blob: aa3f9d7e8d5ca455387efa22d4a9d3a079a56f0c [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
Tom Herbert7a6ae712015-12-14 11:19:47 -080042/* The interface for checksum offload between the stack and networking drivers
43 * is as follows...
44 *
45 * A. IP checksum related features
46 *
47 * Drivers advertise checksum offload capabilities in the features of a device.
48 * From the stack's point of view these are capabilities offered by the driver,
49 * a driver typically only advertises features that it is capable of offloading
50 * to its device.
51 *
52 * The checksum related features are:
53 *
54 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
55 * IP (one's complement) checksum for any combination
56 * of protocols or protocol layering. The checksum is
57 * computed and set in a packet per the CHECKSUM_PARTIAL
58 * interface (see below).
59 *
60 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
61 * TCP or UDP packets over IPv4. These are specifically
62 * unencapsulated packets of the form IPv4|TCP or
63 * IPv4|UDP where the Protocol field in the IPv4 header
64 * is TCP or UDP. The IPv4 header may contain IP options
65 * This feature cannot be set in features for a device
66 * with NETIF_F_HW_CSUM also set. This feature is being
67 * DEPRECATED (see below).
68 *
69 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
70 * TCP or UDP packets over IPv6. These are specifically
71 * unencapsulated packets of the form IPv6|TCP or
72 * IPv4|UDP where the Next Header field in the IPv6
73 * header is either TCP or UDP. IPv6 extension headers
74 * are not supported with this feature. This feature
75 * cannot be set in features for a device with
76 * NETIF_F_HW_CSUM also set. This feature is being
77 * DEPRECATED (see below).
78 *
79 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
80 * This flag is used only used to disable the RX checksum
81 * feature for a device. The stack will accept receive
82 * checksum indication in packets received on a device
83 * regardless of whether NETIF_F_RXCSUM is set.
84 *
85 * B. Checksumming of received packets by device. Indication of checksum
86 * verification is in set skb->ip_summed. Possible values are:
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010087 *
88 * CHECKSUM_NONE:
89 *
Tom Herbert7a6ae712015-12-14 11:19:47 -080090 * Device did not checksum this packet e.g. due to lack of capabilities.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010091 * The packet contains full (though not verified) checksum in packet but
92 * not in skb->csum. Thus, skb->csum is undefined in this case.
93 *
94 * CHECKSUM_UNNECESSARY:
95 *
96 * The hardware you're dealing with doesn't calculate the full checksum
97 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
Tom Herbert77cffe22014-08-27 21:26:46 -070098 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
99 * if their checksums are okay. skb->csum is still undefined in this case
Tom Herbert7a6ae712015-12-14 11:19:47 -0800100 * though. A driver or device must never modify the checksum field in the
101 * packet even if checksum is verified.
Tom Herbert77cffe22014-08-27 21:26:46 -0700102 *
103 * CHECKSUM_UNNECESSARY is applicable to following protocols:
104 * TCP: IPv6 and IPv4.
105 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
106 * zero UDP checksum for either IPv4 or IPv6, the networking stack
107 * may perform further validation in this case.
108 * GRE: only if the checksum is present in the header.
109 * SCTP: indicates the CRC in SCTP header has been validated.
110 *
111 * skb->csum_level indicates the number of consecutive checksums found in
112 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
113 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
114 * and a device is able to verify the checksums for UDP (possibly zero),
115 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
116 * two. If the device were only able to verify the UDP checksum and not
117 * GRE, either because it doesn't support GRE checksum of because GRE
118 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
119 * not considered in this case).
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100120 *
121 * CHECKSUM_COMPLETE:
122 *
123 * This is the most generic way. The device supplied checksum of the _whole_
124 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
125 * hardware doesn't need to parse L3/L4 headers to implement this.
126 *
127 * Note: Even if device supports only some protocols, but is able to produce
128 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
129 *
130 * CHECKSUM_PARTIAL:
131 *
Tom Herbert6edec0e2015-02-10 16:30:28 -0800132 * A checksum is set up to be offloaded to a device as described in the
133 * output description for CHECKSUM_PARTIAL. This may occur on a packet
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100134 * received directly from another Linux OS, e.g., a virtualized Linux kernel
Tom Herbert6edec0e2015-02-10 16:30:28 -0800135 * on the same host, or it may be set in the input path in GRO or remote
136 * checksum offload. For the purposes of checksum verification, the checksum
137 * referred to by skb->csum_start + skb->csum_offset and any preceding
138 * checksums in the packet are considered verified. Any checksums in the
139 * packet that are after the checksum being offloaded are not considered to
140 * be verified.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100141 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800142 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
143 * in the skb->ip_summed for a packet. Values are:
144 *
145 * CHECKSUM_PARTIAL:
146 *
147 * The driver is required to checksum the packet as seen by hard_start_xmit()
148 * from skb->csum_start up to the end, and to record/write the checksum at
149 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
150 * csum_start and csum_offset values are valid values given the length and
151 * offset of the packet, however they should not attempt to validate that the
152 * checksum refers to a legitimate transport layer checksum-- it is the
153 * purview of the stack to validate that csum_start and csum_offset are set
154 * correctly.
155 *
156 * When the stack requests checksum offload for a packet, the driver MUST
157 * ensure that the checksum is set correctly. A driver can either offload the
158 * checksum calculation to the device, or call skb_checksum_help (in the case
159 * that the device does not support offload for a particular checksum).
160 *
161 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
162 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
163 * checksum offload capability. If a device has limited checksum capabilities
164 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as
165 * described above) a helper function can be called to resolve
166 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper
167 * function takes a spec argument that describes the protocol layer that is
168 * supported for checksum offload and can be called for each packet. If a
169 * packet does not match the specification for offload, skb_checksum_help
170 * is called to resolve the checksum.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100171 *
172 * CHECKSUM_NONE:
173 *
174 * The skb was already checksummed by the protocol, or a checksum is not
175 * required.
176 *
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100177 * CHECKSUM_UNNECESSARY:
178 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800179 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
180 * output.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100181 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800182 * CHECKSUM_COMPLETE:
183 * Not used in checksum output. If a driver observes a packet with this value
184 * set in skbuff, if should treat as CHECKSUM_NONE being set.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100185 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800186 * D. Non-IP checksum (CRC) offloads
187 *
188 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
189 * offloading the SCTP CRC in a packet. To perform this offload the stack
190 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
191 * accordingly. Note the there is no indication in the skbuff that the
192 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports
193 * both IP checksum offload and SCTP CRC offload must verify which offload
194 * is configured for a packet presumably by inspecting packet headers.
195 *
196 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
197 * offloading the FCOE CRC in a packet. To perform this offload the stack
198 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
199 * accordingly. Note the there is no indication in the skbuff that the
200 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
201 * both IP checksum offload and FCOE CRC offload must verify which offload
202 * is configured for a packet presumably by inspecting packet headers.
203 *
204 * E. Checksumming on output with GSO.
205 *
206 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
207 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
208 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
209 * part of the GSO operation is implied. If a checksum is being offloaded
210 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
211 * are set to refer to the outermost checksum being offload (two offloaded
212 * checksums are possible with UDP encapsulation).
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100213 */
214
Herbert Xu60476372007-04-09 11:59:39 -0700215/* Don't change this without changing skb_csum_unnecessary! */
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100216#define CHECKSUM_NONE 0
217#define CHECKSUM_UNNECESSARY 1
218#define CHECKSUM_COMPLETE 2
219#define CHECKSUM_PARTIAL 3
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
Tom Herbert77cffe22014-08-27 21:26:46 -0700221/* Maximum value in skb->csum_level */
222#define SKB_MAX_CSUM_LEVEL 3
223
Tobias Klauser0bec8c82014-07-22 12:06:23 +0200224#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
David S. Millerfc910a22007-03-25 20:27:59 -0700225#define SKB_WITH_OVERHEAD(X) \
Herbert Xudeea84b2007-10-21 16:27:46 -0700226 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
David S. Millerfc910a22007-03-25 20:27:59 -0700227#define SKB_MAX_ORDER(X, ORDER) \
228 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
230#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
231
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000232/* return minimum truesize of one skb containing X bytes of data */
233#define SKB_TRUESIZE(X) ((X) + \
234 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
235 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237struct net_device;
David Howells716ea3a2007-04-02 20:19:53 -0700238struct scatterlist;
Jens Axboe9c55e012007-11-06 23:30:13 -0800239struct pipe_inode_info;
Herbert Xua8f820aa2014-11-07 21:22:22 +0800240struct iov_iter;
Alexander Duyckfd11a832014-12-09 19:40:49 -0800241struct napi_struct;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700243#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244struct nf_conntrack {
245 atomic_t use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246};
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700247#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200249#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250struct nf_bridge_info {
Eric Dumazetbf1ac5c2012-04-18 23:19:25 +0000251 atomic_t use;
Florian Westphal3eaf4022015-04-02 14:31:44 +0200252 enum {
253 BRNF_PROTO_UNCHANGED,
254 BRNF_PROTO_8021Q,
255 BRNF_PROTO_PPPOE
Florian Westphal7fb48c52015-05-03 22:05:28 +0200256 } orig_proto:8;
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200257 u8 pkt_otherhost:1;
258 u8 in_prerouting:1;
259 u8 bridged_dnat:1;
Bernhard Thaler411ffb42015-05-30 15:28:28 +0200260 __u16 frag_max_size;
Eric Dumazetbf1ac5c2012-04-18 23:19:25 +0000261 struct net_device *physindev;
Florian Westphal63cdbc02015-09-14 17:06:27 +0200262
263 /* always valid & non-NULL from FORWARD on, for physdev match */
264 struct net_device *physoutdev;
Florian Westphal7fb48c52015-05-03 22:05:28 +0200265 union {
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200266 /* prerouting: detect dnat in orig/reply direction */
Bernhard Thaler72b31f72015-05-30 15:27:40 +0200267 __be32 ipv4_daddr;
268 struct in6_addr ipv6_daddr;
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200269
270 /* after prerouting + nat detected: store original source
271 * mac since neigh resolution overwrites it, only used while
272 * skb is out in neigh layer.
273 */
274 char neigh_header[8];
Bernhard Thaler72b31f72015-05-30 15:27:40 +0200275 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276};
277#endif
278
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279struct sk_buff_head {
280 /* These two members must be first. */
281 struct sk_buff *next;
282 struct sk_buff *prev;
283
284 __u32 qlen;
285 spinlock_t lock;
286};
287
288struct sk_buff;
289
Ian Campbell9d4dde52011-12-22 23:39:14 +0000290/* To allow 64K frame to be packed as single skb without frag_list we
291 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
292 * buffers which do not start on a page boundary.
293 *
294 * Since GRO uses frags we allocate at least 16 regardless of page
295 * size.
Anton Blancharda715dea2011-03-27 14:57:26 +0000296 */
Ian Campbell9d4dde52011-12-22 23:39:14 +0000297#if (65536/PAGE_SIZE + 1) < 16
David S. Millereec00952011-03-29 23:34:08 -0700298#define MAX_SKB_FRAGS 16UL
Anton Blancharda715dea2011-03-27 14:57:26 +0000299#else
Ian Campbell9d4dde52011-12-22 23:39:14 +0000300#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
Anton Blancharda715dea2011-03-27 14:57:26 +0000301#endif
Hans Westgaard Ry5f74f822016-02-03 09:26:57 +0100302extern int sysctl_max_skb_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
Marcelo Ricardo Leitner3953c462016-06-02 15:05:40 -0300304/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
305 * segment using its current segmentation instead.
306 */
307#define GSO_BY_FRAGS 0xFFFF
308
Linus Torvalds1da177e2005-04-16 15:20:36 -0700309typedef struct skb_frag_struct skb_frag_t;
310
311struct skb_frag_struct {
Ian Campbella8605c62011-10-19 23:01:49 +0000312 struct {
313 struct page *p;
314 } page;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000315#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
David S. Millera309bb02007-07-30 18:47:03 -0700316 __u32 page_offset;
317 __u32 size;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000318#else
319 __u16 page_offset;
320 __u16 size;
321#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700322};
323
Eric Dumazet9e903e02011-10-18 21:00:24 +0000324static inline unsigned int skb_frag_size(const skb_frag_t *frag)
325{
326 return frag->size;
327}
328
329static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
330{
331 frag->size = size;
332}
333
334static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
335{
336 frag->size += delta;
337}
338
339static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
340{
341 frag->size -= delta;
342}
343
Patrick Ohlyac45f602009-02-12 05:03:37 +0000344#define HAVE_HW_TIME_STAMP
345
346/**
Randy Dunlapd3a21be2009-03-02 03:15:58 -0800347 * struct skb_shared_hwtstamps - hardware time stamps
Patrick Ohlyac45f602009-02-12 05:03:37 +0000348 * @hwtstamp: hardware time stamp transformed into duration
349 * since arbitrary point in time
Patrick Ohlyac45f602009-02-12 05:03:37 +0000350 *
351 * Software time stamps generated by ktime_get_real() are stored in
Willem de Bruijn4d276eb2014-07-25 18:01:32 -0400352 * skb->tstamp.
Patrick Ohlyac45f602009-02-12 05:03:37 +0000353 *
354 * hwtstamps can only be compared against other hwtstamps from
355 * the same device.
356 *
357 * This structure is attached to packets as part of the
358 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
359 */
360struct skb_shared_hwtstamps {
361 ktime_t hwtstamp;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000362};
363
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000364/* Definitions for tx_flags in struct skb_shared_info */
365enum {
366 /* generate hardware time stamp */
367 SKBTX_HW_TSTAMP = 1 << 0,
368
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400369 /* generate software time stamp when queueing packet to NIC */
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000370 SKBTX_SW_TSTAMP = 1 << 1,
371
372 /* device driver is going to provide hardware time stamp */
373 SKBTX_IN_PROGRESS = 1 << 2,
374
Shirley Maa6686f22011-07-06 12:22:12 +0000375 /* device driver supports TX zero-copy buffers */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000376 SKBTX_DEV_ZEROCOPY = 1 << 3,
Johannes Berg6e3e9392011-11-09 10:15:42 +0100377
378 /* generate wifi status information (where possible) */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000379 SKBTX_WIFI_STATUS = 1 << 4,
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000380
381 /* This indicates at least one fragment might be overwritten
382 * (as in vmsplice(), sendfile() ...)
383 * If we need to compute a TX checksum, we'll need to copy
384 * all frags to avoid possible bad checksum
385 */
386 SKBTX_SHARED_FRAG = 1 << 5,
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400387
388 /* generate software time stamp when entering packet scheduling */
389 SKBTX_SCHED_TSTAMP = 1 << 6,
Shirley Maa6686f22011-07-06 12:22:12 +0000390};
391
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400392#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
Soheil Hassas Yeganeh0a2cf202016-04-27 23:39:01 -0400393 SKBTX_SCHED_TSTAMP)
Willem de Bruijnf24b9be2014-08-04 22:11:45 -0400394#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
395
Shirley Maa6686f22011-07-06 12:22:12 +0000396/*
397 * The callback notifies userspace to release buffers when skb DMA is done in
398 * lower device, the skb last reference should be 0 when calling this.
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000399 * The zerocopy_success argument is true if zero copy transmit occurred,
400 * false on data copy or out of memory error caused by data copy attempt.
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000401 * The ctx field is used to track device context.
402 * The desc field is used to track userspace buffer index.
Shirley Maa6686f22011-07-06 12:22:12 +0000403 */
404struct ubuf_info {
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000405 void (*callback)(struct ubuf_info *, bool zerocopy_success);
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000406 void *ctx;
Shirley Maa6686f22011-07-06 12:22:12 +0000407 unsigned long desc;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000408};
409
Linus Torvalds1da177e2005-04-16 15:20:36 -0700410/* This data is invariant across clones and lives at
411 * the end of the header data, ie. at skb->end.
412 */
413struct skb_shared_info {
Ian Campbell9f42f122012-01-05 07:13:39 +0000414 unsigned char nr_frags;
415 __u8 tx_flags;
Herbert Xu79671682006-06-22 02:40:14 -0700416 unsigned short gso_size;
417 /* Warning: this field is not always filled in (UFO)! */
418 unsigned short gso_segs;
419 unsigned short gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700420 struct sk_buff *frag_list;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000421 struct skb_shared_hwtstamps hwtstamps;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400422 u32 tskey;
Ian Campbell9f42f122012-01-05 07:13:39 +0000423 __be32 ip6_frag_id;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700424
425 /*
426 * Warning : all fields before dataref are cleared in __alloc_skb()
427 */
428 atomic_t dataref;
429
Johann Baudy69e3c752009-05-18 22:11:22 -0700430 /* Intermediate layers must ensure that destructor_arg
431 * remains valid until skb destructor */
432 void * destructor_arg;
Shirley Maa6686f22011-07-06 12:22:12 +0000433
Eric Dumazetfed66382010-07-22 19:09:08 +0000434 /* must be last field, see pskb_expand_head() */
435 skb_frag_t frags[MAX_SKB_FRAGS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700436};
437
438/* We divide dataref into two halves. The higher 16 bits hold references
439 * to the payload part of skb->data. The lower 16 bits hold references to
Patrick McHardy334a8132007-06-25 04:35:20 -0700440 * the entire skb->data. A clone of a headerless skb holds the length of
441 * the header in skb->hdr_len.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700442 *
443 * All users must obey the rule that the skb->data reference count must be
444 * greater than or equal to the payload reference count.
445 *
446 * Holding a reference to the payload part means that the user does not
447 * care about modifications to the header part of skb->data.
448 */
449#define SKB_DATAREF_SHIFT 16
450#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
451
David S. Millerd179cd12005-08-17 14:57:30 -0700452
453enum {
Vijay Subramanianc8753d52014-10-02 10:00:43 -0700454 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
455 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
456 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
David S. Millerd179cd12005-08-17 14:57:30 -0700457};
458
Herbert Xu79671682006-06-22 02:40:14 -0700459enum {
460 SKB_GSO_TCPV4 = 1 << 0,
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700461 SKB_GSO_UDP = 1 << 1,
Herbert Xu576a30e2006-06-27 13:22:38 -0700462
463 /* This indicates the skb is from an untrusted source. */
464 SKB_GSO_DODGY = 1 << 2,
Michael Chanb0da85372006-06-29 12:30:00 -0700465
466 /* This indicates the tcp segment has CWR set. */
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700467 SKB_GSO_TCP_ECN = 1 << 3,
468
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400469 SKB_GSO_TCP_FIXEDID = 1 << 4,
Chris Leech01d5b2f2009-02-27 14:06:49 -0800470
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400471 SKB_GSO_TCPV6 = 1 << 5,
Pravin B Shelar68c33162013-02-14 14:02:41 +0000472
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400473 SKB_GSO_FCOE = 1 << 6,
Pravin B Shelar73136262013-03-07 13:21:51 +0000474
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400475 SKB_GSO_GRE = 1 << 7,
Simon Horman0d89d202013-05-23 21:02:52 +0000476
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400477 SKB_GSO_GRE_CSUM = 1 << 8,
Eric Dumazetcb32f512013-10-19 11:42:57 -0700478
Tom Herbert7e133182016-05-18 09:06:10 -0700479 SKB_GSO_IPXIP4 = 1 << 9,
Eric Dumazet61c1db72013-10-20 20:47:30 -0700480
Tom Herbert7e133182016-05-18 09:06:10 -0700481 SKB_GSO_IPXIP6 = 1 << 10,
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700482
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400483 SKB_GSO_UDP_TUNNEL = 1 << 11,
Tom Herbert4749c092014-06-04 17:20:23 -0700484
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400485 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
486
Alexander Duyck802ab552016-04-10 21:45:03 -0400487 SKB_GSO_PARTIAL = 1 << 13,
488
489 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
Herbert Xu79671682006-06-22 02:40:14 -0700490};
491
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700492#if BITS_PER_LONG > 32
493#define NET_SKBUFF_DATA_USES_OFFSET 1
494#endif
495
496#ifdef NET_SKBUFF_DATA_USES_OFFSET
497typedef unsigned int sk_buff_data_t;
498#else
499typedef unsigned char *sk_buff_data_t;
500#endif
501
Eric Dumazet363ec39232014-02-26 14:02:11 -0800502/**
503 * struct skb_mstamp - multi resolution time stamps
504 * @stamp_us: timestamp in us resolution
505 * @stamp_jiffies: timestamp in jiffies
506 */
507struct skb_mstamp {
508 union {
509 u64 v64;
510 struct {
511 u32 stamp_us;
512 u32 stamp_jiffies;
513 };
514 };
515};
516
517/**
518 * skb_mstamp_get - get current timestamp
519 * @cl: place to store timestamps
520 */
521static inline void skb_mstamp_get(struct skb_mstamp *cl)
522{
523 u64 val = local_clock();
524
525 do_div(val, NSEC_PER_USEC);
526 cl->stamp_us = (u32)val;
527 cl->stamp_jiffies = (u32)jiffies;
528}
529
530/**
531 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
532 * @t1: pointer to newest sample
533 * @t0: pointer to oldest sample
534 */
535static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
536 const struct skb_mstamp *t0)
537{
538 s32 delta_us = t1->stamp_us - t0->stamp_us;
539 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
540
541 /* If delta_us is negative, this might be because interval is too big,
542 * or local_clock() drift is too big : fallback using jiffies.
543 */
544 if (delta_us <= 0 ||
545 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
546
547 delta_us = jiffies_to_usecs(delta_jiffies);
548
549 return delta_us;
550}
551
Yuchung Cheng625a5e12015-10-16 21:57:45 -0700552static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
553 const struct skb_mstamp *t0)
554{
555 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
556
557 if (!diff)
558 diff = t1->stamp_us - t0->stamp_us;
559 return diff > 0;
560}
Eric Dumazet363ec39232014-02-26 14:02:11 -0800561
Linus Torvalds1da177e2005-04-16 15:20:36 -0700562/**
563 * struct sk_buff - socket buffer
564 * @next: Next buffer in list
565 * @prev: Previous buffer in list
Eric Dumazet363ec39232014-02-26 14:02:11 -0800566 * @tstamp: Time we arrived/left
Eric Dumazet56b17422014-11-03 08:19:53 -0800567 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700568 * @sk: Socket we are owned by
Linus Torvalds1da177e2005-04-16 15:20:36 -0700569 * @dev: Device we arrived on/are leaving by
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700570 * @cb: Control buffer. Free for use by every layer. Put private vars here
Eric Dumazet7fee2262010-05-11 23:19:48 +0000571 * @_skb_refdst: destination entry (with norefcount bit)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700572 * @sp: the security path, used for xfrm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700573 * @len: Length of actual data
574 * @data_len: Data length
575 * @mac_len: Length of link layer header
Patrick McHardy334a8132007-06-25 04:35:20 -0700576 * @hdr_len: writable header length of cloned skb
Herbert Xu663ead32007-04-09 11:59:07 -0700577 * @csum: Checksum (must include start/offset pair)
578 * @csum_start: Offset from skb->head where checksumming should start
579 * @csum_offset: Offset from csum_start where checksum should be stored
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700580 * @priority: Packet queueing priority
WANG Cong60ff7462014-05-04 16:39:18 -0700581 * @ignore_df: allow local fragmentation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700582 * @cloned: Head may be cloned (check refcnt to be sure)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700583 * @ip_summed: Driver fed us an IP checksum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700584 * @nohdr: Payload reference only, must not modify header
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700585 * @nfctinfo: Relationship of this skb to the connection
Linus Torvalds1da177e2005-04-16 15:20:36 -0700586 * @pkt_type: Packet class
Randy Dunlapc83c2482005-10-18 22:07:41 -0700587 * @fclone: skbuff clone status
Randy Dunlapc83c2482005-10-18 22:07:41 -0700588 * @ipvs_property: skbuff is owned by ipvs
Randy Dunlap31729362008-02-18 20:52:13 -0800589 * @peeked: this packet has been seen already, so stats have been
590 * done for it, don't do them again
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700591 * @nf_trace: netfilter packet trace flag
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700592 * @protocol: Packet protocol from driver
593 * @destructor: Destruct function
594 * @nfct: Associated connection, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700595 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
Eric Dumazet8964be42009-11-20 15:35:04 -0800596 * @skb_iif: ifindex of device we arrived on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700597 * @tc_index: Traffic control index
598 * @tc_verd: traffic control verdict
Tom Herbert61b905d2014-03-24 15:34:47 -0700599 * @hash: the packet hash
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700600 * @queue_mapping: Queue mapping for multiqueue devices
David S. Miller0b725a22014-08-25 15:51:53 -0700601 * @xmit_more: More SKBs are pending for this queue
Randy Dunlap553a5672008-04-20 10:51:01 -0700602 * @ndisc_nodetype: router type (from link layer)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700603 * @ooo_okay: allow the mapping of a socket to a queue to be changed
Tom Herbert61b905d2014-03-24 15:34:47 -0700604 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
Changli Gao4ca24622011-08-19 07:26:44 -0700605 * ports.
Tom Herberta3b18dd2014-07-01 21:33:17 -0700606 * @sw_hash: indicates hash was computed in software stack
Johannes Berg6e3e9392011-11-09 10:15:42 +0100607 * @wifi_acked_valid: wifi_acked was set
608 * @wifi_acked: whether frame was acked on wifi or not
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000609 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
Eliezer Tamir06021292013-06-10 11:39:50 +0300610 * @napi_id: id of the NAPI struct this skb came from
James Morris984bc162006-06-09 00:29:17 -0700611 * @secmark: security marking
Scott Feldman0c4f6912015-07-18 18:24:48 -0700612 * @offload_fwd_mark: fwding offload mark
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700613 * @mark: Generic packet mark
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000614 * @vlan_proto: vlan encapsulation protocol
Patrick McHardy6aa895b02008-07-14 22:49:06 -0700615 * @vlan_tci: vlan tag control information
Simon Horman0d89d202013-05-23 21:02:52 +0000616 * @inner_protocol: Protocol (encapsulation)
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000617 * @inner_transport_header: Inner transport layer header (encapsulation)
618 * @inner_network_header: Network layer header (encapsulation)
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000619 * @inner_mac_header: Link layer header (encapsulation)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700620 * @transport_header: Transport layer header
621 * @network_header: Network layer header
622 * @mac_header: Link layer header
623 * @tail: Tail pointer
624 * @end: End pointer
625 * @head: Head of buffer
626 * @data: Data head pointer
627 * @truesize: Buffer size
628 * @users: User count - see {datagram,tcp}.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700629 */
630
631struct sk_buff {
Eric Dumazet363ec39232014-02-26 14:02:11 -0800632 union {
Eric Dumazet56b17422014-11-03 08:19:53 -0800633 struct {
634 /* These two members must be first. */
635 struct sk_buff *next;
636 struct sk_buff *prev;
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000637
Eric Dumazet56b17422014-11-03 08:19:53 -0800638 union {
639 ktime_t tstamp;
640 struct skb_mstamp skb_mstamp;
641 };
642 };
643 struct rb_node rbnode; /* used in netem & tcp stack */
644 };
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000645 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647
Linus Torvalds1da177e2005-04-16 15:20:36 -0700648 /*
649 * This is the control buffer. It is free to use for every
650 * layer. Please put your private variables there. If you
651 * want to keep them across layers you have to do a skb_clone()
652 * first. This is owned by whoever has the skb queued ATM.
653 */
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000654 char cb[48] __aligned(8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655
Eric Dumazet7fee2262010-05-11 23:19:48 +0000656 unsigned long _skb_refdst;
Eric Dumazetb1937222014-09-28 22:18:47 -0700657 void (*destructor)(struct sk_buff *skb);
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000658#ifdef CONFIG_XFRM
659 struct sec_path *sp;
660#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700661#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
662 struct nf_conntrack *nfct;
663#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200664#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Eric Dumazetb1937222014-09-28 22:18:47 -0700665 struct nf_bridge_info *nf_bridge;
666#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700667 unsigned int len,
Patrick McHardy334a8132007-06-25 04:35:20 -0700668 data_len;
669 __u16 mac_len,
670 hdr_len;
Eric Dumazetb1937222014-09-28 22:18:47 -0700671
672 /* Following fields are _not_ copied in __copy_skb_header()
673 * Note that queue_mapping is here mostly to fill a hole.
674 */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200675 kmemcheck_bitfield_begin(flags1);
Eric Dumazetb1937222014-09-28 22:18:47 -0700676 __u16 queue_mapping;
677 __u8 cloned:1,
Harald Welte6869c4d2005-08-09 19:24:19 -0700678 nohdr:1,
Patrick McHardyb84f4cc2005-11-20 21:19:21 -0800679 fclone:2,
Herbert Xua59322b2007-12-05 01:53:40 -0800680 peeked:1,
Eric Dumazetb1937222014-09-28 22:18:47 -0700681 head_frag:1,
682 xmit_more:1;
683 /* one bit hole */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200684 kmemcheck_bitfield_end(flags1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700685
Eric Dumazetb1937222014-09-28 22:18:47 -0700686 /* fields enclosed in headers_start/headers_end are copied
687 * using a single memcpy() in __copy_skb_header()
688 */
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700689 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700690 __u32 headers_start[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700691 /* public: */
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200692
693/* if you move pkt_type around you also must adapt those constants */
694#ifdef __BIG_ENDIAN_BITFIELD
695#define PKT_TYPE_MAX (7 << 5)
696#else
697#define PKT_TYPE_MAX 7
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100698#endif
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200699#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
700
701 __u8 __pkt_type_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700702 __u8 pkt_type:3;
703 __u8 pfmemalloc:1;
704 __u8 ignore_df:1;
705 __u8 nfctinfo:3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700706
Eric Dumazetb1937222014-09-28 22:18:47 -0700707 __u8 nf_trace:1;
708 __u8 ip_summed:2;
709 __u8 ooo_okay:1;
710 __u8 l4_hash:1;
711 __u8 sw_hash:1;
712 __u8 wifi_acked_valid:1;
713 __u8 wifi_acked:1;
714
715 __u8 no_fcs:1;
716 /* Indicates the inner headers are valid in the skbuff. */
717 __u8 encapsulation:1;
718 __u8 encap_hdr_csum:1;
719 __u8 csum_valid:1;
720 __u8 csum_complete_sw:1;
721 __u8 csum_level:2;
722 __u8 csum_bad:1;
723
724#ifdef CONFIG_IPV6_NDISC_NODETYPE
725 __u8 ndisc_nodetype:2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700727 __u8 ipvs_property:1;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700728 __u8 inner_protocol_type:1;
Tom Herberte585f232014-11-04 09:06:54 -0800729 __u8 remcsum_offload:1;
730 /* 3 or 5 bit hole */
Alexander Duyck4031ae62012-01-27 06:22:53 +0000731
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732#ifdef CONFIG_NET_SCHED
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700733 __u16 tc_index; /* traffic control index */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734#ifdef CONFIG_NET_CLS_ACT
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700735 __u16 tc_verd; /* traffic control verdict */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700737#endif
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200738
Eric Dumazetb1937222014-09-28 22:18:47 -0700739 union {
740 __wsum csum;
741 struct {
742 __u16 csum_start;
743 __u16 csum_offset;
744 };
745 };
746 __u32 priority;
747 int skb_iif;
748 __u32 hash;
749 __be16 vlan_proto;
750 __u16 vlan_tci;
Eric Dumazet2bd82482015-02-03 23:48:24 -0800751#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
752 union {
753 unsigned int napi_id;
754 unsigned int sender_cpu;
755 };
Chris Leech97fc2f02006-05-23 17:55:33 -0700756#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700757 union {
James Morris984bc162006-06-09 00:29:17 -0700758#ifdef CONFIG_NETWORK_SECMARK
Scott Feldman0c4f6912015-07-18 18:24:48 -0700759 __u32 secmark;
James Morris984bc162006-06-09 00:29:17 -0700760#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700761#ifdef CONFIG_NET_SWITCHDEV
762 __u32 offload_fwd_mark;
763#endif
764 };
765
Neil Horman3b885782009-10-12 13:26:31 -0700766 union {
767 __u32 mark;
Eric Dumazet16fad692013-03-14 05:40:32 +0000768 __u32 reserved_tailroom;
Neil Horman3b885782009-10-12 13:26:31 -0700769 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700770
Tom Herbert8bce6d72014-09-29 20:22:29 -0700771 union {
772 __be16 inner_protocol;
773 __u8 inner_ipproto;
774 };
775
Simon Horman1a37e412013-05-23 21:02:51 +0000776 __u16 inner_transport_header;
777 __u16 inner_network_header;
778 __u16 inner_mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700779
780 __be16 protocol;
Simon Horman1a37e412013-05-23 21:02:51 +0000781 __u16 transport_header;
782 __u16 network_header;
783 __u16 mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700784
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700785 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700786 __u32 headers_end[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700787 /* public: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700788
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 /* These elements must be at the end, see alloc_skb() for details. */
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700790 sk_buff_data_t tail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700791 sk_buff_data_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700792 unsigned char *head,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700793 *data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700794 unsigned int truesize;
795 atomic_t users;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700796};
797
798#ifdef __KERNEL__
799/*
800 * Handling routines are only of interest to the kernel
801 */
802#include <linux/slab.h>
803
Linus Torvalds1da177e2005-04-16 15:20:36 -0700804
Mel Gormanc93bdd02012-07-31 16:44:19 -0700805#define SKB_ALLOC_FCLONE 0x01
806#define SKB_ALLOC_RX 0x02
Alexander Duyckfd11a832014-12-09 19:40:49 -0800807#define SKB_ALLOC_NAPI 0x04
Mel Gormanc93bdd02012-07-31 16:44:19 -0700808
809/* Returns true if the skb was allocated from PFMEMALLOC reserves */
810static inline bool skb_pfmemalloc(const struct sk_buff *skb)
811{
812 return unlikely(skb->pfmemalloc);
813}
814
Eric Dumazet7fee2262010-05-11 23:19:48 +0000815/*
816 * skb might have a dst pointer attached, refcounted or not.
817 * _skb_refdst low order bit is set if refcount was _not_ taken
818 */
819#define SKB_DST_NOREF 1UL
820#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
821
822/**
823 * skb_dst - returns skb dst_entry
824 * @skb: buffer
825 *
826 * Returns skb dst_entry, regardless of reference taken or not.
827 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000828static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
829{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000830 /* If refdst was not refcounted, check we still are in a
831 * rcu_read_lock section
832 */
833 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
834 !rcu_read_lock_held() &&
835 !rcu_read_lock_bh_held());
836 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
Eric Dumazetadf30902009-06-02 05:19:30 +0000837}
838
Eric Dumazet7fee2262010-05-11 23:19:48 +0000839/**
840 * skb_dst_set - sets skb dst
841 * @skb: buffer
842 * @dst: dst entry
843 *
844 * Sets skb dst, assuming a reference was taken on dst and should
845 * be released by skb_dst_drop()
846 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000847static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
848{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000849 skb->_skb_refdst = (unsigned long)dst;
850}
851
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200852/**
853 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
854 * @skb: buffer
855 * @dst: dst entry
856 *
857 * Sets skb dst, assuming a reference was not taken on dst.
858 * If dst entry is cached, we do not take reference and dst_release
859 * will be avoided by refdst_drop. If dst entry is not cached, we take
860 * reference, so that last dst_release can destroy the dst immediately.
861 */
862static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
863{
Hannes Frederic Sowadbfc4fb2014-12-06 19:19:42 +0100864 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
865 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200866}
Eric Dumazet7fee2262010-05-11 23:19:48 +0000867
868/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300869 * skb_dst_is_noref - Test if skb dst isn't refcounted
Eric Dumazet7fee2262010-05-11 23:19:48 +0000870 * @skb: buffer
871 */
872static inline bool skb_dst_is_noref(const struct sk_buff *skb)
873{
874 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
Eric Dumazetadf30902009-06-02 05:19:30 +0000875}
876
Eric Dumazet511c3f92009-06-02 05:14:27 +0000877static inline struct rtable *skb_rtable(const struct sk_buff *skb)
878{
Eric Dumazetadf30902009-06-02 05:19:30 +0000879 return (struct rtable *)skb_dst(skb);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000880}
881
Joe Perches7965bd42013-09-26 14:48:15 -0700882void kfree_skb(struct sk_buff *skb);
883void kfree_skb_list(struct sk_buff *segs);
884void skb_tx_error(struct sk_buff *skb);
885void consume_skb(struct sk_buff *skb);
886void __kfree_skb(struct sk_buff *skb);
Eric Dumazetd7e88832012-04-30 08:10:34 +0000887extern struct kmem_cache *skbuff_head_cache;
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000888
Joe Perches7965bd42013-09-26 14:48:15 -0700889void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
890bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
891 bool *fragstolen, int *delta_truesize);
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000892
Joe Perches7965bd42013-09-26 14:48:15 -0700893struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
894 int node);
Eric Dumazet2ea2f622015-04-24 16:05:01 -0700895struct sk_buff *__build_skb(void *data, unsigned int frag_size);
Joe Perches7965bd42013-09-26 14:48:15 -0700896struct sk_buff *build_skb(void *data, unsigned int frag_size);
David S. Millerd179cd12005-08-17 14:57:30 -0700897static inline struct sk_buff *alloc_skb(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100898 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700899{
Eric Dumazet564824b2010-10-11 19:05:25 +0000900 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700901}
902
Eric Dumazet2e4e4412014-09-17 04:49:49 -0700903struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
904 unsigned long data_len,
905 int max_page_order,
906 int *errcode,
907 gfp_t gfp_mask);
908
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700909/* Layout of fast clones : [skb1][skb2][fclone_ref] */
910struct sk_buff_fclones {
911 struct sk_buff skb1;
912
913 struct sk_buff skb2;
914
915 atomic_t fclone_ref;
916};
917
918/**
919 * skb_fclone_busy - check if fclone is busy
920 * @skb: buffer
921 *
Masanari Iidabda13fe2015-12-13 16:53:02 +0900922 * Returns true if skb is a fast clone, and its clone is not freed.
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700923 * Some drivers call skb_orphan() in their ndo_start_xmit(),
924 * so we also check that this didnt happen.
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700925 */
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700926static inline bool skb_fclone_busy(const struct sock *sk,
927 const struct sk_buff *skb)
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700928{
929 const struct sk_buff_fclones *fclones;
930
931 fclones = container_of(skb, struct sk_buff_fclones, skb1);
932
933 return skb->fclone == SKB_FCLONE_ORIG &&
Eric Dumazet6ffe75e2014-12-03 17:04:39 -0800934 atomic_read(&fclones->fclone_ref) > 1 &&
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700935 fclones->skb2.sk == sk;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700936}
937
David S. Millerd179cd12005-08-17 14:57:30 -0700938static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100939 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700940{
Mel Gormanc93bdd02012-07-31 16:44:19 -0700941 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700942}
943
Joe Perches7965bd42013-09-26 14:48:15 -0700944struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000945static inline struct sk_buff *alloc_skb_head(gfp_t priority)
946{
947 return __alloc_skb_head(priority, -1);
948}
949
Joe Perches7965bd42013-09-26 14:48:15 -0700950struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
951int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
952struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
953struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
Octavian Purdilabad93e92014-06-12 01:36:26 +0300954struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
955 gfp_t gfp_mask, bool fclone);
956static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
957 gfp_t gfp_mask)
958{
959 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
960}
Eric Dumazet117632e2011-12-03 21:39:53 +0000961
Joe Perches7965bd42013-09-26 14:48:15 -0700962int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
963struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
964 unsigned int headroom);
965struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
966 int newtailroom, gfp_t priority);
Fan Du25a91d82014-01-18 09:54:23 +0800967int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
968 int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -0700969int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
970 int len);
971int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
972int skb_pad(struct sk_buff *skb, int pad);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000973#define dev_kfree_skb(a) consume_skb(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700974
Joe Perches7965bd42013-09-26 14:48:15 -0700975int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
976 int getfrag(void *from, char *to, int offset,
977 int len, int odd, struct sk_buff *skb),
978 void *from, int length);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700979
Hannes Frederic Sowabe12a1f2015-05-21 16:59:58 +0200980int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
981 int offset, size_t size);
982
Eric Dumazetd94d9fe2009-11-04 09:50:58 -0800983struct skb_seq_state {
Thomas Graf677e90e2005-06-23 20:59:51 -0700984 __u32 lower_offset;
985 __u32 upper_offset;
986 __u32 frag_idx;
987 __u32 stepped_offset;
988 struct sk_buff *root_skb;
989 struct sk_buff *cur_skb;
990 __u8 *frag_data;
991};
992
Joe Perches7965bd42013-09-26 14:48:15 -0700993void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
994 unsigned int to, struct skb_seq_state *st);
995unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
996 struct skb_seq_state *st);
997void skb_abort_seq_read(struct skb_seq_state *st);
Thomas Graf677e90e2005-06-23 20:59:51 -0700998
Joe Perches7965bd42013-09-26 14:48:15 -0700999unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
Bojan Prtvar059a2442015-02-22 11:46:35 +01001000 unsigned int to, struct ts_config *config);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07001001
Tom Herbert09323cc2013-12-15 22:16:19 -08001002/*
1003 * Packet hash types specify the type of hash in skb_set_hash.
1004 *
1005 * Hash types refer to the protocol layer addresses which are used to
1006 * construct a packet's hash. The hashes are used to differentiate or identify
1007 * flows of the protocol layer for the hash type. Hash types are either
1008 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1009 *
1010 * Properties of hashes:
1011 *
1012 * 1) Two packets in different flows have different hash values
1013 * 2) Two packets in the same flow should have the same hash value
1014 *
1015 * A hash at a higher layer is considered to be more specific. A driver should
1016 * set the most specific hash possible.
1017 *
1018 * A driver cannot indicate a more specific hash than the layer at which a hash
1019 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1020 *
1021 * A driver may indicate a hash level which is less specific than the
1022 * actual layer the hash was computed on. For instance, a hash computed
1023 * at L4 may be considered an L3 hash. This should only be done if the
1024 * driver can't unambiguously determine that the HW computed the hash at
1025 * the higher layer. Note that the "should" in the second property above
1026 * permits this.
1027 */
1028enum pkt_hash_types {
1029 PKT_HASH_TYPE_NONE, /* Undefined type */
1030 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1031 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1032 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1033};
1034
Tom Herbertbcc83832015-09-01 09:24:24 -07001035static inline void skb_clear_hash(struct sk_buff *skb)
1036{
1037 skb->hash = 0;
1038 skb->sw_hash = 0;
1039 skb->l4_hash = 0;
1040}
1041
1042static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1043{
1044 if (!skb->l4_hash)
1045 skb_clear_hash(skb);
1046}
1047
1048static inline void
1049__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1050{
1051 skb->l4_hash = is_l4;
1052 skb->sw_hash = is_sw;
1053 skb->hash = hash;
1054}
1055
Tom Herbert09323cc2013-12-15 22:16:19 -08001056static inline void
1057skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1058{
Tom Herbertbcc83832015-09-01 09:24:24 -07001059 /* Used by drivers to set hash from HW */
1060 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1061}
1062
1063static inline void
1064__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1065{
1066 __skb_set_hash(skb, hash, true, is_l4);
Tom Herbert09323cc2013-12-15 22:16:19 -08001067}
1068
Tom Herberte5276932015-09-01 09:24:23 -07001069void __skb_get_hash(struct sk_buff *skb);
1070u32 skb_get_poff(const struct sk_buff *skb);
1071u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1072 const struct flow_keys *keys, int hlen);
1073__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1074 void *data, int hlen_proto);
1075
1076static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1077 int thoff, u8 ip_proto)
1078{
1079 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1080}
1081
1082void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1083 const struct flow_dissector_key *key,
1084 unsigned int key_count);
1085
1086bool __skb_flow_dissect(const struct sk_buff *skb,
1087 struct flow_dissector *flow_dissector,
1088 void *target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001089 void *data, __be16 proto, int nhoff, int hlen,
1090 unsigned int flags);
Tom Herberte5276932015-09-01 09:24:23 -07001091
1092static inline bool skb_flow_dissect(const struct sk_buff *skb,
1093 struct flow_dissector *flow_dissector,
Tom Herbertcd79a232015-09-01 09:24:27 -07001094 void *target_container, unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001095{
1096 return __skb_flow_dissect(skb, flow_dissector, target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001097 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001098}
1099
1100static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
Tom Herbertcd79a232015-09-01 09:24:27 -07001101 struct flow_keys *flow,
1102 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001103{
1104 memset(flow, 0, sizeof(*flow));
1105 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001106 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001107}
1108
1109static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1110 void *data, __be16 proto,
Tom Herbertcd79a232015-09-01 09:24:27 -07001111 int nhoff, int hlen,
1112 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001113{
1114 memset(flow, 0, sizeof(*flow));
1115 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001116 data, proto, nhoff, hlen, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001117}
1118
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001119static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001120{
Tom Herberta3b18dd2014-07-01 21:33:17 -07001121 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001122 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001123
Tom Herbert61b905d2014-03-24 15:34:47 -07001124 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001125}
1126
David S. Miller20a17bf2015-09-01 21:19:17 -07001127__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
Tom Herbertf70ea012015-07-31 16:52:10 -07001128
David S. Miller20a17bf2015-09-01 21:19:17 -07001129static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
Tom Herbertf70ea012015-07-31 16:52:10 -07001130{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001131 if (!skb->l4_hash && !skb->sw_hash) {
1132 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001133 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001134
Tom Herbertde4c1f82015-09-01 18:11:04 -07001135 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001136 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001137
1138 return skb->hash;
1139}
1140
David S. Miller20a17bf2015-09-01 21:19:17 -07001141__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
Tom Herbertf70ea012015-07-31 16:52:10 -07001142
David S. Miller20a17bf2015-09-01 21:19:17 -07001143static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
Tom Herbertf70ea012015-07-31 16:52:10 -07001144{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001145 if (!skb->l4_hash && !skb->sw_hash) {
1146 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001147 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001148
Tom Herbertde4c1f82015-09-01 18:11:04 -07001149 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001150 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001151
1152 return skb->hash;
1153}
1154
Tom Herbert50fb7992015-05-01 11:30:12 -07001155__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1156
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001157static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1158{
Tom Herbert61b905d2014-03-24 15:34:47 -07001159 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001160}
1161
Tom Herbert3df7a742013-12-15 22:16:29 -08001162static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1163{
Tom Herbert61b905d2014-03-24 15:34:47 -07001164 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -07001165 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -07001166 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -08001167};
1168
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001169#ifdef NET_SKBUFF_DATA_USES_OFFSET
1170static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1171{
1172 return skb->head + skb->end;
1173}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001174
1175static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1176{
1177 return skb->end;
1178}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001179#else
1180static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1181{
1182 return skb->end;
1183}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001184
1185static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1186{
1187 return skb->end - skb->head;
1188}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001189#endif
1190
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001192#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001193
Patrick Ohlyac45f602009-02-12 05:03:37 +00001194static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1195{
1196 return &skb_shinfo(skb)->hwtstamps;
1197}
1198
Linus Torvalds1da177e2005-04-16 15:20:36 -07001199/**
1200 * skb_queue_empty - check if a queue is empty
1201 * @list: queue head
1202 *
1203 * Returns true if the queue is empty, false otherwise.
1204 */
1205static inline int skb_queue_empty(const struct sk_buff_head *list)
1206{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001207 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001208}
1209
1210/**
David S. Millerfc7ebb22008-09-23 00:34:07 -07001211 * skb_queue_is_last - check if skb is the last entry in the queue
1212 * @list: queue head
1213 * @skb: buffer
1214 *
1215 * Returns true if @skb is the last buffer on the list.
1216 */
1217static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1218 const struct sk_buff *skb)
1219{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001220 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -07001221}
1222
1223/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001224 * skb_queue_is_first - check if skb is the first entry in the queue
1225 * @list: queue head
1226 * @skb: buffer
1227 *
1228 * Returns true if @skb is the first buffer on the list.
1229 */
1230static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1231 const struct sk_buff *skb)
1232{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001233 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001234}
1235
1236/**
David S. Miller249c8b42008-09-23 00:44:42 -07001237 * skb_queue_next - return the next packet in the queue
1238 * @list: queue head
1239 * @skb: current buffer
1240 *
1241 * Return the next packet in @list after @skb. It is only valid to
1242 * call this if skb_queue_is_last() evaluates to false.
1243 */
1244static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1245 const struct sk_buff *skb)
1246{
1247 /* This BUG_ON may seem severe, but if we just return then we
1248 * are going to dereference garbage.
1249 */
1250 BUG_ON(skb_queue_is_last(list, skb));
1251 return skb->next;
1252}
1253
1254/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001255 * skb_queue_prev - return the prev packet in the queue
1256 * @list: queue head
1257 * @skb: current buffer
1258 *
1259 * Return the prev packet in @list before @skb. It is only valid to
1260 * call this if skb_queue_is_first() evaluates to false.
1261 */
1262static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1263 const struct sk_buff *skb)
1264{
1265 /* This BUG_ON may seem severe, but if we just return then we
1266 * are going to dereference garbage.
1267 */
1268 BUG_ON(skb_queue_is_first(list, skb));
1269 return skb->prev;
1270}
1271
1272/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001273 * skb_get - reference buffer
1274 * @skb: buffer to reference
1275 *
1276 * Makes another reference to a socket buffer and returns a pointer
1277 * to the buffer.
1278 */
1279static inline struct sk_buff *skb_get(struct sk_buff *skb)
1280{
1281 atomic_inc(&skb->users);
1282 return skb;
1283}
1284
1285/*
1286 * If users == 1, we are the only owner and are can avoid redundant
1287 * atomic change.
1288 */
1289
1290/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001291 * skb_cloned - is the buffer a clone
1292 * @skb: buffer to check
1293 *
1294 * Returns true if the buffer was generated with skb_clone() and is
1295 * one of multiple shared copies of the buffer. Cloned buffers are
1296 * shared data so must not be written to under normal circumstances.
1297 */
1298static inline int skb_cloned(const struct sk_buff *skb)
1299{
1300 return skb->cloned &&
1301 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1302}
1303
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001304static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1305{
Mel Gormand0164ad2015-11-06 16:28:21 -08001306 might_sleep_if(gfpflags_allow_blocking(pri));
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001307
1308 if (skb_cloned(skb))
1309 return pskb_expand_head(skb, 0, 0, pri);
1310
1311 return 0;
1312}
1313
Linus Torvalds1da177e2005-04-16 15:20:36 -07001314/**
1315 * skb_header_cloned - is the header a clone
1316 * @skb: buffer to check
1317 *
1318 * Returns true if modifying the header part of the buffer requires
1319 * the data to be copied.
1320 */
1321static inline int skb_header_cloned(const struct sk_buff *skb)
1322{
1323 int dataref;
1324
1325 if (!skb->cloned)
1326 return 0;
1327
1328 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1329 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1330 return dataref != 1;
1331}
1332
Eric Dumazet9580bf22016-04-30 10:19:29 -07001333static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1334{
1335 might_sleep_if(gfpflags_allow_blocking(pri));
1336
1337 if (skb_header_cloned(skb))
1338 return pskb_expand_head(skb, 0, 0, pri);
1339
1340 return 0;
1341}
1342
Linus Torvalds1da177e2005-04-16 15:20:36 -07001343/**
1344 * skb_header_release - release reference to header
1345 * @skb: buffer to operate on
1346 *
1347 * Drop a reference to the header part of the buffer. This is done
1348 * by acquiring a payload reference. You must not read from the header
1349 * part of skb->data after this.
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001350 * Note : Check if you can use __skb_header_release() instead.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001351 */
1352static inline void skb_header_release(struct sk_buff *skb)
1353{
1354 BUG_ON(skb->nohdr);
1355 skb->nohdr = 1;
1356 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1357}
1358
1359/**
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001360 * __skb_header_release - release reference to header
1361 * @skb: buffer to operate on
1362 *
1363 * Variant of skb_header_release() assuming skb is private to caller.
1364 * We can avoid one atomic operation.
1365 */
1366static inline void __skb_header_release(struct sk_buff *skb)
1367{
1368 skb->nohdr = 1;
1369 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1370}
1371
1372
1373/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001374 * skb_shared - is the buffer shared
1375 * @skb: buffer to check
1376 *
1377 * Returns true if more than one person has a reference to this
1378 * buffer.
1379 */
1380static inline int skb_shared(const struct sk_buff *skb)
1381{
1382 return atomic_read(&skb->users) != 1;
1383}
1384
1385/**
1386 * skb_share_check - check if buffer is shared and if so clone it
1387 * @skb: buffer to check
1388 * @pri: priority for memory allocation
1389 *
1390 * If the buffer is shared the buffer is cloned and the old copy
1391 * drops a reference. A new clone with a single reference is returned.
1392 * If the buffer is not shared the original buffer is returned. When
1393 * being called from interrupt status or with spinlocks held pri must
1394 * be GFP_ATOMIC.
1395 *
1396 * NULL is returned on a memory allocation failure.
1397 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001398static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001399{
Mel Gormand0164ad2015-11-06 16:28:21 -08001400 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001401 if (skb_shared(skb)) {
1402 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001403
1404 if (likely(nskb))
1405 consume_skb(skb);
1406 else
1407 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001408 skb = nskb;
1409 }
1410 return skb;
1411}
1412
1413/*
1414 * Copy shared buffers into a new sk_buff. We effectively do COW on
1415 * packets to handle cases where we have a local reader and forward
1416 * and a couple of other messy ones. The normal one is tcpdumping
1417 * a packet thats being forwarded.
1418 */
1419
1420/**
1421 * skb_unshare - make a copy of a shared buffer
1422 * @skb: buffer to check
1423 * @pri: priority for memory allocation
1424 *
1425 * If the socket buffer is a clone then this function creates a new
1426 * copy of the data, drops a reference count on the old copy and returns
1427 * the new copy with the reference count at 1. If the buffer is not a clone
1428 * the original buffer is returned. When called with a spinlock held or
1429 * from interrupt state @pri must be %GFP_ATOMIC
1430 *
1431 * %NULL is returned on a memory allocation failure.
1432 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001433static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001434 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001435{
Mel Gormand0164ad2015-11-06 16:28:21 -08001436 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001437 if (skb_cloned(skb)) {
1438 struct sk_buff *nskb = skb_copy(skb, pri);
Alexander Aring31eff812014-10-10 23:10:47 +02001439
1440 /* Free our shared copy */
1441 if (likely(nskb))
1442 consume_skb(skb);
1443 else
1444 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001445 skb = nskb;
1446 }
1447 return skb;
1448}
1449
1450/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001451 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452 * @list_: list to peek at
1453 *
1454 * Peek an &sk_buff. Unlike most other operations you _MUST_
1455 * be careful with this one. A peek leaves the buffer on the
1456 * list and someone else may run off with it. You must hold
1457 * the appropriate locks or have a private queue to do this.
1458 *
1459 * Returns %NULL for an empty list or a pointer to the head element.
1460 * The reference count is not incremented and the reference is therefore
1461 * volatile. Use with caution.
1462 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001463static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001464{
Eric Dumazet18d07002012-04-30 16:31:46 +00001465 struct sk_buff *skb = list_->next;
1466
1467 if (skb == (struct sk_buff *)list_)
1468 skb = NULL;
1469 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001470}
1471
1472/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001473 * skb_peek_next - peek skb following the given one from a queue
1474 * @skb: skb to start from
1475 * @list_: list to peek at
1476 *
1477 * Returns %NULL when the end of the list is met or a pointer to the
1478 * next element. The reference count is not incremented and the
1479 * reference is therefore volatile. Use with caution.
1480 */
1481static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1482 const struct sk_buff_head *list_)
1483{
1484 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001485
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001486 if (next == (struct sk_buff *)list_)
1487 next = NULL;
1488 return next;
1489}
1490
1491/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001492 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493 * @list_: list to peek at
1494 *
1495 * Peek an &sk_buff. Unlike most other operations you _MUST_
1496 * be careful with this one. A peek leaves the buffer on the
1497 * list and someone else may run off with it. You must hold
1498 * the appropriate locks or have a private queue to do this.
1499 *
1500 * Returns %NULL for an empty list or a pointer to the tail element.
1501 * The reference count is not incremented and the reference is therefore
1502 * volatile. Use with caution.
1503 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001504static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505{
Eric Dumazet18d07002012-04-30 16:31:46 +00001506 struct sk_buff *skb = list_->prev;
1507
1508 if (skb == (struct sk_buff *)list_)
1509 skb = NULL;
1510 return skb;
1511
Linus Torvalds1da177e2005-04-16 15:20:36 -07001512}
1513
1514/**
1515 * skb_queue_len - get queue length
1516 * @list_: list to measure
1517 *
1518 * Return the length of an &sk_buff queue.
1519 */
1520static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1521{
1522 return list_->qlen;
1523}
1524
David S. Miller67fed452008-09-21 22:36:24 -07001525/**
1526 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1527 * @list: queue to initialize
1528 *
1529 * This initializes only the list and queue length aspects of
1530 * an sk_buff_head object. This allows to initialize the list
1531 * aspects of an sk_buff_head without reinitializing things like
1532 * the spinlock. It can also be used for on-stack sk_buff_head
1533 * objects where the spinlock is known to not be used.
1534 */
1535static inline void __skb_queue_head_init(struct sk_buff_head *list)
1536{
1537 list->prev = list->next = (struct sk_buff *)list;
1538 list->qlen = 0;
1539}
1540
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001541/*
1542 * This function creates a split out lock class for each invocation;
1543 * this is needed for now since a whole lot of users of the skb-queue
1544 * infrastructure in drivers have different locking usage (in hardirq)
1545 * than the networking core (in softirq only). In the long run either the
1546 * network layer or drivers should need annotation to consolidate the
1547 * main types of usage into 3 classes.
1548 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549static inline void skb_queue_head_init(struct sk_buff_head *list)
1550{
1551 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001552 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553}
1554
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001555static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1556 struct lock_class_key *class)
1557{
1558 skb_queue_head_init(list);
1559 lockdep_set_class(&list->lock, class);
1560}
1561
Linus Torvalds1da177e2005-04-16 15:20:36 -07001562/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001563 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001564 *
1565 * The "__skb_xxxx()" functions are the non-atomic ones that
1566 * can only be called with interrupts disabled.
1567 */
Joe Perches7965bd42013-09-26 14:48:15 -07001568void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1569 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001570static inline void __skb_insert(struct sk_buff *newsk,
1571 struct sk_buff *prev, struct sk_buff *next,
1572 struct sk_buff_head *list)
1573{
1574 newsk->next = next;
1575 newsk->prev = prev;
1576 next->prev = prev->next = newsk;
1577 list->qlen++;
1578}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001579
David S. Miller67fed452008-09-21 22:36:24 -07001580static inline void __skb_queue_splice(const struct sk_buff_head *list,
1581 struct sk_buff *prev,
1582 struct sk_buff *next)
1583{
1584 struct sk_buff *first = list->next;
1585 struct sk_buff *last = list->prev;
1586
1587 first->prev = prev;
1588 prev->next = first;
1589
1590 last->next = next;
1591 next->prev = last;
1592}
1593
1594/**
1595 * skb_queue_splice - join two skb lists, this is designed for stacks
1596 * @list: the new list to add
1597 * @head: the place to add it in the first list
1598 */
1599static inline void skb_queue_splice(const struct sk_buff_head *list,
1600 struct sk_buff_head *head)
1601{
1602 if (!skb_queue_empty(list)) {
1603 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001604 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001605 }
1606}
1607
1608/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001609 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001610 * @list: the new list to add
1611 * @head: the place to add it in the first list
1612 *
1613 * The list at @list is reinitialised
1614 */
1615static inline void skb_queue_splice_init(struct sk_buff_head *list,
1616 struct sk_buff_head *head)
1617{
1618 if (!skb_queue_empty(list)) {
1619 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001620 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001621 __skb_queue_head_init(list);
1622 }
1623}
1624
1625/**
1626 * skb_queue_splice_tail - join two skb lists, each list being a queue
1627 * @list: the new list to add
1628 * @head: the place to add it in the first list
1629 */
1630static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1631 struct sk_buff_head *head)
1632{
1633 if (!skb_queue_empty(list)) {
1634 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001635 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001636 }
1637}
1638
1639/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001640 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001641 * @list: the new list to add
1642 * @head: the place to add it in the first list
1643 *
1644 * Each of the lists is a queue.
1645 * The list at @list is reinitialised
1646 */
1647static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1648 struct sk_buff_head *head)
1649{
1650 if (!skb_queue_empty(list)) {
1651 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001652 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001653 __skb_queue_head_init(list);
1654 }
1655}
1656
Linus Torvalds1da177e2005-04-16 15:20:36 -07001657/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001658 * __skb_queue_after - queue a buffer at the list head
1659 * @list: list to use
1660 * @prev: place after this buffer
1661 * @newsk: buffer to queue
1662 *
1663 * Queue a buffer int the middle of a list. This function takes no locks
1664 * and you must therefore hold required locks before calling it.
1665 *
1666 * A buffer cannot be placed on two lists at the same time.
1667 */
1668static inline void __skb_queue_after(struct sk_buff_head *list,
1669 struct sk_buff *prev,
1670 struct sk_buff *newsk)
1671{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001672 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001673}
1674
Joe Perches7965bd42013-09-26 14:48:15 -07001675void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1676 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001677
Gerrit Renkerf5572852008-04-14 00:05:28 -07001678static inline void __skb_queue_before(struct sk_buff_head *list,
1679 struct sk_buff *next,
1680 struct sk_buff *newsk)
1681{
1682 __skb_insert(newsk, next->prev, next, list);
1683}
1684
Stephen Hemminger300ce172005-10-30 13:47:34 -08001685/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001686 * __skb_queue_head - queue a buffer at the list head
1687 * @list: list to use
1688 * @newsk: buffer to queue
1689 *
1690 * Queue a buffer at the start of a list. This function takes no locks
1691 * and you must therefore hold required locks before calling it.
1692 *
1693 * A buffer cannot be placed on two lists at the same time.
1694 */
Joe Perches7965bd42013-09-26 14:48:15 -07001695void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001696static inline void __skb_queue_head(struct sk_buff_head *list,
1697 struct sk_buff *newsk)
1698{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001699 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700}
1701
1702/**
1703 * __skb_queue_tail - queue a buffer at the list tail
1704 * @list: list to use
1705 * @newsk: buffer to queue
1706 *
1707 * Queue a buffer at the end of a list. This function takes no locks
1708 * and you must therefore hold required locks before calling it.
1709 *
1710 * A buffer cannot be placed on two lists at the same time.
1711 */
Joe Perches7965bd42013-09-26 14:48:15 -07001712void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713static inline void __skb_queue_tail(struct sk_buff_head *list,
1714 struct sk_buff *newsk)
1715{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001716 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001717}
1718
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 * remove sk_buff from list. _Must_ be called atomically, and with
1721 * the list known..
1722 */
Joe Perches7965bd42013-09-26 14:48:15 -07001723void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001724static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1725{
1726 struct sk_buff *next, *prev;
1727
1728 list->qlen--;
1729 next = skb->next;
1730 prev = skb->prev;
1731 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001732 next->prev = prev;
1733 prev->next = next;
1734}
1735
Gerrit Renkerf525c062008-04-14 00:04:12 -07001736/**
1737 * __skb_dequeue - remove from the head of the queue
1738 * @list: list to dequeue from
1739 *
1740 * Remove the head of the list. This function does not take any locks
1741 * so must be used with appropriate locks held only. The head item is
1742 * returned or %NULL if the list is empty.
1743 */
Joe Perches7965bd42013-09-26 14:48:15 -07001744struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001745static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1746{
1747 struct sk_buff *skb = skb_peek(list);
1748 if (skb)
1749 __skb_unlink(skb, list);
1750 return skb;
1751}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001752
1753/**
1754 * __skb_dequeue_tail - remove from the tail of the queue
1755 * @list: list to dequeue from
1756 *
1757 * Remove the tail of the list. This function does not take any locks
1758 * so must be used with appropriate locks held only. The tail item is
1759 * returned or %NULL if the list is empty.
1760 */
Joe Perches7965bd42013-09-26 14:48:15 -07001761struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001762static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1763{
1764 struct sk_buff *skb = skb_peek_tail(list);
1765 if (skb)
1766 __skb_unlink(skb, list);
1767 return skb;
1768}
1769
1770
David S. Millerbdcc0922012-03-07 20:53:36 -05001771static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001772{
1773 return skb->data_len;
1774}
1775
1776static inline unsigned int skb_headlen(const struct sk_buff *skb)
1777{
1778 return skb->len - skb->data_len;
1779}
1780
1781static inline int skb_pagelen(const struct sk_buff *skb)
1782{
1783 int i, len = 0;
1784
1785 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001786 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001787 return len + skb_headlen(skb);
1788}
1789
Ian Campbell131ea662011-08-19 06:25:00 +00001790/**
1791 * __skb_fill_page_desc - initialise a paged fragment in an skb
1792 * @skb: buffer containing fragment to be initialised
1793 * @i: paged fragment index to initialise
1794 * @page: the page to use for this fragment
1795 * @off: the offset to the data with @page
1796 * @size: the length of the data
1797 *
1798 * Initialises the @i'th fragment of @skb to point to &size bytes at
1799 * offset @off within @page.
1800 *
1801 * Does not take any additional reference on the fragment.
1802 */
1803static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1804 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001805{
1806 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1807
Mel Gormanc48a11c2012-07-31 16:44:23 -07001808 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001809 * Propagate page pfmemalloc to the skb if we can. The problem is
1810 * that not all callers have unique ownership of the page but rely
1811 * on page_is_pfmemalloc doing the right thing(tm).
Mel Gormanc48a11c2012-07-31 16:44:23 -07001812 */
Ian Campbella8605c62011-10-19 23:01:49 +00001813 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001814 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001815 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001816
1817 page = compound_head(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001818 if (page_is_pfmemalloc(page))
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001819 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001820}
1821
1822/**
1823 * skb_fill_page_desc - initialise a paged fragment in an skb
1824 * @skb: buffer containing fragment to be initialised
1825 * @i: paged fragment index to initialise
1826 * @page: the page to use for this fragment
1827 * @off: the offset to the data with @page
1828 * @size: the length of the data
1829 *
1830 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001831 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001832 * addition updates @skb such that @i is the last fragment.
1833 *
1834 * Does not take any additional reference on the fragment.
1835 */
1836static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1837 struct page *page, int off, int size)
1838{
1839 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840 skb_shinfo(skb)->nr_frags = i + 1;
1841}
1842
Joe Perches7965bd42013-09-26 14:48:15 -07001843void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1844 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001845
Jason Wangf8e617e2013-11-01 14:07:47 +08001846void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1847 unsigned int truesize);
1848
Linus Torvalds1da177e2005-04-16 15:20:36 -07001849#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001850#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1852
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001853#ifdef NET_SKBUFF_DATA_USES_OFFSET
1854static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1855{
1856 return skb->head + skb->tail;
1857}
1858
1859static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1860{
1861 skb->tail = skb->data - skb->head;
1862}
1863
1864static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1865{
1866 skb_reset_tail_pointer(skb);
1867 skb->tail += offset;
1868}
Simon Horman7cc46192013-05-28 20:34:29 +00001869
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001870#else /* NET_SKBUFF_DATA_USES_OFFSET */
1871static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1872{
1873 return skb->tail;
1874}
1875
1876static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1877{
1878 skb->tail = skb->data;
1879}
1880
1881static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1882{
1883 skb->tail = skb->data + offset;
1884}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001885
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001886#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1887
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888/*
1889 * Add data to an sk_buff
1890 */
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001891unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001892unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001893static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1894{
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001895 unsigned char *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001896 SKB_LINEAR_ASSERT(skb);
1897 skb->tail += len;
1898 skb->len += len;
1899 return tmp;
1900}
1901
Joe Perches7965bd42013-09-26 14:48:15 -07001902unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001903static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1904{
1905 skb->data -= len;
1906 skb->len += len;
1907 return skb->data;
1908}
1909
Joe Perches7965bd42013-09-26 14:48:15 -07001910unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1912{
1913 skb->len -= len;
1914 BUG_ON(skb->len < skb->data_len);
1915 return skb->data += len;
1916}
1917
David S. Miller47d29642010-05-02 02:21:44 -07001918static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1919{
1920 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1921}
1922
Joe Perches7965bd42013-09-26 14:48:15 -07001923unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924
1925static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1926{
1927 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07001928 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001929 return NULL;
1930 skb->len -= len;
1931 return skb->data += len;
1932}
1933
1934static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1935{
1936 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1937}
1938
1939static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1940{
1941 if (likely(len <= skb_headlen(skb)))
1942 return 1;
1943 if (unlikely(len > skb->len))
1944 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07001945 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001946}
1947
1948/**
1949 * skb_headroom - bytes at buffer head
1950 * @skb: buffer to check
1951 *
1952 * Return the number of bytes of free space at the head of an &sk_buff.
1953 */
Chuck Leverc2636b42007-10-23 21:07:32 -07001954static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955{
1956 return skb->data - skb->head;
1957}
1958
1959/**
1960 * skb_tailroom - bytes at buffer end
1961 * @skb: buffer to check
1962 *
1963 * Return the number of bytes of free space at the tail of an sk_buff
1964 */
1965static inline int skb_tailroom(const struct sk_buff *skb)
1966{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001967 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001968}
1969
1970/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00001971 * skb_availroom - bytes at buffer end
1972 * @skb: buffer to check
1973 *
1974 * Return the number of bytes of free space at the tail of an sk_buff
1975 * allocated by sk_stream_alloc()
1976 */
1977static inline int skb_availroom(const struct sk_buff *skb)
1978{
Eric Dumazet16fad692013-03-14 05:40:32 +00001979 if (skb_is_nonlinear(skb))
1980 return 0;
1981
1982 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00001983}
1984
1985/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001986 * skb_reserve - adjust headroom
1987 * @skb: buffer to alter
1988 * @len: bytes to move
1989 *
1990 * Increase the headroom of an empty &sk_buff by reducing the tail
1991 * room. This is only allowed for an empty buffer.
1992 */
David S. Miller8243126c2006-01-17 02:54:21 -08001993static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001994{
1995 skb->data += len;
1996 skb->tail += len;
1997}
1998
Benjamin Poirier1837b2e2016-02-29 15:03:33 -08001999/**
2000 * skb_tailroom_reserve - adjust reserved_tailroom
2001 * @skb: buffer to alter
2002 * @mtu: maximum amount of headlen permitted
2003 * @needed_tailroom: minimum amount of reserved_tailroom
2004 *
2005 * Set reserved_tailroom so that headlen can be as large as possible but
2006 * not larger than mtu and tailroom cannot be smaller than
2007 * needed_tailroom.
2008 * The required headroom should already have been reserved before using
2009 * this function.
2010 */
2011static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2012 unsigned int needed_tailroom)
2013{
2014 SKB_LINEAR_ASSERT(skb);
2015 if (mtu < skb_tailroom(skb) - needed_tailroom)
2016 /* use at most mtu */
2017 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2018 else
2019 /* use up to all available space */
2020 skb->reserved_tailroom = needed_tailroom;
2021}
2022
Tom Herbert8bce6d72014-09-29 20:22:29 -07002023#define ENCAP_TYPE_ETHER 0
2024#define ENCAP_TYPE_IPPROTO 1
2025
2026static inline void skb_set_inner_protocol(struct sk_buff *skb,
2027 __be16 protocol)
2028{
2029 skb->inner_protocol = protocol;
2030 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2031}
2032
2033static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2034 __u8 ipproto)
2035{
2036 skb->inner_ipproto = ipproto;
2037 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2038}
2039
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002040static inline void skb_reset_inner_headers(struct sk_buff *skb)
2041{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002042 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002043 skb->inner_network_header = skb->network_header;
2044 skb->inner_transport_header = skb->transport_header;
2045}
2046
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00002047static inline void skb_reset_mac_len(struct sk_buff *skb)
2048{
2049 skb->mac_len = skb->network_header - skb->mac_header;
2050}
2051
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002052static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2053 *skb)
2054{
2055 return skb->head + skb->inner_transport_header;
2056}
2057
Tom Herbert55dc5a92015-12-14 11:19:40 -08002058static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2059{
2060 return skb_inner_transport_header(skb) - skb->data;
2061}
2062
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002063static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2064{
2065 skb->inner_transport_header = skb->data - skb->head;
2066}
2067
2068static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2069 const int offset)
2070{
2071 skb_reset_inner_transport_header(skb);
2072 skb->inner_transport_header += offset;
2073}
2074
2075static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2076{
2077 return skb->head + skb->inner_network_header;
2078}
2079
2080static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2081{
2082 skb->inner_network_header = skb->data - skb->head;
2083}
2084
2085static inline void skb_set_inner_network_header(struct sk_buff *skb,
2086 const int offset)
2087{
2088 skb_reset_inner_network_header(skb);
2089 skb->inner_network_header += offset;
2090}
2091
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002092static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2093{
2094 return skb->head + skb->inner_mac_header;
2095}
2096
2097static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2098{
2099 skb->inner_mac_header = skb->data - skb->head;
2100}
2101
2102static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2103 const int offset)
2104{
2105 skb_reset_inner_mac_header(skb);
2106 skb->inner_mac_header += offset;
2107}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002108static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2109{
Cong Wang35d04612013-05-29 15:16:05 +08002110 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002111}
2112
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002113static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2114{
2115 return skb->head + skb->transport_header;
2116}
2117
2118static inline void skb_reset_transport_header(struct sk_buff *skb)
2119{
2120 skb->transport_header = skb->data - skb->head;
2121}
2122
2123static inline void skb_set_transport_header(struct sk_buff *skb,
2124 const int offset)
2125{
2126 skb_reset_transport_header(skb);
2127 skb->transport_header += offset;
2128}
2129
2130static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2131{
2132 return skb->head + skb->network_header;
2133}
2134
2135static inline void skb_reset_network_header(struct sk_buff *skb)
2136{
2137 skb->network_header = skb->data - skb->head;
2138}
2139
2140static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2141{
2142 skb_reset_network_header(skb);
2143 skb->network_header += offset;
2144}
2145
2146static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2147{
2148 return skb->head + skb->mac_header;
2149}
2150
2151static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2152{
Cong Wang35d04612013-05-29 15:16:05 +08002153 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002154}
2155
2156static inline void skb_reset_mac_header(struct sk_buff *skb)
2157{
2158 skb->mac_header = skb->data - skb->head;
2159}
2160
2161static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2162{
2163 skb_reset_mac_header(skb);
2164 skb->mac_header += offset;
2165}
2166
Timo Teräs0e3da5b2013-12-16 11:02:09 +02002167static inline void skb_pop_mac_header(struct sk_buff *skb)
2168{
2169 skb->mac_header = skb->network_header;
2170}
2171
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002172static inline void skb_probe_transport_header(struct sk_buff *skb,
2173 const int offset_hint)
2174{
2175 struct flow_keys keys;
2176
2177 if (skb_transport_header_was_set(skb))
2178 return;
Tom Herbertcd79a232015-09-01 09:24:27 -07002179 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
Tom Herbert42aecaa2015-06-04 09:16:39 -07002180 skb_set_transport_header(skb, keys.control.thoff);
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002181 else
2182 skb_set_transport_header(skb, offset_hint);
2183}
2184
Eric Dumazet03606892012-02-23 10:55:02 +00002185static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2186{
2187 if (skb_mac_header_was_set(skb)) {
2188 const unsigned char *old_mac = skb_mac_header(skb);
2189
2190 skb_set_mac_header(skb, -skb->mac_len);
2191 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2192 }
2193}
2194
Michał Mirosław04fb4512010-12-14 15:24:08 +00002195static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2196{
2197 return skb->csum_start - skb_headroom(skb);
2198}
2199
Alexander Duyck08b64fc2016-02-05 15:27:49 -08002200static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2201{
2202 return skb->head + skb->csum_start;
2203}
2204
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002205static inline int skb_transport_offset(const struct sk_buff *skb)
2206{
2207 return skb_transport_header(skb) - skb->data;
2208}
2209
2210static inline u32 skb_network_header_len(const struct sk_buff *skb)
2211{
2212 return skb->transport_header - skb->network_header;
2213}
2214
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002215static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2216{
2217 return skb->inner_transport_header - skb->inner_network_header;
2218}
2219
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002220static inline int skb_network_offset(const struct sk_buff *skb)
2221{
2222 return skb_network_header(skb) - skb->data;
2223}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03002224
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002225static inline int skb_inner_network_offset(const struct sk_buff *skb)
2226{
2227 return skb_inner_network_header(skb) - skb->data;
2228}
2229
Changli Gaof9599ce2010-08-04 04:43:44 +00002230static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2231{
2232 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2233}
2234
Linus Torvalds1da177e2005-04-16 15:20:36 -07002235/*
2236 * CPUs often take a performance hit when accessing unaligned memory
2237 * locations. The actual performance hit varies, it can be small if the
2238 * hardware handles it or large if we have to take an exception and fix it
2239 * in software.
2240 *
2241 * Since an ethernet header is 14 bytes network drivers often end up with
2242 * the IP header at an unaligned offset. The IP header can be aligned by
2243 * shifting the start of the packet by 2 bytes. Drivers should do this
2244 * with:
2245 *
Tobias Klauser8660c122009-07-13 22:48:16 +00002246 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 *
2248 * The downside to this alignment of the IP header is that the DMA is now
2249 * unaligned. On some architectures the cost of an unaligned DMA is high
2250 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00002251 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002252 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2253 * to be overridden.
2254 */
2255#ifndef NET_IP_ALIGN
2256#define NET_IP_ALIGN 2
2257#endif
2258
Anton Blanchard025be812006-03-31 02:27:06 -08002259/*
2260 * The networking layer reserves some headroom in skb data (via
2261 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2262 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08002263 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08002264 *
2265 * Unfortunately this headroom changes the DMA alignment of the resulting
2266 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2267 * on some architectures. An architecture can override this value,
2268 * perhaps setting it to a cacheline in size (since that will maintain
2269 * cacheline alignment of the DMA). It must be a power of 2.
2270 *
David S. Millerd6301d32009-02-08 19:24:13 -08002271 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08002272 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07002273 *
2274 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2275 * to reduce average number of cache lines per packet.
2276 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07002277 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08002278 */
2279#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07002280#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08002281#endif
2282
Joe Perches7965bd42013-09-26 14:48:15 -07002283int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002284
2285static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2286{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00002287 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07002288 WARN_ON(1);
2289 return;
2290 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002291 skb->len = len;
2292 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002293}
2294
Joe Perches7965bd42013-09-26 14:48:15 -07002295void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002296
2297static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2298{
Herbert Xu3cc0e872006-06-09 16:13:38 -07002299 if (skb->data_len)
2300 return ___pskb_trim(skb, len);
2301 __skb_trim(skb, len);
2302 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002303}
2304
2305static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2306{
2307 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2308}
2309
2310/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07002311 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2312 * @skb: buffer to alter
2313 * @len: new length
2314 *
2315 * This is identical to pskb_trim except that the caller knows that
2316 * the skb is not cloned so we should never get an error due to out-
2317 * of-memory.
2318 */
2319static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2320{
2321 int err = pskb_trim(skb, len);
2322 BUG_ON(err);
2323}
2324
2325/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002326 * skb_orphan - orphan a buffer
2327 * @skb: buffer to orphan
2328 *
2329 * If a buffer currently has an owner then we call the owner's
2330 * destructor function and make the @skb unowned. The buffer continues
2331 * to exist but is no longer charged to its former owner.
2332 */
2333static inline void skb_orphan(struct sk_buff *skb)
2334{
Eric Dumazetc34a7612013-07-30 16:11:15 -07002335 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002336 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002337 skb->destructor = NULL;
2338 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07002339 } else {
2340 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002341 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002342}
2343
2344/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002345 * skb_orphan_frags - orphan the frags contained in a buffer
2346 * @skb: buffer to orphan frags from
2347 * @gfp_mask: allocation mask for replacement pages
2348 *
2349 * For each frag in the SKB which needs a destructor (i.e. has an
2350 * owner) create a copy of that frag and release the original
2351 * page by calling the destructor.
2352 */
2353static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2354{
2355 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2356 return 0;
2357 return skb_copy_ubufs(skb, gfp_mask);
2358}
2359
2360/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002361 * __skb_queue_purge - empty a list
2362 * @list: list to empty
2363 *
2364 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2365 * the list and one reference dropped. This function does not take the
2366 * list lock and the caller must hold the relevant locks to use it.
2367 */
Joe Perches7965bd42013-09-26 14:48:15 -07002368void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002369static inline void __skb_queue_purge(struct sk_buff_head *list)
2370{
2371 struct sk_buff *skb;
2372 while ((skb = __skb_dequeue(list)) != NULL)
2373 kfree_skb(skb);
2374}
2375
Joe Perches7965bd42013-09-26 14:48:15 -07002376void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002377
Joe Perches7965bd42013-09-26 14:48:15 -07002378struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2379 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002380
2381/**
2382 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2383 * @dev: network device to receive on
2384 * @length: length to allocate
2385 *
2386 * Allocate a new &sk_buff and assign it a usage count of one. The
2387 * buffer has unspecified headroom built in. Users should allocate
2388 * the headroom they think they need without accounting for the
2389 * built in space. The built in space is used for optimisations.
2390 *
2391 * %NULL is returned if there is no free memory. Although this function
2392 * allocates memory it can be called from an interrupt.
2393 */
2394static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002395 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002396{
2397 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2398}
2399
Eric Dumazet6f532612012-05-18 05:12:12 +00002400/* legacy helper around __netdev_alloc_skb() */
2401static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2402 gfp_t gfp_mask)
2403{
2404 return __netdev_alloc_skb(NULL, length, gfp_mask);
2405}
2406
2407/* legacy helper around netdev_alloc_skb() */
2408static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2409{
2410 return netdev_alloc_skb(NULL, length);
2411}
2412
2413
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002414static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2415 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002416{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002417 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002418
2419 if (NET_IP_ALIGN && skb)
2420 skb_reserve(skb, NET_IP_ALIGN);
2421 return skb;
2422}
2423
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002424static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2425 unsigned int length)
2426{
2427 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2428}
2429
Alexander Duyck181edb22015-05-06 21:12:03 -07002430static inline void skb_free_frag(void *addr)
2431{
2432 __free_page_frag(addr);
2433}
2434
Alexander Duyckffde7322014-12-09 19:40:42 -08002435void *napi_alloc_frag(unsigned int fragsz);
Alexander Duyckfd11a832014-12-09 19:40:49 -08002436struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2437 unsigned int length, gfp_t gfp_mask);
2438static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2439 unsigned int length)
2440{
2441 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2442}
Jesper Dangaard Brouer795bb1c2016-02-08 13:14:59 +01002443void napi_consume_skb(struct sk_buff *skb, int budget);
2444
2445void __kfree_skb_flush(void);
Jesper Dangaard Brouer15fad712016-02-08 13:15:04 +01002446void __kfree_skb_defer(struct sk_buff *skb);
Alexander Duyckffde7322014-12-09 19:40:42 -08002447
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002448/**
Alexander Duyck71dfda52014-11-11 09:26:34 -08002449 * __dev_alloc_pages - allocate page for network Rx
2450 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2451 * @order: size of the allocation
2452 *
2453 * Allocate a new page.
2454 *
2455 * %NULL is returned if there is no free memory.
2456*/
2457static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2458 unsigned int order)
2459{
2460 /* This piece of code contains several assumptions.
2461 * 1. This is for device Rx, therefor a cold page is preferred.
2462 * 2. The expectation is the user wants a compound page.
2463 * 3. If requesting a order 0 page it will not be compound
2464 * due to the check to see if order has a value in prep_new_page
2465 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2466 * code in gfp_to_alloc_flags that should be enforcing this.
2467 */
2468 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2469
2470 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2471}
2472
2473static inline struct page *dev_alloc_pages(unsigned int order)
2474{
Neil Horman95829b32016-05-19 11:30:54 -04002475 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
Alexander Duyck71dfda52014-11-11 09:26:34 -08002476}
2477
2478/**
2479 * __dev_alloc_page - allocate a page for network Rx
2480 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2481 *
2482 * Allocate a new page.
2483 *
2484 * %NULL is returned if there is no free memory.
2485 */
2486static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2487{
2488 return __dev_alloc_pages(gfp_mask, 0);
2489}
2490
2491static inline struct page *dev_alloc_page(void)
2492{
Neil Horman95829b32016-05-19 11:30:54 -04002493 return dev_alloc_pages(0);
Alexander Duyck71dfda52014-11-11 09:26:34 -08002494}
2495
2496/**
Mel Gorman06140022012-07-31 16:44:24 -07002497 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2498 * @page: The page that was allocated from skb_alloc_page
2499 * @skb: The skb that may need pfmemalloc set
2500 */
2501static inline void skb_propagate_pfmemalloc(struct page *page,
2502 struct sk_buff *skb)
2503{
Michal Hocko2f064f32015-08-21 14:11:51 -07002504 if (page_is_pfmemalloc(page))
Mel Gorman06140022012-07-31 16:44:24 -07002505 skb->pfmemalloc = true;
2506}
2507
Eric Dumazet564824b2010-10-11 19:05:25 +00002508/**
Masanari Iidae2278672014-02-18 22:54:36 +09002509 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002510 * @frag: the paged fragment
2511 *
2512 * Returns the &struct page associated with @frag.
2513 */
2514static inline struct page *skb_frag_page(const skb_frag_t *frag)
2515{
Ian Campbella8605c62011-10-19 23:01:49 +00002516 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002517}
2518
2519/**
2520 * __skb_frag_ref - take an addition reference on a paged fragment.
2521 * @frag: the paged fragment
2522 *
2523 * Takes an additional reference on the paged fragment @frag.
2524 */
2525static inline void __skb_frag_ref(skb_frag_t *frag)
2526{
2527 get_page(skb_frag_page(frag));
2528}
2529
2530/**
2531 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2532 * @skb: the buffer
2533 * @f: the fragment offset.
2534 *
2535 * Takes an additional reference on the @f'th paged fragment of @skb.
2536 */
2537static inline void skb_frag_ref(struct sk_buff *skb, int f)
2538{
2539 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2540}
2541
2542/**
2543 * __skb_frag_unref - release a reference on a paged fragment.
2544 * @frag: the paged fragment
2545 *
2546 * Releases a reference on the paged fragment @frag.
2547 */
2548static inline void __skb_frag_unref(skb_frag_t *frag)
2549{
2550 put_page(skb_frag_page(frag));
2551}
2552
2553/**
2554 * skb_frag_unref - release a reference on a paged fragment of an skb.
2555 * @skb: the buffer
2556 * @f: the fragment offset
2557 *
2558 * Releases a reference on the @f'th paged fragment of @skb.
2559 */
2560static inline void skb_frag_unref(struct sk_buff *skb, int f)
2561{
2562 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2563}
2564
2565/**
2566 * skb_frag_address - gets the address of the data contained in a paged fragment
2567 * @frag: the paged fragment buffer
2568 *
2569 * Returns the address of the data within @frag. The page must already
2570 * be mapped.
2571 */
2572static inline void *skb_frag_address(const skb_frag_t *frag)
2573{
2574 return page_address(skb_frag_page(frag)) + frag->page_offset;
2575}
2576
2577/**
2578 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2579 * @frag: the paged fragment buffer
2580 *
2581 * Returns the address of the data within @frag. Checks that the page
2582 * is mapped and returns %NULL otherwise.
2583 */
2584static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2585{
2586 void *ptr = page_address(skb_frag_page(frag));
2587 if (unlikely(!ptr))
2588 return NULL;
2589
2590 return ptr + frag->page_offset;
2591}
2592
2593/**
2594 * __skb_frag_set_page - sets the page contained in a paged fragment
2595 * @frag: the paged fragment
2596 * @page: the page to set
2597 *
2598 * Sets the fragment @frag to contain @page.
2599 */
2600static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2601{
Ian Campbella8605c62011-10-19 23:01:49 +00002602 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002603}
2604
2605/**
2606 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2607 * @skb: the buffer
2608 * @f: the fragment offset
2609 * @page: the page to set
2610 *
2611 * Sets the @f'th fragment of @skb to contain @page.
2612 */
2613static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2614 struct page *page)
2615{
2616 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2617}
2618
Eric Dumazet400dfd32013-10-17 16:27:07 -07002619bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2620
Ian Campbell131ea662011-08-19 06:25:00 +00002621/**
2622 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002623 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002624 * @frag: the paged fragment to map
2625 * @offset: the offset within the fragment (starting at the
2626 * fragment's own offset)
2627 * @size: the number of bytes to map
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002628 * @dir: the direction of the mapping (%PCI_DMA_*)
Ian Campbell131ea662011-08-19 06:25:00 +00002629 *
2630 * Maps the page associated with @frag to @device.
2631 */
2632static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2633 const skb_frag_t *frag,
2634 size_t offset, size_t size,
2635 enum dma_data_direction dir)
2636{
2637 return dma_map_page(dev, skb_frag_page(frag),
2638 frag->page_offset + offset, size, dir);
2639}
2640
Eric Dumazet117632e2011-12-03 21:39:53 +00002641static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2642 gfp_t gfp_mask)
2643{
2644 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2645}
2646
Octavian Purdilabad93e92014-06-12 01:36:26 +03002647
2648static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2649 gfp_t gfp_mask)
2650{
2651 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2652}
2653
2654
Ian Campbell131ea662011-08-19 06:25:00 +00002655/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002656 * skb_clone_writable - is the header of a clone writable
2657 * @skb: buffer to check
2658 * @len: length up to which to write
2659 *
2660 * Returns true if modifying the header part of the cloned buffer
2661 * does not requires the data to be copied.
2662 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002663static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002664{
2665 return !skb_header_cloned(skb) &&
2666 skb_headroom(skb) + len <= skb->hdr_len;
2667}
2668
Daniel Borkmann36976492016-02-19 23:05:25 +01002669static inline int skb_try_make_writable(struct sk_buff *skb,
2670 unsigned int write_len)
2671{
2672 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2673 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2674}
2675
Herbert Xud9cc2042007-09-16 16:21:16 -07002676static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2677 int cloned)
2678{
2679 int delta = 0;
2680
Herbert Xud9cc2042007-09-16 16:21:16 -07002681 if (headroom > skb_headroom(skb))
2682 delta = headroom - skb_headroom(skb);
2683
2684 if (delta || cloned)
2685 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2686 GFP_ATOMIC);
2687 return 0;
2688}
2689
Patrick McHardy334a8132007-06-25 04:35:20 -07002690/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691 * skb_cow - copy header of skb when it is required
2692 * @skb: buffer to cow
2693 * @headroom: needed headroom
2694 *
2695 * If the skb passed lacks sufficient headroom or its data part
2696 * is shared, data is reallocated. If reallocation fails, an error
2697 * is returned and original skb is not changed.
2698 *
2699 * The result is skb with writable area skb->head...skb->tail
2700 * and at least @headroom of space at head.
2701 */
2702static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2703{
Herbert Xud9cc2042007-09-16 16:21:16 -07002704 return __skb_cow(skb, headroom, skb_cloned(skb));
2705}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002706
Herbert Xud9cc2042007-09-16 16:21:16 -07002707/**
2708 * skb_cow_head - skb_cow but only making the head writable
2709 * @skb: buffer to cow
2710 * @headroom: needed headroom
2711 *
2712 * This function is identical to skb_cow except that we replace the
2713 * skb_cloned check by skb_header_cloned. It should be used when
2714 * you only need to push on some header and do not need to modify
2715 * the data.
2716 */
2717static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2718{
2719 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720}
2721
2722/**
2723 * skb_padto - pad an skbuff up to a minimal size
2724 * @skb: buffer to pad
2725 * @len: minimal length
2726 *
2727 * Pads up a buffer to ensure the trailing bytes exist and are
2728 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002729 * is untouched. Otherwise it is extended. Returns zero on
2730 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002731 */
Herbert Xu5b057c62006-06-23 02:06:41 -07002732static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733{
2734 unsigned int size = skb->len;
2735 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002736 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002737 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002738}
2739
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002740/**
2741 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2742 * @skb: buffer to pad
2743 * @len: minimal length
2744 *
2745 * Pads up a buffer to ensure the trailing bytes exist and are
2746 * blanked. If the buffer already contains sufficient data it
2747 * is untouched. Otherwise it is extended. Returns zero on
2748 * success. The skb is freed on error.
2749 */
2750static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2751{
2752 unsigned int size = skb->len;
2753
2754 if (unlikely(size < len)) {
2755 len -= size;
2756 if (skb_pad(skb, len))
2757 return -ENOMEM;
2758 __skb_put(skb, len);
2759 }
2760 return 0;
2761}
2762
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763static inline int skb_add_data(struct sk_buff *skb,
Al Viroaf2b0402014-11-27 21:44:24 -05002764 struct iov_iter *from, int copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765{
2766 const int off = skb->len;
2767
2768 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viroaf2b0402014-11-27 21:44:24 -05002769 __wsum csum = 0;
2770 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2771 &csum, from) == copy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002772 skb->csum = csum_block_add(skb->csum, csum, off);
2773 return 0;
2774 }
Al Viroaf2b0402014-11-27 21:44:24 -05002775 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776 return 0;
2777
2778 __skb_trim(skb, off);
2779 return -EFAULT;
2780}
2781
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002782static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2783 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002784{
2785 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002786 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787
Ian Campbellea2ab692011-08-22 23:44:58 +00002788 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00002789 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002790 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002791 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002792}
2793
Herbert Xu364c6ba2006-06-09 16:10:40 -07002794static inline int __skb_linearize(struct sk_buff *skb)
2795{
2796 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2797}
2798
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799/**
2800 * skb_linearize - convert paged skb to linear one
2801 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802 *
2803 * If there is no free memory -ENOMEM is returned, otherwise zero
2804 * is returned and the old skb data released.
2805 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07002806static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002807{
Herbert Xu364c6ba2006-06-09 16:10:40 -07002808 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2809}
2810
2811/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00002812 * skb_has_shared_frag - can any frag be overwritten
2813 * @skb: buffer to test
2814 *
2815 * Return true if the skb has at least one frag that might be modified
2816 * by an external entity (as in vmsplice()/sendfile())
2817 */
2818static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2819{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00002820 return skb_is_nonlinear(skb) &&
2821 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002822}
2823
2824/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07002825 * skb_linearize_cow - make sure skb is linear and writable
2826 * @skb: buffer to process
2827 *
2828 * If there is no free memory -ENOMEM is returned, otherwise zero
2829 * is returned and the old skb data released.
2830 */
2831static inline int skb_linearize_cow(struct sk_buff *skb)
2832{
2833 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2834 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835}
2836
2837/**
2838 * skb_postpull_rcsum - update checksum for received skb after pull
2839 * @skb: buffer to update
2840 * @start: start of data before pull
2841 * @len: length of data pulled
2842 *
2843 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07002844 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2845 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002846 */
2847
2848static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f2006-03-20 22:43:56 -08002849 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002850{
Patrick McHardy84fa7932006-08-29 16:44:56 -07002851 if (skb->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002852 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002853 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
Pravin B Shelar31b33df2015-09-28 17:24:25 -07002854 skb_checksum_start_offset(skb) < 0)
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002855 skb->ip_summed = CHECKSUM_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002856}
2857
Herbert Xucbb042f2006-03-20 22:43:56 -08002858unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2859
Daniel Borkmannf8ffad692016-01-07 15:50:23 +01002860static inline void skb_postpush_rcsum(struct sk_buff *skb,
2861 const void *start, unsigned int len)
2862{
2863 /* For performing the reverse operation to skb_postpull_rcsum(),
2864 * we can instead of ...
2865 *
2866 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0));
2867 *
2868 * ... just use this equivalent version here to save a few
2869 * instructions. Feeding csum of 0 in csum_partial() and later
2870 * on adding skb->csum is equivalent to feed skb->csum in the
2871 * first place.
2872 */
2873 if (skb->ip_summed == CHECKSUM_COMPLETE)
2874 skb->csum = csum_partial(start, len, skb->csum);
2875}
2876
David S. Miller7ce5a272013-12-02 17:26:05 -05002877/**
2878 * pskb_trim_rcsum - trim received skb and update checksum
2879 * @skb: buffer to trim
2880 * @len: new length
2881 *
2882 * This is exactly the same as pskb_trim except that it ensures the
2883 * checksum of received packets are still valid after the operation.
2884 */
2885
2886static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2887{
2888 if (likely(len >= skb->len))
2889 return 0;
2890 if (skb->ip_summed == CHECKSUM_COMPLETE)
2891 skb->ip_summed = CHECKSUM_NONE;
2892 return __pskb_trim(skb, len);
2893}
2894
Linus Torvalds1da177e2005-04-16 15:20:36 -07002895#define skb_queue_walk(queue, skb) \
2896 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002897 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002898 skb = skb->next)
2899
James Chapman46f89142007-04-30 00:07:31 -07002900#define skb_queue_walk_safe(queue, skb, tmp) \
2901 for (skb = (queue)->next, tmp = skb->next; \
2902 skb != (struct sk_buff *)(queue); \
2903 skb = tmp, tmp = skb->next)
2904
David S. Miller1164f522008-09-23 00:49:44 -07002905#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002906 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07002907 skb = skb->next)
2908
2909#define skb_queue_walk_from_safe(queue, skb, tmp) \
2910 for (tmp = skb->next; \
2911 skb != (struct sk_buff *)(queue); \
2912 skb = tmp, tmp = skb->next)
2913
Stephen Hemminger300ce172005-10-30 13:47:34 -08002914#define skb_queue_reverse_walk(queue, skb) \
2915 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002916 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08002917 skb = skb->prev)
2918
David S. Miller686a2952011-01-20 22:47:32 -08002919#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2920 for (skb = (queue)->prev, tmp = skb->prev; \
2921 skb != (struct sk_buff *)(queue); \
2922 skb = tmp, tmp = skb->prev)
2923
2924#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2925 for (tmp = skb->prev; \
2926 skb != (struct sk_buff *)(queue); \
2927 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002928
David S. Miller21dc3302010-08-23 00:13:46 -07002929static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07002930{
2931 return skb_shinfo(skb)->frag_list != NULL;
2932}
2933
2934static inline void skb_frag_list_init(struct sk_buff *skb)
2935{
2936 skb_shinfo(skb)->frag_list = NULL;
2937}
2938
David S. Milleree039872009-06-09 00:17:13 -07002939#define skb_walk_frags(skb, iter) \
2940 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2941
Rainer Weikusatea3793e2015-12-06 21:11:34 +00002942
2943int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
2944 const struct sk_buff *skb);
2945struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
2946 int *peeked, int *off, int *err,
2947 struct sk_buff **last);
Joe Perches7965bd42013-09-26 14:48:15 -07002948struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2949 int *peeked, int *off, int *err);
2950struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2951 int *err);
2952unsigned int datagram_poll(struct file *file, struct socket *sock,
2953 struct poll_table_struct *wait);
Al Viroc0371da2014-11-24 10:42:55 -05002954int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2955 struct iov_iter *to, int size);
David S. Miller51f3d022014-11-05 16:46:40 -05002956static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2957 struct msghdr *msg, int size)
2958{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002959 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
David S. Miller51f3d022014-11-05 16:46:40 -05002960}
Al Viroe5a4b0b2014-11-24 18:17:55 -05002961int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2962 struct msghdr *msg);
Al Viro3a654f92014-06-19 14:15:22 -04002963int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2964 struct iov_iter *from, int len);
Al Viro3a654f92014-06-19 14:15:22 -04002965int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
Joe Perches7965bd42013-09-26 14:48:15 -07002966void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
samanthakumar627d2d62016-04-05 12:41:16 -04002967void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
2968static inline void skb_free_datagram_locked(struct sock *sk,
2969 struct sk_buff *skb)
2970{
2971 __skb_free_datagram_locked(sk, skb, 0);
2972}
Joe Perches7965bd42013-09-26 14:48:15 -07002973int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07002974int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2975int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2976__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2977 int len, __wsum csum);
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002978ssize_t skb_socket_splice(struct sock *sk,
2979 struct pipe_inode_info *pipe,
2980 struct splice_pipe_desc *spd);
2981int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
Joe Perches7965bd42013-09-26 14:48:15 -07002982 struct pipe_inode_info *pipe, unsigned int len,
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002983 unsigned int flags,
2984 ssize_t (*splice_cb)(struct sock *,
2985 struct pipe_inode_info *,
2986 struct splice_pipe_desc *));
Joe Perches7965bd42013-09-26 14:48:15 -07002987void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01002988unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002989int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2990 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07002991void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2992int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2993void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01002994unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Marcelo Ricardo Leitnerae7ef812016-06-02 15:05:41 -03002995bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
Joe Perches7965bd42013-09-26 14:48:15 -07002996struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Vlad Yasevich0d5501c2014-08-08 14:42:13 -04002997struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
Jiri Pirkoe2195122014-11-19 14:05:01 +01002998int skb_ensure_writable(struct sk_buff *skb, int write_len);
Jiri Pirko93515d52014-11-19 14:05:02 +01002999int skb_vlan_pop(struct sk_buff *skb);
3000int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
Sowmini Varadhan6fa01cc2016-04-22 18:36:35 -07003001struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3002 gfp_t gfp);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03003003
Al Viro6ce8e9c2014-04-06 21:25:44 -04003004static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3005{
Al Viro21226ab2014-11-28 15:48:29 -05003006 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro6ce8e9c2014-04-06 21:25:44 -04003007}
3008
Al Viro7eab8d92014-04-06 21:51:23 -04003009static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3010{
Al Viroe5a4b0b2014-11-24 18:17:55 -05003011 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro7eab8d92014-04-06 21:51:23 -04003012}
3013
Daniel Borkmann2817a332013-10-30 11:50:51 +01003014struct skb_checksum_ops {
3015 __wsum (*update)(const void *mem, int len, __wsum wsum);
3016 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3017};
3018
3019__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3020 __wsum csum, const struct skb_checksum_ops *ops);
3021__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3022 __wsum csum);
3023
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003024static inline void * __must_check
3025__skb_header_pointer(const struct sk_buff *skb, int offset,
3026 int len, void *data, int hlen, void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027{
Patrick McHardy55820ee2005-07-05 14:08:10 -07003028 if (hlen - offset >= len)
David S. Miller690e36e2014-08-23 12:13:41 -07003029 return data + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003030
David S. Miller690e36e2014-08-23 12:13:41 -07003031 if (!skb ||
3032 skb_copy_bits(skb, offset, buffer, len) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003033 return NULL;
3034
3035 return buffer;
3036}
3037
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003038static inline void * __must_check
3039skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
David S. Miller690e36e2014-08-23 12:13:41 -07003040{
3041 return __skb_header_pointer(skb, offset, len, skb->data,
3042 skb_headlen(skb), buffer);
3043}
3044
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01003045/**
3046 * skb_needs_linearize - check if we need to linearize a given skb
3047 * depending on the given device features.
3048 * @skb: socket buffer to check
3049 * @features: net device features
3050 *
3051 * Returns true if either:
3052 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3053 * 2. skb is fragmented and the device does not support SG.
3054 */
3055static inline bool skb_needs_linearize(struct sk_buff *skb,
3056 netdev_features_t features)
3057{
3058 return skb_is_nonlinear(skb) &&
3059 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3060 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3061}
3062
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03003063static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3064 void *to,
3065 const unsigned int len)
3066{
3067 memcpy(to, skb->data, len);
3068}
3069
3070static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3071 const int offset, void *to,
3072 const unsigned int len)
3073{
3074 memcpy(to, skb->data + offset, len);
3075}
3076
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03003077static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3078 const void *from,
3079 const unsigned int len)
3080{
3081 memcpy(skb->data, from, len);
3082}
3083
3084static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3085 const int offset,
3086 const void *from,
3087 const unsigned int len)
3088{
3089 memcpy(skb->data + offset, from, len);
3090}
3091
Joe Perches7965bd42013-09-26 14:48:15 -07003092void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003093
Patrick Ohlyac45f602009-02-12 05:03:37 +00003094static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3095{
3096 return skb->tstamp;
3097}
3098
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003099/**
3100 * skb_get_timestamp - get timestamp from a skb
3101 * @skb: skb to get stamp from
3102 * @stamp: pointer to struct timeval to store stamp in
3103 *
3104 * Timestamps are stored in the skb as offsets to a base timestamp.
3105 * This function converts the offset back to a struct timeval and stores
3106 * it in stamp.
3107 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00003108static inline void skb_get_timestamp(const struct sk_buff *skb,
3109 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003110{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003111 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003112}
3113
Patrick Ohlyac45f602009-02-12 05:03:37 +00003114static inline void skb_get_timestampns(const struct sk_buff *skb,
3115 struct timespec *stamp)
3116{
3117 *stamp = ktime_to_timespec(skb->tstamp);
3118}
3119
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003120static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003121{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003122 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003123}
3124
Stephen Hemminger164891a2007-04-23 22:26:16 -07003125static inline ktime_t net_timedelta(ktime_t t)
3126{
3127 return ktime_sub(ktime_get_real(), t);
3128}
3129
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07003130static inline ktime_t net_invalid_timestamp(void)
3131{
3132 return ktime_set(0, 0);
3133}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003134
Alexander Duyck62bccb82014-09-04 13:31:35 -04003135struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3136
Richard Cochranc1f19b52010-07-17 08:49:36 +00003137#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3138
Joe Perches7965bd42013-09-26 14:48:15 -07003139void skb_clone_tx_timestamp(struct sk_buff *skb);
3140bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00003141
3142#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3143
3144static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3145{
3146}
3147
3148static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3149{
3150 return false;
3151}
3152
3153#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3154
3155/**
3156 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3157 *
Richard Cochranda92b192011-10-21 00:49:15 +00003158 * PHY drivers may accept clones of transmitted packets for
3159 * timestamping via their phy_driver.txtstamp method. These drivers
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003160 * must call this function to return the skb back to the stack with a
3161 * timestamp.
Richard Cochranda92b192011-10-21 00:49:15 +00003162 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00003163 * @skb: clone of the the original outgoing packet
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003164 * @hwtstamps: hardware time stamps
Richard Cochranc1f19b52010-07-17 08:49:36 +00003165 *
3166 */
3167void skb_complete_tx_timestamp(struct sk_buff *skb,
3168 struct skb_shared_hwtstamps *hwtstamps);
3169
Willem de Bruijne7fd2882014-08-04 22:11:48 -04003170void __skb_tstamp_tx(struct sk_buff *orig_skb,
3171 struct skb_shared_hwtstamps *hwtstamps,
3172 struct sock *sk, int tstype);
3173
Patrick Ohlyac45f602009-02-12 05:03:37 +00003174/**
3175 * skb_tstamp_tx - queue clone of skb with send time stamps
3176 * @orig_skb: the original outgoing packet
3177 * @hwtstamps: hardware time stamps, may be NULL if not available
3178 *
3179 * If the skb has a socket associated, then this function clones the
3180 * skb (thus sharing the actual data and optional structures), stores
3181 * the optional hardware time stamping information (if non NULL) or
3182 * generates a software time stamp (otherwise), then queues the clone
3183 * to the error queue of the socket. Errors are silently ignored.
3184 */
Joe Perches7965bd42013-09-26 14:48:15 -07003185void skb_tstamp_tx(struct sk_buff *orig_skb,
3186 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00003187
Richard Cochran4507a712010-07-17 08:48:28 +00003188static inline void sw_tx_timestamp(struct sk_buff *skb)
3189{
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003190 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3191 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
Richard Cochran4507a712010-07-17 08:48:28 +00003192 skb_tstamp_tx(skb, NULL);
3193}
3194
3195/**
3196 * skb_tx_timestamp() - Driver hook for transmit timestamping
3197 *
3198 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00003199 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00003200 *
David S. Miller73409f32013-12-27 13:04:33 -05003201 * Specifically, one should make absolutely sure that this function is
3202 * called before TX completion of this packet can trigger. Otherwise
3203 * the packet could potentially already be freed.
3204 *
Richard Cochran4507a712010-07-17 08:48:28 +00003205 * @skb: A socket buffer.
3206 */
3207static inline void skb_tx_timestamp(struct sk_buff *skb)
3208{
Richard Cochranc1f19b52010-07-17 08:49:36 +00003209 skb_clone_tx_timestamp(skb);
Richard Cochran4507a712010-07-17 08:48:28 +00003210 sw_tx_timestamp(skb);
3211}
3212
Johannes Berg6e3e9392011-11-09 10:15:42 +01003213/**
3214 * skb_complete_wifi_ack - deliver skb with wifi status
3215 *
3216 * @skb: the original outgoing packet
3217 * @acked: ack status
3218 *
3219 */
3220void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3221
Joe Perches7965bd42013-09-26 14:48:15 -07003222__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3223__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003224
Herbert Xu60476372007-04-09 11:59:39 -07003225static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3226{
Tom Herbert6edec0e2015-02-10 16:30:28 -08003227 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3228 skb->csum_valid ||
3229 (skb->ip_summed == CHECKSUM_PARTIAL &&
3230 skb_checksum_start_offset(skb) >= 0));
Herbert Xu60476372007-04-09 11:59:39 -07003231}
3232
Herbert Xufb286bb2005-11-10 13:01:24 -08003233/**
3234 * skb_checksum_complete - Calculate checksum of an entire packet
3235 * @skb: packet to process
3236 *
3237 * This function calculates the checksum over the entire packet plus
3238 * the value of skb->csum. The latter can be used to supply the
3239 * checksum of a pseudo header as used by TCP/UDP. It returns the
3240 * checksum.
3241 *
3242 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3243 * this function can be used to verify that checksum on received
3244 * packets. In that case the function should return zero if the
3245 * checksum is correct. In particular, this function will return zero
3246 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3247 * hardware has already verified the correctness of the checksum.
3248 */
Al Viro4381ca32007-07-15 21:00:11 +01003249static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08003250{
Herbert Xu60476372007-04-09 11:59:39 -07003251 return skb_csum_unnecessary(skb) ?
3252 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003253}
3254
Tom Herbert77cffe22014-08-27 21:26:46 -07003255static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3256{
3257 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3258 if (skb->csum_level == 0)
3259 skb->ip_summed = CHECKSUM_NONE;
3260 else
3261 skb->csum_level--;
3262 }
3263}
3264
3265static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3266{
3267 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3268 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3269 skb->csum_level++;
3270 } else if (skb->ip_summed == CHECKSUM_NONE) {
3271 skb->ip_summed = CHECKSUM_UNNECESSARY;
3272 skb->csum_level = 0;
3273 }
3274}
3275
Tom Herbert5a212322014-08-31 15:12:41 -07003276static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3277{
3278 /* Mark current checksum as bad (typically called from GRO
3279 * path). In the case that ip_summed is CHECKSUM_NONE
3280 * this must be the first checksum encountered in the packet.
3281 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3282 * checksum after the last one validated. For UDP, a zero
3283 * checksum can not be marked as bad.
3284 */
3285
3286 if (skb->ip_summed == CHECKSUM_NONE ||
3287 skb->ip_summed == CHECKSUM_UNNECESSARY)
3288 skb->csum_bad = 1;
3289}
3290
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003291/* Check if we need to perform checksum complete validation.
3292 *
3293 * Returns true if checksum complete is needed, false otherwise
3294 * (either checksum is unnecessary or zero checksum is allowed).
3295 */
3296static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3297 bool zero_okay,
3298 __sum16 check)
3299{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003300 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3301 skb->csum_valid = 1;
Tom Herbert77cffe22014-08-27 21:26:46 -07003302 __skb_decr_checksum_unnecessary(skb);
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003303 return false;
3304 }
3305
3306 return true;
3307}
3308
3309/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3310 * in checksum_init.
3311 */
3312#define CHECKSUM_BREAK 76
3313
Tom Herbert4e18b9a2015-04-20 14:10:04 -07003314/* Unset checksum-complete
3315 *
3316 * Unset checksum complete can be done when packet is being modified
3317 * (uncompressed for instance) and checksum-complete value is
3318 * invalidated.
3319 */
3320static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3321{
3322 if (skb->ip_summed == CHECKSUM_COMPLETE)
3323 skb->ip_summed = CHECKSUM_NONE;
3324}
3325
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003326/* Validate (init) checksum based on checksum complete.
3327 *
3328 * Return values:
3329 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3330 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3331 * checksum is stored in skb->csum for use in __skb_checksum_complete
3332 * non-zero: value of invalid checksum
3333 *
3334 */
3335static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3336 bool complete,
3337 __wsum psum)
3338{
3339 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3340 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003341 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003342 return 0;
3343 }
Tom Herbert5a212322014-08-31 15:12:41 -07003344 } else if (skb->csum_bad) {
3345 /* ip_summed == CHECKSUM_NONE in this case */
Eric Dumazetc91d4602015-05-15 05:48:07 -07003346 return (__force __sum16)1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003347 }
3348
3349 skb->csum = psum;
3350
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003351 if (complete || skb->len <= CHECKSUM_BREAK) {
3352 __sum16 csum;
3353
3354 csum = __skb_checksum_complete(skb);
3355 skb->csum_valid = !csum;
3356 return csum;
3357 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003358
3359 return 0;
3360}
3361
3362static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3363{
3364 return 0;
3365}
3366
3367/* Perform checksum validate (init). Note that this is a macro since we only
3368 * want to calculate the pseudo header which is an input function if necessary.
3369 * First we try to validate without any computation (checksum unnecessary) and
3370 * then calculate based on checksum complete calling the function to compute
3371 * pseudo header.
3372 *
3373 * Return values:
3374 * 0: checksum is validated or try to in skb_checksum_complete
3375 * non-zero: value of invalid checksum
3376 */
3377#define __skb_checksum_validate(skb, proto, complete, \
3378 zero_okay, check, compute_pseudo) \
3379({ \
3380 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003381 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003382 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3383 __ret = __skb_checksum_validate_complete(skb, \
3384 complete, compute_pseudo(skb, proto)); \
3385 __ret; \
3386})
3387
3388#define skb_checksum_init(skb, proto, compute_pseudo) \
3389 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3390
3391#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3392 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3393
3394#define skb_checksum_validate(skb, proto, compute_pseudo) \
3395 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3396
3397#define skb_checksum_validate_zero_check(skb, proto, check, \
3398 compute_pseudo) \
Sabrina Dubroca096a4cf2015-02-06 18:54:19 +01003399 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003400
3401#define skb_checksum_simple_validate(skb) \
3402 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3403
Tom Herbertd96535a2014-08-31 15:12:42 -07003404static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3405{
3406 return (skb->ip_summed == CHECKSUM_NONE &&
3407 skb->csum_valid && !skb->csum_bad);
3408}
3409
3410static inline void __skb_checksum_convert(struct sk_buff *skb,
3411 __sum16 check, __wsum pseudo)
3412{
3413 skb->csum = ~pseudo;
3414 skb->ip_summed = CHECKSUM_COMPLETE;
3415}
3416
3417#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3418do { \
3419 if (__skb_checksum_convert_check(skb)) \
3420 __skb_checksum_convert(skb, check, \
3421 compute_pseudo(skb, proto)); \
3422} while (0)
3423
Tom Herbert15e23962015-02-10 16:30:31 -08003424static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3425 u16 start, u16 offset)
3426{
3427 skb->ip_summed = CHECKSUM_PARTIAL;
3428 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3429 skb->csum_offset = offset - start;
3430}
3431
Tom Herbertdcdc8992015-02-02 16:07:34 -08003432/* Update skbuf and packet to reflect the remote checksum offload operation.
3433 * When called, ptr indicates the starting point for skb->csum when
3434 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3435 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3436 */
3437static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
Tom Herbert15e23962015-02-10 16:30:31 -08003438 int start, int offset, bool nopartial)
Tom Herbertdcdc8992015-02-02 16:07:34 -08003439{
3440 __wsum delta;
3441
Tom Herbert15e23962015-02-10 16:30:31 -08003442 if (!nopartial) {
3443 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3444 return;
3445 }
3446
Tom Herbertdcdc8992015-02-02 16:07:34 -08003447 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3448 __skb_checksum_complete(skb);
3449 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3450 }
3451
3452 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3453
3454 /* Adjust skb->csum since we changed the packet */
3455 skb->csum = csum_add(skb->csum, delta);
3456}
3457
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003458#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07003459void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3461{
3462 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07003463 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003464}
3465static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3466{
3467 if (nfct)
3468 atomic_inc(&nfct->use);
3469}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003470#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003471#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003472static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3473{
3474 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3475 kfree(nf_bridge);
3476}
3477static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3478{
3479 if (nf_bridge)
3480 atomic_inc(&nf_bridge->use);
3481}
3482#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08003483static inline void nf_reset(struct sk_buff *skb)
3484{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003485#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003486 nf_conntrack_put(skb->nfct);
3487 skb->nfct = NULL;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003488#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003489#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003490 nf_bridge_put(skb->nf_bridge);
3491 skb->nf_bridge = NULL;
3492#endif
3493}
3494
Patrick McHardy124dff02013-04-05 20:42:05 +02003495static inline void nf_reset_trace(struct sk_buff *skb)
3496{
Florian Westphal478b3602014-02-15 23:48:45 +01003497#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00003498 skb->nf_trace = 0;
3499#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003500}
3501
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003502/* Note: This doesn't put any conntrack and bridge info in dst. */
Eric Dumazetb1937222014-09-28 22:18:47 -07003503static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3504 bool copy)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003505{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003506#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003507 dst->nfct = src->nfct;
3508 nf_conntrack_get(src->nfct);
Eric Dumazetb1937222014-09-28 22:18:47 -07003509 if (copy)
3510 dst->nfctinfo = src->nfctinfo;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003511#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003512#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003513 dst->nf_bridge = src->nf_bridge;
3514 nf_bridge_get(src->nf_bridge);
3515#endif
Florian Westphal478b3602014-02-15 23:48:45 +01003516#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Eric Dumazetb1937222014-09-28 22:18:47 -07003517 if (copy)
3518 dst->nf_trace = src->nf_trace;
Florian Westphal478b3602014-02-15 23:48:45 +01003519#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003520}
3521
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003522static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3523{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003524#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003525 nf_conntrack_put(dst->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003526#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003527#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003528 nf_bridge_put(dst->nf_bridge);
3529#endif
Eric Dumazetb1937222014-09-28 22:18:47 -07003530 __nf_copy(dst, src, true);
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003531}
3532
James Morris984bc162006-06-09 00:29:17 -07003533#ifdef CONFIG_NETWORK_SECMARK
3534static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3535{
3536 to->secmark = from->secmark;
3537}
3538
3539static inline void skb_init_secmark(struct sk_buff *skb)
3540{
3541 skb->secmark = 0;
3542}
3543#else
3544static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3545{ }
3546
3547static inline void skb_init_secmark(struct sk_buff *skb)
3548{ }
3549#endif
3550
Eric W. Biederman574f7192014-04-01 12:20:24 -07003551static inline bool skb_irq_freeable(const struct sk_buff *skb)
3552{
3553 return !skb->destructor &&
3554#if IS_ENABLED(CONFIG_XFRM)
3555 !skb->sp &&
3556#endif
3557#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3558 !skb->nfct &&
3559#endif
3560 !skb->_skb_refdst &&
3561 !skb_has_frag_list(skb);
3562}
3563
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003564static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3565{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003566 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003567}
3568
Stephen Hemminger92477442009-03-21 13:39:26 -07003569static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003570{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003571 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003572}
3573
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003574static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3575{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003576 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003577}
3578
David S. Millerd5a9e242009-01-27 16:22:11 -08003579static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3580{
3581 skb->queue_mapping = rx_queue + 1;
3582}
3583
Stephen Hemminger92477442009-03-21 13:39:26 -07003584static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003585{
3586 return skb->queue_mapping - 1;
3587}
3588
Stephen Hemminger92477442009-03-21 13:39:26 -07003589static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003590{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003591 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003592}
3593
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003594static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3595{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003596#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003597 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003598#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003599 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003600#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003601}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003602
Pravin B Shelar68c33162013-02-14 14:02:41 +00003603/* Keeps track of mac header offset relative to skb->head.
3604 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3605 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003606 * tunnel skb it points to outer mac header.
3607 * Keeps track of level of encapsulation of network headers.
3608 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003609struct skb_gso_cb {
Alexander Duyck802ab552016-04-10 21:45:03 -04003610 union {
3611 int mac_offset;
3612 int data_offset;
3613 };
Eric Dumazet3347c962013-10-19 11:42:56 -07003614 int encap_level;
Alexander Duyck76443452016-02-05 15:27:37 -08003615 __wsum csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003616 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003617};
Konstantin Khlebnikov9207f9d2016-01-08 15:21:46 +03003618#define SKB_SGO_CB_OFFSET 32
3619#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
Pravin B Shelar68c33162013-02-14 14:02:41 +00003620
3621static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3622{
3623 return (skb_mac_header(inner_skb) - inner_skb->head) -
3624 SKB_GSO_CB(inner_skb)->mac_offset;
3625}
3626
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003627static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3628{
3629 int new_headroom, headroom;
3630 int ret;
3631
3632 headroom = skb_headroom(skb);
3633 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3634 if (ret)
3635 return ret;
3636
3637 new_headroom = skb_headroom(skb);
3638 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3639 return 0;
3640}
3641
Alexander Duyck08b64fc2016-02-05 15:27:49 -08003642static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3643{
3644 /* Do not update partial checksums if remote checksum is enabled. */
3645 if (skb->remcsum_offload)
3646 return;
3647
3648 SKB_GSO_CB(skb)->csum = res;
3649 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3650}
3651
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003652/* Compute the checksum for a gso segment. First compute the checksum value
3653 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3654 * then add in skb->csum (checksum from csum_start to end of packet).
3655 * skb->csum and csum_start are then updated to reflect the checksum of the
3656 * resultant packet starting from the transport header-- the resultant checksum
3657 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3658 * header.
3659 */
3660static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3661{
Alexander Duyck76443452016-02-05 15:27:37 -08003662 unsigned char *csum_start = skb_transport_header(skb);
3663 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3664 __wsum partial = SKB_GSO_CB(skb)->csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003665
Alexander Duyck76443452016-02-05 15:27:37 -08003666 SKB_GSO_CB(skb)->csum = res;
3667 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003668
Alexander Duyck76443452016-02-05 15:27:37 -08003669 return csum_fold(csum_partial(csum_start, plen, partial));
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003670}
3671
David S. Millerbdcc0922012-03-07 20:53:36 -05003672static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07003673{
3674 return skb_shinfo(skb)->gso_size;
3675}
3676
Eric Dumazet36a8f392013-09-29 01:21:32 -07003677/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05003678static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02003679{
3680 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3681}
3682
Joe Perches7965bd42013-09-26 14:48:15 -07003683void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07003684
3685static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3686{
3687 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3688 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003689 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3690
Alexander Duyckb78462e2010-06-02 12:24:37 +00003691 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3692 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07003693 __skb_warn_lro_forwarding(skb);
3694 return true;
3695 }
3696 return false;
3697}
3698
Herbert Xu35fc92a2007-03-26 23:22:20 -07003699static inline void skb_forward_csum(struct sk_buff *skb)
3700{
3701 /* Unfortunately we don't support this one. Any brave souls? */
3702 if (skb->ip_summed == CHECKSUM_COMPLETE)
3703 skb->ip_summed = CHECKSUM_NONE;
3704}
3705
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003706/**
3707 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3708 * @skb: skb to check
3709 *
3710 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3711 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3712 * use this helper, to document places where we make this assertion.
3713 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003714static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003715{
3716#ifdef DEBUG
3717 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3718#endif
3719}
3720
Rusty Russellf35d9d82008-02-04 23:49:54 -05003721bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00003722
Paul Durranted1f50c2014-01-09 10:02:46 +00003723int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
Linus Lüssing9afd85c2015-05-02 14:01:07 +02003724struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3725 unsigned int transport_len,
3726 __sum16(*skb_chkf)(struct sk_buff *skb));
Paul Durranted1f50c2014-01-09 10:02:46 +00003727
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00003728/**
3729 * skb_head_is_locked - Determine if the skb->head is locked down
3730 * @skb: skb to check
3731 *
3732 * The head on skbs build around a head frag can be removed if they are
3733 * not cloned. This function returns true if the skb head is locked down
3734 * due to either being allocated via kmalloc, or by being a clone with
3735 * multiple references to the head.
3736 */
3737static inline bool skb_head_is_locked(const struct sk_buff *skb)
3738{
3739 return !skb->head_frag || skb_cloned(skb);
3740}
Florian Westphalfe6cc552014-02-13 23:09:12 +01003741
3742/**
3743 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3744 *
3745 * @skb: GSO skb
3746 *
3747 * skb_gso_network_seglen is used to determine the real size of the
3748 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3749 *
3750 * The MAC/L2 header is not accounted for.
3751 */
3752static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3753{
3754 unsigned int hdr_len = skb_transport_header(skb) -
3755 skb_network_header(skb);
3756 return hdr_len + skb_gso_transport_seglen(skb);
3757}
Thomas Grafee122c72015-07-21 10:43:58 +02003758
Edward Cree179bc672016-02-11 20:48:04 +00003759/* Local Checksum Offload.
3760 * Compute outer checksum based on the assumption that the
3761 * inner checksum will be offloaded later.
Edward Creee8ae7b02016-02-11 21:03:37 +00003762 * See Documentation/networking/checksum-offloads.txt for
3763 * explanation of how this works.
Edward Cree179bc672016-02-11 20:48:04 +00003764 * Fill in outer checksum adjustment (e.g. with sum of outer
3765 * pseudo-header) before calling.
3766 * Also ensure that inner checksum is in linear data area.
3767 */
3768static inline __wsum lco_csum(struct sk_buff *skb)
3769{
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003770 unsigned char *csum_start = skb_checksum_start(skb);
3771 unsigned char *l4_hdr = skb_transport_header(skb);
3772 __wsum partial;
Edward Cree179bc672016-02-11 20:48:04 +00003773
3774 /* Start with complement of inner checksum adjustment */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003775 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3776 skb->csum_offset));
3777
Edward Cree179bc672016-02-11 20:48:04 +00003778 /* Add in checksum of our headers (incl. outer checksum
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003779 * adjustment filled in by caller) and return result.
Edward Cree179bc672016-02-11 20:48:04 +00003780 */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003781 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
Edward Cree179bc672016-02-11 20:48:04 +00003782}
3783
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784#endif /* __KERNEL__ */
3785#endif /* _LINUX_SKBUFF_H */