blob: 5fba16658f9dd25feb06a93283e1b45317ff3740 [file] [log] [blame]
Linus Torvalds1da177e2005-04-16 15:20:36 -07001/*
2 * Definitions for the 'struct sk_buff' memory handlers.
3 *
4 * Authors:
5 * Alan Cox, <gw4pts@gw4pts.ampr.org>
6 * Florian La Roche, <rzsfl@rz.uni-sb.de>
7 *
8 * This program is free software; you can redistribute it and/or
9 * modify it under the terms of the GNU General Public License
10 * as published by the Free Software Foundation; either version
11 * 2 of the License, or (at your option) any later version.
12 */
13
14#ifndef _LINUX_SKBUFF_H
15#define _LINUX_SKBUFF_H
16
Linus Torvalds1da177e2005-04-16 15:20:36 -070017#include <linux/kernel.h>
Vegard Nossumfe55f6d2008-08-30 12:16:35 +020018#include <linux/kmemcheck.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070019#include <linux/compiler.h>
20#include <linux/time.h>
Paul Gortmaker187f1882011-11-23 20:12:59 -050021#include <linux/bug.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070022#include <linux/cache.h>
Eric Dumazet56b17422014-11-03 08:19:53 -080023#include <linux/rbtree.h>
David S. Miller51f3d022014-11-05 16:46:40 -050024#include <linux/socket.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070025
Arun Sharma600634972011-07-26 16:09:06 -070026#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070027#include <asm/types.h>
28#include <linux/spinlock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070029#include <linux/net.h>
Thomas Graf3fc7e8a2005-06-23 21:00:17 -070030#include <linux/textsearch.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070031#include <net/checksum.h>
Al Viroa80958f2006-12-04 20:41:19 +000032#include <linux/rcupdate.h>
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -070033#include <linux/hrtimer.h>
Ian Campbell131ea662011-08-19 06:25:00 +000034#include <linux/dma-mapping.h>
Michał Mirosławc8f44af2011-11-15 15:29:55 +000035#include <linux/netdev_features.h>
Eric Dumazet363ec39232014-02-26 14:02:11 -080036#include <linux/sched.h>
Jiri Pirko1bd758e2015-05-12 14:56:07 +020037#include <net/flow_dissector.h>
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +020038#include <linux/splice.h>
Bernhard Thaler72b31f72015-05-30 15:27:40 +020039#include <linux/in6.h>
Tom Herbertf70ea012015-07-31 16:52:10 -070040#include <net/flow.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070041
Tom Herbert7a6ae712015-12-14 11:19:47 -080042/* The interface for checksum offload between the stack and networking drivers
43 * is as follows...
44 *
45 * A. IP checksum related features
46 *
47 * Drivers advertise checksum offload capabilities in the features of a device.
48 * From the stack's point of view these are capabilities offered by the driver,
49 * a driver typically only advertises features that it is capable of offloading
50 * to its device.
51 *
52 * The checksum related features are:
53 *
54 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
55 * IP (one's complement) checksum for any combination
56 * of protocols or protocol layering. The checksum is
57 * computed and set in a packet per the CHECKSUM_PARTIAL
58 * interface (see below).
59 *
60 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
61 * TCP or UDP packets over IPv4. These are specifically
62 * unencapsulated packets of the form IPv4|TCP or
63 * IPv4|UDP where the Protocol field in the IPv4 header
64 * is TCP or UDP. The IPv4 header may contain IP options
65 * This feature cannot be set in features for a device
66 * with NETIF_F_HW_CSUM also set. This feature is being
67 * DEPRECATED (see below).
68 *
69 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
70 * TCP or UDP packets over IPv6. These are specifically
71 * unencapsulated packets of the form IPv6|TCP or
72 * IPv4|UDP where the Next Header field in the IPv6
73 * header is either TCP or UDP. IPv6 extension headers
74 * are not supported with this feature. This feature
75 * cannot be set in features for a device with
76 * NETIF_F_HW_CSUM also set. This feature is being
77 * DEPRECATED (see below).
78 *
79 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
80 * This flag is used only used to disable the RX checksum
81 * feature for a device. The stack will accept receive
82 * checksum indication in packets received on a device
83 * regardless of whether NETIF_F_RXCSUM is set.
84 *
85 * B. Checksumming of received packets by device. Indication of checksum
86 * verification is in set skb->ip_summed. Possible values are:
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010087 *
88 * CHECKSUM_NONE:
89 *
Tom Herbert7a6ae712015-12-14 11:19:47 -080090 * Device did not checksum this packet e.g. due to lack of capabilities.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010091 * The packet contains full (though not verified) checksum in packet but
92 * not in skb->csum. Thus, skb->csum is undefined in this case.
93 *
94 * CHECKSUM_UNNECESSARY:
95 *
96 * The hardware you're dealing with doesn't calculate the full checksum
97 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
Tom Herbert77cffe22014-08-27 21:26:46 -070098 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
99 * if their checksums are okay. skb->csum is still undefined in this case
Tom Herbert7a6ae712015-12-14 11:19:47 -0800100 * though. A driver or device must never modify the checksum field in the
101 * packet even if checksum is verified.
Tom Herbert77cffe22014-08-27 21:26:46 -0700102 *
103 * CHECKSUM_UNNECESSARY is applicable to following protocols:
104 * TCP: IPv6 and IPv4.
105 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
106 * zero UDP checksum for either IPv4 or IPv6, the networking stack
107 * may perform further validation in this case.
108 * GRE: only if the checksum is present in the header.
109 * SCTP: indicates the CRC in SCTP header has been validated.
110 *
111 * skb->csum_level indicates the number of consecutive checksums found in
112 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
113 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
114 * and a device is able to verify the checksums for UDP (possibly zero),
115 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
116 * two. If the device were only able to verify the UDP checksum and not
117 * GRE, either because it doesn't support GRE checksum of because GRE
118 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
119 * not considered in this case).
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100120 *
121 * CHECKSUM_COMPLETE:
122 *
123 * This is the most generic way. The device supplied checksum of the _whole_
124 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
125 * hardware doesn't need to parse L3/L4 headers to implement this.
126 *
127 * Note: Even if device supports only some protocols, but is able to produce
128 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
129 *
130 * CHECKSUM_PARTIAL:
131 *
Tom Herbert6edec0e2015-02-10 16:30:28 -0800132 * A checksum is set up to be offloaded to a device as described in the
133 * output description for CHECKSUM_PARTIAL. This may occur on a packet
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100134 * received directly from another Linux OS, e.g., a virtualized Linux kernel
Tom Herbert6edec0e2015-02-10 16:30:28 -0800135 * on the same host, or it may be set in the input path in GRO or remote
136 * checksum offload. For the purposes of checksum verification, the checksum
137 * referred to by skb->csum_start + skb->csum_offset and any preceding
138 * checksums in the packet are considered verified. Any checksums in the
139 * packet that are after the checksum being offloaded are not considered to
140 * be verified.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100141 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800142 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
143 * in the skb->ip_summed for a packet. Values are:
144 *
145 * CHECKSUM_PARTIAL:
146 *
147 * The driver is required to checksum the packet as seen by hard_start_xmit()
148 * from skb->csum_start up to the end, and to record/write the checksum at
149 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
150 * csum_start and csum_offset values are valid values given the length and
151 * offset of the packet, however they should not attempt to validate that the
152 * checksum refers to a legitimate transport layer checksum-- it is the
153 * purview of the stack to validate that csum_start and csum_offset are set
154 * correctly.
155 *
156 * When the stack requests checksum offload for a packet, the driver MUST
157 * ensure that the checksum is set correctly. A driver can either offload the
158 * checksum calculation to the device, or call skb_checksum_help (in the case
159 * that the device does not support offload for a particular checksum).
160 *
161 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
162 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
163 * checksum offload capability. If a device has limited checksum capabilities
164 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as
165 * described above) a helper function can be called to resolve
166 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper
167 * function takes a spec argument that describes the protocol layer that is
168 * supported for checksum offload and can be called for each packet. If a
169 * packet does not match the specification for offload, skb_checksum_help
170 * is called to resolve the checksum.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100171 *
172 * CHECKSUM_NONE:
173 *
174 * The skb was already checksummed by the protocol, or a checksum is not
175 * required.
176 *
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100177 * CHECKSUM_UNNECESSARY:
178 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800179 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
180 * output.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100181 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800182 * CHECKSUM_COMPLETE:
183 * Not used in checksum output. If a driver observes a packet with this value
184 * set in skbuff, if should treat as CHECKSUM_NONE being set.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100185 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800186 * D. Non-IP checksum (CRC) offloads
187 *
188 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
189 * offloading the SCTP CRC in a packet. To perform this offload the stack
190 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
191 * accordingly. Note the there is no indication in the skbuff that the
192 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports
193 * both IP checksum offload and SCTP CRC offload must verify which offload
194 * is configured for a packet presumably by inspecting packet headers.
195 *
196 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
197 * offloading the FCOE CRC in a packet. To perform this offload the stack
198 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
199 * accordingly. Note the there is no indication in the skbuff that the
200 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
201 * both IP checksum offload and FCOE CRC offload must verify which offload
202 * is configured for a packet presumably by inspecting packet headers.
203 *
204 * E. Checksumming on output with GSO.
205 *
206 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
207 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
208 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
209 * part of the GSO operation is implied. If a checksum is being offloaded
210 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
211 * are set to refer to the outermost checksum being offload (two offloaded
212 * checksums are possible with UDP encapsulation).
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100213 */
214
Herbert Xu60476372007-04-09 11:59:39 -0700215/* Don't change this without changing skb_csum_unnecessary! */
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100216#define CHECKSUM_NONE 0
217#define CHECKSUM_UNNECESSARY 1
218#define CHECKSUM_COMPLETE 2
219#define CHECKSUM_PARTIAL 3
Linus Torvalds1da177e2005-04-16 15:20:36 -0700220
Tom Herbert77cffe22014-08-27 21:26:46 -0700221/* Maximum value in skb->csum_level */
222#define SKB_MAX_CSUM_LEVEL 3
223
Tobias Klauser0bec8c82014-07-22 12:06:23 +0200224#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
David S. Millerfc910a22007-03-25 20:27:59 -0700225#define SKB_WITH_OVERHEAD(X) \
Herbert Xudeea84b2007-10-21 16:27:46 -0700226 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
David S. Millerfc910a22007-03-25 20:27:59 -0700227#define SKB_MAX_ORDER(X, ORDER) \
228 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700229#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
230#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
231
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000232/* return minimum truesize of one skb containing X bytes of data */
233#define SKB_TRUESIZE(X) ((X) + \
234 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
235 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
236
Linus Torvalds1da177e2005-04-16 15:20:36 -0700237struct net_device;
David Howells716ea3a2007-04-02 20:19:53 -0700238struct scatterlist;
Jens Axboe9c55e012007-11-06 23:30:13 -0800239struct pipe_inode_info;
Herbert Xua8f820aa2014-11-07 21:22:22 +0800240struct iov_iter;
Alexander Duyckfd11a832014-12-09 19:40:49 -0800241struct napi_struct;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700243#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700244struct nf_conntrack {
245 atomic_t use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700246};
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700247#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700248
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200249#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700250struct nf_bridge_info {
Eric Dumazetbf1ac5c2012-04-18 23:19:25 +0000251 atomic_t use;
Florian Westphal3eaf4022015-04-02 14:31:44 +0200252 enum {
253 BRNF_PROTO_UNCHANGED,
254 BRNF_PROTO_8021Q,
255 BRNF_PROTO_PPPOE
Florian Westphal7fb48c52015-05-03 22:05:28 +0200256 } orig_proto:8;
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200257 u8 pkt_otherhost:1;
258 u8 in_prerouting:1;
259 u8 bridged_dnat:1;
Bernhard Thaler411ffb42015-05-30 15:28:28 +0200260 __u16 frag_max_size;
Eric Dumazetbf1ac5c2012-04-18 23:19:25 +0000261 struct net_device *physindev;
Florian Westphal63cdbc02015-09-14 17:06:27 +0200262
263 /* always valid & non-NULL from FORWARD on, for physdev match */
264 struct net_device *physoutdev;
Florian Westphal7fb48c52015-05-03 22:05:28 +0200265 union {
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200266 /* prerouting: detect dnat in orig/reply direction */
Bernhard Thaler72b31f72015-05-30 15:27:40 +0200267 __be32 ipv4_daddr;
268 struct in6_addr ipv6_daddr;
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200269
270 /* after prerouting + nat detected: store original source
271 * mac since neigh resolution overwrites it, only used while
272 * skb is out in neigh layer.
273 */
274 char neigh_header[8];
Bernhard Thaler72b31f72015-05-30 15:27:40 +0200275 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700276};
277#endif
278
Linus Torvalds1da177e2005-04-16 15:20:36 -0700279struct sk_buff_head {
280 /* These two members must be first. */
281 struct sk_buff *next;
282 struct sk_buff *prev;
283
284 __u32 qlen;
285 spinlock_t lock;
286};
287
288struct sk_buff;
289
Ian Campbell9d4dde52011-12-22 23:39:14 +0000290/* To allow 64K frame to be packed as single skb without frag_list we
291 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
292 * buffers which do not start on a page boundary.
293 *
294 * Since GRO uses frags we allocate at least 16 regardless of page
295 * size.
Anton Blancharda715dea2011-03-27 14:57:26 +0000296 */
Ian Campbell9d4dde52011-12-22 23:39:14 +0000297#if (65536/PAGE_SIZE + 1) < 16
David S. Millereec00952011-03-29 23:34:08 -0700298#define MAX_SKB_FRAGS 16UL
Anton Blancharda715dea2011-03-27 14:57:26 +0000299#else
Ian Campbell9d4dde52011-12-22 23:39:14 +0000300#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
Anton Blancharda715dea2011-03-27 14:57:26 +0000301#endif
Hans Westgaard Ry5f74f822016-02-03 09:26:57 +0100302extern int sysctl_max_skb_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700303
304typedef struct skb_frag_struct skb_frag_t;
305
306struct skb_frag_struct {
Ian Campbella8605c62011-10-19 23:01:49 +0000307 struct {
308 struct page *p;
309 } page;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000310#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
David S. Millera309bb02007-07-30 18:47:03 -0700311 __u32 page_offset;
312 __u32 size;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000313#else
314 __u16 page_offset;
315 __u16 size;
316#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700317};
318
Eric Dumazet9e903e02011-10-18 21:00:24 +0000319static inline unsigned int skb_frag_size(const skb_frag_t *frag)
320{
321 return frag->size;
322}
323
324static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
325{
326 frag->size = size;
327}
328
329static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
330{
331 frag->size += delta;
332}
333
334static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
335{
336 frag->size -= delta;
337}
338
Patrick Ohlyac45f602009-02-12 05:03:37 +0000339#define HAVE_HW_TIME_STAMP
340
341/**
Randy Dunlapd3a21be2009-03-02 03:15:58 -0800342 * struct skb_shared_hwtstamps - hardware time stamps
Patrick Ohlyac45f602009-02-12 05:03:37 +0000343 * @hwtstamp: hardware time stamp transformed into duration
344 * since arbitrary point in time
Patrick Ohlyac45f602009-02-12 05:03:37 +0000345 *
346 * Software time stamps generated by ktime_get_real() are stored in
Willem de Bruijn4d276eb2014-07-25 18:01:32 -0400347 * skb->tstamp.
Patrick Ohlyac45f602009-02-12 05:03:37 +0000348 *
349 * hwtstamps can only be compared against other hwtstamps from
350 * the same device.
351 *
352 * This structure is attached to packets as part of the
353 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
354 */
355struct skb_shared_hwtstamps {
356 ktime_t hwtstamp;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000357};
358
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000359/* Definitions for tx_flags in struct skb_shared_info */
360enum {
361 /* generate hardware time stamp */
362 SKBTX_HW_TSTAMP = 1 << 0,
363
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400364 /* generate software time stamp when queueing packet to NIC */
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000365 SKBTX_SW_TSTAMP = 1 << 1,
366
367 /* device driver is going to provide hardware time stamp */
368 SKBTX_IN_PROGRESS = 1 << 2,
369
Shirley Maa6686f22011-07-06 12:22:12 +0000370 /* device driver supports TX zero-copy buffers */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000371 SKBTX_DEV_ZEROCOPY = 1 << 3,
Johannes Berg6e3e9392011-11-09 10:15:42 +0100372
373 /* generate wifi status information (where possible) */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000374 SKBTX_WIFI_STATUS = 1 << 4,
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000375
376 /* This indicates at least one fragment might be overwritten
377 * (as in vmsplice(), sendfile() ...)
378 * If we need to compute a TX checksum, we'll need to copy
379 * all frags to avoid possible bad checksum
380 */
381 SKBTX_SHARED_FRAG = 1 << 5,
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400382
383 /* generate software time stamp when entering packet scheduling */
384 SKBTX_SCHED_TSTAMP = 1 << 6,
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400385
386 /* generate software timestamp on peer data acknowledgment */
387 SKBTX_ACK_TSTAMP = 1 << 7,
Shirley Maa6686f22011-07-06 12:22:12 +0000388};
389
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400390#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
391 SKBTX_SCHED_TSTAMP | \
392 SKBTX_ACK_TSTAMP)
Willem de Bruijnf24b9be2014-08-04 22:11:45 -0400393#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
394
Shirley Maa6686f22011-07-06 12:22:12 +0000395/*
396 * The callback notifies userspace to release buffers when skb DMA is done in
397 * lower device, the skb last reference should be 0 when calling this.
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000398 * The zerocopy_success argument is true if zero copy transmit occurred,
399 * false on data copy or out of memory error caused by data copy attempt.
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000400 * The ctx field is used to track device context.
401 * The desc field is used to track userspace buffer index.
Shirley Maa6686f22011-07-06 12:22:12 +0000402 */
403struct ubuf_info {
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000404 void (*callback)(struct ubuf_info *, bool zerocopy_success);
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000405 void *ctx;
Shirley Maa6686f22011-07-06 12:22:12 +0000406 unsigned long desc;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000407};
408
Linus Torvalds1da177e2005-04-16 15:20:36 -0700409/* This data is invariant across clones and lives at
410 * the end of the header data, ie. at skb->end.
411 */
412struct skb_shared_info {
Ian Campbell9f42f122012-01-05 07:13:39 +0000413 unsigned char nr_frags;
414 __u8 tx_flags;
Herbert Xu79671682006-06-22 02:40:14 -0700415 unsigned short gso_size;
416 /* Warning: this field is not always filled in (UFO)! */
417 unsigned short gso_segs;
418 unsigned short gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700419 struct sk_buff *frag_list;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000420 struct skb_shared_hwtstamps hwtstamps;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400421 u32 tskey;
Ian Campbell9f42f122012-01-05 07:13:39 +0000422 __be32 ip6_frag_id;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700423
424 /*
425 * Warning : all fields before dataref are cleared in __alloc_skb()
426 */
427 atomic_t dataref;
428
Johann Baudy69e3c752009-05-18 22:11:22 -0700429 /* Intermediate layers must ensure that destructor_arg
430 * remains valid until skb destructor */
431 void * destructor_arg;
Shirley Maa6686f22011-07-06 12:22:12 +0000432
Eric Dumazetfed66382010-07-22 19:09:08 +0000433 /* must be last field, see pskb_expand_head() */
434 skb_frag_t frags[MAX_SKB_FRAGS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700435};
436
437/* We divide dataref into two halves. The higher 16 bits hold references
438 * to the payload part of skb->data. The lower 16 bits hold references to
Patrick McHardy334a8132007-06-25 04:35:20 -0700439 * the entire skb->data. A clone of a headerless skb holds the length of
440 * the header in skb->hdr_len.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700441 *
442 * All users must obey the rule that the skb->data reference count must be
443 * greater than or equal to the payload reference count.
444 *
445 * Holding a reference to the payload part means that the user does not
446 * care about modifications to the header part of skb->data.
447 */
448#define SKB_DATAREF_SHIFT 16
449#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
450
David S. Millerd179cd12005-08-17 14:57:30 -0700451
452enum {
Vijay Subramanianc8753d52014-10-02 10:00:43 -0700453 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
454 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
455 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
David S. Millerd179cd12005-08-17 14:57:30 -0700456};
457
Herbert Xu79671682006-06-22 02:40:14 -0700458enum {
459 SKB_GSO_TCPV4 = 1 << 0,
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700460 SKB_GSO_UDP = 1 << 1,
Herbert Xu576a30e2006-06-27 13:22:38 -0700461
462 /* This indicates the skb is from an untrusted source. */
463 SKB_GSO_DODGY = 1 << 2,
Michael Chanb0da85372006-06-29 12:30:00 -0700464
465 /* This indicates the tcp segment has CWR set. */
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700466 SKB_GSO_TCP_ECN = 1 << 3,
467
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400468 SKB_GSO_TCP_FIXEDID = 1 << 4,
Chris Leech01d5b2f2009-02-27 14:06:49 -0800469
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400470 SKB_GSO_TCPV6 = 1 << 5,
Pravin B Shelar68c33162013-02-14 14:02:41 +0000471
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400472 SKB_GSO_FCOE = 1 << 6,
Pravin B Shelar73136262013-03-07 13:21:51 +0000473
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400474 SKB_GSO_GRE = 1 << 7,
Simon Horman0d89d202013-05-23 21:02:52 +0000475
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400476 SKB_GSO_GRE_CSUM = 1 << 8,
Eric Dumazetcb32f512013-10-19 11:42:57 -0700477
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400478 SKB_GSO_IPIP = 1 << 9,
Eric Dumazet61c1db72013-10-20 20:47:30 -0700479
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400480 SKB_GSO_SIT = 1 << 10,
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700481
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400482 SKB_GSO_UDP_TUNNEL = 1 << 11,
Tom Herbert4749c092014-06-04 17:20:23 -0700483
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400484 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12,
485
486 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
Herbert Xu79671682006-06-22 02:40:14 -0700487};
488
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700489#if BITS_PER_LONG > 32
490#define NET_SKBUFF_DATA_USES_OFFSET 1
491#endif
492
493#ifdef NET_SKBUFF_DATA_USES_OFFSET
494typedef unsigned int sk_buff_data_t;
495#else
496typedef unsigned char *sk_buff_data_t;
497#endif
498
Eric Dumazet363ec39232014-02-26 14:02:11 -0800499/**
500 * struct skb_mstamp - multi resolution time stamps
501 * @stamp_us: timestamp in us resolution
502 * @stamp_jiffies: timestamp in jiffies
503 */
504struct skb_mstamp {
505 union {
506 u64 v64;
507 struct {
508 u32 stamp_us;
509 u32 stamp_jiffies;
510 };
511 };
512};
513
514/**
515 * skb_mstamp_get - get current timestamp
516 * @cl: place to store timestamps
517 */
518static inline void skb_mstamp_get(struct skb_mstamp *cl)
519{
520 u64 val = local_clock();
521
522 do_div(val, NSEC_PER_USEC);
523 cl->stamp_us = (u32)val;
524 cl->stamp_jiffies = (u32)jiffies;
525}
526
527/**
528 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
529 * @t1: pointer to newest sample
530 * @t0: pointer to oldest sample
531 */
532static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
533 const struct skb_mstamp *t0)
534{
535 s32 delta_us = t1->stamp_us - t0->stamp_us;
536 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
537
538 /* If delta_us is negative, this might be because interval is too big,
539 * or local_clock() drift is too big : fallback using jiffies.
540 */
541 if (delta_us <= 0 ||
542 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
543
544 delta_us = jiffies_to_usecs(delta_jiffies);
545
546 return delta_us;
547}
548
Yuchung Cheng625a5e12015-10-16 21:57:45 -0700549static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
550 const struct skb_mstamp *t0)
551{
552 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
553
554 if (!diff)
555 diff = t1->stamp_us - t0->stamp_us;
556 return diff > 0;
557}
Eric Dumazet363ec39232014-02-26 14:02:11 -0800558
Linus Torvalds1da177e2005-04-16 15:20:36 -0700559/**
560 * struct sk_buff - socket buffer
561 * @next: Next buffer in list
562 * @prev: Previous buffer in list
Eric Dumazet363ec39232014-02-26 14:02:11 -0800563 * @tstamp: Time we arrived/left
Eric Dumazet56b17422014-11-03 08:19:53 -0800564 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700565 * @sk: Socket we are owned by
Linus Torvalds1da177e2005-04-16 15:20:36 -0700566 * @dev: Device we arrived on/are leaving by
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700567 * @cb: Control buffer. Free for use by every layer. Put private vars here
Eric Dumazet7fee2262010-05-11 23:19:48 +0000568 * @_skb_refdst: destination entry (with norefcount bit)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700569 * @sp: the security path, used for xfrm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700570 * @len: Length of actual data
571 * @data_len: Data length
572 * @mac_len: Length of link layer header
Patrick McHardy334a8132007-06-25 04:35:20 -0700573 * @hdr_len: writable header length of cloned skb
Herbert Xu663ead32007-04-09 11:59:07 -0700574 * @csum: Checksum (must include start/offset pair)
575 * @csum_start: Offset from skb->head where checksumming should start
576 * @csum_offset: Offset from csum_start where checksum should be stored
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700577 * @priority: Packet queueing priority
WANG Cong60ff7462014-05-04 16:39:18 -0700578 * @ignore_df: allow local fragmentation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700579 * @cloned: Head may be cloned (check refcnt to be sure)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700580 * @ip_summed: Driver fed us an IP checksum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700581 * @nohdr: Payload reference only, must not modify header
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700582 * @nfctinfo: Relationship of this skb to the connection
Linus Torvalds1da177e2005-04-16 15:20:36 -0700583 * @pkt_type: Packet class
Randy Dunlapc83c2482005-10-18 22:07:41 -0700584 * @fclone: skbuff clone status
Randy Dunlapc83c2482005-10-18 22:07:41 -0700585 * @ipvs_property: skbuff is owned by ipvs
Randy Dunlap31729362008-02-18 20:52:13 -0800586 * @peeked: this packet has been seen already, so stats have been
587 * done for it, don't do them again
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700588 * @nf_trace: netfilter packet trace flag
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700589 * @protocol: Packet protocol from driver
590 * @destructor: Destruct function
591 * @nfct: Associated connection, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
Eric Dumazet8964be42009-11-20 15:35:04 -0800593 * @skb_iif: ifindex of device we arrived on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700594 * @tc_index: Traffic control index
595 * @tc_verd: traffic control verdict
Tom Herbert61b905d2014-03-24 15:34:47 -0700596 * @hash: the packet hash
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700597 * @queue_mapping: Queue mapping for multiqueue devices
David S. Miller0b725a22014-08-25 15:51:53 -0700598 * @xmit_more: More SKBs are pending for this queue
Randy Dunlap553a5672008-04-20 10:51:01 -0700599 * @ndisc_nodetype: router type (from link layer)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700600 * @ooo_okay: allow the mapping of a socket to a queue to be changed
Tom Herbert61b905d2014-03-24 15:34:47 -0700601 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
Changli Gao4ca24622011-08-19 07:26:44 -0700602 * ports.
Tom Herberta3b18dd2014-07-01 21:33:17 -0700603 * @sw_hash: indicates hash was computed in software stack
Johannes Berg6e3e9392011-11-09 10:15:42 +0100604 * @wifi_acked_valid: wifi_acked was set
605 * @wifi_acked: whether frame was acked on wifi or not
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000606 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
Eliezer Tamir06021292013-06-10 11:39:50 +0300607 * @napi_id: id of the NAPI struct this skb came from
James Morris984bc162006-06-09 00:29:17 -0700608 * @secmark: security marking
Scott Feldman0c4f6912015-07-18 18:24:48 -0700609 * @offload_fwd_mark: fwding offload mark
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700610 * @mark: Generic packet mark
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000611 * @vlan_proto: vlan encapsulation protocol
Patrick McHardy6aa895b2008-07-14 22:49:06 -0700612 * @vlan_tci: vlan tag control information
Simon Horman0d89d202013-05-23 21:02:52 +0000613 * @inner_protocol: Protocol (encapsulation)
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000614 * @inner_transport_header: Inner transport layer header (encapsulation)
615 * @inner_network_header: Network layer header (encapsulation)
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000616 * @inner_mac_header: Link layer header (encapsulation)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700617 * @transport_header: Transport layer header
618 * @network_header: Network layer header
619 * @mac_header: Link layer header
620 * @tail: Tail pointer
621 * @end: End pointer
622 * @head: Head of buffer
623 * @data: Data head pointer
624 * @truesize: Buffer size
625 * @users: User count - see {datagram,tcp}.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700626 */
627
628struct sk_buff {
Eric Dumazet363ec39232014-02-26 14:02:11 -0800629 union {
Eric Dumazet56b17422014-11-03 08:19:53 -0800630 struct {
631 /* These two members must be first. */
632 struct sk_buff *next;
633 struct sk_buff *prev;
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000634
Eric Dumazet56b17422014-11-03 08:19:53 -0800635 union {
636 ktime_t tstamp;
637 struct skb_mstamp skb_mstamp;
638 };
639 };
640 struct rb_node rbnode; /* used in netem & tcp stack */
641 };
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000642 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700643 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700644
Linus Torvalds1da177e2005-04-16 15:20:36 -0700645 /*
646 * This is the control buffer. It is free to use for every
647 * layer. Please put your private variables there. If you
648 * want to keep them across layers you have to do a skb_clone()
649 * first. This is owned by whoever has the skb queued ATM.
650 */
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000651 char cb[48] __aligned(8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700652
Eric Dumazet7fee2262010-05-11 23:19:48 +0000653 unsigned long _skb_refdst;
Eric Dumazetb1937222014-09-28 22:18:47 -0700654 void (*destructor)(struct sk_buff *skb);
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000655#ifdef CONFIG_XFRM
656 struct sec_path *sp;
657#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700658#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
659 struct nf_conntrack *nfct;
660#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200661#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Eric Dumazetb1937222014-09-28 22:18:47 -0700662 struct nf_bridge_info *nf_bridge;
663#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700664 unsigned int len,
Patrick McHardy334a8132007-06-25 04:35:20 -0700665 data_len;
666 __u16 mac_len,
667 hdr_len;
Eric Dumazetb1937222014-09-28 22:18:47 -0700668
669 /* Following fields are _not_ copied in __copy_skb_header()
670 * Note that queue_mapping is here mostly to fill a hole.
671 */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200672 kmemcheck_bitfield_begin(flags1);
Eric Dumazetb1937222014-09-28 22:18:47 -0700673 __u16 queue_mapping;
674 __u8 cloned:1,
Harald Welte6869c4d2005-08-09 19:24:19 -0700675 nohdr:1,
Patrick McHardyb84f4cc2005-11-20 21:19:21 -0800676 fclone:2,
Herbert Xua59322b2007-12-05 01:53:40 -0800677 peeked:1,
Eric Dumazetb1937222014-09-28 22:18:47 -0700678 head_frag:1,
679 xmit_more:1;
680 /* one bit hole */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200681 kmemcheck_bitfield_end(flags1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682
Eric Dumazetb1937222014-09-28 22:18:47 -0700683 /* fields enclosed in headers_start/headers_end are copied
684 * using a single memcpy() in __copy_skb_header()
685 */
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700686 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700687 __u32 headers_start[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700688 /* public: */
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200689
690/* if you move pkt_type around you also must adapt those constants */
691#ifdef __BIG_ENDIAN_BITFIELD
692#define PKT_TYPE_MAX (7 << 5)
693#else
694#define PKT_TYPE_MAX 7
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100695#endif
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200696#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
697
698 __u8 __pkt_type_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700699 __u8 pkt_type:3;
700 __u8 pfmemalloc:1;
701 __u8 ignore_df:1;
702 __u8 nfctinfo:3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700703
Eric Dumazetb1937222014-09-28 22:18:47 -0700704 __u8 nf_trace:1;
705 __u8 ip_summed:2;
706 __u8 ooo_okay:1;
707 __u8 l4_hash:1;
708 __u8 sw_hash:1;
709 __u8 wifi_acked_valid:1;
710 __u8 wifi_acked:1;
711
712 __u8 no_fcs:1;
713 /* Indicates the inner headers are valid in the skbuff. */
714 __u8 encapsulation:1;
715 __u8 encap_hdr_csum:1;
716 __u8 csum_valid:1;
717 __u8 csum_complete_sw:1;
718 __u8 csum_level:2;
719 __u8 csum_bad:1;
720
721#ifdef CONFIG_IPV6_NDISC_NODETYPE
722 __u8 ndisc_nodetype:2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700723#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700724 __u8 ipvs_property:1;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700725 __u8 inner_protocol_type:1;
Tom Herberte585f232014-11-04 09:06:54 -0800726 __u8 remcsum_offload:1;
727 /* 3 or 5 bit hole */
Alexander Duyck4031ae62012-01-27 06:22:53 +0000728
Linus Torvalds1da177e2005-04-16 15:20:36 -0700729#ifdef CONFIG_NET_SCHED
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700730 __u16 tc_index; /* traffic control index */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731#ifdef CONFIG_NET_CLS_ACT
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700732 __u16 tc_verd; /* traffic control verdict */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700733#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700734#endif
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200735
Eric Dumazetb1937222014-09-28 22:18:47 -0700736 union {
737 __wsum csum;
738 struct {
739 __u16 csum_start;
740 __u16 csum_offset;
741 };
742 };
743 __u32 priority;
744 int skb_iif;
745 __u32 hash;
746 __be16 vlan_proto;
747 __u16 vlan_tci;
Eric Dumazet2bd82482015-02-03 23:48:24 -0800748#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
749 union {
750 unsigned int napi_id;
751 unsigned int sender_cpu;
752 };
Chris Leech97fc2f02006-05-23 17:55:33 -0700753#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700754 union {
James Morris984bc162006-06-09 00:29:17 -0700755#ifdef CONFIG_NETWORK_SECMARK
Scott Feldman0c4f6912015-07-18 18:24:48 -0700756 __u32 secmark;
James Morris984bc162006-06-09 00:29:17 -0700757#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700758#ifdef CONFIG_NET_SWITCHDEV
759 __u32 offload_fwd_mark;
760#endif
761 };
762
Neil Horman3b885782009-10-12 13:26:31 -0700763 union {
764 __u32 mark;
Eric Dumazet16fad692013-03-14 05:40:32 +0000765 __u32 reserved_tailroom;
Neil Horman3b885782009-10-12 13:26:31 -0700766 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700767
Tom Herbert8bce6d72014-09-29 20:22:29 -0700768 union {
769 __be16 inner_protocol;
770 __u8 inner_ipproto;
771 };
772
Simon Horman1a37e412013-05-23 21:02:51 +0000773 __u16 inner_transport_header;
774 __u16 inner_network_header;
775 __u16 inner_mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700776
777 __be16 protocol;
Simon Horman1a37e412013-05-23 21:02:51 +0000778 __u16 transport_header;
779 __u16 network_header;
780 __u16 mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700781
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700782 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700783 __u32 headers_end[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700784 /* public: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700785
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 /* These elements must be at the end, see alloc_skb() for details. */
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700787 sk_buff_data_t tail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700788 sk_buff_data_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700789 unsigned char *head,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700790 *data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700791 unsigned int truesize;
792 atomic_t users;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793};
794
795#ifdef __KERNEL__
796/*
797 * Handling routines are only of interest to the kernel
798 */
799#include <linux/slab.h>
800
Linus Torvalds1da177e2005-04-16 15:20:36 -0700801
Mel Gormanc93bdd02012-07-31 16:44:19 -0700802#define SKB_ALLOC_FCLONE 0x01
803#define SKB_ALLOC_RX 0x02
Alexander Duyckfd11a832014-12-09 19:40:49 -0800804#define SKB_ALLOC_NAPI 0x04
Mel Gormanc93bdd02012-07-31 16:44:19 -0700805
806/* Returns true if the skb was allocated from PFMEMALLOC reserves */
807static inline bool skb_pfmemalloc(const struct sk_buff *skb)
808{
809 return unlikely(skb->pfmemalloc);
810}
811
Eric Dumazet7fee2262010-05-11 23:19:48 +0000812/*
813 * skb might have a dst pointer attached, refcounted or not.
814 * _skb_refdst low order bit is set if refcount was _not_ taken
815 */
816#define SKB_DST_NOREF 1UL
817#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
818
819/**
820 * skb_dst - returns skb dst_entry
821 * @skb: buffer
822 *
823 * Returns skb dst_entry, regardless of reference taken or not.
824 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000825static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
826{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000827 /* If refdst was not refcounted, check we still are in a
828 * rcu_read_lock section
829 */
830 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
831 !rcu_read_lock_held() &&
832 !rcu_read_lock_bh_held());
833 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
Eric Dumazetadf30902009-06-02 05:19:30 +0000834}
835
Eric Dumazet7fee2262010-05-11 23:19:48 +0000836/**
837 * skb_dst_set - sets skb dst
838 * @skb: buffer
839 * @dst: dst entry
840 *
841 * Sets skb dst, assuming a reference was taken on dst and should
842 * be released by skb_dst_drop()
843 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000844static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
845{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000846 skb->_skb_refdst = (unsigned long)dst;
847}
848
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200849/**
850 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
851 * @skb: buffer
852 * @dst: dst entry
853 *
854 * Sets skb dst, assuming a reference was not taken on dst.
855 * If dst entry is cached, we do not take reference and dst_release
856 * will be avoided by refdst_drop. If dst entry is not cached, we take
857 * reference, so that last dst_release can destroy the dst immediately.
858 */
859static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
860{
Hannes Frederic Sowadbfc4fb2014-12-06 19:19:42 +0100861 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
862 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200863}
Eric Dumazet7fee2262010-05-11 23:19:48 +0000864
865/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300866 * skb_dst_is_noref - Test if skb dst isn't refcounted
Eric Dumazet7fee2262010-05-11 23:19:48 +0000867 * @skb: buffer
868 */
869static inline bool skb_dst_is_noref(const struct sk_buff *skb)
870{
871 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
Eric Dumazetadf30902009-06-02 05:19:30 +0000872}
873
Eric Dumazet511c3f92009-06-02 05:14:27 +0000874static inline struct rtable *skb_rtable(const struct sk_buff *skb)
875{
Eric Dumazetadf30902009-06-02 05:19:30 +0000876 return (struct rtable *)skb_dst(skb);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000877}
878
Joe Perches7965bd42013-09-26 14:48:15 -0700879void kfree_skb(struct sk_buff *skb);
880void kfree_skb_list(struct sk_buff *segs);
881void skb_tx_error(struct sk_buff *skb);
882void consume_skb(struct sk_buff *skb);
883void __kfree_skb(struct sk_buff *skb);
Eric Dumazetd7e88832012-04-30 08:10:34 +0000884extern struct kmem_cache *skbuff_head_cache;
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000885
Joe Perches7965bd42013-09-26 14:48:15 -0700886void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
887bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
888 bool *fragstolen, int *delta_truesize);
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000889
Joe Perches7965bd42013-09-26 14:48:15 -0700890struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
891 int node);
Eric Dumazet2ea2f622015-04-24 16:05:01 -0700892struct sk_buff *__build_skb(void *data, unsigned int frag_size);
Joe Perches7965bd42013-09-26 14:48:15 -0700893struct sk_buff *build_skb(void *data, unsigned int frag_size);
David S. Millerd179cd12005-08-17 14:57:30 -0700894static inline struct sk_buff *alloc_skb(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100895 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700896{
Eric Dumazet564824b2010-10-11 19:05:25 +0000897 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700898}
899
Eric Dumazet2e4e4412014-09-17 04:49:49 -0700900struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
901 unsigned long data_len,
902 int max_page_order,
903 int *errcode,
904 gfp_t gfp_mask);
905
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700906/* Layout of fast clones : [skb1][skb2][fclone_ref] */
907struct sk_buff_fclones {
908 struct sk_buff skb1;
909
910 struct sk_buff skb2;
911
912 atomic_t fclone_ref;
913};
914
915/**
916 * skb_fclone_busy - check if fclone is busy
917 * @skb: buffer
918 *
Masanari Iidabda13fe2015-12-13 16:53:02 +0900919 * Returns true if skb is a fast clone, and its clone is not freed.
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700920 * Some drivers call skb_orphan() in their ndo_start_xmit(),
921 * so we also check that this didnt happen.
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700922 */
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700923static inline bool skb_fclone_busy(const struct sock *sk,
924 const struct sk_buff *skb)
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700925{
926 const struct sk_buff_fclones *fclones;
927
928 fclones = container_of(skb, struct sk_buff_fclones, skb1);
929
930 return skb->fclone == SKB_FCLONE_ORIG &&
Eric Dumazet6ffe75e2014-12-03 17:04:39 -0800931 atomic_read(&fclones->fclone_ref) > 1 &&
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700932 fclones->skb2.sk == sk;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700933}
934
David S. Millerd179cd12005-08-17 14:57:30 -0700935static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100936 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700937{
Mel Gormanc93bdd02012-07-31 16:44:19 -0700938 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700939}
940
Joe Perches7965bd42013-09-26 14:48:15 -0700941struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000942static inline struct sk_buff *alloc_skb_head(gfp_t priority)
943{
944 return __alloc_skb_head(priority, -1);
945}
946
Joe Perches7965bd42013-09-26 14:48:15 -0700947struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
948int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
949struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
950struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
Octavian Purdilabad93e92014-06-12 01:36:26 +0300951struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
952 gfp_t gfp_mask, bool fclone);
953static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
954 gfp_t gfp_mask)
955{
956 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
957}
Eric Dumazet117632e2011-12-03 21:39:53 +0000958
Joe Perches7965bd42013-09-26 14:48:15 -0700959int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
960struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
961 unsigned int headroom);
962struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
963 int newtailroom, gfp_t priority);
Fan Du25a91d82014-01-18 09:54:23 +0800964int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
965 int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -0700966int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
967 int len);
968int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
969int skb_pad(struct sk_buff *skb, int pad);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000970#define dev_kfree_skb(a) consume_skb(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700971
Joe Perches7965bd42013-09-26 14:48:15 -0700972int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
973 int getfrag(void *from, char *to, int offset,
974 int len, int odd, struct sk_buff *skb),
975 void *from, int length);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700976
Hannes Frederic Sowabe12a1f2015-05-21 16:59:58 +0200977int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
978 int offset, size_t size);
979
Eric Dumazetd94d9fe2009-11-04 09:50:58 -0800980struct skb_seq_state {
Thomas Graf677e90e2005-06-23 20:59:51 -0700981 __u32 lower_offset;
982 __u32 upper_offset;
983 __u32 frag_idx;
984 __u32 stepped_offset;
985 struct sk_buff *root_skb;
986 struct sk_buff *cur_skb;
987 __u8 *frag_data;
988};
989
Joe Perches7965bd42013-09-26 14:48:15 -0700990void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
991 unsigned int to, struct skb_seq_state *st);
992unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
993 struct skb_seq_state *st);
994void skb_abort_seq_read(struct skb_seq_state *st);
Thomas Graf677e90e2005-06-23 20:59:51 -0700995
Joe Perches7965bd42013-09-26 14:48:15 -0700996unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
Bojan Prtvar059a2442015-02-22 11:46:35 +0100997 unsigned int to, struct ts_config *config);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -0700998
Tom Herbert09323cc2013-12-15 22:16:19 -0800999/*
1000 * Packet hash types specify the type of hash in skb_set_hash.
1001 *
1002 * Hash types refer to the protocol layer addresses which are used to
1003 * construct a packet's hash. The hashes are used to differentiate or identify
1004 * flows of the protocol layer for the hash type. Hash types are either
1005 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1006 *
1007 * Properties of hashes:
1008 *
1009 * 1) Two packets in different flows have different hash values
1010 * 2) Two packets in the same flow should have the same hash value
1011 *
1012 * A hash at a higher layer is considered to be more specific. A driver should
1013 * set the most specific hash possible.
1014 *
1015 * A driver cannot indicate a more specific hash than the layer at which a hash
1016 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1017 *
1018 * A driver may indicate a hash level which is less specific than the
1019 * actual layer the hash was computed on. For instance, a hash computed
1020 * at L4 may be considered an L3 hash. This should only be done if the
1021 * driver can't unambiguously determine that the HW computed the hash at
1022 * the higher layer. Note that the "should" in the second property above
1023 * permits this.
1024 */
1025enum pkt_hash_types {
1026 PKT_HASH_TYPE_NONE, /* Undefined type */
1027 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1028 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1029 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1030};
1031
Tom Herbertbcc83832015-09-01 09:24:24 -07001032static inline void skb_clear_hash(struct sk_buff *skb)
1033{
1034 skb->hash = 0;
1035 skb->sw_hash = 0;
1036 skb->l4_hash = 0;
1037}
1038
1039static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1040{
1041 if (!skb->l4_hash)
1042 skb_clear_hash(skb);
1043}
1044
1045static inline void
1046__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1047{
1048 skb->l4_hash = is_l4;
1049 skb->sw_hash = is_sw;
1050 skb->hash = hash;
1051}
1052
Tom Herbert09323cc2013-12-15 22:16:19 -08001053static inline void
1054skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1055{
Tom Herbertbcc83832015-09-01 09:24:24 -07001056 /* Used by drivers to set hash from HW */
1057 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1058}
1059
1060static inline void
1061__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1062{
1063 __skb_set_hash(skb, hash, true, is_l4);
Tom Herbert09323cc2013-12-15 22:16:19 -08001064}
1065
Tom Herberte5276932015-09-01 09:24:23 -07001066void __skb_get_hash(struct sk_buff *skb);
1067u32 skb_get_poff(const struct sk_buff *skb);
1068u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1069 const struct flow_keys *keys, int hlen);
1070__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1071 void *data, int hlen_proto);
1072
1073static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1074 int thoff, u8 ip_proto)
1075{
1076 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1077}
1078
1079void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1080 const struct flow_dissector_key *key,
1081 unsigned int key_count);
1082
1083bool __skb_flow_dissect(const struct sk_buff *skb,
1084 struct flow_dissector *flow_dissector,
1085 void *target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001086 void *data, __be16 proto, int nhoff, int hlen,
1087 unsigned int flags);
Tom Herberte5276932015-09-01 09:24:23 -07001088
1089static inline bool skb_flow_dissect(const struct sk_buff *skb,
1090 struct flow_dissector *flow_dissector,
Tom Herbertcd79a232015-09-01 09:24:27 -07001091 void *target_container, unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001092{
1093 return __skb_flow_dissect(skb, flow_dissector, target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001094 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001095}
1096
1097static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
Tom Herbertcd79a232015-09-01 09:24:27 -07001098 struct flow_keys *flow,
1099 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001100{
1101 memset(flow, 0, sizeof(*flow));
1102 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001103 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001104}
1105
1106static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1107 void *data, __be16 proto,
Tom Herbertcd79a232015-09-01 09:24:27 -07001108 int nhoff, int hlen,
1109 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001110{
1111 memset(flow, 0, sizeof(*flow));
1112 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001113 data, proto, nhoff, hlen, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001114}
1115
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001116static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001117{
Tom Herberta3b18dd2014-07-01 21:33:17 -07001118 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001119 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001120
Tom Herbert61b905d2014-03-24 15:34:47 -07001121 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001122}
1123
David S. Miller20a17bf2015-09-01 21:19:17 -07001124__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
Tom Herbertf70ea012015-07-31 16:52:10 -07001125
David S. Miller20a17bf2015-09-01 21:19:17 -07001126static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
Tom Herbertf70ea012015-07-31 16:52:10 -07001127{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001128 if (!skb->l4_hash && !skb->sw_hash) {
1129 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001130 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001131
Tom Herbertde4c1f82015-09-01 18:11:04 -07001132 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001133 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001134
1135 return skb->hash;
1136}
1137
David S. Miller20a17bf2015-09-01 21:19:17 -07001138__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
Tom Herbertf70ea012015-07-31 16:52:10 -07001139
David S. Miller20a17bf2015-09-01 21:19:17 -07001140static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
Tom Herbertf70ea012015-07-31 16:52:10 -07001141{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001142 if (!skb->l4_hash && !skb->sw_hash) {
1143 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001144 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001145
Tom Herbertde4c1f82015-09-01 18:11:04 -07001146 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001147 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001148
1149 return skb->hash;
1150}
1151
Tom Herbert50fb7992015-05-01 11:30:12 -07001152__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1153
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001154static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1155{
Tom Herbert61b905d2014-03-24 15:34:47 -07001156 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001157}
1158
Tom Herbert3df7a742013-12-15 22:16:29 -08001159static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1160{
Tom Herbert61b905d2014-03-24 15:34:47 -07001161 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -07001162 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -07001163 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -08001164};
1165
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001166#ifdef NET_SKBUFF_DATA_USES_OFFSET
1167static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1168{
1169 return skb->head + skb->end;
1170}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001171
1172static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1173{
1174 return skb->end;
1175}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001176#else
1177static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1178{
1179 return skb->end;
1180}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001181
1182static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1183{
1184 return skb->end - skb->head;
1185}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001186#endif
1187
Linus Torvalds1da177e2005-04-16 15:20:36 -07001188/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001189#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001190
Patrick Ohlyac45f602009-02-12 05:03:37 +00001191static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1192{
1193 return &skb_shinfo(skb)->hwtstamps;
1194}
1195
Linus Torvalds1da177e2005-04-16 15:20:36 -07001196/**
1197 * skb_queue_empty - check if a queue is empty
1198 * @list: queue head
1199 *
1200 * Returns true if the queue is empty, false otherwise.
1201 */
1202static inline int skb_queue_empty(const struct sk_buff_head *list)
1203{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001204 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001205}
1206
1207/**
David S. Millerfc7ebb22008-09-23 00:34:07 -07001208 * skb_queue_is_last - check if skb is the last entry in the queue
1209 * @list: queue head
1210 * @skb: buffer
1211 *
1212 * Returns true if @skb is the last buffer on the list.
1213 */
1214static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1215 const struct sk_buff *skb)
1216{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001217 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -07001218}
1219
1220/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001221 * skb_queue_is_first - check if skb is the first entry in the queue
1222 * @list: queue head
1223 * @skb: buffer
1224 *
1225 * Returns true if @skb is the first buffer on the list.
1226 */
1227static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1228 const struct sk_buff *skb)
1229{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001230 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001231}
1232
1233/**
David S. Miller249c8b42008-09-23 00:44:42 -07001234 * skb_queue_next - return the next packet in the queue
1235 * @list: queue head
1236 * @skb: current buffer
1237 *
1238 * Return the next packet in @list after @skb. It is only valid to
1239 * call this if skb_queue_is_last() evaluates to false.
1240 */
1241static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1242 const struct sk_buff *skb)
1243{
1244 /* This BUG_ON may seem severe, but if we just return then we
1245 * are going to dereference garbage.
1246 */
1247 BUG_ON(skb_queue_is_last(list, skb));
1248 return skb->next;
1249}
1250
1251/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001252 * skb_queue_prev - return the prev packet in the queue
1253 * @list: queue head
1254 * @skb: current buffer
1255 *
1256 * Return the prev packet in @list before @skb. It is only valid to
1257 * call this if skb_queue_is_first() evaluates to false.
1258 */
1259static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1260 const struct sk_buff *skb)
1261{
1262 /* This BUG_ON may seem severe, but if we just return then we
1263 * are going to dereference garbage.
1264 */
1265 BUG_ON(skb_queue_is_first(list, skb));
1266 return skb->prev;
1267}
1268
1269/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001270 * skb_get - reference buffer
1271 * @skb: buffer to reference
1272 *
1273 * Makes another reference to a socket buffer and returns a pointer
1274 * to the buffer.
1275 */
1276static inline struct sk_buff *skb_get(struct sk_buff *skb)
1277{
1278 atomic_inc(&skb->users);
1279 return skb;
1280}
1281
1282/*
1283 * If users == 1, we are the only owner and are can avoid redundant
1284 * atomic change.
1285 */
1286
1287/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001288 * skb_cloned - is the buffer a clone
1289 * @skb: buffer to check
1290 *
1291 * Returns true if the buffer was generated with skb_clone() and is
1292 * one of multiple shared copies of the buffer. Cloned buffers are
1293 * shared data so must not be written to under normal circumstances.
1294 */
1295static inline int skb_cloned(const struct sk_buff *skb)
1296{
1297 return skb->cloned &&
1298 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1299}
1300
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001301static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1302{
Mel Gormand0164ad2015-11-06 16:28:21 -08001303 might_sleep_if(gfpflags_allow_blocking(pri));
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001304
1305 if (skb_cloned(skb))
1306 return pskb_expand_head(skb, 0, 0, pri);
1307
1308 return 0;
1309}
1310
Linus Torvalds1da177e2005-04-16 15:20:36 -07001311/**
1312 * skb_header_cloned - is the header a clone
1313 * @skb: buffer to check
1314 *
1315 * Returns true if modifying the header part of the buffer requires
1316 * the data to be copied.
1317 */
1318static inline int skb_header_cloned(const struct sk_buff *skb)
1319{
1320 int dataref;
1321
1322 if (!skb->cloned)
1323 return 0;
1324
1325 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1326 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1327 return dataref != 1;
1328}
1329
1330/**
1331 * skb_header_release - release reference to header
1332 * @skb: buffer to operate on
1333 *
1334 * Drop a reference to the header part of the buffer. This is done
1335 * by acquiring a payload reference. You must not read from the header
1336 * part of skb->data after this.
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001337 * Note : Check if you can use __skb_header_release() instead.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001338 */
1339static inline void skb_header_release(struct sk_buff *skb)
1340{
1341 BUG_ON(skb->nohdr);
1342 skb->nohdr = 1;
1343 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1344}
1345
1346/**
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001347 * __skb_header_release - release reference to header
1348 * @skb: buffer to operate on
1349 *
1350 * Variant of skb_header_release() assuming skb is private to caller.
1351 * We can avoid one atomic operation.
1352 */
1353static inline void __skb_header_release(struct sk_buff *skb)
1354{
1355 skb->nohdr = 1;
1356 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1357}
1358
1359
1360/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001361 * skb_shared - is the buffer shared
1362 * @skb: buffer to check
1363 *
1364 * Returns true if more than one person has a reference to this
1365 * buffer.
1366 */
1367static inline int skb_shared(const struct sk_buff *skb)
1368{
1369 return atomic_read(&skb->users) != 1;
1370}
1371
1372/**
1373 * skb_share_check - check if buffer is shared and if so clone it
1374 * @skb: buffer to check
1375 * @pri: priority for memory allocation
1376 *
1377 * If the buffer is shared the buffer is cloned and the old copy
1378 * drops a reference. A new clone with a single reference is returned.
1379 * If the buffer is not shared the original buffer is returned. When
1380 * being called from interrupt status or with spinlocks held pri must
1381 * be GFP_ATOMIC.
1382 *
1383 * NULL is returned on a memory allocation failure.
1384 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001385static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001386{
Mel Gormand0164ad2015-11-06 16:28:21 -08001387 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001388 if (skb_shared(skb)) {
1389 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001390
1391 if (likely(nskb))
1392 consume_skb(skb);
1393 else
1394 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001395 skb = nskb;
1396 }
1397 return skb;
1398}
1399
1400/*
1401 * Copy shared buffers into a new sk_buff. We effectively do COW on
1402 * packets to handle cases where we have a local reader and forward
1403 * and a couple of other messy ones. The normal one is tcpdumping
1404 * a packet thats being forwarded.
1405 */
1406
1407/**
1408 * skb_unshare - make a copy of a shared buffer
1409 * @skb: buffer to check
1410 * @pri: priority for memory allocation
1411 *
1412 * If the socket buffer is a clone then this function creates a new
1413 * copy of the data, drops a reference count on the old copy and returns
1414 * the new copy with the reference count at 1. If the buffer is not a clone
1415 * the original buffer is returned. When called with a spinlock held or
1416 * from interrupt state @pri must be %GFP_ATOMIC
1417 *
1418 * %NULL is returned on a memory allocation failure.
1419 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001420static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001421 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001422{
Mel Gormand0164ad2015-11-06 16:28:21 -08001423 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001424 if (skb_cloned(skb)) {
1425 struct sk_buff *nskb = skb_copy(skb, pri);
Alexander Aring31eff812014-10-10 23:10:47 +02001426
1427 /* Free our shared copy */
1428 if (likely(nskb))
1429 consume_skb(skb);
1430 else
1431 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432 skb = nskb;
1433 }
1434 return skb;
1435}
1436
1437/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001438 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439 * @list_: list to peek at
1440 *
1441 * Peek an &sk_buff. Unlike most other operations you _MUST_
1442 * be careful with this one. A peek leaves the buffer on the
1443 * list and someone else may run off with it. You must hold
1444 * the appropriate locks or have a private queue to do this.
1445 *
1446 * Returns %NULL for an empty list or a pointer to the head element.
1447 * The reference count is not incremented and the reference is therefore
1448 * volatile. Use with caution.
1449 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001450static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001451{
Eric Dumazet18d07002012-04-30 16:31:46 +00001452 struct sk_buff *skb = list_->next;
1453
1454 if (skb == (struct sk_buff *)list_)
1455 skb = NULL;
1456 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001457}
1458
1459/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001460 * skb_peek_next - peek skb following the given one from a queue
1461 * @skb: skb to start from
1462 * @list_: list to peek at
1463 *
1464 * Returns %NULL when the end of the list is met or a pointer to the
1465 * next element. The reference count is not incremented and the
1466 * reference is therefore volatile. Use with caution.
1467 */
1468static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1469 const struct sk_buff_head *list_)
1470{
1471 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001472
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001473 if (next == (struct sk_buff *)list_)
1474 next = NULL;
1475 return next;
1476}
1477
1478/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001479 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 * @list_: list to peek at
1481 *
1482 * Peek an &sk_buff. Unlike most other operations you _MUST_
1483 * be careful with this one. A peek leaves the buffer on the
1484 * list and someone else may run off with it. You must hold
1485 * the appropriate locks or have a private queue to do this.
1486 *
1487 * Returns %NULL for an empty list or a pointer to the tail element.
1488 * The reference count is not incremented and the reference is therefore
1489 * volatile. Use with caution.
1490 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001491static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001492{
Eric Dumazet18d07002012-04-30 16:31:46 +00001493 struct sk_buff *skb = list_->prev;
1494
1495 if (skb == (struct sk_buff *)list_)
1496 skb = NULL;
1497 return skb;
1498
Linus Torvalds1da177e2005-04-16 15:20:36 -07001499}
1500
1501/**
1502 * skb_queue_len - get queue length
1503 * @list_: list to measure
1504 *
1505 * Return the length of an &sk_buff queue.
1506 */
1507static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1508{
1509 return list_->qlen;
1510}
1511
David S. Miller67fed452008-09-21 22:36:24 -07001512/**
1513 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1514 * @list: queue to initialize
1515 *
1516 * This initializes only the list and queue length aspects of
1517 * an sk_buff_head object. This allows to initialize the list
1518 * aspects of an sk_buff_head without reinitializing things like
1519 * the spinlock. It can also be used for on-stack sk_buff_head
1520 * objects where the spinlock is known to not be used.
1521 */
1522static inline void __skb_queue_head_init(struct sk_buff_head *list)
1523{
1524 list->prev = list->next = (struct sk_buff *)list;
1525 list->qlen = 0;
1526}
1527
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001528/*
1529 * This function creates a split out lock class for each invocation;
1530 * this is needed for now since a whole lot of users of the skb-queue
1531 * infrastructure in drivers have different locking usage (in hardirq)
1532 * than the networking core (in softirq only). In the long run either the
1533 * network layer or drivers should need annotation to consolidate the
1534 * main types of usage into 3 classes.
1535 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536static inline void skb_queue_head_init(struct sk_buff_head *list)
1537{
1538 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001539 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001540}
1541
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001542static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1543 struct lock_class_key *class)
1544{
1545 skb_queue_head_init(list);
1546 lockdep_set_class(&list->lock, class);
1547}
1548
Linus Torvalds1da177e2005-04-16 15:20:36 -07001549/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001550 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 *
1552 * The "__skb_xxxx()" functions are the non-atomic ones that
1553 * can only be called with interrupts disabled.
1554 */
Joe Perches7965bd42013-09-26 14:48:15 -07001555void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1556 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001557static inline void __skb_insert(struct sk_buff *newsk,
1558 struct sk_buff *prev, struct sk_buff *next,
1559 struct sk_buff_head *list)
1560{
1561 newsk->next = next;
1562 newsk->prev = prev;
1563 next->prev = prev->next = newsk;
1564 list->qlen++;
1565}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001566
David S. Miller67fed452008-09-21 22:36:24 -07001567static inline void __skb_queue_splice(const struct sk_buff_head *list,
1568 struct sk_buff *prev,
1569 struct sk_buff *next)
1570{
1571 struct sk_buff *first = list->next;
1572 struct sk_buff *last = list->prev;
1573
1574 first->prev = prev;
1575 prev->next = first;
1576
1577 last->next = next;
1578 next->prev = last;
1579}
1580
1581/**
1582 * skb_queue_splice - join two skb lists, this is designed for stacks
1583 * @list: the new list to add
1584 * @head: the place to add it in the first list
1585 */
1586static inline void skb_queue_splice(const struct sk_buff_head *list,
1587 struct sk_buff_head *head)
1588{
1589 if (!skb_queue_empty(list)) {
1590 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001591 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001592 }
1593}
1594
1595/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001596 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001597 * @list: the new list to add
1598 * @head: the place to add it in the first list
1599 *
1600 * The list at @list is reinitialised
1601 */
1602static inline void skb_queue_splice_init(struct sk_buff_head *list,
1603 struct sk_buff_head *head)
1604{
1605 if (!skb_queue_empty(list)) {
1606 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001607 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001608 __skb_queue_head_init(list);
1609 }
1610}
1611
1612/**
1613 * skb_queue_splice_tail - join two skb lists, each list being a queue
1614 * @list: the new list to add
1615 * @head: the place to add it in the first list
1616 */
1617static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1618 struct sk_buff_head *head)
1619{
1620 if (!skb_queue_empty(list)) {
1621 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001622 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001623 }
1624}
1625
1626/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001627 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001628 * @list: the new list to add
1629 * @head: the place to add it in the first list
1630 *
1631 * Each of the lists is a queue.
1632 * The list at @list is reinitialised
1633 */
1634static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1635 struct sk_buff_head *head)
1636{
1637 if (!skb_queue_empty(list)) {
1638 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001639 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001640 __skb_queue_head_init(list);
1641 }
1642}
1643
Linus Torvalds1da177e2005-04-16 15:20:36 -07001644/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001645 * __skb_queue_after - queue a buffer at the list head
1646 * @list: list to use
1647 * @prev: place after this buffer
1648 * @newsk: buffer to queue
1649 *
1650 * Queue a buffer int the middle of a list. This function takes no locks
1651 * and you must therefore hold required locks before calling it.
1652 *
1653 * A buffer cannot be placed on two lists at the same time.
1654 */
1655static inline void __skb_queue_after(struct sk_buff_head *list,
1656 struct sk_buff *prev,
1657 struct sk_buff *newsk)
1658{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001659 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001660}
1661
Joe Perches7965bd42013-09-26 14:48:15 -07001662void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1663 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001664
Gerrit Renkerf5572852008-04-14 00:05:28 -07001665static inline void __skb_queue_before(struct sk_buff_head *list,
1666 struct sk_buff *next,
1667 struct sk_buff *newsk)
1668{
1669 __skb_insert(newsk, next->prev, next, list);
1670}
1671
Stephen Hemminger300ce172005-10-30 13:47:34 -08001672/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001673 * __skb_queue_head - queue a buffer at the list head
1674 * @list: list to use
1675 * @newsk: buffer to queue
1676 *
1677 * Queue a buffer at the start of a list. This function takes no locks
1678 * and you must therefore hold required locks before calling it.
1679 *
1680 * A buffer cannot be placed on two lists at the same time.
1681 */
Joe Perches7965bd42013-09-26 14:48:15 -07001682void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001683static inline void __skb_queue_head(struct sk_buff_head *list,
1684 struct sk_buff *newsk)
1685{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001686 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001687}
1688
1689/**
1690 * __skb_queue_tail - queue a buffer at the list tail
1691 * @list: list to use
1692 * @newsk: buffer to queue
1693 *
1694 * Queue a buffer at the end of a list. This function takes no locks
1695 * and you must therefore hold required locks before calling it.
1696 *
1697 * A buffer cannot be placed on two lists at the same time.
1698 */
Joe Perches7965bd42013-09-26 14:48:15 -07001699void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001700static inline void __skb_queue_tail(struct sk_buff_head *list,
1701 struct sk_buff *newsk)
1702{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001703 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001704}
1705
Linus Torvalds1da177e2005-04-16 15:20:36 -07001706/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707 * remove sk_buff from list. _Must_ be called atomically, and with
1708 * the list known..
1709 */
Joe Perches7965bd42013-09-26 14:48:15 -07001710void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001711static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1712{
1713 struct sk_buff *next, *prev;
1714
1715 list->qlen--;
1716 next = skb->next;
1717 prev = skb->prev;
1718 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001719 next->prev = prev;
1720 prev->next = next;
1721}
1722
Gerrit Renkerf525c062008-04-14 00:04:12 -07001723/**
1724 * __skb_dequeue - remove from the head of the queue
1725 * @list: list to dequeue from
1726 *
1727 * Remove the head of the list. This function does not take any locks
1728 * so must be used with appropriate locks held only. The head item is
1729 * returned or %NULL if the list is empty.
1730 */
Joe Perches7965bd42013-09-26 14:48:15 -07001731struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001732static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1733{
1734 struct sk_buff *skb = skb_peek(list);
1735 if (skb)
1736 __skb_unlink(skb, list);
1737 return skb;
1738}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001739
1740/**
1741 * __skb_dequeue_tail - remove from the tail of the queue
1742 * @list: list to dequeue from
1743 *
1744 * Remove the tail of the list. This function does not take any locks
1745 * so must be used with appropriate locks held only. The tail item is
1746 * returned or %NULL if the list is empty.
1747 */
Joe Perches7965bd42013-09-26 14:48:15 -07001748struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001749static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1750{
1751 struct sk_buff *skb = skb_peek_tail(list);
1752 if (skb)
1753 __skb_unlink(skb, list);
1754 return skb;
1755}
1756
1757
David S. Millerbdcc0922012-03-07 20:53:36 -05001758static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001759{
1760 return skb->data_len;
1761}
1762
1763static inline unsigned int skb_headlen(const struct sk_buff *skb)
1764{
1765 return skb->len - skb->data_len;
1766}
1767
1768static inline int skb_pagelen(const struct sk_buff *skb)
1769{
1770 int i, len = 0;
1771
1772 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001773 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001774 return len + skb_headlen(skb);
1775}
1776
Ian Campbell131ea662011-08-19 06:25:00 +00001777/**
1778 * __skb_fill_page_desc - initialise a paged fragment in an skb
1779 * @skb: buffer containing fragment to be initialised
1780 * @i: paged fragment index to initialise
1781 * @page: the page to use for this fragment
1782 * @off: the offset to the data with @page
1783 * @size: the length of the data
1784 *
1785 * Initialises the @i'th fragment of @skb to point to &size bytes at
1786 * offset @off within @page.
1787 *
1788 * Does not take any additional reference on the fragment.
1789 */
1790static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1791 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792{
1793 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1794
Mel Gormanc48a11c2012-07-31 16:44:23 -07001795 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001796 * Propagate page pfmemalloc to the skb if we can. The problem is
1797 * that not all callers have unique ownership of the page but rely
1798 * on page_is_pfmemalloc doing the right thing(tm).
Mel Gormanc48a11c2012-07-31 16:44:23 -07001799 */
Ian Campbella8605c62011-10-19 23:01:49 +00001800 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001801 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001802 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001803
1804 page = compound_head(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001805 if (page_is_pfmemalloc(page))
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001806 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001807}
1808
1809/**
1810 * skb_fill_page_desc - initialise a paged fragment in an skb
1811 * @skb: buffer containing fragment to be initialised
1812 * @i: paged fragment index to initialise
1813 * @page: the page to use for this fragment
1814 * @off: the offset to the data with @page
1815 * @size: the length of the data
1816 *
1817 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001818 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001819 * addition updates @skb such that @i is the last fragment.
1820 *
1821 * Does not take any additional reference on the fragment.
1822 */
1823static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1824 struct page *page, int off, int size)
1825{
1826 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001827 skb_shinfo(skb)->nr_frags = i + 1;
1828}
1829
Joe Perches7965bd42013-09-26 14:48:15 -07001830void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1831 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001832
Jason Wangf8e617e2013-11-01 14:07:47 +08001833void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1834 unsigned int truesize);
1835
Linus Torvalds1da177e2005-04-16 15:20:36 -07001836#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001837#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1839
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001840#ifdef NET_SKBUFF_DATA_USES_OFFSET
1841static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1842{
1843 return skb->head + skb->tail;
1844}
1845
1846static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1847{
1848 skb->tail = skb->data - skb->head;
1849}
1850
1851static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1852{
1853 skb_reset_tail_pointer(skb);
1854 skb->tail += offset;
1855}
Simon Horman7cc46192013-05-28 20:34:29 +00001856
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001857#else /* NET_SKBUFF_DATA_USES_OFFSET */
1858static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1859{
1860 return skb->tail;
1861}
1862
1863static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1864{
1865 skb->tail = skb->data;
1866}
1867
1868static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1869{
1870 skb->tail = skb->data + offset;
1871}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001872
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001873#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1874
Linus Torvalds1da177e2005-04-16 15:20:36 -07001875/*
1876 * Add data to an sk_buff
1877 */
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001878unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001879unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001880static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1881{
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001882 unsigned char *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883 SKB_LINEAR_ASSERT(skb);
1884 skb->tail += len;
1885 skb->len += len;
1886 return tmp;
1887}
1888
Joe Perches7965bd42013-09-26 14:48:15 -07001889unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1891{
1892 skb->data -= len;
1893 skb->len += len;
1894 return skb->data;
1895}
1896
Joe Perches7965bd42013-09-26 14:48:15 -07001897unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001898static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1899{
1900 skb->len -= len;
1901 BUG_ON(skb->len < skb->data_len);
1902 return skb->data += len;
1903}
1904
David S. Miller47d29642010-05-02 02:21:44 -07001905static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1906{
1907 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1908}
1909
Joe Perches7965bd42013-09-26 14:48:15 -07001910unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001911
1912static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1913{
1914 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07001915 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916 return NULL;
1917 skb->len -= len;
1918 return skb->data += len;
1919}
1920
1921static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1922{
1923 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1924}
1925
1926static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1927{
1928 if (likely(len <= skb_headlen(skb)))
1929 return 1;
1930 if (unlikely(len > skb->len))
1931 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07001932 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001933}
1934
1935/**
1936 * skb_headroom - bytes at buffer head
1937 * @skb: buffer to check
1938 *
1939 * Return the number of bytes of free space at the head of an &sk_buff.
1940 */
Chuck Leverc2636b42007-10-23 21:07:32 -07001941static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001942{
1943 return skb->data - skb->head;
1944}
1945
1946/**
1947 * skb_tailroom - bytes at buffer end
1948 * @skb: buffer to check
1949 *
1950 * Return the number of bytes of free space at the tail of an sk_buff
1951 */
1952static inline int skb_tailroom(const struct sk_buff *skb)
1953{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001954 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955}
1956
1957/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00001958 * skb_availroom - bytes at buffer end
1959 * @skb: buffer to check
1960 *
1961 * Return the number of bytes of free space at the tail of an sk_buff
1962 * allocated by sk_stream_alloc()
1963 */
1964static inline int skb_availroom(const struct sk_buff *skb)
1965{
Eric Dumazet16fad692013-03-14 05:40:32 +00001966 if (skb_is_nonlinear(skb))
1967 return 0;
1968
1969 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00001970}
1971
1972/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001973 * skb_reserve - adjust headroom
1974 * @skb: buffer to alter
1975 * @len: bytes to move
1976 *
1977 * Increase the headroom of an empty &sk_buff by reducing the tail
1978 * room. This is only allowed for an empty buffer.
1979 */
David S. Miller8243126c2006-01-17 02:54:21 -08001980static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001981{
1982 skb->data += len;
1983 skb->tail += len;
1984}
1985
Benjamin Poirier1837b2e2016-02-29 15:03:33 -08001986/**
1987 * skb_tailroom_reserve - adjust reserved_tailroom
1988 * @skb: buffer to alter
1989 * @mtu: maximum amount of headlen permitted
1990 * @needed_tailroom: minimum amount of reserved_tailroom
1991 *
1992 * Set reserved_tailroom so that headlen can be as large as possible but
1993 * not larger than mtu and tailroom cannot be smaller than
1994 * needed_tailroom.
1995 * The required headroom should already have been reserved before using
1996 * this function.
1997 */
1998static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
1999 unsigned int needed_tailroom)
2000{
2001 SKB_LINEAR_ASSERT(skb);
2002 if (mtu < skb_tailroom(skb) - needed_tailroom)
2003 /* use at most mtu */
2004 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2005 else
2006 /* use up to all available space */
2007 skb->reserved_tailroom = needed_tailroom;
2008}
2009
Tom Herbert8bce6d72014-09-29 20:22:29 -07002010#define ENCAP_TYPE_ETHER 0
2011#define ENCAP_TYPE_IPPROTO 1
2012
2013static inline void skb_set_inner_protocol(struct sk_buff *skb,
2014 __be16 protocol)
2015{
2016 skb->inner_protocol = protocol;
2017 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2018}
2019
2020static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2021 __u8 ipproto)
2022{
2023 skb->inner_ipproto = ipproto;
2024 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2025}
2026
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002027static inline void skb_reset_inner_headers(struct sk_buff *skb)
2028{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002029 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002030 skb->inner_network_header = skb->network_header;
2031 skb->inner_transport_header = skb->transport_header;
2032}
2033
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00002034static inline void skb_reset_mac_len(struct sk_buff *skb)
2035{
2036 skb->mac_len = skb->network_header - skb->mac_header;
2037}
2038
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002039static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2040 *skb)
2041{
2042 return skb->head + skb->inner_transport_header;
2043}
2044
Tom Herbert55dc5a92015-12-14 11:19:40 -08002045static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2046{
2047 return skb_inner_transport_header(skb) - skb->data;
2048}
2049
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002050static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2051{
2052 skb->inner_transport_header = skb->data - skb->head;
2053}
2054
2055static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2056 const int offset)
2057{
2058 skb_reset_inner_transport_header(skb);
2059 skb->inner_transport_header += offset;
2060}
2061
2062static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2063{
2064 return skb->head + skb->inner_network_header;
2065}
2066
2067static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2068{
2069 skb->inner_network_header = skb->data - skb->head;
2070}
2071
2072static inline void skb_set_inner_network_header(struct sk_buff *skb,
2073 const int offset)
2074{
2075 skb_reset_inner_network_header(skb);
2076 skb->inner_network_header += offset;
2077}
2078
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002079static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2080{
2081 return skb->head + skb->inner_mac_header;
2082}
2083
2084static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2085{
2086 skb->inner_mac_header = skb->data - skb->head;
2087}
2088
2089static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2090 const int offset)
2091{
2092 skb_reset_inner_mac_header(skb);
2093 skb->inner_mac_header += offset;
2094}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002095static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2096{
Cong Wang35d04612013-05-29 15:16:05 +08002097 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002098}
2099
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002100static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2101{
2102 return skb->head + skb->transport_header;
2103}
2104
2105static inline void skb_reset_transport_header(struct sk_buff *skb)
2106{
2107 skb->transport_header = skb->data - skb->head;
2108}
2109
2110static inline void skb_set_transport_header(struct sk_buff *skb,
2111 const int offset)
2112{
2113 skb_reset_transport_header(skb);
2114 skb->transport_header += offset;
2115}
2116
2117static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2118{
2119 return skb->head + skb->network_header;
2120}
2121
2122static inline void skb_reset_network_header(struct sk_buff *skb)
2123{
2124 skb->network_header = skb->data - skb->head;
2125}
2126
2127static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2128{
2129 skb_reset_network_header(skb);
2130 skb->network_header += offset;
2131}
2132
2133static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2134{
2135 return skb->head + skb->mac_header;
2136}
2137
2138static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2139{
Cong Wang35d04612013-05-29 15:16:05 +08002140 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002141}
2142
2143static inline void skb_reset_mac_header(struct sk_buff *skb)
2144{
2145 skb->mac_header = skb->data - skb->head;
2146}
2147
2148static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2149{
2150 skb_reset_mac_header(skb);
2151 skb->mac_header += offset;
2152}
2153
Timo Teräs0e3da5b2013-12-16 11:02:09 +02002154static inline void skb_pop_mac_header(struct sk_buff *skb)
2155{
2156 skb->mac_header = skb->network_header;
2157}
2158
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002159static inline void skb_probe_transport_header(struct sk_buff *skb,
2160 const int offset_hint)
2161{
2162 struct flow_keys keys;
2163
2164 if (skb_transport_header_was_set(skb))
2165 return;
Tom Herbertcd79a232015-09-01 09:24:27 -07002166 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
Tom Herbert42aecaa2015-06-04 09:16:39 -07002167 skb_set_transport_header(skb, keys.control.thoff);
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002168 else
2169 skb_set_transport_header(skb, offset_hint);
2170}
2171
Eric Dumazet03606892012-02-23 10:55:02 +00002172static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2173{
2174 if (skb_mac_header_was_set(skb)) {
2175 const unsigned char *old_mac = skb_mac_header(skb);
2176
2177 skb_set_mac_header(skb, -skb->mac_len);
2178 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2179 }
2180}
2181
Michał Mirosław04fb4512010-12-14 15:24:08 +00002182static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2183{
2184 return skb->csum_start - skb_headroom(skb);
2185}
2186
Alexander Duyck08b64fc2016-02-05 15:27:49 -08002187static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2188{
2189 return skb->head + skb->csum_start;
2190}
2191
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002192static inline int skb_transport_offset(const struct sk_buff *skb)
2193{
2194 return skb_transport_header(skb) - skb->data;
2195}
2196
2197static inline u32 skb_network_header_len(const struct sk_buff *skb)
2198{
2199 return skb->transport_header - skb->network_header;
2200}
2201
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002202static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2203{
2204 return skb->inner_transport_header - skb->inner_network_header;
2205}
2206
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002207static inline int skb_network_offset(const struct sk_buff *skb)
2208{
2209 return skb_network_header(skb) - skb->data;
2210}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03002211
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002212static inline int skb_inner_network_offset(const struct sk_buff *skb)
2213{
2214 return skb_inner_network_header(skb) - skb->data;
2215}
2216
Changli Gaof9599ce2010-08-04 04:43:44 +00002217static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2218{
2219 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2220}
2221
Linus Torvalds1da177e2005-04-16 15:20:36 -07002222/*
2223 * CPUs often take a performance hit when accessing unaligned memory
2224 * locations. The actual performance hit varies, it can be small if the
2225 * hardware handles it or large if we have to take an exception and fix it
2226 * in software.
2227 *
2228 * Since an ethernet header is 14 bytes network drivers often end up with
2229 * the IP header at an unaligned offset. The IP header can be aligned by
2230 * shifting the start of the packet by 2 bytes. Drivers should do this
2231 * with:
2232 *
Tobias Klauser8660c122009-07-13 22:48:16 +00002233 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002234 *
2235 * The downside to this alignment of the IP header is that the DMA is now
2236 * unaligned. On some architectures the cost of an unaligned DMA is high
2237 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00002238 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002239 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2240 * to be overridden.
2241 */
2242#ifndef NET_IP_ALIGN
2243#define NET_IP_ALIGN 2
2244#endif
2245
Anton Blanchard025be812006-03-31 02:27:06 -08002246/*
2247 * The networking layer reserves some headroom in skb data (via
2248 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2249 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08002250 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08002251 *
2252 * Unfortunately this headroom changes the DMA alignment of the resulting
2253 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2254 * on some architectures. An architecture can override this value,
2255 * perhaps setting it to a cacheline in size (since that will maintain
2256 * cacheline alignment of the DMA). It must be a power of 2.
2257 *
David S. Millerd6301d32009-02-08 19:24:13 -08002258 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08002259 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07002260 *
2261 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2262 * to reduce average number of cache lines per packet.
2263 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07002264 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08002265 */
2266#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07002267#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08002268#endif
2269
Joe Perches7965bd42013-09-26 14:48:15 -07002270int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002271
2272static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2273{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00002274 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07002275 WARN_ON(1);
2276 return;
2277 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002278 skb->len = len;
2279 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002280}
2281
Joe Perches7965bd42013-09-26 14:48:15 -07002282void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002283
2284static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2285{
Herbert Xu3cc0e872006-06-09 16:13:38 -07002286 if (skb->data_len)
2287 return ___pskb_trim(skb, len);
2288 __skb_trim(skb, len);
2289 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290}
2291
2292static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2293{
2294 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2295}
2296
2297/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07002298 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2299 * @skb: buffer to alter
2300 * @len: new length
2301 *
2302 * This is identical to pskb_trim except that the caller knows that
2303 * the skb is not cloned so we should never get an error due to out-
2304 * of-memory.
2305 */
2306static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2307{
2308 int err = pskb_trim(skb, len);
2309 BUG_ON(err);
2310}
2311
2312/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002313 * skb_orphan - orphan a buffer
2314 * @skb: buffer to orphan
2315 *
2316 * If a buffer currently has an owner then we call the owner's
2317 * destructor function and make the @skb unowned. The buffer continues
2318 * to exist but is no longer charged to its former owner.
2319 */
2320static inline void skb_orphan(struct sk_buff *skb)
2321{
Eric Dumazetc34a7612013-07-30 16:11:15 -07002322 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002323 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002324 skb->destructor = NULL;
2325 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07002326 } else {
2327 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002328 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002329}
2330
2331/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002332 * skb_orphan_frags - orphan the frags contained in a buffer
2333 * @skb: buffer to orphan frags from
2334 * @gfp_mask: allocation mask for replacement pages
2335 *
2336 * For each frag in the SKB which needs a destructor (i.e. has an
2337 * owner) create a copy of that frag and release the original
2338 * page by calling the destructor.
2339 */
2340static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2341{
2342 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2343 return 0;
2344 return skb_copy_ubufs(skb, gfp_mask);
2345}
2346
2347/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002348 * __skb_queue_purge - empty a list
2349 * @list: list to empty
2350 *
2351 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2352 * the list and one reference dropped. This function does not take the
2353 * list lock and the caller must hold the relevant locks to use it.
2354 */
Joe Perches7965bd42013-09-26 14:48:15 -07002355void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002356static inline void __skb_queue_purge(struct sk_buff_head *list)
2357{
2358 struct sk_buff *skb;
2359 while ((skb = __skb_dequeue(list)) != NULL)
2360 kfree_skb(skb);
2361}
2362
Joe Perches7965bd42013-09-26 14:48:15 -07002363void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002364
Joe Perches7965bd42013-09-26 14:48:15 -07002365struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2366 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002367
2368/**
2369 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2370 * @dev: network device to receive on
2371 * @length: length to allocate
2372 *
2373 * Allocate a new &sk_buff and assign it a usage count of one. The
2374 * buffer has unspecified headroom built in. Users should allocate
2375 * the headroom they think they need without accounting for the
2376 * built in space. The built in space is used for optimisations.
2377 *
2378 * %NULL is returned if there is no free memory. Although this function
2379 * allocates memory it can be called from an interrupt.
2380 */
2381static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002382 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002383{
2384 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2385}
2386
Eric Dumazet6f532612012-05-18 05:12:12 +00002387/* legacy helper around __netdev_alloc_skb() */
2388static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2389 gfp_t gfp_mask)
2390{
2391 return __netdev_alloc_skb(NULL, length, gfp_mask);
2392}
2393
2394/* legacy helper around netdev_alloc_skb() */
2395static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2396{
2397 return netdev_alloc_skb(NULL, length);
2398}
2399
2400
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002401static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2402 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002403{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002404 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002405
2406 if (NET_IP_ALIGN && skb)
2407 skb_reserve(skb, NET_IP_ALIGN);
2408 return skb;
2409}
2410
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002411static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2412 unsigned int length)
2413{
2414 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2415}
2416
Alexander Duyck181edb22015-05-06 21:12:03 -07002417static inline void skb_free_frag(void *addr)
2418{
2419 __free_page_frag(addr);
2420}
2421
Alexander Duyckffde7322014-12-09 19:40:42 -08002422void *napi_alloc_frag(unsigned int fragsz);
Alexander Duyckfd11a832014-12-09 19:40:49 -08002423struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2424 unsigned int length, gfp_t gfp_mask);
2425static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2426 unsigned int length)
2427{
2428 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2429}
Jesper Dangaard Brouer795bb1c2016-02-08 13:14:59 +01002430void napi_consume_skb(struct sk_buff *skb, int budget);
2431
2432void __kfree_skb_flush(void);
Jesper Dangaard Brouer15fad712016-02-08 13:15:04 +01002433void __kfree_skb_defer(struct sk_buff *skb);
Alexander Duyckffde7322014-12-09 19:40:42 -08002434
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002435/**
Alexander Duyck71dfda52014-11-11 09:26:34 -08002436 * __dev_alloc_pages - allocate page for network Rx
2437 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2438 * @order: size of the allocation
2439 *
2440 * Allocate a new page.
2441 *
2442 * %NULL is returned if there is no free memory.
2443*/
2444static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2445 unsigned int order)
2446{
2447 /* This piece of code contains several assumptions.
2448 * 1. This is for device Rx, therefor a cold page is preferred.
2449 * 2. The expectation is the user wants a compound page.
2450 * 3. If requesting a order 0 page it will not be compound
2451 * due to the check to see if order has a value in prep_new_page
2452 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2453 * code in gfp_to_alloc_flags that should be enforcing this.
2454 */
2455 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2456
2457 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2458}
2459
2460static inline struct page *dev_alloc_pages(unsigned int order)
2461{
2462 return __dev_alloc_pages(GFP_ATOMIC, order);
2463}
2464
2465/**
2466 * __dev_alloc_page - allocate a page for network Rx
2467 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2468 *
2469 * Allocate a new page.
2470 *
2471 * %NULL is returned if there is no free memory.
2472 */
2473static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2474{
2475 return __dev_alloc_pages(gfp_mask, 0);
2476}
2477
2478static inline struct page *dev_alloc_page(void)
2479{
2480 return __dev_alloc_page(GFP_ATOMIC);
2481}
2482
2483/**
Mel Gorman06140022012-07-31 16:44:24 -07002484 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2485 * @page: The page that was allocated from skb_alloc_page
2486 * @skb: The skb that may need pfmemalloc set
2487 */
2488static inline void skb_propagate_pfmemalloc(struct page *page,
2489 struct sk_buff *skb)
2490{
Michal Hocko2f064f32015-08-21 14:11:51 -07002491 if (page_is_pfmemalloc(page))
Mel Gorman06140022012-07-31 16:44:24 -07002492 skb->pfmemalloc = true;
2493}
2494
Eric Dumazet564824b2010-10-11 19:05:25 +00002495/**
Masanari Iidae2278672014-02-18 22:54:36 +09002496 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002497 * @frag: the paged fragment
2498 *
2499 * Returns the &struct page associated with @frag.
2500 */
2501static inline struct page *skb_frag_page(const skb_frag_t *frag)
2502{
Ian Campbella8605c62011-10-19 23:01:49 +00002503 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002504}
2505
2506/**
2507 * __skb_frag_ref - take an addition reference on a paged fragment.
2508 * @frag: the paged fragment
2509 *
2510 * Takes an additional reference on the paged fragment @frag.
2511 */
2512static inline void __skb_frag_ref(skb_frag_t *frag)
2513{
2514 get_page(skb_frag_page(frag));
2515}
2516
2517/**
2518 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2519 * @skb: the buffer
2520 * @f: the fragment offset.
2521 *
2522 * Takes an additional reference on the @f'th paged fragment of @skb.
2523 */
2524static inline void skb_frag_ref(struct sk_buff *skb, int f)
2525{
2526 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2527}
2528
2529/**
2530 * __skb_frag_unref - release a reference on a paged fragment.
2531 * @frag: the paged fragment
2532 *
2533 * Releases a reference on the paged fragment @frag.
2534 */
2535static inline void __skb_frag_unref(skb_frag_t *frag)
2536{
2537 put_page(skb_frag_page(frag));
2538}
2539
2540/**
2541 * skb_frag_unref - release a reference on a paged fragment of an skb.
2542 * @skb: the buffer
2543 * @f: the fragment offset
2544 *
2545 * Releases a reference on the @f'th paged fragment of @skb.
2546 */
2547static inline void skb_frag_unref(struct sk_buff *skb, int f)
2548{
2549 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2550}
2551
2552/**
2553 * skb_frag_address - gets the address of the data contained in a paged fragment
2554 * @frag: the paged fragment buffer
2555 *
2556 * Returns the address of the data within @frag. The page must already
2557 * be mapped.
2558 */
2559static inline void *skb_frag_address(const skb_frag_t *frag)
2560{
2561 return page_address(skb_frag_page(frag)) + frag->page_offset;
2562}
2563
2564/**
2565 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2566 * @frag: the paged fragment buffer
2567 *
2568 * Returns the address of the data within @frag. Checks that the page
2569 * is mapped and returns %NULL otherwise.
2570 */
2571static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2572{
2573 void *ptr = page_address(skb_frag_page(frag));
2574 if (unlikely(!ptr))
2575 return NULL;
2576
2577 return ptr + frag->page_offset;
2578}
2579
2580/**
2581 * __skb_frag_set_page - sets the page contained in a paged fragment
2582 * @frag: the paged fragment
2583 * @page: the page to set
2584 *
2585 * Sets the fragment @frag to contain @page.
2586 */
2587static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2588{
Ian Campbella8605c62011-10-19 23:01:49 +00002589 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002590}
2591
2592/**
2593 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2594 * @skb: the buffer
2595 * @f: the fragment offset
2596 * @page: the page to set
2597 *
2598 * Sets the @f'th fragment of @skb to contain @page.
2599 */
2600static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2601 struct page *page)
2602{
2603 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2604}
2605
Eric Dumazet400dfd32013-10-17 16:27:07 -07002606bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2607
Ian Campbell131ea662011-08-19 06:25:00 +00002608/**
2609 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002610 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002611 * @frag: the paged fragment to map
2612 * @offset: the offset within the fragment (starting at the
2613 * fragment's own offset)
2614 * @size: the number of bytes to map
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002615 * @dir: the direction of the mapping (%PCI_DMA_*)
Ian Campbell131ea662011-08-19 06:25:00 +00002616 *
2617 * Maps the page associated with @frag to @device.
2618 */
2619static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2620 const skb_frag_t *frag,
2621 size_t offset, size_t size,
2622 enum dma_data_direction dir)
2623{
2624 return dma_map_page(dev, skb_frag_page(frag),
2625 frag->page_offset + offset, size, dir);
2626}
2627
Eric Dumazet117632e2011-12-03 21:39:53 +00002628static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2629 gfp_t gfp_mask)
2630{
2631 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2632}
2633
Octavian Purdilabad93e92014-06-12 01:36:26 +03002634
2635static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2636 gfp_t gfp_mask)
2637{
2638 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2639}
2640
2641
Ian Campbell131ea662011-08-19 06:25:00 +00002642/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002643 * skb_clone_writable - is the header of a clone writable
2644 * @skb: buffer to check
2645 * @len: length up to which to write
2646 *
2647 * Returns true if modifying the header part of the cloned buffer
2648 * does not requires the data to be copied.
2649 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002650static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002651{
2652 return !skb_header_cloned(skb) &&
2653 skb_headroom(skb) + len <= skb->hdr_len;
2654}
2655
Daniel Borkmann36976492016-02-19 23:05:25 +01002656static inline int skb_try_make_writable(struct sk_buff *skb,
2657 unsigned int write_len)
2658{
2659 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2660 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2661}
2662
Herbert Xud9cc2042007-09-16 16:21:16 -07002663static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2664 int cloned)
2665{
2666 int delta = 0;
2667
Herbert Xud9cc2042007-09-16 16:21:16 -07002668 if (headroom > skb_headroom(skb))
2669 delta = headroom - skb_headroom(skb);
2670
2671 if (delta || cloned)
2672 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2673 GFP_ATOMIC);
2674 return 0;
2675}
2676
Patrick McHardy334a8132007-06-25 04:35:20 -07002677/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002678 * skb_cow - copy header of skb when it is required
2679 * @skb: buffer to cow
2680 * @headroom: needed headroom
2681 *
2682 * If the skb passed lacks sufficient headroom or its data part
2683 * is shared, data is reallocated. If reallocation fails, an error
2684 * is returned and original skb is not changed.
2685 *
2686 * The result is skb with writable area skb->head...skb->tail
2687 * and at least @headroom of space at head.
2688 */
2689static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2690{
Herbert Xud9cc2042007-09-16 16:21:16 -07002691 return __skb_cow(skb, headroom, skb_cloned(skb));
2692}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002693
Herbert Xud9cc2042007-09-16 16:21:16 -07002694/**
2695 * skb_cow_head - skb_cow but only making the head writable
2696 * @skb: buffer to cow
2697 * @headroom: needed headroom
2698 *
2699 * This function is identical to skb_cow except that we replace the
2700 * skb_cloned check by skb_header_cloned. It should be used when
2701 * you only need to push on some header and do not need to modify
2702 * the data.
2703 */
2704static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2705{
2706 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002707}
2708
2709/**
2710 * skb_padto - pad an skbuff up to a minimal size
2711 * @skb: buffer to pad
2712 * @len: minimal length
2713 *
2714 * Pads up a buffer to ensure the trailing bytes exist and are
2715 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002716 * is untouched. Otherwise it is extended. Returns zero on
2717 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718 */
Herbert Xu5b057c62006-06-23 02:06:41 -07002719static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002720{
2721 unsigned int size = skb->len;
2722 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002723 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002724 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725}
2726
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002727/**
2728 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2729 * @skb: buffer to pad
2730 * @len: minimal length
2731 *
2732 * Pads up a buffer to ensure the trailing bytes exist and are
2733 * blanked. If the buffer already contains sufficient data it
2734 * is untouched. Otherwise it is extended. Returns zero on
2735 * success. The skb is freed on error.
2736 */
2737static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2738{
2739 unsigned int size = skb->len;
2740
2741 if (unlikely(size < len)) {
2742 len -= size;
2743 if (skb_pad(skb, len))
2744 return -ENOMEM;
2745 __skb_put(skb, len);
2746 }
2747 return 0;
2748}
2749
Linus Torvalds1da177e2005-04-16 15:20:36 -07002750static inline int skb_add_data(struct sk_buff *skb,
Al Viroaf2b0402014-11-27 21:44:24 -05002751 struct iov_iter *from, int copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752{
2753 const int off = skb->len;
2754
2755 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viroaf2b0402014-11-27 21:44:24 -05002756 __wsum csum = 0;
2757 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2758 &csum, from) == copy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002759 skb->csum = csum_block_add(skb->csum, csum, off);
2760 return 0;
2761 }
Al Viroaf2b0402014-11-27 21:44:24 -05002762 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002763 return 0;
2764
2765 __skb_trim(skb, off);
2766 return -EFAULT;
2767}
2768
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002769static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2770 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002771{
2772 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002773 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002774
Ian Campbellea2ab692011-08-22 23:44:58 +00002775 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00002776 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002777 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002778 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002779}
2780
Herbert Xu364c6ba2006-06-09 16:10:40 -07002781static inline int __skb_linearize(struct sk_buff *skb)
2782{
2783 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2784}
2785
Linus Torvalds1da177e2005-04-16 15:20:36 -07002786/**
2787 * skb_linearize - convert paged skb to linear one
2788 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07002789 *
2790 * If there is no free memory -ENOMEM is returned, otherwise zero
2791 * is returned and the old skb data released.
2792 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07002793static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002794{
Herbert Xu364c6ba2006-06-09 16:10:40 -07002795 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2796}
2797
2798/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00002799 * skb_has_shared_frag - can any frag be overwritten
2800 * @skb: buffer to test
2801 *
2802 * Return true if the skb has at least one frag that might be modified
2803 * by an external entity (as in vmsplice()/sendfile())
2804 */
2805static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2806{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00002807 return skb_is_nonlinear(skb) &&
2808 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002809}
2810
2811/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07002812 * skb_linearize_cow - make sure skb is linear and writable
2813 * @skb: buffer to process
2814 *
2815 * If there is no free memory -ENOMEM is returned, otherwise zero
2816 * is returned and the old skb data released.
2817 */
2818static inline int skb_linearize_cow(struct sk_buff *skb)
2819{
2820 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2821 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002822}
2823
2824/**
2825 * skb_postpull_rcsum - update checksum for received skb after pull
2826 * @skb: buffer to update
2827 * @start: start of data before pull
2828 * @len: length of data pulled
2829 *
2830 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07002831 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2832 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002833 */
2834
2835static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f2006-03-20 22:43:56 -08002836 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002837{
Patrick McHardy84fa7932006-08-29 16:44:56 -07002838 if (skb->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002839 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002840 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
Pravin B Shelar31b33df2015-09-28 17:24:25 -07002841 skb_checksum_start_offset(skb) < 0)
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002842 skb->ip_summed = CHECKSUM_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002843}
2844
Herbert Xucbb042f2006-03-20 22:43:56 -08002845unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2846
Daniel Borkmannf8ffad62016-01-07 15:50:23 +01002847static inline void skb_postpush_rcsum(struct sk_buff *skb,
2848 const void *start, unsigned int len)
2849{
2850 /* For performing the reverse operation to skb_postpull_rcsum(),
2851 * we can instead of ...
2852 *
2853 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0));
2854 *
2855 * ... just use this equivalent version here to save a few
2856 * instructions. Feeding csum of 0 in csum_partial() and later
2857 * on adding skb->csum is equivalent to feed skb->csum in the
2858 * first place.
2859 */
2860 if (skb->ip_summed == CHECKSUM_COMPLETE)
2861 skb->csum = csum_partial(start, len, skb->csum);
2862}
2863
David S. Miller7ce5a272013-12-02 17:26:05 -05002864/**
2865 * pskb_trim_rcsum - trim received skb and update checksum
2866 * @skb: buffer to trim
2867 * @len: new length
2868 *
2869 * This is exactly the same as pskb_trim except that it ensures the
2870 * checksum of received packets are still valid after the operation.
2871 */
2872
2873static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2874{
2875 if (likely(len >= skb->len))
2876 return 0;
2877 if (skb->ip_summed == CHECKSUM_COMPLETE)
2878 skb->ip_summed = CHECKSUM_NONE;
2879 return __pskb_trim(skb, len);
2880}
2881
Linus Torvalds1da177e2005-04-16 15:20:36 -07002882#define skb_queue_walk(queue, skb) \
2883 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002884 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002885 skb = skb->next)
2886
James Chapman46f89142007-04-30 00:07:31 -07002887#define skb_queue_walk_safe(queue, skb, tmp) \
2888 for (skb = (queue)->next, tmp = skb->next; \
2889 skb != (struct sk_buff *)(queue); \
2890 skb = tmp, tmp = skb->next)
2891
David S. Miller1164f522008-09-23 00:49:44 -07002892#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002893 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07002894 skb = skb->next)
2895
2896#define skb_queue_walk_from_safe(queue, skb, tmp) \
2897 for (tmp = skb->next; \
2898 skb != (struct sk_buff *)(queue); \
2899 skb = tmp, tmp = skb->next)
2900
Stephen Hemminger300ce172005-10-30 13:47:34 -08002901#define skb_queue_reverse_walk(queue, skb) \
2902 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002903 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08002904 skb = skb->prev)
2905
David S. Miller686a2952011-01-20 22:47:32 -08002906#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2907 for (skb = (queue)->prev, tmp = skb->prev; \
2908 skb != (struct sk_buff *)(queue); \
2909 skb = tmp, tmp = skb->prev)
2910
2911#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2912 for (tmp = skb->prev; \
2913 skb != (struct sk_buff *)(queue); \
2914 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002915
David S. Miller21dc3302010-08-23 00:13:46 -07002916static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07002917{
2918 return skb_shinfo(skb)->frag_list != NULL;
2919}
2920
2921static inline void skb_frag_list_init(struct sk_buff *skb)
2922{
2923 skb_shinfo(skb)->frag_list = NULL;
2924}
2925
David S. Milleree039872009-06-09 00:17:13 -07002926#define skb_walk_frags(skb, iter) \
2927 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2928
Rainer Weikusatea3793e2015-12-06 21:11:34 +00002929
2930int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
2931 const struct sk_buff *skb);
2932struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
2933 int *peeked, int *off, int *err,
2934 struct sk_buff **last);
Joe Perches7965bd42013-09-26 14:48:15 -07002935struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2936 int *peeked, int *off, int *err);
2937struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2938 int *err);
2939unsigned int datagram_poll(struct file *file, struct socket *sock,
2940 struct poll_table_struct *wait);
Al Viroc0371da2014-11-24 10:42:55 -05002941int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2942 struct iov_iter *to, int size);
David S. Miller51f3d022014-11-05 16:46:40 -05002943static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2944 struct msghdr *msg, int size)
2945{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002946 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
David S. Miller51f3d022014-11-05 16:46:40 -05002947}
Al Viroe5a4b0b2014-11-24 18:17:55 -05002948int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2949 struct msghdr *msg);
Al Viro3a654f92014-06-19 14:15:22 -04002950int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2951 struct iov_iter *from, int len);
Al Viro3a654f92014-06-19 14:15:22 -04002952int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
Joe Perches7965bd42013-09-26 14:48:15 -07002953void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
samanthakumar627d2d62016-04-05 12:41:16 -04002954void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
2955static inline void skb_free_datagram_locked(struct sock *sk,
2956 struct sk_buff *skb)
2957{
2958 __skb_free_datagram_locked(sk, skb, 0);
2959}
Joe Perches7965bd42013-09-26 14:48:15 -07002960int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07002961int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2962int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2963__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2964 int len, __wsum csum);
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002965ssize_t skb_socket_splice(struct sock *sk,
2966 struct pipe_inode_info *pipe,
2967 struct splice_pipe_desc *spd);
2968int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
Joe Perches7965bd42013-09-26 14:48:15 -07002969 struct pipe_inode_info *pipe, unsigned int len,
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002970 unsigned int flags,
2971 ssize_t (*splice_cb)(struct sock *,
2972 struct pipe_inode_info *,
2973 struct splice_pipe_desc *));
Joe Perches7965bd42013-09-26 14:48:15 -07002974void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01002975unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002976int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2977 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07002978void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2979int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2980void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01002981unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -07002982struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Vlad Yasevich0d5501c2014-08-08 14:42:13 -04002983struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
Jiri Pirkoe2195122014-11-19 14:05:01 +01002984int skb_ensure_writable(struct sk_buff *skb, int write_len);
Jiri Pirko93515d52014-11-19 14:05:02 +01002985int skb_vlan_pop(struct sk_buff *skb);
2986int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002987
Al Viro6ce8e9c2014-04-06 21:25:44 -04002988static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2989{
Al Viro21226ab2014-11-28 15:48:29 -05002990 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro6ce8e9c2014-04-06 21:25:44 -04002991}
2992
Al Viro7eab8d92014-04-06 21:51:23 -04002993static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2994{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002995 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro7eab8d92014-04-06 21:51:23 -04002996}
2997
Daniel Borkmann2817a332013-10-30 11:50:51 +01002998struct skb_checksum_ops {
2999 __wsum (*update)(const void *mem, int len, __wsum wsum);
3000 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3001};
3002
3003__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3004 __wsum csum, const struct skb_checksum_ops *ops);
3005__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3006 __wsum csum);
3007
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003008static inline void * __must_check
3009__skb_header_pointer(const struct sk_buff *skb, int offset,
3010 int len, void *data, int hlen, void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011{
Patrick McHardy55820ee2005-07-05 14:08:10 -07003012 if (hlen - offset >= len)
David S. Miller690e36e2014-08-23 12:13:41 -07003013 return data + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014
David S. Miller690e36e2014-08-23 12:13:41 -07003015 if (!skb ||
3016 skb_copy_bits(skb, offset, buffer, len) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003017 return NULL;
3018
3019 return buffer;
3020}
3021
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003022static inline void * __must_check
3023skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
David S. Miller690e36e2014-08-23 12:13:41 -07003024{
3025 return __skb_header_pointer(skb, offset, len, skb->data,
3026 skb_headlen(skb), buffer);
3027}
3028
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01003029/**
3030 * skb_needs_linearize - check if we need to linearize a given skb
3031 * depending on the given device features.
3032 * @skb: socket buffer to check
3033 * @features: net device features
3034 *
3035 * Returns true if either:
3036 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3037 * 2. skb is fragmented and the device does not support SG.
3038 */
3039static inline bool skb_needs_linearize(struct sk_buff *skb,
3040 netdev_features_t features)
3041{
3042 return skb_is_nonlinear(skb) &&
3043 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3044 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3045}
3046
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03003047static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3048 void *to,
3049 const unsigned int len)
3050{
3051 memcpy(to, skb->data, len);
3052}
3053
3054static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3055 const int offset, void *to,
3056 const unsigned int len)
3057{
3058 memcpy(to, skb->data + offset, len);
3059}
3060
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03003061static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3062 const void *from,
3063 const unsigned int len)
3064{
3065 memcpy(skb->data, from, len);
3066}
3067
3068static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3069 const int offset,
3070 const void *from,
3071 const unsigned int len)
3072{
3073 memcpy(skb->data + offset, from, len);
3074}
3075
Joe Perches7965bd42013-09-26 14:48:15 -07003076void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077
Patrick Ohlyac45f602009-02-12 05:03:37 +00003078static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3079{
3080 return skb->tstamp;
3081}
3082
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003083/**
3084 * skb_get_timestamp - get timestamp from a skb
3085 * @skb: skb to get stamp from
3086 * @stamp: pointer to struct timeval to store stamp in
3087 *
3088 * Timestamps are stored in the skb as offsets to a base timestamp.
3089 * This function converts the offset back to a struct timeval and stores
3090 * it in stamp.
3091 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00003092static inline void skb_get_timestamp(const struct sk_buff *skb,
3093 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003094{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003095 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003096}
3097
Patrick Ohlyac45f602009-02-12 05:03:37 +00003098static inline void skb_get_timestampns(const struct sk_buff *skb,
3099 struct timespec *stamp)
3100{
3101 *stamp = ktime_to_timespec(skb->tstamp);
3102}
3103
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003104static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003105{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003106 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003107}
3108
Stephen Hemminger164891a2007-04-23 22:26:16 -07003109static inline ktime_t net_timedelta(ktime_t t)
3110{
3111 return ktime_sub(ktime_get_real(), t);
3112}
3113
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07003114static inline ktime_t net_invalid_timestamp(void)
3115{
3116 return ktime_set(0, 0);
3117}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003118
Alexander Duyck62bccb82014-09-04 13:31:35 -04003119struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3120
Richard Cochranc1f19b52010-07-17 08:49:36 +00003121#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3122
Joe Perches7965bd42013-09-26 14:48:15 -07003123void skb_clone_tx_timestamp(struct sk_buff *skb);
3124bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00003125
3126#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3127
3128static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3129{
3130}
3131
3132static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3133{
3134 return false;
3135}
3136
3137#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3138
3139/**
3140 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3141 *
Richard Cochranda92b192011-10-21 00:49:15 +00003142 * PHY drivers may accept clones of transmitted packets for
3143 * timestamping via their phy_driver.txtstamp method. These drivers
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003144 * must call this function to return the skb back to the stack with a
3145 * timestamp.
Richard Cochranda92b192011-10-21 00:49:15 +00003146 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00003147 * @skb: clone of the the original outgoing packet
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003148 * @hwtstamps: hardware time stamps
Richard Cochranc1f19b52010-07-17 08:49:36 +00003149 *
3150 */
3151void skb_complete_tx_timestamp(struct sk_buff *skb,
3152 struct skb_shared_hwtstamps *hwtstamps);
3153
Willem de Bruijne7fd2882014-08-04 22:11:48 -04003154void __skb_tstamp_tx(struct sk_buff *orig_skb,
3155 struct skb_shared_hwtstamps *hwtstamps,
3156 struct sock *sk, int tstype);
3157
Patrick Ohlyac45f602009-02-12 05:03:37 +00003158/**
3159 * skb_tstamp_tx - queue clone of skb with send time stamps
3160 * @orig_skb: the original outgoing packet
3161 * @hwtstamps: hardware time stamps, may be NULL if not available
3162 *
3163 * If the skb has a socket associated, then this function clones the
3164 * skb (thus sharing the actual data and optional structures), stores
3165 * the optional hardware time stamping information (if non NULL) or
3166 * generates a software time stamp (otherwise), then queues the clone
3167 * to the error queue of the socket. Errors are silently ignored.
3168 */
Joe Perches7965bd42013-09-26 14:48:15 -07003169void skb_tstamp_tx(struct sk_buff *orig_skb,
3170 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00003171
Richard Cochran4507a712010-07-17 08:48:28 +00003172static inline void sw_tx_timestamp(struct sk_buff *skb)
3173{
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003174 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3175 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
Richard Cochran4507a712010-07-17 08:48:28 +00003176 skb_tstamp_tx(skb, NULL);
3177}
3178
3179/**
3180 * skb_tx_timestamp() - Driver hook for transmit timestamping
3181 *
3182 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00003183 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00003184 *
David S. Miller73409f32013-12-27 13:04:33 -05003185 * Specifically, one should make absolutely sure that this function is
3186 * called before TX completion of this packet can trigger. Otherwise
3187 * the packet could potentially already be freed.
3188 *
Richard Cochran4507a712010-07-17 08:48:28 +00003189 * @skb: A socket buffer.
3190 */
3191static inline void skb_tx_timestamp(struct sk_buff *skb)
3192{
Richard Cochranc1f19b52010-07-17 08:49:36 +00003193 skb_clone_tx_timestamp(skb);
Richard Cochran4507a712010-07-17 08:48:28 +00003194 sw_tx_timestamp(skb);
3195}
3196
Johannes Berg6e3e9392011-11-09 10:15:42 +01003197/**
3198 * skb_complete_wifi_ack - deliver skb with wifi status
3199 *
3200 * @skb: the original outgoing packet
3201 * @acked: ack status
3202 *
3203 */
3204void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3205
Joe Perches7965bd42013-09-26 14:48:15 -07003206__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3207__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003208
Herbert Xu60476372007-04-09 11:59:39 -07003209static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3210{
Tom Herbert6edec0e2015-02-10 16:30:28 -08003211 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3212 skb->csum_valid ||
3213 (skb->ip_summed == CHECKSUM_PARTIAL &&
3214 skb_checksum_start_offset(skb) >= 0));
Herbert Xu60476372007-04-09 11:59:39 -07003215}
3216
Herbert Xufb286bb2005-11-10 13:01:24 -08003217/**
3218 * skb_checksum_complete - Calculate checksum of an entire packet
3219 * @skb: packet to process
3220 *
3221 * This function calculates the checksum over the entire packet plus
3222 * the value of skb->csum. The latter can be used to supply the
3223 * checksum of a pseudo header as used by TCP/UDP. It returns the
3224 * checksum.
3225 *
3226 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3227 * this function can be used to verify that checksum on received
3228 * packets. In that case the function should return zero if the
3229 * checksum is correct. In particular, this function will return zero
3230 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3231 * hardware has already verified the correctness of the checksum.
3232 */
Al Viro4381ca32007-07-15 21:00:11 +01003233static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08003234{
Herbert Xu60476372007-04-09 11:59:39 -07003235 return skb_csum_unnecessary(skb) ?
3236 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003237}
3238
Tom Herbert77cffe22014-08-27 21:26:46 -07003239static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3240{
3241 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3242 if (skb->csum_level == 0)
3243 skb->ip_summed = CHECKSUM_NONE;
3244 else
3245 skb->csum_level--;
3246 }
3247}
3248
3249static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3250{
3251 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3252 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3253 skb->csum_level++;
3254 } else if (skb->ip_summed == CHECKSUM_NONE) {
3255 skb->ip_summed = CHECKSUM_UNNECESSARY;
3256 skb->csum_level = 0;
3257 }
3258}
3259
Tom Herbert5a212322014-08-31 15:12:41 -07003260static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3261{
3262 /* Mark current checksum as bad (typically called from GRO
3263 * path). In the case that ip_summed is CHECKSUM_NONE
3264 * this must be the first checksum encountered in the packet.
3265 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3266 * checksum after the last one validated. For UDP, a zero
3267 * checksum can not be marked as bad.
3268 */
3269
3270 if (skb->ip_summed == CHECKSUM_NONE ||
3271 skb->ip_summed == CHECKSUM_UNNECESSARY)
3272 skb->csum_bad = 1;
3273}
3274
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003275/* Check if we need to perform checksum complete validation.
3276 *
3277 * Returns true if checksum complete is needed, false otherwise
3278 * (either checksum is unnecessary or zero checksum is allowed).
3279 */
3280static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3281 bool zero_okay,
3282 __sum16 check)
3283{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003284 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3285 skb->csum_valid = 1;
Tom Herbert77cffe22014-08-27 21:26:46 -07003286 __skb_decr_checksum_unnecessary(skb);
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003287 return false;
3288 }
3289
3290 return true;
3291}
3292
3293/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3294 * in checksum_init.
3295 */
3296#define CHECKSUM_BREAK 76
3297
Tom Herbert4e18b9a2015-04-20 14:10:04 -07003298/* Unset checksum-complete
3299 *
3300 * Unset checksum complete can be done when packet is being modified
3301 * (uncompressed for instance) and checksum-complete value is
3302 * invalidated.
3303 */
3304static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3305{
3306 if (skb->ip_summed == CHECKSUM_COMPLETE)
3307 skb->ip_summed = CHECKSUM_NONE;
3308}
3309
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003310/* Validate (init) checksum based on checksum complete.
3311 *
3312 * Return values:
3313 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3314 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3315 * checksum is stored in skb->csum for use in __skb_checksum_complete
3316 * non-zero: value of invalid checksum
3317 *
3318 */
3319static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3320 bool complete,
3321 __wsum psum)
3322{
3323 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3324 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003325 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003326 return 0;
3327 }
Tom Herbert5a212322014-08-31 15:12:41 -07003328 } else if (skb->csum_bad) {
3329 /* ip_summed == CHECKSUM_NONE in this case */
Eric Dumazetc91d4602015-05-15 05:48:07 -07003330 return (__force __sum16)1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003331 }
3332
3333 skb->csum = psum;
3334
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003335 if (complete || skb->len <= CHECKSUM_BREAK) {
3336 __sum16 csum;
3337
3338 csum = __skb_checksum_complete(skb);
3339 skb->csum_valid = !csum;
3340 return csum;
3341 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003342
3343 return 0;
3344}
3345
3346static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3347{
3348 return 0;
3349}
3350
3351/* Perform checksum validate (init). Note that this is a macro since we only
3352 * want to calculate the pseudo header which is an input function if necessary.
3353 * First we try to validate without any computation (checksum unnecessary) and
3354 * then calculate based on checksum complete calling the function to compute
3355 * pseudo header.
3356 *
3357 * Return values:
3358 * 0: checksum is validated or try to in skb_checksum_complete
3359 * non-zero: value of invalid checksum
3360 */
3361#define __skb_checksum_validate(skb, proto, complete, \
3362 zero_okay, check, compute_pseudo) \
3363({ \
3364 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003365 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003366 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3367 __ret = __skb_checksum_validate_complete(skb, \
3368 complete, compute_pseudo(skb, proto)); \
3369 __ret; \
3370})
3371
3372#define skb_checksum_init(skb, proto, compute_pseudo) \
3373 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3374
3375#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3376 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3377
3378#define skb_checksum_validate(skb, proto, compute_pseudo) \
3379 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3380
3381#define skb_checksum_validate_zero_check(skb, proto, check, \
3382 compute_pseudo) \
Sabrina Dubroca096a4cf2015-02-06 18:54:19 +01003383 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003384
3385#define skb_checksum_simple_validate(skb) \
3386 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3387
Tom Herbertd96535a2014-08-31 15:12:42 -07003388static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3389{
3390 return (skb->ip_summed == CHECKSUM_NONE &&
3391 skb->csum_valid && !skb->csum_bad);
3392}
3393
3394static inline void __skb_checksum_convert(struct sk_buff *skb,
3395 __sum16 check, __wsum pseudo)
3396{
3397 skb->csum = ~pseudo;
3398 skb->ip_summed = CHECKSUM_COMPLETE;
3399}
3400
3401#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3402do { \
3403 if (__skb_checksum_convert_check(skb)) \
3404 __skb_checksum_convert(skb, check, \
3405 compute_pseudo(skb, proto)); \
3406} while (0)
3407
Tom Herbert15e23962015-02-10 16:30:31 -08003408static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3409 u16 start, u16 offset)
3410{
3411 skb->ip_summed = CHECKSUM_PARTIAL;
3412 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3413 skb->csum_offset = offset - start;
3414}
3415
Tom Herbertdcdc8992015-02-02 16:07:34 -08003416/* Update skbuf and packet to reflect the remote checksum offload operation.
3417 * When called, ptr indicates the starting point for skb->csum when
3418 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3419 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3420 */
3421static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
Tom Herbert15e23962015-02-10 16:30:31 -08003422 int start, int offset, bool nopartial)
Tom Herbertdcdc8992015-02-02 16:07:34 -08003423{
3424 __wsum delta;
3425
Tom Herbert15e23962015-02-10 16:30:31 -08003426 if (!nopartial) {
3427 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3428 return;
3429 }
3430
Tom Herbertdcdc8992015-02-02 16:07:34 -08003431 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3432 __skb_checksum_complete(skb);
3433 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3434 }
3435
3436 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3437
3438 /* Adjust skb->csum since we changed the packet */
3439 skb->csum = csum_add(skb->csum, delta);
3440}
3441
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003442#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07003443void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003444static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3445{
3446 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07003447 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003448}
3449static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3450{
3451 if (nfct)
3452 atomic_inc(&nfct->use);
3453}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003454#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003455#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003456static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3457{
3458 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3459 kfree(nf_bridge);
3460}
3461static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3462{
3463 if (nf_bridge)
3464 atomic_inc(&nf_bridge->use);
3465}
3466#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08003467static inline void nf_reset(struct sk_buff *skb)
3468{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003469#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003470 nf_conntrack_put(skb->nfct);
3471 skb->nfct = NULL;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003472#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003473#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003474 nf_bridge_put(skb->nf_bridge);
3475 skb->nf_bridge = NULL;
3476#endif
3477}
3478
Patrick McHardy124dff02013-04-05 20:42:05 +02003479static inline void nf_reset_trace(struct sk_buff *skb)
3480{
Florian Westphal478b3602014-02-15 23:48:45 +01003481#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00003482 skb->nf_trace = 0;
3483#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003484}
3485
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003486/* Note: This doesn't put any conntrack and bridge info in dst. */
Eric Dumazetb1937222014-09-28 22:18:47 -07003487static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3488 bool copy)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003489{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003490#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003491 dst->nfct = src->nfct;
3492 nf_conntrack_get(src->nfct);
Eric Dumazetb1937222014-09-28 22:18:47 -07003493 if (copy)
3494 dst->nfctinfo = src->nfctinfo;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003495#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003496#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003497 dst->nf_bridge = src->nf_bridge;
3498 nf_bridge_get(src->nf_bridge);
3499#endif
Florian Westphal478b3602014-02-15 23:48:45 +01003500#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Eric Dumazetb1937222014-09-28 22:18:47 -07003501 if (copy)
3502 dst->nf_trace = src->nf_trace;
Florian Westphal478b3602014-02-15 23:48:45 +01003503#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003504}
3505
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003506static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3507{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003508#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003509 nf_conntrack_put(dst->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003510#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003511#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003512 nf_bridge_put(dst->nf_bridge);
3513#endif
Eric Dumazetb1937222014-09-28 22:18:47 -07003514 __nf_copy(dst, src, true);
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003515}
3516
James Morris984bc162006-06-09 00:29:17 -07003517#ifdef CONFIG_NETWORK_SECMARK
3518static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3519{
3520 to->secmark = from->secmark;
3521}
3522
3523static inline void skb_init_secmark(struct sk_buff *skb)
3524{
3525 skb->secmark = 0;
3526}
3527#else
3528static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3529{ }
3530
3531static inline void skb_init_secmark(struct sk_buff *skb)
3532{ }
3533#endif
3534
Eric W. Biederman574f7192014-04-01 12:20:24 -07003535static inline bool skb_irq_freeable(const struct sk_buff *skb)
3536{
3537 return !skb->destructor &&
3538#if IS_ENABLED(CONFIG_XFRM)
3539 !skb->sp &&
3540#endif
3541#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3542 !skb->nfct &&
3543#endif
3544 !skb->_skb_refdst &&
3545 !skb_has_frag_list(skb);
3546}
3547
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003548static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3549{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003550 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003551}
3552
Stephen Hemminger92477442009-03-21 13:39:26 -07003553static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003554{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003555 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003556}
3557
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003558static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3559{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003560 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003561}
3562
David S. Millerd5a9e242009-01-27 16:22:11 -08003563static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3564{
3565 skb->queue_mapping = rx_queue + 1;
3566}
3567
Stephen Hemminger92477442009-03-21 13:39:26 -07003568static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003569{
3570 return skb->queue_mapping - 1;
3571}
3572
Stephen Hemminger92477442009-03-21 13:39:26 -07003573static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003574{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003575 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003576}
3577
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003578static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3579{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003580#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003581 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003582#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003583 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003584#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003585}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003586
Pravin B Shelar68c33162013-02-14 14:02:41 +00003587/* Keeps track of mac header offset relative to skb->head.
3588 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3589 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003590 * tunnel skb it points to outer mac header.
3591 * Keeps track of level of encapsulation of network headers.
3592 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003593struct skb_gso_cb {
Eric Dumazet3347c962013-10-19 11:42:56 -07003594 int mac_offset;
3595 int encap_level;
Alexander Duyck76443452016-02-05 15:27:37 -08003596 __wsum csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003597 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003598};
Konstantin Khlebnikov9207f9d2016-01-08 15:21:46 +03003599#define SKB_SGO_CB_OFFSET 32
3600#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
Pravin B Shelar68c33162013-02-14 14:02:41 +00003601
3602static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3603{
3604 return (skb_mac_header(inner_skb) - inner_skb->head) -
3605 SKB_GSO_CB(inner_skb)->mac_offset;
3606}
3607
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003608static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3609{
3610 int new_headroom, headroom;
3611 int ret;
3612
3613 headroom = skb_headroom(skb);
3614 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3615 if (ret)
3616 return ret;
3617
3618 new_headroom = skb_headroom(skb);
3619 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3620 return 0;
3621}
3622
Alexander Duyck08b64fc2016-02-05 15:27:49 -08003623static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3624{
3625 /* Do not update partial checksums if remote checksum is enabled. */
3626 if (skb->remcsum_offload)
3627 return;
3628
3629 SKB_GSO_CB(skb)->csum = res;
3630 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3631}
3632
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003633/* Compute the checksum for a gso segment. First compute the checksum value
3634 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3635 * then add in skb->csum (checksum from csum_start to end of packet).
3636 * skb->csum and csum_start are then updated to reflect the checksum of the
3637 * resultant packet starting from the transport header-- the resultant checksum
3638 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3639 * header.
3640 */
3641static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3642{
Alexander Duyck76443452016-02-05 15:27:37 -08003643 unsigned char *csum_start = skb_transport_header(skb);
3644 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3645 __wsum partial = SKB_GSO_CB(skb)->csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003646
Alexander Duyck76443452016-02-05 15:27:37 -08003647 SKB_GSO_CB(skb)->csum = res;
3648 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003649
Alexander Duyck76443452016-02-05 15:27:37 -08003650 return csum_fold(csum_partial(csum_start, plen, partial));
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003651}
3652
David S. Millerbdcc0922012-03-07 20:53:36 -05003653static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07003654{
3655 return skb_shinfo(skb)->gso_size;
3656}
3657
Eric Dumazet36a8f392013-09-29 01:21:32 -07003658/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05003659static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02003660{
3661 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3662}
3663
Joe Perches7965bd42013-09-26 14:48:15 -07003664void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07003665
3666static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3667{
3668 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3669 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003670 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3671
Alexander Duyckb78462e2010-06-02 12:24:37 +00003672 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3673 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07003674 __skb_warn_lro_forwarding(skb);
3675 return true;
3676 }
3677 return false;
3678}
3679
Herbert Xu35fc92a2007-03-26 23:22:20 -07003680static inline void skb_forward_csum(struct sk_buff *skb)
3681{
3682 /* Unfortunately we don't support this one. Any brave souls? */
3683 if (skb->ip_summed == CHECKSUM_COMPLETE)
3684 skb->ip_summed = CHECKSUM_NONE;
3685}
3686
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003687/**
3688 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3689 * @skb: skb to check
3690 *
3691 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3692 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3693 * use this helper, to document places where we make this assertion.
3694 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003695static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003696{
3697#ifdef DEBUG
3698 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3699#endif
3700}
3701
Rusty Russellf35d9d82008-02-04 23:49:54 -05003702bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00003703
Paul Durranted1f50c2014-01-09 10:02:46 +00003704int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
Linus Lüssing9afd85c2015-05-02 14:01:07 +02003705struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3706 unsigned int transport_len,
3707 __sum16(*skb_chkf)(struct sk_buff *skb));
Paul Durranted1f50c2014-01-09 10:02:46 +00003708
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00003709/**
3710 * skb_head_is_locked - Determine if the skb->head is locked down
3711 * @skb: skb to check
3712 *
3713 * The head on skbs build around a head frag can be removed if they are
3714 * not cloned. This function returns true if the skb head is locked down
3715 * due to either being allocated via kmalloc, or by being a clone with
3716 * multiple references to the head.
3717 */
3718static inline bool skb_head_is_locked(const struct sk_buff *skb)
3719{
3720 return !skb->head_frag || skb_cloned(skb);
3721}
Florian Westphalfe6cc552014-02-13 23:09:12 +01003722
3723/**
3724 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3725 *
3726 * @skb: GSO skb
3727 *
3728 * skb_gso_network_seglen is used to determine the real size of the
3729 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3730 *
3731 * The MAC/L2 header is not accounted for.
3732 */
3733static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3734{
3735 unsigned int hdr_len = skb_transport_header(skb) -
3736 skb_network_header(skb);
3737 return hdr_len + skb_gso_transport_seglen(skb);
3738}
Thomas Grafee122c72015-07-21 10:43:58 +02003739
Edward Cree179bc672016-02-11 20:48:04 +00003740/* Local Checksum Offload.
3741 * Compute outer checksum based on the assumption that the
3742 * inner checksum will be offloaded later.
Edward Creee8ae7b02016-02-11 21:03:37 +00003743 * See Documentation/networking/checksum-offloads.txt for
3744 * explanation of how this works.
Edward Cree179bc672016-02-11 20:48:04 +00003745 * Fill in outer checksum adjustment (e.g. with sum of outer
3746 * pseudo-header) before calling.
3747 * Also ensure that inner checksum is in linear data area.
3748 */
3749static inline __wsum lco_csum(struct sk_buff *skb)
3750{
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003751 unsigned char *csum_start = skb_checksum_start(skb);
3752 unsigned char *l4_hdr = skb_transport_header(skb);
3753 __wsum partial;
Edward Cree179bc672016-02-11 20:48:04 +00003754
3755 /* Start with complement of inner checksum adjustment */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003756 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
3757 skb->csum_offset));
3758
Edward Cree179bc672016-02-11 20:48:04 +00003759 /* Add in checksum of our headers (incl. outer checksum
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003760 * adjustment filled in by caller) and return result.
Edward Cree179bc672016-02-11 20:48:04 +00003761 */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08003762 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
Edward Cree179bc672016-02-11 20:48:04 +00003763}
3764
Linus Torvalds1da177e2005-04-16 15:20:36 -07003765#endif /* __KERNEL__ */
3766#endif /* _LINUX_SKBUFF_H */