blob: ee38a41274759f279be1c0752a7fab63fac517c8 [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,
Shirley Maa6686f22011-07-06 12:22:12 +0000385};
386
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400387#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
Soheil Hassas Yeganeh0a2cf202016-04-27 23:39:01 -0400388 SKBTX_SCHED_TSTAMP)
Willem de Bruijnf24b9be2014-08-04 22:11:45 -0400389#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
390
Shirley Maa6686f22011-07-06 12:22:12 +0000391/*
392 * The callback notifies userspace to release buffers when skb DMA is done in
393 * lower device, the skb last reference should be 0 when calling this.
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000394 * The zerocopy_success argument is true if zero copy transmit occurred,
395 * false on data copy or out of memory error caused by data copy attempt.
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000396 * The ctx field is used to track device context.
397 * The desc field is used to track userspace buffer index.
Shirley Maa6686f22011-07-06 12:22:12 +0000398 */
399struct ubuf_info {
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000400 void (*callback)(struct ubuf_info *, bool zerocopy_success);
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000401 void *ctx;
Shirley Maa6686f22011-07-06 12:22:12 +0000402 unsigned long desc;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000403};
404
Linus Torvalds1da177e2005-04-16 15:20:36 -0700405/* This data is invariant across clones and lives at
406 * the end of the header data, ie. at skb->end.
407 */
408struct skb_shared_info {
Ian Campbell9f42f122012-01-05 07:13:39 +0000409 unsigned char nr_frags;
410 __u8 tx_flags;
Herbert Xu79671682006-06-22 02:40:14 -0700411 unsigned short gso_size;
412 /* Warning: this field is not always filled in (UFO)! */
413 unsigned short gso_segs;
414 unsigned short gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700415 struct sk_buff *frag_list;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000416 struct skb_shared_hwtstamps hwtstamps;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400417 u32 tskey;
Ian Campbell9f42f122012-01-05 07:13:39 +0000418 __be32 ip6_frag_id;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700419
420 /*
421 * Warning : all fields before dataref are cleared in __alloc_skb()
422 */
423 atomic_t dataref;
424
Johann Baudy69e3c752009-05-18 22:11:22 -0700425 /* Intermediate layers must ensure that destructor_arg
426 * remains valid until skb destructor */
427 void * destructor_arg;
Shirley Maa6686f22011-07-06 12:22:12 +0000428
Eric Dumazetfed66382010-07-22 19:09:08 +0000429 /* must be last field, see pskb_expand_head() */
430 skb_frag_t frags[MAX_SKB_FRAGS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700431};
432
433/* We divide dataref into two halves. The higher 16 bits hold references
434 * to the payload part of skb->data. The lower 16 bits hold references to
Patrick McHardy334a8132007-06-25 04:35:20 -0700435 * the entire skb->data. A clone of a headerless skb holds the length of
436 * the header in skb->hdr_len.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700437 *
438 * All users must obey the rule that the skb->data reference count must be
439 * greater than or equal to the payload reference count.
440 *
441 * Holding a reference to the payload part means that the user does not
442 * care about modifications to the header part of skb->data.
443 */
444#define SKB_DATAREF_SHIFT 16
445#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
446
David S. Millerd179cd12005-08-17 14:57:30 -0700447
448enum {
Vijay Subramanianc8753d52014-10-02 10:00:43 -0700449 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
450 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
451 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
David S. Millerd179cd12005-08-17 14:57:30 -0700452};
453
Herbert Xu79671682006-06-22 02:40:14 -0700454enum {
455 SKB_GSO_TCPV4 = 1 << 0,
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700456 SKB_GSO_UDP = 1 << 1,
Herbert Xu576a30e2006-06-27 13:22:38 -0700457
458 /* This indicates the skb is from an untrusted source. */
459 SKB_GSO_DODGY = 1 << 2,
Michael Chanb0da85372006-06-29 12:30:00 -0700460
461 /* This indicates the tcp segment has CWR set. */
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700462 SKB_GSO_TCP_ECN = 1 << 3,
463
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400464 SKB_GSO_TCP_FIXEDID = 1 << 4,
Chris Leech01d5b2f2009-02-27 14:06:49 -0800465
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400466 SKB_GSO_TCPV6 = 1 << 5,
Pravin B Shelar68c33162013-02-14 14:02:41 +0000467
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400468 SKB_GSO_FCOE = 1 << 6,
Pravin B Shelar73136262013-03-07 13:21:51 +0000469
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400470 SKB_GSO_GRE = 1 << 7,
Simon Horman0d89d202013-05-23 21:02:52 +0000471
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400472 SKB_GSO_GRE_CSUM = 1 << 8,
Eric Dumazetcb32f512013-10-19 11:42:57 -0700473
Tom Herbert7e133182016-05-18 09:06:10 -0700474 SKB_GSO_IPXIP4 = 1 << 9,
Eric Dumazet61c1db72013-10-20 20:47:30 -0700475
Tom Herbert7e133182016-05-18 09:06:10 -0700476 SKB_GSO_IPXIP6 = 1 << 10,
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700477
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400478 SKB_GSO_UDP_TUNNEL = 1 << 11,
Tom Herbert4749c092014-06-04 17:20:23 -0700479
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400480 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
481
Alexander Duyck802ab552016-04-10 21:45:03 -0400482 SKB_GSO_PARTIAL = 1 << 13,
483
484 SKB_GSO_TUNNEL_REMCSUM = 1 << 14,
Herbert Xu79671682006-06-22 02:40:14 -0700485};
486
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700487#if BITS_PER_LONG > 32
488#define NET_SKBUFF_DATA_USES_OFFSET 1
489#endif
490
491#ifdef NET_SKBUFF_DATA_USES_OFFSET
492typedef unsigned int sk_buff_data_t;
493#else
494typedef unsigned char *sk_buff_data_t;
495#endif
496
Eric Dumazet363ec39232014-02-26 14:02:11 -0800497/**
498 * struct skb_mstamp - multi resolution time stamps
499 * @stamp_us: timestamp in us resolution
500 * @stamp_jiffies: timestamp in jiffies
501 */
502struct skb_mstamp {
503 union {
504 u64 v64;
505 struct {
506 u32 stamp_us;
507 u32 stamp_jiffies;
508 };
509 };
510};
511
512/**
513 * skb_mstamp_get - get current timestamp
514 * @cl: place to store timestamps
515 */
516static inline void skb_mstamp_get(struct skb_mstamp *cl)
517{
518 u64 val = local_clock();
519
520 do_div(val, NSEC_PER_USEC);
521 cl->stamp_us = (u32)val;
522 cl->stamp_jiffies = (u32)jiffies;
523}
524
525/**
526 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
527 * @t1: pointer to newest sample
528 * @t0: pointer to oldest sample
529 */
530static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
531 const struct skb_mstamp *t0)
532{
533 s32 delta_us = t1->stamp_us - t0->stamp_us;
534 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
535
536 /* If delta_us is negative, this might be because interval is too big,
537 * or local_clock() drift is too big : fallback using jiffies.
538 */
539 if (delta_us <= 0 ||
540 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
541
542 delta_us = jiffies_to_usecs(delta_jiffies);
543
544 return delta_us;
545}
546
Yuchung Cheng625a5e12015-10-16 21:57:45 -0700547static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
548 const struct skb_mstamp *t0)
549{
550 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
551
552 if (!diff)
553 diff = t1->stamp_us - t0->stamp_us;
554 return diff > 0;
555}
Eric Dumazet363ec39232014-02-26 14:02:11 -0800556
Linus Torvalds1da177e2005-04-16 15:20:36 -0700557/**
558 * struct sk_buff - socket buffer
559 * @next: Next buffer in list
560 * @prev: Previous buffer in list
Eric Dumazet363ec39232014-02-26 14:02:11 -0800561 * @tstamp: Time we arrived/left
Eric Dumazet56b17422014-11-03 08:19:53 -0800562 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700563 * @sk: Socket we are owned by
Linus Torvalds1da177e2005-04-16 15:20:36 -0700564 * @dev: Device we arrived on/are leaving by
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700565 * @cb: Control buffer. Free for use by every layer. Put private vars here
Eric Dumazet7fee2262010-05-11 23:19:48 +0000566 * @_skb_refdst: destination entry (with norefcount bit)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700567 * @sp: the security path, used for xfrm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700568 * @len: Length of actual data
569 * @data_len: Data length
570 * @mac_len: Length of link layer header
Patrick McHardy334a8132007-06-25 04:35:20 -0700571 * @hdr_len: writable header length of cloned skb
Herbert Xu663ead32007-04-09 11:59:07 -0700572 * @csum: Checksum (must include start/offset pair)
573 * @csum_start: Offset from skb->head where checksumming should start
574 * @csum_offset: Offset from csum_start where checksum should be stored
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700575 * @priority: Packet queueing priority
WANG Cong60ff7462014-05-04 16:39:18 -0700576 * @ignore_df: allow local fragmentation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700577 * @cloned: Head may be cloned (check refcnt to be sure)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700578 * @ip_summed: Driver fed us an IP checksum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 * @nohdr: Payload reference only, must not modify header
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700580 * @nfctinfo: Relationship of this skb to the connection
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 * @pkt_type: Packet class
Randy Dunlapc83c2482005-10-18 22:07:41 -0700582 * @fclone: skbuff clone status
Randy Dunlapc83c2482005-10-18 22:07:41 -0700583 * @ipvs_property: skbuff is owned by ipvs
Randy Dunlap31729362008-02-18 20:52:13 -0800584 * @peeked: this packet has been seen already, so stats have been
585 * done for it, don't do them again
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700586 * @nf_trace: netfilter packet trace flag
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700587 * @protocol: Packet protocol from driver
588 * @destructor: Destruct function
589 * @nfct: Associated connection, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700590 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
Eric Dumazet8964be42009-11-20 15:35:04 -0800591 * @skb_iif: ifindex of device we arrived on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 * @tc_index: Traffic control index
593 * @tc_verd: traffic control verdict
Tom Herbert61b905d2014-03-24 15:34:47 -0700594 * @hash: the packet hash
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700595 * @queue_mapping: Queue mapping for multiqueue devices
David S. Miller0b725a22014-08-25 15:51:53 -0700596 * @xmit_more: More SKBs are pending for this queue
Randy Dunlap553a5672008-04-20 10:51:01 -0700597 * @ndisc_nodetype: router type (from link layer)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700598 * @ooo_okay: allow the mapping of a socket to a queue to be changed
Tom Herbert61b905d2014-03-24 15:34:47 -0700599 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
Changli Gao4ca24622011-08-19 07:26:44 -0700600 * ports.
Tom Herberta3b18dd2014-07-01 21:33:17 -0700601 * @sw_hash: indicates hash was computed in software stack
Johannes Berg6e3e9392011-11-09 10:15:42 +0100602 * @wifi_acked_valid: wifi_acked was set
603 * @wifi_acked: whether frame was acked on wifi or not
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000604 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
Eliezer Tamir06021292013-06-10 11:39:50 +0300605 * @napi_id: id of the NAPI struct this skb came from
James Morris984bc162006-06-09 00:29:17 -0700606 * @secmark: security marking
Scott Feldman0c4f6912015-07-18 18:24:48 -0700607 * @offload_fwd_mark: fwding offload mark
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700608 * @mark: Generic packet mark
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000609 * @vlan_proto: vlan encapsulation protocol
Patrick McHardy6aa895b2008-07-14 22:49:06 -0700610 * @vlan_tci: vlan tag control information
Simon Horman0d89d202013-05-23 21:02:52 +0000611 * @inner_protocol: Protocol (encapsulation)
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000612 * @inner_transport_header: Inner transport layer header (encapsulation)
613 * @inner_network_header: Network layer header (encapsulation)
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000614 * @inner_mac_header: Link layer header (encapsulation)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700615 * @transport_header: Transport layer header
616 * @network_header: Network layer header
617 * @mac_header: Link layer header
618 * @tail: Tail pointer
619 * @end: End pointer
620 * @head: Head of buffer
621 * @data: Data head pointer
622 * @truesize: Buffer size
623 * @users: User count - see {datagram,tcp}.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 */
625
626struct sk_buff {
Eric Dumazet363ec39232014-02-26 14:02:11 -0800627 union {
Eric Dumazet56b17422014-11-03 08:19:53 -0800628 struct {
629 /* These two members must be first. */
630 struct sk_buff *next;
631 struct sk_buff *prev;
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000632
Eric Dumazet56b17422014-11-03 08:19:53 -0800633 union {
634 ktime_t tstamp;
635 struct skb_mstamp skb_mstamp;
636 };
637 };
638 struct rb_node rbnode; /* used in netem & tcp stack */
639 };
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000640 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 /*
644 * This is the control buffer. It is free to use for every
645 * layer. Please put your private variables there. If you
646 * want to keep them across layers you have to do a skb_clone()
647 * first. This is owned by whoever has the skb queued ATM.
648 */
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000649 char cb[48] __aligned(8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700650
Eric Dumazet7fee2262010-05-11 23:19:48 +0000651 unsigned long _skb_refdst;
Eric Dumazetb1937222014-09-28 22:18:47 -0700652 void (*destructor)(struct sk_buff *skb);
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000653#ifdef CONFIG_XFRM
654 struct sec_path *sp;
655#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700656#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
657 struct nf_conntrack *nfct;
658#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200659#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Eric Dumazetb1937222014-09-28 22:18:47 -0700660 struct nf_bridge_info *nf_bridge;
661#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700662 unsigned int len,
Patrick McHardy334a8132007-06-25 04:35:20 -0700663 data_len;
664 __u16 mac_len,
665 hdr_len;
Eric Dumazetb1937222014-09-28 22:18:47 -0700666
667 /* Following fields are _not_ copied in __copy_skb_header()
668 * Note that queue_mapping is here mostly to fill a hole.
669 */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200670 kmemcheck_bitfield_begin(flags1);
Eric Dumazetb1937222014-09-28 22:18:47 -0700671 __u16 queue_mapping;
672 __u8 cloned:1,
Harald Welte6869c4d2005-08-09 19:24:19 -0700673 nohdr:1,
Patrick McHardyb84f4cc2005-11-20 21:19:21 -0800674 fclone:2,
Herbert Xua59322b2007-12-05 01:53:40 -0800675 peeked:1,
Eric Dumazetb1937222014-09-28 22:18:47 -0700676 head_frag:1,
677 xmit_more:1;
678 /* one bit hole */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200679 kmemcheck_bitfield_end(flags1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700680
Eric Dumazetb1937222014-09-28 22:18:47 -0700681 /* fields enclosed in headers_start/headers_end are copied
682 * using a single memcpy() in __copy_skb_header()
683 */
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700684 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700685 __u32 headers_start[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700686 /* public: */
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200687
688/* if you move pkt_type around you also must adapt those constants */
689#ifdef __BIG_ENDIAN_BITFIELD
690#define PKT_TYPE_MAX (7 << 5)
691#else
692#define PKT_TYPE_MAX 7
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100693#endif
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200694#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
695
696 __u8 __pkt_type_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700697 __u8 pkt_type:3;
698 __u8 pfmemalloc:1;
699 __u8 ignore_df:1;
700 __u8 nfctinfo:3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700701
Eric Dumazetb1937222014-09-28 22:18:47 -0700702 __u8 nf_trace:1;
703 __u8 ip_summed:2;
704 __u8 ooo_okay:1;
705 __u8 l4_hash:1;
706 __u8 sw_hash:1;
707 __u8 wifi_acked_valid:1;
708 __u8 wifi_acked:1;
709
710 __u8 no_fcs:1;
711 /* Indicates the inner headers are valid in the skbuff. */
712 __u8 encapsulation:1;
713 __u8 encap_hdr_csum:1;
714 __u8 csum_valid:1;
715 __u8 csum_complete_sw:1;
716 __u8 csum_level:2;
717 __u8 csum_bad:1;
718
719#ifdef CONFIG_IPV6_NDISC_NODETYPE
720 __u8 ndisc_nodetype:2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700721#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700722 __u8 ipvs_property:1;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700723 __u8 inner_protocol_type:1;
Tom Herberte585f232014-11-04 09:06:54 -0800724 __u8 remcsum_offload:1;
725 /* 3 or 5 bit hole */
Alexander Duyck4031ae62012-01-27 06:22:53 +0000726
Linus Torvalds1da177e2005-04-16 15:20:36 -0700727#ifdef CONFIG_NET_SCHED
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700728 __u16 tc_index; /* traffic control index */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729#ifdef CONFIG_NET_CLS_ACT
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700730 __u16 tc_verd; /* traffic control verdict */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700732#endif
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200733
Eric Dumazetb1937222014-09-28 22:18:47 -0700734 union {
735 __wsum csum;
736 struct {
737 __u16 csum_start;
738 __u16 csum_offset;
739 };
740 };
741 __u32 priority;
742 int skb_iif;
743 __u32 hash;
744 __be16 vlan_proto;
745 __u16 vlan_tci;
Eric Dumazet2bd82482015-02-03 23:48:24 -0800746#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
747 union {
748 unsigned int napi_id;
749 unsigned int sender_cpu;
750 };
Chris Leech97fc2f02006-05-23 17:55:33 -0700751#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700752 union {
James Morris984bc162006-06-09 00:29:17 -0700753#ifdef CONFIG_NETWORK_SECMARK
Scott Feldman0c4f6912015-07-18 18:24:48 -0700754 __u32 secmark;
James Morris984bc162006-06-09 00:29:17 -0700755#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700756#ifdef CONFIG_NET_SWITCHDEV
757 __u32 offload_fwd_mark;
758#endif
759 };
760
Neil Horman3b885782009-10-12 13:26:31 -0700761 union {
762 __u32 mark;
Eric Dumazet16fad692013-03-14 05:40:32 +0000763 __u32 reserved_tailroom;
Neil Horman3b885782009-10-12 13:26:31 -0700764 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700765
Tom Herbert8bce6d72014-09-29 20:22:29 -0700766 union {
767 __be16 inner_protocol;
768 __u8 inner_ipproto;
769 };
770
Simon Horman1a37e412013-05-23 21:02:51 +0000771 __u16 inner_transport_header;
772 __u16 inner_network_header;
773 __u16 inner_mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700774
775 __be16 protocol;
Simon Horman1a37e412013-05-23 21:02:51 +0000776 __u16 transport_header;
777 __u16 network_header;
778 __u16 mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700779
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700780 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700781 __u32 headers_end[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700782 /* public: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784 /* These elements must be at the end, see alloc_skb() for details. */
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700785 sk_buff_data_t tail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700786 sk_buff_data_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700787 unsigned char *head,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700788 *data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700789 unsigned int truesize;
790 atomic_t users;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791};
792
793#ifdef __KERNEL__
794/*
795 * Handling routines are only of interest to the kernel
796 */
797#include <linux/slab.h>
798
Linus Torvalds1da177e2005-04-16 15:20:36 -0700799
Mel Gormanc93bdd02012-07-31 16:44:19 -0700800#define SKB_ALLOC_FCLONE 0x01
801#define SKB_ALLOC_RX 0x02
Alexander Duyckfd11a832014-12-09 19:40:49 -0800802#define SKB_ALLOC_NAPI 0x04
Mel Gormanc93bdd02012-07-31 16:44:19 -0700803
804/* Returns true if the skb was allocated from PFMEMALLOC reserves */
805static inline bool skb_pfmemalloc(const struct sk_buff *skb)
806{
807 return unlikely(skb->pfmemalloc);
808}
809
Eric Dumazet7fee2262010-05-11 23:19:48 +0000810/*
811 * skb might have a dst pointer attached, refcounted or not.
812 * _skb_refdst low order bit is set if refcount was _not_ taken
813 */
814#define SKB_DST_NOREF 1UL
815#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
816
817/**
818 * skb_dst - returns skb dst_entry
819 * @skb: buffer
820 *
821 * Returns skb dst_entry, regardless of reference taken or not.
822 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000823static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
824{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000825 /* If refdst was not refcounted, check we still are in a
826 * rcu_read_lock section
827 */
828 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
829 !rcu_read_lock_held() &&
830 !rcu_read_lock_bh_held());
831 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
Eric Dumazetadf30902009-06-02 05:19:30 +0000832}
833
Eric Dumazet7fee2262010-05-11 23:19:48 +0000834/**
835 * skb_dst_set - sets skb dst
836 * @skb: buffer
837 * @dst: dst entry
838 *
839 * Sets skb dst, assuming a reference was taken on dst and should
840 * be released by skb_dst_drop()
841 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000842static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
843{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000844 skb->_skb_refdst = (unsigned long)dst;
845}
846
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200847/**
848 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
849 * @skb: buffer
850 * @dst: dst entry
851 *
852 * Sets skb dst, assuming a reference was not taken on dst.
853 * If dst entry is cached, we do not take reference and dst_release
854 * will be avoided by refdst_drop. If dst entry is not cached, we take
855 * reference, so that last dst_release can destroy the dst immediately.
856 */
857static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
858{
Hannes Frederic Sowadbfc4fb2014-12-06 19:19:42 +0100859 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
860 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200861}
Eric Dumazet7fee2262010-05-11 23:19:48 +0000862
863/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300864 * skb_dst_is_noref - Test if skb dst isn't refcounted
Eric Dumazet7fee2262010-05-11 23:19:48 +0000865 * @skb: buffer
866 */
867static inline bool skb_dst_is_noref(const struct sk_buff *skb)
868{
869 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
Eric Dumazetadf30902009-06-02 05:19:30 +0000870}
871
Eric Dumazet511c3f92009-06-02 05:14:27 +0000872static inline struct rtable *skb_rtable(const struct sk_buff *skb)
873{
Eric Dumazetadf30902009-06-02 05:19:30 +0000874 return (struct rtable *)skb_dst(skb);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000875}
876
Joe Perches7965bd42013-09-26 14:48:15 -0700877void kfree_skb(struct sk_buff *skb);
878void kfree_skb_list(struct sk_buff *segs);
879void skb_tx_error(struct sk_buff *skb);
880void consume_skb(struct sk_buff *skb);
881void __kfree_skb(struct sk_buff *skb);
Eric Dumazetd7e88832012-04-30 08:10:34 +0000882extern struct kmem_cache *skbuff_head_cache;
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000883
Joe Perches7965bd42013-09-26 14:48:15 -0700884void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
885bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
886 bool *fragstolen, int *delta_truesize);
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000887
Joe Perches7965bd42013-09-26 14:48:15 -0700888struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
889 int node);
Eric Dumazet2ea2f622015-04-24 16:05:01 -0700890struct sk_buff *__build_skb(void *data, unsigned int frag_size);
Joe Perches7965bd42013-09-26 14:48:15 -0700891struct sk_buff *build_skb(void *data, unsigned int frag_size);
David S. Millerd179cd12005-08-17 14:57:30 -0700892static inline struct sk_buff *alloc_skb(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100893 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700894{
Eric Dumazet564824b2010-10-11 19:05:25 +0000895 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700896}
897
Eric Dumazet2e4e4412014-09-17 04:49:49 -0700898struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
899 unsigned long data_len,
900 int max_page_order,
901 int *errcode,
902 gfp_t gfp_mask);
903
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700904/* Layout of fast clones : [skb1][skb2][fclone_ref] */
905struct sk_buff_fclones {
906 struct sk_buff skb1;
907
908 struct sk_buff skb2;
909
910 atomic_t fclone_ref;
911};
912
913/**
914 * skb_fclone_busy - check if fclone is busy
915 * @skb: buffer
916 *
Masanari Iidabda13fe2015-12-13 16:53:02 +0900917 * Returns true if skb is a fast clone, and its clone is not freed.
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700918 * Some drivers call skb_orphan() in their ndo_start_xmit(),
919 * so we also check that this didnt happen.
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700920 */
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700921static inline bool skb_fclone_busy(const struct sock *sk,
922 const struct sk_buff *skb)
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700923{
924 const struct sk_buff_fclones *fclones;
925
926 fclones = container_of(skb, struct sk_buff_fclones, skb1);
927
928 return skb->fclone == SKB_FCLONE_ORIG &&
Eric Dumazet6ffe75e2014-12-03 17:04:39 -0800929 atomic_read(&fclones->fclone_ref) > 1 &&
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700930 fclones->skb2.sk == sk;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700931}
932
David S. Millerd179cd12005-08-17 14:57:30 -0700933static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100934 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700935{
Mel Gormanc93bdd02012-07-31 16:44:19 -0700936 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700937}
938
Joe Perches7965bd42013-09-26 14:48:15 -0700939struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000940static inline struct sk_buff *alloc_skb_head(gfp_t priority)
941{
942 return __alloc_skb_head(priority, -1);
943}
944
Joe Perches7965bd42013-09-26 14:48:15 -0700945struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
946int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
947struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
948struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
Octavian Purdilabad93e92014-06-12 01:36:26 +0300949struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
950 gfp_t gfp_mask, bool fclone);
951static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
952 gfp_t gfp_mask)
953{
954 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
955}
Eric Dumazet117632e2011-12-03 21:39:53 +0000956
Joe Perches7965bd42013-09-26 14:48:15 -0700957int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
958struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
959 unsigned int headroom);
960struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
961 int newtailroom, gfp_t priority);
Fan Du25a91d82014-01-18 09:54:23 +0800962int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
963 int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -0700964int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
965 int len);
966int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
967int skb_pad(struct sk_buff *skb, int pad);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000968#define dev_kfree_skb(a) consume_skb(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700969
Joe Perches7965bd42013-09-26 14:48:15 -0700970int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
971 int getfrag(void *from, char *to, int offset,
972 int len, int odd, struct sk_buff *skb),
973 void *from, int length);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700974
Hannes Frederic Sowabe12a1f2015-05-21 16:59:58 +0200975int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
976 int offset, size_t size);
977
Eric Dumazetd94d9fe2009-11-04 09:50:58 -0800978struct skb_seq_state {
Thomas Graf677e90e2005-06-23 20:59:51 -0700979 __u32 lower_offset;
980 __u32 upper_offset;
981 __u32 frag_idx;
982 __u32 stepped_offset;
983 struct sk_buff *root_skb;
984 struct sk_buff *cur_skb;
985 __u8 *frag_data;
986};
987
Joe Perches7965bd42013-09-26 14:48:15 -0700988void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
989 unsigned int to, struct skb_seq_state *st);
990unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
991 struct skb_seq_state *st);
992void skb_abort_seq_read(struct skb_seq_state *st);
Thomas Graf677e90e2005-06-23 20:59:51 -0700993
Joe Perches7965bd42013-09-26 14:48:15 -0700994unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
Bojan Prtvar059a2442015-02-22 11:46:35 +0100995 unsigned int to, struct ts_config *config);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -0700996
Tom Herbert09323cc2013-12-15 22:16:19 -0800997/*
998 * Packet hash types specify the type of hash in skb_set_hash.
999 *
1000 * Hash types refer to the protocol layer addresses which are used to
1001 * construct a packet's hash. The hashes are used to differentiate or identify
1002 * flows of the protocol layer for the hash type. Hash types are either
1003 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1004 *
1005 * Properties of hashes:
1006 *
1007 * 1) Two packets in different flows have different hash values
1008 * 2) Two packets in the same flow should have the same hash value
1009 *
1010 * A hash at a higher layer is considered to be more specific. A driver should
1011 * set the most specific hash possible.
1012 *
1013 * A driver cannot indicate a more specific hash than the layer at which a hash
1014 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1015 *
1016 * A driver may indicate a hash level which is less specific than the
1017 * actual layer the hash was computed on. For instance, a hash computed
1018 * at L4 may be considered an L3 hash. This should only be done if the
1019 * driver can't unambiguously determine that the HW computed the hash at
1020 * the higher layer. Note that the "should" in the second property above
1021 * permits this.
1022 */
1023enum pkt_hash_types {
1024 PKT_HASH_TYPE_NONE, /* Undefined type */
1025 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1026 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1027 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1028};
1029
Tom Herbertbcc83832015-09-01 09:24:24 -07001030static inline void skb_clear_hash(struct sk_buff *skb)
1031{
1032 skb->hash = 0;
1033 skb->sw_hash = 0;
1034 skb->l4_hash = 0;
1035}
1036
1037static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1038{
1039 if (!skb->l4_hash)
1040 skb_clear_hash(skb);
1041}
1042
1043static inline void
1044__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1045{
1046 skb->l4_hash = is_l4;
1047 skb->sw_hash = is_sw;
1048 skb->hash = hash;
1049}
1050
Tom Herbert09323cc2013-12-15 22:16:19 -08001051static inline void
1052skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1053{
Tom Herbertbcc83832015-09-01 09:24:24 -07001054 /* Used by drivers to set hash from HW */
1055 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1056}
1057
1058static inline void
1059__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1060{
1061 __skb_set_hash(skb, hash, true, is_l4);
Tom Herbert09323cc2013-12-15 22:16:19 -08001062}
1063
Tom Herberte5276932015-09-01 09:24:23 -07001064void __skb_get_hash(struct sk_buff *skb);
1065u32 skb_get_poff(const struct sk_buff *skb);
1066u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1067 const struct flow_keys *keys, int hlen);
1068__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1069 void *data, int hlen_proto);
1070
1071static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1072 int thoff, u8 ip_proto)
1073{
1074 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1075}
1076
1077void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1078 const struct flow_dissector_key *key,
1079 unsigned int key_count);
1080
1081bool __skb_flow_dissect(const struct sk_buff *skb,
1082 struct flow_dissector *flow_dissector,
1083 void *target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001084 void *data, __be16 proto, int nhoff, int hlen,
1085 unsigned int flags);
Tom Herberte5276932015-09-01 09:24:23 -07001086
1087static inline bool skb_flow_dissect(const struct sk_buff *skb,
1088 struct flow_dissector *flow_dissector,
Tom Herbertcd79a232015-09-01 09:24:27 -07001089 void *target_container, unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001090{
1091 return __skb_flow_dissect(skb, flow_dissector, target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001092 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001093}
1094
1095static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
Tom Herbertcd79a232015-09-01 09:24:27 -07001096 struct flow_keys *flow,
1097 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001098{
1099 memset(flow, 0, sizeof(*flow));
1100 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001101 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001102}
1103
1104static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1105 void *data, __be16 proto,
Tom Herbertcd79a232015-09-01 09:24:27 -07001106 int nhoff, int hlen,
1107 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001108{
1109 memset(flow, 0, sizeof(*flow));
1110 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001111 data, proto, nhoff, hlen, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001112}
1113
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001114static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001115{
Tom Herberta3b18dd2014-07-01 21:33:17 -07001116 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001117 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001118
Tom Herbert61b905d2014-03-24 15:34:47 -07001119 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001120}
1121
David S. Miller20a17bf2015-09-01 21:19:17 -07001122__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
Tom Herbertf70ea012015-07-31 16:52:10 -07001123
David S. Miller20a17bf2015-09-01 21:19:17 -07001124static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
Tom Herbertf70ea012015-07-31 16:52:10 -07001125{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001126 if (!skb->l4_hash && !skb->sw_hash) {
1127 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001128 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001129
Tom Herbertde4c1f82015-09-01 18:11:04 -07001130 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001131 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001132
1133 return skb->hash;
1134}
1135
David S. Miller20a17bf2015-09-01 21:19:17 -07001136__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
Tom Herbertf70ea012015-07-31 16:52:10 -07001137
David S. Miller20a17bf2015-09-01 21:19:17 -07001138static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
Tom Herbertf70ea012015-07-31 16:52:10 -07001139{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001140 if (!skb->l4_hash && !skb->sw_hash) {
1141 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001142 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001143
Tom Herbertde4c1f82015-09-01 18:11:04 -07001144 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001145 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001146
1147 return skb->hash;
1148}
1149
Tom Herbert50fb7992015-05-01 11:30:12 -07001150__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1151
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001152static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1153{
Tom Herbert61b905d2014-03-24 15:34:47 -07001154 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001155}
1156
Tom Herbert3df7a742013-12-15 22:16:29 -08001157static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1158{
Tom Herbert61b905d2014-03-24 15:34:47 -07001159 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -07001160 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -07001161 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -08001162};
1163
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001164#ifdef NET_SKBUFF_DATA_USES_OFFSET
1165static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1166{
1167 return skb->head + skb->end;
1168}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001169
1170static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1171{
1172 return skb->end;
1173}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001174#else
1175static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1176{
1177 return skb->end;
1178}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001179
1180static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1181{
1182 return skb->end - skb->head;
1183}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001184#endif
1185
Linus Torvalds1da177e2005-04-16 15:20:36 -07001186/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001187#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188
Patrick Ohlyac45f602009-02-12 05:03:37 +00001189static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1190{
1191 return &skb_shinfo(skb)->hwtstamps;
1192}
1193
Linus Torvalds1da177e2005-04-16 15:20:36 -07001194/**
1195 * skb_queue_empty - check if a queue is empty
1196 * @list: queue head
1197 *
1198 * Returns true if the queue is empty, false otherwise.
1199 */
1200static inline int skb_queue_empty(const struct sk_buff_head *list)
1201{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001202 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001203}
1204
1205/**
David S. Millerfc7ebb22008-09-23 00:34:07 -07001206 * skb_queue_is_last - check if skb is the last entry in the queue
1207 * @list: queue head
1208 * @skb: buffer
1209 *
1210 * Returns true if @skb is the last buffer on the list.
1211 */
1212static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1213 const struct sk_buff *skb)
1214{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001215 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -07001216}
1217
1218/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001219 * skb_queue_is_first - check if skb is the first entry in the queue
1220 * @list: queue head
1221 * @skb: buffer
1222 *
1223 * Returns true if @skb is the first buffer on the list.
1224 */
1225static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1226 const struct sk_buff *skb)
1227{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001228 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001229}
1230
1231/**
David S. Miller249c8b42008-09-23 00:44:42 -07001232 * skb_queue_next - return the next packet in the queue
1233 * @list: queue head
1234 * @skb: current buffer
1235 *
1236 * Return the next packet in @list after @skb. It is only valid to
1237 * call this if skb_queue_is_last() evaluates to false.
1238 */
1239static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1240 const struct sk_buff *skb)
1241{
1242 /* This BUG_ON may seem severe, but if we just return then we
1243 * are going to dereference garbage.
1244 */
1245 BUG_ON(skb_queue_is_last(list, skb));
1246 return skb->next;
1247}
1248
1249/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001250 * skb_queue_prev - return the prev packet in the queue
1251 * @list: queue head
1252 * @skb: current buffer
1253 *
1254 * Return the prev packet in @list before @skb. It is only valid to
1255 * call this if skb_queue_is_first() evaluates to false.
1256 */
1257static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1258 const struct sk_buff *skb)
1259{
1260 /* This BUG_ON may seem severe, but if we just return then we
1261 * are going to dereference garbage.
1262 */
1263 BUG_ON(skb_queue_is_first(list, skb));
1264 return skb->prev;
1265}
1266
1267/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001268 * skb_get - reference buffer
1269 * @skb: buffer to reference
1270 *
1271 * Makes another reference to a socket buffer and returns a pointer
1272 * to the buffer.
1273 */
1274static inline struct sk_buff *skb_get(struct sk_buff *skb)
1275{
1276 atomic_inc(&skb->users);
1277 return skb;
1278}
1279
1280/*
1281 * If users == 1, we are the only owner and are can avoid redundant
1282 * atomic change.
1283 */
1284
1285/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001286 * skb_cloned - is the buffer a clone
1287 * @skb: buffer to check
1288 *
1289 * Returns true if the buffer was generated with skb_clone() and is
1290 * one of multiple shared copies of the buffer. Cloned buffers are
1291 * shared data so must not be written to under normal circumstances.
1292 */
1293static inline int skb_cloned(const struct sk_buff *skb)
1294{
1295 return skb->cloned &&
1296 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1297}
1298
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001299static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1300{
Mel Gormand0164ad2015-11-06 16:28:21 -08001301 might_sleep_if(gfpflags_allow_blocking(pri));
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001302
1303 if (skb_cloned(skb))
1304 return pskb_expand_head(skb, 0, 0, pri);
1305
1306 return 0;
1307}
1308
Linus Torvalds1da177e2005-04-16 15:20:36 -07001309/**
1310 * skb_header_cloned - is the header a clone
1311 * @skb: buffer to check
1312 *
1313 * Returns true if modifying the header part of the buffer requires
1314 * the data to be copied.
1315 */
1316static inline int skb_header_cloned(const struct sk_buff *skb)
1317{
1318 int dataref;
1319
1320 if (!skb->cloned)
1321 return 0;
1322
1323 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1324 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1325 return dataref != 1;
1326}
1327
Eric Dumazet9580bf22016-04-30 10:19:29 -07001328static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1329{
1330 might_sleep_if(gfpflags_allow_blocking(pri));
1331
1332 if (skb_header_cloned(skb))
1333 return pskb_expand_head(skb, 0, 0, pri);
1334
1335 return 0;
1336}
1337
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338/**
1339 * skb_header_release - release reference to header
1340 * @skb: buffer to operate on
1341 *
1342 * Drop a reference to the header part of the buffer. This is done
1343 * by acquiring a payload reference. You must not read from the header
1344 * part of skb->data after this.
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001345 * Note : Check if you can use __skb_header_release() instead.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001346 */
1347static inline void skb_header_release(struct sk_buff *skb)
1348{
1349 BUG_ON(skb->nohdr);
1350 skb->nohdr = 1;
1351 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1352}
1353
1354/**
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001355 * __skb_header_release - release reference to header
1356 * @skb: buffer to operate on
1357 *
1358 * Variant of skb_header_release() assuming skb is private to caller.
1359 * We can avoid one atomic operation.
1360 */
1361static inline void __skb_header_release(struct sk_buff *skb)
1362{
1363 skb->nohdr = 1;
1364 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1365}
1366
1367
1368/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001369 * skb_shared - is the buffer shared
1370 * @skb: buffer to check
1371 *
1372 * Returns true if more than one person has a reference to this
1373 * buffer.
1374 */
1375static inline int skb_shared(const struct sk_buff *skb)
1376{
1377 return atomic_read(&skb->users) != 1;
1378}
1379
1380/**
1381 * skb_share_check - check if buffer is shared and if so clone it
1382 * @skb: buffer to check
1383 * @pri: priority for memory allocation
1384 *
1385 * If the buffer is shared the buffer is cloned and the old copy
1386 * drops a reference. A new clone with a single reference is returned.
1387 * If the buffer is not shared the original buffer is returned. When
1388 * being called from interrupt status or with spinlocks held pri must
1389 * be GFP_ATOMIC.
1390 *
1391 * NULL is returned on a memory allocation failure.
1392 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001393static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001394{
Mel Gormand0164ad2015-11-06 16:28:21 -08001395 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 if (skb_shared(skb)) {
1397 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001398
1399 if (likely(nskb))
1400 consume_skb(skb);
1401 else
1402 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001403 skb = nskb;
1404 }
1405 return skb;
1406}
1407
1408/*
1409 * Copy shared buffers into a new sk_buff. We effectively do COW on
1410 * packets to handle cases where we have a local reader and forward
1411 * and a couple of other messy ones. The normal one is tcpdumping
1412 * a packet thats being forwarded.
1413 */
1414
1415/**
1416 * skb_unshare - make a copy of a shared buffer
1417 * @skb: buffer to check
1418 * @pri: priority for memory allocation
1419 *
1420 * If the socket buffer is a clone then this function creates a new
1421 * copy of the data, drops a reference count on the old copy and returns
1422 * the new copy with the reference count at 1. If the buffer is not a clone
1423 * the original buffer is returned. When called with a spinlock held or
1424 * from interrupt state @pri must be %GFP_ATOMIC
1425 *
1426 * %NULL is returned on a memory allocation failure.
1427 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001428static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001429 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001430{
Mel Gormand0164ad2015-11-06 16:28:21 -08001431 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 if (skb_cloned(skb)) {
1433 struct sk_buff *nskb = skb_copy(skb, pri);
Alexander Aring31eff812014-10-10 23:10:47 +02001434
1435 /* Free our shared copy */
1436 if (likely(nskb))
1437 consume_skb(skb);
1438 else
1439 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 skb = nskb;
1441 }
1442 return skb;
1443}
1444
1445/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001446 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001447 * @list_: list to peek at
1448 *
1449 * Peek an &sk_buff. Unlike most other operations you _MUST_
1450 * be careful with this one. A peek leaves the buffer on the
1451 * list and someone else may run off with it. You must hold
1452 * the appropriate locks or have a private queue to do this.
1453 *
1454 * Returns %NULL for an empty list or a pointer to the head element.
1455 * The reference count is not incremented and the reference is therefore
1456 * volatile. Use with caution.
1457 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001458static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001459{
Eric Dumazet18d07002012-04-30 16:31:46 +00001460 struct sk_buff *skb = list_->next;
1461
1462 if (skb == (struct sk_buff *)list_)
1463 skb = NULL;
1464 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001465}
1466
1467/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001468 * skb_peek_next - peek skb following the given one from a queue
1469 * @skb: skb to start from
1470 * @list_: list to peek at
1471 *
1472 * Returns %NULL when the end of the list is met or a pointer to the
1473 * next element. The reference count is not incremented and the
1474 * reference is therefore volatile. Use with caution.
1475 */
1476static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1477 const struct sk_buff_head *list_)
1478{
1479 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001480
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001481 if (next == (struct sk_buff *)list_)
1482 next = NULL;
1483 return next;
1484}
1485
1486/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001487 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001488 * @list_: list to peek at
1489 *
1490 * Peek an &sk_buff. Unlike most other operations you _MUST_
1491 * be careful with this one. A peek leaves the buffer on the
1492 * list and someone else may run off with it. You must hold
1493 * the appropriate locks or have a private queue to do this.
1494 *
1495 * Returns %NULL for an empty list or a pointer to the tail element.
1496 * The reference count is not incremented and the reference is therefore
1497 * volatile. Use with caution.
1498 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001499static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500{
Eric Dumazet18d07002012-04-30 16:31:46 +00001501 struct sk_buff *skb = list_->prev;
1502
1503 if (skb == (struct sk_buff *)list_)
1504 skb = NULL;
1505 return skb;
1506
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507}
1508
1509/**
1510 * skb_queue_len - get queue length
1511 * @list_: list to measure
1512 *
1513 * Return the length of an &sk_buff queue.
1514 */
1515static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1516{
1517 return list_->qlen;
1518}
1519
David S. Miller67fed452008-09-21 22:36:24 -07001520/**
1521 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1522 * @list: queue to initialize
1523 *
1524 * This initializes only the list and queue length aspects of
1525 * an sk_buff_head object. This allows to initialize the list
1526 * aspects of an sk_buff_head without reinitializing things like
1527 * the spinlock. It can also be used for on-stack sk_buff_head
1528 * objects where the spinlock is known to not be used.
1529 */
1530static inline void __skb_queue_head_init(struct sk_buff_head *list)
1531{
1532 list->prev = list->next = (struct sk_buff *)list;
1533 list->qlen = 0;
1534}
1535
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001536/*
1537 * This function creates a split out lock class for each invocation;
1538 * this is needed for now since a whole lot of users of the skb-queue
1539 * infrastructure in drivers have different locking usage (in hardirq)
1540 * than the networking core (in softirq only). In the long run either the
1541 * network layer or drivers should need annotation to consolidate the
1542 * main types of usage into 3 classes.
1543 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544static inline void skb_queue_head_init(struct sk_buff_head *list)
1545{
1546 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001547 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001548}
1549
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001550static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1551 struct lock_class_key *class)
1552{
1553 skb_queue_head_init(list);
1554 lockdep_set_class(&list->lock, class);
1555}
1556
Linus Torvalds1da177e2005-04-16 15:20:36 -07001557/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001558 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001559 *
1560 * The "__skb_xxxx()" functions are the non-atomic ones that
1561 * can only be called with interrupts disabled.
1562 */
Joe Perches7965bd42013-09-26 14:48:15 -07001563void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1564 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001565static inline void __skb_insert(struct sk_buff *newsk,
1566 struct sk_buff *prev, struct sk_buff *next,
1567 struct sk_buff_head *list)
1568{
1569 newsk->next = next;
1570 newsk->prev = prev;
1571 next->prev = prev->next = newsk;
1572 list->qlen++;
1573}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001574
David S. Miller67fed452008-09-21 22:36:24 -07001575static inline void __skb_queue_splice(const struct sk_buff_head *list,
1576 struct sk_buff *prev,
1577 struct sk_buff *next)
1578{
1579 struct sk_buff *first = list->next;
1580 struct sk_buff *last = list->prev;
1581
1582 first->prev = prev;
1583 prev->next = first;
1584
1585 last->next = next;
1586 next->prev = last;
1587}
1588
1589/**
1590 * skb_queue_splice - join two skb lists, this is designed for stacks
1591 * @list: the new list to add
1592 * @head: the place to add it in the first list
1593 */
1594static inline void skb_queue_splice(const struct sk_buff_head *list,
1595 struct sk_buff_head *head)
1596{
1597 if (!skb_queue_empty(list)) {
1598 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001599 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001600 }
1601}
1602
1603/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001604 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001605 * @list: the new list to add
1606 * @head: the place to add it in the first list
1607 *
1608 * The list at @list is reinitialised
1609 */
1610static inline void skb_queue_splice_init(struct sk_buff_head *list,
1611 struct sk_buff_head *head)
1612{
1613 if (!skb_queue_empty(list)) {
1614 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001615 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001616 __skb_queue_head_init(list);
1617 }
1618}
1619
1620/**
1621 * skb_queue_splice_tail - join two skb lists, each list being a queue
1622 * @list: the new list to add
1623 * @head: the place to add it in the first list
1624 */
1625static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1626 struct sk_buff_head *head)
1627{
1628 if (!skb_queue_empty(list)) {
1629 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001630 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001631 }
1632}
1633
1634/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001635 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001636 * @list: the new list to add
1637 * @head: the place to add it in the first list
1638 *
1639 * Each of the lists is a queue.
1640 * The list at @list is reinitialised
1641 */
1642static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1643 struct sk_buff_head *head)
1644{
1645 if (!skb_queue_empty(list)) {
1646 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001647 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001648 __skb_queue_head_init(list);
1649 }
1650}
1651
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001653 * __skb_queue_after - queue a buffer at the list head
1654 * @list: list to use
1655 * @prev: place after this buffer
1656 * @newsk: buffer to queue
1657 *
1658 * Queue a buffer int the middle of a list. This function takes no locks
1659 * and you must therefore hold required locks before calling it.
1660 *
1661 * A buffer cannot be placed on two lists at the same time.
1662 */
1663static inline void __skb_queue_after(struct sk_buff_head *list,
1664 struct sk_buff *prev,
1665 struct sk_buff *newsk)
1666{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001667 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001668}
1669
Joe Perches7965bd42013-09-26 14:48:15 -07001670void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1671 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001672
Gerrit Renkerf5572852008-04-14 00:05:28 -07001673static inline void __skb_queue_before(struct sk_buff_head *list,
1674 struct sk_buff *next,
1675 struct sk_buff *newsk)
1676{
1677 __skb_insert(newsk, next->prev, next, list);
1678}
1679
Stephen Hemminger300ce172005-10-30 13:47:34 -08001680/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001681 * __skb_queue_head - queue a buffer at the list head
1682 * @list: list to use
1683 * @newsk: buffer to queue
1684 *
1685 * Queue a buffer at the start of a list. This function takes no locks
1686 * and you must therefore hold required locks before calling it.
1687 *
1688 * A buffer cannot be placed on two lists at the same time.
1689 */
Joe Perches7965bd42013-09-26 14:48:15 -07001690void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001691static inline void __skb_queue_head(struct sk_buff_head *list,
1692 struct sk_buff *newsk)
1693{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001694 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001695}
1696
1697/**
1698 * __skb_queue_tail - queue a buffer at the list tail
1699 * @list: list to use
1700 * @newsk: buffer to queue
1701 *
1702 * Queue a buffer at the end of a list. This function takes no locks
1703 * and you must therefore hold required locks before calling it.
1704 *
1705 * A buffer cannot be placed on two lists at the same time.
1706 */
Joe Perches7965bd42013-09-26 14:48:15 -07001707void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708static inline void __skb_queue_tail(struct sk_buff_head *list,
1709 struct sk_buff *newsk)
1710{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001711 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712}
1713
Linus Torvalds1da177e2005-04-16 15:20:36 -07001714/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001715 * remove sk_buff from list. _Must_ be called atomically, and with
1716 * the list known..
1717 */
Joe Perches7965bd42013-09-26 14:48:15 -07001718void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1720{
1721 struct sk_buff *next, *prev;
1722
1723 list->qlen--;
1724 next = skb->next;
1725 prev = skb->prev;
1726 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001727 next->prev = prev;
1728 prev->next = next;
1729}
1730
Gerrit Renkerf525c062008-04-14 00:04:12 -07001731/**
1732 * __skb_dequeue - remove from the head of the queue
1733 * @list: list to dequeue from
1734 *
1735 * Remove the head of the list. This function does not take any locks
1736 * so must be used with appropriate locks held only. The head item is
1737 * returned or %NULL if the list is empty.
1738 */
Joe Perches7965bd42013-09-26 14:48:15 -07001739struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001740static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1741{
1742 struct sk_buff *skb = skb_peek(list);
1743 if (skb)
1744 __skb_unlink(skb, list);
1745 return skb;
1746}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001747
1748/**
1749 * __skb_dequeue_tail - remove from the tail of the queue
1750 * @list: list to dequeue from
1751 *
1752 * Remove the tail of the list. This function does not take any locks
1753 * so must be used with appropriate locks held only. The tail item is
1754 * returned or %NULL if the list is empty.
1755 */
Joe Perches7965bd42013-09-26 14:48:15 -07001756struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001757static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1758{
1759 struct sk_buff *skb = skb_peek_tail(list);
1760 if (skb)
1761 __skb_unlink(skb, list);
1762 return skb;
1763}
1764
1765
David S. Millerbdcc0922012-03-07 20:53:36 -05001766static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001767{
1768 return skb->data_len;
1769}
1770
1771static inline unsigned int skb_headlen(const struct sk_buff *skb)
1772{
1773 return skb->len - skb->data_len;
1774}
1775
1776static inline int skb_pagelen(const struct sk_buff *skb)
1777{
1778 int i, len = 0;
1779
1780 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001781 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001782 return len + skb_headlen(skb);
1783}
1784
Ian Campbell131ea662011-08-19 06:25:00 +00001785/**
1786 * __skb_fill_page_desc - initialise a paged fragment in an skb
1787 * @skb: buffer containing fragment to be initialised
1788 * @i: paged fragment index to initialise
1789 * @page: the page to use for this fragment
1790 * @off: the offset to the data with @page
1791 * @size: the length of the data
1792 *
1793 * Initialises the @i'th fragment of @skb to point to &size bytes at
1794 * offset @off within @page.
1795 *
1796 * Does not take any additional reference on the fragment.
1797 */
1798static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1799 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001800{
1801 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1802
Mel Gormanc48a11c2012-07-31 16:44:23 -07001803 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001804 * Propagate page pfmemalloc to the skb if we can. The problem is
1805 * that not all callers have unique ownership of the page but rely
1806 * on page_is_pfmemalloc doing the right thing(tm).
Mel Gormanc48a11c2012-07-31 16:44:23 -07001807 */
Ian Campbella8605c62011-10-19 23:01:49 +00001808 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001809 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001810 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001811
1812 page = compound_head(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001813 if (page_is_pfmemalloc(page))
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001814 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001815}
1816
1817/**
1818 * skb_fill_page_desc - initialise a paged fragment in an skb
1819 * @skb: buffer containing fragment to be initialised
1820 * @i: paged fragment index to initialise
1821 * @page: the page to use for this fragment
1822 * @off: the offset to the data with @page
1823 * @size: the length of the data
1824 *
1825 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001826 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001827 * addition updates @skb such that @i is the last fragment.
1828 *
1829 * Does not take any additional reference on the fragment.
1830 */
1831static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1832 struct page *page, int off, int size)
1833{
1834 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001835 skb_shinfo(skb)->nr_frags = i + 1;
1836}
1837
Joe Perches7965bd42013-09-26 14:48:15 -07001838void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1839 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001840
Jason Wangf8e617e2013-11-01 14:07:47 +08001841void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1842 unsigned int truesize);
1843
Linus Torvalds1da177e2005-04-16 15:20:36 -07001844#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001845#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001846#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1847
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001848#ifdef NET_SKBUFF_DATA_USES_OFFSET
1849static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1850{
1851 return skb->head + skb->tail;
1852}
1853
1854static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1855{
1856 skb->tail = skb->data - skb->head;
1857}
1858
1859static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1860{
1861 skb_reset_tail_pointer(skb);
1862 skb->tail += offset;
1863}
Simon Horman7cc46192013-05-28 20:34:29 +00001864
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001865#else /* NET_SKBUFF_DATA_USES_OFFSET */
1866static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1867{
1868 return skb->tail;
1869}
1870
1871static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1872{
1873 skb->tail = skb->data;
1874}
1875
1876static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1877{
1878 skb->tail = skb->data + offset;
1879}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001880
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001881#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1882
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883/*
1884 * Add data to an sk_buff
1885 */
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001886unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001887unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1889{
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001890 unsigned char *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891 SKB_LINEAR_ASSERT(skb);
1892 skb->tail += len;
1893 skb->len += len;
1894 return tmp;
1895}
1896
Joe Perches7965bd42013-09-26 14:48:15 -07001897unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1899{
1900 skb->data -= len;
1901 skb->len += len;
1902 return skb->data;
1903}
1904
Joe Perches7965bd42013-09-26 14:48:15 -07001905unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001906static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1907{
1908 skb->len -= len;
1909 BUG_ON(skb->len < skb->data_len);
1910 return skb->data += len;
1911}
1912
David S. Miller47d29642010-05-02 02:21:44 -07001913static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1914{
1915 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1916}
1917
Joe Perches7965bd42013-09-26 14:48:15 -07001918unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001919
1920static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1921{
1922 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07001923 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001924 return NULL;
1925 skb->len -= len;
1926 return skb->data += len;
1927}
1928
1929static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1930{
1931 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1932}
1933
1934static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1935{
1936 if (likely(len <= skb_headlen(skb)))
1937 return 1;
1938 if (unlikely(len > skb->len))
1939 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07001940 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001941}
1942
1943/**
1944 * skb_headroom - bytes at buffer head
1945 * @skb: buffer to check
1946 *
1947 * Return the number of bytes of free space at the head of an &sk_buff.
1948 */
Chuck Leverc2636b42007-10-23 21:07:32 -07001949static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001950{
1951 return skb->data - skb->head;
1952}
1953
1954/**
1955 * skb_tailroom - bytes at buffer end
1956 * @skb: buffer to check
1957 *
1958 * Return the number of bytes of free space at the tail of an sk_buff
1959 */
1960static inline int skb_tailroom(const struct sk_buff *skb)
1961{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001962 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001963}
1964
1965/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00001966 * skb_availroom - bytes at buffer end
1967 * @skb: buffer to check
1968 *
1969 * Return the number of bytes of free space at the tail of an sk_buff
1970 * allocated by sk_stream_alloc()
1971 */
1972static inline int skb_availroom(const struct sk_buff *skb)
1973{
Eric Dumazet16fad692013-03-14 05:40:32 +00001974 if (skb_is_nonlinear(skb))
1975 return 0;
1976
1977 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00001978}
1979
1980/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981 * skb_reserve - adjust headroom
1982 * @skb: buffer to alter
1983 * @len: bytes to move
1984 *
1985 * Increase the headroom of an empty &sk_buff by reducing the tail
1986 * room. This is only allowed for an empty buffer.
1987 */
David S. Miller8243126c2006-01-17 02:54:21 -08001988static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001989{
1990 skb->data += len;
1991 skb->tail += len;
1992}
1993
Benjamin Poirier1837b2e2016-02-29 15:03:33 -08001994/**
1995 * skb_tailroom_reserve - adjust reserved_tailroom
1996 * @skb: buffer to alter
1997 * @mtu: maximum amount of headlen permitted
1998 * @needed_tailroom: minimum amount of reserved_tailroom
1999 *
2000 * Set reserved_tailroom so that headlen can be as large as possible but
2001 * not larger than mtu and tailroom cannot be smaller than
2002 * needed_tailroom.
2003 * The required headroom should already have been reserved before using
2004 * this function.
2005 */
2006static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2007 unsigned int needed_tailroom)
2008{
2009 SKB_LINEAR_ASSERT(skb);
2010 if (mtu < skb_tailroom(skb) - needed_tailroom)
2011 /* use at most mtu */
2012 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2013 else
2014 /* use up to all available space */
2015 skb->reserved_tailroom = needed_tailroom;
2016}
2017
Tom Herbert8bce6d72014-09-29 20:22:29 -07002018#define ENCAP_TYPE_ETHER 0
2019#define ENCAP_TYPE_IPPROTO 1
2020
2021static inline void skb_set_inner_protocol(struct sk_buff *skb,
2022 __be16 protocol)
2023{
2024 skb->inner_protocol = protocol;
2025 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2026}
2027
2028static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2029 __u8 ipproto)
2030{
2031 skb->inner_ipproto = ipproto;
2032 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2033}
2034
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002035static inline void skb_reset_inner_headers(struct sk_buff *skb)
2036{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002037 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002038 skb->inner_network_header = skb->network_header;
2039 skb->inner_transport_header = skb->transport_header;
2040}
2041
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00002042static inline void skb_reset_mac_len(struct sk_buff *skb)
2043{
2044 skb->mac_len = skb->network_header - skb->mac_header;
2045}
2046
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002047static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2048 *skb)
2049{
2050 return skb->head + skb->inner_transport_header;
2051}
2052
Tom Herbert55dc5a92015-12-14 11:19:40 -08002053static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2054{
2055 return skb_inner_transport_header(skb) - skb->data;
2056}
2057
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002058static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2059{
2060 skb->inner_transport_header = skb->data - skb->head;
2061}
2062
2063static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2064 const int offset)
2065{
2066 skb_reset_inner_transport_header(skb);
2067 skb->inner_transport_header += offset;
2068}
2069
2070static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2071{
2072 return skb->head + skb->inner_network_header;
2073}
2074
2075static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2076{
2077 skb->inner_network_header = skb->data - skb->head;
2078}
2079
2080static inline void skb_set_inner_network_header(struct sk_buff *skb,
2081 const int offset)
2082{
2083 skb_reset_inner_network_header(skb);
2084 skb->inner_network_header += offset;
2085}
2086
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002087static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2088{
2089 return skb->head + skb->inner_mac_header;
2090}
2091
2092static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2093{
2094 skb->inner_mac_header = skb->data - skb->head;
2095}
2096
2097static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2098 const int offset)
2099{
2100 skb_reset_inner_mac_header(skb);
2101 skb->inner_mac_header += offset;
2102}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002103static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2104{
Cong Wang35d04612013-05-29 15:16:05 +08002105 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002106}
2107
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002108static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2109{
2110 return skb->head + skb->transport_header;
2111}
2112
2113static inline void skb_reset_transport_header(struct sk_buff *skb)
2114{
2115 skb->transport_header = skb->data - skb->head;
2116}
2117
2118static inline void skb_set_transport_header(struct sk_buff *skb,
2119 const int offset)
2120{
2121 skb_reset_transport_header(skb);
2122 skb->transport_header += offset;
2123}
2124
2125static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2126{
2127 return skb->head + skb->network_header;
2128}
2129
2130static inline void skb_reset_network_header(struct sk_buff *skb)
2131{
2132 skb->network_header = skb->data - skb->head;
2133}
2134
2135static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2136{
2137 skb_reset_network_header(skb);
2138 skb->network_header += offset;
2139}
2140
2141static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2142{
2143 return skb->head + skb->mac_header;
2144}
2145
2146static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2147{
Cong Wang35d04612013-05-29 15:16:05 +08002148 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002149}
2150
2151static inline void skb_reset_mac_header(struct sk_buff *skb)
2152{
2153 skb->mac_header = skb->data - skb->head;
2154}
2155
2156static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2157{
2158 skb_reset_mac_header(skb);
2159 skb->mac_header += offset;
2160}
2161
Timo Teräs0e3da5b2013-12-16 11:02:09 +02002162static inline void skb_pop_mac_header(struct sk_buff *skb)
2163{
2164 skb->mac_header = skb->network_header;
2165}
2166
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002167static inline void skb_probe_transport_header(struct sk_buff *skb,
2168 const int offset_hint)
2169{
2170 struct flow_keys keys;
2171
2172 if (skb_transport_header_was_set(skb))
2173 return;
Tom Herbertcd79a232015-09-01 09:24:27 -07002174 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
Tom Herbert42aecaa2015-06-04 09:16:39 -07002175 skb_set_transport_header(skb, keys.control.thoff);
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002176 else
2177 skb_set_transport_header(skb, offset_hint);
2178}
2179
Eric Dumazet03606892012-02-23 10:55:02 +00002180static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2181{
2182 if (skb_mac_header_was_set(skb)) {
2183 const unsigned char *old_mac = skb_mac_header(skb);
2184
2185 skb_set_mac_header(skb, -skb->mac_len);
2186 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2187 }
2188}
2189
Michał Mirosław04fb4512010-12-14 15:24:08 +00002190static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2191{
2192 return skb->csum_start - skb_headroom(skb);
2193}
2194
Alexander Duyck08b64fc2016-02-05 15:27:49 -08002195static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2196{
2197 return skb->head + skb->csum_start;
2198}
2199
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002200static inline int skb_transport_offset(const struct sk_buff *skb)
2201{
2202 return skb_transport_header(skb) - skb->data;
2203}
2204
2205static inline u32 skb_network_header_len(const struct sk_buff *skb)
2206{
2207 return skb->transport_header - skb->network_header;
2208}
2209
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002210static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2211{
2212 return skb->inner_transport_header - skb->inner_network_header;
2213}
2214
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002215static inline int skb_network_offset(const struct sk_buff *skb)
2216{
2217 return skb_network_header(skb) - skb->data;
2218}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03002219
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002220static inline int skb_inner_network_offset(const struct sk_buff *skb)
2221{
2222 return skb_inner_network_header(skb) - skb->data;
2223}
2224
Changli Gaof9599ce2010-08-04 04:43:44 +00002225static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2226{
2227 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2228}
2229
Linus Torvalds1da177e2005-04-16 15:20:36 -07002230/*
2231 * CPUs often take a performance hit when accessing unaligned memory
2232 * locations. The actual performance hit varies, it can be small if the
2233 * hardware handles it or large if we have to take an exception and fix it
2234 * in software.
2235 *
2236 * Since an ethernet header is 14 bytes network drivers often end up with
2237 * the IP header at an unaligned offset. The IP header can be aligned by
2238 * shifting the start of the packet by 2 bytes. Drivers should do this
2239 * with:
2240 *
Tobias Klauser8660c122009-07-13 22:48:16 +00002241 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002242 *
2243 * The downside to this alignment of the IP header is that the DMA is now
2244 * unaligned. On some architectures the cost of an unaligned DMA is high
2245 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00002246 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002247 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2248 * to be overridden.
2249 */
2250#ifndef NET_IP_ALIGN
2251#define NET_IP_ALIGN 2
2252#endif
2253
Anton Blanchard025be812006-03-31 02:27:06 -08002254/*
2255 * The networking layer reserves some headroom in skb data (via
2256 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2257 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08002258 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08002259 *
2260 * Unfortunately this headroom changes the DMA alignment of the resulting
2261 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2262 * on some architectures. An architecture can override this value,
2263 * perhaps setting it to a cacheline in size (since that will maintain
2264 * cacheline alignment of the DMA). It must be a power of 2.
2265 *
David S. Millerd6301d32009-02-08 19:24:13 -08002266 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08002267 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07002268 *
2269 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2270 * to reduce average number of cache lines per packet.
2271 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07002272 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08002273 */
2274#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07002275#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08002276#endif
2277
Joe Perches7965bd42013-09-26 14:48:15 -07002278int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002279
2280static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2281{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00002282 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07002283 WARN_ON(1);
2284 return;
2285 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002286 skb->len = len;
2287 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002288}
2289
Joe Perches7965bd42013-09-26 14:48:15 -07002290void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002291
2292static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2293{
Herbert Xu3cc0e872006-06-09 16:13:38 -07002294 if (skb->data_len)
2295 return ___pskb_trim(skb, len);
2296 __skb_trim(skb, len);
2297 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002298}
2299
2300static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2301{
2302 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2303}
2304
2305/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07002306 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2307 * @skb: buffer to alter
2308 * @len: new length
2309 *
2310 * This is identical to pskb_trim except that the caller knows that
2311 * the skb is not cloned so we should never get an error due to out-
2312 * of-memory.
2313 */
2314static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2315{
2316 int err = pskb_trim(skb, len);
2317 BUG_ON(err);
2318}
2319
2320/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002321 * skb_orphan - orphan a buffer
2322 * @skb: buffer to orphan
2323 *
2324 * If a buffer currently has an owner then we call the owner's
2325 * destructor function and make the @skb unowned. The buffer continues
2326 * to exist but is no longer charged to its former owner.
2327 */
2328static inline void skb_orphan(struct sk_buff *skb)
2329{
Eric Dumazetc34a7612013-07-30 16:11:15 -07002330 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002331 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002332 skb->destructor = NULL;
2333 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07002334 } else {
2335 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002336 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002337}
2338
2339/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002340 * skb_orphan_frags - orphan the frags contained in a buffer
2341 * @skb: buffer to orphan frags from
2342 * @gfp_mask: allocation mask for replacement pages
2343 *
2344 * For each frag in the SKB which needs a destructor (i.e. has an
2345 * owner) create a copy of that frag and release the original
2346 * page by calling the destructor.
2347 */
2348static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2349{
2350 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2351 return 0;
2352 return skb_copy_ubufs(skb, gfp_mask);
2353}
2354
2355/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356 * __skb_queue_purge - empty a list
2357 * @list: list to empty
2358 *
2359 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2360 * the list and one reference dropped. This function does not take the
2361 * list lock and the caller must hold the relevant locks to use it.
2362 */
Joe Perches7965bd42013-09-26 14:48:15 -07002363void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364static inline void __skb_queue_purge(struct sk_buff_head *list)
2365{
2366 struct sk_buff *skb;
2367 while ((skb = __skb_dequeue(list)) != NULL)
2368 kfree_skb(skb);
2369}
2370
Joe Perches7965bd42013-09-26 14:48:15 -07002371void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002372
Joe Perches7965bd42013-09-26 14:48:15 -07002373struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2374 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002375
2376/**
2377 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2378 * @dev: network device to receive on
2379 * @length: length to allocate
2380 *
2381 * Allocate a new &sk_buff and assign it a usage count of one. The
2382 * buffer has unspecified headroom built in. Users should allocate
2383 * the headroom they think they need without accounting for the
2384 * built in space. The built in space is used for optimisations.
2385 *
2386 * %NULL is returned if there is no free memory. Although this function
2387 * allocates memory it can be called from an interrupt.
2388 */
2389static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002390 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002391{
2392 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2393}
2394
Eric Dumazet6f532612012-05-18 05:12:12 +00002395/* legacy helper around __netdev_alloc_skb() */
2396static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2397 gfp_t gfp_mask)
2398{
2399 return __netdev_alloc_skb(NULL, length, gfp_mask);
2400}
2401
2402/* legacy helper around netdev_alloc_skb() */
2403static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2404{
2405 return netdev_alloc_skb(NULL, length);
2406}
2407
2408
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002409static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2410 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002411{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002412 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002413
2414 if (NET_IP_ALIGN && skb)
2415 skb_reserve(skb, NET_IP_ALIGN);
2416 return skb;
2417}
2418
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002419static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2420 unsigned int length)
2421{
2422 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2423}
2424
Alexander Duyck181edb22015-05-06 21:12:03 -07002425static inline void skb_free_frag(void *addr)
2426{
2427 __free_page_frag(addr);
2428}
2429
Alexander Duyckffde7322014-12-09 19:40:42 -08002430void *napi_alloc_frag(unsigned int fragsz);
Alexander Duyckfd11a832014-12-09 19:40:49 -08002431struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2432 unsigned int length, gfp_t gfp_mask);
2433static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2434 unsigned int length)
2435{
2436 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2437}
Jesper Dangaard Brouer795bb1c2016-02-08 13:14:59 +01002438void napi_consume_skb(struct sk_buff *skb, int budget);
2439
2440void __kfree_skb_flush(void);
Jesper Dangaard Brouer15fad712016-02-08 13:15:04 +01002441void __kfree_skb_defer(struct sk_buff *skb);
Alexander Duyckffde7322014-12-09 19:40:42 -08002442
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002443/**
Alexander Duyck71dfda52014-11-11 09:26:34 -08002444 * __dev_alloc_pages - allocate page for network Rx
2445 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2446 * @order: size of the allocation
2447 *
2448 * Allocate a new page.
2449 *
2450 * %NULL is returned if there is no free memory.
2451*/
2452static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2453 unsigned int order)
2454{
2455 /* This piece of code contains several assumptions.
2456 * 1. This is for device Rx, therefor a cold page is preferred.
2457 * 2. The expectation is the user wants a compound page.
2458 * 3. If requesting a order 0 page it will not be compound
2459 * due to the check to see if order has a value in prep_new_page
2460 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2461 * code in gfp_to_alloc_flags that should be enforcing this.
2462 */
2463 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2464
2465 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2466}
2467
2468static inline struct page *dev_alloc_pages(unsigned int order)
2469{
Neil Horman95829b32016-05-19 11:30:54 -04002470 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
Alexander Duyck71dfda52014-11-11 09:26:34 -08002471}
2472
2473/**
2474 * __dev_alloc_page - allocate a page for network Rx
2475 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2476 *
2477 * Allocate a new page.
2478 *
2479 * %NULL is returned if there is no free memory.
2480 */
2481static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2482{
2483 return __dev_alloc_pages(gfp_mask, 0);
2484}
2485
2486static inline struct page *dev_alloc_page(void)
2487{
Neil Horman95829b32016-05-19 11:30:54 -04002488 return dev_alloc_pages(0);
Alexander Duyck71dfda52014-11-11 09:26:34 -08002489}
2490
2491/**
Mel Gorman06140022012-07-31 16:44:24 -07002492 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2493 * @page: The page that was allocated from skb_alloc_page
2494 * @skb: The skb that may need pfmemalloc set
2495 */
2496static inline void skb_propagate_pfmemalloc(struct page *page,
2497 struct sk_buff *skb)
2498{
Michal Hocko2f064f32015-08-21 14:11:51 -07002499 if (page_is_pfmemalloc(page))
Mel Gorman06140022012-07-31 16:44:24 -07002500 skb->pfmemalloc = true;
2501}
2502
Eric Dumazet564824b2010-10-11 19:05:25 +00002503/**
Masanari Iidae2278672014-02-18 22:54:36 +09002504 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002505 * @frag: the paged fragment
2506 *
2507 * Returns the &struct page associated with @frag.
2508 */
2509static inline struct page *skb_frag_page(const skb_frag_t *frag)
2510{
Ian Campbella8605c62011-10-19 23:01:49 +00002511 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002512}
2513
2514/**
2515 * __skb_frag_ref - take an addition reference on a paged fragment.
2516 * @frag: the paged fragment
2517 *
2518 * Takes an additional reference on the paged fragment @frag.
2519 */
2520static inline void __skb_frag_ref(skb_frag_t *frag)
2521{
2522 get_page(skb_frag_page(frag));
2523}
2524
2525/**
2526 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2527 * @skb: the buffer
2528 * @f: the fragment offset.
2529 *
2530 * Takes an additional reference on the @f'th paged fragment of @skb.
2531 */
2532static inline void skb_frag_ref(struct sk_buff *skb, int f)
2533{
2534 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2535}
2536
2537/**
2538 * __skb_frag_unref - release a reference on a paged fragment.
2539 * @frag: the paged fragment
2540 *
2541 * Releases a reference on the paged fragment @frag.
2542 */
2543static inline void __skb_frag_unref(skb_frag_t *frag)
2544{
2545 put_page(skb_frag_page(frag));
2546}
2547
2548/**
2549 * skb_frag_unref - release a reference on a paged fragment of an skb.
2550 * @skb: the buffer
2551 * @f: the fragment offset
2552 *
2553 * Releases a reference on the @f'th paged fragment of @skb.
2554 */
2555static inline void skb_frag_unref(struct sk_buff *skb, int f)
2556{
2557 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2558}
2559
2560/**
2561 * skb_frag_address - gets the address of the data contained in a paged fragment
2562 * @frag: the paged fragment buffer
2563 *
2564 * Returns the address of the data within @frag. The page must already
2565 * be mapped.
2566 */
2567static inline void *skb_frag_address(const skb_frag_t *frag)
2568{
2569 return page_address(skb_frag_page(frag)) + frag->page_offset;
2570}
2571
2572/**
2573 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2574 * @frag: the paged fragment buffer
2575 *
2576 * Returns the address of the data within @frag. Checks that the page
2577 * is mapped and returns %NULL otherwise.
2578 */
2579static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2580{
2581 void *ptr = page_address(skb_frag_page(frag));
2582 if (unlikely(!ptr))
2583 return NULL;
2584
2585 return ptr + frag->page_offset;
2586}
2587
2588/**
2589 * __skb_frag_set_page - sets the page contained in a paged fragment
2590 * @frag: the paged fragment
2591 * @page: the page to set
2592 *
2593 * Sets the fragment @frag to contain @page.
2594 */
2595static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2596{
Ian Campbella8605c62011-10-19 23:01:49 +00002597 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002598}
2599
2600/**
2601 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2602 * @skb: the buffer
2603 * @f: the fragment offset
2604 * @page: the page to set
2605 *
2606 * Sets the @f'th fragment of @skb to contain @page.
2607 */
2608static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2609 struct page *page)
2610{
2611 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2612}
2613
Eric Dumazet400dfd32013-10-17 16:27:07 -07002614bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2615
Ian Campbell131ea662011-08-19 06:25:00 +00002616/**
2617 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002618 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002619 * @frag: the paged fragment to map
2620 * @offset: the offset within the fragment (starting at the
2621 * fragment's own offset)
2622 * @size: the number of bytes to map
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002623 * @dir: the direction of the mapping (%PCI_DMA_*)
Ian Campbell131ea662011-08-19 06:25:00 +00002624 *
2625 * Maps the page associated with @frag to @device.
2626 */
2627static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2628 const skb_frag_t *frag,
2629 size_t offset, size_t size,
2630 enum dma_data_direction dir)
2631{
2632 return dma_map_page(dev, skb_frag_page(frag),
2633 frag->page_offset + offset, size, dir);
2634}
2635
Eric Dumazet117632e2011-12-03 21:39:53 +00002636static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2637 gfp_t gfp_mask)
2638{
2639 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2640}
2641
Octavian Purdilabad93e92014-06-12 01:36:26 +03002642
2643static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2644 gfp_t gfp_mask)
2645{
2646 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2647}
2648
2649
Ian Campbell131ea662011-08-19 06:25:00 +00002650/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002651 * skb_clone_writable - is the header of a clone writable
2652 * @skb: buffer to check
2653 * @len: length up to which to write
2654 *
2655 * Returns true if modifying the header part of the cloned buffer
2656 * does not requires the data to be copied.
2657 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002658static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002659{
2660 return !skb_header_cloned(skb) &&
2661 skb_headroom(skb) + len <= skb->hdr_len;
2662}
2663
Daniel Borkmann36976492016-02-19 23:05:25 +01002664static inline int skb_try_make_writable(struct sk_buff *skb,
2665 unsigned int write_len)
2666{
2667 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2668 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2669}
2670
Herbert Xud9cc2042007-09-16 16:21:16 -07002671static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2672 int cloned)
2673{
2674 int delta = 0;
2675
Herbert Xud9cc2042007-09-16 16:21:16 -07002676 if (headroom > skb_headroom(skb))
2677 delta = headroom - skb_headroom(skb);
2678
2679 if (delta || cloned)
2680 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2681 GFP_ATOMIC);
2682 return 0;
2683}
2684
Patrick McHardy334a8132007-06-25 04:35:20 -07002685/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686 * skb_cow - copy header of skb when it is required
2687 * @skb: buffer to cow
2688 * @headroom: needed headroom
2689 *
2690 * If the skb passed lacks sufficient headroom or its data part
2691 * is shared, data is reallocated. If reallocation fails, an error
2692 * is returned and original skb is not changed.
2693 *
2694 * The result is skb with writable area skb->head...skb->tail
2695 * and at least @headroom of space at head.
2696 */
2697static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2698{
Herbert Xud9cc2042007-09-16 16:21:16 -07002699 return __skb_cow(skb, headroom, skb_cloned(skb));
2700}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002701
Herbert Xud9cc2042007-09-16 16:21:16 -07002702/**
2703 * skb_cow_head - skb_cow but only making the head writable
2704 * @skb: buffer to cow
2705 * @headroom: needed headroom
2706 *
2707 * This function is identical to skb_cow except that we replace the
2708 * skb_cloned check by skb_header_cloned. It should be used when
2709 * you only need to push on some header and do not need to modify
2710 * the data.
2711 */
2712static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2713{
2714 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002715}
2716
2717/**
2718 * skb_padto - pad an skbuff up to a minimal size
2719 * @skb: buffer to pad
2720 * @len: minimal length
2721 *
2722 * Pads up a buffer to ensure the trailing bytes exist and are
2723 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002724 * is untouched. Otherwise it is extended. Returns zero on
2725 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002726 */
Herbert Xu5b057c62006-06-23 02:06:41 -07002727static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002728{
2729 unsigned int size = skb->len;
2730 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002731 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002732 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002733}
2734
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002735/**
2736 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2737 * @skb: buffer to pad
2738 * @len: minimal length
2739 *
2740 * Pads up a buffer to ensure the trailing bytes exist and are
2741 * blanked. If the buffer already contains sufficient data it
2742 * is untouched. Otherwise it is extended. Returns zero on
2743 * success. The skb is freed on error.
2744 */
2745static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2746{
2747 unsigned int size = skb->len;
2748
2749 if (unlikely(size < len)) {
2750 len -= size;
2751 if (skb_pad(skb, len))
2752 return -ENOMEM;
2753 __skb_put(skb, len);
2754 }
2755 return 0;
2756}
2757
Linus Torvalds1da177e2005-04-16 15:20:36 -07002758static inline int skb_add_data(struct sk_buff *skb,
Al Viroaf2b0402014-11-27 21:44:24 -05002759 struct iov_iter *from, int copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760{
2761 const int off = skb->len;
2762
2763 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viroaf2b0402014-11-27 21:44:24 -05002764 __wsum csum = 0;
2765 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2766 &csum, from) == copy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002767 skb->csum = csum_block_add(skb->csum, csum, off);
2768 return 0;
2769 }
Al Viroaf2b0402014-11-27 21:44:24 -05002770 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771 return 0;
2772
2773 __skb_trim(skb, off);
2774 return -EFAULT;
2775}
2776
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002777static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2778 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779{
2780 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002781 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002782
Ian Campbellea2ab692011-08-22 23:44:58 +00002783 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00002784 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002785 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002786 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002787}
2788
Herbert Xu364c6ba2006-06-09 16:10:40 -07002789static inline int __skb_linearize(struct sk_buff *skb)
2790{
2791 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2792}
2793
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794/**
2795 * skb_linearize - convert paged skb to linear one
2796 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07002797 *
2798 * If there is no free memory -ENOMEM is returned, otherwise zero
2799 * is returned and the old skb data released.
2800 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07002801static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002802{
Herbert Xu364c6ba2006-06-09 16:10:40 -07002803 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2804}
2805
2806/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00002807 * skb_has_shared_frag - can any frag be overwritten
2808 * @skb: buffer to test
2809 *
2810 * Return true if the skb has at least one frag that might be modified
2811 * by an external entity (as in vmsplice()/sendfile())
2812 */
2813static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2814{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00002815 return skb_is_nonlinear(skb) &&
2816 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002817}
2818
2819/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07002820 * skb_linearize_cow - make sure skb is linear and writable
2821 * @skb: buffer to process
2822 *
2823 * If there is no free memory -ENOMEM is returned, otherwise zero
2824 * is returned and the old skb data released.
2825 */
2826static inline int skb_linearize_cow(struct sk_buff *skb)
2827{
2828 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2829 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002830}
2831
2832/**
2833 * skb_postpull_rcsum - update checksum for received skb after pull
2834 * @skb: buffer to update
2835 * @start: start of data before pull
2836 * @len: length of data pulled
2837 *
2838 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07002839 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2840 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002841 */
2842
2843static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f2006-03-20 22:43:56 -08002844 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002845{
Patrick McHardy84fa7932006-08-29 16:44:56 -07002846 if (skb->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002847 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002848 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
Pravin B Shelar31b33df2015-09-28 17:24:25 -07002849 skb_checksum_start_offset(skb) < 0)
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002850 skb->ip_summed = CHECKSUM_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851}
2852
Herbert Xucbb042f2006-03-20 22:43:56 -08002853unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2854
Daniel Borkmannf8ffad62016-01-07 15:50:23 +01002855static inline void skb_postpush_rcsum(struct sk_buff *skb,
2856 const void *start, unsigned int len)
2857{
2858 /* For performing the reverse operation to skb_postpull_rcsum(),
2859 * we can instead of ...
2860 *
2861 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0));
2862 *
2863 * ... just use this equivalent version here to save a few
2864 * instructions. Feeding csum of 0 in csum_partial() and later
2865 * on adding skb->csum is equivalent to feed skb->csum in the
2866 * first place.
2867 */
2868 if (skb->ip_summed == CHECKSUM_COMPLETE)
2869 skb->csum = csum_partial(start, len, skb->csum);
2870}
2871
David S. Miller7ce5a272013-12-02 17:26:05 -05002872/**
2873 * pskb_trim_rcsum - trim received skb and update checksum
2874 * @skb: buffer to trim
2875 * @len: new length
2876 *
2877 * This is exactly the same as pskb_trim except that it ensures the
2878 * checksum of received packets are still valid after the operation.
2879 */
2880
2881static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2882{
2883 if (likely(len >= skb->len))
2884 return 0;
2885 if (skb->ip_summed == CHECKSUM_COMPLETE)
2886 skb->ip_summed = CHECKSUM_NONE;
2887 return __pskb_trim(skb, len);
2888}
2889
Linus Torvalds1da177e2005-04-16 15:20:36 -07002890#define skb_queue_walk(queue, skb) \
2891 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002892 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002893 skb = skb->next)
2894
James Chapman46f89142007-04-30 00:07:31 -07002895#define skb_queue_walk_safe(queue, skb, tmp) \
2896 for (skb = (queue)->next, tmp = skb->next; \
2897 skb != (struct sk_buff *)(queue); \
2898 skb = tmp, tmp = skb->next)
2899
David S. Miller1164f522008-09-23 00:49:44 -07002900#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002901 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07002902 skb = skb->next)
2903
2904#define skb_queue_walk_from_safe(queue, skb, tmp) \
2905 for (tmp = skb->next; \
2906 skb != (struct sk_buff *)(queue); \
2907 skb = tmp, tmp = skb->next)
2908
Stephen Hemminger300ce172005-10-30 13:47:34 -08002909#define skb_queue_reverse_walk(queue, skb) \
2910 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002911 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08002912 skb = skb->prev)
2913
David S. Miller686a2952011-01-20 22:47:32 -08002914#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2915 for (skb = (queue)->prev, tmp = skb->prev; \
2916 skb != (struct sk_buff *)(queue); \
2917 skb = tmp, tmp = skb->prev)
2918
2919#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2920 for (tmp = skb->prev; \
2921 skb != (struct sk_buff *)(queue); \
2922 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002923
David S. Miller21dc3302010-08-23 00:13:46 -07002924static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07002925{
2926 return skb_shinfo(skb)->frag_list != NULL;
2927}
2928
2929static inline void skb_frag_list_init(struct sk_buff *skb)
2930{
2931 skb_shinfo(skb)->frag_list = NULL;
2932}
2933
David S. Milleree039872009-06-09 00:17:13 -07002934#define skb_walk_frags(skb, iter) \
2935 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2936
Rainer Weikusatea3793e2015-12-06 21:11:34 +00002937
2938int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
2939 const struct sk_buff *skb);
2940struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
2941 int *peeked, int *off, int *err,
2942 struct sk_buff **last);
Joe Perches7965bd42013-09-26 14:48:15 -07002943struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2944 int *peeked, int *off, int *err);
2945struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2946 int *err);
2947unsigned int datagram_poll(struct file *file, struct socket *sock,
2948 struct poll_table_struct *wait);
Al Viroc0371da2014-11-24 10:42:55 -05002949int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2950 struct iov_iter *to, int size);
David S. Miller51f3d022014-11-05 16:46:40 -05002951static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2952 struct msghdr *msg, int size)
2953{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002954 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
David S. Miller51f3d022014-11-05 16:46:40 -05002955}
Al Viroe5a4b0b2014-11-24 18:17:55 -05002956int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2957 struct msghdr *msg);
Al Viro3a654f92014-06-19 14:15:22 -04002958int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2959 struct iov_iter *from, int len);
Al Viro3a654f92014-06-19 14:15:22 -04002960int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
Joe Perches7965bd42013-09-26 14:48:15 -07002961void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
samanthakumar627d2d62016-04-05 12:41:16 -04002962void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
2963static inline void skb_free_datagram_locked(struct sock *sk,
2964 struct sk_buff *skb)
2965{
2966 __skb_free_datagram_locked(sk, skb, 0);
2967}
Joe Perches7965bd42013-09-26 14:48:15 -07002968int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07002969int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2970int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2971__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2972 int len, __wsum csum);
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002973ssize_t skb_socket_splice(struct sock *sk,
2974 struct pipe_inode_info *pipe,
2975 struct splice_pipe_desc *spd);
2976int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
Joe Perches7965bd42013-09-26 14:48:15 -07002977 struct pipe_inode_info *pipe, unsigned int len,
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002978 unsigned int flags,
2979 ssize_t (*splice_cb)(struct sock *,
2980 struct pipe_inode_info *,
2981 struct splice_pipe_desc *));
Joe Perches7965bd42013-09-26 14:48:15 -07002982void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01002983unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002984int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2985 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07002986void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2987int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2988void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01002989unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -07002990struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Vlad Yasevich0d5501c2014-08-08 14:42:13 -04002991struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
Jiri Pirkoe2195122014-11-19 14:05:01 +01002992int skb_ensure_writable(struct sk_buff *skb, int write_len);
Jiri Pirko93515d52014-11-19 14:05:02 +01002993int skb_vlan_pop(struct sk_buff *skb);
2994int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
Sowmini Varadhan6fa01cc2016-04-22 18:36:35 -07002995struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
2996 gfp_t gfp);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002997
Al Viro6ce8e9c2014-04-06 21:25:44 -04002998static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2999{
Al Viro21226ab2014-11-28 15:48:29 -05003000 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro6ce8e9c2014-04-06 21:25:44 -04003001}
3002
Al Viro7eab8d92014-04-06 21:51:23 -04003003static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3004{
Al Viroe5a4b0b2014-11-24 18:17:55 -05003005 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro7eab8d92014-04-06 21:51:23 -04003006}
3007
Daniel Borkmann2817a332013-10-30 11:50:51 +01003008struct skb_checksum_ops {
3009 __wsum (*update)(const void *mem, int len, __wsum wsum);
3010 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3011};
3012
3013__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3014 __wsum csum, const struct skb_checksum_ops *ops);
3015__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3016 __wsum csum);
3017
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003018static inline void * __must_check
3019__skb_header_pointer(const struct sk_buff *skb, int offset,
3020 int len, void *data, int hlen, void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021{
Patrick McHardy55820ee2005-07-05 14:08:10 -07003022 if (hlen - offset >= len)
David S. Miller690e36e2014-08-23 12:13:41 -07003023 return data + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003024
David S. Miller690e36e2014-08-23 12:13:41 -07003025 if (!skb ||
3026 skb_copy_bits(skb, offset, buffer, len) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003027 return NULL;
3028
3029 return buffer;
3030}
3031
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003032static inline void * __must_check
3033skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
David S. Miller690e36e2014-08-23 12:13:41 -07003034{
3035 return __skb_header_pointer(skb, offset, len, skb->data,
3036 skb_headlen(skb), buffer);
3037}
3038
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01003039/**
3040 * skb_needs_linearize - check if we need to linearize a given skb
3041 * depending on the given device features.
3042 * @skb: socket buffer to check
3043 * @features: net device features
3044 *
3045 * Returns true if either:
3046 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3047 * 2. skb is fragmented and the device does not support SG.
3048 */
3049static inline bool skb_needs_linearize(struct sk_buff *skb,
3050 netdev_features_t features)
3051{
3052 return skb_is_nonlinear(skb) &&
3053 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3054 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3055}
3056
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03003057static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3058 void *to,
3059 const unsigned int len)
3060{
3061 memcpy(to, skb->data, len);
3062}
3063
3064static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3065 const int offset, void *to,
3066 const unsigned int len)
3067{
3068 memcpy(to, skb->data + offset, len);
3069}
3070
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03003071static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3072 const void *from,
3073 const unsigned int len)
3074{
3075 memcpy(skb->data, from, len);
3076}
3077
3078static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3079 const int offset,
3080 const void *from,
3081 const unsigned int len)
3082{
3083 memcpy(skb->data + offset, from, len);
3084}
3085
Joe Perches7965bd42013-09-26 14:48:15 -07003086void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003087
Patrick Ohlyac45f602009-02-12 05:03:37 +00003088static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3089{
3090 return skb->tstamp;
3091}
3092
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003093/**
3094 * skb_get_timestamp - get timestamp from a skb
3095 * @skb: skb to get stamp from
3096 * @stamp: pointer to struct timeval to store stamp in
3097 *
3098 * Timestamps are stored in the skb as offsets to a base timestamp.
3099 * This function converts the offset back to a struct timeval and stores
3100 * it in stamp.
3101 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00003102static inline void skb_get_timestamp(const struct sk_buff *skb,
3103 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003104{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003105 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003106}
3107
Patrick Ohlyac45f602009-02-12 05:03:37 +00003108static inline void skb_get_timestampns(const struct sk_buff *skb,
3109 struct timespec *stamp)
3110{
3111 *stamp = ktime_to_timespec(skb->tstamp);
3112}
3113
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003114static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003115{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003116 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003117}
3118
Stephen Hemminger164891a2007-04-23 22:26:16 -07003119static inline ktime_t net_timedelta(ktime_t t)
3120{
3121 return ktime_sub(ktime_get_real(), t);
3122}
3123
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07003124static inline ktime_t net_invalid_timestamp(void)
3125{
3126 return ktime_set(0, 0);
3127}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003128
Alexander Duyck62bccb82014-09-04 13:31:35 -04003129struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3130
Richard Cochranc1f19b52010-07-17 08:49:36 +00003131#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3132
Joe Perches7965bd42013-09-26 14:48:15 -07003133void skb_clone_tx_timestamp(struct sk_buff *skb);
3134bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00003135
3136#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3137
3138static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3139{
3140}
3141
3142static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3143{
3144 return false;
3145}
3146
3147#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3148
3149/**
3150 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3151 *
Richard Cochranda92b192011-10-21 00:49:15 +00003152 * PHY drivers may accept clones of transmitted packets for
3153 * timestamping via their phy_driver.txtstamp method. These drivers
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003154 * must call this function to return the skb back to the stack with a
3155 * timestamp.
Richard Cochranda92b192011-10-21 00:49:15 +00003156 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00003157 * @skb: clone of the the original outgoing packet
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003158 * @hwtstamps: hardware time stamps
Richard Cochranc1f19b52010-07-17 08:49:36 +00003159 *
3160 */
3161void skb_complete_tx_timestamp(struct sk_buff *skb,
3162 struct skb_shared_hwtstamps *hwtstamps);
3163
Willem de Bruijne7fd2882014-08-04 22:11:48 -04003164void __skb_tstamp_tx(struct sk_buff *orig_skb,
3165 struct skb_shared_hwtstamps *hwtstamps,
3166 struct sock *sk, int tstype);
3167
Patrick Ohlyac45f602009-02-12 05:03:37 +00003168/**
3169 * skb_tstamp_tx - queue clone of skb with send time stamps
3170 * @orig_skb: the original outgoing packet
3171 * @hwtstamps: hardware time stamps, may be NULL if not available
3172 *
3173 * If the skb has a socket associated, then this function clones the
3174 * skb (thus sharing the actual data and optional structures), stores
3175 * the optional hardware time stamping information (if non NULL) or
3176 * generates a software time stamp (otherwise), then queues the clone
3177 * to the error queue of the socket. Errors are silently ignored.
3178 */
Joe Perches7965bd42013-09-26 14:48:15 -07003179void skb_tstamp_tx(struct sk_buff *orig_skb,
3180 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00003181
Richard Cochran4507a712010-07-17 08:48:28 +00003182static inline void sw_tx_timestamp(struct sk_buff *skb)
3183{
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003184 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3185 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
Richard Cochran4507a712010-07-17 08:48:28 +00003186 skb_tstamp_tx(skb, NULL);
3187}
3188
3189/**
3190 * skb_tx_timestamp() - Driver hook for transmit timestamping
3191 *
3192 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00003193 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00003194 *
David S. Miller73409f32013-12-27 13:04:33 -05003195 * Specifically, one should make absolutely sure that this function is
3196 * called before TX completion of this packet can trigger. Otherwise
3197 * the packet could potentially already be freed.
3198 *
Richard Cochran4507a712010-07-17 08:48:28 +00003199 * @skb: A socket buffer.
3200 */
3201static inline void skb_tx_timestamp(struct sk_buff *skb)
3202{
Richard Cochranc1f19b52010-07-17 08:49:36 +00003203 skb_clone_tx_timestamp(skb);
Richard Cochran4507a712010-07-17 08:48:28 +00003204 sw_tx_timestamp(skb);
3205}
3206
Johannes Berg6e3e9392011-11-09 10:15:42 +01003207/**
3208 * skb_complete_wifi_ack - deliver skb with wifi status
3209 *
3210 * @skb: the original outgoing packet
3211 * @acked: ack status
3212 *
3213 */
3214void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3215
Joe Perches7965bd42013-09-26 14:48:15 -07003216__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3217__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003218
Herbert Xu60476372007-04-09 11:59:39 -07003219static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3220{
Tom Herbert6edec0e2015-02-10 16:30:28 -08003221 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3222 skb->csum_valid ||
3223 (skb->ip_summed == CHECKSUM_PARTIAL &&
3224 skb_checksum_start_offset(skb) >= 0));
Herbert Xu60476372007-04-09 11:59:39 -07003225}
3226
Herbert Xufb286bb2005-11-10 13:01:24 -08003227/**
3228 * skb_checksum_complete - Calculate checksum of an entire packet
3229 * @skb: packet to process
3230 *
3231 * This function calculates the checksum over the entire packet plus
3232 * the value of skb->csum. The latter can be used to supply the
3233 * checksum of a pseudo header as used by TCP/UDP. It returns the
3234 * checksum.
3235 *
3236 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3237 * this function can be used to verify that checksum on received
3238 * packets. In that case the function should return zero if the
3239 * checksum is correct. In particular, this function will return zero
3240 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3241 * hardware has already verified the correctness of the checksum.
3242 */
Al Viro4381ca32007-07-15 21:00:11 +01003243static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08003244{
Herbert Xu60476372007-04-09 11:59:39 -07003245 return skb_csum_unnecessary(skb) ?
3246 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003247}
3248
Tom Herbert77cffe22014-08-27 21:26:46 -07003249static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3250{
3251 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3252 if (skb->csum_level == 0)
3253 skb->ip_summed = CHECKSUM_NONE;
3254 else
3255 skb->csum_level--;
3256 }
3257}
3258
3259static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3260{
3261 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3262 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3263 skb->csum_level++;
3264 } else if (skb->ip_summed == CHECKSUM_NONE) {
3265 skb->ip_summed = CHECKSUM_UNNECESSARY;
3266 skb->csum_level = 0;
3267 }
3268}
3269
Tom Herbert5a212322014-08-31 15:12:41 -07003270static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3271{
3272 /* Mark current checksum as bad (typically called from GRO
3273 * path). In the case that ip_summed is CHECKSUM_NONE
3274 * this must be the first checksum encountered in the packet.
3275 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3276 * checksum after the last one validated. For UDP, a zero
3277 * checksum can not be marked as bad.
3278 */
3279
3280 if (skb->ip_summed == CHECKSUM_NONE ||
3281 skb->ip_summed == CHECKSUM_UNNECESSARY)
3282 skb->csum_bad = 1;
3283}
3284
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003285/* Check if we need to perform checksum complete validation.
3286 *
3287 * Returns true if checksum complete is needed, false otherwise
3288 * (either checksum is unnecessary or zero checksum is allowed).
3289 */
3290static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3291 bool zero_okay,
3292 __sum16 check)
3293{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003294 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3295 skb->csum_valid = 1;
Tom Herbert77cffe22014-08-27 21:26:46 -07003296 __skb_decr_checksum_unnecessary(skb);
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003297 return false;
3298 }
3299
3300 return true;
3301}
3302
3303/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3304 * in checksum_init.
3305 */
3306#define CHECKSUM_BREAK 76
3307
Tom Herbert4e18b9a2015-04-20 14:10:04 -07003308/* Unset checksum-complete
3309 *
3310 * Unset checksum complete can be done when packet is being modified
3311 * (uncompressed for instance) and checksum-complete value is
3312 * invalidated.
3313 */
3314static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3315{
3316 if (skb->ip_summed == CHECKSUM_COMPLETE)
3317 skb->ip_summed = CHECKSUM_NONE;
3318}
3319
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003320/* Validate (init) checksum based on checksum complete.
3321 *
3322 * Return values:
3323 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3324 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3325 * checksum is stored in skb->csum for use in __skb_checksum_complete
3326 * non-zero: value of invalid checksum
3327 *
3328 */
3329static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3330 bool complete,
3331 __wsum psum)
3332{
3333 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3334 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003335 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003336 return 0;
3337 }
Tom Herbert5a212322014-08-31 15:12:41 -07003338 } else if (skb->csum_bad) {
3339 /* ip_summed == CHECKSUM_NONE in this case */
Eric Dumazetc91d4602015-05-15 05:48:07 -07003340 return (__force __sum16)1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003341 }
3342
3343 skb->csum = psum;
3344
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003345 if (complete || skb->len <= CHECKSUM_BREAK) {
3346 __sum16 csum;
3347
3348 csum = __skb_checksum_complete(skb);
3349 skb->csum_valid = !csum;
3350 return csum;
3351 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003352
3353 return 0;
3354}
3355
3356static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3357{
3358 return 0;
3359}
3360
3361/* Perform checksum validate (init). Note that this is a macro since we only
3362 * want to calculate the pseudo header which is an input function if necessary.
3363 * First we try to validate without any computation (checksum unnecessary) and
3364 * then calculate based on checksum complete calling the function to compute
3365 * pseudo header.
3366 *
3367 * Return values:
3368 * 0: checksum is validated or try to in skb_checksum_complete
3369 * non-zero: value of invalid checksum
3370 */
3371#define __skb_checksum_validate(skb, proto, complete, \
3372 zero_okay, check, compute_pseudo) \
3373({ \
3374 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003375 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003376 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3377 __ret = __skb_checksum_validate_complete(skb, \
3378 complete, compute_pseudo(skb, proto)); \
3379 __ret; \
3380})
3381
3382#define skb_checksum_init(skb, proto, compute_pseudo) \
3383 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3384
3385#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3386 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3387
3388#define skb_checksum_validate(skb, proto, compute_pseudo) \
3389 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3390
3391#define skb_checksum_validate_zero_check(skb, proto, check, \
3392 compute_pseudo) \
Sabrina Dubroca096a4cf2015-02-06 18:54:19 +01003393 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003394
3395#define skb_checksum_simple_validate(skb) \
3396 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3397
Tom Herbertd96535a2014-08-31 15:12:42 -07003398static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3399{
3400 return (skb->ip_summed == CHECKSUM_NONE &&
3401 skb->csum_valid && !skb->csum_bad);
3402}
3403
3404static inline void __skb_checksum_convert(struct sk_buff *skb,
3405 __sum16 check, __wsum pseudo)
3406{
3407 skb->csum = ~pseudo;
3408 skb->ip_summed = CHECKSUM_COMPLETE;
3409}
3410
3411#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3412do { \
3413 if (__skb_checksum_convert_check(skb)) \
3414 __skb_checksum_convert(skb, check, \
3415 compute_pseudo(skb, proto)); \
3416} while (0)
3417
Tom Herbert15e23962015-02-10 16:30:31 -08003418static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3419 u16 start, u16 offset)
3420{
3421 skb->ip_summed = CHECKSUM_PARTIAL;
3422 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3423 skb->csum_offset = offset - start;
3424}
3425
Tom Herbertdcdc8992015-02-02 16:07:34 -08003426/* Update skbuf and packet to reflect the remote checksum offload operation.
3427 * When called, ptr indicates the starting point for skb->csum when
3428 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3429 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3430 */
3431static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
Tom Herbert15e23962015-02-10 16:30:31 -08003432 int start, int offset, bool nopartial)
Tom Herbertdcdc8992015-02-02 16:07:34 -08003433{
3434 __wsum delta;
3435
Tom Herbert15e23962015-02-10 16:30:31 -08003436 if (!nopartial) {
3437 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3438 return;
3439 }
3440
Tom Herbertdcdc8992015-02-02 16:07:34 -08003441 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3442 __skb_checksum_complete(skb);
3443 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3444 }
3445
3446 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3447
3448 /* Adjust skb->csum since we changed the packet */
3449 skb->csum = csum_add(skb->csum, delta);
3450}
3451
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003452#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07003453void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003454static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3455{
3456 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07003457 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003458}
3459static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3460{
3461 if (nfct)
3462 atomic_inc(&nfct->use);
3463}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003464#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003465#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003466static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3467{
3468 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3469 kfree(nf_bridge);
3470}
3471static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3472{
3473 if (nf_bridge)
3474 atomic_inc(&nf_bridge->use);
3475}
3476#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08003477static inline void nf_reset(struct sk_buff *skb)
3478{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003479#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003480 nf_conntrack_put(skb->nfct);
3481 skb->nfct = NULL;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003482#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003483#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003484 nf_bridge_put(skb->nf_bridge);
3485 skb->nf_bridge = NULL;
3486#endif
3487}
3488
Patrick McHardy124dff02013-04-05 20:42:05 +02003489static inline void nf_reset_trace(struct sk_buff *skb)
3490{
Florian Westphal478b3602014-02-15 23:48:45 +01003491#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00003492 skb->nf_trace = 0;
3493#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003494}
3495
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003496/* Note: This doesn't put any conntrack and bridge info in dst. */
Eric Dumazetb1937222014-09-28 22:18:47 -07003497static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3498 bool copy)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003499{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003500#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003501 dst->nfct = src->nfct;
3502 nf_conntrack_get(src->nfct);
Eric Dumazetb1937222014-09-28 22:18:47 -07003503 if (copy)
3504 dst->nfctinfo = src->nfctinfo;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003505#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003506#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003507 dst->nf_bridge = src->nf_bridge;
3508 nf_bridge_get(src->nf_bridge);
3509#endif
Florian Westphal478b3602014-02-15 23:48:45 +01003510#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Eric Dumazetb1937222014-09-28 22:18:47 -07003511 if (copy)
3512 dst->nf_trace = src->nf_trace;
Florian Westphal478b3602014-02-15 23:48:45 +01003513#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003514}
3515
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003516static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3517{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003518#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003519 nf_conntrack_put(dst->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003520#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003521#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003522 nf_bridge_put(dst->nf_bridge);
3523#endif
Eric Dumazetb1937222014-09-28 22:18:47 -07003524 __nf_copy(dst, src, true);
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003525}
3526
James Morris984bc162006-06-09 00:29:17 -07003527#ifdef CONFIG_NETWORK_SECMARK
3528static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3529{
3530 to->secmark = from->secmark;
3531}
3532
3533static inline void skb_init_secmark(struct sk_buff *skb)
3534{
3535 skb->secmark = 0;
3536}
3537#else
3538static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3539{ }
3540
3541static inline void skb_init_secmark(struct sk_buff *skb)
3542{ }
3543#endif
3544
Eric W. Biederman574f7192014-04-01 12:20:24 -07003545static inline bool skb_irq_freeable(const struct sk_buff *skb)
3546{
3547 return !skb->destructor &&
3548#if IS_ENABLED(CONFIG_XFRM)
3549 !skb->sp &&
3550#endif
3551#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3552 !skb->nfct &&
3553#endif
3554 !skb->_skb_refdst &&
3555 !skb_has_frag_list(skb);
3556}
3557
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003558static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3559{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003560 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003561}
3562
Stephen Hemminger92477442009-03-21 13:39:26 -07003563static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003564{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003565 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003566}
3567
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003568static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3569{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003570 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003571}
3572
David S. Millerd5a9e242009-01-27 16:22:11 -08003573static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3574{
3575 skb->queue_mapping = rx_queue + 1;
3576}
3577
Stephen Hemminger92477442009-03-21 13:39:26 -07003578static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003579{
3580 return skb->queue_mapping - 1;
3581}
3582
Stephen Hemminger92477442009-03-21 13:39:26 -07003583static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003584{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003585 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003586}
3587
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003588static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3589{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003590#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003591 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003592#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003593 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003594#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003595}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003596
Pravin B Shelar68c33162013-02-14 14:02:41 +00003597/* Keeps track of mac header offset relative to skb->head.
3598 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3599 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003600 * tunnel skb it points to outer mac header.
3601 * Keeps track of level of encapsulation of network headers.
3602 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003603struct skb_gso_cb {
Alexander Duyck802ab552016-04-10 21:45:03 -04003604 union {
3605 int mac_offset;
3606 int data_offset;
3607 };
Eric Dumazet3347c962013-10-19 11:42:56 -07003608 int encap_level;
Alexander Duyck76443452016-02-05 15:27:37 -08003609 __wsum csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003610 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003611};
Konstantin Khlebnikov9207f9d2016-01-08 15:21:46 +03003612#define SKB_SGO_CB_OFFSET 32
3613#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
Pravin B Shelar68c33162013-02-14 14:02:41 +00003614
3615static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3616{
3617 return (skb_mac_header(inner_skb) - inner_skb->head) -
3618 SKB_GSO_CB(inner_skb)->mac_offset;
3619}
3620
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003621static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3622{
3623 int new_headroom, headroom;
3624 int ret;
3625
3626 headroom = skb_headroom(skb);
3627 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3628 if (ret)
3629 return ret;
3630
3631 new_headroom = skb_headroom(skb);
3632 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3633 return 0;
3634}
3635
Alexander Duyck08b64fc2016-02-05 15:27:49 -08003636static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3637{
3638 /* Do not update partial checksums if remote checksum is enabled. */
3639 if (skb->remcsum_offload)
3640 return;
3641
3642 SKB_GSO_CB(skb)->csum = res;
3643 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3644}
3645
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003646/* Compute the checksum for a gso segment. First compute the checksum value
3647 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3648 * then add in skb->csum (checksum from csum_start to end of packet).
3649 * skb->csum and csum_start are then updated to reflect the checksum of the
3650 * resultant packet starting from the transport header-- the resultant checksum
3651 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3652 * header.
3653 */
3654static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3655{
Alexander Duyck76443452016-02-05 15:27:37 -08003656 unsigned char *csum_start = skb_transport_header(skb);
3657 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3658 __wsum partial = SKB_GSO_CB(skb)->csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003659
Alexander Duyck76443452016-02-05 15:27:37 -08003660 SKB_GSO_CB(skb)->csum = res;
3661 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003662
Alexander Duyck76443452016-02-05 15:27:37 -08003663 return csum_fold(csum_partial(csum_start, plen, partial));
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003664}
3665
David S. Millerbdcc0922012-03-07 20:53:36 -05003666static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07003667{
3668 return skb_shinfo(skb)->gso_size;
3669}
3670
Eric Dumazet36a8f392013-09-29 01:21:32 -07003671/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05003672static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02003673{
3674 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3675}
3676
Joe Perches7965bd42013-09-26 14:48:15 -07003677void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07003678
3679static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3680{
3681 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3682 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003683 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3684
Alexander Duyckb78462e2010-06-02 12:24:37 +00003685 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3686 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07003687 __skb_warn_lro_forwarding(skb);
3688 return true;
3689 }
3690 return false;
3691}
3692
Herbert Xu35fc92a2007-03-26 23:22:20 -07003693static inline void skb_forward_csum(struct sk_buff *skb)
3694{
3695 /* Unfortunately we don't support this one. Any brave souls? */
3696 if (skb->ip_summed == CHECKSUM_COMPLETE)
3697 skb->ip_summed = CHECKSUM_NONE;
3698}
3699
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003700/**
3701 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3702 * @skb: skb to check
3703 *
3704 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3705 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3706 * use this helper, to document places where we make this assertion.
3707 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003708static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003709{
3710#ifdef DEBUG
3711 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3712#endif
3713}
3714
Rusty Russellf35d9d82008-02-04 23:49:54 -05003715bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00003716
Paul Durranted1f50c2014-01-09 10:02:46 +00003717int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
Linus Lüssing9afd85c2015-05-02 14:01:07 +02003718struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3719 unsigned int transport_len,
3720 __sum16(*skb_chkf)(struct sk_buff *skb));
Paul Durranted1f50c2014-01-09 10:02:46 +00003721
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00003722/**
3723 * skb_head_is_locked - Determine if the skb->head is locked down
3724 * @skb: skb to check
3725 *
3726 * The head on skbs build around a head frag can be removed if they are
3727 * not cloned. This function returns true if the skb head is locked down
3728 * due to either being allocated via kmalloc, or by being a clone with
3729 * multiple references to the head.
3730 */
3731static inline bool skb_head_is_locked(const struct sk_buff *skb)
3732{
3733 return !skb->head_frag || skb_cloned(skb);
3734}
Florian Westphalfe6cc552014-02-13 23:09:12 +01003735
3736/**
3737 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3738 *
3739 * @skb: GSO skb
3740 *
3741 * skb_gso_network_seglen is used to determine the real size of the
3742 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3743 *
3744 * The MAC/L2 header is not accounted for.
3745 */
3746static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3747{
3748 unsigned int hdr_len = skb_transport_header(skb) -
3749 skb_network_header(skb);
3750 return hdr_len + skb_gso_transport_seglen(skb);
3751}
Thomas Grafee122c72015-07-21 10:43:58 +02003752
Edward Cree179bc672016-02-11 20:48:04 +00003753/* Local Checksum Offload.
3754 * Compute outer checksum based on the assumption that the
3755 * inner checksum will be offloaded later.
Edward Creee8ae7b02016-02-11 21:03:37 +00003756 * See Documentation/networking/checksum-offloads.txt for
3757 * explanation of how this works.
Edward Cree179bc672016-02-11 20:48:04 +00003758 * Fill in outer checksum adjustment (e.g. with sum of outer
3759 * pseudo-header) before calling.
3760 * Also ensure that inner checksum is in linear data area.
3761 */
3762static inline __wsum lco_csum(struct sk_buff *skb)
3763{
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003764 unsigned char *csum_start = skb_checksum_start(skb);
3765 unsigned char *l4_hdr = skb_transport_header(skb);
3766 __wsum partial;
Edward Cree179bc672016-02-11 20:48:04 +00003767
3768 /* Start with complement of inner checksum adjustment */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003769 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3770 skb->csum_offset));
3771
Edward Cree179bc672016-02-11 20:48:04 +00003772 /* Add in checksum of our headers (incl. outer checksum
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003773 * adjustment filled in by caller) and return result.
Edward Cree179bc672016-02-11 20:48:04 +00003774 */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003775 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
Edward Cree179bc672016-02-11 20:48:04 +00003776}
3777
Linus Torvalds1da177e2005-04-16 15:20:36 -07003778#endif /* __KERNEL__ */
3779#endif /* _LINUX_SKBUFF_H */