blob: a1ce63979ad8873fb8309196acbd2590e9d7cecf [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
304typedef struct skb_frag_struct skb_frag_t;
305
306struct skb_frag_struct {
Ian Campbella8605c62011-10-19 23:01:49 +0000307 struct {
308 struct page *p;
309 } page;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000310#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
David S. Millera309bb02007-07-30 18:47:03 -0700311 __u32 page_offset;
312 __u32 size;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000313#else
314 __u16 page_offset;
315 __u16 size;
316#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317};
318
Eric Dumazet9e903e02011-10-18 21:00:24 +0000319static inline unsigned int skb_frag_size(const skb_frag_t *frag)
320{
321 return frag->size;
322}
323
324static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
325{
326 frag->size = size;
327}
328
329static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
330{
331 frag->size += delta;
332}
333
334static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
335{
336 frag->size -= delta;
337}
338
Patrick Ohlyac45f602009-02-12 05:03:37 +0000339#define HAVE_HW_TIME_STAMP
340
341/**
Randy Dunlapd3a21be2009-03-02 03:15:58 -0800342 * struct skb_shared_hwtstamps - hardware time stamps
Patrick Ohlyac45f602009-02-12 05:03:37 +0000343 * @hwtstamp: hardware time stamp transformed into duration
344 * since arbitrary point in time
Patrick Ohlyac45f602009-02-12 05:03:37 +0000345 *
346 * Software time stamps generated by ktime_get_real() are stored in
Willem de Bruijn4d276eb2014-07-25 18:01:32 -0400347 * skb->tstamp.
Patrick Ohlyac45f602009-02-12 05:03:37 +0000348 *
349 * hwtstamps can only be compared against other hwtstamps from
350 * the same device.
351 *
352 * This structure is attached to packets as part of the
353 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
354 */
355struct skb_shared_hwtstamps {
356 ktime_t hwtstamp;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000357};
358
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000359/* Definitions for tx_flags in struct skb_shared_info */
360enum {
361 /* generate hardware time stamp */
362 SKBTX_HW_TSTAMP = 1 << 0,
363
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400364 /* generate software time stamp when queueing packet to NIC */
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000365 SKBTX_SW_TSTAMP = 1 << 1,
366
367 /* device driver is going to provide hardware time stamp */
368 SKBTX_IN_PROGRESS = 1 << 2,
369
Shirley Maa6686f22011-07-06 12:22:12 +0000370 /* device driver supports TX zero-copy buffers */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000371 SKBTX_DEV_ZEROCOPY = 1 << 3,
Johannes Berg6e3e9392011-11-09 10:15:42 +0100372
373 /* generate wifi status information (where possible) */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000374 SKBTX_WIFI_STATUS = 1 << 4,
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000375
376 /* This indicates at least one fragment might be overwritten
377 * (as in vmsplice(), sendfile() ...)
378 * If we need to compute a TX checksum, we'll need to copy
379 * all frags to avoid possible bad checksum
380 */
381 SKBTX_SHARED_FRAG = 1 << 5,
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400382
383 /* generate software time stamp when entering packet scheduling */
384 SKBTX_SCHED_TSTAMP = 1 << 6,
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400385
386 /* generate software timestamp on peer data acknowledgment */
387 SKBTX_ACK_TSTAMP = 1 << 7,
Shirley Maa6686f22011-07-06 12:22:12 +0000388};
389
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400390#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
391 SKBTX_SCHED_TSTAMP | \
392 SKBTX_ACK_TSTAMP)
Willem de Bruijnf24b9be2014-08-04 22:11:45 -0400393#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
394
Shirley Maa6686f22011-07-06 12:22:12 +0000395/*
396 * The callback notifies userspace to release buffers when skb DMA is done in
397 * lower device, the skb last reference should be 0 when calling this.
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000398 * The zerocopy_success argument is true if zero copy transmit occurred,
399 * false on data copy or out of memory error caused by data copy attempt.
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000400 * The ctx field is used to track device context.
401 * The desc field is used to track userspace buffer index.
Shirley Maa6686f22011-07-06 12:22:12 +0000402 */
403struct ubuf_info {
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000404 void (*callback)(struct ubuf_info *, bool zerocopy_success);
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000405 void *ctx;
Shirley Maa6686f22011-07-06 12:22:12 +0000406 unsigned long desc;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000407};
408
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409/* This data is invariant across clones and lives at
410 * the end of the header data, ie. at skb->end.
411 */
412struct skb_shared_info {
Ian Campbell9f42f122012-01-05 07:13:39 +0000413 unsigned char nr_frags;
414 __u8 tx_flags;
Herbert Xu79671682006-06-22 02:40:14 -0700415 unsigned short gso_size;
416 /* Warning: this field is not always filled in (UFO)! */
417 unsigned short gso_segs;
418 unsigned short gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 struct sk_buff *frag_list;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000420 struct skb_shared_hwtstamps hwtstamps;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400421 u32 tskey;
Ian Campbell9f42f122012-01-05 07:13:39 +0000422 __be32 ip6_frag_id;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700423
424 /*
425 * Warning : all fields before dataref are cleared in __alloc_skb()
426 */
427 atomic_t dataref;
428
Johann Baudy69e3c752009-05-18 22:11:22 -0700429 /* Intermediate layers must ensure that destructor_arg
430 * remains valid until skb destructor */
431 void * destructor_arg;
Shirley Maa6686f22011-07-06 12:22:12 +0000432
Eric Dumazetfed66382010-07-22 19:09:08 +0000433 /* must be last field, see pskb_expand_head() */
434 skb_frag_t frags[MAX_SKB_FRAGS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435};
436
437/* We divide dataref into two halves. The higher 16 bits hold references
438 * to the payload part of skb->data. The lower 16 bits hold references to
Patrick McHardy334a8132007-06-25 04:35:20 -0700439 * the entire skb->data. A clone of a headerless skb holds the length of
440 * the header in skb->hdr_len.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 *
442 * All users must obey the rule that the skb->data reference count must be
443 * greater than or equal to the payload reference count.
444 *
445 * Holding a reference to the payload part means that the user does not
446 * care about modifications to the header part of skb->data.
447 */
448#define SKB_DATAREF_SHIFT 16
449#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
450
David S. Millerd179cd12005-08-17 14:57:30 -0700451
452enum {
Vijay Subramanianc8753d52014-10-02 10:00:43 -0700453 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
454 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
455 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
David S. Millerd179cd12005-08-17 14:57:30 -0700456};
457
Herbert Xu79671682006-06-22 02:40:14 -0700458enum {
459 SKB_GSO_TCPV4 = 1 << 0,
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700460 SKB_GSO_UDP = 1 << 1,
Herbert Xu576a30e2006-06-27 13:22:38 -0700461
462 /* This indicates the skb is from an untrusted source. */
463 SKB_GSO_DODGY = 1 << 2,
Michael Chanb0da85372006-06-29 12:30:00 -0700464
465 /* This indicates the tcp segment has CWR set. */
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700466 SKB_GSO_TCP_ECN = 1 << 3,
467
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400468 SKB_GSO_TCP_FIXEDID = 1 << 4,
Chris Leech01d5b2f2009-02-27 14:06:49 -0800469
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400470 SKB_GSO_TCPV6 = 1 << 5,
Pravin B Shelar68c33162013-02-14 14:02:41 +0000471
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400472 SKB_GSO_FCOE = 1 << 6,
Pravin B Shelar73136262013-03-07 13:21:51 +0000473
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400474 SKB_GSO_GRE = 1 << 7,
Simon Horman0d89d202013-05-23 21:02:52 +0000475
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400476 SKB_GSO_GRE_CSUM = 1 << 8,
Eric Dumazetcb32f512013-10-19 11:42:57 -0700477
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400478 SKB_GSO_IPIP = 1 << 9,
Eric Dumazet61c1db72013-10-20 20:47:30 -0700479
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400480 SKB_GSO_SIT = 1 << 10,
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700481
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400482 SKB_GSO_UDP_TUNNEL = 1 << 11,
Tom Herbert4749c092014-06-04 17:20:23 -0700483
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400484 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
485
Alexander Duyck802ab552016-04-10 21:45:03 -0400486 SKB_GSO_PARTIAL = 1 << 13,
487
488 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
Herbert Xu79671682006-06-22 02:40:14 -0700489};
490
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700491#if BITS_PER_LONG > 32
492#define NET_SKBUFF_DATA_USES_OFFSET 1
493#endif
494
495#ifdef NET_SKBUFF_DATA_USES_OFFSET
496typedef unsigned int sk_buff_data_t;
497#else
498typedef unsigned char *sk_buff_data_t;
499#endif
500
Eric Dumazet363ec39232014-02-26 14:02:11 -0800501/**
502 * struct skb_mstamp - multi resolution time stamps
503 * @stamp_us: timestamp in us resolution
504 * @stamp_jiffies: timestamp in jiffies
505 */
506struct skb_mstamp {
507 union {
508 u64 v64;
509 struct {
510 u32 stamp_us;
511 u32 stamp_jiffies;
512 };
513 };
514};
515
516/**
517 * skb_mstamp_get - get current timestamp
518 * @cl: place to store timestamps
519 */
520static inline void skb_mstamp_get(struct skb_mstamp *cl)
521{
522 u64 val = local_clock();
523
524 do_div(val, NSEC_PER_USEC);
525 cl->stamp_us = (u32)val;
526 cl->stamp_jiffies = (u32)jiffies;
527}
528
529/**
530 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
531 * @t1: pointer to newest sample
532 * @t0: pointer to oldest sample
533 */
534static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
535 const struct skb_mstamp *t0)
536{
537 s32 delta_us = t1->stamp_us - t0->stamp_us;
538 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
539
540 /* If delta_us is negative, this might be because interval is too big,
541 * or local_clock() drift is too big : fallback using jiffies.
542 */
543 if (delta_us <= 0 ||
544 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
545
546 delta_us = jiffies_to_usecs(delta_jiffies);
547
548 return delta_us;
549}
550
Yuchung Cheng625a5e12015-10-16 21:57:45 -0700551static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
552 const struct skb_mstamp *t0)
553{
554 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
555
556 if (!diff)
557 diff = t1->stamp_us - t0->stamp_us;
558 return diff > 0;
559}
Eric Dumazet363ec39232014-02-26 14:02:11 -0800560
Linus Torvalds1da177e2005-04-16 15:20:36 -0700561/**
562 * struct sk_buff - socket buffer
563 * @next: Next buffer in list
564 * @prev: Previous buffer in list
Eric Dumazet363ec39232014-02-26 14:02:11 -0800565 * @tstamp: Time we arrived/left
Eric Dumazet56b17422014-11-03 08:19:53 -0800566 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700567 * @sk: Socket we are owned by
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 * @dev: Device we arrived on/are leaving by
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700569 * @cb: Control buffer. Free for use by every layer. Put private vars here
Eric Dumazet7fee2262010-05-11 23:19:48 +0000570 * @_skb_refdst: destination entry (with norefcount bit)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700571 * @sp: the security path, used for xfrm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700572 * @len: Length of actual data
573 * @data_len: Data length
574 * @mac_len: Length of link layer header
Patrick McHardy334a8132007-06-25 04:35:20 -0700575 * @hdr_len: writable header length of cloned skb
Herbert Xu663ead32007-04-09 11:59:07 -0700576 * @csum: Checksum (must include start/offset pair)
577 * @csum_start: Offset from skb->head where checksumming should start
578 * @csum_offset: Offset from csum_start where checksum should be stored
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700579 * @priority: Packet queueing priority
WANG Cong60ff7462014-05-04 16:39:18 -0700580 * @ignore_df: allow local fragmentation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 * @cloned: Head may be cloned (check refcnt to be sure)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700582 * @ip_summed: Driver fed us an IP checksum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 * @nohdr: Payload reference only, must not modify header
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700584 * @nfctinfo: Relationship of this skb to the connection
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585 * @pkt_type: Packet class
Randy Dunlapc83c2482005-10-18 22:07:41 -0700586 * @fclone: skbuff clone status
Randy Dunlapc83c2482005-10-18 22:07:41 -0700587 * @ipvs_property: skbuff is owned by ipvs
Randy Dunlap31729362008-02-18 20:52:13 -0800588 * @peeked: this packet has been seen already, so stats have been
589 * done for it, don't do them again
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700590 * @nf_trace: netfilter packet trace flag
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700591 * @protocol: Packet protocol from driver
592 * @destructor: Destruct function
593 * @nfct: Associated connection, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
Eric Dumazet8964be42009-11-20 15:35:04 -0800595 * @skb_iif: ifindex of device we arrived on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 * @tc_index: Traffic control index
597 * @tc_verd: traffic control verdict
Tom Herbert61b905d2014-03-24 15:34:47 -0700598 * @hash: the packet hash
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700599 * @queue_mapping: Queue mapping for multiqueue devices
David S. Miller0b725a22014-08-25 15:51:53 -0700600 * @xmit_more: More SKBs are pending for this queue
Randy Dunlap553a5672008-04-20 10:51:01 -0700601 * @ndisc_nodetype: router type (from link layer)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700602 * @ooo_okay: allow the mapping of a socket to a queue to be changed
Tom Herbert61b905d2014-03-24 15:34:47 -0700603 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
Changli Gao4ca24622011-08-19 07:26:44 -0700604 * ports.
Tom Herberta3b18dd2014-07-01 21:33:17 -0700605 * @sw_hash: indicates hash was computed in software stack
Johannes Berg6e3e9392011-11-09 10:15:42 +0100606 * @wifi_acked_valid: wifi_acked was set
607 * @wifi_acked: whether frame was acked on wifi or not
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000608 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
Eliezer Tamir06021292013-06-10 11:39:50 +0300609 * @napi_id: id of the NAPI struct this skb came from
James Morris984bc162006-06-09 00:29:17 -0700610 * @secmark: security marking
Scott Feldman0c4f6912015-07-18 18:24:48 -0700611 * @offload_fwd_mark: fwding offload mark
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700612 * @mark: Generic packet mark
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000613 * @vlan_proto: vlan encapsulation protocol
Patrick McHardy6aa895b2008-07-14 22:49:06 -0700614 * @vlan_tci: vlan tag control information
Simon Horman0d89d202013-05-23 21:02:52 +0000615 * @inner_protocol: Protocol (encapsulation)
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000616 * @inner_transport_header: Inner transport layer header (encapsulation)
617 * @inner_network_header: Network layer header (encapsulation)
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000618 * @inner_mac_header: Link layer header (encapsulation)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700619 * @transport_header: Transport layer header
620 * @network_header: Network layer header
621 * @mac_header: Link layer header
622 * @tail: Tail pointer
623 * @end: End pointer
624 * @head: Head of buffer
625 * @data: Data head pointer
626 * @truesize: Buffer size
627 * @users: User count - see {datagram,tcp}.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700628 */
629
630struct sk_buff {
Eric Dumazet363ec39232014-02-26 14:02:11 -0800631 union {
Eric Dumazet56b17422014-11-03 08:19:53 -0800632 struct {
633 /* These two members must be first. */
634 struct sk_buff *next;
635 struct sk_buff *prev;
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000636
Eric Dumazet56b17422014-11-03 08:19:53 -0800637 union {
638 ktime_t tstamp;
639 struct skb_mstamp skb_mstamp;
640 };
641 };
642 struct rb_node rbnode; /* used in netem & tcp stack */
643 };
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000644 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700646
Linus Torvalds1da177e2005-04-16 15:20:36 -0700647 /*
648 * This is the control buffer. It is free to use for every
649 * layer. Please put your private variables there. If you
650 * want to keep them across layers you have to do a skb_clone()
651 * first. This is owned by whoever has the skb queued ATM.
652 */
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000653 char cb[48] __aligned(8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700654
Eric Dumazet7fee2262010-05-11 23:19:48 +0000655 unsigned long _skb_refdst;
Eric Dumazetb1937222014-09-28 22:18:47 -0700656 void (*destructor)(struct sk_buff *skb);
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000657#ifdef CONFIG_XFRM
658 struct sec_path *sp;
659#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700660#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
661 struct nf_conntrack *nfct;
662#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200663#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Eric Dumazetb1937222014-09-28 22:18:47 -0700664 struct nf_bridge_info *nf_bridge;
665#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700666 unsigned int len,
Patrick McHardy334a8132007-06-25 04:35:20 -0700667 data_len;
668 __u16 mac_len,
669 hdr_len;
Eric Dumazetb1937222014-09-28 22:18:47 -0700670
671 /* Following fields are _not_ copied in __copy_skb_header()
672 * Note that queue_mapping is here mostly to fill a hole.
673 */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200674 kmemcheck_bitfield_begin(flags1);
Eric Dumazetb1937222014-09-28 22:18:47 -0700675 __u16 queue_mapping;
676 __u8 cloned:1,
Harald Welte6869c4d2005-08-09 19:24:19 -0700677 nohdr:1,
Patrick McHardyb84f4cc2005-11-20 21:19:21 -0800678 fclone:2,
Herbert Xua59322b2007-12-05 01:53:40 -0800679 peeked:1,
Eric Dumazetb1937222014-09-28 22:18:47 -0700680 head_frag:1,
681 xmit_more:1;
682 /* one bit hole */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200683 kmemcheck_bitfield_end(flags1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700684
Eric Dumazetb1937222014-09-28 22:18:47 -0700685 /* fields enclosed in headers_start/headers_end are copied
686 * using a single memcpy() in __copy_skb_header()
687 */
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700688 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700689 __u32 headers_start[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700690 /* public: */
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200691
692/* if you move pkt_type around you also must adapt those constants */
693#ifdef __BIG_ENDIAN_BITFIELD
694#define PKT_TYPE_MAX (7 << 5)
695#else
696#define PKT_TYPE_MAX 7
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100697#endif
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200698#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
699
700 __u8 __pkt_type_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700701 __u8 pkt_type:3;
702 __u8 pfmemalloc:1;
703 __u8 ignore_df:1;
704 __u8 nfctinfo:3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700705
Eric Dumazetb1937222014-09-28 22:18:47 -0700706 __u8 nf_trace:1;
707 __u8 ip_summed:2;
708 __u8 ooo_okay:1;
709 __u8 l4_hash:1;
710 __u8 sw_hash:1;
711 __u8 wifi_acked_valid:1;
712 __u8 wifi_acked:1;
713
714 __u8 no_fcs:1;
715 /* Indicates the inner headers are valid in the skbuff. */
716 __u8 encapsulation:1;
717 __u8 encap_hdr_csum:1;
718 __u8 csum_valid:1;
719 __u8 csum_complete_sw:1;
720 __u8 csum_level:2;
721 __u8 csum_bad:1;
722
723#ifdef CONFIG_IPV6_NDISC_NODETYPE
724 __u8 ndisc_nodetype:2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700725#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700726 __u8 ipvs_property:1;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700727 __u8 inner_protocol_type:1;
Tom Herberte585f232014-11-04 09:06:54 -0800728 __u8 remcsum_offload:1;
729 /* 3 or 5 bit hole */
Alexander Duyck4031ae62012-01-27 06:22:53 +0000730
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731#ifdef CONFIG_NET_SCHED
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700732 __u16 tc_index; /* traffic control index */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733#ifdef CONFIG_NET_CLS_ACT
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700734 __u16 tc_verd; /* traffic control verdict */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700736#endif
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200737
Eric Dumazetb1937222014-09-28 22:18:47 -0700738 union {
739 __wsum csum;
740 struct {
741 __u16 csum_start;
742 __u16 csum_offset;
743 };
744 };
745 __u32 priority;
746 int skb_iif;
747 __u32 hash;
748 __be16 vlan_proto;
749 __u16 vlan_tci;
Eric Dumazet2bd82482015-02-03 23:48:24 -0800750#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
751 union {
752 unsigned int napi_id;
753 unsigned int sender_cpu;
754 };
Chris Leech97fc2f02006-05-23 17:55:33 -0700755#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700756 union {
James Morris984bc162006-06-09 00:29:17 -0700757#ifdef CONFIG_NETWORK_SECMARK
Scott Feldman0c4f6912015-07-18 18:24:48 -0700758 __u32 secmark;
James Morris984bc162006-06-09 00:29:17 -0700759#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700760#ifdef CONFIG_NET_SWITCHDEV
761 __u32 offload_fwd_mark;
762#endif
763 };
764
Neil Horman3b885782009-10-12 13:26:31 -0700765 union {
766 __u32 mark;
Eric Dumazet16fad692013-03-14 05:40:32 +0000767 __u32 reserved_tailroom;
Neil Horman3b885782009-10-12 13:26:31 -0700768 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769
Tom Herbert8bce6d72014-09-29 20:22:29 -0700770 union {
771 __be16 inner_protocol;
772 __u8 inner_ipproto;
773 };
774
Simon Horman1a37e412013-05-23 21:02:51 +0000775 __u16 inner_transport_header;
776 __u16 inner_network_header;
777 __u16 inner_mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700778
779 __be16 protocol;
Simon Horman1a37e412013-05-23 21:02:51 +0000780 __u16 transport_header;
781 __u16 network_header;
782 __u16 mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700783
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700784 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700785 __u32 headers_end[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700786 /* public: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700787
Linus Torvalds1da177e2005-04-16 15:20:36 -0700788 /* These elements must be at the end, see alloc_skb() for details. */
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700789 sk_buff_data_t tail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700790 sk_buff_data_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791 unsigned char *head,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700792 *data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700793 unsigned int truesize;
794 atomic_t users;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700795};
796
797#ifdef __KERNEL__
798/*
799 * Handling routines are only of interest to the kernel
800 */
801#include <linux/slab.h>
802
Linus Torvalds1da177e2005-04-16 15:20:36 -0700803
Mel Gormanc93bdd02012-07-31 16:44:19 -0700804#define SKB_ALLOC_FCLONE 0x01
805#define SKB_ALLOC_RX 0x02
Alexander Duyckfd11a832014-12-09 19:40:49 -0800806#define SKB_ALLOC_NAPI 0x04
Mel Gormanc93bdd02012-07-31 16:44:19 -0700807
808/* Returns true if the skb was allocated from PFMEMALLOC reserves */
809static inline bool skb_pfmemalloc(const struct sk_buff *skb)
810{
811 return unlikely(skb->pfmemalloc);
812}
813
Eric Dumazet7fee2262010-05-11 23:19:48 +0000814/*
815 * skb might have a dst pointer attached, refcounted or not.
816 * _skb_refdst low order bit is set if refcount was _not_ taken
817 */
818#define SKB_DST_NOREF 1UL
819#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
820
821/**
822 * skb_dst - returns skb dst_entry
823 * @skb: buffer
824 *
825 * Returns skb dst_entry, regardless of reference taken or not.
826 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000827static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
828{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000829 /* If refdst was not refcounted, check we still are in a
830 * rcu_read_lock section
831 */
832 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
833 !rcu_read_lock_held() &&
834 !rcu_read_lock_bh_held());
835 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
Eric Dumazetadf30902009-06-02 05:19:30 +0000836}
837
Eric Dumazet7fee2262010-05-11 23:19:48 +0000838/**
839 * skb_dst_set - sets skb dst
840 * @skb: buffer
841 * @dst: dst entry
842 *
843 * Sets skb dst, assuming a reference was taken on dst and should
844 * be released by skb_dst_drop()
845 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000846static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
847{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000848 skb->_skb_refdst = (unsigned long)dst;
849}
850
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200851/**
852 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
853 * @skb: buffer
854 * @dst: dst entry
855 *
856 * Sets skb dst, assuming a reference was not taken on dst.
857 * If dst entry is cached, we do not take reference and dst_release
858 * will be avoided by refdst_drop. If dst entry is not cached, we take
859 * reference, so that last dst_release can destroy the dst immediately.
860 */
861static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
862{
Hannes Frederic Sowadbfc4fb2014-12-06 19:19:42 +0100863 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
864 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200865}
Eric Dumazet7fee2262010-05-11 23:19:48 +0000866
867/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300868 * skb_dst_is_noref - Test if skb dst isn't refcounted
Eric Dumazet7fee2262010-05-11 23:19:48 +0000869 * @skb: buffer
870 */
871static inline bool skb_dst_is_noref(const struct sk_buff *skb)
872{
873 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
Eric Dumazetadf30902009-06-02 05:19:30 +0000874}
875
Eric Dumazet511c3f92009-06-02 05:14:27 +0000876static inline struct rtable *skb_rtable(const struct sk_buff *skb)
877{
Eric Dumazetadf30902009-06-02 05:19:30 +0000878 return (struct rtable *)skb_dst(skb);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000879}
880
Joe Perches7965bd42013-09-26 14:48:15 -0700881void kfree_skb(struct sk_buff *skb);
882void kfree_skb_list(struct sk_buff *segs);
883void skb_tx_error(struct sk_buff *skb);
884void consume_skb(struct sk_buff *skb);
885void __kfree_skb(struct sk_buff *skb);
Eric Dumazetd7e88832012-04-30 08:10:34 +0000886extern struct kmem_cache *skbuff_head_cache;
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000887
Joe Perches7965bd42013-09-26 14:48:15 -0700888void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
889bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
890 bool *fragstolen, int *delta_truesize);
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000891
Joe Perches7965bd42013-09-26 14:48:15 -0700892struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
893 int node);
Eric Dumazet2ea2f622015-04-24 16:05:01 -0700894struct sk_buff *__build_skb(void *data, unsigned int frag_size);
Joe Perches7965bd42013-09-26 14:48:15 -0700895struct sk_buff *build_skb(void *data, unsigned int frag_size);
David S. Millerd179cd12005-08-17 14:57:30 -0700896static inline struct sk_buff *alloc_skb(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100897 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700898{
Eric Dumazet564824b2010-10-11 19:05:25 +0000899 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700900}
901
Eric Dumazet2e4e4412014-09-17 04:49:49 -0700902struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
903 unsigned long data_len,
904 int max_page_order,
905 int *errcode,
906 gfp_t gfp_mask);
907
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700908/* Layout of fast clones : [skb1][skb2][fclone_ref] */
909struct sk_buff_fclones {
910 struct sk_buff skb1;
911
912 struct sk_buff skb2;
913
914 atomic_t fclone_ref;
915};
916
917/**
918 * skb_fclone_busy - check if fclone is busy
919 * @skb: buffer
920 *
Masanari Iidabda13fe2015-12-13 16:53:02 +0900921 * Returns true if skb is a fast clone, and its clone is not freed.
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700922 * Some drivers call skb_orphan() in their ndo_start_xmit(),
923 * so we also check that this didnt happen.
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700924 */
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700925static inline bool skb_fclone_busy(const struct sock *sk,
926 const struct sk_buff *skb)
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700927{
928 const struct sk_buff_fclones *fclones;
929
930 fclones = container_of(skb, struct sk_buff_fclones, skb1);
931
932 return skb->fclone == SKB_FCLONE_ORIG &&
Eric Dumazet6ffe75e2014-12-03 17:04:39 -0800933 atomic_read(&fclones->fclone_ref) > 1 &&
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700934 fclones->skb2.sk == sk;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700935}
936
David S. Millerd179cd12005-08-17 14:57:30 -0700937static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100938 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700939{
Mel Gormanc93bdd02012-07-31 16:44:19 -0700940 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700941}
942
Joe Perches7965bd42013-09-26 14:48:15 -0700943struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000944static inline struct sk_buff *alloc_skb_head(gfp_t priority)
945{
946 return __alloc_skb_head(priority, -1);
947}
948
Joe Perches7965bd42013-09-26 14:48:15 -0700949struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
950int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
951struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
952struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
Octavian Purdilabad93e92014-06-12 01:36:26 +0300953struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
954 gfp_t gfp_mask, bool fclone);
955static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
956 gfp_t gfp_mask)
957{
958 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
959}
Eric Dumazet117632e2011-12-03 21:39:53 +0000960
Joe Perches7965bd42013-09-26 14:48:15 -0700961int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
962struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
963 unsigned int headroom);
964struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
965 int newtailroom, gfp_t priority);
Fan Du25a91d82014-01-18 09:54:23 +0800966int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
967 int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -0700968int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
969 int len);
970int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
971int skb_pad(struct sk_buff *skb, int pad);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000972#define dev_kfree_skb(a) consume_skb(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700973
Joe Perches7965bd42013-09-26 14:48:15 -0700974int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
975 int getfrag(void *from, char *to, int offset,
976 int len, int odd, struct sk_buff *skb),
977 void *from, int length);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700978
Hannes Frederic Sowabe12a1f2015-05-21 16:59:58 +0200979int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
980 int offset, size_t size);
981
Eric Dumazetd94d9fe2009-11-04 09:50:58 -0800982struct skb_seq_state {
Thomas Graf677e90e2005-06-23 20:59:51 -0700983 __u32 lower_offset;
984 __u32 upper_offset;
985 __u32 frag_idx;
986 __u32 stepped_offset;
987 struct sk_buff *root_skb;
988 struct sk_buff *cur_skb;
989 __u8 *frag_data;
990};
991
Joe Perches7965bd42013-09-26 14:48:15 -0700992void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
993 unsigned int to, struct skb_seq_state *st);
994unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
995 struct skb_seq_state *st);
996void skb_abort_seq_read(struct skb_seq_state *st);
Thomas Graf677e90e2005-06-23 20:59:51 -0700997
Joe Perches7965bd42013-09-26 14:48:15 -0700998unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
Bojan Prtvar059a2442015-02-22 11:46:35 +0100999 unsigned int to, struct ts_config *config);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07001000
Tom Herbert09323cc2013-12-15 22:16:19 -08001001/*
1002 * Packet hash types specify the type of hash in skb_set_hash.
1003 *
1004 * Hash types refer to the protocol layer addresses which are used to
1005 * construct a packet's hash. The hashes are used to differentiate or identify
1006 * flows of the protocol layer for the hash type. Hash types are either
1007 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1008 *
1009 * Properties of hashes:
1010 *
1011 * 1) Two packets in different flows have different hash values
1012 * 2) Two packets in the same flow should have the same hash value
1013 *
1014 * A hash at a higher layer is considered to be more specific. A driver should
1015 * set the most specific hash possible.
1016 *
1017 * A driver cannot indicate a more specific hash than the layer at which a hash
1018 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1019 *
1020 * A driver may indicate a hash level which is less specific than the
1021 * actual layer the hash was computed on. For instance, a hash computed
1022 * at L4 may be considered an L3 hash. This should only be done if the
1023 * driver can't unambiguously determine that the HW computed the hash at
1024 * the higher layer. Note that the "should" in the second property above
1025 * permits this.
1026 */
1027enum pkt_hash_types {
1028 PKT_HASH_TYPE_NONE, /* Undefined type */
1029 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1030 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1031 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1032};
1033
Tom Herbertbcc83832015-09-01 09:24:24 -07001034static inline void skb_clear_hash(struct sk_buff *skb)
1035{
1036 skb->hash = 0;
1037 skb->sw_hash = 0;
1038 skb->l4_hash = 0;
1039}
1040
1041static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1042{
1043 if (!skb->l4_hash)
1044 skb_clear_hash(skb);
1045}
1046
1047static inline void
1048__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1049{
1050 skb->l4_hash = is_l4;
1051 skb->sw_hash = is_sw;
1052 skb->hash = hash;
1053}
1054
Tom Herbert09323cc2013-12-15 22:16:19 -08001055static inline void
1056skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1057{
Tom Herbertbcc83832015-09-01 09:24:24 -07001058 /* Used by drivers to set hash from HW */
1059 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1060}
1061
1062static inline void
1063__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1064{
1065 __skb_set_hash(skb, hash, true, is_l4);
Tom Herbert09323cc2013-12-15 22:16:19 -08001066}
1067
Tom Herberte5276932015-09-01 09:24:23 -07001068void __skb_get_hash(struct sk_buff *skb);
1069u32 skb_get_poff(const struct sk_buff *skb);
1070u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1071 const struct flow_keys *keys, int hlen);
1072__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1073 void *data, int hlen_proto);
1074
1075static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1076 int thoff, u8 ip_proto)
1077{
1078 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1079}
1080
1081void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1082 const struct flow_dissector_key *key,
1083 unsigned int key_count);
1084
1085bool __skb_flow_dissect(const struct sk_buff *skb,
1086 struct flow_dissector *flow_dissector,
1087 void *target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001088 void *data, __be16 proto, int nhoff, int hlen,
1089 unsigned int flags);
Tom Herberte5276932015-09-01 09:24:23 -07001090
1091static inline bool skb_flow_dissect(const struct sk_buff *skb,
1092 struct flow_dissector *flow_dissector,
Tom Herbertcd79a232015-09-01 09:24:27 -07001093 void *target_container, unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001094{
1095 return __skb_flow_dissect(skb, flow_dissector, target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001096 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001097}
1098
1099static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
Tom Herbertcd79a232015-09-01 09:24:27 -07001100 struct flow_keys *flow,
1101 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001102{
1103 memset(flow, 0, sizeof(*flow));
1104 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001105 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001106}
1107
1108static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1109 void *data, __be16 proto,
Tom Herbertcd79a232015-09-01 09:24:27 -07001110 int nhoff, int hlen,
1111 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001112{
1113 memset(flow, 0, sizeof(*flow));
1114 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001115 data, proto, nhoff, hlen, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001116}
1117
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001118static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001119{
Tom Herberta3b18dd2014-07-01 21:33:17 -07001120 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001121 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001122
Tom Herbert61b905d2014-03-24 15:34:47 -07001123 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001124}
1125
David S. Miller20a17bf2015-09-01 21:19:17 -07001126__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
Tom Herbertf70ea012015-07-31 16:52:10 -07001127
David S. Miller20a17bf2015-09-01 21:19:17 -07001128static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
Tom Herbertf70ea012015-07-31 16:52:10 -07001129{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001130 if (!skb->l4_hash && !skb->sw_hash) {
1131 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001132 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001133
Tom Herbertde4c1f82015-09-01 18:11:04 -07001134 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001135 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001136
1137 return skb->hash;
1138}
1139
David S. Miller20a17bf2015-09-01 21:19:17 -07001140__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
Tom Herbertf70ea012015-07-31 16:52:10 -07001141
David S. Miller20a17bf2015-09-01 21:19:17 -07001142static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
Tom Herbertf70ea012015-07-31 16:52:10 -07001143{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001144 if (!skb->l4_hash && !skb->sw_hash) {
1145 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001146 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001147
Tom Herbertde4c1f82015-09-01 18:11:04 -07001148 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001149 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001150
1151 return skb->hash;
1152}
1153
Tom Herbert50fb7992015-05-01 11:30:12 -07001154__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1155
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001156static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1157{
Tom Herbert61b905d2014-03-24 15:34:47 -07001158 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001159}
1160
Tom Herbert3df7a742013-12-15 22:16:29 -08001161static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1162{
Tom Herbert61b905d2014-03-24 15:34:47 -07001163 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -07001164 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -07001165 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -08001166};
1167
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001168#ifdef NET_SKBUFF_DATA_USES_OFFSET
1169static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1170{
1171 return skb->head + skb->end;
1172}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001173
1174static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1175{
1176 return skb->end;
1177}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001178#else
1179static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1180{
1181 return skb->end;
1182}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001183
1184static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1185{
1186 return skb->end - skb->head;
1187}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001188#endif
1189
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001191#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001192
Patrick Ohlyac45f602009-02-12 05:03:37 +00001193static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1194{
1195 return &skb_shinfo(skb)->hwtstamps;
1196}
1197
Linus Torvalds1da177e2005-04-16 15:20:36 -07001198/**
1199 * skb_queue_empty - check if a queue is empty
1200 * @list: queue head
1201 *
1202 * Returns true if the queue is empty, false otherwise.
1203 */
1204static inline int skb_queue_empty(const struct sk_buff_head *list)
1205{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001206 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001207}
1208
1209/**
David S. Millerfc7ebb22008-09-23 00:34:07 -07001210 * skb_queue_is_last - check if skb is the last entry in the queue
1211 * @list: queue head
1212 * @skb: buffer
1213 *
1214 * Returns true if @skb is the last buffer on the list.
1215 */
1216static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1217 const struct sk_buff *skb)
1218{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001219 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -07001220}
1221
1222/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001223 * skb_queue_is_first - check if skb is the first entry in the queue
1224 * @list: queue head
1225 * @skb: buffer
1226 *
1227 * Returns true if @skb is the first buffer on the list.
1228 */
1229static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1230 const struct sk_buff *skb)
1231{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001232 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001233}
1234
1235/**
David S. Miller249c8b42008-09-23 00:44:42 -07001236 * skb_queue_next - return the next packet in the queue
1237 * @list: queue head
1238 * @skb: current buffer
1239 *
1240 * Return the next packet in @list after @skb. It is only valid to
1241 * call this if skb_queue_is_last() evaluates to false.
1242 */
1243static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1244 const struct sk_buff *skb)
1245{
1246 /* This BUG_ON may seem severe, but if we just return then we
1247 * are going to dereference garbage.
1248 */
1249 BUG_ON(skb_queue_is_last(list, skb));
1250 return skb->next;
1251}
1252
1253/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001254 * skb_queue_prev - return the prev packet in the queue
1255 * @list: queue head
1256 * @skb: current buffer
1257 *
1258 * Return the prev packet in @list before @skb. It is only valid to
1259 * call this if skb_queue_is_first() evaluates to false.
1260 */
1261static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1262 const struct sk_buff *skb)
1263{
1264 /* This BUG_ON may seem severe, but if we just return then we
1265 * are going to dereference garbage.
1266 */
1267 BUG_ON(skb_queue_is_first(list, skb));
1268 return skb->prev;
1269}
1270
1271/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272 * skb_get - reference buffer
1273 * @skb: buffer to reference
1274 *
1275 * Makes another reference to a socket buffer and returns a pointer
1276 * to the buffer.
1277 */
1278static inline struct sk_buff *skb_get(struct sk_buff *skb)
1279{
1280 atomic_inc(&skb->users);
1281 return skb;
1282}
1283
1284/*
1285 * If users == 1, we are the only owner and are can avoid redundant
1286 * atomic change.
1287 */
1288
1289/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001290 * skb_cloned - is the buffer a clone
1291 * @skb: buffer to check
1292 *
1293 * Returns true if the buffer was generated with skb_clone() and is
1294 * one of multiple shared copies of the buffer. Cloned buffers are
1295 * shared data so must not be written to under normal circumstances.
1296 */
1297static inline int skb_cloned(const struct sk_buff *skb)
1298{
1299 return skb->cloned &&
1300 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1301}
1302
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001303static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1304{
Mel Gormand0164ad2015-11-06 16:28:21 -08001305 might_sleep_if(gfpflags_allow_blocking(pri));
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001306
1307 if (skb_cloned(skb))
1308 return pskb_expand_head(skb, 0, 0, pri);
1309
1310 return 0;
1311}
1312
Linus Torvalds1da177e2005-04-16 15:20:36 -07001313/**
1314 * skb_header_cloned - is the header a clone
1315 * @skb: buffer to check
1316 *
1317 * Returns true if modifying the header part of the buffer requires
1318 * the data to be copied.
1319 */
1320static inline int skb_header_cloned(const struct sk_buff *skb)
1321{
1322 int dataref;
1323
1324 if (!skb->cloned)
1325 return 0;
1326
1327 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1328 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1329 return dataref != 1;
1330}
1331
1332/**
1333 * skb_header_release - release reference to header
1334 * @skb: buffer to operate on
1335 *
1336 * Drop a reference to the header part of the buffer. This is done
1337 * by acquiring a payload reference. You must not read from the header
1338 * part of skb->data after this.
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001339 * Note : Check if you can use __skb_header_release() instead.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001340 */
1341static inline void skb_header_release(struct sk_buff *skb)
1342{
1343 BUG_ON(skb->nohdr);
1344 skb->nohdr = 1;
1345 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1346}
1347
1348/**
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001349 * __skb_header_release - release reference to header
1350 * @skb: buffer to operate on
1351 *
1352 * Variant of skb_header_release() assuming skb is private to caller.
1353 * We can avoid one atomic operation.
1354 */
1355static inline void __skb_header_release(struct sk_buff *skb)
1356{
1357 skb->nohdr = 1;
1358 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1359}
1360
1361
1362/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001363 * skb_shared - is the buffer shared
1364 * @skb: buffer to check
1365 *
1366 * Returns true if more than one person has a reference to this
1367 * buffer.
1368 */
1369static inline int skb_shared(const struct sk_buff *skb)
1370{
1371 return atomic_read(&skb->users) != 1;
1372}
1373
1374/**
1375 * skb_share_check - check if buffer is shared and if so clone it
1376 * @skb: buffer to check
1377 * @pri: priority for memory allocation
1378 *
1379 * If the buffer is shared the buffer is cloned and the old copy
1380 * drops a reference. A new clone with a single reference is returned.
1381 * If the buffer is not shared the original buffer is returned. When
1382 * being called from interrupt status or with spinlocks held pri must
1383 * be GFP_ATOMIC.
1384 *
1385 * NULL is returned on a memory allocation failure.
1386 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001387static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388{
Mel Gormand0164ad2015-11-06 16:28:21 -08001389 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001390 if (skb_shared(skb)) {
1391 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001392
1393 if (likely(nskb))
1394 consume_skb(skb);
1395 else
1396 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001397 skb = nskb;
1398 }
1399 return skb;
1400}
1401
1402/*
1403 * Copy shared buffers into a new sk_buff. We effectively do COW on
1404 * packets to handle cases where we have a local reader and forward
1405 * and a couple of other messy ones. The normal one is tcpdumping
1406 * a packet thats being forwarded.
1407 */
1408
1409/**
1410 * skb_unshare - make a copy of a shared buffer
1411 * @skb: buffer to check
1412 * @pri: priority for memory allocation
1413 *
1414 * If the socket buffer is a clone then this function creates a new
1415 * copy of the data, drops a reference count on the old copy and returns
1416 * the new copy with the reference count at 1. If the buffer is not a clone
1417 * the original buffer is returned. When called with a spinlock held or
1418 * from interrupt state @pri must be %GFP_ATOMIC
1419 *
1420 * %NULL is returned on a memory allocation failure.
1421 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001422static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001423 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424{
Mel Gormand0164ad2015-11-06 16:28:21 -08001425 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001426 if (skb_cloned(skb)) {
1427 struct sk_buff *nskb = skb_copy(skb, pri);
Alexander Aring31eff812014-10-10 23:10:47 +02001428
1429 /* Free our shared copy */
1430 if (likely(nskb))
1431 consume_skb(skb);
1432 else
1433 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001434 skb = nskb;
1435 }
1436 return skb;
1437}
1438
1439/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001440 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001441 * @list_: list to peek at
1442 *
1443 * Peek an &sk_buff. Unlike most other operations you _MUST_
1444 * be careful with this one. A peek leaves the buffer on the
1445 * list and someone else may run off with it. You must hold
1446 * the appropriate locks or have a private queue to do this.
1447 *
1448 * Returns %NULL for an empty list or a pointer to the head element.
1449 * The reference count is not incremented and the reference is therefore
1450 * volatile. Use with caution.
1451 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001452static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001453{
Eric Dumazet18d07002012-04-30 16:31:46 +00001454 struct sk_buff *skb = list_->next;
1455
1456 if (skb == (struct sk_buff *)list_)
1457 skb = NULL;
1458 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459}
1460
1461/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001462 * skb_peek_next - peek skb following the given one from a queue
1463 * @skb: skb to start from
1464 * @list_: list to peek at
1465 *
1466 * Returns %NULL when the end of the list is met or a pointer to the
1467 * next element. The reference count is not incremented and the
1468 * reference is therefore volatile. Use with caution.
1469 */
1470static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1471 const struct sk_buff_head *list_)
1472{
1473 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001474
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001475 if (next == (struct sk_buff *)list_)
1476 next = NULL;
1477 return next;
1478}
1479
1480/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001481 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482 * @list_: list to peek at
1483 *
1484 * Peek an &sk_buff. Unlike most other operations you _MUST_
1485 * be careful with this one. A peek leaves the buffer on the
1486 * list and someone else may run off with it. You must hold
1487 * the appropriate locks or have a private queue to do this.
1488 *
1489 * Returns %NULL for an empty list or a pointer to the tail element.
1490 * The reference count is not incremented and the reference is therefore
1491 * volatile. Use with caution.
1492 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001493static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001494{
Eric Dumazet18d07002012-04-30 16:31:46 +00001495 struct sk_buff *skb = list_->prev;
1496
1497 if (skb == (struct sk_buff *)list_)
1498 skb = NULL;
1499 return skb;
1500
Linus Torvalds1da177e2005-04-16 15:20:36 -07001501}
1502
1503/**
1504 * skb_queue_len - get queue length
1505 * @list_: list to measure
1506 *
1507 * Return the length of an &sk_buff queue.
1508 */
1509static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1510{
1511 return list_->qlen;
1512}
1513
David S. Miller67fed452008-09-21 22:36:24 -07001514/**
1515 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1516 * @list: queue to initialize
1517 *
1518 * This initializes only the list and queue length aspects of
1519 * an sk_buff_head object. This allows to initialize the list
1520 * aspects of an sk_buff_head without reinitializing things like
1521 * the spinlock. It can also be used for on-stack sk_buff_head
1522 * objects where the spinlock is known to not be used.
1523 */
1524static inline void __skb_queue_head_init(struct sk_buff_head *list)
1525{
1526 list->prev = list->next = (struct sk_buff *)list;
1527 list->qlen = 0;
1528}
1529
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001530/*
1531 * This function creates a split out lock class for each invocation;
1532 * this is needed for now since a whole lot of users of the skb-queue
1533 * infrastructure in drivers have different locking usage (in hardirq)
1534 * than the networking core (in softirq only). In the long run either the
1535 * network layer or drivers should need annotation to consolidate the
1536 * main types of usage into 3 classes.
1537 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001538static inline void skb_queue_head_init(struct sk_buff_head *list)
1539{
1540 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001541 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001542}
1543
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001544static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1545 struct lock_class_key *class)
1546{
1547 skb_queue_head_init(list);
1548 lockdep_set_class(&list->lock, class);
1549}
1550
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001552 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001553 *
1554 * The "__skb_xxxx()" functions are the non-atomic ones that
1555 * can only be called with interrupts disabled.
1556 */
Joe Perches7965bd42013-09-26 14:48:15 -07001557void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1558 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001559static inline void __skb_insert(struct sk_buff *newsk,
1560 struct sk_buff *prev, struct sk_buff *next,
1561 struct sk_buff_head *list)
1562{
1563 newsk->next = next;
1564 newsk->prev = prev;
1565 next->prev = prev->next = newsk;
1566 list->qlen++;
1567}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001568
David S. Miller67fed452008-09-21 22:36:24 -07001569static inline void __skb_queue_splice(const struct sk_buff_head *list,
1570 struct sk_buff *prev,
1571 struct sk_buff *next)
1572{
1573 struct sk_buff *first = list->next;
1574 struct sk_buff *last = list->prev;
1575
1576 first->prev = prev;
1577 prev->next = first;
1578
1579 last->next = next;
1580 next->prev = last;
1581}
1582
1583/**
1584 * skb_queue_splice - join two skb lists, this is designed for stacks
1585 * @list: the new list to add
1586 * @head: the place to add it in the first list
1587 */
1588static inline void skb_queue_splice(const struct sk_buff_head *list,
1589 struct sk_buff_head *head)
1590{
1591 if (!skb_queue_empty(list)) {
1592 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001593 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001594 }
1595}
1596
1597/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001598 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001599 * @list: the new list to add
1600 * @head: the place to add it in the first list
1601 *
1602 * The list at @list is reinitialised
1603 */
1604static inline void skb_queue_splice_init(struct sk_buff_head *list,
1605 struct sk_buff_head *head)
1606{
1607 if (!skb_queue_empty(list)) {
1608 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001609 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001610 __skb_queue_head_init(list);
1611 }
1612}
1613
1614/**
1615 * skb_queue_splice_tail - join two skb lists, each list being a queue
1616 * @list: the new list to add
1617 * @head: the place to add it in the first list
1618 */
1619static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1620 struct sk_buff_head *head)
1621{
1622 if (!skb_queue_empty(list)) {
1623 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001624 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001625 }
1626}
1627
1628/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001629 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001630 * @list: the new list to add
1631 * @head: the place to add it in the first list
1632 *
1633 * Each of the lists is a queue.
1634 * The list at @list is reinitialised
1635 */
1636static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1637 struct sk_buff_head *head)
1638{
1639 if (!skb_queue_empty(list)) {
1640 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001641 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001642 __skb_queue_head_init(list);
1643 }
1644}
1645
Linus Torvalds1da177e2005-04-16 15:20:36 -07001646/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001647 * __skb_queue_after - queue a buffer at the list head
1648 * @list: list to use
1649 * @prev: place after this buffer
1650 * @newsk: buffer to queue
1651 *
1652 * Queue a buffer int the middle of a list. This function takes no locks
1653 * and you must therefore hold required locks before calling it.
1654 *
1655 * A buffer cannot be placed on two lists at the same time.
1656 */
1657static inline void __skb_queue_after(struct sk_buff_head *list,
1658 struct sk_buff *prev,
1659 struct sk_buff *newsk)
1660{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001661 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001662}
1663
Joe Perches7965bd42013-09-26 14:48:15 -07001664void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1665 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001666
Gerrit Renkerf5572852008-04-14 00:05:28 -07001667static inline void __skb_queue_before(struct sk_buff_head *list,
1668 struct sk_buff *next,
1669 struct sk_buff *newsk)
1670{
1671 __skb_insert(newsk, next->prev, next, list);
1672}
1673
Stephen Hemminger300ce172005-10-30 13:47:34 -08001674/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001675 * __skb_queue_head - queue a buffer at the list head
1676 * @list: list to use
1677 * @newsk: buffer to queue
1678 *
1679 * Queue a buffer at the start of a list. This function takes no locks
1680 * and you must therefore hold required locks before calling it.
1681 *
1682 * A buffer cannot be placed on two lists at the same time.
1683 */
Joe Perches7965bd42013-09-26 14:48:15 -07001684void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685static inline void __skb_queue_head(struct sk_buff_head *list,
1686 struct sk_buff *newsk)
1687{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001688 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001689}
1690
1691/**
1692 * __skb_queue_tail - queue a buffer at the list tail
1693 * @list: list to use
1694 * @newsk: buffer to queue
1695 *
1696 * Queue a buffer at the end of a list. This function takes no locks
1697 * and you must therefore hold required locks before calling it.
1698 *
1699 * A buffer cannot be placed on two lists at the same time.
1700 */
Joe Perches7965bd42013-09-26 14:48:15 -07001701void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001702static inline void __skb_queue_tail(struct sk_buff_head *list,
1703 struct sk_buff *newsk)
1704{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001705 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706}
1707
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001709 * remove sk_buff from list. _Must_ be called atomically, and with
1710 * the list known..
1711 */
Joe Perches7965bd42013-09-26 14:48:15 -07001712void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001713static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1714{
1715 struct sk_buff *next, *prev;
1716
1717 list->qlen--;
1718 next = skb->next;
1719 prev = skb->prev;
1720 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001721 next->prev = prev;
1722 prev->next = next;
1723}
1724
Gerrit Renkerf525c062008-04-14 00:04:12 -07001725/**
1726 * __skb_dequeue - remove from the head of the queue
1727 * @list: list to dequeue from
1728 *
1729 * Remove the head of the list. This function does not take any locks
1730 * so must be used with appropriate locks held only. The head item is
1731 * returned or %NULL if the list is empty.
1732 */
Joe Perches7965bd42013-09-26 14:48:15 -07001733struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001734static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1735{
1736 struct sk_buff *skb = skb_peek(list);
1737 if (skb)
1738 __skb_unlink(skb, list);
1739 return skb;
1740}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001741
1742/**
1743 * __skb_dequeue_tail - remove from the tail of the queue
1744 * @list: list to dequeue from
1745 *
1746 * Remove the tail of the list. This function does not take any locks
1747 * so must be used with appropriate locks held only. The tail item is
1748 * returned or %NULL if the list is empty.
1749 */
Joe Perches7965bd42013-09-26 14:48:15 -07001750struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001751static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1752{
1753 struct sk_buff *skb = skb_peek_tail(list);
1754 if (skb)
1755 __skb_unlink(skb, list);
1756 return skb;
1757}
1758
1759
David S. Millerbdcc0922012-03-07 20:53:36 -05001760static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001761{
1762 return skb->data_len;
1763}
1764
1765static inline unsigned int skb_headlen(const struct sk_buff *skb)
1766{
1767 return skb->len - skb->data_len;
1768}
1769
1770static inline int skb_pagelen(const struct sk_buff *skb)
1771{
1772 int i, len = 0;
1773
1774 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001775 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001776 return len + skb_headlen(skb);
1777}
1778
Ian Campbell131ea662011-08-19 06:25:00 +00001779/**
1780 * __skb_fill_page_desc - initialise a paged fragment in an skb
1781 * @skb: buffer containing fragment to be initialised
1782 * @i: paged fragment index to initialise
1783 * @page: the page to use for this fragment
1784 * @off: the offset to the data with @page
1785 * @size: the length of the data
1786 *
1787 * Initialises the @i'th fragment of @skb to point to &size bytes at
1788 * offset @off within @page.
1789 *
1790 * Does not take any additional reference on the fragment.
1791 */
1792static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1793 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001794{
1795 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1796
Mel Gormanc48a11c2012-07-31 16:44:23 -07001797 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001798 * Propagate page pfmemalloc to the skb if we can. The problem is
1799 * that not all callers have unique ownership of the page but rely
1800 * on page_is_pfmemalloc doing the right thing(tm).
Mel Gormanc48a11c2012-07-31 16:44:23 -07001801 */
Ian Campbella8605c62011-10-19 23:01:49 +00001802 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001803 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001804 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001805
1806 page = compound_head(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001807 if (page_is_pfmemalloc(page))
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001808 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001809}
1810
1811/**
1812 * skb_fill_page_desc - initialise a paged fragment in an skb
1813 * @skb: buffer containing fragment to be initialised
1814 * @i: paged fragment index to initialise
1815 * @page: the page to use for this fragment
1816 * @off: the offset to the data with @page
1817 * @size: the length of the data
1818 *
1819 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001820 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001821 * addition updates @skb such that @i is the last fragment.
1822 *
1823 * Does not take any additional reference on the fragment.
1824 */
1825static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1826 struct page *page, int off, int size)
1827{
1828 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001829 skb_shinfo(skb)->nr_frags = i + 1;
1830}
1831
Joe Perches7965bd42013-09-26 14:48:15 -07001832void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1833 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001834
Jason Wangf8e617e2013-11-01 14:07:47 +08001835void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1836 unsigned int truesize);
1837
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001839#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001840#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1841
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001842#ifdef NET_SKBUFF_DATA_USES_OFFSET
1843static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1844{
1845 return skb->head + skb->tail;
1846}
1847
1848static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1849{
1850 skb->tail = skb->data - skb->head;
1851}
1852
1853static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1854{
1855 skb_reset_tail_pointer(skb);
1856 skb->tail += offset;
1857}
Simon Horman7cc46192013-05-28 20:34:29 +00001858
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001859#else /* NET_SKBUFF_DATA_USES_OFFSET */
1860static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1861{
1862 return skb->tail;
1863}
1864
1865static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1866{
1867 skb->tail = skb->data;
1868}
1869
1870static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1871{
1872 skb->tail = skb->data + offset;
1873}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001874
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001875#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1876
Linus Torvalds1da177e2005-04-16 15:20:36 -07001877/*
1878 * Add data to an sk_buff
1879 */
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001880unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001881unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001882static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1883{
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001884 unsigned char *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001885 SKB_LINEAR_ASSERT(skb);
1886 skb->tail += len;
1887 skb->len += len;
1888 return tmp;
1889}
1890
Joe Perches7965bd42013-09-26 14:48:15 -07001891unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1893{
1894 skb->data -= len;
1895 skb->len += len;
1896 return skb->data;
1897}
1898
Joe Perches7965bd42013-09-26 14:48:15 -07001899unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001900static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1901{
1902 skb->len -= len;
1903 BUG_ON(skb->len < skb->data_len);
1904 return skb->data += len;
1905}
1906
David S. Miller47d29642010-05-02 02:21:44 -07001907static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1908{
1909 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1910}
1911
Joe Perches7965bd42013-09-26 14:48:15 -07001912unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001913
1914static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1915{
1916 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07001917 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 return NULL;
1919 skb->len -= len;
1920 return skb->data += len;
1921}
1922
1923static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1924{
1925 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1926}
1927
1928static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1929{
1930 if (likely(len <= skb_headlen(skb)))
1931 return 1;
1932 if (unlikely(len > skb->len))
1933 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07001934 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001935}
1936
1937/**
1938 * skb_headroom - bytes at buffer head
1939 * @skb: buffer to check
1940 *
1941 * Return the number of bytes of free space at the head of an &sk_buff.
1942 */
Chuck Leverc2636b42007-10-23 21:07:32 -07001943static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944{
1945 return skb->data - skb->head;
1946}
1947
1948/**
1949 * skb_tailroom - bytes at buffer end
1950 * @skb: buffer to check
1951 *
1952 * Return the number of bytes of free space at the tail of an sk_buff
1953 */
1954static inline int skb_tailroom(const struct sk_buff *skb)
1955{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001956 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001957}
1958
1959/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00001960 * skb_availroom - 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 * allocated by sk_stream_alloc()
1965 */
1966static inline int skb_availroom(const struct sk_buff *skb)
1967{
Eric Dumazet16fad692013-03-14 05:40:32 +00001968 if (skb_is_nonlinear(skb))
1969 return 0;
1970
1971 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00001972}
1973
1974/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001975 * skb_reserve - adjust headroom
1976 * @skb: buffer to alter
1977 * @len: bytes to move
1978 *
1979 * Increase the headroom of an empty &sk_buff by reducing the tail
1980 * room. This is only allowed for an empty buffer.
1981 */
David S. Miller8243126c2006-01-17 02:54:21 -08001982static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001983{
1984 skb->data += len;
1985 skb->tail += len;
1986}
1987
Benjamin Poirier1837b2e2016-02-29 15:03:33 -08001988/**
1989 * skb_tailroom_reserve - adjust reserved_tailroom
1990 * @skb: buffer to alter
1991 * @mtu: maximum amount of headlen permitted
1992 * @needed_tailroom: minimum amount of reserved_tailroom
1993 *
1994 * Set reserved_tailroom so that headlen can be as large as possible but
1995 * not larger than mtu and tailroom cannot be smaller than
1996 * needed_tailroom.
1997 * The required headroom should already have been reserved before using
1998 * this function.
1999 */
2000static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2001 unsigned int needed_tailroom)
2002{
2003 SKB_LINEAR_ASSERT(skb);
2004 if (mtu < skb_tailroom(skb) - needed_tailroom)
2005 /* use at most mtu */
2006 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2007 else
2008 /* use up to all available space */
2009 skb->reserved_tailroom = needed_tailroom;
2010}
2011
Tom Herbert8bce6d72014-09-29 20:22:29 -07002012#define ENCAP_TYPE_ETHER 0
2013#define ENCAP_TYPE_IPPROTO 1
2014
2015static inline void skb_set_inner_protocol(struct sk_buff *skb,
2016 __be16 protocol)
2017{
2018 skb->inner_protocol = protocol;
2019 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2020}
2021
2022static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2023 __u8 ipproto)
2024{
2025 skb->inner_ipproto = ipproto;
2026 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2027}
2028
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002029static inline void skb_reset_inner_headers(struct sk_buff *skb)
2030{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002031 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002032 skb->inner_network_header = skb->network_header;
2033 skb->inner_transport_header = skb->transport_header;
2034}
2035
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00002036static inline void skb_reset_mac_len(struct sk_buff *skb)
2037{
2038 skb->mac_len = skb->network_header - skb->mac_header;
2039}
2040
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002041static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2042 *skb)
2043{
2044 return skb->head + skb->inner_transport_header;
2045}
2046
Tom Herbert55dc5a92015-12-14 11:19:40 -08002047static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2048{
2049 return skb_inner_transport_header(skb) - skb->data;
2050}
2051
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002052static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2053{
2054 skb->inner_transport_header = skb->data - skb->head;
2055}
2056
2057static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2058 const int offset)
2059{
2060 skb_reset_inner_transport_header(skb);
2061 skb->inner_transport_header += offset;
2062}
2063
2064static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2065{
2066 return skb->head + skb->inner_network_header;
2067}
2068
2069static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2070{
2071 skb->inner_network_header = skb->data - skb->head;
2072}
2073
2074static inline void skb_set_inner_network_header(struct sk_buff *skb,
2075 const int offset)
2076{
2077 skb_reset_inner_network_header(skb);
2078 skb->inner_network_header += offset;
2079}
2080
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002081static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2082{
2083 return skb->head + skb->inner_mac_header;
2084}
2085
2086static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2087{
2088 skb->inner_mac_header = skb->data - skb->head;
2089}
2090
2091static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2092 const int offset)
2093{
2094 skb_reset_inner_mac_header(skb);
2095 skb->inner_mac_header += offset;
2096}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002097static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2098{
Cong Wang35d04612013-05-29 15:16:05 +08002099 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002100}
2101
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002102static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2103{
2104 return skb->head + skb->transport_header;
2105}
2106
2107static inline void skb_reset_transport_header(struct sk_buff *skb)
2108{
2109 skb->transport_header = skb->data - skb->head;
2110}
2111
2112static inline void skb_set_transport_header(struct sk_buff *skb,
2113 const int offset)
2114{
2115 skb_reset_transport_header(skb);
2116 skb->transport_header += offset;
2117}
2118
2119static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2120{
2121 return skb->head + skb->network_header;
2122}
2123
2124static inline void skb_reset_network_header(struct sk_buff *skb)
2125{
2126 skb->network_header = skb->data - skb->head;
2127}
2128
2129static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2130{
2131 skb_reset_network_header(skb);
2132 skb->network_header += offset;
2133}
2134
2135static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2136{
2137 return skb->head + skb->mac_header;
2138}
2139
2140static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2141{
Cong Wang35d04612013-05-29 15:16:05 +08002142 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002143}
2144
2145static inline void skb_reset_mac_header(struct sk_buff *skb)
2146{
2147 skb->mac_header = skb->data - skb->head;
2148}
2149
2150static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2151{
2152 skb_reset_mac_header(skb);
2153 skb->mac_header += offset;
2154}
2155
Timo Teräs0e3da5b2013-12-16 11:02:09 +02002156static inline void skb_pop_mac_header(struct sk_buff *skb)
2157{
2158 skb->mac_header = skb->network_header;
2159}
2160
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002161static inline void skb_probe_transport_header(struct sk_buff *skb,
2162 const int offset_hint)
2163{
2164 struct flow_keys keys;
2165
2166 if (skb_transport_header_was_set(skb))
2167 return;
Tom Herbertcd79a232015-09-01 09:24:27 -07002168 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
Tom Herbert42aecaa2015-06-04 09:16:39 -07002169 skb_set_transport_header(skb, keys.control.thoff);
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002170 else
2171 skb_set_transport_header(skb, offset_hint);
2172}
2173
Eric Dumazet03606892012-02-23 10:55:02 +00002174static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2175{
2176 if (skb_mac_header_was_set(skb)) {
2177 const unsigned char *old_mac = skb_mac_header(skb);
2178
2179 skb_set_mac_header(skb, -skb->mac_len);
2180 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2181 }
2182}
2183
Michał Mirosław04fb4512010-12-14 15:24:08 +00002184static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2185{
2186 return skb->csum_start - skb_headroom(skb);
2187}
2188
Alexander Duyck08b64fc2016-02-05 15:27:49 -08002189static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2190{
2191 return skb->head + skb->csum_start;
2192}
2193
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002194static inline int skb_transport_offset(const struct sk_buff *skb)
2195{
2196 return skb_transport_header(skb) - skb->data;
2197}
2198
2199static inline u32 skb_network_header_len(const struct sk_buff *skb)
2200{
2201 return skb->transport_header - skb->network_header;
2202}
2203
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002204static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2205{
2206 return skb->inner_transport_header - skb->inner_network_header;
2207}
2208
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002209static inline int skb_network_offset(const struct sk_buff *skb)
2210{
2211 return skb_network_header(skb) - skb->data;
2212}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03002213
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002214static inline int skb_inner_network_offset(const struct sk_buff *skb)
2215{
2216 return skb_inner_network_header(skb) - skb->data;
2217}
2218
Changli Gaof9599ce2010-08-04 04:43:44 +00002219static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2220{
2221 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2222}
2223
Linus Torvalds1da177e2005-04-16 15:20:36 -07002224/*
2225 * CPUs often take a performance hit when accessing unaligned memory
2226 * locations. The actual performance hit varies, it can be small if the
2227 * hardware handles it or large if we have to take an exception and fix it
2228 * in software.
2229 *
2230 * Since an ethernet header is 14 bytes network drivers often end up with
2231 * the IP header at an unaligned offset. The IP header can be aligned by
2232 * shifting the start of the packet by 2 bytes. Drivers should do this
2233 * with:
2234 *
Tobias Klauser8660c122009-07-13 22:48:16 +00002235 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002236 *
2237 * The downside to this alignment of the IP header is that the DMA is now
2238 * unaligned. On some architectures the cost of an unaligned DMA is high
2239 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00002240 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002241 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2242 * to be overridden.
2243 */
2244#ifndef NET_IP_ALIGN
2245#define NET_IP_ALIGN 2
2246#endif
2247
Anton Blanchard025be812006-03-31 02:27:06 -08002248/*
2249 * The networking layer reserves some headroom in skb data (via
2250 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2251 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08002252 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08002253 *
2254 * Unfortunately this headroom changes the DMA alignment of the resulting
2255 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2256 * on some architectures. An architecture can override this value,
2257 * perhaps setting it to a cacheline in size (since that will maintain
2258 * cacheline alignment of the DMA). It must be a power of 2.
2259 *
David S. Millerd6301d32009-02-08 19:24:13 -08002260 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08002261 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07002262 *
2263 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2264 * to reduce average number of cache lines per packet.
2265 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07002266 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08002267 */
2268#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07002269#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08002270#endif
2271
Joe Perches7965bd42013-09-26 14:48:15 -07002272int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002273
2274static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2275{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00002276 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07002277 WARN_ON(1);
2278 return;
2279 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002280 skb->len = len;
2281 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002282}
2283
Joe Perches7965bd42013-09-26 14:48:15 -07002284void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002285
2286static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2287{
Herbert Xu3cc0e872006-06-09 16:13:38 -07002288 if (skb->data_len)
2289 return ___pskb_trim(skb, len);
2290 __skb_trim(skb, len);
2291 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002292}
2293
2294static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2295{
2296 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2297}
2298
2299/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07002300 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2301 * @skb: buffer to alter
2302 * @len: new length
2303 *
2304 * This is identical to pskb_trim except that the caller knows that
2305 * the skb is not cloned so we should never get an error due to out-
2306 * of-memory.
2307 */
2308static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2309{
2310 int err = pskb_trim(skb, len);
2311 BUG_ON(err);
2312}
2313
2314/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002315 * skb_orphan - orphan a buffer
2316 * @skb: buffer to orphan
2317 *
2318 * If a buffer currently has an owner then we call the owner's
2319 * destructor function and make the @skb unowned. The buffer continues
2320 * to exist but is no longer charged to its former owner.
2321 */
2322static inline void skb_orphan(struct sk_buff *skb)
2323{
Eric Dumazetc34a7612013-07-30 16:11:15 -07002324 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002326 skb->destructor = NULL;
2327 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07002328 } else {
2329 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002330 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331}
2332
2333/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002334 * skb_orphan_frags - orphan the frags contained in a buffer
2335 * @skb: buffer to orphan frags from
2336 * @gfp_mask: allocation mask for replacement pages
2337 *
2338 * For each frag in the SKB which needs a destructor (i.e. has an
2339 * owner) create a copy of that frag and release the original
2340 * page by calling the destructor.
2341 */
2342static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2343{
2344 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2345 return 0;
2346 return skb_copy_ubufs(skb, gfp_mask);
2347}
2348
2349/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002350 * __skb_queue_purge - empty a list
2351 * @list: list to empty
2352 *
2353 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2354 * the list and one reference dropped. This function does not take the
2355 * list lock and the caller must hold the relevant locks to use it.
2356 */
Joe Perches7965bd42013-09-26 14:48:15 -07002357void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002358static inline void __skb_queue_purge(struct sk_buff_head *list)
2359{
2360 struct sk_buff *skb;
2361 while ((skb = __skb_dequeue(list)) != NULL)
2362 kfree_skb(skb);
2363}
2364
Joe Perches7965bd42013-09-26 14:48:15 -07002365void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002366
Joe Perches7965bd42013-09-26 14:48:15 -07002367struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2368 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002369
2370/**
2371 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2372 * @dev: network device to receive on
2373 * @length: length to allocate
2374 *
2375 * Allocate a new &sk_buff and assign it a usage count of one. The
2376 * buffer has unspecified headroom built in. Users should allocate
2377 * the headroom they think they need without accounting for the
2378 * built in space. The built in space is used for optimisations.
2379 *
2380 * %NULL is returned if there is no free memory. Although this function
2381 * allocates memory it can be called from an interrupt.
2382 */
2383static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002384 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002385{
2386 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2387}
2388
Eric Dumazet6f532612012-05-18 05:12:12 +00002389/* legacy helper around __netdev_alloc_skb() */
2390static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2391 gfp_t gfp_mask)
2392{
2393 return __netdev_alloc_skb(NULL, length, gfp_mask);
2394}
2395
2396/* legacy helper around netdev_alloc_skb() */
2397static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2398{
2399 return netdev_alloc_skb(NULL, length);
2400}
2401
2402
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002403static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2404 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002405{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002406 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002407
2408 if (NET_IP_ALIGN && skb)
2409 skb_reserve(skb, NET_IP_ALIGN);
2410 return skb;
2411}
2412
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002413static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2414 unsigned int length)
2415{
2416 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2417}
2418
Alexander Duyck181edb22015-05-06 21:12:03 -07002419static inline void skb_free_frag(void *addr)
2420{
2421 __free_page_frag(addr);
2422}
2423
Alexander Duyckffde7322014-12-09 19:40:42 -08002424void *napi_alloc_frag(unsigned int fragsz);
Alexander Duyckfd11a832014-12-09 19:40:49 -08002425struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2426 unsigned int length, gfp_t gfp_mask);
2427static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2428 unsigned int length)
2429{
2430 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2431}
Jesper Dangaard Brouer795bb1c2016-02-08 13:14:59 +01002432void napi_consume_skb(struct sk_buff *skb, int budget);
2433
2434void __kfree_skb_flush(void);
Jesper Dangaard Brouer15fad712016-02-08 13:15:04 +01002435void __kfree_skb_defer(struct sk_buff *skb);
Alexander Duyckffde7322014-12-09 19:40:42 -08002436
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002437/**
Alexander Duyck71dfda52014-11-11 09:26:34 -08002438 * __dev_alloc_pages - allocate page for network Rx
2439 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2440 * @order: size of the allocation
2441 *
2442 * Allocate a new page.
2443 *
2444 * %NULL is returned if there is no free memory.
2445*/
2446static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2447 unsigned int order)
2448{
2449 /* This piece of code contains several assumptions.
2450 * 1. This is for device Rx, therefor a cold page is preferred.
2451 * 2. The expectation is the user wants a compound page.
2452 * 3. If requesting a order 0 page it will not be compound
2453 * due to the check to see if order has a value in prep_new_page
2454 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2455 * code in gfp_to_alloc_flags that should be enforcing this.
2456 */
2457 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2458
2459 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2460}
2461
2462static inline struct page *dev_alloc_pages(unsigned int order)
2463{
2464 return __dev_alloc_pages(GFP_ATOMIC, order);
2465}
2466
2467/**
2468 * __dev_alloc_page - allocate a page for network Rx
2469 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2470 *
2471 * Allocate a new page.
2472 *
2473 * %NULL is returned if there is no free memory.
2474 */
2475static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2476{
2477 return __dev_alloc_pages(gfp_mask, 0);
2478}
2479
2480static inline struct page *dev_alloc_page(void)
2481{
2482 return __dev_alloc_page(GFP_ATOMIC);
2483}
2484
2485/**
Mel Gorman06140022012-07-31 16:44:24 -07002486 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2487 * @page: The page that was allocated from skb_alloc_page
2488 * @skb: The skb that may need pfmemalloc set
2489 */
2490static inline void skb_propagate_pfmemalloc(struct page *page,
2491 struct sk_buff *skb)
2492{
Michal Hocko2f064f32015-08-21 14:11:51 -07002493 if (page_is_pfmemalloc(page))
Mel Gorman06140022012-07-31 16:44:24 -07002494 skb->pfmemalloc = true;
2495}
2496
Eric Dumazet564824b2010-10-11 19:05:25 +00002497/**
Masanari Iidae2278672014-02-18 22:54:36 +09002498 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002499 * @frag: the paged fragment
2500 *
2501 * Returns the &struct page associated with @frag.
2502 */
2503static inline struct page *skb_frag_page(const skb_frag_t *frag)
2504{
Ian Campbella8605c62011-10-19 23:01:49 +00002505 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002506}
2507
2508/**
2509 * __skb_frag_ref - take an addition reference on a paged fragment.
2510 * @frag: the paged fragment
2511 *
2512 * Takes an additional reference on the paged fragment @frag.
2513 */
2514static inline void __skb_frag_ref(skb_frag_t *frag)
2515{
2516 get_page(skb_frag_page(frag));
2517}
2518
2519/**
2520 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2521 * @skb: the buffer
2522 * @f: the fragment offset.
2523 *
2524 * Takes an additional reference on the @f'th paged fragment of @skb.
2525 */
2526static inline void skb_frag_ref(struct sk_buff *skb, int f)
2527{
2528 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2529}
2530
2531/**
2532 * __skb_frag_unref - release a reference on a paged fragment.
2533 * @frag: the paged fragment
2534 *
2535 * Releases a reference on the paged fragment @frag.
2536 */
2537static inline void __skb_frag_unref(skb_frag_t *frag)
2538{
2539 put_page(skb_frag_page(frag));
2540}
2541
2542/**
2543 * skb_frag_unref - release a reference on a paged fragment of an skb.
2544 * @skb: the buffer
2545 * @f: the fragment offset
2546 *
2547 * Releases a reference on the @f'th paged fragment of @skb.
2548 */
2549static inline void skb_frag_unref(struct sk_buff *skb, int f)
2550{
2551 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2552}
2553
2554/**
2555 * skb_frag_address - gets the address of the data contained in a paged fragment
2556 * @frag: the paged fragment buffer
2557 *
2558 * Returns the address of the data within @frag. The page must already
2559 * be mapped.
2560 */
2561static inline void *skb_frag_address(const skb_frag_t *frag)
2562{
2563 return page_address(skb_frag_page(frag)) + frag->page_offset;
2564}
2565
2566/**
2567 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2568 * @frag: the paged fragment buffer
2569 *
2570 * Returns the address of the data within @frag. Checks that the page
2571 * is mapped and returns %NULL otherwise.
2572 */
2573static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2574{
2575 void *ptr = page_address(skb_frag_page(frag));
2576 if (unlikely(!ptr))
2577 return NULL;
2578
2579 return ptr + frag->page_offset;
2580}
2581
2582/**
2583 * __skb_frag_set_page - sets the page contained in a paged fragment
2584 * @frag: the paged fragment
2585 * @page: the page to set
2586 *
2587 * Sets the fragment @frag to contain @page.
2588 */
2589static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2590{
Ian Campbella8605c62011-10-19 23:01:49 +00002591 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002592}
2593
2594/**
2595 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2596 * @skb: the buffer
2597 * @f: the fragment offset
2598 * @page: the page to set
2599 *
2600 * Sets the @f'th fragment of @skb to contain @page.
2601 */
2602static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2603 struct page *page)
2604{
2605 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2606}
2607
Eric Dumazet400dfd32013-10-17 16:27:07 -07002608bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2609
Ian Campbell131ea662011-08-19 06:25:00 +00002610/**
2611 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002612 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002613 * @frag: the paged fragment to map
2614 * @offset: the offset within the fragment (starting at the
2615 * fragment's own offset)
2616 * @size: the number of bytes to map
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002617 * @dir: the direction of the mapping (%PCI_DMA_*)
Ian Campbell131ea662011-08-19 06:25:00 +00002618 *
2619 * Maps the page associated with @frag to @device.
2620 */
2621static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2622 const skb_frag_t *frag,
2623 size_t offset, size_t size,
2624 enum dma_data_direction dir)
2625{
2626 return dma_map_page(dev, skb_frag_page(frag),
2627 frag->page_offset + offset, size, dir);
2628}
2629
Eric Dumazet117632e2011-12-03 21:39:53 +00002630static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2631 gfp_t gfp_mask)
2632{
2633 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2634}
2635
Octavian Purdilabad93e92014-06-12 01:36:26 +03002636
2637static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2638 gfp_t gfp_mask)
2639{
2640 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2641}
2642
2643
Ian Campbell131ea662011-08-19 06:25:00 +00002644/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002645 * skb_clone_writable - is the header of a clone writable
2646 * @skb: buffer to check
2647 * @len: length up to which to write
2648 *
2649 * Returns true if modifying the header part of the cloned buffer
2650 * does not requires the data to be copied.
2651 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002652static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002653{
2654 return !skb_header_cloned(skb) &&
2655 skb_headroom(skb) + len <= skb->hdr_len;
2656}
2657
Daniel Borkmann36976492016-02-19 23:05:25 +01002658static inline int skb_try_make_writable(struct sk_buff *skb,
2659 unsigned int write_len)
2660{
2661 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2662 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2663}
2664
Herbert Xud9cc2042007-09-16 16:21:16 -07002665static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2666 int cloned)
2667{
2668 int delta = 0;
2669
Herbert Xud9cc2042007-09-16 16:21:16 -07002670 if (headroom > skb_headroom(skb))
2671 delta = headroom - skb_headroom(skb);
2672
2673 if (delta || cloned)
2674 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2675 GFP_ATOMIC);
2676 return 0;
2677}
2678
Patrick McHardy334a8132007-06-25 04:35:20 -07002679/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002680 * skb_cow - copy header of skb when it is required
2681 * @skb: buffer to cow
2682 * @headroom: needed headroom
2683 *
2684 * If the skb passed lacks sufficient headroom or its data part
2685 * is shared, data is reallocated. If reallocation fails, an error
2686 * is returned and original skb is not changed.
2687 *
2688 * The result is skb with writable area skb->head...skb->tail
2689 * and at least @headroom of space at head.
2690 */
2691static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2692{
Herbert Xud9cc2042007-09-16 16:21:16 -07002693 return __skb_cow(skb, headroom, skb_cloned(skb));
2694}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002695
Herbert Xud9cc2042007-09-16 16:21:16 -07002696/**
2697 * skb_cow_head - skb_cow but only making the head writable
2698 * @skb: buffer to cow
2699 * @headroom: needed headroom
2700 *
2701 * This function is identical to skb_cow except that we replace the
2702 * skb_cloned check by skb_header_cloned. It should be used when
2703 * you only need to push on some header and do not need to modify
2704 * the data.
2705 */
2706static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2707{
2708 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002709}
2710
2711/**
2712 * skb_padto - pad an skbuff up to a minimal size
2713 * @skb: buffer to pad
2714 * @len: minimal length
2715 *
2716 * Pads up a buffer to ensure the trailing bytes exist and are
2717 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002718 * is untouched. Otherwise it is extended. Returns zero on
2719 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720 */
Herbert Xu5b057c62006-06-23 02:06:41 -07002721static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002722{
2723 unsigned int size = skb->len;
2724 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002725 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002726 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002727}
2728
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002729/**
2730 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2731 * @skb: buffer to pad
2732 * @len: minimal length
2733 *
2734 * Pads up a buffer to ensure the trailing bytes exist and are
2735 * blanked. If the buffer already contains sufficient data it
2736 * is untouched. Otherwise it is extended. Returns zero on
2737 * success. The skb is freed on error.
2738 */
2739static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2740{
2741 unsigned int size = skb->len;
2742
2743 if (unlikely(size < len)) {
2744 len -= size;
2745 if (skb_pad(skb, len))
2746 return -ENOMEM;
2747 __skb_put(skb, len);
2748 }
2749 return 0;
2750}
2751
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752static inline int skb_add_data(struct sk_buff *skb,
Al Viroaf2b0402014-11-27 21:44:24 -05002753 struct iov_iter *from, int copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002754{
2755 const int off = skb->len;
2756
2757 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viroaf2b0402014-11-27 21:44:24 -05002758 __wsum csum = 0;
2759 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2760 &csum, from) == copy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002761 skb->csum = csum_block_add(skb->csum, csum, off);
2762 return 0;
2763 }
Al Viroaf2b0402014-11-27 21:44:24 -05002764 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002765 return 0;
2766
2767 __skb_trim(skb, off);
2768 return -EFAULT;
2769}
2770
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002771static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2772 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002773{
2774 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002775 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002776
Ian Campbellea2ab692011-08-22 23:44:58 +00002777 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00002778 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002780 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002781}
2782
Herbert Xu364c6ba2006-06-09 16:10:40 -07002783static inline int __skb_linearize(struct sk_buff *skb)
2784{
2785 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2786}
2787
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788/**
2789 * skb_linearize - convert paged skb to linear one
2790 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07002791 *
2792 * If there is no free memory -ENOMEM is returned, otherwise zero
2793 * is returned and the old skb data released.
2794 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07002795static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002796{
Herbert Xu364c6ba2006-06-09 16:10:40 -07002797 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2798}
2799
2800/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00002801 * skb_has_shared_frag - can any frag be overwritten
2802 * @skb: buffer to test
2803 *
2804 * Return true if the skb has at least one frag that might be modified
2805 * by an external entity (as in vmsplice()/sendfile())
2806 */
2807static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2808{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00002809 return skb_is_nonlinear(skb) &&
2810 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002811}
2812
2813/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07002814 * skb_linearize_cow - make sure skb is linear and writable
2815 * @skb: buffer to process
2816 *
2817 * If there is no free memory -ENOMEM is returned, otherwise zero
2818 * is returned and the old skb data released.
2819 */
2820static inline int skb_linearize_cow(struct sk_buff *skb)
2821{
2822 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2823 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002824}
2825
2826/**
2827 * skb_postpull_rcsum - update checksum for received skb after pull
2828 * @skb: buffer to update
2829 * @start: start of data before pull
2830 * @len: length of data pulled
2831 *
2832 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07002833 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2834 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002835 */
2836
2837static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f2006-03-20 22:43:56 -08002838 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839{
Patrick McHardy84fa7932006-08-29 16:44:56 -07002840 if (skb->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002842 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
Pravin B Shelar31b33df2015-09-28 17:24:25 -07002843 skb_checksum_start_offset(skb) < 0)
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002844 skb->ip_summed = CHECKSUM_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845}
2846
Herbert Xucbb042f2006-03-20 22:43:56 -08002847unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2848
Daniel Borkmannf8ffad62016-01-07 15:50:23 +01002849static inline void skb_postpush_rcsum(struct sk_buff *skb,
2850 const void *start, unsigned int len)
2851{
2852 /* For performing the reverse operation to skb_postpull_rcsum(),
2853 * we can instead of ...
2854 *
2855 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0));
2856 *
2857 * ... just use this equivalent version here to save a few
2858 * instructions. Feeding csum of 0 in csum_partial() and later
2859 * on adding skb->csum is equivalent to feed skb->csum in the
2860 * first place.
2861 */
2862 if (skb->ip_summed == CHECKSUM_COMPLETE)
2863 skb->csum = csum_partial(start, len, skb->csum);
2864}
2865
David S. Miller7ce5a272013-12-02 17:26:05 -05002866/**
2867 * pskb_trim_rcsum - trim received skb and update checksum
2868 * @skb: buffer to trim
2869 * @len: new length
2870 *
2871 * This is exactly the same as pskb_trim except that it ensures the
2872 * checksum of received packets are still valid after the operation.
2873 */
2874
2875static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2876{
2877 if (likely(len >= skb->len))
2878 return 0;
2879 if (skb->ip_summed == CHECKSUM_COMPLETE)
2880 skb->ip_summed = CHECKSUM_NONE;
2881 return __pskb_trim(skb, len);
2882}
2883
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884#define skb_queue_walk(queue, skb) \
2885 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002886 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002887 skb = skb->next)
2888
James Chapman46f89142007-04-30 00:07:31 -07002889#define skb_queue_walk_safe(queue, skb, tmp) \
2890 for (skb = (queue)->next, tmp = skb->next; \
2891 skb != (struct sk_buff *)(queue); \
2892 skb = tmp, tmp = skb->next)
2893
David S. Miller1164f522008-09-23 00:49:44 -07002894#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002895 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07002896 skb = skb->next)
2897
2898#define skb_queue_walk_from_safe(queue, skb, tmp) \
2899 for (tmp = skb->next; \
2900 skb != (struct sk_buff *)(queue); \
2901 skb = tmp, tmp = skb->next)
2902
Stephen Hemminger300ce172005-10-30 13:47:34 -08002903#define skb_queue_reverse_walk(queue, skb) \
2904 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002905 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08002906 skb = skb->prev)
2907
David S. Miller686a2952011-01-20 22:47:32 -08002908#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2909 for (skb = (queue)->prev, tmp = skb->prev; \
2910 skb != (struct sk_buff *)(queue); \
2911 skb = tmp, tmp = skb->prev)
2912
2913#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2914 for (tmp = skb->prev; \
2915 skb != (struct sk_buff *)(queue); \
2916 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002917
David S. Miller21dc3302010-08-23 00:13:46 -07002918static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07002919{
2920 return skb_shinfo(skb)->frag_list != NULL;
2921}
2922
2923static inline void skb_frag_list_init(struct sk_buff *skb)
2924{
2925 skb_shinfo(skb)->frag_list = NULL;
2926}
2927
David S. Milleree039872009-06-09 00:17:13 -07002928#define skb_walk_frags(skb, iter) \
2929 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2930
Rainer Weikusatea3793e2015-12-06 21:11:34 +00002931
2932int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
2933 const struct sk_buff *skb);
2934struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
2935 int *peeked, int *off, int *err,
2936 struct sk_buff **last);
Joe Perches7965bd42013-09-26 14:48:15 -07002937struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2938 int *peeked, int *off, int *err);
2939struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2940 int *err);
2941unsigned int datagram_poll(struct file *file, struct socket *sock,
2942 struct poll_table_struct *wait);
Al Viroc0371da2014-11-24 10:42:55 -05002943int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2944 struct iov_iter *to, int size);
David S. Miller51f3d022014-11-05 16:46:40 -05002945static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2946 struct msghdr *msg, int size)
2947{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002948 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
David S. Miller51f3d022014-11-05 16:46:40 -05002949}
Al Viroe5a4b0b2014-11-24 18:17:55 -05002950int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2951 struct msghdr *msg);
Al Viro3a654f92014-06-19 14:15:22 -04002952int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2953 struct iov_iter *from, int len);
Al Viro3a654f92014-06-19 14:15:22 -04002954int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
Joe Perches7965bd42013-09-26 14:48:15 -07002955void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
samanthakumar627d2d62016-04-05 12:41:16 -04002956void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
2957static inline void skb_free_datagram_locked(struct sock *sk,
2958 struct sk_buff *skb)
2959{
2960 __skb_free_datagram_locked(sk, skb, 0);
2961}
Joe Perches7965bd42013-09-26 14:48:15 -07002962int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07002963int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2964int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2965__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2966 int len, __wsum csum);
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002967ssize_t skb_socket_splice(struct sock *sk,
2968 struct pipe_inode_info *pipe,
2969 struct splice_pipe_desc *spd);
2970int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
Joe Perches7965bd42013-09-26 14:48:15 -07002971 struct pipe_inode_info *pipe, unsigned int len,
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002972 unsigned int flags,
2973 ssize_t (*splice_cb)(struct sock *,
2974 struct pipe_inode_info *,
2975 struct splice_pipe_desc *));
Joe Perches7965bd42013-09-26 14:48:15 -07002976void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01002977unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002978int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2979 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07002980void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2981int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2982void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01002983unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -07002984struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Vlad Yasevich0d5501c2014-08-08 14:42:13 -04002985struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
Jiri Pirkoe2195122014-11-19 14:05:01 +01002986int skb_ensure_writable(struct sk_buff *skb, int write_len);
Jiri Pirko93515d52014-11-19 14:05:02 +01002987int skb_vlan_pop(struct sk_buff *skb);
2988int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
Sowmini Varadhan6fa01cc2016-04-22 18:36:35 -07002989struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
2990 gfp_t gfp);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002991
Al Viro6ce8e9c2014-04-06 21:25:44 -04002992static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2993{
Al Viro21226ab2014-11-28 15:48:29 -05002994 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro6ce8e9c2014-04-06 21:25:44 -04002995}
2996
Al Viro7eab8d92014-04-06 21:51:23 -04002997static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2998{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002999 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro7eab8d92014-04-06 21:51:23 -04003000}
3001
Daniel Borkmann2817a332013-10-30 11:50:51 +01003002struct skb_checksum_ops {
3003 __wsum (*update)(const void *mem, int len, __wsum wsum);
3004 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3005};
3006
3007__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3008 __wsum csum, const struct skb_checksum_ops *ops);
3009__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3010 __wsum csum);
3011
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003012static inline void * __must_check
3013__skb_header_pointer(const struct sk_buff *skb, int offset,
3014 int len, void *data, int hlen, void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003015{
Patrick McHardy55820ee2005-07-05 14:08:10 -07003016 if (hlen - offset >= len)
David S. Miller690e36e2014-08-23 12:13:41 -07003017 return data + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018
David S. Miller690e36e2014-08-23 12:13:41 -07003019 if (!skb ||
3020 skb_copy_bits(skb, offset, buffer, len) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 return NULL;
3022
3023 return buffer;
3024}
3025
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003026static inline void * __must_check
3027skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
David S. Miller690e36e2014-08-23 12:13:41 -07003028{
3029 return __skb_header_pointer(skb, offset, len, skb->data,
3030 skb_headlen(skb), buffer);
3031}
3032
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01003033/**
3034 * skb_needs_linearize - check if we need to linearize a given skb
3035 * depending on the given device features.
3036 * @skb: socket buffer to check
3037 * @features: net device features
3038 *
3039 * Returns true if either:
3040 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3041 * 2. skb is fragmented and the device does not support SG.
3042 */
3043static inline bool skb_needs_linearize(struct sk_buff *skb,
3044 netdev_features_t features)
3045{
3046 return skb_is_nonlinear(skb) &&
3047 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3048 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3049}
3050
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03003051static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3052 void *to,
3053 const unsigned int len)
3054{
3055 memcpy(to, skb->data, len);
3056}
3057
3058static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3059 const int offset, void *to,
3060 const unsigned int len)
3061{
3062 memcpy(to, skb->data + offset, len);
3063}
3064
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03003065static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3066 const void *from,
3067 const unsigned int len)
3068{
3069 memcpy(skb->data, from, len);
3070}
3071
3072static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3073 const int offset,
3074 const void *from,
3075 const unsigned int len)
3076{
3077 memcpy(skb->data + offset, from, len);
3078}
3079
Joe Perches7965bd42013-09-26 14:48:15 -07003080void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003081
Patrick Ohlyac45f602009-02-12 05:03:37 +00003082static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3083{
3084 return skb->tstamp;
3085}
3086
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003087/**
3088 * skb_get_timestamp - get timestamp from a skb
3089 * @skb: skb to get stamp from
3090 * @stamp: pointer to struct timeval to store stamp in
3091 *
3092 * Timestamps are stored in the skb as offsets to a base timestamp.
3093 * This function converts the offset back to a struct timeval and stores
3094 * it in stamp.
3095 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00003096static inline void skb_get_timestamp(const struct sk_buff *skb,
3097 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003098{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003099 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003100}
3101
Patrick Ohlyac45f602009-02-12 05:03:37 +00003102static inline void skb_get_timestampns(const struct sk_buff *skb,
3103 struct timespec *stamp)
3104{
3105 *stamp = ktime_to_timespec(skb->tstamp);
3106}
3107
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003108static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003109{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003110 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003111}
3112
Stephen Hemminger164891a2007-04-23 22:26:16 -07003113static inline ktime_t net_timedelta(ktime_t t)
3114{
3115 return ktime_sub(ktime_get_real(), t);
3116}
3117
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07003118static inline ktime_t net_invalid_timestamp(void)
3119{
3120 return ktime_set(0, 0);
3121}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003122
Alexander Duyck62bccb82014-09-04 13:31:35 -04003123struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3124
Richard Cochranc1f19b52010-07-17 08:49:36 +00003125#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3126
Joe Perches7965bd42013-09-26 14:48:15 -07003127void skb_clone_tx_timestamp(struct sk_buff *skb);
3128bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00003129
3130#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3131
3132static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3133{
3134}
3135
3136static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3137{
3138 return false;
3139}
3140
3141#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3142
3143/**
3144 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3145 *
Richard Cochranda92b192011-10-21 00:49:15 +00003146 * PHY drivers may accept clones of transmitted packets for
3147 * timestamping via their phy_driver.txtstamp method. These drivers
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003148 * must call this function to return the skb back to the stack with a
3149 * timestamp.
Richard Cochranda92b192011-10-21 00:49:15 +00003150 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00003151 * @skb: clone of the the original outgoing packet
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003152 * @hwtstamps: hardware time stamps
Richard Cochranc1f19b52010-07-17 08:49:36 +00003153 *
3154 */
3155void skb_complete_tx_timestamp(struct sk_buff *skb,
3156 struct skb_shared_hwtstamps *hwtstamps);
3157
Willem de Bruijne7fd2882014-08-04 22:11:48 -04003158void __skb_tstamp_tx(struct sk_buff *orig_skb,
3159 struct skb_shared_hwtstamps *hwtstamps,
3160 struct sock *sk, int tstype);
3161
Patrick Ohlyac45f602009-02-12 05:03:37 +00003162/**
3163 * skb_tstamp_tx - queue clone of skb with send time stamps
3164 * @orig_skb: the original outgoing packet
3165 * @hwtstamps: hardware time stamps, may be NULL if not available
3166 *
3167 * If the skb has a socket associated, then this function clones the
3168 * skb (thus sharing the actual data and optional structures), stores
3169 * the optional hardware time stamping information (if non NULL) or
3170 * generates a software time stamp (otherwise), then queues the clone
3171 * to the error queue of the socket. Errors are silently ignored.
3172 */
Joe Perches7965bd42013-09-26 14:48:15 -07003173void skb_tstamp_tx(struct sk_buff *orig_skb,
3174 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00003175
Richard Cochran4507a712010-07-17 08:48:28 +00003176static inline void sw_tx_timestamp(struct sk_buff *skb)
3177{
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003178 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3179 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
Richard Cochran4507a712010-07-17 08:48:28 +00003180 skb_tstamp_tx(skb, NULL);
3181}
3182
3183/**
3184 * skb_tx_timestamp() - Driver hook for transmit timestamping
3185 *
3186 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00003187 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00003188 *
David S. Miller73409f32013-12-27 13:04:33 -05003189 * Specifically, one should make absolutely sure that this function is
3190 * called before TX completion of this packet can trigger. Otherwise
3191 * the packet could potentially already be freed.
3192 *
Richard Cochran4507a712010-07-17 08:48:28 +00003193 * @skb: A socket buffer.
3194 */
3195static inline void skb_tx_timestamp(struct sk_buff *skb)
3196{
Richard Cochranc1f19b52010-07-17 08:49:36 +00003197 skb_clone_tx_timestamp(skb);
Richard Cochran4507a712010-07-17 08:48:28 +00003198 sw_tx_timestamp(skb);
3199}
3200
Johannes Berg6e3e9392011-11-09 10:15:42 +01003201/**
3202 * skb_complete_wifi_ack - deliver skb with wifi status
3203 *
3204 * @skb: the original outgoing packet
3205 * @acked: ack status
3206 *
3207 */
3208void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3209
Joe Perches7965bd42013-09-26 14:48:15 -07003210__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3211__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003212
Herbert Xu60476372007-04-09 11:59:39 -07003213static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3214{
Tom Herbert6edec0e2015-02-10 16:30:28 -08003215 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3216 skb->csum_valid ||
3217 (skb->ip_summed == CHECKSUM_PARTIAL &&
3218 skb_checksum_start_offset(skb) >= 0));
Herbert Xu60476372007-04-09 11:59:39 -07003219}
3220
Herbert Xufb286bb2005-11-10 13:01:24 -08003221/**
3222 * skb_checksum_complete - Calculate checksum of an entire packet
3223 * @skb: packet to process
3224 *
3225 * This function calculates the checksum over the entire packet plus
3226 * the value of skb->csum. The latter can be used to supply the
3227 * checksum of a pseudo header as used by TCP/UDP. It returns the
3228 * checksum.
3229 *
3230 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3231 * this function can be used to verify that checksum on received
3232 * packets. In that case the function should return zero if the
3233 * checksum is correct. In particular, this function will return zero
3234 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3235 * hardware has already verified the correctness of the checksum.
3236 */
Al Viro4381ca32007-07-15 21:00:11 +01003237static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08003238{
Herbert Xu60476372007-04-09 11:59:39 -07003239 return skb_csum_unnecessary(skb) ?
3240 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003241}
3242
Tom Herbert77cffe22014-08-27 21:26:46 -07003243static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3244{
3245 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3246 if (skb->csum_level == 0)
3247 skb->ip_summed = CHECKSUM_NONE;
3248 else
3249 skb->csum_level--;
3250 }
3251}
3252
3253static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3254{
3255 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3256 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3257 skb->csum_level++;
3258 } else if (skb->ip_summed == CHECKSUM_NONE) {
3259 skb->ip_summed = CHECKSUM_UNNECESSARY;
3260 skb->csum_level = 0;
3261 }
3262}
3263
Tom Herbert5a212322014-08-31 15:12:41 -07003264static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3265{
3266 /* Mark current checksum as bad (typically called from GRO
3267 * path). In the case that ip_summed is CHECKSUM_NONE
3268 * this must be the first checksum encountered in the packet.
3269 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3270 * checksum after the last one validated. For UDP, a zero
3271 * checksum can not be marked as bad.
3272 */
3273
3274 if (skb->ip_summed == CHECKSUM_NONE ||
3275 skb->ip_summed == CHECKSUM_UNNECESSARY)
3276 skb->csum_bad = 1;
3277}
3278
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003279/* Check if we need to perform checksum complete validation.
3280 *
3281 * Returns true if checksum complete is needed, false otherwise
3282 * (either checksum is unnecessary or zero checksum is allowed).
3283 */
3284static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3285 bool zero_okay,
3286 __sum16 check)
3287{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003288 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3289 skb->csum_valid = 1;
Tom Herbert77cffe22014-08-27 21:26:46 -07003290 __skb_decr_checksum_unnecessary(skb);
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003291 return false;
3292 }
3293
3294 return true;
3295}
3296
3297/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3298 * in checksum_init.
3299 */
3300#define CHECKSUM_BREAK 76
3301
Tom Herbert4e18b9a2015-04-20 14:10:04 -07003302/* Unset checksum-complete
3303 *
3304 * Unset checksum complete can be done when packet is being modified
3305 * (uncompressed for instance) and checksum-complete value is
3306 * invalidated.
3307 */
3308static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3309{
3310 if (skb->ip_summed == CHECKSUM_COMPLETE)
3311 skb->ip_summed = CHECKSUM_NONE;
3312}
3313
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003314/* Validate (init) checksum based on checksum complete.
3315 *
3316 * Return values:
3317 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3318 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3319 * checksum is stored in skb->csum for use in __skb_checksum_complete
3320 * non-zero: value of invalid checksum
3321 *
3322 */
3323static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3324 bool complete,
3325 __wsum psum)
3326{
3327 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3328 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003329 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003330 return 0;
3331 }
Tom Herbert5a212322014-08-31 15:12:41 -07003332 } else if (skb->csum_bad) {
3333 /* ip_summed == CHECKSUM_NONE in this case */
Eric Dumazetc91d4602015-05-15 05:48:07 -07003334 return (__force __sum16)1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003335 }
3336
3337 skb->csum = psum;
3338
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003339 if (complete || skb->len <= CHECKSUM_BREAK) {
3340 __sum16 csum;
3341
3342 csum = __skb_checksum_complete(skb);
3343 skb->csum_valid = !csum;
3344 return csum;
3345 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003346
3347 return 0;
3348}
3349
3350static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3351{
3352 return 0;
3353}
3354
3355/* Perform checksum validate (init). Note that this is a macro since we only
3356 * want to calculate the pseudo header which is an input function if necessary.
3357 * First we try to validate without any computation (checksum unnecessary) and
3358 * then calculate based on checksum complete calling the function to compute
3359 * pseudo header.
3360 *
3361 * Return values:
3362 * 0: checksum is validated or try to in skb_checksum_complete
3363 * non-zero: value of invalid checksum
3364 */
3365#define __skb_checksum_validate(skb, proto, complete, \
3366 zero_okay, check, compute_pseudo) \
3367({ \
3368 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003369 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003370 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3371 __ret = __skb_checksum_validate_complete(skb, \
3372 complete, compute_pseudo(skb, proto)); \
3373 __ret; \
3374})
3375
3376#define skb_checksum_init(skb, proto, compute_pseudo) \
3377 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3378
3379#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3380 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3381
3382#define skb_checksum_validate(skb, proto, compute_pseudo) \
3383 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3384
3385#define skb_checksum_validate_zero_check(skb, proto, check, \
3386 compute_pseudo) \
Sabrina Dubroca096a4cf2015-02-06 18:54:19 +01003387 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003388
3389#define skb_checksum_simple_validate(skb) \
3390 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3391
Tom Herbertd96535a2014-08-31 15:12:42 -07003392static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3393{
3394 return (skb->ip_summed == CHECKSUM_NONE &&
3395 skb->csum_valid && !skb->csum_bad);
3396}
3397
3398static inline void __skb_checksum_convert(struct sk_buff *skb,
3399 __sum16 check, __wsum pseudo)
3400{
3401 skb->csum = ~pseudo;
3402 skb->ip_summed = CHECKSUM_COMPLETE;
3403}
3404
3405#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3406do { \
3407 if (__skb_checksum_convert_check(skb)) \
3408 __skb_checksum_convert(skb, check, \
3409 compute_pseudo(skb, proto)); \
3410} while (0)
3411
Tom Herbert15e23962015-02-10 16:30:31 -08003412static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3413 u16 start, u16 offset)
3414{
3415 skb->ip_summed = CHECKSUM_PARTIAL;
3416 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3417 skb->csum_offset = offset - start;
3418}
3419
Tom Herbertdcdc8992015-02-02 16:07:34 -08003420/* Update skbuf and packet to reflect the remote checksum offload operation.
3421 * When called, ptr indicates the starting point for skb->csum when
3422 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3423 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3424 */
3425static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
Tom Herbert15e23962015-02-10 16:30:31 -08003426 int start, int offset, bool nopartial)
Tom Herbertdcdc8992015-02-02 16:07:34 -08003427{
3428 __wsum delta;
3429
Tom Herbert15e23962015-02-10 16:30:31 -08003430 if (!nopartial) {
3431 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3432 return;
3433 }
3434
Tom Herbertdcdc8992015-02-02 16:07:34 -08003435 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3436 __skb_checksum_complete(skb);
3437 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3438 }
3439
3440 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3441
3442 /* Adjust skb->csum since we changed the packet */
3443 skb->csum = csum_add(skb->csum, delta);
3444}
3445
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003446#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07003447void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3449{
3450 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07003451 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003452}
3453static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3454{
3455 if (nfct)
3456 atomic_inc(&nfct->use);
3457}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003458#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003459#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003460static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3461{
3462 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3463 kfree(nf_bridge);
3464}
3465static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3466{
3467 if (nf_bridge)
3468 atomic_inc(&nf_bridge->use);
3469}
3470#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08003471static inline void nf_reset(struct sk_buff *skb)
3472{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003473#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003474 nf_conntrack_put(skb->nfct);
3475 skb->nfct = NULL;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003476#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003477#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003478 nf_bridge_put(skb->nf_bridge);
3479 skb->nf_bridge = NULL;
3480#endif
3481}
3482
Patrick McHardy124dff02013-04-05 20:42:05 +02003483static inline void nf_reset_trace(struct sk_buff *skb)
3484{
Florian Westphal478b3602014-02-15 23:48:45 +01003485#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00003486 skb->nf_trace = 0;
3487#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003488}
3489
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003490/* Note: This doesn't put any conntrack and bridge info in dst. */
Eric Dumazetb1937222014-09-28 22:18:47 -07003491static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3492 bool copy)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003493{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003494#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003495 dst->nfct = src->nfct;
3496 nf_conntrack_get(src->nfct);
Eric Dumazetb1937222014-09-28 22:18:47 -07003497 if (copy)
3498 dst->nfctinfo = src->nfctinfo;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003499#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003500#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003501 dst->nf_bridge = src->nf_bridge;
3502 nf_bridge_get(src->nf_bridge);
3503#endif
Florian Westphal478b3602014-02-15 23:48:45 +01003504#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Eric Dumazetb1937222014-09-28 22:18:47 -07003505 if (copy)
3506 dst->nf_trace = src->nf_trace;
Florian Westphal478b3602014-02-15 23:48:45 +01003507#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003508}
3509
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003510static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3511{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003512#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003513 nf_conntrack_put(dst->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003514#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003515#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003516 nf_bridge_put(dst->nf_bridge);
3517#endif
Eric Dumazetb1937222014-09-28 22:18:47 -07003518 __nf_copy(dst, src, true);
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003519}
3520
James Morris984bc162006-06-09 00:29:17 -07003521#ifdef CONFIG_NETWORK_SECMARK
3522static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3523{
3524 to->secmark = from->secmark;
3525}
3526
3527static inline void skb_init_secmark(struct sk_buff *skb)
3528{
3529 skb->secmark = 0;
3530}
3531#else
3532static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3533{ }
3534
3535static inline void skb_init_secmark(struct sk_buff *skb)
3536{ }
3537#endif
3538
Eric W. Biederman574f7192014-04-01 12:20:24 -07003539static inline bool skb_irq_freeable(const struct sk_buff *skb)
3540{
3541 return !skb->destructor &&
3542#if IS_ENABLED(CONFIG_XFRM)
3543 !skb->sp &&
3544#endif
3545#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3546 !skb->nfct &&
3547#endif
3548 !skb->_skb_refdst &&
3549 !skb_has_frag_list(skb);
3550}
3551
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003552static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3553{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003554 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003555}
3556
Stephen Hemminger92477442009-03-21 13:39:26 -07003557static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003558{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003559 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003560}
3561
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003562static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3563{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003564 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003565}
3566
David S. Millerd5a9e242009-01-27 16:22:11 -08003567static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3568{
3569 skb->queue_mapping = rx_queue + 1;
3570}
3571
Stephen Hemminger92477442009-03-21 13:39:26 -07003572static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003573{
3574 return skb->queue_mapping - 1;
3575}
3576
Stephen Hemminger92477442009-03-21 13:39:26 -07003577static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003578{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003579 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003580}
3581
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003582static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3583{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003584#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003585 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003586#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003587 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003588#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003589}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003590
Pravin B Shelar68c33162013-02-14 14:02:41 +00003591/* Keeps track of mac header offset relative to skb->head.
3592 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3593 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003594 * tunnel skb it points to outer mac header.
3595 * Keeps track of level of encapsulation of network headers.
3596 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003597struct skb_gso_cb {
Alexander Duyck802ab552016-04-10 21:45:03 -04003598 union {
3599 int mac_offset;
3600 int data_offset;
3601 };
Eric Dumazet3347c962013-10-19 11:42:56 -07003602 int encap_level;
Alexander Duyck76443452016-02-05 15:27:37 -08003603 __wsum csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003604 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003605};
Konstantin Khlebnikov9207f9d2016-01-08 15:21:46 +03003606#define SKB_SGO_CB_OFFSET 32
3607#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
Pravin B Shelar68c33162013-02-14 14:02:41 +00003608
3609static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3610{
3611 return (skb_mac_header(inner_skb) - inner_skb->head) -
3612 SKB_GSO_CB(inner_skb)->mac_offset;
3613}
3614
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003615static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3616{
3617 int new_headroom, headroom;
3618 int ret;
3619
3620 headroom = skb_headroom(skb);
3621 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3622 if (ret)
3623 return ret;
3624
3625 new_headroom = skb_headroom(skb);
3626 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3627 return 0;
3628}
3629
Alexander Duyck08b64fc2016-02-05 15:27:49 -08003630static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3631{
3632 /* Do not update partial checksums if remote checksum is enabled. */
3633 if (skb->remcsum_offload)
3634 return;
3635
3636 SKB_GSO_CB(skb)->csum = res;
3637 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3638}
3639
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003640/* Compute the checksum for a gso segment. First compute the checksum value
3641 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3642 * then add in skb->csum (checksum from csum_start to end of packet).
3643 * skb->csum and csum_start are then updated to reflect the checksum of the
3644 * resultant packet starting from the transport header-- the resultant checksum
3645 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3646 * header.
3647 */
3648static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3649{
Alexander Duyck76443452016-02-05 15:27:37 -08003650 unsigned char *csum_start = skb_transport_header(skb);
3651 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3652 __wsum partial = SKB_GSO_CB(skb)->csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003653
Alexander Duyck76443452016-02-05 15:27:37 -08003654 SKB_GSO_CB(skb)->csum = res;
3655 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003656
Alexander Duyck76443452016-02-05 15:27:37 -08003657 return csum_fold(csum_partial(csum_start, plen, partial));
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003658}
3659
David S. Millerbdcc0922012-03-07 20:53:36 -05003660static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07003661{
3662 return skb_shinfo(skb)->gso_size;
3663}
3664
Eric Dumazet36a8f392013-09-29 01:21:32 -07003665/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05003666static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02003667{
3668 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3669}
3670
Joe Perches7965bd42013-09-26 14:48:15 -07003671void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07003672
3673static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3674{
3675 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3676 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003677 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3678
Alexander Duyckb78462e2010-06-02 12:24:37 +00003679 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3680 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07003681 __skb_warn_lro_forwarding(skb);
3682 return true;
3683 }
3684 return false;
3685}
3686
Herbert Xu35fc92a2007-03-26 23:22:20 -07003687static inline void skb_forward_csum(struct sk_buff *skb)
3688{
3689 /* Unfortunately we don't support this one. Any brave souls? */
3690 if (skb->ip_summed == CHECKSUM_COMPLETE)
3691 skb->ip_summed = CHECKSUM_NONE;
3692}
3693
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003694/**
3695 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3696 * @skb: skb to check
3697 *
3698 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3699 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3700 * use this helper, to document places where we make this assertion.
3701 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003702static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003703{
3704#ifdef DEBUG
3705 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3706#endif
3707}
3708
Rusty Russellf35d9d82008-02-04 23:49:54 -05003709bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00003710
Paul Durranted1f50c2014-01-09 10:02:46 +00003711int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
Linus Lüssing9afd85c2015-05-02 14:01:07 +02003712struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3713 unsigned int transport_len,
3714 __sum16(*skb_chkf)(struct sk_buff *skb));
Paul Durranted1f50c2014-01-09 10:02:46 +00003715
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00003716/**
3717 * skb_head_is_locked - Determine if the skb->head is locked down
3718 * @skb: skb to check
3719 *
3720 * The head on skbs build around a head frag can be removed if they are
3721 * not cloned. This function returns true if the skb head is locked down
3722 * due to either being allocated via kmalloc, or by being a clone with
3723 * multiple references to the head.
3724 */
3725static inline bool skb_head_is_locked(const struct sk_buff *skb)
3726{
3727 return !skb->head_frag || skb_cloned(skb);
3728}
Florian Westphalfe6cc552014-02-13 23:09:12 +01003729
3730/**
3731 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3732 *
3733 * @skb: GSO skb
3734 *
3735 * skb_gso_network_seglen is used to determine the real size of the
3736 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3737 *
3738 * The MAC/L2 header is not accounted for.
3739 */
3740static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3741{
3742 unsigned int hdr_len = skb_transport_header(skb) -
3743 skb_network_header(skb);
3744 return hdr_len + skb_gso_transport_seglen(skb);
3745}
Thomas Grafee122c72015-07-21 10:43:58 +02003746
Edward Cree179bc672016-02-11 20:48:04 +00003747/* Local Checksum Offload.
3748 * Compute outer checksum based on the assumption that the
3749 * inner checksum will be offloaded later.
Edward Creee8ae7b02016-02-11 21:03:37 +00003750 * See Documentation/networking/checksum-offloads.txt for
3751 * explanation of how this works.
Edward Cree179bc672016-02-11 20:48:04 +00003752 * Fill in outer checksum adjustment (e.g. with sum of outer
3753 * pseudo-header) before calling.
3754 * Also ensure that inner checksum is in linear data area.
3755 */
3756static inline __wsum lco_csum(struct sk_buff *skb)
3757{
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003758 unsigned char *csum_start = skb_checksum_start(skb);
3759 unsigned char *l4_hdr = skb_transport_header(skb);
3760 __wsum partial;
Edward Cree179bc672016-02-11 20:48:04 +00003761
3762 /* Start with complement of inner checksum adjustment */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003763 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3764 skb->csum_offset));
3765
Edward Cree179bc672016-02-11 20:48:04 +00003766 /* Add in checksum of our headers (incl. outer checksum
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003767 * adjustment filled in by caller) and return result.
Edward Cree179bc672016-02-11 20:48:04 +00003768 */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003769 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
Edward Cree179bc672016-02-11 20:48:04 +00003770}
3771
Linus Torvalds1da177e2005-04-16 15:20:36 -07003772#endif /* __KERNEL__ */
3773#endif /* _LINUX_SKBUFF_H */