),
1051 flow_keys_have_l4(&keys));
1052 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001053
1054 return skb->hash;
1055}
1056
Tom Herbert50fb7992015-05-01 11:30:12 -07001057__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1058
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001059static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1060{
Tom Herbert61b905d2014-03-24 15:34:47 -07001061 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001062}
1063
Tom Herbert3df7a742013-12-15 22:16:29 -08001064static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1065{
Tom Herbert61b905d2014-03-24 15:34:47 -07001066 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -07001067 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -07001068 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -08001069};
1070
Eric Dumazetc29390c2015-03-11 18:42:02 -07001071static inline void skb_sender_cpu_clear(struct sk_buff *skb)
1072{
1073#ifdef CONFIG_XPS
1074 skb->sender_cpu = 0;
1075#endif
1076}
1077
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001078#ifdef NET_SKBUFF_DATA_USES_OFFSET
1079static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1080{
1081 return skb->head + skb->end;
1082}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001083
1084static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1085{
1086 return skb->end;
1087}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001088#else
1089static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1090{
1091 return skb->end;
1092}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001093
1094static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1095{
1096 return skb->end - skb->head;
1097}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001098#endif
1099
Linus Torvalds1da177e2005-04-16 15:20:36 -07001100/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001101#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001102
Patrick Ohlyac45f602009-02-12 05:03:37 +00001103static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1104{
1105 return &skb_shinfo(skb)->hwtstamps;
1106}
1107
Linus Torvalds1da177e2005-04-16 15:20:36 -07001108/**
1109 * skb_queue_empty - check if a queue is empty
1110 * @list: queue head
1111 *
1112 * Returns true if the queue is empty, false otherwise.
1113 */
1114static inline int skb_queue_empty(const struct sk_buff_head *list)
1115{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001116 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117}
1118
1119/**
David S. Millerfc7ebb22008-09-23 00:34:07 -07001120 * skb_queue_is_last - check if skb is the last entry in the queue
1121 * @list: queue head
1122 * @skb: buffer
1123 *
1124 * Returns true if @skb is the last buffer on the list.
1125 */
1126static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1127 const struct sk_buff *skb)
1128{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001129 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -07001130}
1131
1132/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001133 * skb_queue_is_first - check if skb is the first entry in the queue
1134 * @list: queue head
1135 * @skb: buffer
1136 *
1137 * Returns true if @skb is the first buffer on the list.
1138 */
1139static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1140 const struct sk_buff *skb)
1141{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001142 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001143}
1144
1145/**
David S. Miller249c8b42008-09-23 00:44:42 -07001146 * skb_queue_next - return the next packet in the queue
1147 * @list: queue head
1148 * @skb: current buffer
1149 *
1150 * Return the next packet in @list after @skb. It is only valid to
1151 * call this if skb_queue_is_last() evaluates to false.
1152 */
1153static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1154 const struct sk_buff *skb)
1155{
1156 /* This BUG_ON may seem severe, but if we just return then we
1157 * are going to dereference garbage.
1158 */
1159 BUG_ON(skb_queue_is_last(list, skb));
1160 return skb->next;
1161}
1162
1163/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001164 * skb_queue_prev - return the prev packet in the queue
1165 * @list: queue head
1166 * @skb: current buffer
1167 *
1168 * Return the prev packet in @list before @skb. It is only valid to
1169 * call this if skb_queue_is_first() evaluates to false.
1170 */
1171static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1172 const struct sk_buff *skb)
1173{
1174 /* This BUG_ON may seem severe, but if we just return then we
1175 * are going to dereference garbage.
1176 */
1177 BUG_ON(skb_queue_is_first(list, skb));
1178 return skb->prev;
1179}
1180
1181/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001182 * skb_get - reference buffer
1183 * @skb: buffer to reference
1184 *
1185 * Makes another reference to a socket buffer and returns a pointer
1186 * to the buffer.
1187 */
1188static inline struct sk_buff *skb_get(struct sk_buff *skb)
1189{
1190 atomic_inc(&skb->users);
1191 return skb;
1192}
1193
1194/*
1195 * If users == 1, we are the only owner and are can avoid redundant
1196 * atomic change.
1197 */
1198
1199/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001200 * skb_cloned - is the buffer a clone
1201 * @skb: buffer to check
1202 *
1203 * Returns true if the buffer was generated with skb_clone() and is
1204 * one of multiple shared copies of the buffer. Cloned buffers are
1205 * shared data so must not be written to under normal circumstances.
1206 */
1207static inline int skb_cloned(const struct sk_buff *skb)
1208{
1209 return skb->cloned &&
1210 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1211}
1212
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001213static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1214{
1215 might_sleep_if(pri & __GFP_WAIT);
1216
1217 if (skb_cloned(skb))
1218 return pskb_expand_head(skb, 0, 0, pri);
1219
1220 return 0;
1221}
1222
Linus Torvalds1da177e2005-04-16 15:20:36 -07001223/**
1224 * skb_header_cloned - is the header a clone
1225 * @skb: buffer to check
1226 *
1227 * Returns true if modifying the header part of the buffer requires
1228 * the data to be copied.
1229 */
1230static inline int skb_header_cloned(const struct sk_buff *skb)
1231{
1232 int dataref;
1233
1234 if (!skb->cloned)
1235 return 0;
1236
1237 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1238 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1239 return dataref != 1;
1240}
1241
1242/**
1243 * skb_header_release - release reference to header
1244 * @skb: buffer to operate on
1245 *
1246 * Drop a reference to the header part of the buffer. This is done
1247 * by acquiring a payload reference. You must not read from the header
1248 * part of skb->data after this.
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001249 * Note : Check if you can use __skb_header_release() instead.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001250 */
1251static inline void skb_header_release(struct sk_buff *skb)
1252{
1253 BUG_ON(skb->nohdr);
1254 skb->nohdr = 1;
1255 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1256}
1257
1258/**
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001259 * __skb_header_release - release reference to header
1260 * @skb: buffer to operate on
1261 *
1262 * Variant of skb_header_release() assuming skb is private to caller.
1263 * We can avoid one atomic operation.
1264 */
1265static inline void __skb_header_release(struct sk_buff *skb)
1266{
1267 skb->nohdr = 1;
1268 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1269}
1270
1271
1272/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 * skb_shared - is the buffer shared
1274 * @skb: buffer to check
1275 *
1276 * Returns true if more than one person has a reference to this
1277 * buffer.
1278 */
1279static inline int skb_shared(const struct sk_buff *skb)
1280{
1281 return atomic_read(&skb->users) != 1;
1282}
1283
1284/**
1285 * skb_share_check - check if buffer is shared and if so clone it
1286 * @skb: buffer to check
1287 * @pri: priority for memory allocation
1288 *
1289 * If the buffer is shared the buffer is cloned and the old copy
1290 * drops a reference. A new clone with a single reference is returned.
1291 * If the buffer is not shared the original buffer is returned. When
1292 * being called from interrupt status or with spinlocks held pri must
1293 * be GFP_ATOMIC.
1294 *
1295 * NULL is returned on a memory allocation failure.
1296 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001297static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001298{
1299 might_sleep_if(pri & __GFP_WAIT);
1300 if (skb_shared(skb)) {
1301 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001302
1303 if (likely(nskb))
1304 consume_skb(skb);
1305 else
1306 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307 skb = nskb;
1308 }
1309 return skb;
1310}
1311
1312/*
1313 * Copy shared buffers into a new sk_buff. We effectively do COW on
1314 * packets to handle cases where we have a local reader and forward
1315 * and a couple of other messy ones. The normal one is tcpdumping
1316 * a packet thats being forwarded.
1317 */
1318
1319/**
1320 * skb_unshare - make a copy of a shared buffer
1321 * @skb: buffer to check
1322 * @pri: priority for memory allocation
1323 *
1324 * If the socket buffer is a clone then this function creates a new
1325 * copy of the data, drops a reference count on the old copy and returns
1326 * the new copy with the reference count at 1. If the buffer is not a clone
1327 * the original buffer is returned. When called with a spinlock held or
1328 * from interrupt state @pri must be %GFP_ATOMIC
1329 *
1330 * %NULL is returned on a memory allocation failure.
1331 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001332static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001333 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001334{
1335 might_sleep_if(pri & __GFP_WAIT);
1336 if (skb_cloned(skb)) {
1337 struct sk_buff *nskb = skb_copy(skb, pri);
Alexander Aring31eff812014-10-10 23:10:47 +02001338
1339 /* Free our shared copy */
1340 if (likely(nskb))
1341 consume_skb(skb);
1342 else
1343 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001344 skb = nskb;
1345 }
1346 return skb;
1347}
1348
1349/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001350 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 * @list_: list to peek at
1352 *
1353 * Peek an &sk_buff. Unlike most other operations you _MUST_
1354 * be careful with this one. A peek leaves the buffer on the
1355 * list and someone else may run off with it. You must hold
1356 * the appropriate locks or have a private queue to do this.
1357 *
1358 * Returns %NULL for an empty list or a pointer to the head element.
1359 * The reference count is not incremented and the reference is therefore
1360 * volatile. Use with caution.
1361 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001362static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363{
Eric Dumazet18d07002012-04-30 16:31:46 +00001364 struct sk_buff *skb = list_->next;
1365
1366 if (skb == (struct sk_buff *)list_)
1367 skb = NULL;
1368 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369}
1370
1371/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001372 * skb_peek_next - peek skb following the given one from a queue
1373 * @skb: skb to start from
1374 * @list_: list to peek at
1375 *
1376 * Returns %NULL when the end of the list is met or a pointer to the
1377 * next element. The reference count is not incremented and the
1378 * reference is therefore volatile. Use with caution.
1379 */
1380static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1381 const struct sk_buff_head *list_)
1382{
1383 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001384
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001385 if (next == (struct sk_buff *)list_)
1386 next = NULL;
1387 return next;
1388}
1389
1390/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001391 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001392 * @list_: list to peek at
1393 *
1394 * Peek an &sk_buff. Unlike most other operations you _MUST_
1395 * be careful with this one. A peek leaves the buffer on the
1396 * list and someone else may run off with it. You must hold
1397 * the appropriate locks or have a private queue to do this.
1398 *
1399 * Returns %NULL for an empty list or a pointer to the tail element.
1400 * The reference count is not incremented and the reference is therefore
1401 * volatile. Use with caution.
1402 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001403static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001404{
Eric Dumazet18d07002012-04-30 16:31:46 +00001405 struct sk_buff *skb = list_->prev;
1406
1407 if (skb == (struct sk_buff *)list_)
1408 skb = NULL;
1409 return skb;
1410
Linus Torvalds1da177e2005-04-16 15:20:36 -07001411}
1412
1413/**
1414 * skb_queue_len - get queue length
1415 * @list_: list to measure
1416 *
1417 * Return the length of an &sk_buff queue.
1418 */
1419static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1420{
1421 return list_->qlen;
1422}
1423
David S. Miller67fed452008-09-21 22:36:24 -07001424/**
1425 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1426 * @list: queue to initialize
1427 *
1428 * This initializes only the list and queue length aspects of
1429 * an sk_buff_head object. This allows to initialize the list
1430 * aspects of an sk_buff_head without reinitializing things like
1431 * the spinlock. It can also be used for on-stack sk_buff_head
1432 * objects where the spinlock is known to not be used.
1433 */
1434static inline void __skb_queue_head_init(struct sk_buff_head *list)
1435{
1436 list->prev = list->next = (struct sk_buff *)list;
1437 list->qlen = 0;
1438}
1439
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001440/*
1441 * This function creates a split out lock class for each invocation;
1442 * this is needed for now since a whole lot of users of the skb-queue
1443 * infrastructure in drivers have different locking usage (in hardirq)
1444 * than the networking core (in softirq only). In the long run either the
1445 * network layer or drivers should need annotation to consolidate the
1446 * main types of usage into 3 classes.
1447 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001448static inline void skb_queue_head_init(struct sk_buff_head *list)
1449{
1450 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001451 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452}
1453
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001454static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1455 struct lock_class_key *class)
1456{
1457 skb_queue_head_init(list);
1458 lockdep_set_class(&list->lock, class);
1459}
1460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001462 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001463 *
1464 * The "__skb_xxxx()" functions are the non-atomic ones that
1465 * can only be called with interrupts disabled.
1466 */
Joe Perches7965bd42013-09-26 14:48:15 -07001467void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1468 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001469static inline void __skb_insert(struct sk_buff *newsk,
1470 struct sk_buff *prev, struct sk_buff *next,
1471 struct sk_buff_head *list)
1472{
1473 newsk->next = next;
1474 newsk->prev = prev;
1475 next->prev = prev->next = newsk;
1476 list->qlen++;
1477}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001478
David S. Miller67fed452008-09-21 22:36:24 -07001479static inline void __skb_queue_splice(const struct sk_buff_head *list,
1480 struct sk_buff *prev,
1481 struct sk_buff *next)
1482{
1483 struct sk_buff *first = list->next;
1484 struct sk_buff *last = list->prev;
1485
1486 first->prev = prev;
1487 prev->next = first;
1488
1489 last->next = next;
1490 next->prev = last;
1491}
1492
1493/**
1494 * skb_queue_splice - join two skb lists, this is designed for stacks
1495 * @list: the new list to add
1496 * @head: the place to add it in the first list
1497 */
1498static inline void skb_queue_splice(const struct sk_buff_head *list,
1499 struct sk_buff_head *head)
1500{
1501 if (!skb_queue_empty(list)) {
1502 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001503 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001504 }
1505}
1506
1507/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001508 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001509 * @list: the new list to add
1510 * @head: the place to add it in the first list
1511 *
1512 * The list at @list is reinitialised
1513 */
1514static inline void skb_queue_splice_init(struct sk_buff_head *list,
1515 struct sk_buff_head *head)
1516{
1517 if (!skb_queue_empty(list)) {
1518 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001519 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001520 __skb_queue_head_init(list);
1521 }
1522}
1523
1524/**
1525 * skb_queue_splice_tail - join two skb lists, each list being a queue
1526 * @list: the new list to add
1527 * @head: the place to add it in the first list
1528 */
1529static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1530 struct sk_buff_head *head)
1531{
1532 if (!skb_queue_empty(list)) {
1533 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001534 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001535 }
1536}
1537
1538/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001539 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001540 * @list: the new list to add
1541 * @head: the place to add it in the first list
1542 *
1543 * Each of the lists is a queue.
1544 * The list at @list is reinitialised
1545 */
1546static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1547 struct sk_buff_head *head)
1548{
1549 if (!skb_queue_empty(list)) {
1550 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001551 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001552 __skb_queue_head_init(list);
1553 }
1554}
1555
Linus Torvalds1da177e2005-04-16 15:20:36 -07001556/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001557 * __skb_queue_after - queue a buffer at the list head
1558 * @list: list to use
1559 * @prev: place after this buffer
1560 * @newsk: buffer to queue
1561 *
1562 * Queue a buffer int the middle of a list. This function takes no locks
1563 * and you must therefore hold required locks before calling it.
1564 *
1565 * A buffer cannot be placed on two lists at the same time.
1566 */
1567static inline void __skb_queue_after(struct sk_buff_head *list,
1568 struct sk_buff *prev,
1569 struct sk_buff *newsk)
1570{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001571 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001572}
1573
Joe Perches7965bd42013-09-26 14:48:15 -07001574void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1575 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001576
Gerrit Renkerf5572852008-04-14 00:05:28 -07001577static inline void __skb_queue_before(struct sk_buff_head *list,
1578 struct sk_buff *next,
1579 struct sk_buff *newsk)
1580{
1581 __skb_insert(newsk, next->prev, next, list);
1582}
1583
Stephen Hemminger300ce172005-10-30 13:47:34 -08001584/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001585 * __skb_queue_head - queue a buffer at the list head
1586 * @list: list to use
1587 * @newsk: buffer to queue
1588 *
1589 * Queue a buffer at the start of a list. This function takes no locks
1590 * and you must therefore hold required locks before calling it.
1591 *
1592 * A buffer cannot be placed on two lists at the same time.
1593 */
Joe Perches7965bd42013-09-26 14:48:15 -07001594void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001595static inline void __skb_queue_head(struct sk_buff_head *list,
1596 struct sk_buff *newsk)
1597{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001598 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599}
1600
1601/**
1602 * __skb_queue_tail - queue a buffer at the list tail
1603 * @list: list to use
1604 * @newsk: buffer to queue
1605 *
1606 * Queue a buffer at the end of a list. This function takes no locks
1607 * and you must therefore hold required locks before calling it.
1608 *
1609 * A buffer cannot be placed on two lists at the same time.
1610 */
Joe Perches7965bd42013-09-26 14:48:15 -07001611void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001612static inline void __skb_queue_tail(struct sk_buff_head *list,
1613 struct sk_buff *newsk)
1614{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001615 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001616}
1617
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001619 * remove sk_buff from list. _Must_ be called atomically, and with
1620 * the list known..
1621 */
Joe Perches7965bd42013-09-26 14:48:15 -07001622void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001623static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1624{
1625 struct sk_buff *next, *prev;
1626
1627 list->qlen--;
1628 next = skb->next;
1629 prev = skb->prev;
1630 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001631 next->prev = prev;
1632 prev->next = next;
1633}
1634
Gerrit Renkerf525c062008-04-14 00:04:12 -07001635/**
1636 * __skb_dequeue - remove from the head of the queue
1637 * @list: list to dequeue from
1638 *
1639 * Remove the head of the list. This function does not take any locks
1640 * so must be used with appropriate locks held only. The head item is
1641 * returned or %NULL if the list is empty.
1642 */
Joe Perches7965bd42013-09-26 14:48:15 -07001643struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001644static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1645{
1646 struct sk_buff *skb = skb_peek(list);
1647 if (skb)
1648 __skb_unlink(skb, list);
1649 return skb;
1650}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001651
1652/**
1653 * __skb_dequeue_tail - remove from the tail of the queue
1654 * @list: list to dequeue from
1655 *
1656 * Remove the tail of the list. This function does not take any locks
1657 * so must be used with appropriate locks held only. The tail item is
1658 * returned or %NULL if the list is empty.
1659 */
Joe Perches7965bd42013-09-26 14:48:15 -07001660struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1662{
1663 struct sk_buff *skb = skb_peek_tail(list);
1664 if (skb)
1665 __skb_unlink(skb, list);
1666 return skb;
1667}
1668
1669
David S. Millerbdcc0922012-03-07 20:53:36 -05001670static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001671{
1672 return skb->data_len;
1673}
1674
1675static inline unsigned int skb_headlen(const struct sk_buff *skb)
1676{
1677 return skb->len - skb->data_len;
1678}
1679
1680static inline int skb_pagelen(const struct sk_buff *skb)
1681{
1682 int i, len = 0;
1683
1684 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001685 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 return len + skb_headlen(skb);
1687}
1688
Ian Campbell131ea662011-08-19 06:25:00 +00001689/**
1690 * __skb_fill_page_desc - initialise a paged fragment in an skb
1691 * @skb: buffer containing fragment to be initialised
1692 * @i: paged fragment index to initialise
1693 * @page: the page to use for this fragment
1694 * @off: the offset to the data with @page
1695 * @size: the length of the data
1696 *
1697 * Initialises the @i'th fragment of @skb to point to &size bytes at
1698 * offset @off within @page.
1699 *
1700 * Does not take any additional reference on the fragment.
1701 */
1702static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1703 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704{
1705 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1706
Mel Gormanc48a11c2012-07-31 16:44:23 -07001707 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001708 * Propagate page pfmemalloc to the skb if we can. The problem is
1709 * that not all callers have unique ownership of the page but rely
1710 * on page_is_pfmemalloc doing the right thing(tm).
Mel Gormanc48a11c2012-07-31 16:44:23 -07001711 */
Ian Campbella8605c62011-10-19 23:01:49 +00001712 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001714 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001715
1716 page = compound_head(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001717 if (page_is_pfmemalloc(page))
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001718 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001719}
1720
1721/**
1722 * skb_fill_page_desc - initialise a paged fragment in an skb
1723 * @skb: buffer containing fragment to be initialised
1724 * @i: paged fragment index to initialise
1725 * @page: the page to use for this fragment
1726 * @off: the offset to the data with @page
1727 * @size: the length of the data
1728 *
1729 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001730 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001731 * addition updates @skb such that @i is the last fragment.
1732 *
1733 * Does not take any additional reference on the fragment.
1734 */
1735static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1736 struct page *page, int off, int size)
1737{
1738 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739 skb_shinfo(skb)->nr_frags = i + 1;
1740}
1741
Joe Perches7965bd42013-09-26 14:48:15 -07001742void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1743 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001744
Jason Wangf8e617e2013-11-01 14:07:47 +08001745void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1746 unsigned int truesize);
1747
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001749#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1751
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001752#ifdef NET_SKBUFF_DATA_USES_OFFSET
1753static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1754{
1755 return skb->head + skb->tail;
1756}
1757
1758static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1759{
1760 skb->tail = skb->data - skb->head;
1761}
1762
1763static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1764{
1765 skb_reset_tail_pointer(skb);
1766 skb->tail += offset;
1767}
Simon Horman7cc46192013-05-28 20:34:29 +00001768
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001769#else /* NET_SKBUFF_DATA_USES_OFFSET */
1770static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1771{
1772 return skb->tail;
1773}
1774
1775static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1776{
1777 skb->tail = skb->data;
1778}
1779
1780static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1781{
1782 skb->tail = skb->data + offset;
1783}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001784
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001785#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1786
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787/*
1788 * Add data to an sk_buff
1789 */
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001790unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001791unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1793{
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001794 unsigned char *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795 SKB_LINEAR_ASSERT(skb);
1796 skb->tail += len;
1797 skb->len += len;
1798 return tmp;
1799}
1800
Joe Perches7965bd42013-09-26 14:48:15 -07001801unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1803{
1804 skb->data -= len;
1805 skb->len += len;
1806 return skb->data;
1807}
1808
Joe Perches7965bd42013-09-26 14:48:15 -07001809unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001810static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1811{
1812 skb->len -= len;
1813 BUG_ON(skb->len < skb->data_len);
1814 return skb->data += len;
1815}
1816
David S. Miller47d29642010-05-02 02:21:44 -07001817static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1818{
1819 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1820}
1821
Joe Perches7965bd42013-09-26 14:48:15 -07001822unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823
1824static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1825{
1826 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07001827 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 return NULL;
1829 skb->len -= len;
1830 return skb->data += len;
1831}
1832
1833static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1834{
1835 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1836}
1837
1838static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1839{
1840 if (likely(len <= skb_headlen(skb)))
1841 return 1;
1842 if (unlikely(len > skb->len))
1843 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07001844 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001845}
1846
1847/**
1848 * skb_headroom - bytes at buffer head
1849 * @skb: buffer to check
1850 *
1851 * Return the number of bytes of free space at the head of an &sk_buff.
1852 */
Chuck Leverc2636b42007-10-23 21:07:32 -07001853static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854{
1855 return skb->data - skb->head;
1856}
1857
1858/**
1859 * skb_tailroom - bytes at buffer end
1860 * @skb: buffer to check
1861 *
1862 * Return the number of bytes of free space at the tail of an sk_buff
1863 */
1864static inline int skb_tailroom(const struct sk_buff *skb)
1865{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001866 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001867}
1868
1869/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00001870 * skb_availroom - bytes at buffer end
1871 * @skb: buffer to check
1872 *
1873 * Return the number of bytes of free space at the tail of an sk_buff
1874 * allocated by sk_stream_alloc()
1875 */
1876static inline int skb_availroom(const struct sk_buff *skb)
1877{
Eric Dumazet16fad692013-03-14 05:40:32 +00001878 if (skb_is_nonlinear(skb))
1879 return 0;
1880
1881 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00001882}
1883
1884/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 * skb_reserve - adjust headroom
1886 * @skb: buffer to alter
1887 * @len: bytes to move
1888 *
1889 * Increase the headroom of an empty &sk_buff by reducing the tail
1890 * room. This is only allowed for an empty buffer.
1891 */
David S. Miller8243126c2006-01-17 02:54:21 -08001892static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893{
1894 skb->data += len;
1895 skb->tail += len;
1896}
1897
Tom Herbert8bce6d72014-09-29 20:22:29 -07001898#define ENCAP_TYPE_ETHER 0
1899#define ENCAP_TYPE_IPPROTO 1
1900
1901static inline void skb_set_inner_protocol(struct sk_buff *skb,
1902 __be16 protocol)
1903{
1904 skb->inner_protocol = protocol;
1905 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1906}
1907
1908static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1909 __u8 ipproto)
1910{
1911 skb->inner_ipproto = ipproto;
1912 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
1913}
1914
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001915static inline void skb_reset_inner_headers(struct sk_buff *skb)
1916{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00001917 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001918 skb->inner_network_header = skb->network_header;
1919 skb->inner_transport_header = skb->transport_header;
1920}
1921
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00001922static inline void skb_reset_mac_len(struct sk_buff *skb)
1923{
1924 skb->mac_len = skb->network_header - skb->mac_header;
1925}
1926
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001927static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1928 *skb)
1929{
1930 return skb->head + skb->inner_transport_header;
1931}
1932
1933static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1934{
1935 skb->inner_transport_header = skb->data - skb->head;
1936}
1937
1938static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1939 const int offset)
1940{
1941 skb_reset_inner_transport_header(skb);
1942 skb->inner_transport_header += offset;
1943}
1944
1945static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1946{
1947 return skb->head + skb->inner_network_header;
1948}
1949
1950static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1951{
1952 skb->inner_network_header = skb->data - skb->head;
1953}
1954
1955static inline void skb_set_inner_network_header(struct sk_buff *skb,
1956 const int offset)
1957{
1958 skb_reset_inner_network_header(skb);
1959 skb->inner_network_header += offset;
1960}
1961
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00001962static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1963{
1964 return skb->head + skb->inner_mac_header;
1965}
1966
1967static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1968{
1969 skb->inner_mac_header = skb->data - skb->head;
1970}
1971
1972static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1973 const int offset)
1974{
1975 skb_reset_inner_mac_header(skb);
1976 skb->inner_mac_header += offset;
1977}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00001978static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1979{
Cong Wang35d04612013-05-29 15:16:05 +08001980 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00001981}
1982
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07001983static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1984{
1985 return skb->head + skb->transport_header;
1986}
1987
1988static inline void skb_reset_transport_header(struct sk_buff *skb)
1989{
1990 skb->transport_header = skb->data - skb->head;
1991}
1992
1993static inline void skb_set_transport_header(struct sk_buff *skb,
1994 const int offset)
1995{
1996 skb_reset_transport_header(skb);
1997 skb->transport_header += offset;
1998}
1999
2000static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2001{
2002 return skb->head + skb->network_header;
2003}
2004
2005static inline void skb_reset_network_header(struct sk_buff *skb)
2006{
2007 skb->network_header = skb->data - skb->head;
2008}
2009
2010static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2011{
2012 skb_reset_network_header(skb);
2013 skb->network_header += offset;
2014}
2015
2016static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2017{
2018 return skb->head + skb->mac_header;
2019}
2020
2021static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2022{
Cong Wang35d04612013-05-29 15:16:05 +08002023 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002024}
2025
2026static inline void skb_reset_mac_header(struct sk_buff *skb)
2027{
2028 skb->mac_header = skb->data - skb->head;
2029}
2030
2031static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2032{
2033 skb_reset_mac_header(skb);
2034 skb->mac_header += offset;
2035}
2036
Timo Teräs0e3da5b2013-12-16 11:02:09 +02002037static inline void skb_pop_mac_header(struct sk_buff *skb)
2038{
2039 skb->mac_header = skb->network_header;
2040}
2041
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002042static inline void skb_probe_transport_header(struct sk_buff *skb,
2043 const int offset_hint)
2044{
2045 struct flow_keys keys;
2046
2047 if (skb_transport_header_was_set(skb))
2048 return;
Jiri Pirko06635a32015-05-12 14:56:16 +02002049 else if (skb_flow_dissect_flow_keys(skb, &keys))
Tom Herbert42aecaa2015-06-04 09:16:39 -07002050 skb_set_transport_header(skb, keys.control.thoff);
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002051 else
2052 skb_set_transport_header(skb, offset_hint);
2053}
2054
Eric Dumazet03606892012-02-23 10:55:02 +00002055static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2056{
2057 if (skb_mac_header_was_set(skb)) {
2058 const unsigned char *old_mac = skb_mac_header(skb);
2059
2060 skb_set_mac_header(skb, -skb->mac_len);
2061 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2062 }
2063}
2064
Michał Mirosław04fb4512010-12-14 15:24:08 +00002065static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2066{
2067 return skb->csum_start - skb_headroom(skb);
2068}
2069
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002070static inline int skb_transport_offset(const struct sk_buff *skb)
2071{
2072 return skb_transport_header(skb) - skb->data;
2073}
2074
2075static inline u32 skb_network_header_len(const struct sk_buff *skb)
2076{
2077 return skb->transport_header - skb->network_header;
2078}
2079
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002080static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2081{
2082 return skb->inner_transport_header - skb->inner_network_header;
2083}
2084
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002085static inline int skb_network_offset(const struct sk_buff *skb)
2086{
2087 return skb_network_header(skb) - skb->data;
2088}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03002089
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002090static inline int skb_inner_network_offset(const struct sk_buff *skb)
2091{
2092 return skb_inner_network_header(skb) - skb->data;
2093}
2094
Changli Gaof9599ce2010-08-04 04:43:44 +00002095static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2096{
2097 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2098}
2099
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100/*
2101 * CPUs often take a performance hit when accessing unaligned memory
2102 * locations. The actual performance hit varies, it can be small if the
2103 * hardware handles it or large if we have to take an exception and fix it
2104 * in software.
2105 *
2106 * Since an ethernet header is 14 bytes network drivers often end up with
2107 * the IP header at an unaligned offset. The IP header can be aligned by
2108 * shifting the start of the packet by 2 bytes. Drivers should do this
2109 * with:
2110 *
Tobias Klauser8660c122009-07-13 22:48:16 +00002111 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002112 *
2113 * The downside to this alignment of the IP header is that the DMA is now
2114 * unaligned. On some architectures the cost of an unaligned DMA is high
2115 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00002116 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002117 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2118 * to be overridden.
2119 */
2120#ifndef NET_IP_ALIGN
2121#define NET_IP_ALIGN 2
2122#endif
2123
Anton Blanchard025be812006-03-31 02:27:06 -08002124/*
2125 * The networking layer reserves some headroom in skb data (via
2126 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2127 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08002128 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08002129 *
2130 * Unfortunately this headroom changes the DMA alignment of the resulting
2131 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2132 * on some architectures. An architecture can override this value,
2133 * perhaps setting it to a cacheline in size (since that will maintain
2134 * cacheline alignment of the DMA). It must be a power of 2.
2135 *
David S. Millerd6301d32009-02-08 19:24:13 -08002136 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08002137 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07002138 *
2139 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2140 * to reduce average number of cache lines per packet.
2141 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07002142 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08002143 */
2144#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07002145#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08002146#endif
2147
Joe Perches7965bd42013-09-26 14:48:15 -07002148int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002149
2150static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2151{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00002152 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07002153 WARN_ON(1);
2154 return;
2155 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002156 skb->len = len;
2157 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158}
2159
Joe Perches7965bd42013-09-26 14:48:15 -07002160void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002161
2162static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2163{
Herbert Xu3cc0e872006-06-09 16:13:38 -07002164 if (skb->data_len)
2165 return ___pskb_trim(skb, len);
2166 __skb_trim(skb, len);
2167 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002168}
2169
2170static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2171{
2172 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2173}
2174
2175/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07002176 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2177 * @skb: buffer to alter
2178 * @len: new length
2179 *
2180 * This is identical to pskb_trim except that the caller knows that
2181 * the skb is not cloned so we should never get an error due to out-
2182 * of-memory.
2183 */
2184static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2185{
2186 int err = pskb_trim(skb, len);
2187 BUG_ON(err);
2188}
2189
2190/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002191 * skb_orphan - orphan a buffer
2192 * @skb: buffer to orphan
2193 *
2194 * If a buffer currently has an owner then we call the owner's
2195 * destructor function and make the @skb unowned. The buffer continues
2196 * to exist but is no longer charged to its former owner.
2197 */
2198static inline void skb_orphan(struct sk_buff *skb)
2199{
Eric Dumazetc34a7612013-07-30 16:11:15 -07002200 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002201 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002202 skb->destructor = NULL;
2203 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07002204 } else {
2205 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002206 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002207}
2208
2209/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002210 * skb_orphan_frags - orphan the frags contained in a buffer
2211 * @skb: buffer to orphan frags from
2212 * @gfp_mask: allocation mask for replacement pages
2213 *
2214 * For each frag in the SKB which needs a destructor (i.e. has an
2215 * owner) create a copy of that frag and release the original
2216 * page by calling the destructor.
2217 */
2218static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2219{
2220 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2221 return 0;
2222 return skb_copy_ubufs(skb, gfp_mask);
2223}
2224
2225/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002226 * __skb_queue_purge - empty a list
2227 * @list: list to empty
2228 *
2229 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2230 * the list and one reference dropped. This function does not take the
2231 * list lock and the caller must hold the relevant locks to use it.
2232 */
Joe Perches7965bd42013-09-26 14:48:15 -07002233void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234static inline void __skb_queue_purge(struct sk_buff_head *list)
2235{
2236 struct sk_buff *skb;
2237 while ((skb = __skb_dequeue(list)) != NULL)
2238 kfree_skb(skb);
2239}
2240
Joe Perches7965bd42013-09-26 14:48:15 -07002241void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242
Joe Perches7965bd42013-09-26 14:48:15 -07002243struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2244 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002245
2246/**
2247 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2248 * @dev: network device to receive on
2249 * @length: length to allocate
2250 *
2251 * Allocate a new &sk_buff and assign it a usage count of one. The
2252 * buffer has unspecified headroom built in. Users should allocate
2253 * the headroom they think they need without accounting for the
2254 * built in space. The built in space is used for optimisations.
2255 *
2256 * %NULL is returned if there is no free memory. Although this function
2257 * allocates memory it can be called from an interrupt.
2258 */
2259static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002260 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002261{
2262 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2263}
2264
Eric Dumazet6f532612012-05-18 05:12:12 +00002265/* legacy helper around __netdev_alloc_skb() */
2266static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2267 gfp_t gfp_mask)
2268{
2269 return __netdev_alloc_skb(NULL, length, gfp_mask);
2270}
2271
2272/* legacy helper around netdev_alloc_skb() */
2273static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2274{
2275 return netdev_alloc_skb(NULL, length);
2276}
2277
2278
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002279static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2280 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002281{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002282 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002283
2284 if (NET_IP_ALIGN && skb)
2285 skb_reserve(skb, NET_IP_ALIGN);
2286 return skb;
2287}
2288
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002289static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2290 unsigned int length)
2291{
2292 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2293}
2294
Alexander Duyck181edb22015-05-06 21:12:03 -07002295static inline void skb_free_frag(void *addr)
2296{
2297 __free_page_frag(addr);
2298}
2299
Alexander Duyckffde7322014-12-09 19:40:42 -08002300void *napi_alloc_frag(unsigned int fragsz);
Alexander Duyckfd11a832014-12-09 19:40:49 -08002301struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2302 unsigned int length, gfp_t gfp_mask);
2303static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2304 unsigned int length)
2305{
2306 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2307}
Alexander Duyckffde7322014-12-09 19:40:42 -08002308
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002309/**
Alexander Duyck71dfda52014-11-11 09:26:34 -08002310 * __dev_alloc_pages - allocate page for network Rx
2311 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2312 * @order: size of the allocation
2313 *
2314 * Allocate a new page.
2315 *
2316 * %NULL is returned if there is no free memory.
2317*/
2318static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2319 unsigned int order)
2320{
2321 /* This piece of code contains several assumptions.
2322 * 1. This is for device Rx, therefor a cold page is preferred.
2323 * 2. The expectation is the user wants a compound page.
2324 * 3. If requesting a order 0 page it will not be compound
2325 * due to the check to see if order has a value in prep_new_page
2326 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2327 * code in gfp_to_alloc_flags that should be enforcing this.
2328 */
2329 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2330
2331 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2332}
2333
2334static inline struct page *dev_alloc_pages(unsigned int order)
2335{
2336 return __dev_alloc_pages(GFP_ATOMIC, order);
2337}
2338
2339/**
2340 * __dev_alloc_page - allocate a page for network Rx
2341 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2342 *
2343 * Allocate a new page.
2344 *
2345 * %NULL is returned if there is no free memory.
2346 */
2347static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2348{
2349 return __dev_alloc_pages(gfp_mask, 0);
2350}
2351
2352static inline struct page *dev_alloc_page(void)
2353{
2354 return __dev_alloc_page(GFP_ATOMIC);
2355}
2356
2357/**
Mel Gorman06140022012-07-31 16:44:24 -07002358 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2359 * @page: The page that was allocated from skb_alloc_page
2360 * @skb: The skb that may need pfmemalloc set
2361 */
2362static inline void skb_propagate_pfmemalloc(struct page *page,
2363 struct sk_buff *skb)
2364{
Michal Hocko2f064f32015-08-21 14:11:51 -07002365 if (page_is_pfmemalloc(page))
Mel Gorman06140022012-07-31 16:44:24 -07002366 skb->pfmemalloc = true;
2367}
2368
Eric Dumazet564824b2010-10-11 19:05:25 +00002369/**
Masanari Iidae2278672014-02-18 22:54:36 +09002370 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002371 * @frag: the paged fragment
2372 *
2373 * Returns the &struct page associated with @frag.
2374 */
2375static inline struct page *skb_frag_page(const skb_frag_t *frag)
2376{
Ian Campbella8605c62011-10-19 23:01:49 +00002377 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002378}
2379
2380/**
2381 * __skb_frag_ref - take an addition reference on a paged fragment.
2382 * @frag: the paged fragment
2383 *
2384 * Takes an additional reference on the paged fragment @frag.
2385 */
2386static inline void __skb_frag_ref(skb_frag_t *frag)
2387{
2388 get_page(skb_frag_page(frag));
2389}
2390
2391/**
2392 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2393 * @skb: the buffer
2394 * @f: the fragment offset.
2395 *
2396 * Takes an additional reference on the @f'th paged fragment of @skb.
2397 */
2398static inline void skb_frag_ref(struct sk_buff *skb, int f)
2399{
2400 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2401}
2402
2403/**
2404 * __skb_frag_unref - release a reference on a paged fragment.
2405 * @frag: the paged fragment
2406 *
2407 * Releases a reference on the paged fragment @frag.
2408 */
2409static inline void __skb_frag_unref(skb_frag_t *frag)
2410{
2411 put_page(skb_frag_page(frag));
2412}
2413
2414/**
2415 * skb_frag_unref - release a reference on a paged fragment of an skb.
2416 * @skb: the buffer
2417 * @f: the fragment offset
2418 *
2419 * Releases a reference on the @f'th paged fragment of @skb.
2420 */
2421static inline void skb_frag_unref(struct sk_buff *skb, int f)
2422{
2423 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2424}
2425
2426/**
2427 * skb_frag_address - gets the address of the data contained in a paged fragment
2428 * @frag: the paged fragment buffer
2429 *
2430 * Returns the address of the data within @frag. The page must already
2431 * be mapped.
2432 */
2433static inline void *skb_frag_address(const skb_frag_t *frag)
2434{
2435 return page_address(skb_frag_page(frag)) + frag->page_offset;
2436}
2437
2438/**
2439 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2440 * @frag: the paged fragment buffer
2441 *
2442 * Returns the address of the data within @frag. Checks that the page
2443 * is mapped and returns %NULL otherwise.
2444 */
2445static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2446{
2447 void *ptr = page_address(skb_frag_page(frag));
2448 if (unlikely(!ptr))
2449 return NULL;
2450
2451 return ptr + frag->page_offset;
2452}
2453
2454/**
2455 * __skb_frag_set_page - sets the page contained in a paged fragment
2456 * @frag: the paged fragment
2457 * @page: the page to set
2458 *
2459 * Sets the fragment @frag to contain @page.
2460 */
2461static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2462{
Ian Campbella8605c62011-10-19 23:01:49 +00002463 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002464}
2465
2466/**
2467 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2468 * @skb: the buffer
2469 * @f: the fragment offset
2470 * @page: the page to set
2471 *
2472 * Sets the @f'th fragment of @skb to contain @page.
2473 */
2474static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2475 struct page *page)
2476{
2477 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2478}
2479
Eric Dumazet400dfd32013-10-17 16:27:07 -07002480bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2481
Ian Campbell131ea662011-08-19 06:25:00 +00002482/**
2483 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002484 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002485 * @frag: the paged fragment to map
2486 * @offset: the offset within the fragment (starting at the
2487 * fragment's own offset)
2488 * @size: the number of bytes to map
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002489 * @dir: the direction of the mapping (%PCI_DMA_*)
Ian Campbell131ea662011-08-19 06:25:00 +00002490 *
2491 * Maps the page associated with @frag to @device.
2492 */
2493static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2494 const skb_frag_t *frag,
2495 size_t offset, size_t size,
2496 enum dma_data_direction dir)
2497{
2498 return dma_map_page(dev, skb_frag_page(frag),
2499 frag->page_offset + offset, size, dir);
2500}
2501
Eric Dumazet117632e2011-12-03 21:39:53 +00002502static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2503 gfp_t gfp_mask)
2504{
2505 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2506}
2507
Octavian Purdilabad93e92014-06-12 01:36:26 +03002508
2509static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2510 gfp_t gfp_mask)
2511{
2512 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2513}
2514
2515
Ian Campbell131ea662011-08-19 06:25:00 +00002516/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002517 * skb_clone_writable - is the header of a clone writable
2518 * @skb: buffer to check
2519 * @len: length up to which to write
2520 *
2521 * Returns true if modifying the header part of the cloned buffer
2522 * does not requires the data to be copied.
2523 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002524static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002525{
2526 return !skb_header_cloned(skb) &&
2527 skb_headroom(skb) + len <= skb->hdr_len;
2528}
2529
Herbert Xud9cc2042007-09-16 16:21:16 -07002530static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2531 int cloned)
2532{
2533 int delta = 0;
2534
Herbert Xud9cc2042007-09-16 16:21:16 -07002535 if (headroom > skb_headroom(skb))
2536 delta = headroom - skb_headroom(skb);
2537
2538 if (delta || cloned)
2539 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2540 GFP_ATOMIC);
2541 return 0;
2542}
2543
Patrick McHardy334a8132007-06-25 04:35:20 -07002544/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002545 * skb_cow - copy header of skb when it is required
2546 * @skb: buffer to cow
2547 * @headroom: needed headroom
2548 *
2549 * If the skb passed lacks sufficient headroom or its data part
2550 * is shared, data is reallocated. If reallocation fails, an error
2551 * is returned and original skb is not changed.
2552 *
2553 * The result is skb with writable area skb->head...skb->tail
2554 * and at least @headroom of space at head.
2555 */
2556static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2557{
Herbert Xud9cc2042007-09-16 16:21:16 -07002558 return __skb_cow(skb, headroom, skb_cloned(skb));
2559}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560
Herbert Xud9cc2042007-09-16 16:21:16 -07002561/**
2562 * skb_cow_head - skb_cow but only making the head writable
2563 * @skb: buffer to cow
2564 * @headroom: needed headroom
2565 *
2566 * This function is identical to skb_cow except that we replace the
2567 * skb_cloned check by skb_header_cloned. It should be used when
2568 * you only need to push on some header and do not need to modify
2569 * the data.
2570 */
2571static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2572{
2573 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002574}
2575
2576/**
2577 * skb_padto - pad an skbuff up to a minimal size
2578 * @skb: buffer to pad
2579 * @len: minimal length
2580 *
2581 * Pads up a buffer to ensure the trailing bytes exist and are
2582 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002583 * is untouched. Otherwise it is extended. Returns zero on
2584 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002585 */
Herbert Xu5b057c62006-06-23 02:06:41 -07002586static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002587{
2588 unsigned int size = skb->len;
2589 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002590 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002591 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002592}
2593
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002594/**
2595 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2596 * @skb: buffer to pad
2597 * @len: minimal length
2598 *
2599 * Pads up a buffer to ensure the trailing bytes exist and are
2600 * blanked. If the buffer already contains sufficient data it
2601 * is untouched. Otherwise it is extended. Returns zero on
2602 * success. The skb is freed on error.
2603 */
2604static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2605{
2606 unsigned int size = skb->len;
2607
2608 if (unlikely(size < len)) {
2609 len -= size;
2610 if (skb_pad(skb, len))
2611 return -ENOMEM;
2612 __skb_put(skb, len);
2613 }
2614 return 0;
2615}
2616
Linus Torvalds1da177e2005-04-16 15:20:36 -07002617static inline int skb_add_data(struct sk_buff *skb,
Al Viroaf2b0402014-11-27 21:44:24 -05002618 struct iov_iter *from, int copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002619{
2620 const int off = skb->len;
2621
2622 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viroaf2b0402014-11-27 21:44:24 -05002623 __wsum csum = 0;
2624 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2625 &csum, from) == copy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626 skb->csum = csum_block_add(skb->csum, csum, off);
2627 return 0;
2628 }
Al Viroaf2b0402014-11-27 21:44:24 -05002629 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002630 return 0;
2631
2632 __skb_trim(skb, off);
2633 return -EFAULT;
2634}
2635
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002636static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2637 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002638{
2639 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002640 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002641
Ian Campbellea2ab692011-08-22 23:44:58 +00002642 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00002643 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002645 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002646}
2647
Herbert Xu364c6ba2006-06-09 16:10:40 -07002648static inline int __skb_linearize(struct sk_buff *skb)
2649{
2650 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2651}
2652
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653/**
2654 * skb_linearize - convert paged skb to linear one
2655 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07002656 *
2657 * If there is no free memory -ENOMEM is returned, otherwise zero
2658 * is returned and the old skb data released.
2659 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07002660static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002661{
Herbert Xu364c6ba2006-06-09 16:10:40 -07002662 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2663}
2664
2665/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00002666 * skb_has_shared_frag - can any frag be overwritten
2667 * @skb: buffer to test
2668 *
2669 * Return true if the skb has at least one frag that might be modified
2670 * by an external entity (as in vmsplice()/sendfile())
2671 */
2672static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2673{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00002674 return skb_is_nonlinear(skb) &&
2675 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002676}
2677
2678/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07002679 * skb_linearize_cow - make sure skb is linear and writable
2680 * @skb: buffer to process
2681 *
2682 * If there is no free memory -ENOMEM is returned, otherwise zero
2683 * is returned and the old skb data released.
2684 */
2685static inline int skb_linearize_cow(struct sk_buff *skb)
2686{
2687 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2688 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002689}
2690
2691/**
2692 * skb_postpull_rcsum - update checksum for received skb after pull
2693 * @skb: buffer to update
2694 * @start: start of data before pull
2695 * @len: length of data pulled
2696 *
2697 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07002698 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2699 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002700 */
2701
2702static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f2006-03-20 22:43:56 -08002703 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002704{
Patrick McHardy84fa7932006-08-29 16:44:56 -07002705 if (skb->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
2707}
2708
Herbert Xucbb042f2006-03-20 22:43:56 -08002709unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2710
David S. Miller7ce5a272013-12-02 17:26:05 -05002711/**
2712 * pskb_trim_rcsum - trim received skb and update checksum
2713 * @skb: buffer to trim
2714 * @len: new length
2715 *
2716 * This is exactly the same as pskb_trim except that it ensures the
2717 * checksum of received packets are still valid after the operation.
2718 */
2719
2720static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2721{
2722 if (likely(len >= skb->len))
2723 return 0;
2724 if (skb->ip_summed == CHECKSUM_COMPLETE)
2725 skb->ip_summed = CHECKSUM_NONE;
2726 return __pskb_trim(skb, len);
2727}
2728
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729#define skb_queue_walk(queue, skb) \
2730 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002731 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002732 skb = skb->next)
2733
James Chapman46f89142007-04-30 00:07:31 -07002734#define skb_queue_walk_safe(queue, skb, tmp) \
2735 for (skb = (queue)->next, tmp = skb->next; \
2736 skb != (struct sk_buff *)(queue); \
2737 skb = tmp, tmp = skb->next)
2738
David S. Miller1164f522008-09-23 00:49:44 -07002739#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002740 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07002741 skb = skb->next)
2742
2743#define skb_queue_walk_from_safe(queue, skb, tmp) \
2744 for (tmp = skb->next; \
2745 skb != (struct sk_buff *)(queue); \
2746 skb = tmp, tmp = skb->next)
2747
Stephen Hemminger300ce172005-10-30 13:47:34 -08002748#define skb_queue_reverse_walk(queue, skb) \
2749 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002750 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08002751 skb = skb->prev)
2752
David S. Miller686a2952011-01-20 22:47:32 -08002753#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2754 for (skb = (queue)->prev, tmp = skb->prev; \
2755 skb != (struct sk_buff *)(queue); \
2756 skb = tmp, tmp = skb->prev)
2757
2758#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2759 for (tmp = skb->prev; \
2760 skb != (struct sk_buff *)(queue); \
2761 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002762
David S. Miller21dc3302010-08-23 00:13:46 -07002763static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07002764{
2765 return skb_shinfo(skb)->frag_list != NULL;
2766}
2767
2768static inline void skb_frag_list_init(struct sk_buff *skb)
2769{
2770 skb_shinfo(skb)->frag_list = NULL;
2771}
2772
David S. Milleree039872009-06-09 00:17:13 -07002773#define skb_walk_frags(skb, iter) \
2774 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2775
Joe Perches7965bd42013-09-26 14:48:15 -07002776struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2777 int *peeked, int *off, int *err);
2778struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2779 int *err);
2780unsigned int datagram_poll(struct file *file, struct socket *sock,
2781 struct poll_table_struct *wait);
Al Viroc0371da2014-11-24 10:42:55 -05002782int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2783 struct iov_iter *to, int size);
David S. Miller51f3d022014-11-05 16:46:40 -05002784static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2785 struct msghdr *msg, int size)
2786{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002787 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
David S. Miller51f3d022014-11-05 16:46:40 -05002788}
Al Viroe5a4b0b2014-11-24 18:17:55 -05002789int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2790 struct msghdr *msg);
Al Viro3a654f92014-06-19 14:15:22 -04002791int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2792 struct iov_iter *from, int len);
Al Viro3a654f92014-06-19 14:15:22 -04002793int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
Joe Perches7965bd42013-09-26 14:48:15 -07002794void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2795void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2796int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07002797int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2798int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2799__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2800 int len, __wsum csum);
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002801ssize_t skb_socket_splice(struct sock *sk,
2802 struct pipe_inode_info *pipe,
2803 struct splice_pipe_desc *spd);
2804int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
Joe Perches7965bd42013-09-26 14:48:15 -07002805 struct pipe_inode_info *pipe, unsigned int len,
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002806 unsigned int flags,
2807 ssize_t (*splice_cb)(struct sock *,
2808 struct pipe_inode_info *,
2809 struct splice_pipe_desc *));
Joe Perches7965bd42013-09-26 14:48:15 -07002810void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01002811unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002812int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2813 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07002814void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2815int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2816void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01002817unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -07002818struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Vlad Yasevich0d5501c2014-08-08 14:42:13 -04002819struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
Jiri Pirkoe2195122014-11-19 14:05:01 +01002820int skb_ensure_writable(struct sk_buff *skb, int write_len);
Jiri Pirko93515d52014-11-19 14:05:02 +01002821int skb_vlan_pop(struct sk_buff *skb);
2822int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002823
Al Viro6ce8e9c2014-04-06 21:25:44 -04002824static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2825{
Al Viro21226ab2014-11-28 15:48:29 -05002826 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro6ce8e9c2014-04-06 21:25:44 -04002827}
2828
Al Viro7eab8d92014-04-06 21:51:23 -04002829static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2830{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002831 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro7eab8d92014-04-06 21:51:23 -04002832}
2833
Daniel Borkmann2817a332013-10-30 11:50:51 +01002834struct skb_checksum_ops {
2835 __wsum (*update)(const void *mem, int len, __wsum wsum);
2836 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2837};
2838
2839__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2840 __wsum csum, const struct skb_checksum_ops *ops);
2841__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2842 __wsum csum);
2843
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07002844static inline void * __must_check
2845__skb_header_pointer(const struct sk_buff *skb, int offset,
2846 int len, void *data, int hlen, void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847{
Patrick McHardy55820ee2005-07-05 14:08:10 -07002848 if (hlen - offset >= len)
David S. Miller690e36e2014-08-23 12:13:41 -07002849 return data + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850
David S. Miller690e36e2014-08-23 12:13:41 -07002851 if (!skb ||
2852 skb_copy_bits(skb, offset, buffer, len) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002853 return NULL;
2854
2855 return buffer;
2856}
2857
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07002858static inline void * __must_check
2859skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
David S. Miller690e36e2014-08-23 12:13:41 -07002860{
2861 return __skb_header_pointer(skb, offset, len, skb->data,
2862 skb_headlen(skb), buffer);
2863}
2864
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01002865/**
2866 * skb_needs_linearize - check if we need to linearize a given skb
2867 * depending on the given device features.
2868 * @skb: socket buffer to check
2869 * @features: net device features
2870 *
2871 * Returns true if either:
2872 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2873 * 2. skb is fragmented and the device does not support SG.
2874 */
2875static inline bool skb_needs_linearize(struct sk_buff *skb,
2876 netdev_features_t features)
2877{
2878 return skb_is_nonlinear(skb) &&
2879 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2880 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2881}
2882
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002883static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2884 void *to,
2885 const unsigned int len)
2886{
2887 memcpy(to, skb->data, len);
2888}
2889
2890static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2891 const int offset, void *to,
2892 const unsigned int len)
2893{
2894 memcpy(to, skb->data + offset, len);
2895}
2896
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03002897static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2898 const void *from,
2899 const unsigned int len)
2900{
2901 memcpy(skb->data, from, len);
2902}
2903
2904static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2905 const int offset,
2906 const void *from,
2907 const unsigned int len)
2908{
2909 memcpy(skb->data + offset, from, len);
2910}
2911
Joe Perches7965bd42013-09-26 14:48:15 -07002912void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913
Patrick Ohlyac45f602009-02-12 05:03:37 +00002914static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2915{
2916 return skb->tstamp;
2917}
2918
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002919/**
2920 * skb_get_timestamp - get timestamp from a skb
2921 * @skb: skb to get stamp from
2922 * @stamp: pointer to struct timeval to store stamp in
2923 *
2924 * Timestamps are stored in the skb as offsets to a base timestamp.
2925 * This function converts the offset back to a struct timeval and stores
2926 * it in stamp.
2927 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00002928static inline void skb_get_timestamp(const struct sk_buff *skb,
2929 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002930{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002931 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002932}
2933
Patrick Ohlyac45f602009-02-12 05:03:37 +00002934static inline void skb_get_timestampns(const struct sk_buff *skb,
2935 struct timespec *stamp)
2936{
2937 *stamp = ktime_to_timespec(skb->tstamp);
2938}
2939
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002940static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002941{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002942 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002943}
2944
Stephen Hemminger164891a2007-04-23 22:26:16 -07002945static inline ktime_t net_timedelta(ktime_t t)
2946{
2947 return ktime_sub(ktime_get_real(), t);
2948}
2949
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07002950static inline ktime_t net_invalid_timestamp(void)
2951{
2952 return ktime_set(0, 0);
2953}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002954
Alexander Duyck62bccb82014-09-04 13:31:35 -04002955struct sk_buff *skb_clone_sk(struct sk_buff *skb);
2956
Richard Cochranc1f19b52010-07-17 08:49:36 +00002957#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2958
Joe Perches7965bd42013-09-26 14:48:15 -07002959void skb_clone_tx_timestamp(struct sk_buff *skb);
2960bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00002961
2962#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2963
2964static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2965{
2966}
2967
2968static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2969{
2970 return false;
2971}
2972
2973#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2974
2975/**
2976 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2977 *
Richard Cochranda92b192011-10-21 00:49:15 +00002978 * PHY drivers may accept clones of transmitted packets for
2979 * timestamping via their phy_driver.txtstamp method. These drivers
Benjamin Poirier7a76a022015-08-07 09:32:21 -07002980 * must call this function to return the skb back to the stack with a
2981 * timestamp.
Richard Cochranda92b192011-10-21 00:49:15 +00002982 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00002983 * @skb: clone of the the original outgoing packet
Benjamin Poirier7a76a022015-08-07 09:32:21 -07002984 * @hwtstamps: hardware time stamps
Richard Cochranc1f19b52010-07-17 08:49:36 +00002985 *
2986 */
2987void skb_complete_tx_timestamp(struct sk_buff *skb,
2988 struct skb_shared_hwtstamps *hwtstamps);
2989
Willem de Bruijne7fd2882014-08-04 22:11:48 -04002990void __skb_tstamp_tx(struct sk_buff *orig_skb,
2991 struct skb_shared_hwtstamps *hwtstamps,
2992 struct sock *sk, int tstype);
2993
Patrick Ohlyac45f602009-02-12 05:03:37 +00002994/**
2995 * skb_tstamp_tx - queue clone of skb with send time stamps
2996 * @orig_skb: the original outgoing packet
2997 * @hwtstamps: hardware time stamps, may be NULL if not available
2998 *
2999 * If the skb has a socket associated, then this function clones the
3000 * skb (thus sharing the actual data and optional structures), stores
3001 * the optional hardware time stamping information (if non NULL) or
3002 * generates a software time stamp (otherwise), then queues the clone
3003 * to the error queue of the socket. Errors are silently ignored.
3004 */
Joe Perches7965bd42013-09-26 14:48:15 -07003005void skb_tstamp_tx(struct sk_buff *orig_skb,
3006 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00003007
Richard Cochran4507a712010-07-17 08:48:28 +00003008static inline void sw_tx_timestamp(struct sk_buff *skb)
3009{
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003010 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3011 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
Richard Cochran4507a712010-07-17 08:48:28 +00003012 skb_tstamp_tx(skb, NULL);
3013}
3014
3015/**
3016 * skb_tx_timestamp() - Driver hook for transmit timestamping
3017 *
3018 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00003019 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00003020 *
David S. Miller73409f32013-12-27 13:04:33 -05003021 * Specifically, one should make absolutely sure that this function is
3022 * called before TX completion of this packet can trigger. Otherwise
3023 * the packet could potentially already be freed.
3024 *
Richard Cochran4507a712010-07-17 08:48:28 +00003025 * @skb: A socket buffer.
3026 */
3027static inline void skb_tx_timestamp(struct sk_buff *skb)
3028{
Richard Cochranc1f19b52010-07-17 08:49:36 +00003029 skb_clone_tx_timestamp(skb);
Richard Cochran4507a712010-07-17 08:48:28 +00003030 sw_tx_timestamp(skb);
3031}
3032
Johannes Berg6e3e9392011-11-09 10:15:42 +01003033/**
3034 * skb_complete_wifi_ack - deliver skb with wifi status
3035 *
3036 * @skb: the original outgoing packet
3037 * @acked: ack status
3038 *
3039 */
3040void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3041
Joe Perches7965bd42013-09-26 14:48:15 -07003042__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3043__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003044
Herbert Xu60476372007-04-09 11:59:39 -07003045static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3046{
Tom Herbert6edec0e2015-02-10 16:30:28 -08003047 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3048 skb->csum_valid ||
3049 (skb->ip_summed == CHECKSUM_PARTIAL &&
3050 skb_checksum_start_offset(skb) >= 0));
Herbert Xu60476372007-04-09 11:59:39 -07003051}
3052
Herbert Xufb286bb2005-11-10 13:01:24 -08003053/**
3054 * skb_checksum_complete - Calculate checksum of an entire packet
3055 * @skb: packet to process
3056 *
3057 * This function calculates the checksum over the entire packet plus
3058 * the value of skb->csum. The latter can be used to supply the
3059 * checksum of a pseudo header as used by TCP/UDP. It returns the
3060 * checksum.
3061 *
3062 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3063 * this function can be used to verify that checksum on received
3064 * packets. In that case the function should return zero if the
3065 * checksum is correct. In particular, this function will return zero
3066 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3067 * hardware has already verified the correctness of the checksum.
3068 */
Al Viro4381ca32007-07-15 21:00:11 +01003069static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08003070{
Herbert Xu60476372007-04-09 11:59:39 -07003071 return skb_csum_unnecessary(skb) ?
3072 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003073}
3074
Tom Herbert77cffe22014-08-27 21:26:46 -07003075static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3076{
3077 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3078 if (skb->csum_level == 0)
3079 skb->ip_summed = CHECKSUM_NONE;
3080 else
3081 skb->csum_level--;
3082 }
3083}
3084
3085static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3086{
3087 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3088 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3089 skb->csum_level++;
3090 } else if (skb->ip_summed == CHECKSUM_NONE) {
3091 skb->ip_summed = CHECKSUM_UNNECESSARY;
3092 skb->csum_level = 0;
3093 }
3094}
3095
Tom Herbert5a212322014-08-31 15:12:41 -07003096static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3097{
3098 /* Mark current checksum as bad (typically called from GRO
3099 * path). In the case that ip_summed is CHECKSUM_NONE
3100 * this must be the first checksum encountered in the packet.
3101 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3102 * checksum after the last one validated. For UDP, a zero
3103 * checksum can not be marked as bad.
3104 */
3105
3106 if (skb->ip_summed == CHECKSUM_NONE ||
3107 skb->ip_summed == CHECKSUM_UNNECESSARY)
3108 skb->csum_bad = 1;
3109}
3110
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003111/* Check if we need to perform checksum complete validation.
3112 *
3113 * Returns true if checksum complete is needed, false otherwise
3114 * (either checksum is unnecessary or zero checksum is allowed).
3115 */
3116static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3117 bool zero_okay,
3118 __sum16 check)
3119{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003120 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3121 skb->csum_valid = 1;
Tom Herbert77cffe22014-08-27 21:26:46 -07003122 __skb_decr_checksum_unnecessary(skb);
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003123 return false;
3124 }
3125
3126 return true;
3127}
3128
3129/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3130 * in checksum_init.
3131 */
3132#define CHECKSUM_BREAK 76
3133
Tom Herbert4e18b9a2015-04-20 14:10:04 -07003134/* Unset checksum-complete
3135 *
3136 * Unset checksum complete can be done when packet is being modified
3137 * (uncompressed for instance) and checksum-complete value is
3138 * invalidated.
3139 */
3140static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3141{
3142 if (skb->ip_summed == CHECKSUM_COMPLETE)
3143 skb->ip_summed = CHECKSUM_NONE;
3144}
3145
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003146/* Validate (init) checksum based on checksum complete.
3147 *
3148 * Return values:
3149 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3150 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3151 * checksum is stored in skb->csum for use in __skb_checksum_complete
3152 * non-zero: value of invalid checksum
3153 *
3154 */
3155static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3156 bool complete,
3157 __wsum psum)
3158{
3159 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3160 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003161 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003162 return 0;
3163 }
Tom Herbert5a212322014-08-31 15:12:41 -07003164 } else if (skb->csum_bad) {
3165 /* ip_summed == CHECKSUM_NONE in this case */
Eric Dumazetc91d4602015-05-15 05:48:07 -07003166 return (__force __sum16)1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003167 }
3168
3169 skb->csum = psum;
3170
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003171 if (complete || skb->len <= CHECKSUM_BREAK) {
3172 __sum16 csum;
3173
3174 csum = __skb_checksum_complete(skb);
3175 skb->csum_valid = !csum;
3176 return csum;
3177 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003178
3179 return 0;
3180}
3181
3182static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3183{
3184 return 0;
3185}
3186
3187/* Perform checksum validate (init). Note that this is a macro since we only
3188 * want to calculate the pseudo header which is an input function if necessary.
3189 * First we try to validate without any computation (checksum unnecessary) and
3190 * then calculate based on checksum complete calling the function to compute
3191 * pseudo header.
3192 *
3193 * Return values:
3194 * 0: checksum is validated or try to in skb_checksum_complete
3195 * non-zero: value of invalid checksum
3196 */
3197#define __skb_checksum_validate(skb, proto, complete, \
3198 zero_okay, check, compute_pseudo) \
3199({ \
3200 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003201 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003202 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3203 __ret = __skb_checksum_validate_complete(skb, \
3204 complete, compute_pseudo(skb, proto)); \
3205 __ret; \
3206})
3207
3208#define skb_checksum_init(skb, proto, compute_pseudo) \
3209 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3210
3211#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3212 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3213
3214#define skb_checksum_validate(skb, proto, compute_pseudo) \
3215 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3216
3217#define skb_checksum_validate_zero_check(skb, proto, check, \
3218 compute_pseudo) \
Sabrina Dubroca096a4cf2015-02-06 18:54:19 +01003219 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003220
3221#define skb_checksum_simple_validate(skb) \
3222 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3223
Tom Herbertd96535a2014-08-31 15:12:42 -07003224static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3225{
3226 return (skb->ip_summed == CHECKSUM_NONE &&
3227 skb->csum_valid && !skb->csum_bad);
3228}
3229
3230static inline void __skb_checksum_convert(struct sk_buff *skb,
3231 __sum16 check, __wsum pseudo)
3232{
3233 skb->csum = ~pseudo;
3234 skb->ip_summed = CHECKSUM_COMPLETE;
3235}
3236
3237#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3238do { \
3239 if (__skb_checksum_convert_check(skb)) \
3240 __skb_checksum_convert(skb, check, \
3241 compute_pseudo(skb, proto)); \
3242} while (0)
3243
Tom Herbert15e23962015-02-10 16:30:31 -08003244static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3245 u16 start, u16 offset)
3246{
3247 skb->ip_summed = CHECKSUM_PARTIAL;
3248 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3249 skb->csum_offset = offset - start;
3250}
3251
Tom Herbertdcdc8992015-02-02 16:07:34 -08003252/* Update skbuf and packet to reflect the remote checksum offload operation.
3253 * When called, ptr indicates the starting point for skb->csum when
3254 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3255 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3256 */
3257static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
Tom Herbert15e23962015-02-10 16:30:31 -08003258 int start, int offset, bool nopartial)
Tom Herbertdcdc8992015-02-02 16:07:34 -08003259{
3260 __wsum delta;
3261
Tom Herbert15e23962015-02-10 16:30:31 -08003262 if (!nopartial) {
3263 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3264 return;
3265 }
3266
Tom Herbertdcdc8992015-02-02 16:07:34 -08003267 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3268 __skb_checksum_complete(skb);
3269 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3270 }
3271
3272 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3273
3274 /* Adjust skb->csum since we changed the packet */
3275 skb->csum = csum_add(skb->csum, delta);
3276}
3277
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003278#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07003279void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003280static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3281{
3282 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07003283 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003284}
3285static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3286{
3287 if (nfct)
3288 atomic_inc(&nfct->use);
3289}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003290#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003291#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003292static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3293{
3294 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3295 kfree(nf_bridge);
3296}
3297static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3298{
3299 if (nf_bridge)
3300 atomic_inc(&nf_bridge->use);
3301}
3302#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08003303static inline void nf_reset(struct sk_buff *skb)
3304{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003305#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003306 nf_conntrack_put(skb->nfct);
3307 skb->nfct = NULL;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003308#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003309#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003310 nf_bridge_put(skb->nf_bridge);
3311 skb->nf_bridge = NULL;
3312#endif
3313}
3314
Patrick McHardy124dff02013-04-05 20:42:05 +02003315static inline void nf_reset_trace(struct sk_buff *skb)
3316{
Florian Westphal478b3602014-02-15 23:48:45 +01003317#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00003318 skb->nf_trace = 0;
3319#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003320}
3321
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003322/* Note: This doesn't put any conntrack and bridge info in dst. */
Eric Dumazetb1937222014-09-28 22:18:47 -07003323static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3324 bool copy)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003325{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003326#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003327 dst->nfct = src->nfct;
3328 nf_conntrack_get(src->nfct);
Eric Dumazetb1937222014-09-28 22:18:47 -07003329 if (copy)
3330 dst->nfctinfo = src->nfctinfo;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003331#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003332#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003333 dst->nf_bridge = src->nf_bridge;
3334 nf_bridge_get(src->nf_bridge);
3335#endif
Florian Westphal478b3602014-02-15 23:48:45 +01003336#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Eric Dumazetb1937222014-09-28 22:18:47 -07003337 if (copy)
3338 dst->nf_trace = src->nf_trace;
Florian Westphal478b3602014-02-15 23:48:45 +01003339#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003340}
3341
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003342static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3343{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003344#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003345 nf_conntrack_put(dst->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003346#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003347#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003348 nf_bridge_put(dst->nf_bridge);
3349#endif
Eric Dumazetb1937222014-09-28 22:18:47 -07003350 __nf_copy(dst, src, true);
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003351}
3352
James Morris984bc162006-06-09 00:29:17 -07003353#ifdef CONFIG_NETWORK_SECMARK
3354static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3355{
3356 to->secmark = from->secmark;
3357}
3358
3359static inline void skb_init_secmark(struct sk_buff *skb)
3360{
3361 skb->secmark = 0;
3362}
3363#else
3364static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3365{ }
3366
3367static inline void skb_init_secmark(struct sk_buff *skb)
3368{ }
3369#endif
3370
Eric W. Biederman574f7192014-04-01 12:20:24 -07003371static inline bool skb_irq_freeable(const struct sk_buff *skb)
3372{
3373 return !skb->destructor &&
3374#if IS_ENABLED(CONFIG_XFRM)
3375 !skb->sp &&
3376#endif
3377#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3378 !skb->nfct &&
3379#endif
3380 !skb->_skb_refdst &&
3381 !skb_has_frag_list(skb);
3382}
3383
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003384static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3385{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003386 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003387}
3388
Stephen Hemminger92477442009-03-21 13:39:26 -07003389static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003390{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003391 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003392}
3393
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003394static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3395{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003396 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003397}
3398
David S. Millerd5a9e242009-01-27 16:22:11 -08003399static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3400{
3401 skb->queue_mapping = rx_queue + 1;
3402}
3403
Stephen Hemminger92477442009-03-21 13:39:26 -07003404static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003405{
3406 return skb->queue_mapping - 1;
3407}
3408
Stephen Hemminger92477442009-03-21 13:39:26 -07003409static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003410{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003411 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003412}
3413
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003414static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3415{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003416#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003417 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003418#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003419 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003420#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003421}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003422
Pravin B Shelar68c33162013-02-14 14:02:41 +00003423/* Keeps track of mac header offset relative to skb->head.
3424 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3425 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003426 * tunnel skb it points to outer mac header.
3427 * Keeps track of level of encapsulation of network headers.
3428 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003429struct skb_gso_cb {
Eric Dumazet3347c962013-10-19 11:42:56 -07003430 int mac_offset;
3431 int encap_level;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003432 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003433};
3434#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3435
3436static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3437{
3438 return (skb_mac_header(inner_skb) - inner_skb->head) -
3439 SKB_GSO_CB(inner_skb)->mac_offset;
3440}
3441
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003442static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3443{
3444 int new_headroom, headroom;
3445 int ret;
3446
3447 headroom = skb_headroom(skb);
3448 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3449 if (ret)
3450 return ret;
3451
3452 new_headroom = skb_headroom(skb);
3453 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3454 return 0;
3455}
3456
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003457/* Compute the checksum for a gso segment. First compute the checksum value
3458 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3459 * then add in skb->csum (checksum from csum_start to end of packet).
3460 * skb->csum and csum_start are then updated to reflect the checksum of the
3461 * resultant packet starting from the transport header-- the resultant checksum
3462 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3463 * header.
3464 */
3465static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3466{
3467 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
Eric Dumazetc91d4602015-05-15 05:48:07 -07003468 skb_transport_offset(skb);
3469 __wsum partial;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003470
Eric Dumazetc91d4602015-05-15 05:48:07 -07003471 partial = csum_partial(skb_transport_header(skb), plen, skb->csum);
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003472 skb->csum = res;
3473 SKB_GSO_CB(skb)->csum_start -= plen;
3474
Eric Dumazetc91d4602015-05-15 05:48:07 -07003475 return csum_fold(partial);
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003476}
3477
David S. Millerbdcc0922012-03-07 20:53:36 -05003478static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07003479{
3480 return skb_shinfo(skb)->gso_size;
3481}
3482
Eric Dumazet36a8f392013-09-29 01:21:32 -07003483/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05003484static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02003485{
3486 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3487}
3488
Joe Perches7965bd42013-09-26 14:48:15 -07003489void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07003490
3491static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3492{
3493 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3494 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003495 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3496
Alexander Duyckb78462e2010-06-02 12:24:37 +00003497 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3498 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07003499 __skb_warn_lro_forwarding(skb);
3500 return true;
3501 }
3502 return false;
3503}
3504
Herbert Xu35fc92a2007-03-26 23:22:20 -07003505static inline void skb_forward_csum(struct sk_buff *skb)
3506{
3507 /* Unfortunately we don't support this one. Any brave souls? */
3508 if (skb->ip_summed == CHECKSUM_COMPLETE)
3509 skb->ip_summed = CHECKSUM_NONE;
3510}
3511
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003512/**
3513 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3514 * @skb: skb to check
3515 *
3516 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3517 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3518 * use this helper, to document places where we make this assertion.
3519 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003520static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003521{
3522#ifdef DEBUG
3523 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3524#endif
3525}
3526
Rusty Russellf35d9d82008-02-04 23:49:54 -05003527bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00003528
Paul Durranted1f50c2014-01-09 10:02:46 +00003529int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
Linus Lüssing9afd85c2015-05-02 14:01:07 +02003530struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3531 unsigned int transport_len,
3532 __sum16(*skb_chkf)(struct sk_buff *skb));
Paul Durranted1f50c2014-01-09 10:02:46 +00003533
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00003534/**
3535 * skb_head_is_locked - Determine if the skb->head is locked down
3536 * @skb: skb to check
3537 *
3538 * The head on skbs build around a head frag can be removed if they are
3539 * not cloned. This function returns true if the skb head is locked down
3540 * due to either being allocated via kmalloc, or by being a clone with
3541 * multiple references to the head.
3542 */
3543static inline bool skb_head_is_locked(const struct sk_buff *skb)
3544{
3545 return !skb->head_frag || skb_cloned(skb);
3546}
Florian Westphalfe6cc552014-02-13 23:09:12 +01003547
3548/**
3549 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3550 *
3551 * @skb: GSO skb
3552 *
3553 * skb_gso_network_seglen is used to determine the real size of the
3554 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3555 *
3556 * The MAC/L2 header is not accounted for.
3557 */
3558static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3559{
3560 unsigned int hdr_len = skb_transport_header(skb) -
3561 skb_network_header(skb);
3562 return hdr_len + skb_gso_transport_seglen(skb);
3563}
Thomas Grafee122c72015-07-21 10:43:58 +02003564
Linus Torvalds1da177e2005-04-16 15:20:36 -07003565#endif /* __KERNEL__ */
3566#endif /* _LINUX_SKBUFF_H */