blob: af4f6ac025b6b360be51363d4e0fc3652b76040f [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 Westphal63cdbc02015-09-14 17:06:27 +0200182
183 /* always valid & non-NULL from FORWARD on, for physdev match */
184 struct net_device *physoutdev;
Florian Westphal7fb48c52015-05-03 22:05:28 +0200185 union {
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200186 /* prerouting: detect dnat in orig/reply direction */
Bernhard Thaler72b31f72015-05-30 15:27:40 +0200187 __be32 ipv4_daddr;
188 struct in6_addr ipv6_daddr;
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200189
190 /* after prerouting + nat detected: store original source
191 * mac since neigh resolution overwrites it, only used while
192 * skb is out in neigh layer.
193 */
194 char neigh_header[8];
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
Yuchung Cheng625a5e12015-10-16 21:57:45 -0700466static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
467 const struct skb_mstamp *t0)
468{
469 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
470
471 if (!diff)
472 diff = t1->stamp_us - t0->stamp_us;
473 return diff > 0;
474}
Eric Dumazet363ec39232014-02-26 14:02:11 -0800475
Linus Torvalds1da177e2005-04-16 15:20:36 -0700476/**
477 * struct sk_buff - socket buffer
478 * @next: Next buffer in list
479 * @prev: Previous buffer in list
Eric Dumazet363ec39232014-02-26 14:02:11 -0800480 * @tstamp: Time we arrived/left
Eric Dumazet56b17422014-11-03 08:19:53 -0800481 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700482 * @sk: Socket we are owned by
Linus Torvalds1da177e2005-04-16 15:20:36 -0700483 * @dev: Device we arrived on/are leaving by
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700484 * @cb: Control buffer. Free for use by every layer. Put private vars here
Eric Dumazet7fee2262010-05-11 23:19:48 +0000485 * @_skb_refdst: destination entry (with norefcount bit)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700486 * @sp: the security path, used for xfrm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700487 * @len: Length of actual data
488 * @data_len: Data length
489 * @mac_len: Length of link layer header
Patrick McHardy334a8132007-06-25 04:35:20 -0700490 * @hdr_len: writable header length of cloned skb
Herbert Xu663ead32007-04-09 11:59:07 -0700491 * @csum: Checksum (must include start/offset pair)
492 * @csum_start: Offset from skb->head where checksumming should start
493 * @csum_offset: Offset from csum_start where checksum should be stored
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700494 * @priority: Packet queueing priority
WANG Cong60ff7462014-05-04 16:39:18 -0700495 * @ignore_df: allow local fragmentation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700496 * @cloned: Head may be cloned (check refcnt to be sure)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700497 * @ip_summed: Driver fed us an IP checksum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700498 * @nohdr: Payload reference only, must not modify header
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700499 * @nfctinfo: Relationship of this skb to the connection
Linus Torvalds1da177e2005-04-16 15:20:36 -0700500 * @pkt_type: Packet class
Randy Dunlapc83c2482005-10-18 22:07:41 -0700501 * @fclone: skbuff clone status
Randy Dunlapc83c2482005-10-18 22:07:41 -0700502 * @ipvs_property: skbuff is owned by ipvs
Randy Dunlap31729362008-02-18 20:52:13 -0800503 * @peeked: this packet has been seen already, so stats have been
504 * done for it, don't do them again
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700505 * @nf_trace: netfilter packet trace flag
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700506 * @protocol: Packet protocol from driver
507 * @destructor: Destruct function
508 * @nfct: Associated connection, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700509 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
Eric Dumazet8964be42009-11-20 15:35:04 -0800510 * @skb_iif: ifindex of device we arrived on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700511 * @tc_index: Traffic control index
512 * @tc_verd: traffic control verdict
Tom Herbert61b905d2014-03-24 15:34:47 -0700513 * @hash: the packet hash
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700514 * @queue_mapping: Queue mapping for multiqueue devices
David S. Miller0b725a22014-08-25 15:51:53 -0700515 * @xmit_more: More SKBs are pending for this queue
Randy Dunlap553a5672008-04-20 10:51:01 -0700516 * @ndisc_nodetype: router type (from link layer)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700517 * @ooo_okay: allow the mapping of a socket to a queue to be changed
Tom Herbert61b905d2014-03-24 15:34:47 -0700518 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
Changli Gao4ca24622011-08-19 07:26:44 -0700519 * ports.
Tom Herberta3b18dd2014-07-01 21:33:17 -0700520 * @sw_hash: indicates hash was computed in software stack
Johannes Berg6e3e9392011-11-09 10:15:42 +0100521 * @wifi_acked_valid: wifi_acked was set
522 * @wifi_acked: whether frame was acked on wifi or not
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000523 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
Eliezer Tamir06021292013-06-10 11:39:50 +0300524 * @napi_id: id of the NAPI struct this skb came from
James Morris984bc162006-06-09 00:29:17 -0700525 * @secmark: security marking
Scott Feldman0c4f6912015-07-18 18:24:48 -0700526 * @offload_fwd_mark: fwding offload mark
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700527 * @mark: Generic packet mark
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000528 * @vlan_proto: vlan encapsulation protocol
Patrick McHardy6aa895b2008-07-14 22:49:06 -0700529 * @vlan_tci: vlan tag control information
Simon Horman0d89d202013-05-23 21:02:52 +0000530 * @inner_protocol: Protocol (encapsulation)
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000531 * @inner_transport_header: Inner transport layer header (encapsulation)
532 * @inner_network_header: Network layer header (encapsulation)
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000533 * @inner_mac_header: Link layer header (encapsulation)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700534 * @transport_header: Transport layer header
535 * @network_header: Network layer header
536 * @mac_header: Link layer header
537 * @tail: Tail pointer
538 * @end: End pointer
539 * @head: Head of buffer
540 * @data: Data head pointer
541 * @truesize: Buffer size
542 * @users: User count - see {datagram,tcp}.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700543 */
544
545struct sk_buff {
Eric Dumazet363ec39232014-02-26 14:02:11 -0800546 union {
Eric Dumazet56b17422014-11-03 08:19:53 -0800547 struct {
548 /* These two members must be first. */
549 struct sk_buff *next;
550 struct sk_buff *prev;
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000551
Eric Dumazet56b17422014-11-03 08:19:53 -0800552 union {
553 ktime_t tstamp;
554 struct skb_mstamp skb_mstamp;
555 };
556 };
557 struct rb_node rbnode; /* used in netem & tcp stack */
558 };
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000559 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700560 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562 /*
563 * This is the control buffer. It is free to use for every
564 * layer. Please put your private variables there. If you
565 * want to keep them across layers you have to do a skb_clone()
566 * first. This is owned by whoever has the skb queued ATM.
567 */
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000568 char cb[48] __aligned(8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569
Eric Dumazet7fee2262010-05-11 23:19:48 +0000570 unsigned long _skb_refdst;
Eric Dumazetb1937222014-09-28 22:18:47 -0700571 void (*destructor)(struct sk_buff *skb);
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000572#ifdef CONFIG_XFRM
573 struct sec_path *sp;
574#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700575#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
576 struct nf_conntrack *nfct;
577#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200578#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Eric Dumazetb1937222014-09-28 22:18:47 -0700579 struct nf_bridge_info *nf_bridge;
580#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 unsigned int len,
Patrick McHardy334a8132007-06-25 04:35:20 -0700582 data_len;
583 __u16 mac_len,
584 hdr_len;
Eric Dumazetb1937222014-09-28 22:18:47 -0700585
586 /* Following fields are _not_ copied in __copy_skb_header()
587 * Note that queue_mapping is here mostly to fill a hole.
588 */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200589 kmemcheck_bitfield_begin(flags1);
Eric Dumazetb1937222014-09-28 22:18:47 -0700590 __u16 queue_mapping;
591 __u8 cloned:1,
Harald Welte6869c4d2005-08-09 19:24:19 -0700592 nohdr:1,
Patrick McHardyb84f4cc2005-11-20 21:19:21 -0800593 fclone:2,
Herbert Xua59322b2007-12-05 01:53:40 -0800594 peeked:1,
Eric Dumazetb1937222014-09-28 22:18:47 -0700595 head_frag:1,
596 xmit_more:1;
597 /* one bit hole */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200598 kmemcheck_bitfield_end(flags1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700599
Eric Dumazetb1937222014-09-28 22:18:47 -0700600 /* fields enclosed in headers_start/headers_end are copied
601 * using a single memcpy() in __copy_skb_header()
602 */
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700603 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700604 __u32 headers_start[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700605 /* public: */
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200606
607/* if you move pkt_type around you also must adapt those constants */
608#ifdef __BIG_ENDIAN_BITFIELD
609#define PKT_TYPE_MAX (7 << 5)
610#else
611#define PKT_TYPE_MAX 7
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100612#endif
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200613#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
614
615 __u8 __pkt_type_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700616 __u8 pkt_type:3;
617 __u8 pfmemalloc:1;
618 __u8 ignore_df:1;
619 __u8 nfctinfo:3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700620
Eric Dumazetb1937222014-09-28 22:18:47 -0700621 __u8 nf_trace:1;
622 __u8 ip_summed:2;
623 __u8 ooo_okay:1;
624 __u8 l4_hash:1;
625 __u8 sw_hash:1;
626 __u8 wifi_acked_valid:1;
627 __u8 wifi_acked:1;
628
629 __u8 no_fcs:1;
630 /* Indicates the inner headers are valid in the skbuff. */
631 __u8 encapsulation:1;
632 __u8 encap_hdr_csum:1;
633 __u8 csum_valid:1;
634 __u8 csum_complete_sw:1;
635 __u8 csum_level:2;
636 __u8 csum_bad:1;
637
638#ifdef CONFIG_IPV6_NDISC_NODETYPE
639 __u8 ndisc_nodetype:2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700641 __u8 ipvs_property:1;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700642 __u8 inner_protocol_type:1;
Tom Herberte585f232014-11-04 09:06:54 -0800643 __u8 remcsum_offload:1;
644 /* 3 or 5 bit hole */
Alexander Duyck4031ae62012-01-27 06:22:53 +0000645
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646#ifdef CONFIG_NET_SCHED
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700647 __u16 tc_index; /* traffic control index */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648#ifdef CONFIG_NET_CLS_ACT
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700649 __u16 tc_verd; /* traffic control verdict */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700651#endif
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200652
Eric Dumazetb1937222014-09-28 22:18:47 -0700653 union {
654 __wsum csum;
655 struct {
656 __u16 csum_start;
657 __u16 csum_offset;
658 };
659 };
660 __u32 priority;
661 int skb_iif;
662 __u32 hash;
663 __be16 vlan_proto;
664 __u16 vlan_tci;
Eric Dumazet2bd82482015-02-03 23:48:24 -0800665#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
666 union {
667 unsigned int napi_id;
668 unsigned int sender_cpu;
669 };
Chris Leech97fc2f02006-05-23 17:55:33 -0700670#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700671 union {
James Morris984bc162006-06-09 00:29:17 -0700672#ifdef CONFIG_NETWORK_SECMARK
Scott Feldman0c4f6912015-07-18 18:24:48 -0700673 __u32 secmark;
James Morris984bc162006-06-09 00:29:17 -0700674#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700675#ifdef CONFIG_NET_SWITCHDEV
676 __u32 offload_fwd_mark;
677#endif
678 };
679
Neil Horman3b885782009-10-12 13:26:31 -0700680 union {
681 __u32 mark;
Eric Dumazet16fad692013-03-14 05:40:32 +0000682 __u32 reserved_tailroom;
Neil Horman3b885782009-10-12 13:26:31 -0700683 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Tom Herbert8bce6d72014-09-29 20:22:29 -0700685 union {
686 __be16 inner_protocol;
687 __u8 inner_ipproto;
688 };
689
Simon Horman1a37e412013-05-23 21:02:51 +0000690 __u16 inner_transport_header;
691 __u16 inner_network_header;
692 __u16 inner_mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700693
694 __be16 protocol;
Simon Horman1a37e412013-05-23 21:02:51 +0000695 __u16 transport_header;
696 __u16 network_header;
697 __u16 mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700698
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700699 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700700 __u32 headers_end[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700701 /* public: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700702
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703 /* These elements must be at the end, see alloc_skb() for details. */
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700704 sk_buff_data_t tail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700705 sk_buff_data_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706 unsigned char *head,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700707 *data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700708 unsigned int truesize;
709 atomic_t users;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700710};
711
712#ifdef __KERNEL__
713/*
714 * Handling routines are only of interest to the kernel
715 */
716#include <linux/slab.h>
717
Linus Torvalds1da177e2005-04-16 15:20:36 -0700718
Mel Gormanc93bdd02012-07-31 16:44:19 -0700719#define SKB_ALLOC_FCLONE 0x01
720#define SKB_ALLOC_RX 0x02
Alexander Duyckfd11a832014-12-09 19:40:49 -0800721#define SKB_ALLOC_NAPI 0x04
Mel Gormanc93bdd02012-07-31 16:44:19 -0700722
723/* Returns true if the skb was allocated from PFMEMALLOC reserves */
724static inline bool skb_pfmemalloc(const struct sk_buff *skb)
725{
726 return unlikely(skb->pfmemalloc);
727}
728
Eric Dumazet7fee2262010-05-11 23:19:48 +0000729/*
730 * skb might have a dst pointer attached, refcounted or not.
731 * _skb_refdst low order bit is set if refcount was _not_ taken
732 */
733#define SKB_DST_NOREF 1UL
734#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
735
736/**
737 * skb_dst - returns skb dst_entry
738 * @skb: buffer
739 *
740 * Returns skb dst_entry, regardless of reference taken or not.
741 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000742static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
743{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000744 /* If refdst was not refcounted, check we still are in a
745 * rcu_read_lock section
746 */
747 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
748 !rcu_read_lock_held() &&
749 !rcu_read_lock_bh_held());
750 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
Eric Dumazetadf30902009-06-02 05:19:30 +0000751}
752
Eric Dumazet7fee2262010-05-11 23:19:48 +0000753/**
754 * skb_dst_set - sets skb dst
755 * @skb: buffer
756 * @dst: dst entry
757 *
758 * Sets skb dst, assuming a reference was taken on dst and should
759 * be released by skb_dst_drop()
760 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000761static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
762{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000763 skb->_skb_refdst = (unsigned long)dst;
764}
765
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200766/**
767 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
768 * @skb: buffer
769 * @dst: dst entry
770 *
771 * Sets skb dst, assuming a reference was not taken on dst.
772 * If dst entry is cached, we do not take reference and dst_release
773 * will be avoided by refdst_drop. If dst entry is not cached, we take
774 * reference, so that last dst_release can destroy the dst immediately.
775 */
776static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
777{
Hannes Frederic Sowadbfc4fb2014-12-06 19:19:42 +0100778 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
779 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200780}
Eric Dumazet7fee2262010-05-11 23:19:48 +0000781
782/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300783 * skb_dst_is_noref - Test if skb dst isn't refcounted
Eric Dumazet7fee2262010-05-11 23:19:48 +0000784 * @skb: buffer
785 */
786static inline bool skb_dst_is_noref(const struct sk_buff *skb)
787{
788 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
Eric Dumazetadf30902009-06-02 05:19:30 +0000789}
790
Eric Dumazet511c3f92009-06-02 05:14:27 +0000791static inline struct rtable *skb_rtable(const struct sk_buff *skb)
792{
Eric Dumazetadf30902009-06-02 05:19:30 +0000793 return (struct rtable *)skb_dst(skb);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000794}
795
Joe Perches7965bd42013-09-26 14:48:15 -0700796void kfree_skb(struct sk_buff *skb);
797void kfree_skb_list(struct sk_buff *segs);
798void skb_tx_error(struct sk_buff *skb);
799void consume_skb(struct sk_buff *skb);
800void __kfree_skb(struct sk_buff *skb);
Eric Dumazetd7e88832012-04-30 08:10:34 +0000801extern struct kmem_cache *skbuff_head_cache;
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000802
Joe Perches7965bd42013-09-26 14:48:15 -0700803void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
804bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
805 bool *fragstolen, int *delta_truesize);
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000806
Joe Perches7965bd42013-09-26 14:48:15 -0700807struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
808 int node);
Eric Dumazet2ea2f622015-04-24 16:05:01 -0700809struct sk_buff *__build_skb(void *data, unsigned int frag_size);
Joe Perches7965bd42013-09-26 14:48:15 -0700810struct sk_buff *build_skb(void *data, unsigned int frag_size);
David S. Millerd179cd12005-08-17 14:57:30 -0700811static inline struct sk_buff *alloc_skb(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100812 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700813{
Eric Dumazet564824b2010-10-11 19:05:25 +0000814 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700815}
816
Eric Dumazet2e4e4412014-09-17 04:49:49 -0700817struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
818 unsigned long data_len,
819 int max_page_order,
820 int *errcode,
821 gfp_t gfp_mask);
822
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700823/* Layout of fast clones : [skb1][skb2][fclone_ref] */
824struct sk_buff_fclones {
825 struct sk_buff skb1;
826
827 struct sk_buff skb2;
828
829 atomic_t fclone_ref;
830};
831
832/**
833 * skb_fclone_busy - check if fclone is busy
834 * @skb: buffer
835 *
Masanari Iidabda13fe2015-12-13 16:53:02 +0900836 * Returns true if skb is a fast clone, and its clone is not freed.
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700837 * Some drivers call skb_orphan() in their ndo_start_xmit(),
838 * so we also check that this didnt happen.
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700839 */
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700840static inline bool skb_fclone_busy(const struct sock *sk,
841 const struct sk_buff *skb)
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700842{
843 const struct sk_buff_fclones *fclones;
844
845 fclones = container_of(skb, struct sk_buff_fclones, skb1);
846
847 return skb->fclone == SKB_FCLONE_ORIG &&
Eric Dumazet6ffe75eb2014-12-03 17:04:39 -0800848 atomic_read(&fclones->fclone_ref) > 1 &&
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700849 fclones->skb2.sk == sk;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700850}
851
David S. Millerd179cd12005-08-17 14:57:30 -0700852static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100853 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700854{
Mel Gormanc93bdd02012-07-31 16:44:19 -0700855 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700856}
857
Joe Perches7965bd42013-09-26 14:48:15 -0700858struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000859static inline struct sk_buff *alloc_skb_head(gfp_t priority)
860{
861 return __alloc_skb_head(priority, -1);
862}
863
Joe Perches7965bd42013-09-26 14:48:15 -0700864struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
865int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
866struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
867struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
Octavian Purdilabad93e92014-06-12 01:36:26 +0300868struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
869 gfp_t gfp_mask, bool fclone);
870static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
871 gfp_t gfp_mask)
872{
873 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
874}
Eric Dumazet117632e2011-12-03 21:39:53 +0000875
Joe Perches7965bd42013-09-26 14:48:15 -0700876int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
877struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
878 unsigned int headroom);
879struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
880 int newtailroom, gfp_t priority);
Fan Du25a91d82014-01-18 09:54:23 +0800881int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
882 int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -0700883int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
884 int len);
885int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
886int skb_pad(struct sk_buff *skb, int pad);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000887#define dev_kfree_skb(a) consume_skb(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700888
Joe Perches7965bd42013-09-26 14:48:15 -0700889int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
890 int getfrag(void *from, char *to, int offset,
891 int len, int odd, struct sk_buff *skb),
892 void *from, int length);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700893
Hannes Frederic Sowabe12a1f2015-05-21 16:59:58 +0200894int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
895 int offset, size_t size);
896
Eric Dumazetd94d9fe2009-11-04 09:50:58 -0800897struct skb_seq_state {
Thomas Graf677e90e2005-06-23 20:59:51 -0700898 __u32 lower_offset;
899 __u32 upper_offset;
900 __u32 frag_idx;
901 __u32 stepped_offset;
902 struct sk_buff *root_skb;
903 struct sk_buff *cur_skb;
904 __u8 *frag_data;
905};
906
Joe Perches7965bd42013-09-26 14:48:15 -0700907void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
908 unsigned int to, struct skb_seq_state *st);
909unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
910 struct skb_seq_state *st);
911void skb_abort_seq_read(struct skb_seq_state *st);
Thomas Graf677e90e2005-06-23 20:59:51 -0700912
Joe Perches7965bd42013-09-26 14:48:15 -0700913unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
Bojan Prtvar059a2442015-02-22 11:46:35 +0100914 unsigned int to, struct ts_config *config);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -0700915
Tom Herbert09323cc2013-12-15 22:16:19 -0800916/*
917 * Packet hash types specify the type of hash in skb_set_hash.
918 *
919 * Hash types refer to the protocol layer addresses which are used to
920 * construct a packet's hash. The hashes are used to differentiate or identify
921 * flows of the protocol layer for the hash type. Hash types are either
922 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
923 *
924 * Properties of hashes:
925 *
926 * 1) Two packets in different flows have different hash values
927 * 2) Two packets in the same flow should have the same hash value
928 *
929 * A hash at a higher layer is considered to be more specific. A driver should
930 * set the most specific hash possible.
931 *
932 * A driver cannot indicate a more specific hash than the layer at which a hash
933 * was computed. For instance an L3 hash cannot be set as an L4 hash.
934 *
935 * A driver may indicate a hash level which is less specific than the
936 * actual layer the hash was computed on. For instance, a hash computed
937 * at L4 may be considered an L3 hash. This should only be done if the
938 * driver can't unambiguously determine that the HW computed the hash at
939 * the higher layer. Note that the "should" in the second property above
940 * permits this.
941 */
942enum pkt_hash_types {
943 PKT_HASH_TYPE_NONE, /* Undefined type */
944 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
945 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
946 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
947};
948
Tom Herbertbcc83832015-09-01 09:24:24 -0700949static inline void skb_clear_hash(struct sk_buff *skb)
950{
951 skb->hash = 0;
952 skb->sw_hash = 0;
953 skb->l4_hash = 0;
954}
955
956static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
957{
958 if (!skb->l4_hash)
959 skb_clear_hash(skb);
960}
961
962static inline void
963__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
964{
965 skb->l4_hash = is_l4;
966 skb->sw_hash = is_sw;
967 skb->hash = hash;
968}
969
Tom Herbert09323cc2013-12-15 22:16:19 -0800970static inline void
971skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
972{
Tom Herbertbcc83832015-09-01 09:24:24 -0700973 /* Used by drivers to set hash from HW */
974 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
975}
976
977static inline void
978__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
979{
980 __skb_set_hash(skb, hash, true, is_l4);
Tom Herbert09323cc2013-12-15 22:16:19 -0800981}
982
Tom Herberte5276932015-09-01 09:24:23 -0700983void __skb_get_hash(struct sk_buff *skb);
984u32 skb_get_poff(const struct sk_buff *skb);
985u32 __skb_get_poff(const struct sk_buff *skb, void *data,
986 const struct flow_keys *keys, int hlen);
987__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
988 void *data, int hlen_proto);
989
990static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
991 int thoff, u8 ip_proto)
992{
993 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
994}
995
996void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
997 const struct flow_dissector_key *key,
998 unsigned int key_count);
999
1000bool __skb_flow_dissect(const struct sk_buff *skb,
1001 struct flow_dissector *flow_dissector,
1002 void *target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001003 void *data, __be16 proto, int nhoff, int hlen,
1004 unsigned int flags);
Tom Herberte5276932015-09-01 09:24:23 -07001005
1006static inline bool skb_flow_dissect(const struct sk_buff *skb,
1007 struct flow_dissector *flow_dissector,
Tom Herbertcd79a232015-09-01 09:24:27 -07001008 void *target_container, unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001009{
1010 return __skb_flow_dissect(skb, flow_dissector, target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001011 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001012}
1013
1014static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
Tom Herbertcd79a232015-09-01 09:24:27 -07001015 struct flow_keys *flow,
1016 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001017{
1018 memset(flow, 0, sizeof(*flow));
1019 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001020 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001021}
1022
1023static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1024 void *data, __be16 proto,
Tom Herbertcd79a232015-09-01 09:24:27 -07001025 int nhoff, int hlen,
1026 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001027{
1028 memset(flow, 0, sizeof(*flow));
1029 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001030 data, proto, nhoff, hlen, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001031}
1032
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001033static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001034{
Tom Herberta3b18dd2014-07-01 21:33:17 -07001035 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001036 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001037
Tom Herbert61b905d2014-03-24 15:34:47 -07001038 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001039}
1040
David S. Miller20a17bf2015-09-01 21:19:17 -07001041__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
Tom Herbertf70ea012015-07-31 16:52:10 -07001042
David S. Miller20a17bf2015-09-01 21:19:17 -07001043static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
Tom Herbertf70ea012015-07-31 16:52:10 -07001044{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001045 if (!skb->l4_hash && !skb->sw_hash) {
1046 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001047 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001048
Tom Herbertde4c1f82015-09-01 18:11:04 -07001049 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001050 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001051
1052 return skb->hash;
1053}
1054
David S. Miller20a17bf2015-09-01 21:19:17 -07001055__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
Tom Herbertf70ea012015-07-31 16:52:10 -07001056
David S. Miller20a17bf2015-09-01 21:19:17 -07001057static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
Tom Herbertf70ea012015-07-31 16:52:10 -07001058{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001059 if (!skb->l4_hash && !skb->sw_hash) {
1060 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001061 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001062
Tom Herbertde4c1f82015-09-01 18:11:04 -07001063 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001064 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001065
1066 return skb->hash;
1067}
1068
Tom Herbert50fb7992015-05-01 11:30:12 -07001069__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1070
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001071static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1072{
Tom Herbert61b905d2014-03-24 15:34:47 -07001073 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001074}
1075
Tom Herbert3df7a742013-12-15 22:16:29 -08001076static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1077{
Tom Herbert61b905d2014-03-24 15:34:47 -07001078 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -07001079 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -07001080 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -08001081};
1082
Eric Dumazetc29390c2015-03-11 18:42:02 -07001083static inline void skb_sender_cpu_clear(struct sk_buff *skb)
1084{
Eric Dumazetc29390c2015-03-11 18:42:02 -07001085}
1086
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001087#ifdef NET_SKBUFF_DATA_USES_OFFSET
1088static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1089{
1090 return skb->head + skb->end;
1091}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001092
1093static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1094{
1095 return skb->end;
1096}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001097#else
1098static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1099{
1100 return skb->end;
1101}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001102
1103static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1104{
1105 return skb->end - skb->head;
1106}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001107#endif
1108
Linus Torvalds1da177e2005-04-16 15:20:36 -07001109/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001110#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001111
Patrick Ohlyac45f602009-02-12 05:03:37 +00001112static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1113{
1114 return &skb_shinfo(skb)->hwtstamps;
1115}
1116
Linus Torvalds1da177e2005-04-16 15:20:36 -07001117/**
1118 * skb_queue_empty - check if a queue is empty
1119 * @list: queue head
1120 *
1121 * Returns true if the queue is empty, false otherwise.
1122 */
1123static inline int skb_queue_empty(const struct sk_buff_head *list)
1124{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001125 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001126}
1127
1128/**
David S. Millerfc7ebb22008-09-23 00:34:07 -07001129 * skb_queue_is_last - check if skb is the last entry in the queue
1130 * @list: queue head
1131 * @skb: buffer
1132 *
1133 * Returns true if @skb is the last buffer on the list.
1134 */
1135static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1136 const struct sk_buff *skb)
1137{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001138 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -07001139}
1140
1141/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001142 * skb_queue_is_first - check if skb is the first entry in the queue
1143 * @list: queue head
1144 * @skb: buffer
1145 *
1146 * Returns true if @skb is the first buffer on the list.
1147 */
1148static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1149 const struct sk_buff *skb)
1150{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001151 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001152}
1153
1154/**
David S. Miller249c8b42008-09-23 00:44:42 -07001155 * skb_queue_next - return the next packet in the queue
1156 * @list: queue head
1157 * @skb: current buffer
1158 *
1159 * Return the next packet in @list after @skb. It is only valid to
1160 * call this if skb_queue_is_last() evaluates to false.
1161 */
1162static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1163 const struct sk_buff *skb)
1164{
1165 /* This BUG_ON may seem severe, but if we just return then we
1166 * are going to dereference garbage.
1167 */
1168 BUG_ON(skb_queue_is_last(list, skb));
1169 return skb->next;
1170}
1171
1172/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001173 * skb_queue_prev - return the prev packet in the queue
1174 * @list: queue head
1175 * @skb: current buffer
1176 *
1177 * Return the prev packet in @list before @skb. It is only valid to
1178 * call this if skb_queue_is_first() evaluates to false.
1179 */
1180static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1181 const struct sk_buff *skb)
1182{
1183 /* This BUG_ON may seem severe, but if we just return then we
1184 * are going to dereference garbage.
1185 */
1186 BUG_ON(skb_queue_is_first(list, skb));
1187 return skb->prev;
1188}
1189
1190/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191 * skb_get - reference buffer
1192 * @skb: buffer to reference
1193 *
1194 * Makes another reference to a socket buffer and returns a pointer
1195 * to the buffer.
1196 */
1197static inline struct sk_buff *skb_get(struct sk_buff *skb)
1198{
1199 atomic_inc(&skb->users);
1200 return skb;
1201}
1202
1203/*
1204 * If users == 1, we are the only owner and are can avoid redundant
1205 * atomic change.
1206 */
1207
1208/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001209 * skb_cloned - is the buffer a clone
1210 * @skb: buffer to check
1211 *
1212 * Returns true if the buffer was generated with skb_clone() and is
1213 * one of multiple shared copies of the buffer. Cloned buffers are
1214 * shared data so must not be written to under normal circumstances.
1215 */
1216static inline int skb_cloned(const struct sk_buff *skb)
1217{
1218 return skb->cloned &&
1219 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1220}
1221
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001222static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1223{
Mel Gormand0164ad2015-11-06 16:28:21 -08001224 might_sleep_if(gfpflags_allow_blocking(pri));
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001225
1226 if (skb_cloned(skb))
1227 return pskb_expand_head(skb, 0, 0, pri);
1228
1229 return 0;
1230}
1231
Linus Torvalds1da177e2005-04-16 15:20:36 -07001232/**
1233 * skb_header_cloned - is the header a clone
1234 * @skb: buffer to check
1235 *
1236 * Returns true if modifying the header part of the buffer requires
1237 * the data to be copied.
1238 */
1239static inline int skb_header_cloned(const struct sk_buff *skb)
1240{
1241 int dataref;
1242
1243 if (!skb->cloned)
1244 return 0;
1245
1246 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1247 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1248 return dataref != 1;
1249}
1250
1251/**
1252 * skb_header_release - release reference to header
1253 * @skb: buffer to operate on
1254 *
1255 * Drop a reference to the header part of the buffer. This is done
1256 * by acquiring a payload reference. You must not read from the header
1257 * part of skb->data after this.
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001258 * Note : Check if you can use __skb_header_release() instead.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001259 */
1260static inline void skb_header_release(struct sk_buff *skb)
1261{
1262 BUG_ON(skb->nohdr);
1263 skb->nohdr = 1;
1264 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1265}
1266
1267/**
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001268 * __skb_header_release - release reference to header
1269 * @skb: buffer to operate on
1270 *
1271 * Variant of skb_header_release() assuming skb is private to caller.
1272 * We can avoid one atomic operation.
1273 */
1274static inline void __skb_header_release(struct sk_buff *skb)
1275{
1276 skb->nohdr = 1;
1277 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1278}
1279
1280
1281/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001282 * skb_shared - is the buffer shared
1283 * @skb: buffer to check
1284 *
1285 * Returns true if more than one person has a reference to this
1286 * buffer.
1287 */
1288static inline int skb_shared(const struct sk_buff *skb)
1289{
1290 return atomic_read(&skb->users) != 1;
1291}
1292
1293/**
1294 * skb_share_check - check if buffer is shared and if so clone it
1295 * @skb: buffer to check
1296 * @pri: priority for memory allocation
1297 *
1298 * If the buffer is shared the buffer is cloned and the old copy
1299 * drops a reference. A new clone with a single reference is returned.
1300 * If the buffer is not shared the original buffer is returned. When
1301 * being called from interrupt status or with spinlocks held pri must
1302 * be GFP_ATOMIC.
1303 *
1304 * NULL is returned on a memory allocation failure.
1305 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001306static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001307{
Mel Gormand0164ad2015-11-06 16:28:21 -08001308 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309 if (skb_shared(skb)) {
1310 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001311
1312 if (likely(nskb))
1313 consume_skb(skb);
1314 else
1315 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001316 skb = nskb;
1317 }
1318 return skb;
1319}
1320
1321/*
1322 * Copy shared buffers into a new sk_buff. We effectively do COW on
1323 * packets to handle cases where we have a local reader and forward
1324 * and a couple of other messy ones. The normal one is tcpdumping
1325 * a packet thats being forwarded.
1326 */
1327
1328/**
1329 * skb_unshare - make a copy of a shared buffer
1330 * @skb: buffer to check
1331 * @pri: priority for memory allocation
1332 *
1333 * If the socket buffer is a clone then this function creates a new
1334 * copy of the data, drops a reference count on the old copy and returns
1335 * the new copy with the reference count at 1. If the buffer is not a clone
1336 * the original buffer is returned. When called with a spinlock held or
1337 * from interrupt state @pri must be %GFP_ATOMIC
1338 *
1339 * %NULL is returned on a memory allocation failure.
1340 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001341static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001342 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343{
Mel Gormand0164ad2015-11-06 16:28:21 -08001344 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001345 if (skb_cloned(skb)) {
1346 struct sk_buff *nskb = skb_copy(skb, pri);
Alexander Aring31eff812014-10-10 23:10:47 +02001347
1348 /* Free our shared copy */
1349 if (likely(nskb))
1350 consume_skb(skb);
1351 else
1352 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001353 skb = nskb;
1354 }
1355 return skb;
1356}
1357
1358/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001359 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001360 * @list_: list to peek at
1361 *
1362 * Peek an &sk_buff. Unlike most other operations you _MUST_
1363 * be careful with this one. A peek leaves the buffer on the
1364 * list and someone else may run off with it. You must hold
1365 * the appropriate locks or have a private queue to do this.
1366 *
1367 * Returns %NULL for an empty list or a pointer to the head element.
1368 * The reference count is not incremented and the reference is therefore
1369 * volatile. Use with caution.
1370 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001371static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001372{
Eric Dumazet18d07002012-04-30 16:31:46 +00001373 struct sk_buff *skb = list_->next;
1374
1375 if (skb == (struct sk_buff *)list_)
1376 skb = NULL;
1377 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001378}
1379
1380/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001381 * skb_peek_next - peek skb following the given one from a queue
1382 * @skb: skb to start from
1383 * @list_: list to peek at
1384 *
1385 * Returns %NULL when the end of the list is met or a pointer to the
1386 * next element. The reference count is not incremented and the
1387 * reference is therefore volatile. Use with caution.
1388 */
1389static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1390 const struct sk_buff_head *list_)
1391{
1392 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001393
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001394 if (next == (struct sk_buff *)list_)
1395 next = NULL;
1396 return next;
1397}
1398
1399/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001400 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 * @list_: list to peek at
1402 *
1403 * Peek an &sk_buff. Unlike most other operations you _MUST_
1404 * be careful with this one. A peek leaves the buffer on the
1405 * list and someone else may run off with it. You must hold
1406 * the appropriate locks or have a private queue to do this.
1407 *
1408 * Returns %NULL for an empty list or a pointer to the tail element.
1409 * The reference count is not incremented and the reference is therefore
1410 * volatile. Use with caution.
1411 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001412static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001413{
Eric Dumazet18d07002012-04-30 16:31:46 +00001414 struct sk_buff *skb = list_->prev;
1415
1416 if (skb == (struct sk_buff *)list_)
1417 skb = NULL;
1418 return skb;
1419
Linus Torvalds1da177e2005-04-16 15:20:36 -07001420}
1421
1422/**
1423 * skb_queue_len - get queue length
1424 * @list_: list to measure
1425 *
1426 * Return the length of an &sk_buff queue.
1427 */
1428static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1429{
1430 return list_->qlen;
1431}
1432
David S. Miller67fed452008-09-21 22:36:24 -07001433/**
1434 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1435 * @list: queue to initialize
1436 *
1437 * This initializes only the list and queue length aspects of
1438 * an sk_buff_head object. This allows to initialize the list
1439 * aspects of an sk_buff_head without reinitializing things like
1440 * the spinlock. It can also be used for on-stack sk_buff_head
1441 * objects where the spinlock is known to not be used.
1442 */
1443static inline void __skb_queue_head_init(struct sk_buff_head *list)
1444{
1445 list->prev = list->next = (struct sk_buff *)list;
1446 list->qlen = 0;
1447}
1448
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001449/*
1450 * This function creates a split out lock class for each invocation;
1451 * this is needed for now since a whole lot of users of the skb-queue
1452 * infrastructure in drivers have different locking usage (in hardirq)
1453 * than the networking core (in softirq only). In the long run either the
1454 * network layer or drivers should need annotation to consolidate the
1455 * main types of usage into 3 classes.
1456 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457static inline void skb_queue_head_init(struct sk_buff_head *list)
1458{
1459 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001460 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461}
1462
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001463static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1464 struct lock_class_key *class)
1465{
1466 skb_queue_head_init(list);
1467 lockdep_set_class(&list->lock, class);
1468}
1469
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001471 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001472 *
1473 * The "__skb_xxxx()" functions are the non-atomic ones that
1474 * can only be called with interrupts disabled.
1475 */
Joe Perches7965bd42013-09-26 14:48:15 -07001476void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1477 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001478static inline void __skb_insert(struct sk_buff *newsk,
1479 struct sk_buff *prev, struct sk_buff *next,
1480 struct sk_buff_head *list)
1481{
1482 newsk->next = next;
1483 newsk->prev = prev;
1484 next->prev = prev->next = newsk;
1485 list->qlen++;
1486}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001487
David S. Miller67fed452008-09-21 22:36:24 -07001488static inline void __skb_queue_splice(const struct sk_buff_head *list,
1489 struct sk_buff *prev,
1490 struct sk_buff *next)
1491{
1492 struct sk_buff *first = list->next;
1493 struct sk_buff *last = list->prev;
1494
1495 first->prev = prev;
1496 prev->next = first;
1497
1498 last->next = next;
1499 next->prev = last;
1500}
1501
1502/**
1503 * skb_queue_splice - join two skb lists, this is designed for stacks
1504 * @list: the new list to add
1505 * @head: the place to add it in the first list
1506 */
1507static inline void skb_queue_splice(const struct sk_buff_head *list,
1508 struct sk_buff_head *head)
1509{
1510 if (!skb_queue_empty(list)) {
1511 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001512 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001513 }
1514}
1515
1516/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001517 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001518 * @list: the new list to add
1519 * @head: the place to add it in the first list
1520 *
1521 * The list at @list is reinitialised
1522 */
1523static inline void skb_queue_splice_init(struct sk_buff_head *list,
1524 struct sk_buff_head *head)
1525{
1526 if (!skb_queue_empty(list)) {
1527 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001528 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001529 __skb_queue_head_init(list);
1530 }
1531}
1532
1533/**
1534 * skb_queue_splice_tail - join two skb lists, each list being a queue
1535 * @list: the new list to add
1536 * @head: the place to add it in the first list
1537 */
1538static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1539 struct sk_buff_head *head)
1540{
1541 if (!skb_queue_empty(list)) {
1542 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001543 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001544 }
1545}
1546
1547/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001548 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001549 * @list: the new list to add
1550 * @head: the place to add it in the first list
1551 *
1552 * Each of the lists is a queue.
1553 * The list at @list is reinitialised
1554 */
1555static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1556 struct sk_buff_head *head)
1557{
1558 if (!skb_queue_empty(list)) {
1559 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001560 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001561 __skb_queue_head_init(list);
1562 }
1563}
1564
Linus Torvalds1da177e2005-04-16 15:20:36 -07001565/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001566 * __skb_queue_after - queue a buffer at the list head
1567 * @list: list to use
1568 * @prev: place after this buffer
1569 * @newsk: buffer to queue
1570 *
1571 * Queue a buffer int the middle of a list. This function takes no locks
1572 * and you must therefore hold required locks before calling it.
1573 *
1574 * A buffer cannot be placed on two lists at the same time.
1575 */
1576static inline void __skb_queue_after(struct sk_buff_head *list,
1577 struct sk_buff *prev,
1578 struct sk_buff *newsk)
1579{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001580 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001581}
1582
Joe Perches7965bd42013-09-26 14:48:15 -07001583void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1584 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001585
Gerrit Renkerf5572852008-04-14 00:05:28 -07001586static inline void __skb_queue_before(struct sk_buff_head *list,
1587 struct sk_buff *next,
1588 struct sk_buff *newsk)
1589{
1590 __skb_insert(newsk, next->prev, next, list);
1591}
1592
Stephen Hemminger300ce172005-10-30 13:47:34 -08001593/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001594 * __skb_queue_head - queue a buffer at the list head
1595 * @list: list to use
1596 * @newsk: buffer to queue
1597 *
1598 * Queue a buffer at the start of a list. This function takes no locks
1599 * and you must therefore hold required locks before calling it.
1600 *
1601 * A buffer cannot be placed on two lists at the same time.
1602 */
Joe Perches7965bd42013-09-26 14:48:15 -07001603void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604static inline void __skb_queue_head(struct sk_buff_head *list,
1605 struct sk_buff *newsk)
1606{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001607 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001608}
1609
1610/**
1611 * __skb_queue_tail - queue a buffer at the list tail
1612 * @list: list to use
1613 * @newsk: buffer to queue
1614 *
1615 * Queue a buffer at the end of a list. This function takes no locks
1616 * and you must therefore hold required locks before calling it.
1617 *
1618 * A buffer cannot be placed on two lists at the same time.
1619 */
Joe Perches7965bd42013-09-26 14:48:15 -07001620void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001621static inline void __skb_queue_tail(struct sk_buff_head *list,
1622 struct sk_buff *newsk)
1623{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001624 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001625}
1626
Linus Torvalds1da177e2005-04-16 15:20:36 -07001627/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001628 * remove sk_buff from list. _Must_ be called atomically, and with
1629 * the list known..
1630 */
Joe Perches7965bd42013-09-26 14:48:15 -07001631void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001632static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1633{
1634 struct sk_buff *next, *prev;
1635
1636 list->qlen--;
1637 next = skb->next;
1638 prev = skb->prev;
1639 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001640 next->prev = prev;
1641 prev->next = next;
1642}
1643
Gerrit Renkerf525c062008-04-14 00:04:12 -07001644/**
1645 * __skb_dequeue - remove from the head of the queue
1646 * @list: list to dequeue from
1647 *
1648 * Remove the head of the list. This function does not take any locks
1649 * so must be used with appropriate locks held only. The head item is
1650 * returned or %NULL if the list is empty.
1651 */
Joe Perches7965bd42013-09-26 14:48:15 -07001652struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001653static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1654{
1655 struct sk_buff *skb = skb_peek(list);
1656 if (skb)
1657 __skb_unlink(skb, list);
1658 return skb;
1659}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001660
1661/**
1662 * __skb_dequeue_tail - remove from the tail of the queue
1663 * @list: list to dequeue from
1664 *
1665 * Remove the tail of the list. This function does not take any locks
1666 * so must be used with appropriate locks held only. The tail item is
1667 * returned or %NULL if the list is empty.
1668 */
Joe Perches7965bd42013-09-26 14:48:15 -07001669struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1671{
1672 struct sk_buff *skb = skb_peek_tail(list);
1673 if (skb)
1674 __skb_unlink(skb, list);
1675 return skb;
1676}
1677
1678
David S. Millerbdcc0922012-03-07 20:53:36 -05001679static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001680{
1681 return skb->data_len;
1682}
1683
1684static inline unsigned int skb_headlen(const struct sk_buff *skb)
1685{
1686 return skb->len - skb->data_len;
1687}
1688
1689static inline int skb_pagelen(const struct sk_buff *skb)
1690{
1691 int i, len = 0;
1692
1693 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001694 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695 return len + skb_headlen(skb);
1696}
1697
Ian Campbell131ea662011-08-19 06:25:00 +00001698/**
1699 * __skb_fill_page_desc - initialise a paged fragment in an skb
1700 * @skb: buffer containing fragment to be initialised
1701 * @i: paged fragment index to initialise
1702 * @page: the page to use for this fragment
1703 * @off: the offset to the data with @page
1704 * @size: the length of the data
1705 *
1706 * Initialises the @i'th fragment of @skb to point to &size bytes at
1707 * offset @off within @page.
1708 *
1709 * Does not take any additional reference on the fragment.
1710 */
1711static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1712 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713{
1714 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1715
Mel Gormanc48a11c2012-07-31 16:44:23 -07001716 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001717 * Propagate page pfmemalloc to the skb if we can. The problem is
1718 * that not all callers have unique ownership of the page but rely
1719 * on page_is_pfmemalloc doing the right thing(tm).
Mel Gormanc48a11c2012-07-31 16:44:23 -07001720 */
Ian Campbella8605c62011-10-19 23:01:49 +00001721 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001722 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001723 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001724
1725 page = compound_head(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001726 if (page_is_pfmemalloc(page))
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001727 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001728}
1729
1730/**
1731 * skb_fill_page_desc - initialise a paged fragment in an skb
1732 * @skb: buffer containing fragment to be initialised
1733 * @i: paged fragment index to initialise
1734 * @page: the page to use for this fragment
1735 * @off: the offset to the data with @page
1736 * @size: the length of the data
1737 *
1738 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001739 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001740 * addition updates @skb such that @i is the last fragment.
1741 *
1742 * Does not take any additional reference on the fragment.
1743 */
1744static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1745 struct page *page, int off, int size)
1746{
1747 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001748 skb_shinfo(skb)->nr_frags = i + 1;
1749}
1750
Joe Perches7965bd42013-09-26 14:48:15 -07001751void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1752 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001753
Jason Wangf8e617e2013-11-01 14:07:47 +08001754void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1755 unsigned int truesize);
1756
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001758#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1760
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001761#ifdef NET_SKBUFF_DATA_USES_OFFSET
1762static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1763{
1764 return skb->head + skb->tail;
1765}
1766
1767static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1768{
1769 skb->tail = skb->data - skb->head;
1770}
1771
1772static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1773{
1774 skb_reset_tail_pointer(skb);
1775 skb->tail += offset;
1776}
Simon Horman7cc46192013-05-28 20:34:29 +00001777
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001778#else /* NET_SKBUFF_DATA_USES_OFFSET */
1779static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1780{
1781 return skb->tail;
1782}
1783
1784static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1785{
1786 skb->tail = skb->data;
1787}
1788
1789static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1790{
1791 skb->tail = skb->data + offset;
1792}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001793
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001794#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1795
Linus Torvalds1da177e2005-04-16 15:20:36 -07001796/*
1797 * Add data to an sk_buff
1798 */
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001799unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001800unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1802{
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001803 unsigned char *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001804 SKB_LINEAR_ASSERT(skb);
1805 skb->tail += len;
1806 skb->len += len;
1807 return tmp;
1808}
1809
Joe Perches7965bd42013-09-26 14:48:15 -07001810unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001811static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1812{
1813 skb->data -= len;
1814 skb->len += len;
1815 return skb->data;
1816}
1817
Joe Perches7965bd42013-09-26 14:48:15 -07001818unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1820{
1821 skb->len -= len;
1822 BUG_ON(skb->len < skb->data_len);
1823 return skb->data += len;
1824}
1825
David S. Miller47d29642010-05-02 02:21:44 -07001826static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1827{
1828 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1829}
1830
Joe Perches7965bd42013-09-26 14:48:15 -07001831unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001832
1833static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1834{
1835 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07001836 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837 return NULL;
1838 skb->len -= len;
1839 return skb->data += len;
1840}
1841
1842static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1843{
1844 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1845}
1846
1847static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1848{
1849 if (likely(len <= skb_headlen(skb)))
1850 return 1;
1851 if (unlikely(len > skb->len))
1852 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07001853 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001854}
1855
1856/**
1857 * skb_headroom - bytes at buffer head
1858 * @skb: buffer to check
1859 *
1860 * Return the number of bytes of free space at the head of an &sk_buff.
1861 */
Chuck Leverc2636b42007-10-23 21:07:32 -07001862static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001863{
1864 return skb->data - skb->head;
1865}
1866
1867/**
1868 * skb_tailroom - bytes at buffer end
1869 * @skb: buffer to check
1870 *
1871 * Return the number of bytes of free space at the tail of an sk_buff
1872 */
1873static inline int skb_tailroom(const struct sk_buff *skb)
1874{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001875 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876}
1877
1878/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00001879 * skb_availroom - bytes at buffer end
1880 * @skb: buffer to check
1881 *
1882 * Return the number of bytes of free space at the tail of an sk_buff
1883 * allocated by sk_stream_alloc()
1884 */
1885static inline int skb_availroom(const struct sk_buff *skb)
1886{
Eric Dumazet16fad692013-03-14 05:40:32 +00001887 if (skb_is_nonlinear(skb))
1888 return 0;
1889
1890 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00001891}
1892
1893/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001894 * skb_reserve - adjust headroom
1895 * @skb: buffer to alter
1896 * @len: bytes to move
1897 *
1898 * Increase the headroom of an empty &sk_buff by reducing the tail
1899 * room. This is only allowed for an empty buffer.
1900 */
David S. Miller8243126c2006-01-17 02:54:21 -08001901static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001902{
1903 skb->data += len;
1904 skb->tail += len;
1905}
1906
Tom Herbert8bce6d72014-09-29 20:22:29 -07001907#define ENCAP_TYPE_ETHER 0
1908#define ENCAP_TYPE_IPPROTO 1
1909
1910static inline void skb_set_inner_protocol(struct sk_buff *skb,
1911 __be16 protocol)
1912{
1913 skb->inner_protocol = protocol;
1914 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1915}
1916
1917static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1918 __u8 ipproto)
1919{
1920 skb->inner_ipproto = ipproto;
1921 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
1922}
1923
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001924static inline void skb_reset_inner_headers(struct sk_buff *skb)
1925{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00001926 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001927 skb->inner_network_header = skb->network_header;
1928 skb->inner_transport_header = skb->transport_header;
1929}
1930
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00001931static inline void skb_reset_mac_len(struct sk_buff *skb)
1932{
1933 skb->mac_len = skb->network_header - skb->mac_header;
1934}
1935
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00001936static inline unsigned char *skb_inner_transport_header(const struct sk_buff
1937 *skb)
1938{
1939 return skb->head + skb->inner_transport_header;
1940}
1941
1942static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
1943{
1944 skb->inner_transport_header = skb->data - skb->head;
1945}
1946
1947static inline void skb_set_inner_transport_header(struct sk_buff *skb,
1948 const int offset)
1949{
1950 skb_reset_inner_transport_header(skb);
1951 skb->inner_transport_header += offset;
1952}
1953
1954static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
1955{
1956 return skb->head + skb->inner_network_header;
1957}
1958
1959static inline void skb_reset_inner_network_header(struct sk_buff *skb)
1960{
1961 skb->inner_network_header = skb->data - skb->head;
1962}
1963
1964static inline void skb_set_inner_network_header(struct sk_buff *skb,
1965 const int offset)
1966{
1967 skb_reset_inner_network_header(skb);
1968 skb->inner_network_header += offset;
1969}
1970
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00001971static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
1972{
1973 return skb->head + skb->inner_mac_header;
1974}
1975
1976static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
1977{
1978 skb->inner_mac_header = skb->data - skb->head;
1979}
1980
1981static inline void skb_set_inner_mac_header(struct sk_buff *skb,
1982 const int offset)
1983{
1984 skb_reset_inner_mac_header(skb);
1985 skb->inner_mac_header += offset;
1986}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00001987static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
1988{
Cong Wang35d04612013-05-29 15:16:05 +08001989 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00001990}
1991
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07001992static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
1993{
1994 return skb->head + skb->transport_header;
1995}
1996
1997static inline void skb_reset_transport_header(struct sk_buff *skb)
1998{
1999 skb->transport_header = skb->data - skb->head;
2000}
2001
2002static inline void skb_set_transport_header(struct sk_buff *skb,
2003 const int offset)
2004{
2005 skb_reset_transport_header(skb);
2006 skb->transport_header += offset;
2007}
2008
2009static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2010{
2011 return skb->head + skb->network_header;
2012}
2013
2014static inline void skb_reset_network_header(struct sk_buff *skb)
2015{
2016 skb->network_header = skb->data - skb->head;
2017}
2018
2019static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2020{
2021 skb_reset_network_header(skb);
2022 skb->network_header += offset;
2023}
2024
2025static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2026{
2027 return skb->head + skb->mac_header;
2028}
2029
2030static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2031{
Cong Wang35d04612013-05-29 15:16:05 +08002032 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002033}
2034
2035static inline void skb_reset_mac_header(struct sk_buff *skb)
2036{
2037 skb->mac_header = skb->data - skb->head;
2038}
2039
2040static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2041{
2042 skb_reset_mac_header(skb);
2043 skb->mac_header += offset;
2044}
2045
Timo Teräs0e3da5b2013-12-16 11:02:09 +02002046static inline void skb_pop_mac_header(struct sk_buff *skb)
2047{
2048 skb->mac_header = skb->network_header;
2049}
2050
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002051static inline void skb_probe_transport_header(struct sk_buff *skb,
2052 const int offset_hint)
2053{
2054 struct flow_keys keys;
2055
2056 if (skb_transport_header_was_set(skb))
2057 return;
Tom Herbertcd79a232015-09-01 09:24:27 -07002058 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
Tom Herbert42aecaa2015-06-04 09:16:39 -07002059 skb_set_transport_header(skb, keys.control.thoff);
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002060 else
2061 skb_set_transport_header(skb, offset_hint);
2062}
2063
Eric Dumazet03606892012-02-23 10:55:02 +00002064static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2065{
2066 if (skb_mac_header_was_set(skb)) {
2067 const unsigned char *old_mac = skb_mac_header(skb);
2068
2069 skb_set_mac_header(skb, -skb->mac_len);
2070 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2071 }
2072}
2073
Michał Mirosław04fb4512010-12-14 15:24:08 +00002074static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2075{
2076 return skb->csum_start - skb_headroom(skb);
2077}
2078
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002079static inline int skb_transport_offset(const struct sk_buff *skb)
2080{
2081 return skb_transport_header(skb) - skb->data;
2082}
2083
2084static inline u32 skb_network_header_len(const struct sk_buff *skb)
2085{
2086 return skb->transport_header - skb->network_header;
2087}
2088
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002089static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2090{
2091 return skb->inner_transport_header - skb->inner_network_header;
2092}
2093
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002094static inline int skb_network_offset(const struct sk_buff *skb)
2095{
2096 return skb_network_header(skb) - skb->data;
2097}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03002098
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002099static inline int skb_inner_network_offset(const struct sk_buff *skb)
2100{
2101 return skb_inner_network_header(skb) - skb->data;
2102}
2103
Changli Gaof9599ce2010-08-04 04:43:44 +00002104static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2105{
2106 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2107}
2108
Linus Torvalds1da177e2005-04-16 15:20:36 -07002109/*
2110 * CPUs often take a performance hit when accessing unaligned memory
2111 * locations. The actual performance hit varies, it can be small if the
2112 * hardware handles it or large if we have to take an exception and fix it
2113 * in software.
2114 *
2115 * Since an ethernet header is 14 bytes network drivers often end up with
2116 * the IP header at an unaligned offset. The IP header can be aligned by
2117 * shifting the start of the packet by 2 bytes. Drivers should do this
2118 * with:
2119 *
Tobias Klauser8660c122009-07-13 22:48:16 +00002120 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002121 *
2122 * The downside to this alignment of the IP header is that the DMA is now
2123 * unaligned. On some architectures the cost of an unaligned DMA is high
2124 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00002125 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002126 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2127 * to be overridden.
2128 */
2129#ifndef NET_IP_ALIGN
2130#define NET_IP_ALIGN 2
2131#endif
2132
Anton Blanchard025be812006-03-31 02:27:06 -08002133/*
2134 * The networking layer reserves some headroom in skb data (via
2135 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2136 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08002137 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08002138 *
2139 * Unfortunately this headroom changes the DMA alignment of the resulting
2140 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2141 * on some architectures. An architecture can override this value,
2142 * perhaps setting it to a cacheline in size (since that will maintain
2143 * cacheline alignment of the DMA). It must be a power of 2.
2144 *
David S. Millerd6301d32009-02-08 19:24:13 -08002145 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08002146 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07002147 *
2148 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2149 * to reduce average number of cache lines per packet.
2150 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07002151 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08002152 */
2153#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07002154#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08002155#endif
2156
Joe Perches7965bd42013-09-26 14:48:15 -07002157int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002158
2159static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2160{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00002161 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07002162 WARN_ON(1);
2163 return;
2164 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002165 skb->len = len;
2166 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002167}
2168
Joe Perches7965bd42013-09-26 14:48:15 -07002169void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002170
2171static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2172{
Herbert Xu3cc0e872006-06-09 16:13:38 -07002173 if (skb->data_len)
2174 return ___pskb_trim(skb, len);
2175 __skb_trim(skb, len);
2176 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002177}
2178
2179static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2180{
2181 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2182}
2183
2184/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07002185 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2186 * @skb: buffer to alter
2187 * @len: new length
2188 *
2189 * This is identical to pskb_trim except that the caller knows that
2190 * the skb is not cloned so we should never get an error due to out-
2191 * of-memory.
2192 */
2193static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2194{
2195 int err = pskb_trim(skb, len);
2196 BUG_ON(err);
2197}
2198
2199/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002200 * skb_orphan - orphan a buffer
2201 * @skb: buffer to orphan
2202 *
2203 * If a buffer currently has an owner then we call the owner's
2204 * destructor function and make the @skb unowned. The buffer continues
2205 * to exist but is no longer charged to its former owner.
2206 */
2207static inline void skb_orphan(struct sk_buff *skb)
2208{
Eric Dumazetc34a7612013-07-30 16:11:15 -07002209 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002210 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002211 skb->destructor = NULL;
2212 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07002213 } else {
2214 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002215 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216}
2217
2218/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002219 * skb_orphan_frags - orphan the frags contained in a buffer
2220 * @skb: buffer to orphan frags from
2221 * @gfp_mask: allocation mask for replacement pages
2222 *
2223 * For each frag in the SKB which needs a destructor (i.e. has an
2224 * owner) create a copy of that frag and release the original
2225 * page by calling the destructor.
2226 */
2227static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2228{
2229 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2230 return 0;
2231 return skb_copy_ubufs(skb, gfp_mask);
2232}
2233
2234/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235 * __skb_queue_purge - empty a list
2236 * @list: list to empty
2237 *
2238 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2239 * the list and one reference dropped. This function does not take the
2240 * list lock and the caller must hold the relevant locks to use it.
2241 */
Joe Perches7965bd42013-09-26 14:48:15 -07002242void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002243static inline void __skb_queue_purge(struct sk_buff_head *list)
2244{
2245 struct sk_buff *skb;
2246 while ((skb = __skb_dequeue(list)) != NULL)
2247 kfree_skb(skb);
2248}
2249
Joe Perches7965bd42013-09-26 14:48:15 -07002250void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002251
Joe Perches7965bd42013-09-26 14:48:15 -07002252struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2253 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002254
2255/**
2256 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2257 * @dev: network device to receive on
2258 * @length: length to allocate
2259 *
2260 * Allocate a new &sk_buff and assign it a usage count of one. The
2261 * buffer has unspecified headroom built in. Users should allocate
2262 * the headroom they think they need without accounting for the
2263 * built in space. The built in space is used for optimisations.
2264 *
2265 * %NULL is returned if there is no free memory. Although this function
2266 * allocates memory it can be called from an interrupt.
2267 */
2268static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002269 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002270{
2271 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2272}
2273
Eric Dumazet6f532612012-05-18 05:12:12 +00002274/* legacy helper around __netdev_alloc_skb() */
2275static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2276 gfp_t gfp_mask)
2277{
2278 return __netdev_alloc_skb(NULL, length, gfp_mask);
2279}
2280
2281/* legacy helper around netdev_alloc_skb() */
2282static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2283{
2284 return netdev_alloc_skb(NULL, length);
2285}
2286
2287
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002288static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2289 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002290{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002291 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002292
2293 if (NET_IP_ALIGN && skb)
2294 skb_reserve(skb, NET_IP_ALIGN);
2295 return skb;
2296}
2297
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002298static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2299 unsigned int length)
2300{
2301 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2302}
2303
Alexander Duyck181edb22015-05-06 21:12:03 -07002304static inline void skb_free_frag(void *addr)
2305{
2306 __free_page_frag(addr);
2307}
2308
Alexander Duyckffde7322014-12-09 19:40:42 -08002309void *napi_alloc_frag(unsigned int fragsz);
Alexander Duyckfd11a832014-12-09 19:40:49 -08002310struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2311 unsigned int length, gfp_t gfp_mask);
2312static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2313 unsigned int length)
2314{
2315 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2316}
Alexander Duyckffde7322014-12-09 19:40:42 -08002317
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002318/**
Alexander Duyck71dfda52014-11-11 09:26:34 -08002319 * __dev_alloc_pages - allocate page for network Rx
2320 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2321 * @order: size of the allocation
2322 *
2323 * Allocate a new page.
2324 *
2325 * %NULL is returned if there is no free memory.
2326*/
2327static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2328 unsigned int order)
2329{
2330 /* This piece of code contains several assumptions.
2331 * 1. This is for device Rx, therefor a cold page is preferred.
2332 * 2. The expectation is the user wants a compound page.
2333 * 3. If requesting a order 0 page it will not be compound
2334 * due to the check to see if order has a value in prep_new_page
2335 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2336 * code in gfp_to_alloc_flags that should be enforcing this.
2337 */
2338 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2339
2340 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2341}
2342
2343static inline struct page *dev_alloc_pages(unsigned int order)
2344{
2345 return __dev_alloc_pages(GFP_ATOMIC, order);
2346}
2347
2348/**
2349 * __dev_alloc_page - allocate a page for network Rx
2350 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2351 *
2352 * Allocate a new page.
2353 *
2354 * %NULL is returned if there is no free memory.
2355 */
2356static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2357{
2358 return __dev_alloc_pages(gfp_mask, 0);
2359}
2360
2361static inline struct page *dev_alloc_page(void)
2362{
2363 return __dev_alloc_page(GFP_ATOMIC);
2364}
2365
2366/**
Mel Gorman06140022012-07-31 16:44:24 -07002367 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2368 * @page: The page that was allocated from skb_alloc_page
2369 * @skb: The skb that may need pfmemalloc set
2370 */
2371static inline void skb_propagate_pfmemalloc(struct page *page,
2372 struct sk_buff *skb)
2373{
Michal Hocko2f064f32015-08-21 14:11:51 -07002374 if (page_is_pfmemalloc(page))
Mel Gorman06140022012-07-31 16:44:24 -07002375 skb->pfmemalloc = true;
2376}
2377
Eric Dumazet564824b2010-10-11 19:05:25 +00002378/**
Masanari Iidae2278672014-02-18 22:54:36 +09002379 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002380 * @frag: the paged fragment
2381 *
2382 * Returns the &struct page associated with @frag.
2383 */
2384static inline struct page *skb_frag_page(const skb_frag_t *frag)
2385{
Ian Campbella8605c62011-10-19 23:01:49 +00002386 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002387}
2388
2389/**
2390 * __skb_frag_ref - take an addition reference on a paged fragment.
2391 * @frag: the paged fragment
2392 *
2393 * Takes an additional reference on the paged fragment @frag.
2394 */
2395static inline void __skb_frag_ref(skb_frag_t *frag)
2396{
2397 get_page(skb_frag_page(frag));
2398}
2399
2400/**
2401 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2402 * @skb: the buffer
2403 * @f: the fragment offset.
2404 *
2405 * Takes an additional reference on the @f'th paged fragment of @skb.
2406 */
2407static inline void skb_frag_ref(struct sk_buff *skb, int f)
2408{
2409 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2410}
2411
2412/**
2413 * __skb_frag_unref - release a reference on a paged fragment.
2414 * @frag: the paged fragment
2415 *
2416 * Releases a reference on the paged fragment @frag.
2417 */
2418static inline void __skb_frag_unref(skb_frag_t *frag)
2419{
2420 put_page(skb_frag_page(frag));
2421}
2422
2423/**
2424 * skb_frag_unref - release a reference on a paged fragment of an skb.
2425 * @skb: the buffer
2426 * @f: the fragment offset
2427 *
2428 * Releases a reference on the @f'th paged fragment of @skb.
2429 */
2430static inline void skb_frag_unref(struct sk_buff *skb, int f)
2431{
2432 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2433}
2434
2435/**
2436 * skb_frag_address - gets the address of the data contained in a paged fragment
2437 * @frag: the paged fragment buffer
2438 *
2439 * Returns the address of the data within @frag. The page must already
2440 * be mapped.
2441 */
2442static inline void *skb_frag_address(const skb_frag_t *frag)
2443{
2444 return page_address(skb_frag_page(frag)) + frag->page_offset;
2445}
2446
2447/**
2448 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2449 * @frag: the paged fragment buffer
2450 *
2451 * Returns the address of the data within @frag. Checks that the page
2452 * is mapped and returns %NULL otherwise.
2453 */
2454static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2455{
2456 void *ptr = page_address(skb_frag_page(frag));
2457 if (unlikely(!ptr))
2458 return NULL;
2459
2460 return ptr + frag->page_offset;
2461}
2462
2463/**
2464 * __skb_frag_set_page - sets the page contained in a paged fragment
2465 * @frag: the paged fragment
2466 * @page: the page to set
2467 *
2468 * Sets the fragment @frag to contain @page.
2469 */
2470static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2471{
Ian Campbella8605c62011-10-19 23:01:49 +00002472 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002473}
2474
2475/**
2476 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2477 * @skb: the buffer
2478 * @f: the fragment offset
2479 * @page: the page to set
2480 *
2481 * Sets the @f'th fragment of @skb to contain @page.
2482 */
2483static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2484 struct page *page)
2485{
2486 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2487}
2488
Eric Dumazet400dfd32013-10-17 16:27:07 -07002489bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2490
Ian Campbell131ea662011-08-19 06:25:00 +00002491/**
2492 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002493 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002494 * @frag: the paged fragment to map
2495 * @offset: the offset within the fragment (starting at the
2496 * fragment's own offset)
2497 * @size: the number of bytes to map
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002498 * @dir: the direction of the mapping (%PCI_DMA_*)
Ian Campbell131ea662011-08-19 06:25:00 +00002499 *
2500 * Maps the page associated with @frag to @device.
2501 */
2502static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2503 const skb_frag_t *frag,
2504 size_t offset, size_t size,
2505 enum dma_data_direction dir)
2506{
2507 return dma_map_page(dev, skb_frag_page(frag),
2508 frag->page_offset + offset, size, dir);
2509}
2510
Eric Dumazet117632e2011-12-03 21:39:53 +00002511static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2512 gfp_t gfp_mask)
2513{
2514 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2515}
2516
Octavian Purdilabad93e92014-06-12 01:36:26 +03002517
2518static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2519 gfp_t gfp_mask)
2520{
2521 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2522}
2523
2524
Ian Campbell131ea662011-08-19 06:25:00 +00002525/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002526 * skb_clone_writable - is the header of a clone writable
2527 * @skb: buffer to check
2528 * @len: length up to which to write
2529 *
2530 * Returns true if modifying the header part of the cloned buffer
2531 * does not requires the data to be copied.
2532 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002533static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002534{
2535 return !skb_header_cloned(skb) &&
2536 skb_headroom(skb) + len <= skb->hdr_len;
2537}
2538
Herbert Xud9cc2042007-09-16 16:21:16 -07002539static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2540 int cloned)
2541{
2542 int delta = 0;
2543
Herbert Xud9cc2042007-09-16 16:21:16 -07002544 if (headroom > skb_headroom(skb))
2545 delta = headroom - skb_headroom(skb);
2546
2547 if (delta || cloned)
2548 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2549 GFP_ATOMIC);
2550 return 0;
2551}
2552
Patrick McHardy334a8132007-06-25 04:35:20 -07002553/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002554 * skb_cow - copy header of skb when it is required
2555 * @skb: buffer to cow
2556 * @headroom: needed headroom
2557 *
2558 * If the skb passed lacks sufficient headroom or its data part
2559 * is shared, data is reallocated. If reallocation fails, an error
2560 * is returned and original skb is not changed.
2561 *
2562 * The result is skb with writable area skb->head...skb->tail
2563 * and at least @headroom of space at head.
2564 */
2565static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2566{
Herbert Xud9cc2042007-09-16 16:21:16 -07002567 return __skb_cow(skb, headroom, skb_cloned(skb));
2568}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002569
Herbert Xud9cc2042007-09-16 16:21:16 -07002570/**
2571 * skb_cow_head - skb_cow but only making the head writable
2572 * @skb: buffer to cow
2573 * @headroom: needed headroom
2574 *
2575 * This function is identical to skb_cow except that we replace the
2576 * skb_cloned check by skb_header_cloned. It should be used when
2577 * you only need to push on some header and do not need to modify
2578 * the data.
2579 */
2580static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2581{
2582 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002583}
2584
2585/**
2586 * skb_padto - pad an skbuff up to a minimal size
2587 * @skb: buffer to pad
2588 * @len: minimal length
2589 *
2590 * Pads up a buffer to ensure the trailing bytes exist and are
2591 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002592 * is untouched. Otherwise it is extended. Returns zero on
2593 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002594 */
Herbert Xu5b057c62006-06-23 02:06:41 -07002595static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002596{
2597 unsigned int size = skb->len;
2598 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002599 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002600 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002601}
2602
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002603/**
2604 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2605 * @skb: buffer to pad
2606 * @len: minimal length
2607 *
2608 * Pads up a buffer to ensure the trailing bytes exist and are
2609 * blanked. If the buffer already contains sufficient data it
2610 * is untouched. Otherwise it is extended. Returns zero on
2611 * success. The skb is freed on error.
2612 */
2613static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2614{
2615 unsigned int size = skb->len;
2616
2617 if (unlikely(size < len)) {
2618 len -= size;
2619 if (skb_pad(skb, len))
2620 return -ENOMEM;
2621 __skb_put(skb, len);
2622 }
2623 return 0;
2624}
2625
Linus Torvalds1da177e2005-04-16 15:20:36 -07002626static inline int skb_add_data(struct sk_buff *skb,
Al Viroaf2b0402014-11-27 21:44:24 -05002627 struct iov_iter *from, int copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002628{
2629 const int off = skb->len;
2630
2631 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viroaf2b0402014-11-27 21:44:24 -05002632 __wsum csum = 0;
2633 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2634 &csum, from) == copy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002635 skb->csum = csum_block_add(skb->csum, csum, off);
2636 return 0;
2637 }
Al Viroaf2b0402014-11-27 21:44:24 -05002638 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002639 return 0;
2640
2641 __skb_trim(skb, off);
2642 return -EFAULT;
2643}
2644
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002645static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2646 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002647{
2648 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002649 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002650
Ian Campbellea2ab692011-08-22 23:44:58 +00002651 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00002652 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002653 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002654 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002655}
2656
Herbert Xu364c6ba2006-06-09 16:10:40 -07002657static inline int __skb_linearize(struct sk_buff *skb)
2658{
2659 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2660}
2661
Linus Torvalds1da177e2005-04-16 15:20:36 -07002662/**
2663 * skb_linearize - convert paged skb to linear one
2664 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07002665 *
2666 * If there is no free memory -ENOMEM is returned, otherwise zero
2667 * is returned and the old skb data released.
2668 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07002669static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002670{
Herbert Xu364c6ba2006-06-09 16:10:40 -07002671 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2672}
2673
2674/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00002675 * skb_has_shared_frag - can any frag be overwritten
2676 * @skb: buffer to test
2677 *
2678 * Return true if the skb has at least one frag that might be modified
2679 * by an external entity (as in vmsplice()/sendfile())
2680 */
2681static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2682{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00002683 return skb_is_nonlinear(skb) &&
2684 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002685}
2686
2687/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07002688 * skb_linearize_cow - make sure skb is linear and writable
2689 * @skb: buffer to process
2690 *
2691 * If there is no free memory -ENOMEM is returned, otherwise zero
2692 * is returned and the old skb data released.
2693 */
2694static inline int skb_linearize_cow(struct sk_buff *skb)
2695{
2696 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2697 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002698}
2699
2700/**
2701 * skb_postpull_rcsum - update checksum for received skb after pull
2702 * @skb: buffer to update
2703 * @start: start of data before pull
2704 * @len: length of data pulled
2705 *
2706 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07002707 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2708 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709 */
2710
2711static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f92006-03-20 22:43:56 -08002712 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002713{
Patrick McHardy84fa7932006-08-29 16:44:56 -07002714 if (skb->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002716 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
Pravin B Shelar31b33df2015-09-28 17:24:25 -07002717 skb_checksum_start_offset(skb) < 0)
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002718 skb->ip_summed = CHECKSUM_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002719}
2720
Herbert Xucbb042f92006-03-20 22:43:56 -08002721unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2722
David S. Miller7ce5a272013-12-02 17:26:05 -05002723/**
2724 * pskb_trim_rcsum - trim received skb and update checksum
2725 * @skb: buffer to trim
2726 * @len: new length
2727 *
2728 * This is exactly the same as pskb_trim except that it ensures the
2729 * checksum of received packets are still valid after the operation.
2730 */
2731
2732static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2733{
2734 if (likely(len >= skb->len))
2735 return 0;
2736 if (skb->ip_summed == CHECKSUM_COMPLETE)
2737 skb->ip_summed = CHECKSUM_NONE;
2738 return __pskb_trim(skb, len);
2739}
2740
Linus Torvalds1da177e2005-04-16 15:20:36 -07002741#define skb_queue_walk(queue, skb) \
2742 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002743 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002744 skb = skb->next)
2745
James Chapman46f89142007-04-30 00:07:31 -07002746#define skb_queue_walk_safe(queue, skb, tmp) \
2747 for (skb = (queue)->next, tmp = skb->next; \
2748 skb != (struct sk_buff *)(queue); \
2749 skb = tmp, tmp = skb->next)
2750
David S. Miller1164f522008-09-23 00:49:44 -07002751#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002752 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07002753 skb = skb->next)
2754
2755#define skb_queue_walk_from_safe(queue, skb, tmp) \
2756 for (tmp = skb->next; \
2757 skb != (struct sk_buff *)(queue); \
2758 skb = tmp, tmp = skb->next)
2759
Stephen Hemminger300ce172005-10-30 13:47:34 -08002760#define skb_queue_reverse_walk(queue, skb) \
2761 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002762 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08002763 skb = skb->prev)
2764
David S. Miller686a2952011-01-20 22:47:32 -08002765#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2766 for (skb = (queue)->prev, tmp = skb->prev; \
2767 skb != (struct sk_buff *)(queue); \
2768 skb = tmp, tmp = skb->prev)
2769
2770#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2771 for (tmp = skb->prev; \
2772 skb != (struct sk_buff *)(queue); \
2773 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774
David S. Miller21dc3302010-08-23 00:13:46 -07002775static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07002776{
2777 return skb_shinfo(skb)->frag_list != NULL;
2778}
2779
2780static inline void skb_frag_list_init(struct sk_buff *skb)
2781{
2782 skb_shinfo(skb)->frag_list = NULL;
2783}
2784
David S. Milleree039872009-06-09 00:17:13 -07002785#define skb_walk_frags(skb, iter) \
2786 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2787
Rainer Weikusatea3793e2015-12-06 21:11:34 +00002788
2789int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
2790 const struct sk_buff *skb);
2791struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
2792 int *peeked, int *off, int *err,
2793 struct sk_buff **last);
Joe Perches7965bd42013-09-26 14:48:15 -07002794struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2795 int *peeked, int *off, int *err);
2796struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2797 int *err);
2798unsigned int datagram_poll(struct file *file, struct socket *sock,
2799 struct poll_table_struct *wait);
Al Viroc0371da2014-11-24 10:42:55 -05002800int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2801 struct iov_iter *to, int size);
David S. Miller51f3d022014-11-05 16:46:40 -05002802static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2803 struct msghdr *msg, int size)
2804{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002805 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
David S. Miller51f3d022014-11-05 16:46:40 -05002806}
Al Viroe5a4b0b2014-11-24 18:17:55 -05002807int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2808 struct msghdr *msg);
Al Viro3a654f92014-06-19 14:15:22 -04002809int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2810 struct iov_iter *from, int len);
Al Viro3a654f92014-06-19 14:15:22 -04002811int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
Joe Perches7965bd42013-09-26 14:48:15 -07002812void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2813void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2814int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07002815int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2816int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2817__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2818 int len, __wsum csum);
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002819ssize_t skb_socket_splice(struct sock *sk,
2820 struct pipe_inode_info *pipe,
2821 struct splice_pipe_desc *spd);
2822int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
Joe Perches7965bd42013-09-26 14:48:15 -07002823 struct pipe_inode_info *pipe, unsigned int len,
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002824 unsigned int flags,
2825 ssize_t (*splice_cb)(struct sock *,
2826 struct pipe_inode_info *,
2827 struct splice_pipe_desc *));
Joe Perches7965bd42013-09-26 14:48:15 -07002828void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01002829unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002830int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2831 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07002832void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2833int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2834void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01002835unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -07002836struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Vlad Yasevich0d5501c2014-08-08 14:42:13 -04002837struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
Jiri Pirkoe2195122014-11-19 14:05:01 +01002838int skb_ensure_writable(struct sk_buff *skb, int write_len);
Jiri Pirko93515d52014-11-19 14:05:02 +01002839int skb_vlan_pop(struct sk_buff *skb);
2840int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002841
Al Viro6ce8e9c2014-04-06 21:25:44 -04002842static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2843{
Al Viro21226ab2014-11-28 15:48:29 -05002844 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro6ce8e9c2014-04-06 21:25:44 -04002845}
2846
Al Viro7eab8d92014-04-06 21:51:23 -04002847static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2848{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002849 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro7eab8d92014-04-06 21:51:23 -04002850}
2851
Daniel Borkmann2817a332013-10-30 11:50:51 +01002852struct skb_checksum_ops {
2853 __wsum (*update)(const void *mem, int len, __wsum wsum);
2854 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2855};
2856
2857__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2858 __wsum csum, const struct skb_checksum_ops *ops);
2859__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2860 __wsum csum);
2861
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07002862static inline void * __must_check
2863__skb_header_pointer(const struct sk_buff *skb, int offset,
2864 int len, void *data, int hlen, void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002865{
Patrick McHardy55820ee2005-07-05 14:08:10 -07002866 if (hlen - offset >= len)
David S. Miller690e36e2014-08-23 12:13:41 -07002867 return data + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002868
David S. Miller690e36e2014-08-23 12:13:41 -07002869 if (!skb ||
2870 skb_copy_bits(skb, offset, buffer, len) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002871 return NULL;
2872
2873 return buffer;
2874}
2875
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07002876static inline void * __must_check
2877skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
David S. Miller690e36e2014-08-23 12:13:41 -07002878{
2879 return __skb_header_pointer(skb, offset, len, skb->data,
2880 skb_headlen(skb), buffer);
2881}
2882
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01002883/**
2884 * skb_needs_linearize - check if we need to linearize a given skb
2885 * depending on the given device features.
2886 * @skb: socket buffer to check
2887 * @features: net device features
2888 *
2889 * Returns true if either:
2890 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2891 * 2. skb is fragmented and the device does not support SG.
2892 */
2893static inline bool skb_needs_linearize(struct sk_buff *skb,
2894 netdev_features_t features)
2895{
2896 return skb_is_nonlinear(skb) &&
2897 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
2898 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
2899}
2900
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03002901static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
2902 void *to,
2903 const unsigned int len)
2904{
2905 memcpy(to, skb->data, len);
2906}
2907
2908static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
2909 const int offset, void *to,
2910 const unsigned int len)
2911{
2912 memcpy(to, skb->data + offset, len);
2913}
2914
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03002915static inline void skb_copy_to_linear_data(struct sk_buff *skb,
2916 const void *from,
2917 const unsigned int len)
2918{
2919 memcpy(skb->data, from, len);
2920}
2921
2922static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
2923 const int offset,
2924 const void *from,
2925 const unsigned int len)
2926{
2927 memcpy(skb->data + offset, from, len);
2928}
2929
Joe Perches7965bd42013-09-26 14:48:15 -07002930void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931
Patrick Ohlyac45f602009-02-12 05:03:37 +00002932static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
2933{
2934 return skb->tstamp;
2935}
2936
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002937/**
2938 * skb_get_timestamp - get timestamp from a skb
2939 * @skb: skb to get stamp from
2940 * @stamp: pointer to struct timeval to store stamp in
2941 *
2942 * Timestamps are stored in the skb as offsets to a base timestamp.
2943 * This function converts the offset back to a struct timeval and stores
2944 * it in stamp.
2945 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00002946static inline void skb_get_timestamp(const struct sk_buff *skb,
2947 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002948{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002949 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002950}
2951
Patrick Ohlyac45f602009-02-12 05:03:37 +00002952static inline void skb_get_timestampns(const struct sk_buff *skb,
2953 struct timespec *stamp)
2954{
2955 *stamp = ktime_to_timespec(skb->tstamp);
2956}
2957
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002958static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002959{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07002960 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002961}
2962
Stephen Hemminger164891a2007-04-23 22:26:16 -07002963static inline ktime_t net_timedelta(ktime_t t)
2964{
2965 return ktime_sub(ktime_get_real(), t);
2966}
2967
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07002968static inline ktime_t net_invalid_timestamp(void)
2969{
2970 return ktime_set(0, 0);
2971}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07002972
Alexander Duyck62bccb82014-09-04 13:31:35 -04002973struct sk_buff *skb_clone_sk(struct sk_buff *skb);
2974
Richard Cochranc1f19b52010-07-17 08:49:36 +00002975#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
2976
Joe Perches7965bd42013-09-26 14:48:15 -07002977void skb_clone_tx_timestamp(struct sk_buff *skb);
2978bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00002979
2980#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
2981
2982static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
2983{
2984}
2985
2986static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
2987{
2988 return false;
2989}
2990
2991#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
2992
2993/**
2994 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
2995 *
Richard Cochranda92b192011-10-21 00:49:15 +00002996 * PHY drivers may accept clones of transmitted packets for
2997 * timestamping via their phy_driver.txtstamp method. These drivers
Benjamin Poirier7a76a022015-08-07 09:32:21 -07002998 * must call this function to return the skb back to the stack with a
2999 * timestamp.
Richard Cochranda92b192011-10-21 00:49:15 +00003000 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00003001 * @skb: clone of the the original outgoing packet
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003002 * @hwtstamps: hardware time stamps
Richard Cochranc1f19b52010-07-17 08:49:36 +00003003 *
3004 */
3005void skb_complete_tx_timestamp(struct sk_buff *skb,
3006 struct skb_shared_hwtstamps *hwtstamps);
3007
Willem de Bruijne7fd2882014-08-04 22:11:48 -04003008void __skb_tstamp_tx(struct sk_buff *orig_skb,
3009 struct skb_shared_hwtstamps *hwtstamps,
3010 struct sock *sk, int tstype);
3011
Patrick Ohlyac45f602009-02-12 05:03:37 +00003012/**
3013 * skb_tstamp_tx - queue clone of skb with send time stamps
3014 * @orig_skb: the original outgoing packet
3015 * @hwtstamps: hardware time stamps, may be NULL if not available
3016 *
3017 * If the skb has a socket associated, then this function clones the
3018 * skb (thus sharing the actual data and optional structures), stores
3019 * the optional hardware time stamping information (if non NULL) or
3020 * generates a software time stamp (otherwise), then queues the clone
3021 * to the error queue of the socket. Errors are silently ignored.
3022 */
Joe Perches7965bd42013-09-26 14:48:15 -07003023void skb_tstamp_tx(struct sk_buff *orig_skb,
3024 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00003025
Richard Cochran4507a712010-07-17 08:48:28 +00003026static inline void sw_tx_timestamp(struct sk_buff *skb)
3027{
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003028 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3029 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
Richard Cochran4507a712010-07-17 08:48:28 +00003030 skb_tstamp_tx(skb, NULL);
3031}
3032
3033/**
3034 * skb_tx_timestamp() - Driver hook for transmit timestamping
3035 *
3036 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00003037 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00003038 *
David S. Miller73409f32013-12-27 13:04:33 -05003039 * Specifically, one should make absolutely sure that this function is
3040 * called before TX completion of this packet can trigger. Otherwise
3041 * the packet could potentially already be freed.
3042 *
Richard Cochran4507a712010-07-17 08:48:28 +00003043 * @skb: A socket buffer.
3044 */
3045static inline void skb_tx_timestamp(struct sk_buff *skb)
3046{
Richard Cochranc1f19b52010-07-17 08:49:36 +00003047 skb_clone_tx_timestamp(skb);
Richard Cochran4507a712010-07-17 08:48:28 +00003048 sw_tx_timestamp(skb);
3049}
3050
Johannes Berg6e3e9392011-11-09 10:15:42 +01003051/**
3052 * skb_complete_wifi_ack - deliver skb with wifi status
3053 *
3054 * @skb: the original outgoing packet
3055 * @acked: ack status
3056 *
3057 */
3058void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3059
Joe Perches7965bd42013-09-26 14:48:15 -07003060__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3061__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003062
Herbert Xu60476372007-04-09 11:59:39 -07003063static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3064{
Tom Herbert6edec0e2015-02-10 16:30:28 -08003065 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3066 skb->csum_valid ||
3067 (skb->ip_summed == CHECKSUM_PARTIAL &&
3068 skb_checksum_start_offset(skb) >= 0));
Herbert Xu60476372007-04-09 11:59:39 -07003069}
3070
Herbert Xufb286bb2005-11-10 13:01:24 -08003071/**
3072 * skb_checksum_complete - Calculate checksum of an entire packet
3073 * @skb: packet to process
3074 *
3075 * This function calculates the checksum over the entire packet plus
3076 * the value of skb->csum. The latter can be used to supply the
3077 * checksum of a pseudo header as used by TCP/UDP. It returns the
3078 * checksum.
3079 *
3080 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3081 * this function can be used to verify that checksum on received
3082 * packets. In that case the function should return zero if the
3083 * checksum is correct. In particular, this function will return zero
3084 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3085 * hardware has already verified the correctness of the checksum.
3086 */
Al Viro4381ca32007-07-15 21:00:11 +01003087static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08003088{
Herbert Xu60476372007-04-09 11:59:39 -07003089 return skb_csum_unnecessary(skb) ?
3090 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003091}
3092
Tom Herbert77cffe22014-08-27 21:26:46 -07003093static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3094{
3095 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3096 if (skb->csum_level == 0)
3097 skb->ip_summed = CHECKSUM_NONE;
3098 else
3099 skb->csum_level--;
3100 }
3101}
3102
3103static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3104{
3105 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3106 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3107 skb->csum_level++;
3108 } else if (skb->ip_summed == CHECKSUM_NONE) {
3109 skb->ip_summed = CHECKSUM_UNNECESSARY;
3110 skb->csum_level = 0;
3111 }
3112}
3113
Tom Herbert5a212322014-08-31 15:12:41 -07003114static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3115{
3116 /* Mark current checksum as bad (typically called from GRO
3117 * path). In the case that ip_summed is CHECKSUM_NONE
3118 * this must be the first checksum encountered in the packet.
3119 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3120 * checksum after the last one validated. For UDP, a zero
3121 * checksum can not be marked as bad.
3122 */
3123
3124 if (skb->ip_summed == CHECKSUM_NONE ||
3125 skb->ip_summed == CHECKSUM_UNNECESSARY)
3126 skb->csum_bad = 1;
3127}
3128
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003129/* Check if we need to perform checksum complete validation.
3130 *
3131 * Returns true if checksum complete is needed, false otherwise
3132 * (either checksum is unnecessary or zero checksum is allowed).
3133 */
3134static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3135 bool zero_okay,
3136 __sum16 check)
3137{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003138 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3139 skb->csum_valid = 1;
Tom Herbert77cffe22014-08-27 21:26:46 -07003140 __skb_decr_checksum_unnecessary(skb);
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003141 return false;
3142 }
3143
3144 return true;
3145}
3146
3147/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3148 * in checksum_init.
3149 */
3150#define CHECKSUM_BREAK 76
3151
Tom Herbert4e18b9a2015-04-20 14:10:04 -07003152/* Unset checksum-complete
3153 *
3154 * Unset checksum complete can be done when packet is being modified
3155 * (uncompressed for instance) and checksum-complete value is
3156 * invalidated.
3157 */
3158static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3159{
3160 if (skb->ip_summed == CHECKSUM_COMPLETE)
3161 skb->ip_summed = CHECKSUM_NONE;
3162}
3163
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003164/* Validate (init) checksum based on checksum complete.
3165 *
3166 * Return values:
3167 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3168 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3169 * checksum is stored in skb->csum for use in __skb_checksum_complete
3170 * non-zero: value of invalid checksum
3171 *
3172 */
3173static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3174 bool complete,
3175 __wsum psum)
3176{
3177 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3178 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003179 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003180 return 0;
3181 }
Tom Herbert5a212322014-08-31 15:12:41 -07003182 } else if (skb->csum_bad) {
3183 /* ip_summed == CHECKSUM_NONE in this case */
Eric Dumazetc91d4602015-05-15 05:48:07 -07003184 return (__force __sum16)1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003185 }
3186
3187 skb->csum = psum;
3188
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003189 if (complete || skb->len <= CHECKSUM_BREAK) {
3190 __sum16 csum;
3191
3192 csum = __skb_checksum_complete(skb);
3193 skb->csum_valid = !csum;
3194 return csum;
3195 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003196
3197 return 0;
3198}
3199
3200static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3201{
3202 return 0;
3203}
3204
3205/* Perform checksum validate (init). Note that this is a macro since we only
3206 * want to calculate the pseudo header which is an input function if necessary.
3207 * First we try to validate without any computation (checksum unnecessary) and
3208 * then calculate based on checksum complete calling the function to compute
3209 * pseudo header.
3210 *
3211 * Return values:
3212 * 0: checksum is validated or try to in skb_checksum_complete
3213 * non-zero: value of invalid checksum
3214 */
3215#define __skb_checksum_validate(skb, proto, complete, \
3216 zero_okay, check, compute_pseudo) \
3217({ \
3218 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003219 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003220 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3221 __ret = __skb_checksum_validate_complete(skb, \
3222 complete, compute_pseudo(skb, proto)); \
3223 __ret; \
3224})
3225
3226#define skb_checksum_init(skb, proto, compute_pseudo) \
3227 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3228
3229#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3230 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3231
3232#define skb_checksum_validate(skb, proto, compute_pseudo) \
3233 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3234
3235#define skb_checksum_validate_zero_check(skb, proto, check, \
3236 compute_pseudo) \
Sabrina Dubroca096a4cf2015-02-06 18:54:19 +01003237 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003238
3239#define skb_checksum_simple_validate(skb) \
3240 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3241
Tom Herbertd96535a2014-08-31 15:12:42 -07003242static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3243{
3244 return (skb->ip_summed == CHECKSUM_NONE &&
3245 skb->csum_valid && !skb->csum_bad);
3246}
3247
3248static inline void __skb_checksum_convert(struct sk_buff *skb,
3249 __sum16 check, __wsum pseudo)
3250{
3251 skb->csum = ~pseudo;
3252 skb->ip_summed = CHECKSUM_COMPLETE;
3253}
3254
3255#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3256do { \
3257 if (__skb_checksum_convert_check(skb)) \
3258 __skb_checksum_convert(skb, check, \
3259 compute_pseudo(skb, proto)); \
3260} while (0)
3261
Tom Herbert15e23962015-02-10 16:30:31 -08003262static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3263 u16 start, u16 offset)
3264{
3265 skb->ip_summed = CHECKSUM_PARTIAL;
3266 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3267 skb->csum_offset = offset - start;
3268}
3269
Tom Herbertdcdc8992015-02-02 16:07:34 -08003270/* Update skbuf and packet to reflect the remote checksum offload operation.
3271 * When called, ptr indicates the starting point for skb->csum when
3272 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3273 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3274 */
3275static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
Tom Herbert15e23962015-02-10 16:30:31 -08003276 int start, int offset, bool nopartial)
Tom Herbertdcdc8992015-02-02 16:07:34 -08003277{
3278 __wsum delta;
3279
Tom Herbert15e23962015-02-10 16:30:31 -08003280 if (!nopartial) {
3281 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3282 return;
3283 }
3284
Tom Herbertdcdc8992015-02-02 16:07:34 -08003285 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3286 __skb_checksum_complete(skb);
3287 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3288 }
3289
3290 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3291
3292 /* Adjust skb->csum since we changed the packet */
3293 skb->csum = csum_add(skb->csum, delta);
3294}
3295
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003296#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07003297void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003298static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3299{
3300 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07003301 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003302}
3303static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3304{
3305 if (nfct)
3306 atomic_inc(&nfct->use);
3307}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003308#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003309#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3311{
3312 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3313 kfree(nf_bridge);
3314}
3315static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3316{
3317 if (nf_bridge)
3318 atomic_inc(&nf_bridge->use);
3319}
3320#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08003321static inline void nf_reset(struct sk_buff *skb)
3322{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003323#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003324 nf_conntrack_put(skb->nfct);
3325 skb->nfct = NULL;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003326#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003327#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003328 nf_bridge_put(skb->nf_bridge);
3329 skb->nf_bridge = NULL;
3330#endif
3331}
3332
Patrick McHardy124dff02013-04-05 20:42:05 +02003333static inline void nf_reset_trace(struct sk_buff *skb)
3334{
Florian Westphal478b3602014-02-15 23:48:45 +01003335#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00003336 skb->nf_trace = 0;
3337#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003338}
3339
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003340/* Note: This doesn't put any conntrack and bridge info in dst. */
Eric Dumazetb1937222014-09-28 22:18:47 -07003341static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3342 bool copy)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003343{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003344#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003345 dst->nfct = src->nfct;
3346 nf_conntrack_get(src->nfct);
Eric Dumazetb1937222014-09-28 22:18:47 -07003347 if (copy)
3348 dst->nfctinfo = src->nfctinfo;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003349#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003350#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003351 dst->nf_bridge = src->nf_bridge;
3352 nf_bridge_get(src->nf_bridge);
3353#endif
Florian Westphal478b3602014-02-15 23:48:45 +01003354#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Eric Dumazetb1937222014-09-28 22:18:47 -07003355 if (copy)
3356 dst->nf_trace = src->nf_trace;
Florian Westphal478b3602014-02-15 23:48:45 +01003357#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003358}
3359
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003360static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3361{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003362#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003363 nf_conntrack_put(dst->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003364#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003365#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003366 nf_bridge_put(dst->nf_bridge);
3367#endif
Eric Dumazetb1937222014-09-28 22:18:47 -07003368 __nf_copy(dst, src, true);
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003369}
3370
James Morris984bc162006-06-09 00:29:17 -07003371#ifdef CONFIG_NETWORK_SECMARK
3372static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3373{
3374 to->secmark = from->secmark;
3375}
3376
3377static inline void skb_init_secmark(struct sk_buff *skb)
3378{
3379 skb->secmark = 0;
3380}
3381#else
3382static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3383{ }
3384
3385static inline void skb_init_secmark(struct sk_buff *skb)
3386{ }
3387#endif
3388
Eric W. Biederman574f7192014-04-01 12:20:24 -07003389static inline bool skb_irq_freeable(const struct sk_buff *skb)
3390{
3391 return !skb->destructor &&
3392#if IS_ENABLED(CONFIG_XFRM)
3393 !skb->sp &&
3394#endif
3395#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3396 !skb->nfct &&
3397#endif
3398 !skb->_skb_refdst &&
3399 !skb_has_frag_list(skb);
3400}
3401
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003402static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3403{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003404 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003405}
3406
Stephen Hemminger92477442009-03-21 13:39:26 -07003407static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003408{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003409 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003410}
3411
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003412static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3413{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003414 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003415}
3416
David S. Millerd5a9e242009-01-27 16:22:11 -08003417static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3418{
3419 skb->queue_mapping = rx_queue + 1;
3420}
3421
Stephen Hemminger92477442009-03-21 13:39:26 -07003422static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003423{
3424 return skb->queue_mapping - 1;
3425}
3426
Stephen Hemminger92477442009-03-21 13:39:26 -07003427static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003428{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003429 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003430}
3431
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003432static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3433{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003434#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003435 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003436#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003437 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003438#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003439}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003440
Pravin B Shelar68c33162013-02-14 14:02:41 +00003441/* Keeps track of mac header offset relative to skb->head.
3442 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3443 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003444 * tunnel skb it points to outer mac header.
3445 * Keeps track of level of encapsulation of network headers.
3446 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003447struct skb_gso_cb {
Eric Dumazet3347c962013-10-19 11:42:56 -07003448 int mac_offset;
3449 int encap_level;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003450 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003451};
3452#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb)
3453
3454static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3455{
3456 return (skb_mac_header(inner_skb) - inner_skb->head) -
3457 SKB_GSO_CB(inner_skb)->mac_offset;
3458}
3459
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003460static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3461{
3462 int new_headroom, headroom;
3463 int ret;
3464
3465 headroom = skb_headroom(skb);
3466 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3467 if (ret)
3468 return ret;
3469
3470 new_headroom = skb_headroom(skb);
3471 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3472 return 0;
3473}
3474
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003475/* Compute the checksum for a gso segment. First compute the checksum value
3476 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3477 * then add in skb->csum (checksum from csum_start to end of packet).
3478 * skb->csum and csum_start are then updated to reflect the checksum of the
3479 * resultant packet starting from the transport header-- the resultant checksum
3480 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3481 * header.
3482 */
3483static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3484{
3485 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) -
Eric Dumazetc91d4602015-05-15 05:48:07 -07003486 skb_transport_offset(skb);
3487 __wsum partial;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003488
Eric Dumazetc91d4602015-05-15 05:48:07 -07003489 partial = csum_partial(skb_transport_header(skb), plen, skb->csum);
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003490 skb->csum = res;
3491 SKB_GSO_CB(skb)->csum_start -= plen;
3492
Eric Dumazetc91d4602015-05-15 05:48:07 -07003493 return csum_fold(partial);
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003494}
3495
David S. Millerbdcc0922012-03-07 20:53:36 -05003496static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07003497{
3498 return skb_shinfo(skb)->gso_size;
3499}
3500
Eric Dumazet36a8f392013-09-29 01:21:32 -07003501/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05003502static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02003503{
3504 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3505}
3506
Joe Perches7965bd42013-09-26 14:48:15 -07003507void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07003508
3509static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3510{
3511 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3512 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003513 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3514
Alexander Duyckb78462e2010-06-02 12:24:37 +00003515 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3516 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07003517 __skb_warn_lro_forwarding(skb);
3518 return true;
3519 }
3520 return false;
3521}
3522
Herbert Xu35fc92a2007-03-26 23:22:20 -07003523static inline void skb_forward_csum(struct sk_buff *skb)
3524{
3525 /* Unfortunately we don't support this one. Any brave souls? */
3526 if (skb->ip_summed == CHECKSUM_COMPLETE)
3527 skb->ip_summed = CHECKSUM_NONE;
3528}
3529
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003530/**
3531 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3532 * @skb: skb to check
3533 *
3534 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3535 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3536 * use this helper, to document places where we make this assertion.
3537 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003538static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003539{
3540#ifdef DEBUG
3541 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3542#endif
3543}
3544
Rusty Russellf35d9d82008-02-04 23:49:54 -05003545bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00003546
Paul Durranted1f50c2014-01-09 10:02:46 +00003547int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
Linus Lüssing9afd85c2015-05-02 14:01:07 +02003548struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3549 unsigned int transport_len,
3550 __sum16(*skb_chkf)(struct sk_buff *skb));
Paul Durranted1f50c2014-01-09 10:02:46 +00003551
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00003552/**
3553 * skb_head_is_locked - Determine if the skb->head is locked down
3554 * @skb: skb to check
3555 *
3556 * The head on skbs build around a head frag can be removed if they are
3557 * not cloned. This function returns true if the skb head is locked down
3558 * due to either being allocated via kmalloc, or by being a clone with
3559 * multiple references to the head.
3560 */
3561static inline bool skb_head_is_locked(const struct sk_buff *skb)
3562{
3563 return !skb->head_frag || skb_cloned(skb);
3564}
Florian Westphalfe6cc552014-02-13 23:09:12 +01003565
3566/**
3567 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3568 *
3569 * @skb: GSO skb
3570 *
3571 * skb_gso_network_seglen is used to determine the real size of the
3572 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3573 *
3574 * The MAC/L2 header is not accounted for.
3575 */
3576static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3577{
3578 unsigned int hdr_len = skb_transport_header(skb) -
3579 skb_network_header(skb);
3580 return hdr_len + skb_gso_transport_seglen(skb);
3581}
Thomas Grafee122c72015-07-21 10:43:58 +02003582
Linus Torvalds1da177e2005-04-16 15:20:36 -07003583#endif /* __KERNEL__ */
3584#endif /* _LINUX_SKBUFF_H */