blob: a8fc2220e8ce99ae251276af5e09375ce6d000c2 [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
Linus Torvalds1da177e2005-04-16 15:20:36 -0700302
303typedef struct skb_frag_struct skb_frag_t;
304
305struct skb_frag_struct {
Ian Campbella8605c62011-10-19 23:01:49 +0000306 struct {
307 struct page *p;
308 } page;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000309#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
David S. Millera309bb02007-07-30 18:47:03 -0700310 __u32 page_offset;
311 __u32 size;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000312#else
313 __u16 page_offset;
314 __u16 size;
315#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700316};
317
Eric Dumazet9e903e02011-10-18 21:00:24 +0000318static inline unsigned int skb_frag_size(const skb_frag_t *frag)
319{
320 return frag->size;
321}
322
323static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
324{
325 frag->size = size;
326}
327
328static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
329{
330 frag->size += delta;
331}
332
333static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
334{
335 frag->size -= delta;
336}
337
Patrick Ohlyac45f602009-02-12 05:03:37 +0000338#define HAVE_HW_TIME_STAMP
339
340/**
Randy Dunlapd3a21be2009-03-02 03:15:58 -0800341 * struct skb_shared_hwtstamps - hardware time stamps
Patrick Ohlyac45f602009-02-12 05:03:37 +0000342 * @hwtstamp: hardware time stamp transformed into duration
343 * since arbitrary point in time
Patrick Ohlyac45f602009-02-12 05:03:37 +0000344 *
345 * Software time stamps generated by ktime_get_real() are stored in
Willem de Bruijn4d276eb2014-07-25 18:01:32 -0400346 * skb->tstamp.
Patrick Ohlyac45f602009-02-12 05:03:37 +0000347 *
348 * hwtstamps can only be compared against other hwtstamps from
349 * the same device.
350 *
351 * This structure is attached to packets as part of the
352 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
353 */
354struct skb_shared_hwtstamps {
355 ktime_t hwtstamp;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000356};
357
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000358/* Definitions for tx_flags in struct skb_shared_info */
359enum {
360 /* generate hardware time stamp */
361 SKBTX_HW_TSTAMP = 1 << 0,
362
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400363 /* generate software time stamp when queueing packet to NIC */
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000364 SKBTX_SW_TSTAMP = 1 << 1,
365
366 /* device driver is going to provide hardware time stamp */
367 SKBTX_IN_PROGRESS = 1 << 2,
368
Shirley Maa6686f22011-07-06 12:22:12 +0000369 /* device driver supports TX zero-copy buffers */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000370 SKBTX_DEV_ZEROCOPY = 1 << 3,
Johannes Berg6e3e9392011-11-09 10:15:42 +0100371
372 /* generate wifi status information (where possible) */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000373 SKBTX_WIFI_STATUS = 1 << 4,
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000374
375 /* This indicates at least one fragment might be overwritten
376 * (as in vmsplice(), sendfile() ...)
377 * If we need to compute a TX checksum, we'll need to copy
378 * all frags to avoid possible bad checksum
379 */
380 SKBTX_SHARED_FRAG = 1 << 5,
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400381
382 /* generate software time stamp when entering packet scheduling */
383 SKBTX_SCHED_TSTAMP = 1 << 6,
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400384
385 /* generate software timestamp on peer data acknowledgment */
386 SKBTX_ACK_TSTAMP = 1 << 7,
Shirley Maa6686f22011-07-06 12:22:12 +0000387};
388
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400389#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
390 SKBTX_SCHED_TSTAMP | \
391 SKBTX_ACK_TSTAMP)
Willem de Bruijnf24b9be2014-08-04 22:11:45 -0400392#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
393
Shirley Maa6686f22011-07-06 12:22:12 +0000394/*
395 * The callback notifies userspace to release buffers when skb DMA is done in
396 * lower device, the skb last reference should be 0 when calling this.
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000397 * The zerocopy_success argument is true if zero copy transmit occurred,
398 * false on data copy or out of memory error caused by data copy attempt.
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000399 * The ctx field is used to track device context.
400 * The desc field is used to track userspace buffer index.
Shirley Maa6686f22011-07-06 12:22:12 +0000401 */
402struct ubuf_info {
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000403 void (*callback)(struct ubuf_info *, bool zerocopy_success);
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000404 void *ctx;
Shirley Maa6686f22011-07-06 12:22:12 +0000405 unsigned long desc;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000406};
407
Linus Torvalds1da177e2005-04-16 15:20:36 -0700408/* This data is invariant across clones and lives at
409 * the end of the header data, ie. at skb->end.
410 */
411struct skb_shared_info {
Ian Campbell9f42f122012-01-05 07:13:39 +0000412 unsigned char nr_frags;
413 __u8 tx_flags;
Herbert Xu79671682006-06-22 02:40:14 -0700414 unsigned short gso_size;
415 /* Warning: this field is not always filled in (UFO)! */
416 unsigned short gso_segs;
417 unsigned short gso_type;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700418 struct sk_buff *frag_list;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000419 struct skb_shared_hwtstamps hwtstamps;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400420 u32 tskey;
Ian Campbell9f42f122012-01-05 07:13:39 +0000421 __be32 ip6_frag_id;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700422
423 /*
424 * Warning : all fields before dataref are cleared in __alloc_skb()
425 */
426 atomic_t dataref;
427
Johann Baudy69e3c752009-05-18 22:11:22 -0700428 /* Intermediate layers must ensure that destructor_arg
429 * remains valid until skb destructor */
430 void * destructor_arg;
Shirley Maa6686f22011-07-06 12:22:12 +0000431
Eric Dumazetfed66382010-07-22 19:09:08 +0000432 /* must be last field, see pskb_expand_head() */
433 skb_frag_t frags[MAX_SKB_FRAGS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700434};
435
436/* We divide dataref into two halves. The higher 16 bits hold references
437 * to the payload part of skb->data. The lower 16 bits hold references to
Patrick McHardy334a8132007-06-25 04:35:20 -0700438 * the entire skb->data. A clone of a headerless skb holds the length of
439 * the header in skb->hdr_len.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700440 *
441 * All users must obey the rule that the skb->data reference count must be
442 * greater than or equal to the payload reference count.
443 *
444 * Holding a reference to the payload part means that the user does not
445 * care about modifications to the header part of skb->data.
446 */
447#define SKB_DATAREF_SHIFT 16
448#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
449
David S. Millerd179cd12005-08-17 14:57:30 -0700450
451enum {
Vijay Subramanianc8753d52014-10-02 10:00:43 -0700452 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
453 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
454 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
David S. Millerd179cd12005-08-17 14:57:30 -0700455};
456
Herbert Xu79671682006-06-22 02:40:14 -0700457enum {
458 SKB_GSO_TCPV4 = 1 << 0,
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700459 SKB_GSO_UDP = 1 << 1,
Herbert Xu576a30e2006-06-27 13:22:38 -0700460
461 /* This indicates the skb is from an untrusted source. */
462 SKB_GSO_DODGY = 1 << 2,
Michael Chanb0da85372006-06-29 12:30:00 -0700463
464 /* This indicates the tcp segment has CWR set. */
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700465 SKB_GSO_TCP_ECN = 1 << 3,
466
467 SKB_GSO_TCPV6 = 1 << 4,
Chris Leech01d5b2f2009-02-27 14:06:49 -0800468
469 SKB_GSO_FCOE = 1 << 5,
Pravin B Shelar68c33162013-02-14 14:02:41 +0000470
471 SKB_GSO_GRE = 1 << 6,
Pravin B Shelar73136262013-03-07 13:21:51 +0000472
Tom Herbert4b282522014-06-14 23:23:52 -0700473 SKB_GSO_GRE_CSUM = 1 << 7,
Simon Horman0d89d202013-05-23 21:02:52 +0000474
Tom Herbert4b282522014-06-14 23:23:52 -0700475 SKB_GSO_IPIP = 1 << 8,
Eric Dumazetcb32f512013-10-19 11:42:57 -0700476
Tom Herbert4b282522014-06-14 23:23:52 -0700477 SKB_GSO_SIT = 1 << 9,
Eric Dumazet61c1db72013-10-20 20:47:30 -0700478
Tom Herbert4b282522014-06-14 23:23:52 -0700479 SKB_GSO_UDP_TUNNEL = 1 << 10,
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700480
481 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
Tom Herbert4749c092014-06-04 17:20:23 -0700482
Pravin B Shelar59b93b42014-11-05 15:27:48 -0800483 SKB_GSO_TUNNEL_REMCSUM = 1 << 12,
Herbert Xu79671682006-06-22 02:40:14 -0700484};
485
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700486#if BITS_PER_LONG > 32
487#define NET_SKBUFF_DATA_USES_OFFSET 1
488#endif
489
490#ifdef NET_SKBUFF_DATA_USES_OFFSET
491typedef unsigned int sk_buff_data_t;
492#else
493typedef unsigned char *sk_buff_data_t;
494#endif
495
Eric Dumazet363ec39232014-02-26 14:02:11 -0800496/**
497 * struct skb_mstamp - multi resolution time stamps
498 * @stamp_us: timestamp in us resolution
499 * @stamp_jiffies: timestamp in jiffies
500 */
501struct skb_mstamp {
502 union {
503 u64 v64;
504 struct {
505 u32 stamp_us;
506 u32 stamp_jiffies;
507 };
508 };
509};
510
511/**
512 * skb_mstamp_get - get current timestamp
513 * @cl: place to store timestamps
514 */
515static inline void skb_mstamp_get(struct skb_mstamp *cl)
516{
517 u64 val = local_clock();
518
519 do_div(val, NSEC_PER_USEC);
520 cl->stamp_us = (u32)val;
521 cl->stamp_jiffies = (u32)jiffies;
522}
523
524/**
525 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp
526 * @t1: pointer to newest sample
527 * @t0: pointer to oldest sample
528 */
529static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1,
530 const struct skb_mstamp *t0)
531{
532 s32 delta_us = t1->stamp_us - t0->stamp_us;
533 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies;
534
535 /* If delta_us is negative, this might be because interval is too big,
536 * or local_clock() drift is too big : fallback using jiffies.
537 */
538 if (delta_us <= 0 ||
539 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ)))
540
541 delta_us = jiffies_to_usecs(delta_jiffies);
542
543 return delta_us;
544}
545
Yuchung Cheng625a5e12015-10-16 21:57:45 -0700546static inline bool skb_mstamp_after(const struct skb_mstamp *t1,
547 const struct skb_mstamp *t0)
548{
549 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies;
550
551 if (!diff)
552 diff = t1->stamp_us - t0->stamp_us;
553 return diff > 0;
554}
Eric Dumazet363ec39232014-02-26 14:02:11 -0800555
Linus Torvalds1da177e2005-04-16 15:20:36 -0700556/**
557 * struct sk_buff - socket buffer
558 * @next: Next buffer in list
559 * @prev: Previous buffer in list
Eric Dumazet363ec39232014-02-26 14:02:11 -0800560 * @tstamp: Time we arrived/left
Eric Dumazet56b17422014-11-03 08:19:53 -0800561 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700562 * @sk: Socket we are owned by
Linus Torvalds1da177e2005-04-16 15:20:36 -0700563 * @dev: Device we arrived on/are leaving by
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700564 * @cb: Control buffer. Free for use by every layer. Put private vars here
Eric Dumazet7fee2262010-05-11 23:19:48 +0000565 * @_skb_refdst: destination entry (with norefcount bit)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700566 * @sp: the security path, used for xfrm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700567 * @len: Length of actual data
568 * @data_len: Data length
569 * @mac_len: Length of link layer header
Patrick McHardy334a8132007-06-25 04:35:20 -0700570 * @hdr_len: writable header length of cloned skb
Herbert Xu663ead32007-04-09 11:59:07 -0700571 * @csum: Checksum (must include start/offset pair)
572 * @csum_start: Offset from skb->head where checksumming should start
573 * @csum_offset: Offset from csum_start where checksum should be stored
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700574 * @priority: Packet queueing priority
WANG Cong60ff7462014-05-04 16:39:18 -0700575 * @ignore_df: allow local fragmentation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700576 * @cloned: Head may be cloned (check refcnt to be sure)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700577 * @ip_summed: Driver fed us an IP checksum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700578 * @nohdr: Payload reference only, must not modify header
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700579 * @nfctinfo: Relationship of this skb to the connection
Linus Torvalds1da177e2005-04-16 15:20:36 -0700580 * @pkt_type: Packet class
Randy Dunlapc83c2482005-10-18 22:07:41 -0700581 * @fclone: skbuff clone status
Randy Dunlapc83c2482005-10-18 22:07:41 -0700582 * @ipvs_property: skbuff is owned by ipvs
Randy Dunlap31729362008-02-18 20:52:13 -0800583 * @peeked: this packet has been seen already, so stats have been
584 * done for it, don't do them again
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700585 * @nf_trace: netfilter packet trace flag
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700586 * @protocol: Packet protocol from driver
587 * @destructor: Destruct function
588 * @nfct: Associated connection, if any
Linus Torvalds1da177e2005-04-16 15:20:36 -0700589 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
Eric Dumazet8964be42009-11-20 15:35:04 -0800590 * @skb_iif: ifindex of device we arrived on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700591 * @tc_index: Traffic control index
592 * @tc_verd: traffic control verdict
Tom Herbert61b905d2014-03-24 15:34:47 -0700593 * @hash: the packet hash
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700594 * @queue_mapping: Queue mapping for multiqueue devices
David S. Miller0b725a22014-08-25 15:51:53 -0700595 * @xmit_more: More SKBs are pending for this queue
Randy Dunlap553a5672008-04-20 10:51:01 -0700596 * @ndisc_nodetype: router type (from link layer)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700597 * @ooo_okay: allow the mapping of a socket to a queue to be changed
Tom Herbert61b905d2014-03-24 15:34:47 -0700598 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
Changli Gao4ca24622011-08-19 07:26:44 -0700599 * ports.
Tom Herberta3b18dd2014-07-01 21:33:17 -0700600 * @sw_hash: indicates hash was computed in software stack
Johannes Berg6e3e9392011-11-09 10:15:42 +0100601 * @wifi_acked_valid: wifi_acked was set
602 * @wifi_acked: whether frame was acked on wifi or not
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000603 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
Eliezer Tamir06021292013-06-10 11:39:50 +0300604 * @napi_id: id of the NAPI struct this skb came from
James Morris984bc162006-06-09 00:29:17 -0700605 * @secmark: security marking
Scott Feldman0c4f6912015-07-18 18:24:48 -0700606 * @offload_fwd_mark: fwding offload mark
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700607 * @mark: Generic packet mark
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000608 * @vlan_proto: vlan encapsulation protocol
Patrick McHardy6aa895b2008-07-14 22:49:06 -0700609 * @vlan_tci: vlan tag control information
Simon Horman0d89d202013-05-23 21:02:52 +0000610 * @inner_protocol: Protocol (encapsulation)
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000611 * @inner_transport_header: Inner transport layer header (encapsulation)
612 * @inner_network_header: Network layer header (encapsulation)
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000613 * @inner_mac_header: Link layer header (encapsulation)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700614 * @transport_header: Transport layer header
615 * @network_header: Network layer header
616 * @mac_header: Link layer header
617 * @tail: Tail pointer
618 * @end: End pointer
619 * @head: Head of buffer
620 * @data: Data head pointer
621 * @truesize: Buffer size
622 * @users: User count - see {datagram,tcp}.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 */
624
625struct sk_buff {
Eric Dumazet363ec39232014-02-26 14:02:11 -0800626 union {
Eric Dumazet56b17422014-11-03 08:19:53 -0800627 struct {
628 /* These two members must be first. */
629 struct sk_buff *next;
630 struct sk_buff *prev;
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000631
Eric Dumazet56b17422014-11-03 08:19:53 -0800632 union {
633 ktime_t tstamp;
634 struct skb_mstamp skb_mstamp;
635 };
636 };
637 struct rb_node rbnode; /* used in netem & tcp stack */
638 };
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000639 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700640 struct net_device *dev;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700641
Linus Torvalds1da177e2005-04-16 15:20:36 -0700642 /*
643 * This is the control buffer. It is free to use for every
644 * layer. Please put your private variables there. If you
645 * want to keep them across layers you have to do a skb_clone()
646 * first. This is owned by whoever has the skb queued ATM.
647 */
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000648 char cb[48] __aligned(8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700649
Eric Dumazet7fee2262010-05-11 23:19:48 +0000650 unsigned long _skb_refdst;
Eric Dumazetb1937222014-09-28 22:18:47 -0700651 void (*destructor)(struct sk_buff *skb);
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000652#ifdef CONFIG_XFRM
653 struct sec_path *sp;
654#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700655#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
656 struct nf_conntrack *nfct;
657#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200658#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Eric Dumazetb1937222014-09-28 22:18:47 -0700659 struct nf_bridge_info *nf_bridge;
660#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700661 unsigned int len,
Patrick McHardy334a8132007-06-25 04:35:20 -0700662 data_len;
663 __u16 mac_len,
664 hdr_len;
Eric Dumazetb1937222014-09-28 22:18:47 -0700665
666 /* Following fields are _not_ copied in __copy_skb_header()
667 * Note that queue_mapping is here mostly to fill a hole.
668 */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200669 kmemcheck_bitfield_begin(flags1);
Eric Dumazetb1937222014-09-28 22:18:47 -0700670 __u16 queue_mapping;
671 __u8 cloned:1,
Harald Welte6869c4d2005-08-09 19:24:19 -0700672 nohdr:1,
Patrick McHardyb84f4cc2005-11-20 21:19:21 -0800673 fclone:2,
Herbert Xua59322b2007-12-05 01:53:40 -0800674 peeked:1,
Eric Dumazetb1937222014-09-28 22:18:47 -0700675 head_frag:1,
676 xmit_more:1;
677 /* one bit hole */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200678 kmemcheck_bitfield_end(flags1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700679
Eric Dumazetb1937222014-09-28 22:18:47 -0700680 /* fields enclosed in headers_start/headers_end are copied
681 * using a single memcpy() in __copy_skb_header()
682 */
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700683 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700684 __u32 headers_start[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700685 /* public: */
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200686
687/* if you move pkt_type around you also must adapt those constants */
688#ifdef __BIG_ENDIAN_BITFIELD
689#define PKT_TYPE_MAX (7 << 5)
690#else
691#define PKT_TYPE_MAX 7
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100692#endif
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200693#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
694
695 __u8 __pkt_type_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700696 __u8 pkt_type:3;
697 __u8 pfmemalloc:1;
698 __u8 ignore_df:1;
699 __u8 nfctinfo:3;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700
Eric Dumazetb1937222014-09-28 22:18:47 -0700701 __u8 nf_trace:1;
702 __u8 ip_summed:2;
703 __u8 ooo_okay:1;
704 __u8 l4_hash:1;
705 __u8 sw_hash:1;
706 __u8 wifi_acked_valid:1;
707 __u8 wifi_acked:1;
708
709 __u8 no_fcs:1;
710 /* Indicates the inner headers are valid in the skbuff. */
711 __u8 encapsulation:1;
712 __u8 encap_hdr_csum:1;
713 __u8 csum_valid:1;
714 __u8 csum_complete_sw:1;
715 __u8 csum_level:2;
716 __u8 csum_bad:1;
717
718#ifdef CONFIG_IPV6_NDISC_NODETYPE
719 __u8 ndisc_nodetype:2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700720#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700721 __u8 ipvs_property:1;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700722 __u8 inner_protocol_type:1;
Tom Herberte585f232014-11-04 09:06:54 -0800723 __u8 remcsum_offload:1;
724 /* 3 or 5 bit hole */
Alexander Duyck4031ae62012-01-27 06:22:53 +0000725
Linus Torvalds1da177e2005-04-16 15:20:36 -0700726#ifdef CONFIG_NET_SCHED
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700727 __u16 tc_index; /* traffic control index */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728#ifdef CONFIG_NET_CLS_ACT
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700729 __u16 tc_verd; /* traffic control verdict */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700730#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700731#endif
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200732
Eric Dumazetb1937222014-09-28 22:18:47 -0700733 union {
734 __wsum csum;
735 struct {
736 __u16 csum_start;
737 __u16 csum_offset;
738 };
739 };
740 __u32 priority;
741 int skb_iif;
742 __u32 hash;
743 __be16 vlan_proto;
744 __u16 vlan_tci;
Eric Dumazet2bd82482015-02-03 23:48:24 -0800745#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
746 union {
747 unsigned int napi_id;
748 unsigned int sender_cpu;
749 };
Chris Leech97fc2f02006-05-23 17:55:33 -0700750#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700751 union {
James Morris984bc162006-06-09 00:29:17 -0700752#ifdef CONFIG_NETWORK_SECMARK
Scott Feldman0c4f6912015-07-18 18:24:48 -0700753 __u32 secmark;
James Morris984bc162006-06-09 00:29:17 -0700754#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700755#ifdef CONFIG_NET_SWITCHDEV
756 __u32 offload_fwd_mark;
757#endif
758 };
759
Neil Horman3b885782009-10-12 13:26:31 -0700760 union {
761 __u32 mark;
Eric Dumazet16fad692013-03-14 05:40:32 +0000762 __u32 reserved_tailroom;
Neil Horman3b885782009-10-12 13:26:31 -0700763 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700764
Tom Herbert8bce6d72014-09-29 20:22:29 -0700765 union {
766 __be16 inner_protocol;
767 __u8 inner_ipproto;
768 };
769
Simon Horman1a37e412013-05-23 21:02:51 +0000770 __u16 inner_transport_header;
771 __u16 inner_network_header;
772 __u16 inner_mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700773
774 __be16 protocol;
Simon Horman1a37e412013-05-23 21:02:51 +0000775 __u16 transport_header;
776 __u16 network_header;
777 __u16 mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700778
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700779 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700780 __u32 headers_end[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700781 /* public: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700782
Linus Torvalds1da177e2005-04-16 15:20:36 -0700783 /* These elements must be at the end, see alloc_skb() for details. */
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700784 sk_buff_data_t tail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700785 sk_buff_data_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786 unsigned char *head,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700787 *data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700788 unsigned int truesize;
789 atomic_t users;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700790};
791
792#ifdef __KERNEL__
793/*
794 * Handling routines are only of interest to the kernel
795 */
796#include <linux/slab.h>
797
Linus Torvalds1da177e2005-04-16 15:20:36 -0700798
Mel Gormanc93bdd02012-07-31 16:44:19 -0700799#define SKB_ALLOC_FCLONE 0x01
800#define SKB_ALLOC_RX 0x02
Alexander Duyckfd11a832014-12-09 19:40:49 -0800801#define SKB_ALLOC_NAPI 0x04
Mel Gormanc93bdd02012-07-31 16:44:19 -0700802
803/* Returns true if the skb was allocated from PFMEMALLOC reserves */
804static inline bool skb_pfmemalloc(const struct sk_buff *skb)
805{
806 return unlikely(skb->pfmemalloc);
807}
808
Eric Dumazet7fee2262010-05-11 23:19:48 +0000809/*
810 * skb might have a dst pointer attached, refcounted or not.
811 * _skb_refdst low order bit is set if refcount was _not_ taken
812 */
813#define SKB_DST_NOREF 1UL
814#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
815
816/**
817 * skb_dst - returns skb dst_entry
818 * @skb: buffer
819 *
820 * Returns skb dst_entry, regardless of reference taken or not.
821 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000822static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
823{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000824 /* If refdst was not refcounted, check we still are in a
825 * rcu_read_lock section
826 */
827 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
828 !rcu_read_lock_held() &&
829 !rcu_read_lock_bh_held());
830 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
Eric Dumazetadf30902009-06-02 05:19:30 +0000831}
832
Eric Dumazet7fee2262010-05-11 23:19:48 +0000833/**
834 * skb_dst_set - sets skb dst
835 * @skb: buffer
836 * @dst: dst entry
837 *
838 * Sets skb dst, assuming a reference was taken on dst and should
839 * be released by skb_dst_drop()
840 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000841static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
842{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000843 skb->_skb_refdst = (unsigned long)dst;
844}
845
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200846/**
847 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
848 * @skb: buffer
849 * @dst: dst entry
850 *
851 * Sets skb dst, assuming a reference was not taken on dst.
852 * If dst entry is cached, we do not take reference and dst_release
853 * will be avoided by refdst_drop. If dst entry is not cached, we take
854 * reference, so that last dst_release can destroy the dst immediately.
855 */
856static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
857{
Hannes Frederic Sowadbfc4fb2014-12-06 19:19:42 +0100858 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
859 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200860}
Eric Dumazet7fee2262010-05-11 23:19:48 +0000861
862/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300863 * skb_dst_is_noref - Test if skb dst isn't refcounted
Eric Dumazet7fee2262010-05-11 23:19:48 +0000864 * @skb: buffer
865 */
866static inline bool skb_dst_is_noref(const struct sk_buff *skb)
867{
868 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
Eric Dumazetadf30902009-06-02 05:19:30 +0000869}
870
Eric Dumazet511c3f92009-06-02 05:14:27 +0000871static inline struct rtable *skb_rtable(const struct sk_buff *skb)
872{
Eric Dumazetadf30902009-06-02 05:19:30 +0000873 return (struct rtable *)skb_dst(skb);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000874}
875
Joe Perches7965bd42013-09-26 14:48:15 -0700876void kfree_skb(struct sk_buff *skb);
877void kfree_skb_list(struct sk_buff *segs);
878void skb_tx_error(struct sk_buff *skb);
879void consume_skb(struct sk_buff *skb);
880void __kfree_skb(struct sk_buff *skb);
Eric Dumazetd7e88832012-04-30 08:10:34 +0000881extern struct kmem_cache *skbuff_head_cache;
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000882
Joe Perches7965bd42013-09-26 14:48:15 -0700883void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
884bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
885 bool *fragstolen, int *delta_truesize);
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000886
Joe Perches7965bd42013-09-26 14:48:15 -0700887struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
888 int node);
Eric Dumazet2ea2f622015-04-24 16:05:01 -0700889struct sk_buff *__build_skb(void *data, unsigned int frag_size);
Joe Perches7965bd42013-09-26 14:48:15 -0700890struct sk_buff *build_skb(void *data, unsigned int frag_size);
David S. Millerd179cd12005-08-17 14:57:30 -0700891static inline struct sk_buff *alloc_skb(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100892 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700893{
Eric Dumazet564824b2010-10-11 19:05:25 +0000894 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700895}
896
Eric Dumazet2e4e4412014-09-17 04:49:49 -0700897struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
898 unsigned long data_len,
899 int max_page_order,
900 int *errcode,
901 gfp_t gfp_mask);
902
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700903/* Layout of fast clones : [skb1][skb2][fclone_ref] */
904struct sk_buff_fclones {
905 struct sk_buff skb1;
906
907 struct sk_buff skb2;
908
909 atomic_t fclone_ref;
910};
911
912/**
913 * skb_fclone_busy - check if fclone is busy
914 * @skb: buffer
915 *
Masanari Iidabda13fe2015-12-13 16:53:02 +0900916 * Returns true if skb is a fast clone, and its clone is not freed.
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700917 * Some drivers call skb_orphan() in their ndo_start_xmit(),
918 * so we also check that this didnt happen.
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700919 */
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700920static inline bool skb_fclone_busy(const struct sock *sk,
921 const struct sk_buff *skb)
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700922{
923 const struct sk_buff_fclones *fclones;
924
925 fclones = container_of(skb, struct sk_buff_fclones, skb1);
926
927 return skb->fclone == SKB_FCLONE_ORIG &&
Eric Dumazet6ffe75eb2014-12-03 17:04:39 -0800928 atomic_read(&fclones->fclone_ref) > 1 &&
Eric Dumazet39bb5e62014-10-30 10:32:34 -0700929 fclones->skb2.sk == sk;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700930}
931
David S. Millerd179cd12005-08-17 14:57:30 -0700932static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100933 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700934{
Mel Gormanc93bdd02012-07-31 16:44:19 -0700935 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700936}
937
Joe Perches7965bd42013-09-26 14:48:15 -0700938struct sk_buff *__alloc_skb_head(gfp_t priority, int node);
Patrick McHardy0ebd0ac2013-04-17 06:46:58 +0000939static inline struct sk_buff *alloc_skb_head(gfp_t priority)
940{
941 return __alloc_skb_head(priority, -1);
942}
943
Joe Perches7965bd42013-09-26 14:48:15 -0700944struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
945int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
946struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
947struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
Octavian Purdilabad93e92014-06-12 01:36:26 +0300948struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
949 gfp_t gfp_mask, bool fclone);
950static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
951 gfp_t gfp_mask)
952{
953 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
954}
Eric Dumazet117632e2011-12-03 21:39:53 +0000955
Joe Perches7965bd42013-09-26 14:48:15 -0700956int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
957struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
958 unsigned int headroom);
959struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
960 int newtailroom, gfp_t priority);
Fan Du25a91d82014-01-18 09:54:23 +0800961int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
962 int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -0700963int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset,
964 int len);
965int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
966int skb_pad(struct sk_buff *skb, int pad);
Neil Hormanead2ceb2009-03-11 09:49:55 +0000967#define dev_kfree_skb(a) consume_skb(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700968
Joe Perches7965bd42013-09-26 14:48:15 -0700969int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
970 int getfrag(void *from, char *to, int offset,
971 int len, int odd, struct sk_buff *skb),
972 void *from, int length);
Ananda Rajue89e9cf2005-10-18 15:46:41 -0700973
Hannes Frederic Sowabe12a1f2015-05-21 16:59:58 +0200974int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
975 int offset, size_t size);
976
Eric Dumazetd94d9fe2009-11-04 09:50:58 -0800977struct skb_seq_state {
Thomas Graf677e90e2005-06-23 20:59:51 -0700978 __u32 lower_offset;
979 __u32 upper_offset;
980 __u32 frag_idx;
981 __u32 stepped_offset;
982 struct sk_buff *root_skb;
983 struct sk_buff *cur_skb;
984 __u8 *frag_data;
985};
986
Joe Perches7965bd42013-09-26 14:48:15 -0700987void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
988 unsigned int to, struct skb_seq_state *st);
989unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
990 struct skb_seq_state *st);
991void skb_abort_seq_read(struct skb_seq_state *st);
Thomas Graf677e90e2005-06-23 20:59:51 -0700992
Joe Perches7965bd42013-09-26 14:48:15 -0700993unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
Bojan Prtvar059a2442015-02-22 11:46:35 +0100994 unsigned int to, struct ts_config *config);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -0700995
Tom Herbert09323cc2013-12-15 22:16:19 -0800996/*
997 * Packet hash types specify the type of hash in skb_set_hash.
998 *
999 * Hash types refer to the protocol layer addresses which are used to
1000 * construct a packet's hash. The hashes are used to differentiate or identify
1001 * flows of the protocol layer for the hash type. Hash types are either
1002 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1003 *
1004 * Properties of hashes:
1005 *
1006 * 1) Two packets in different flows have different hash values
1007 * 2) Two packets in the same flow should have the same hash value
1008 *
1009 * A hash at a higher layer is considered to be more specific. A driver should
1010 * set the most specific hash possible.
1011 *
1012 * A driver cannot indicate a more specific hash than the layer at which a hash
1013 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1014 *
1015 * A driver may indicate a hash level which is less specific than the
1016 * actual layer the hash was computed on. For instance, a hash computed
1017 * at L4 may be considered an L3 hash. This should only be done if the
1018 * driver can't unambiguously determine that the HW computed the hash at
1019 * the higher layer. Note that the "should" in the second property above
1020 * permits this.
1021 */
1022enum pkt_hash_types {
1023 PKT_HASH_TYPE_NONE, /* Undefined type */
1024 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1025 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1026 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1027};
1028
Tom Herbertbcc83832015-09-01 09:24:24 -07001029static inline void skb_clear_hash(struct sk_buff *skb)
1030{
1031 skb->hash = 0;
1032 skb->sw_hash = 0;
1033 skb->l4_hash = 0;
1034}
1035
1036static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1037{
1038 if (!skb->l4_hash)
1039 skb_clear_hash(skb);
1040}
1041
1042static inline void
1043__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1044{
1045 skb->l4_hash = is_l4;
1046 skb->sw_hash = is_sw;
1047 skb->hash = hash;
1048}
1049
Tom Herbert09323cc2013-12-15 22:16:19 -08001050static inline void
1051skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1052{
Tom Herbertbcc83832015-09-01 09:24:24 -07001053 /* Used by drivers to set hash from HW */
1054 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1055}
1056
1057static inline void
1058__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1059{
1060 __skb_set_hash(skb, hash, true, is_l4);
Tom Herbert09323cc2013-12-15 22:16:19 -08001061}
1062
Tom Herberte5276932015-09-01 09:24:23 -07001063void __skb_get_hash(struct sk_buff *skb);
1064u32 skb_get_poff(const struct sk_buff *skb);
1065u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1066 const struct flow_keys *keys, int hlen);
1067__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1068 void *data, int hlen_proto);
1069
1070static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1071 int thoff, u8 ip_proto)
1072{
1073 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1074}
1075
1076void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1077 const struct flow_dissector_key *key,
1078 unsigned int key_count);
1079
1080bool __skb_flow_dissect(const struct sk_buff *skb,
1081 struct flow_dissector *flow_dissector,
1082 void *target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001083 void *data, __be16 proto, int nhoff, int hlen,
1084 unsigned int flags);
Tom Herberte5276932015-09-01 09:24:23 -07001085
1086static inline bool skb_flow_dissect(const struct sk_buff *skb,
1087 struct flow_dissector *flow_dissector,
Tom Herbertcd79a232015-09-01 09:24:27 -07001088 void *target_container, unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001089{
1090 return __skb_flow_dissect(skb, flow_dissector, target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001091 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001092}
1093
1094static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
Tom Herbertcd79a232015-09-01 09:24:27 -07001095 struct flow_keys *flow,
1096 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001097{
1098 memset(flow, 0, sizeof(*flow));
1099 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001100 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001101}
1102
1103static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1104 void *data, __be16 proto,
Tom Herbertcd79a232015-09-01 09:24:27 -07001105 int nhoff, int hlen,
1106 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001107{
1108 memset(flow, 0, sizeof(*flow));
1109 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001110 data, proto, nhoff, hlen, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001111}
1112
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001113static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001114{
Tom Herberta3b18dd2014-07-01 21:33:17 -07001115 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001116 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001117
Tom Herbert61b905d2014-03-24 15:34:47 -07001118 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001119}
1120
David S. Miller20a17bf2015-09-01 21:19:17 -07001121__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6);
Tom Herbertf70ea012015-07-31 16:52:10 -07001122
David S. Miller20a17bf2015-09-01 21:19:17 -07001123static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
Tom Herbertf70ea012015-07-31 16:52:10 -07001124{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001125 if (!skb->l4_hash && !skb->sw_hash) {
1126 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001127 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001128
Tom Herbertde4c1f82015-09-01 18:11:04 -07001129 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001130 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001131
1132 return skb->hash;
1133}
1134
David S. Miller20a17bf2015-09-01 21:19:17 -07001135__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl);
Tom Herbertf70ea012015-07-31 16:52:10 -07001136
David S. Miller20a17bf2015-09-01 21:19:17 -07001137static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4)
Tom Herbertf70ea012015-07-31 16:52:10 -07001138{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001139 if (!skb->l4_hash && !skb->sw_hash) {
1140 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001141 __u32 hash = __get_hash_from_flowi4(fl4, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001142
Tom Herbertde4c1f82015-09-01 18:11:04 -07001143 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001144 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001145
1146 return skb->hash;
1147}
1148
Tom Herbert50fb7992015-05-01 11:30:12 -07001149__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1150
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001151static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1152{
Tom Herbert61b905d2014-03-24 15:34:47 -07001153 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001154}
1155
Tom Herbert3df7a742013-12-15 22:16:29 -08001156static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1157{
Tom Herbert61b905d2014-03-24 15:34:47 -07001158 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -07001159 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -07001160 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -08001161};
1162
Eric Dumazetc29390c2015-03-11 18:42:02 -07001163static inline void skb_sender_cpu_clear(struct sk_buff *skb)
1164{
Eric Dumazetc29390c2015-03-11 18:42:02 -07001165}
1166
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001167#ifdef NET_SKBUFF_DATA_USES_OFFSET
1168static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1169{
1170 return skb->head + skb->end;
1171}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001172
1173static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1174{
1175 return skb->end;
1176}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001177#else
1178static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1179{
1180 return skb->end;
1181}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001182
1183static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1184{
1185 return skb->end - skb->head;
1186}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001187#endif
1188
Linus Torvalds1da177e2005-04-16 15:20:36 -07001189/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001190#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001191
Patrick Ohlyac45f602009-02-12 05:03:37 +00001192static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1193{
1194 return &skb_shinfo(skb)->hwtstamps;
1195}
1196
Linus Torvalds1da177e2005-04-16 15:20:36 -07001197/**
1198 * skb_queue_empty - check if a queue is empty
1199 * @list: queue head
1200 *
1201 * Returns true if the queue is empty, false otherwise.
1202 */
1203static inline int skb_queue_empty(const struct sk_buff_head *list)
1204{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001205 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001206}
1207
1208/**
David S. Millerfc7ebb22008-09-23 00:34:07 -07001209 * skb_queue_is_last - check if skb is the last entry in the queue
1210 * @list: queue head
1211 * @skb: buffer
1212 *
1213 * Returns true if @skb is the last buffer on the list.
1214 */
1215static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1216 const struct sk_buff *skb)
1217{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001218 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -07001219}
1220
1221/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001222 * skb_queue_is_first - check if skb is the first entry in the queue
1223 * @list: queue head
1224 * @skb: buffer
1225 *
1226 * Returns true if @skb is the first buffer on the list.
1227 */
1228static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1229 const struct sk_buff *skb)
1230{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001231 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001232}
1233
1234/**
David S. Miller249c8b42008-09-23 00:44:42 -07001235 * skb_queue_next - return the next packet in the queue
1236 * @list: queue head
1237 * @skb: current buffer
1238 *
1239 * Return the next packet in @list after @skb. It is only valid to
1240 * call this if skb_queue_is_last() evaluates to false.
1241 */
1242static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1243 const struct sk_buff *skb)
1244{
1245 /* This BUG_ON may seem severe, but if we just return then we
1246 * are going to dereference garbage.
1247 */
1248 BUG_ON(skb_queue_is_last(list, skb));
1249 return skb->next;
1250}
1251
1252/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001253 * skb_queue_prev - return the prev packet in the queue
1254 * @list: queue head
1255 * @skb: current buffer
1256 *
1257 * Return the prev packet in @list before @skb. It is only valid to
1258 * call this if skb_queue_is_first() evaluates to false.
1259 */
1260static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1261 const struct sk_buff *skb)
1262{
1263 /* This BUG_ON may seem severe, but if we just return then we
1264 * are going to dereference garbage.
1265 */
1266 BUG_ON(skb_queue_is_first(list, skb));
1267 return skb->prev;
1268}
1269
1270/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001271 * skb_get - reference buffer
1272 * @skb: buffer to reference
1273 *
1274 * Makes another reference to a socket buffer and returns a pointer
1275 * to the buffer.
1276 */
1277static inline struct sk_buff *skb_get(struct sk_buff *skb)
1278{
1279 atomic_inc(&skb->users);
1280 return skb;
1281}
1282
1283/*
1284 * If users == 1, we are the only owner and are can avoid redundant
1285 * atomic change.
1286 */
1287
1288/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001289 * skb_cloned - is the buffer a clone
1290 * @skb: buffer to check
1291 *
1292 * Returns true if the buffer was generated with skb_clone() and is
1293 * one of multiple shared copies of the buffer. Cloned buffers are
1294 * shared data so must not be written to under normal circumstances.
1295 */
1296static inline int skb_cloned(const struct sk_buff *skb)
1297{
1298 return skb->cloned &&
1299 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1300}
1301
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001302static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1303{
Mel Gormand0164ad2015-11-06 16:28:21 -08001304 might_sleep_if(gfpflags_allow_blocking(pri));
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001305
1306 if (skb_cloned(skb))
1307 return pskb_expand_head(skb, 0, 0, pri);
1308
1309 return 0;
1310}
1311
Linus Torvalds1da177e2005-04-16 15:20:36 -07001312/**
1313 * skb_header_cloned - is the header a clone
1314 * @skb: buffer to check
1315 *
1316 * Returns true if modifying the header part of the buffer requires
1317 * the data to be copied.
1318 */
1319static inline int skb_header_cloned(const struct sk_buff *skb)
1320{
1321 int dataref;
1322
1323 if (!skb->cloned)
1324 return 0;
1325
1326 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1327 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1328 return dataref != 1;
1329}
1330
1331/**
1332 * skb_header_release - release reference to header
1333 * @skb: buffer to operate on
1334 *
1335 * Drop a reference to the header part of the buffer. This is done
1336 * by acquiring a payload reference. You must not read from the header
1337 * part of skb->data after this.
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001338 * Note : Check if you can use __skb_header_release() instead.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001339 */
1340static inline void skb_header_release(struct sk_buff *skb)
1341{
1342 BUG_ON(skb->nohdr);
1343 skb->nohdr = 1;
1344 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref);
1345}
1346
1347/**
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001348 * __skb_header_release - release reference to header
1349 * @skb: buffer to operate on
1350 *
1351 * Variant of skb_header_release() assuming skb is private to caller.
1352 * We can avoid one atomic operation.
1353 */
1354static inline void __skb_header_release(struct sk_buff *skb)
1355{
1356 skb->nohdr = 1;
1357 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1358}
1359
1360
1361/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001362 * skb_shared - is the buffer shared
1363 * @skb: buffer to check
1364 *
1365 * Returns true if more than one person has a reference to this
1366 * buffer.
1367 */
1368static inline int skb_shared(const struct sk_buff *skb)
1369{
1370 return atomic_read(&skb->users) != 1;
1371}
1372
1373/**
1374 * skb_share_check - check if buffer is shared and if so clone it
1375 * @skb: buffer to check
1376 * @pri: priority for memory allocation
1377 *
1378 * If the buffer is shared the buffer is cloned and the old copy
1379 * drops a reference. A new clone with a single reference is returned.
1380 * If the buffer is not shared the original buffer is returned. When
1381 * being called from interrupt status or with spinlocks held pri must
1382 * be GFP_ATOMIC.
1383 *
1384 * NULL is returned on a memory allocation failure.
1385 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001386static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001387{
Mel Gormand0164ad2015-11-06 16:28:21 -08001388 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001389 if (skb_shared(skb)) {
1390 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001391
1392 if (likely(nskb))
1393 consume_skb(skb);
1394 else
1395 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001396 skb = nskb;
1397 }
1398 return skb;
1399}
1400
1401/*
1402 * Copy shared buffers into a new sk_buff. We effectively do COW on
1403 * packets to handle cases where we have a local reader and forward
1404 * and a couple of other messy ones. The normal one is tcpdumping
1405 * a packet thats being forwarded.
1406 */
1407
1408/**
1409 * skb_unshare - make a copy of a shared buffer
1410 * @skb: buffer to check
1411 * @pri: priority for memory allocation
1412 *
1413 * If the socket buffer is a clone then this function creates a new
1414 * copy of the data, drops a reference count on the old copy and returns
1415 * the new copy with the reference count at 1. If the buffer is not a clone
1416 * the original buffer is returned. When called with a spinlock held or
1417 * from interrupt state @pri must be %GFP_ATOMIC
1418 *
1419 * %NULL is returned on a memory allocation failure.
1420 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001421static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001422 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001423{
Mel Gormand0164ad2015-11-06 16:28:21 -08001424 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001425 if (skb_cloned(skb)) {
1426 struct sk_buff *nskb = skb_copy(skb, pri);
Alexander Aring31eff812014-10-10 23:10:47 +02001427
1428 /* Free our shared copy */
1429 if (likely(nskb))
1430 consume_skb(skb);
1431 else
1432 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001433 skb = nskb;
1434 }
1435 return skb;
1436}
1437
1438/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001439 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001440 * @list_: list to peek at
1441 *
1442 * Peek an &sk_buff. Unlike most other operations you _MUST_
1443 * be careful with this one. A peek leaves the buffer on the
1444 * list and someone else may run off with it. You must hold
1445 * the appropriate locks or have a private queue to do this.
1446 *
1447 * Returns %NULL for an empty list or a pointer to the head element.
1448 * The reference count is not incremented and the reference is therefore
1449 * volatile. Use with caution.
1450 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001451static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001452{
Eric Dumazet18d07002012-04-30 16:31:46 +00001453 struct sk_buff *skb = list_->next;
1454
1455 if (skb == (struct sk_buff *)list_)
1456 skb = NULL;
1457 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001458}
1459
1460/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001461 * skb_peek_next - peek skb following the given one from a queue
1462 * @skb: skb to start from
1463 * @list_: list to peek at
1464 *
1465 * Returns %NULL when the end of the list is met or a pointer to the
1466 * next element. The reference count is not incremented and the
1467 * reference is therefore volatile. Use with caution.
1468 */
1469static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1470 const struct sk_buff_head *list_)
1471{
1472 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001473
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001474 if (next == (struct sk_buff *)list_)
1475 next = NULL;
1476 return next;
1477}
1478
1479/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001480 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001481 * @list_: list to peek at
1482 *
1483 * Peek an &sk_buff. Unlike most other operations you _MUST_
1484 * be careful with this one. A peek leaves the buffer on the
1485 * list and someone else may run off with it. You must hold
1486 * the appropriate locks or have a private queue to do this.
1487 *
1488 * Returns %NULL for an empty list or a pointer to the tail element.
1489 * The reference count is not incremented and the reference is therefore
1490 * volatile. Use with caution.
1491 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001492static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001493{
Eric Dumazet18d07002012-04-30 16:31:46 +00001494 struct sk_buff *skb = list_->prev;
1495
1496 if (skb == (struct sk_buff *)list_)
1497 skb = NULL;
1498 return skb;
1499
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500}
1501
1502/**
1503 * skb_queue_len - get queue length
1504 * @list_: list to measure
1505 *
1506 * Return the length of an &sk_buff queue.
1507 */
1508static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1509{
1510 return list_->qlen;
1511}
1512
David S. Miller67fed452008-09-21 22:36:24 -07001513/**
1514 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1515 * @list: queue to initialize
1516 *
1517 * This initializes only the list and queue length aspects of
1518 * an sk_buff_head object. This allows to initialize the list
1519 * aspects of an sk_buff_head without reinitializing things like
1520 * the spinlock. It can also be used for on-stack sk_buff_head
1521 * objects where the spinlock is known to not be used.
1522 */
1523static inline void __skb_queue_head_init(struct sk_buff_head *list)
1524{
1525 list->prev = list->next = (struct sk_buff *)list;
1526 list->qlen = 0;
1527}
1528
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001529/*
1530 * This function creates a split out lock class for each invocation;
1531 * this is needed for now since a whole lot of users of the skb-queue
1532 * infrastructure in drivers have different locking usage (in hardirq)
1533 * than the networking core (in softirq only). In the long run either the
1534 * network layer or drivers should need annotation to consolidate the
1535 * main types of usage into 3 classes.
1536 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001537static inline void skb_queue_head_init(struct sk_buff_head *list)
1538{
1539 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001540 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541}
1542
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001543static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1544 struct lock_class_key *class)
1545{
1546 skb_queue_head_init(list);
1547 lockdep_set_class(&list->lock, class);
1548}
1549
Linus Torvalds1da177e2005-04-16 15:20:36 -07001550/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001551 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001552 *
1553 * The "__skb_xxxx()" functions are the non-atomic ones that
1554 * can only be called with interrupts disabled.
1555 */
Joe Perches7965bd42013-09-26 14:48:15 -07001556void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1557 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001558static inline void __skb_insert(struct sk_buff *newsk,
1559 struct sk_buff *prev, struct sk_buff *next,
1560 struct sk_buff_head *list)
1561{
1562 newsk->next = next;
1563 newsk->prev = prev;
1564 next->prev = prev->next = newsk;
1565 list->qlen++;
1566}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001567
David S. Miller67fed452008-09-21 22:36:24 -07001568static inline void __skb_queue_splice(const struct sk_buff_head *list,
1569 struct sk_buff *prev,
1570 struct sk_buff *next)
1571{
1572 struct sk_buff *first = list->next;
1573 struct sk_buff *last = list->prev;
1574
1575 first->prev = prev;
1576 prev->next = first;
1577
1578 last->next = next;
1579 next->prev = last;
1580}
1581
1582/**
1583 * skb_queue_splice - join two skb lists, this is designed for stacks
1584 * @list: the new list to add
1585 * @head: the place to add it in the first list
1586 */
1587static inline void skb_queue_splice(const struct sk_buff_head *list,
1588 struct sk_buff_head *head)
1589{
1590 if (!skb_queue_empty(list)) {
1591 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001592 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001593 }
1594}
1595
1596/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001597 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001598 * @list: the new list to add
1599 * @head: the place to add it in the first list
1600 *
1601 * The list at @list is reinitialised
1602 */
1603static inline void skb_queue_splice_init(struct sk_buff_head *list,
1604 struct sk_buff_head *head)
1605{
1606 if (!skb_queue_empty(list)) {
1607 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001608 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001609 __skb_queue_head_init(list);
1610 }
1611}
1612
1613/**
1614 * skb_queue_splice_tail - join two skb lists, each list being a queue
1615 * @list: the new list to add
1616 * @head: the place to add it in the first list
1617 */
1618static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1619 struct sk_buff_head *head)
1620{
1621 if (!skb_queue_empty(list)) {
1622 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001623 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001624 }
1625}
1626
1627/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001628 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001629 * @list: the new list to add
1630 * @head: the place to add it in the first list
1631 *
1632 * Each of the lists is a queue.
1633 * The list at @list is reinitialised
1634 */
1635static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1636 struct sk_buff_head *head)
1637{
1638 if (!skb_queue_empty(list)) {
1639 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001640 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001641 __skb_queue_head_init(list);
1642 }
1643}
1644
Linus Torvalds1da177e2005-04-16 15:20:36 -07001645/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001646 * __skb_queue_after - queue a buffer at the list head
1647 * @list: list to use
1648 * @prev: place after this buffer
1649 * @newsk: buffer to queue
1650 *
1651 * Queue a buffer int the middle of a list. This function takes no locks
1652 * and you must therefore hold required locks before calling it.
1653 *
1654 * A buffer cannot be placed on two lists at the same time.
1655 */
1656static inline void __skb_queue_after(struct sk_buff_head *list,
1657 struct sk_buff *prev,
1658 struct sk_buff *newsk)
1659{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001660 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001661}
1662
Joe Perches7965bd42013-09-26 14:48:15 -07001663void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1664 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001665
Gerrit Renkerf5572852008-04-14 00:05:28 -07001666static inline void __skb_queue_before(struct sk_buff_head *list,
1667 struct sk_buff *next,
1668 struct sk_buff *newsk)
1669{
1670 __skb_insert(newsk, next->prev, next, list);
1671}
1672
Stephen Hemminger300ce172005-10-30 13:47:34 -08001673/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001674 * __skb_queue_head - queue a buffer at the list head
1675 * @list: list to use
1676 * @newsk: buffer to queue
1677 *
1678 * Queue a buffer at the start of a list. This function takes no locks
1679 * and you must therefore hold required locks before calling it.
1680 *
1681 * A buffer cannot be placed on two lists at the same time.
1682 */
Joe Perches7965bd42013-09-26 14:48:15 -07001683void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001684static inline void __skb_queue_head(struct sk_buff_head *list,
1685 struct sk_buff *newsk)
1686{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001687 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001688}
1689
1690/**
1691 * __skb_queue_tail - queue a buffer at the list tail
1692 * @list: list to use
1693 * @newsk: buffer to queue
1694 *
1695 * Queue a buffer at the end of a list. This function takes no locks
1696 * and you must therefore hold required locks before calling it.
1697 *
1698 * A buffer cannot be placed on two lists at the same time.
1699 */
Joe Perches7965bd42013-09-26 14:48:15 -07001700void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001701static inline void __skb_queue_tail(struct sk_buff_head *list,
1702 struct sk_buff *newsk)
1703{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001704 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001705}
1706
Linus Torvalds1da177e2005-04-16 15:20:36 -07001707/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001708 * remove sk_buff from list. _Must_ be called atomically, and with
1709 * the list known..
1710 */
Joe Perches7965bd42013-09-26 14:48:15 -07001711void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001712static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1713{
1714 struct sk_buff *next, *prev;
1715
1716 list->qlen--;
1717 next = skb->next;
1718 prev = skb->prev;
1719 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001720 next->prev = prev;
1721 prev->next = next;
1722}
1723
Gerrit Renkerf525c062008-04-14 00:04:12 -07001724/**
1725 * __skb_dequeue - remove from the head of the queue
1726 * @list: list to dequeue from
1727 *
1728 * Remove the head of the list. This function does not take any locks
1729 * so must be used with appropriate locks held only. The head item is
1730 * returned or %NULL if the list is empty.
1731 */
Joe Perches7965bd42013-09-26 14:48:15 -07001732struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001733static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1734{
1735 struct sk_buff *skb = skb_peek(list);
1736 if (skb)
1737 __skb_unlink(skb, list);
1738 return skb;
1739}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001740
1741/**
1742 * __skb_dequeue_tail - remove from the tail of the queue
1743 * @list: list to dequeue from
1744 *
1745 * Remove the tail of the list. This function does not take any locks
1746 * so must be used with appropriate locks held only. The tail item is
1747 * returned or %NULL if the list is empty.
1748 */
Joe Perches7965bd42013-09-26 14:48:15 -07001749struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001750static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1751{
1752 struct sk_buff *skb = skb_peek_tail(list);
1753 if (skb)
1754 __skb_unlink(skb, list);
1755 return skb;
1756}
1757
1758
David S. Millerbdcc0922012-03-07 20:53:36 -05001759static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001760{
1761 return skb->data_len;
1762}
1763
1764static inline unsigned int skb_headlen(const struct sk_buff *skb)
1765{
1766 return skb->len - skb->data_len;
1767}
1768
1769static inline int skb_pagelen(const struct sk_buff *skb)
1770{
1771 int i, len = 0;
1772
1773 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001774 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001775 return len + skb_headlen(skb);
1776}
1777
Ian Campbell131ea662011-08-19 06:25:00 +00001778/**
1779 * __skb_fill_page_desc - initialise a paged fragment in an skb
1780 * @skb: buffer containing fragment to be initialised
1781 * @i: paged fragment index to initialise
1782 * @page: the page to use for this fragment
1783 * @off: the offset to the data with @page
1784 * @size: the length of the data
1785 *
1786 * Initialises the @i'th fragment of @skb to point to &size bytes at
1787 * offset @off within @page.
1788 *
1789 * Does not take any additional reference on the fragment.
1790 */
1791static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1792 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001793{
1794 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1795
Mel Gormanc48a11c2012-07-31 16:44:23 -07001796 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001797 * Propagate page pfmemalloc to the skb if we can. The problem is
1798 * that not all callers have unique ownership of the page but rely
1799 * on page_is_pfmemalloc doing the right thing(tm).
Mel Gormanc48a11c2012-07-31 16:44:23 -07001800 */
Ian Campbella8605c62011-10-19 23:01:49 +00001801 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001803 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001804
1805 page = compound_head(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001806 if (page_is_pfmemalloc(page))
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001807 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001808}
1809
1810/**
1811 * skb_fill_page_desc - initialise a paged fragment in an skb
1812 * @skb: buffer containing fragment to be initialised
1813 * @i: paged fragment index to initialise
1814 * @page: the page to use for this fragment
1815 * @off: the offset to the data with @page
1816 * @size: the length of the data
1817 *
1818 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001819 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001820 * addition updates @skb such that @i is the last fragment.
1821 *
1822 * Does not take any additional reference on the fragment.
1823 */
1824static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1825 struct page *page, int off, int size)
1826{
1827 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001828 skb_shinfo(skb)->nr_frags = i + 1;
1829}
1830
Joe Perches7965bd42013-09-26 14:48:15 -07001831void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1832 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001833
Jason Wangf8e617e2013-11-01 14:07:47 +08001834void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1835 unsigned int truesize);
1836
Linus Torvalds1da177e2005-04-16 15:20:36 -07001837#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001838#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001839#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1840
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001841#ifdef NET_SKBUFF_DATA_USES_OFFSET
1842static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1843{
1844 return skb->head + skb->tail;
1845}
1846
1847static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1848{
1849 skb->tail = skb->data - skb->head;
1850}
1851
1852static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1853{
1854 skb_reset_tail_pointer(skb);
1855 skb->tail += offset;
1856}
Simon Horman7cc46192013-05-28 20:34:29 +00001857
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001858#else /* NET_SKBUFF_DATA_USES_OFFSET */
1859static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1860{
1861 return skb->tail;
1862}
1863
1864static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1865{
1866 skb->tail = skb->data;
1867}
1868
1869static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1870{
1871 skb->tail = skb->data + offset;
1872}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001873
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001874#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1875
Linus Torvalds1da177e2005-04-16 15:20:36 -07001876/*
1877 * Add data to an sk_buff
1878 */
Mathias Krause0c7ddf32013-11-07 14:18:24 +01001879unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001880unsigned char *skb_put(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len)
1882{
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001883 unsigned char *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001884 SKB_LINEAR_ASSERT(skb);
1885 skb->tail += len;
1886 skb->len += len;
1887 return tmp;
1888}
1889
Joe Perches7965bd42013-09-26 14:48:15 -07001890unsigned char *skb_push(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001891static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len)
1892{
1893 skb->data -= len;
1894 skb->len += len;
1895 return skb->data;
1896}
1897
Joe Perches7965bd42013-09-26 14:48:15 -07001898unsigned char *skb_pull(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len)
1900{
1901 skb->len -= len;
1902 BUG_ON(skb->len < skb->data_len);
1903 return skb->data += len;
1904}
1905
David S. Miller47d29642010-05-02 02:21:44 -07001906static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len)
1907{
1908 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
1909}
1910
Joe Perches7965bd42013-09-26 14:48:15 -07001911unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001912
1913static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len)
1914{
1915 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07001916 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001917 return NULL;
1918 skb->len -= len;
1919 return skb->data += len;
1920}
1921
1922static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len)
1923{
1924 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
1925}
1926
1927static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
1928{
1929 if (likely(len <= skb_headlen(skb)))
1930 return 1;
1931 if (unlikely(len > skb->len))
1932 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07001933 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001934}
1935
1936/**
1937 * skb_headroom - bytes at buffer head
1938 * @skb: buffer to check
1939 *
1940 * Return the number of bytes of free space at the head of an &sk_buff.
1941 */
Chuck Leverc2636b42007-10-23 21:07:32 -07001942static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001943{
1944 return skb->data - skb->head;
1945}
1946
1947/**
1948 * skb_tailroom - bytes at buffer end
1949 * @skb: buffer to check
1950 *
1951 * Return the number of bytes of free space at the tail of an sk_buff
1952 */
1953static inline int skb_tailroom(const struct sk_buff *skb)
1954{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001955 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001956}
1957
1958/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00001959 * skb_availroom - bytes at buffer end
1960 * @skb: buffer to check
1961 *
1962 * Return the number of bytes of free space at the tail of an sk_buff
1963 * allocated by sk_stream_alloc()
1964 */
1965static inline int skb_availroom(const struct sk_buff *skb)
1966{
Eric Dumazet16fad692013-03-14 05:40:32 +00001967 if (skb_is_nonlinear(skb))
1968 return 0;
1969
1970 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00001971}
1972
1973/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001974 * skb_reserve - adjust headroom
1975 * @skb: buffer to alter
1976 * @len: bytes to move
1977 *
1978 * Increase the headroom of an empty &sk_buff by reducing the tail
1979 * room. This is only allowed for an empty buffer.
1980 */
David S. Miller8243126c2006-01-17 02:54:21 -08001981static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001982{
1983 skb->data += len;
1984 skb->tail += len;
1985}
1986
Tom Herbert8bce6d72014-09-29 20:22:29 -07001987#define ENCAP_TYPE_ETHER 0
1988#define ENCAP_TYPE_IPPROTO 1
1989
1990static inline void skb_set_inner_protocol(struct sk_buff *skb,
1991 __be16 protocol)
1992{
1993 skb->inner_protocol = protocol;
1994 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
1995}
1996
1997static inline void skb_set_inner_ipproto(struct sk_buff *skb,
1998 __u8 ipproto)
1999{
2000 skb->inner_ipproto = ipproto;
2001 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2002}
2003
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002004static inline void skb_reset_inner_headers(struct sk_buff *skb)
2005{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002006 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002007 skb->inner_network_header = skb->network_header;
2008 skb->inner_transport_header = skb->transport_header;
2009}
2010
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00002011static inline void skb_reset_mac_len(struct sk_buff *skb)
2012{
2013 skb->mac_len = skb->network_header - skb->mac_header;
2014}
2015
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002016static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2017 *skb)
2018{
2019 return skb->head + skb->inner_transport_header;
2020}
2021
Tom Herbert55dc5a92015-12-14 11:19:40 -08002022static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2023{
2024 return skb_inner_transport_header(skb) - skb->data;
2025}
2026
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002027static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2028{
2029 skb->inner_transport_header = skb->data - skb->head;
2030}
2031
2032static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2033 const int offset)
2034{
2035 skb_reset_inner_transport_header(skb);
2036 skb->inner_transport_header += offset;
2037}
2038
2039static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2040{
2041 return skb->head + skb->inner_network_header;
2042}
2043
2044static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2045{
2046 skb->inner_network_header = skb->data - skb->head;
2047}
2048
2049static inline void skb_set_inner_network_header(struct sk_buff *skb,
2050 const int offset)
2051{
2052 skb_reset_inner_network_header(skb);
2053 skb->inner_network_header += offset;
2054}
2055
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002056static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2057{
2058 return skb->head + skb->inner_mac_header;
2059}
2060
2061static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2062{
2063 skb->inner_mac_header = skb->data - skb->head;
2064}
2065
2066static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2067 const int offset)
2068{
2069 skb_reset_inner_mac_header(skb);
2070 skb->inner_mac_header += offset;
2071}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002072static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2073{
Cong Wang35d04612013-05-29 15:16:05 +08002074 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002075}
2076
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002077static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2078{
2079 return skb->head + skb->transport_header;
2080}
2081
2082static inline void skb_reset_transport_header(struct sk_buff *skb)
2083{
2084 skb->transport_header = skb->data - skb->head;
2085}
2086
2087static inline void skb_set_transport_header(struct sk_buff *skb,
2088 const int offset)
2089{
2090 skb_reset_transport_header(skb);
2091 skb->transport_header += offset;
2092}
2093
2094static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2095{
2096 return skb->head + skb->network_header;
2097}
2098
2099static inline void skb_reset_network_header(struct sk_buff *skb)
2100{
2101 skb->network_header = skb->data - skb->head;
2102}
2103
2104static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2105{
2106 skb_reset_network_header(skb);
2107 skb->network_header += offset;
2108}
2109
2110static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2111{
2112 return skb->head + skb->mac_header;
2113}
2114
2115static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2116{
Cong Wang35d04612013-05-29 15:16:05 +08002117 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002118}
2119
2120static inline void skb_reset_mac_header(struct sk_buff *skb)
2121{
2122 skb->mac_header = skb->data - skb->head;
2123}
2124
2125static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2126{
2127 skb_reset_mac_header(skb);
2128 skb->mac_header += offset;
2129}
2130
Timo Teräs0e3da5b2013-12-16 11:02:09 +02002131static inline void skb_pop_mac_header(struct sk_buff *skb)
2132{
2133 skb->mac_header = skb->network_header;
2134}
2135
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002136static inline void skb_probe_transport_header(struct sk_buff *skb,
2137 const int offset_hint)
2138{
2139 struct flow_keys keys;
2140
2141 if (skb_transport_header_was_set(skb))
2142 return;
Tom Herbertcd79a232015-09-01 09:24:27 -07002143 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
Tom Herbert42aecaa2015-06-04 09:16:39 -07002144 skb_set_transport_header(skb, keys.control.thoff);
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002145 else
2146 skb_set_transport_header(skb, offset_hint);
2147}
2148
Eric Dumazet03606892012-02-23 10:55:02 +00002149static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2150{
2151 if (skb_mac_header_was_set(skb)) {
2152 const unsigned char *old_mac = skb_mac_header(skb);
2153
2154 skb_set_mac_header(skb, -skb->mac_len);
2155 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2156 }
2157}
2158
Michał Mirosław04fb4512010-12-14 15:24:08 +00002159static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2160{
2161 return skb->csum_start - skb_headroom(skb);
2162}
2163
Alexander Duyck08b64fc2016-02-05 15:27:49 -08002164static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2165{
2166 return skb->head + skb->csum_start;
2167}
2168
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002169static inline int skb_transport_offset(const struct sk_buff *skb)
2170{
2171 return skb_transport_header(skb) - skb->data;
2172}
2173
2174static inline u32 skb_network_header_len(const struct sk_buff *skb)
2175{
2176 return skb->transport_header - skb->network_header;
2177}
2178
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002179static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2180{
2181 return skb->inner_transport_header - skb->inner_network_header;
2182}
2183
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002184static inline int skb_network_offset(const struct sk_buff *skb)
2185{
2186 return skb_network_header(skb) - skb->data;
2187}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03002188
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002189static inline int skb_inner_network_offset(const struct sk_buff *skb)
2190{
2191 return skb_inner_network_header(skb) - skb->data;
2192}
2193
Changli Gaof9599ce2010-08-04 04:43:44 +00002194static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2195{
2196 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2197}
2198
Linus Torvalds1da177e2005-04-16 15:20:36 -07002199/*
2200 * CPUs often take a performance hit when accessing unaligned memory
2201 * locations. The actual performance hit varies, it can be small if the
2202 * hardware handles it or large if we have to take an exception and fix it
2203 * in software.
2204 *
2205 * Since an ethernet header is 14 bytes network drivers often end up with
2206 * the IP header at an unaligned offset. The IP header can be aligned by
2207 * shifting the start of the packet by 2 bytes. Drivers should do this
2208 * with:
2209 *
Tobias Klauser8660c122009-07-13 22:48:16 +00002210 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002211 *
2212 * The downside to this alignment of the IP header is that the DMA is now
2213 * unaligned. On some architectures the cost of an unaligned DMA is high
2214 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00002215 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002216 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2217 * to be overridden.
2218 */
2219#ifndef NET_IP_ALIGN
2220#define NET_IP_ALIGN 2
2221#endif
2222
Anton Blanchard025be812006-03-31 02:27:06 -08002223/*
2224 * The networking layer reserves some headroom in skb data (via
2225 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2226 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08002227 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08002228 *
2229 * Unfortunately this headroom changes the DMA alignment of the resulting
2230 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2231 * on some architectures. An architecture can override this value,
2232 * perhaps setting it to a cacheline in size (since that will maintain
2233 * cacheline alignment of the DMA). It must be a power of 2.
2234 *
David S. Millerd6301d32009-02-08 19:24:13 -08002235 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08002236 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07002237 *
2238 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2239 * to reduce average number of cache lines per packet.
2240 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07002241 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08002242 */
2243#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07002244#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08002245#endif
2246
Joe Perches7965bd42013-09-26 14:48:15 -07002247int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002248
2249static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2250{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00002251 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07002252 WARN_ON(1);
2253 return;
2254 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002255 skb->len = len;
2256 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002257}
2258
Joe Perches7965bd42013-09-26 14:48:15 -07002259void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002260
2261static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2262{
Herbert Xu3cc0e872006-06-09 16:13:38 -07002263 if (skb->data_len)
2264 return ___pskb_trim(skb, len);
2265 __skb_trim(skb, len);
2266 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002267}
2268
2269static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2270{
2271 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2272}
2273
2274/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07002275 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2276 * @skb: buffer to alter
2277 * @len: new length
2278 *
2279 * This is identical to pskb_trim except that the caller knows that
2280 * the skb is not cloned so we should never get an error due to out-
2281 * of-memory.
2282 */
2283static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2284{
2285 int err = pskb_trim(skb, len);
2286 BUG_ON(err);
2287}
2288
2289/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002290 * skb_orphan - orphan a buffer
2291 * @skb: buffer to orphan
2292 *
2293 * If a buffer currently has an owner then we call the owner's
2294 * destructor function and make the @skb unowned. The buffer continues
2295 * to exist but is no longer charged to its former owner.
2296 */
2297static inline void skb_orphan(struct sk_buff *skb)
2298{
Eric Dumazetc34a7612013-07-30 16:11:15 -07002299 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002300 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002301 skb->destructor = NULL;
2302 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07002303 } else {
2304 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002305 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002306}
2307
2308/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002309 * skb_orphan_frags - orphan the frags contained in a buffer
2310 * @skb: buffer to orphan frags from
2311 * @gfp_mask: allocation mask for replacement pages
2312 *
2313 * For each frag in the SKB which needs a destructor (i.e. has an
2314 * owner) create a copy of that frag and release the original
2315 * page by calling the destructor.
2316 */
2317static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2318{
2319 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY)))
2320 return 0;
2321 return skb_copy_ubufs(skb, gfp_mask);
2322}
2323
2324/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002325 * __skb_queue_purge - empty a list
2326 * @list: list to empty
2327 *
2328 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2329 * the list and one reference dropped. This function does not take the
2330 * list lock and the caller must hold the relevant locks to use it.
2331 */
Joe Perches7965bd42013-09-26 14:48:15 -07002332void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002333static inline void __skb_queue_purge(struct sk_buff_head *list)
2334{
2335 struct sk_buff *skb;
2336 while ((skb = __skb_dequeue(list)) != NULL)
2337 kfree_skb(skb);
2338}
2339
Joe Perches7965bd42013-09-26 14:48:15 -07002340void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002341
Joe Perches7965bd42013-09-26 14:48:15 -07002342struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2343 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002344
2345/**
2346 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2347 * @dev: network device to receive on
2348 * @length: length to allocate
2349 *
2350 * Allocate a new &sk_buff and assign it a usage count of one. The
2351 * buffer has unspecified headroom built in. Users should allocate
2352 * the headroom they think they need without accounting for the
2353 * built in space. The built in space is used for optimisations.
2354 *
2355 * %NULL is returned if there is no free memory. Although this function
2356 * allocates memory it can be called from an interrupt.
2357 */
2358static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002359 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002360{
2361 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2362}
2363
Eric Dumazet6f532612012-05-18 05:12:12 +00002364/* legacy helper around __netdev_alloc_skb() */
2365static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2366 gfp_t gfp_mask)
2367{
2368 return __netdev_alloc_skb(NULL, length, gfp_mask);
2369}
2370
2371/* legacy helper around netdev_alloc_skb() */
2372static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2373{
2374 return netdev_alloc_skb(NULL, length);
2375}
2376
2377
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002378static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2379 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002380{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002381 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002382
2383 if (NET_IP_ALIGN && skb)
2384 skb_reserve(skb, NET_IP_ALIGN);
2385 return skb;
2386}
2387
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002388static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2389 unsigned int length)
2390{
2391 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2392}
2393
Alexander Duyck181edb22015-05-06 21:12:03 -07002394static inline void skb_free_frag(void *addr)
2395{
2396 __free_page_frag(addr);
2397}
2398
Alexander Duyckffde7322014-12-09 19:40:42 -08002399void *napi_alloc_frag(unsigned int fragsz);
Alexander Duyckfd11a832014-12-09 19:40:49 -08002400struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2401 unsigned int length, gfp_t gfp_mask);
2402static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2403 unsigned int length)
2404{
2405 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2406}
Alexander Duyckffde7322014-12-09 19:40:42 -08002407
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002408/**
Alexander Duyck71dfda52014-11-11 09:26:34 -08002409 * __dev_alloc_pages - allocate page for network Rx
2410 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2411 * @order: size of the allocation
2412 *
2413 * Allocate a new page.
2414 *
2415 * %NULL is returned if there is no free memory.
2416*/
2417static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2418 unsigned int order)
2419{
2420 /* This piece of code contains several assumptions.
2421 * 1. This is for device Rx, therefor a cold page is preferred.
2422 * 2. The expectation is the user wants a compound page.
2423 * 3. If requesting a order 0 page it will not be compound
2424 * due to the check to see if order has a value in prep_new_page
2425 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2426 * code in gfp_to_alloc_flags that should be enforcing this.
2427 */
2428 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2429
2430 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2431}
2432
2433static inline struct page *dev_alloc_pages(unsigned int order)
2434{
2435 return __dev_alloc_pages(GFP_ATOMIC, order);
2436}
2437
2438/**
2439 * __dev_alloc_page - allocate a page for network Rx
2440 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2441 *
2442 * Allocate a new page.
2443 *
2444 * %NULL is returned if there is no free memory.
2445 */
2446static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2447{
2448 return __dev_alloc_pages(gfp_mask, 0);
2449}
2450
2451static inline struct page *dev_alloc_page(void)
2452{
2453 return __dev_alloc_page(GFP_ATOMIC);
2454}
2455
2456/**
Mel Gorman06140022012-07-31 16:44:24 -07002457 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2458 * @page: The page that was allocated from skb_alloc_page
2459 * @skb: The skb that may need pfmemalloc set
2460 */
2461static inline void skb_propagate_pfmemalloc(struct page *page,
2462 struct sk_buff *skb)
2463{
Michal Hocko2f064f32015-08-21 14:11:51 -07002464 if (page_is_pfmemalloc(page))
Mel Gorman06140022012-07-31 16:44:24 -07002465 skb->pfmemalloc = true;
2466}
2467
Eric Dumazet564824b2010-10-11 19:05:25 +00002468/**
Masanari Iidae2278672014-02-18 22:54:36 +09002469 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002470 * @frag: the paged fragment
2471 *
2472 * Returns the &struct page associated with @frag.
2473 */
2474static inline struct page *skb_frag_page(const skb_frag_t *frag)
2475{
Ian Campbella8605c62011-10-19 23:01:49 +00002476 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002477}
2478
2479/**
2480 * __skb_frag_ref - take an addition reference on a paged fragment.
2481 * @frag: the paged fragment
2482 *
2483 * Takes an additional reference on the paged fragment @frag.
2484 */
2485static inline void __skb_frag_ref(skb_frag_t *frag)
2486{
2487 get_page(skb_frag_page(frag));
2488}
2489
2490/**
2491 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2492 * @skb: the buffer
2493 * @f: the fragment offset.
2494 *
2495 * Takes an additional reference on the @f'th paged fragment of @skb.
2496 */
2497static inline void skb_frag_ref(struct sk_buff *skb, int f)
2498{
2499 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2500}
2501
2502/**
2503 * __skb_frag_unref - release a reference on a paged fragment.
2504 * @frag: the paged fragment
2505 *
2506 * Releases a reference on the paged fragment @frag.
2507 */
2508static inline void __skb_frag_unref(skb_frag_t *frag)
2509{
2510 put_page(skb_frag_page(frag));
2511}
2512
2513/**
2514 * skb_frag_unref - release a reference on a paged fragment of an skb.
2515 * @skb: the buffer
2516 * @f: the fragment offset
2517 *
2518 * Releases a reference on the @f'th paged fragment of @skb.
2519 */
2520static inline void skb_frag_unref(struct sk_buff *skb, int f)
2521{
2522 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2523}
2524
2525/**
2526 * skb_frag_address - gets the address of the data contained in a paged fragment
2527 * @frag: the paged fragment buffer
2528 *
2529 * Returns the address of the data within @frag. The page must already
2530 * be mapped.
2531 */
2532static inline void *skb_frag_address(const skb_frag_t *frag)
2533{
2534 return page_address(skb_frag_page(frag)) + frag->page_offset;
2535}
2536
2537/**
2538 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2539 * @frag: the paged fragment buffer
2540 *
2541 * Returns the address of the data within @frag. Checks that the page
2542 * is mapped and returns %NULL otherwise.
2543 */
2544static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2545{
2546 void *ptr = page_address(skb_frag_page(frag));
2547 if (unlikely(!ptr))
2548 return NULL;
2549
2550 return ptr + frag->page_offset;
2551}
2552
2553/**
2554 * __skb_frag_set_page - sets the page contained in a paged fragment
2555 * @frag: the paged fragment
2556 * @page: the page to set
2557 *
2558 * Sets the fragment @frag to contain @page.
2559 */
2560static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2561{
Ian Campbella8605c62011-10-19 23:01:49 +00002562 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002563}
2564
2565/**
2566 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2567 * @skb: the buffer
2568 * @f: the fragment offset
2569 * @page: the page to set
2570 *
2571 * Sets the @f'th fragment of @skb to contain @page.
2572 */
2573static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2574 struct page *page)
2575{
2576 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2577}
2578
Eric Dumazet400dfd32013-10-17 16:27:07 -07002579bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2580
Ian Campbell131ea662011-08-19 06:25:00 +00002581/**
2582 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002583 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002584 * @frag: the paged fragment to map
2585 * @offset: the offset within the fragment (starting at the
2586 * fragment's own offset)
2587 * @size: the number of bytes to map
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002588 * @dir: the direction of the mapping (%PCI_DMA_*)
Ian Campbell131ea662011-08-19 06:25:00 +00002589 *
2590 * Maps the page associated with @frag to @device.
2591 */
2592static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2593 const skb_frag_t *frag,
2594 size_t offset, size_t size,
2595 enum dma_data_direction dir)
2596{
2597 return dma_map_page(dev, skb_frag_page(frag),
2598 frag->page_offset + offset, size, dir);
2599}
2600
Eric Dumazet117632e2011-12-03 21:39:53 +00002601static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2602 gfp_t gfp_mask)
2603{
2604 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2605}
2606
Octavian Purdilabad93e92014-06-12 01:36:26 +03002607
2608static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2609 gfp_t gfp_mask)
2610{
2611 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2612}
2613
2614
Ian Campbell131ea662011-08-19 06:25:00 +00002615/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002616 * skb_clone_writable - is the header of a clone writable
2617 * @skb: buffer to check
2618 * @len: length up to which to write
2619 *
2620 * Returns true if modifying the header part of the cloned buffer
2621 * does not requires the data to be copied.
2622 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002623static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002624{
2625 return !skb_header_cloned(skb) &&
2626 skb_headroom(skb) + len <= skb->hdr_len;
2627}
2628
Herbert Xud9cc2042007-09-16 16:21:16 -07002629static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2630 int cloned)
2631{
2632 int delta = 0;
2633
Herbert Xud9cc2042007-09-16 16:21:16 -07002634 if (headroom > skb_headroom(skb))
2635 delta = headroom - skb_headroom(skb);
2636
2637 if (delta || cloned)
2638 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2639 GFP_ATOMIC);
2640 return 0;
2641}
2642
Patrick McHardy334a8132007-06-25 04:35:20 -07002643/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002644 * skb_cow - copy header of skb when it is required
2645 * @skb: buffer to cow
2646 * @headroom: needed headroom
2647 *
2648 * If the skb passed lacks sufficient headroom or its data part
2649 * is shared, data is reallocated. If reallocation fails, an error
2650 * is returned and original skb is not changed.
2651 *
2652 * The result is skb with writable area skb->head...skb->tail
2653 * and at least @headroom of space at head.
2654 */
2655static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2656{
Herbert Xud9cc2042007-09-16 16:21:16 -07002657 return __skb_cow(skb, headroom, skb_cloned(skb));
2658}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002659
Herbert Xud9cc2042007-09-16 16:21:16 -07002660/**
2661 * skb_cow_head - skb_cow but only making the head writable
2662 * @skb: buffer to cow
2663 * @headroom: needed headroom
2664 *
2665 * This function is identical to skb_cow except that we replace the
2666 * skb_cloned check by skb_header_cloned. It should be used when
2667 * you only need to push on some header and do not need to modify
2668 * the data.
2669 */
2670static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2671{
2672 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002673}
2674
2675/**
2676 * skb_padto - pad an skbuff up to a minimal size
2677 * @skb: buffer to pad
2678 * @len: minimal length
2679 *
2680 * Pads up a buffer to ensure the trailing bytes exist and are
2681 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002682 * is untouched. Otherwise it is extended. Returns zero on
2683 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002684 */
Herbert Xu5b057c62006-06-23 02:06:41 -07002685static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002686{
2687 unsigned int size = skb->len;
2688 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002689 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002690 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002691}
2692
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002693/**
2694 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2695 * @skb: buffer to pad
2696 * @len: minimal length
2697 *
2698 * Pads up a buffer to ensure the trailing bytes exist and are
2699 * blanked. If the buffer already contains sufficient data it
2700 * is untouched. Otherwise it is extended. Returns zero on
2701 * success. The skb is freed on error.
2702 */
2703static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2704{
2705 unsigned int size = skb->len;
2706
2707 if (unlikely(size < len)) {
2708 len -= size;
2709 if (skb_pad(skb, len))
2710 return -ENOMEM;
2711 __skb_put(skb, len);
2712 }
2713 return 0;
2714}
2715
Linus Torvalds1da177e2005-04-16 15:20:36 -07002716static inline int skb_add_data(struct sk_buff *skb,
Al Viroaf2b0402014-11-27 21:44:24 -05002717 struct iov_iter *from, int copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002718{
2719 const int off = skb->len;
2720
2721 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viroaf2b0402014-11-27 21:44:24 -05002722 __wsum csum = 0;
2723 if (csum_and_copy_from_iter(skb_put(skb, copy), copy,
2724 &csum, from) == copy) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002725 skb->csum = csum_block_add(skb->csum, csum, off);
2726 return 0;
2727 }
Al Viroaf2b0402014-11-27 21:44:24 -05002728 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002729 return 0;
2730
2731 __skb_trim(skb, off);
2732 return -EFAULT;
2733}
2734
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002735static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2736 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002737{
2738 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002739 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002740
Ian Campbellea2ab692011-08-22 23:44:58 +00002741 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00002742 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002743 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002744 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002745}
2746
Herbert Xu364c6ba2006-06-09 16:10:40 -07002747static inline int __skb_linearize(struct sk_buff *skb)
2748{
2749 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
2750}
2751
Linus Torvalds1da177e2005-04-16 15:20:36 -07002752/**
2753 * skb_linearize - convert paged skb to linear one
2754 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07002755 *
2756 * If there is no free memory -ENOMEM is returned, otherwise zero
2757 * is returned and the old skb data released.
2758 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07002759static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002760{
Herbert Xu364c6ba2006-06-09 16:10:40 -07002761 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
2762}
2763
2764/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00002765 * skb_has_shared_frag - can any frag be overwritten
2766 * @skb: buffer to test
2767 *
2768 * Return true if the skb has at least one frag that might be modified
2769 * by an external entity (as in vmsplice()/sendfile())
2770 */
2771static inline bool skb_has_shared_frag(const struct sk_buff *skb)
2772{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00002773 return skb_is_nonlinear(skb) &&
2774 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00002775}
2776
2777/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07002778 * skb_linearize_cow - make sure skb is linear and writable
2779 * @skb: buffer to process
2780 *
2781 * If there is no free memory -ENOMEM is returned, otherwise zero
2782 * is returned and the old skb data released.
2783 */
2784static inline int skb_linearize_cow(struct sk_buff *skb)
2785{
2786 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
2787 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002788}
2789
2790/**
2791 * skb_postpull_rcsum - update checksum for received skb after pull
2792 * @skb: buffer to update
2793 * @start: start of data before pull
2794 * @len: length of data pulled
2795 *
2796 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07002797 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
2798 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002799 */
2800
2801static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f2006-03-20 22:43:56 -08002802 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002803{
Patrick McHardy84fa7932006-08-29 16:44:56 -07002804 if (skb->ip_summed == CHECKSUM_COMPLETE)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002805 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0));
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002806 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
Pravin B Shelar31b33df2015-09-28 17:24:25 -07002807 skb_checksum_start_offset(skb) < 0)
Pravin B Shelar6ae459b2015-09-22 12:57:53 -07002808 skb->ip_summed = CHECKSUM_NONE;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002809}
2810
Herbert Xucbb042f2006-03-20 22:43:56 -08002811unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
2812
Daniel Borkmannf8ffad62016-01-07 15:50:23 +01002813static inline void skb_postpush_rcsum(struct sk_buff *skb,
2814 const void *start, unsigned int len)
2815{
2816 /* For performing the reverse operation to skb_postpull_rcsum(),
2817 * we can instead of ...
2818 *
2819 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0));
2820 *
2821 * ... just use this equivalent version here to save a few
2822 * instructions. Feeding csum of 0 in csum_partial() and later
2823 * on adding skb->csum is equivalent to feed skb->csum in the
2824 * first place.
2825 */
2826 if (skb->ip_summed == CHECKSUM_COMPLETE)
2827 skb->csum = csum_partial(start, len, skb->csum);
2828}
2829
David S. Miller7ce5a272013-12-02 17:26:05 -05002830/**
2831 * pskb_trim_rcsum - trim received skb and update checksum
2832 * @skb: buffer to trim
2833 * @len: new length
2834 *
2835 * This is exactly the same as pskb_trim except that it ensures the
2836 * checksum of received packets are still valid after the operation.
2837 */
2838
2839static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
2840{
2841 if (likely(len >= skb->len))
2842 return 0;
2843 if (skb->ip_summed == CHECKSUM_COMPLETE)
2844 skb->ip_summed = CHECKSUM_NONE;
2845 return __pskb_trim(skb, len);
2846}
2847
Linus Torvalds1da177e2005-04-16 15:20:36 -07002848#define skb_queue_walk(queue, skb) \
2849 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002850 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07002851 skb = skb->next)
2852
James Chapman46f89142007-04-30 00:07:31 -07002853#define skb_queue_walk_safe(queue, skb, tmp) \
2854 for (skb = (queue)->next, tmp = skb->next; \
2855 skb != (struct sk_buff *)(queue); \
2856 skb = tmp, tmp = skb->next)
2857
David S. Miller1164f522008-09-23 00:49:44 -07002858#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002859 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07002860 skb = skb->next)
2861
2862#define skb_queue_walk_from_safe(queue, skb, tmp) \
2863 for (tmp = skb->next; \
2864 skb != (struct sk_buff *)(queue); \
2865 skb = tmp, tmp = skb->next)
2866
Stephen Hemminger300ce172005-10-30 13:47:34 -08002867#define skb_queue_reverse_walk(queue, skb) \
2868 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07002869 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08002870 skb = skb->prev)
2871
David S. Miller686a2952011-01-20 22:47:32 -08002872#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
2873 for (skb = (queue)->prev, tmp = skb->prev; \
2874 skb != (struct sk_buff *)(queue); \
2875 skb = tmp, tmp = skb->prev)
2876
2877#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
2878 for (tmp = skb->prev; \
2879 skb != (struct sk_buff *)(queue); \
2880 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002881
David S. Miller21dc3302010-08-23 00:13:46 -07002882static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07002883{
2884 return skb_shinfo(skb)->frag_list != NULL;
2885}
2886
2887static inline void skb_frag_list_init(struct sk_buff *skb)
2888{
2889 skb_shinfo(skb)->frag_list = NULL;
2890}
2891
David S. Milleree039872009-06-09 00:17:13 -07002892#define skb_walk_frags(skb, iter) \
2893 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
2894
Rainer Weikusatea3793e2015-12-06 21:11:34 +00002895
2896int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
2897 const struct sk_buff *skb);
2898struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
2899 int *peeked, int *off, int *err,
2900 struct sk_buff **last);
Joe Perches7965bd42013-09-26 14:48:15 -07002901struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
2902 int *peeked, int *off, int *err);
2903struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
2904 int *err);
2905unsigned int datagram_poll(struct file *file, struct socket *sock,
2906 struct poll_table_struct *wait);
Al Viroc0371da2014-11-24 10:42:55 -05002907int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
2908 struct iov_iter *to, int size);
David S. Miller51f3d022014-11-05 16:46:40 -05002909static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
2910 struct msghdr *msg, int size)
2911{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002912 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
David S. Miller51f3d022014-11-05 16:46:40 -05002913}
Al Viroe5a4b0b2014-11-24 18:17:55 -05002914int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
2915 struct msghdr *msg);
Al Viro3a654f92014-06-19 14:15:22 -04002916int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
2917 struct iov_iter *from, int len);
Al Viro3a654f92014-06-19 14:15:22 -04002918int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
Joe Perches7965bd42013-09-26 14:48:15 -07002919void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
2920void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb);
2921int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07002922int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
2923int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
2924__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
2925 int len, __wsum csum);
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002926ssize_t skb_socket_splice(struct sock *sk,
2927 struct pipe_inode_info *pipe,
2928 struct splice_pipe_desc *spd);
2929int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
Joe Perches7965bd42013-09-26 14:48:15 -07002930 struct pipe_inode_info *pipe, unsigned int len,
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02002931 unsigned int flags,
2932 ssize_t (*splice_cb)(struct sock *,
2933 struct pipe_inode_info *,
2934 struct splice_pipe_desc *));
Joe Perches7965bd42013-09-26 14:48:15 -07002935void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01002936unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00002937int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
2938 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07002939void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
2940int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
2941void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01002942unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -07002943struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Vlad Yasevich0d5501c2014-08-08 14:42:13 -04002944struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
Jiri Pirkoe2195122014-11-19 14:05:01 +01002945int skb_ensure_writable(struct sk_buff *skb, int write_len);
Jiri Pirko93515d52014-11-19 14:05:02 +01002946int skb_vlan_pop(struct sk_buff *skb);
2947int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03002948
Al Viro6ce8e9c2014-04-06 21:25:44 -04002949static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
2950{
Al Viro21226ab2014-11-28 15:48:29 -05002951 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro6ce8e9c2014-04-06 21:25:44 -04002952}
2953
Al Viro7eab8d92014-04-06 21:51:23 -04002954static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
2955{
Al Viroe5a4b0b2014-11-24 18:17:55 -05002956 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro7eab8d92014-04-06 21:51:23 -04002957}
2958
Daniel Borkmann2817a332013-10-30 11:50:51 +01002959struct skb_checksum_ops {
2960 __wsum (*update)(const void *mem, int len, __wsum wsum);
2961 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
2962};
2963
2964__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
2965 __wsum csum, const struct skb_checksum_ops *ops);
2966__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
2967 __wsum csum);
2968
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07002969static inline void * __must_check
2970__skb_header_pointer(const struct sk_buff *skb, int offset,
2971 int len, void *data, int hlen, void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002972{
Patrick McHardy55820ee2005-07-05 14:08:10 -07002973 if (hlen - offset >= len)
David S. Miller690e36e2014-08-23 12:13:41 -07002974 return data + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975
David S. Miller690e36e2014-08-23 12:13:41 -07002976 if (!skb ||
2977 skb_copy_bits(skb, offset, buffer, len) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002978 return NULL;
2979
2980 return buffer;
2981}
2982
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07002983static inline void * __must_check
2984skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
David S. Miller690e36e2014-08-23 12:13:41 -07002985{
2986 return __skb_header_pointer(skb, offset, len, skb->data,
2987 skb_headlen(skb), buffer);
2988}
2989
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01002990/**
2991 * skb_needs_linearize - check if we need to linearize a given skb
2992 * depending on the given device features.
2993 * @skb: socket buffer to check
2994 * @features: net device features
2995 *
2996 * Returns true if either:
2997 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
2998 * 2. skb is fragmented and the device does not support SG.
2999 */
3000static inline bool skb_needs_linearize(struct sk_buff *skb,
3001 netdev_features_t features)
3002{
3003 return skb_is_nonlinear(skb) &&
3004 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3005 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3006}
3007
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03003008static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3009 void *to,
3010 const unsigned int len)
3011{
3012 memcpy(to, skb->data, len);
3013}
3014
3015static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3016 const int offset, void *to,
3017 const unsigned int len)
3018{
3019 memcpy(to, skb->data + offset, len);
3020}
3021
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03003022static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3023 const void *from,
3024 const unsigned int len)
3025{
3026 memcpy(skb->data, from, len);
3027}
3028
3029static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3030 const int offset,
3031 const void *from,
3032 const unsigned int len)
3033{
3034 memcpy(skb->data + offset, from, len);
3035}
3036
Joe Perches7965bd42013-09-26 14:48:15 -07003037void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003038
Patrick Ohlyac45f602009-02-12 05:03:37 +00003039static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3040{
3041 return skb->tstamp;
3042}
3043
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003044/**
3045 * skb_get_timestamp - get timestamp from a skb
3046 * @skb: skb to get stamp from
3047 * @stamp: pointer to struct timeval to store stamp in
3048 *
3049 * Timestamps are stored in the skb as offsets to a base timestamp.
3050 * This function converts the offset back to a struct timeval and stores
3051 * it in stamp.
3052 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00003053static inline void skb_get_timestamp(const struct sk_buff *skb,
3054 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003055{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003056 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003057}
3058
Patrick Ohlyac45f602009-02-12 05:03:37 +00003059static inline void skb_get_timestampns(const struct sk_buff *skb,
3060 struct timespec *stamp)
3061{
3062 *stamp = ktime_to_timespec(skb->tstamp);
3063}
3064
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003065static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003066{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003067 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003068}
3069
Stephen Hemminger164891a2007-04-23 22:26:16 -07003070static inline ktime_t net_timedelta(ktime_t t)
3071{
3072 return ktime_sub(ktime_get_real(), t);
3073}
3074
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07003075static inline ktime_t net_invalid_timestamp(void)
3076{
3077 return ktime_set(0, 0);
3078}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003079
Alexander Duyck62bccb82014-09-04 13:31:35 -04003080struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3081
Richard Cochranc1f19b52010-07-17 08:49:36 +00003082#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3083
Joe Perches7965bd42013-09-26 14:48:15 -07003084void skb_clone_tx_timestamp(struct sk_buff *skb);
3085bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00003086
3087#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3088
3089static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3090{
3091}
3092
3093static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3094{
3095 return false;
3096}
3097
3098#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3099
3100/**
3101 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3102 *
Richard Cochranda92b192011-10-21 00:49:15 +00003103 * PHY drivers may accept clones of transmitted packets for
3104 * timestamping via their phy_driver.txtstamp method. These drivers
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003105 * must call this function to return the skb back to the stack with a
3106 * timestamp.
Richard Cochranda92b192011-10-21 00:49:15 +00003107 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00003108 * @skb: clone of the the original outgoing packet
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003109 * @hwtstamps: hardware time stamps
Richard Cochranc1f19b52010-07-17 08:49:36 +00003110 *
3111 */
3112void skb_complete_tx_timestamp(struct sk_buff *skb,
3113 struct skb_shared_hwtstamps *hwtstamps);
3114
Willem de Bruijne7fd2882014-08-04 22:11:48 -04003115void __skb_tstamp_tx(struct sk_buff *orig_skb,
3116 struct skb_shared_hwtstamps *hwtstamps,
3117 struct sock *sk, int tstype);
3118
Patrick Ohlyac45f602009-02-12 05:03:37 +00003119/**
3120 * skb_tstamp_tx - queue clone of skb with send time stamps
3121 * @orig_skb: the original outgoing packet
3122 * @hwtstamps: hardware time stamps, may be NULL if not available
3123 *
3124 * If the skb has a socket associated, then this function clones the
3125 * skb (thus sharing the actual data and optional structures), stores
3126 * the optional hardware time stamping information (if non NULL) or
3127 * generates a software time stamp (otherwise), then queues the clone
3128 * to the error queue of the socket. Errors are silently ignored.
3129 */
Joe Perches7965bd42013-09-26 14:48:15 -07003130void skb_tstamp_tx(struct sk_buff *orig_skb,
3131 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00003132
Richard Cochran4507a712010-07-17 08:48:28 +00003133static inline void sw_tx_timestamp(struct sk_buff *skb)
3134{
Oliver Hartkopp2244d072010-08-17 08:59:14 +00003135 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP &&
3136 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS))
Richard Cochran4507a712010-07-17 08:48:28 +00003137 skb_tstamp_tx(skb, NULL);
3138}
3139
3140/**
3141 * skb_tx_timestamp() - Driver hook for transmit timestamping
3142 *
3143 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00003144 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00003145 *
David S. Miller73409f32013-12-27 13:04:33 -05003146 * Specifically, one should make absolutely sure that this function is
3147 * called before TX completion of this packet can trigger. Otherwise
3148 * the packet could potentially already be freed.
3149 *
Richard Cochran4507a712010-07-17 08:48:28 +00003150 * @skb: A socket buffer.
3151 */
3152static inline void skb_tx_timestamp(struct sk_buff *skb)
3153{
Richard Cochranc1f19b52010-07-17 08:49:36 +00003154 skb_clone_tx_timestamp(skb);
Richard Cochran4507a712010-07-17 08:48:28 +00003155 sw_tx_timestamp(skb);
3156}
3157
Johannes Berg6e3e9392011-11-09 10:15:42 +01003158/**
3159 * skb_complete_wifi_ack - deliver skb with wifi status
3160 *
3161 * @skb: the original outgoing packet
3162 * @acked: ack status
3163 *
3164 */
3165void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3166
Joe Perches7965bd42013-09-26 14:48:15 -07003167__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3168__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003169
Herbert Xu60476372007-04-09 11:59:39 -07003170static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3171{
Tom Herbert6edec0e2015-02-10 16:30:28 -08003172 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3173 skb->csum_valid ||
3174 (skb->ip_summed == CHECKSUM_PARTIAL &&
3175 skb_checksum_start_offset(skb) >= 0));
Herbert Xu60476372007-04-09 11:59:39 -07003176}
3177
Herbert Xufb286bb2005-11-10 13:01:24 -08003178/**
3179 * skb_checksum_complete - Calculate checksum of an entire packet
3180 * @skb: packet to process
3181 *
3182 * This function calculates the checksum over the entire packet plus
3183 * the value of skb->csum. The latter can be used to supply the
3184 * checksum of a pseudo header as used by TCP/UDP. It returns the
3185 * checksum.
3186 *
3187 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3188 * this function can be used to verify that checksum on received
3189 * packets. In that case the function should return zero if the
3190 * checksum is correct. In particular, this function will return zero
3191 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3192 * hardware has already verified the correctness of the checksum.
3193 */
Al Viro4381ca32007-07-15 21:00:11 +01003194static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08003195{
Herbert Xu60476372007-04-09 11:59:39 -07003196 return skb_csum_unnecessary(skb) ?
3197 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003198}
3199
Tom Herbert77cffe22014-08-27 21:26:46 -07003200static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3201{
3202 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3203 if (skb->csum_level == 0)
3204 skb->ip_summed = CHECKSUM_NONE;
3205 else
3206 skb->csum_level--;
3207 }
3208}
3209
3210static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3211{
3212 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3213 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3214 skb->csum_level++;
3215 } else if (skb->ip_summed == CHECKSUM_NONE) {
3216 skb->ip_summed = CHECKSUM_UNNECESSARY;
3217 skb->csum_level = 0;
3218 }
3219}
3220
Tom Herbert5a212322014-08-31 15:12:41 -07003221static inline void __skb_mark_checksum_bad(struct sk_buff *skb)
3222{
3223 /* Mark current checksum as bad (typically called from GRO
3224 * path). In the case that ip_summed is CHECKSUM_NONE
3225 * this must be the first checksum encountered in the packet.
3226 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first
3227 * checksum after the last one validated. For UDP, a zero
3228 * checksum can not be marked as bad.
3229 */
3230
3231 if (skb->ip_summed == CHECKSUM_NONE ||
3232 skb->ip_summed == CHECKSUM_UNNECESSARY)
3233 skb->csum_bad = 1;
3234}
3235
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003236/* Check if we need to perform checksum complete validation.
3237 *
3238 * Returns true if checksum complete is needed, false otherwise
3239 * (either checksum is unnecessary or zero checksum is allowed).
3240 */
3241static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3242 bool zero_okay,
3243 __sum16 check)
3244{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003245 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3246 skb->csum_valid = 1;
Tom Herbert77cffe22014-08-27 21:26:46 -07003247 __skb_decr_checksum_unnecessary(skb);
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003248 return false;
3249 }
3250
3251 return true;
3252}
3253
3254/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3255 * in checksum_init.
3256 */
3257#define CHECKSUM_BREAK 76
3258
Tom Herbert4e18b9a2015-04-20 14:10:04 -07003259/* Unset checksum-complete
3260 *
3261 * Unset checksum complete can be done when packet is being modified
3262 * (uncompressed for instance) and checksum-complete value is
3263 * invalidated.
3264 */
3265static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3266{
3267 if (skb->ip_summed == CHECKSUM_COMPLETE)
3268 skb->ip_summed = CHECKSUM_NONE;
3269}
3270
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003271/* Validate (init) checksum based on checksum complete.
3272 *
3273 * Return values:
3274 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3275 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3276 * checksum is stored in skb->csum for use in __skb_checksum_complete
3277 * non-zero: value of invalid checksum
3278 *
3279 */
3280static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3281 bool complete,
3282 __wsum psum)
3283{
3284 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3285 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003286 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003287 return 0;
3288 }
Tom Herbert5a212322014-08-31 15:12:41 -07003289 } else if (skb->csum_bad) {
3290 /* ip_summed == CHECKSUM_NONE in this case */
Eric Dumazetc91d4602015-05-15 05:48:07 -07003291 return (__force __sum16)1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003292 }
3293
3294 skb->csum = psum;
3295
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003296 if (complete || skb->len <= CHECKSUM_BREAK) {
3297 __sum16 csum;
3298
3299 csum = __skb_checksum_complete(skb);
3300 skb->csum_valid = !csum;
3301 return csum;
3302 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003303
3304 return 0;
3305}
3306
3307static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3308{
3309 return 0;
3310}
3311
3312/* Perform checksum validate (init). Note that this is a macro since we only
3313 * want to calculate the pseudo header which is an input function if necessary.
3314 * First we try to validate without any computation (checksum unnecessary) and
3315 * then calculate based on checksum complete calling the function to compute
3316 * pseudo header.
3317 *
3318 * Return values:
3319 * 0: checksum is validated or try to in skb_checksum_complete
3320 * non-zero: value of invalid checksum
3321 */
3322#define __skb_checksum_validate(skb, proto, complete, \
3323 zero_okay, check, compute_pseudo) \
3324({ \
3325 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003326 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003327 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3328 __ret = __skb_checksum_validate_complete(skb, \
3329 complete, compute_pseudo(skb, proto)); \
3330 __ret; \
3331})
3332
3333#define skb_checksum_init(skb, proto, compute_pseudo) \
3334 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3335
3336#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3337 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3338
3339#define skb_checksum_validate(skb, proto, compute_pseudo) \
3340 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3341
3342#define skb_checksum_validate_zero_check(skb, proto, check, \
3343 compute_pseudo) \
Sabrina Dubroca096a4cf2015-02-06 18:54:19 +01003344 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003345
3346#define skb_checksum_simple_validate(skb) \
3347 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3348
Tom Herbertd96535a2014-08-31 15:12:42 -07003349static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3350{
3351 return (skb->ip_summed == CHECKSUM_NONE &&
3352 skb->csum_valid && !skb->csum_bad);
3353}
3354
3355static inline void __skb_checksum_convert(struct sk_buff *skb,
3356 __sum16 check, __wsum pseudo)
3357{
3358 skb->csum = ~pseudo;
3359 skb->ip_summed = CHECKSUM_COMPLETE;
3360}
3361
3362#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3363do { \
3364 if (__skb_checksum_convert_check(skb)) \
3365 __skb_checksum_convert(skb, check, \
3366 compute_pseudo(skb, proto)); \
3367} while (0)
3368
Tom Herbert15e23962015-02-10 16:30:31 -08003369static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3370 u16 start, u16 offset)
3371{
3372 skb->ip_summed = CHECKSUM_PARTIAL;
3373 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3374 skb->csum_offset = offset - start;
3375}
3376
Tom Herbertdcdc8992015-02-02 16:07:34 -08003377/* Update skbuf and packet to reflect the remote checksum offload operation.
3378 * When called, ptr indicates the starting point for skb->csum when
3379 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3380 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3381 */
3382static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
Tom Herbert15e23962015-02-10 16:30:31 -08003383 int start, int offset, bool nopartial)
Tom Herbertdcdc8992015-02-02 16:07:34 -08003384{
3385 __wsum delta;
3386
Tom Herbert15e23962015-02-10 16:30:31 -08003387 if (!nopartial) {
3388 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3389 return;
3390 }
3391
Tom Herbertdcdc8992015-02-02 16:07:34 -08003392 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3393 __skb_checksum_complete(skb);
3394 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3395 }
3396
3397 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3398
3399 /* Adjust skb->csum since we changed the packet */
3400 skb->csum = csum_add(skb->csum, delta);
3401}
3402
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003403#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07003404void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003405static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3406{
3407 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07003408 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003409}
3410static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3411{
3412 if (nfct)
3413 atomic_inc(&nfct->use);
3414}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003415#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003416#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003417static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3418{
3419 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use))
3420 kfree(nf_bridge);
3421}
3422static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3423{
3424 if (nf_bridge)
3425 atomic_inc(&nf_bridge->use);
3426}
3427#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08003428static inline void nf_reset(struct sk_buff *skb)
3429{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003430#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003431 nf_conntrack_put(skb->nfct);
3432 skb->nfct = NULL;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003433#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003434#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003435 nf_bridge_put(skb->nf_bridge);
3436 skb->nf_bridge = NULL;
3437#endif
3438}
3439
Patrick McHardy124dff02013-04-05 20:42:05 +02003440static inline void nf_reset_trace(struct sk_buff *skb)
3441{
Florian Westphal478b3602014-02-15 23:48:45 +01003442#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00003443 skb->nf_trace = 0;
3444#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003445}
3446
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003447/* Note: This doesn't put any conntrack and bridge info in dst. */
Eric Dumazetb1937222014-09-28 22:18:47 -07003448static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3449 bool copy)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003450{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003451#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003452 dst->nfct = src->nfct;
3453 nf_conntrack_get(src->nfct);
Eric Dumazetb1937222014-09-28 22:18:47 -07003454 if (copy)
3455 dst->nfctinfo = src->nfctinfo;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003456#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003457#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003458 dst->nf_bridge = src->nf_bridge;
3459 nf_bridge_get(src->nf_bridge);
3460#endif
Florian Westphal478b3602014-02-15 23:48:45 +01003461#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Eric Dumazetb1937222014-09-28 22:18:47 -07003462 if (copy)
3463 dst->nf_trace = src->nf_trace;
Florian Westphal478b3602014-02-15 23:48:45 +01003464#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003465}
3466
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003467static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3468{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003469#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003470 nf_conntrack_put(dst->nfct);
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003471#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003472#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003473 nf_bridge_put(dst->nf_bridge);
3474#endif
Eric Dumazetb1937222014-09-28 22:18:47 -07003475 __nf_copy(dst, src, true);
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003476}
3477
James Morris984bc162006-06-09 00:29:17 -07003478#ifdef CONFIG_NETWORK_SECMARK
3479static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3480{
3481 to->secmark = from->secmark;
3482}
3483
3484static inline void skb_init_secmark(struct sk_buff *skb)
3485{
3486 skb->secmark = 0;
3487}
3488#else
3489static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3490{ }
3491
3492static inline void skb_init_secmark(struct sk_buff *skb)
3493{ }
3494#endif
3495
Eric W. Biederman574f7192014-04-01 12:20:24 -07003496static inline bool skb_irq_freeable(const struct sk_buff *skb)
3497{
3498 return !skb->destructor &&
3499#if IS_ENABLED(CONFIG_XFRM)
3500 !skb->sp &&
3501#endif
3502#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3503 !skb->nfct &&
3504#endif
3505 !skb->_skb_refdst &&
3506 !skb_has_frag_list(skb);
3507}
3508
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003509static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3510{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003511 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003512}
3513
Stephen Hemminger92477442009-03-21 13:39:26 -07003514static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003515{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003516 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003517}
3518
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003519static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3520{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003521 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003522}
3523
David S. Millerd5a9e242009-01-27 16:22:11 -08003524static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3525{
3526 skb->queue_mapping = rx_queue + 1;
3527}
3528
Stephen Hemminger92477442009-03-21 13:39:26 -07003529static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003530{
3531 return skb->queue_mapping - 1;
3532}
3533
Stephen Hemminger92477442009-03-21 13:39:26 -07003534static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003535{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003536 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003537}
3538
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003539static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3540{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003541#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003542 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003543#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003544 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003545#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003546}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003547
Pravin B Shelar68c33162013-02-14 14:02:41 +00003548/* Keeps track of mac header offset relative to skb->head.
3549 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3550 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003551 * tunnel skb it points to outer mac header.
3552 * Keeps track of level of encapsulation of network headers.
3553 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003554struct skb_gso_cb {
Eric Dumazet3347c962013-10-19 11:42:56 -07003555 int mac_offset;
3556 int encap_level;
Alexander Duyck76443452016-02-05 15:27:37 -08003557 __wsum csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003558 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003559};
Konstantin Khlebnikov9207f9d2016-01-08 15:21:46 +03003560#define SKB_SGO_CB_OFFSET 32
3561#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
Pravin B Shelar68c33162013-02-14 14:02:41 +00003562
3563static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3564{
3565 return (skb_mac_header(inner_skb) - inner_skb->head) -
3566 SKB_GSO_CB(inner_skb)->mac_offset;
3567}
3568
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003569static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3570{
3571 int new_headroom, headroom;
3572 int ret;
3573
3574 headroom = skb_headroom(skb);
3575 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3576 if (ret)
3577 return ret;
3578
3579 new_headroom = skb_headroom(skb);
3580 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3581 return 0;
3582}
3583
Alexander Duyck08b64fc2016-02-05 15:27:49 -08003584static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3585{
3586 /* Do not update partial checksums if remote checksum is enabled. */
3587 if (skb->remcsum_offload)
3588 return;
3589
3590 SKB_GSO_CB(skb)->csum = res;
3591 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3592}
3593
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003594/* Compute the checksum for a gso segment. First compute the checksum value
3595 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3596 * then add in skb->csum (checksum from csum_start to end of packet).
3597 * skb->csum and csum_start are then updated to reflect the checksum of the
3598 * resultant packet starting from the transport header-- the resultant checksum
3599 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3600 * header.
3601 */
3602static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3603{
Alexander Duyck76443452016-02-05 15:27:37 -08003604 unsigned char *csum_start = skb_transport_header(skb);
3605 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3606 __wsum partial = SKB_GSO_CB(skb)->csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003607
Alexander Duyck76443452016-02-05 15:27:37 -08003608 SKB_GSO_CB(skb)->csum = res;
3609 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003610
Alexander Duyck76443452016-02-05 15:27:37 -08003611 return csum_fold(csum_partial(csum_start, plen, partial));
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003612}
3613
David S. Millerbdcc0922012-03-07 20:53:36 -05003614static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07003615{
3616 return skb_shinfo(skb)->gso_size;
3617}
3618
Eric Dumazet36a8f392013-09-29 01:21:32 -07003619/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05003620static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02003621{
3622 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
3623}
3624
Joe Perches7965bd42013-09-26 14:48:15 -07003625void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07003626
3627static inline bool skb_warn_if_lro(const struct sk_buff *skb)
3628{
3629 /* LRO sets gso_size but not gso_type, whereas if GSO is really
3630 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003631 const struct skb_shared_info *shinfo = skb_shinfo(skb);
3632
Alexander Duyckb78462e2010-06-02 12:24:37 +00003633 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
3634 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07003635 __skb_warn_lro_forwarding(skb);
3636 return true;
3637 }
3638 return false;
3639}
3640
Herbert Xu35fc92a2007-03-26 23:22:20 -07003641static inline void skb_forward_csum(struct sk_buff *skb)
3642{
3643 /* Unfortunately we don't support this one. Any brave souls? */
3644 if (skb->ip_summed == CHECKSUM_COMPLETE)
3645 skb->ip_summed = CHECKSUM_NONE;
3646}
3647
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003648/**
3649 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
3650 * @skb: skb to check
3651 *
3652 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
3653 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
3654 * use this helper, to document places where we make this assertion.
3655 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04003656static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07003657{
3658#ifdef DEBUG
3659 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
3660#endif
3661}
3662
Rusty Russellf35d9d82008-02-04 23:49:54 -05003663bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00003664
Paul Durranted1f50c2014-01-09 10:02:46 +00003665int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
Linus Lüssing9afd85c2015-05-02 14:01:07 +02003666struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
3667 unsigned int transport_len,
3668 __sum16(*skb_chkf)(struct sk_buff *skb));
Paul Durranted1f50c2014-01-09 10:02:46 +00003669
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00003670/**
3671 * skb_head_is_locked - Determine if the skb->head is locked down
3672 * @skb: skb to check
3673 *
3674 * The head on skbs build around a head frag can be removed if they are
3675 * not cloned. This function returns true if the skb head is locked down
3676 * due to either being allocated via kmalloc, or by being a clone with
3677 * multiple references to the head.
3678 */
3679static inline bool skb_head_is_locked(const struct sk_buff *skb)
3680{
3681 return !skb->head_frag || skb_cloned(skb);
3682}
Florian Westphalfe6cc552014-02-13 23:09:12 +01003683
3684/**
3685 * skb_gso_network_seglen - Return length of individual segments of a gso packet
3686 *
3687 * @skb: GSO skb
3688 *
3689 * skb_gso_network_seglen is used to determine the real size of the
3690 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
3691 *
3692 * The MAC/L2 header is not accounted for.
3693 */
3694static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
3695{
3696 unsigned int hdr_len = skb_transport_header(skb) -
3697 skb_network_header(skb);
3698 return hdr_len + skb_gso_transport_seglen(skb);
3699}
Thomas Grafee122c72015-07-21 10:43:58 +02003700
Linus Torvalds1da177e2005-04-16 15:20:36 -07003701#endif /* __KERNEL__ */
3702#endif /* _LINUX_SKBUFF_H */