blob: 01a985937867935dd615d355f4a662a9f8674b83 [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>
Eric Dumazetc1d1b432017-08-31 16:48:22 -070025#include <linux/refcount.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070026
Arun Sharma600634972011-07-26 16:09:06 -070027#include <linux/atomic.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070028#include <asm/types.h>
29#include <linux/spinlock.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070030#include <linux/net.h>
Thomas Graf3fc7e8a2005-06-23 21:00:17 -070031#include <linux/textsearch.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070032#include <net/checksum.h>
Al Viroa80958f2006-12-04 20:41:19 +000033#include <linux/rcupdate.h>
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -070034#include <linux/hrtimer.h>
Ian Campbell131ea662011-08-19 06:25:00 +000035#include <linux/dma-mapping.h>
Michał Mirosławc8f44af2011-11-15 15:29:55 +000036#include <linux/netdev_features.h>
Eric Dumazet363ec39232014-02-26 14:02:11 -080037#include <linux/sched.h>
Ingo Molnare6017572017-02-01 16:36:40 +010038#include <linux/sched/clock.h>
Jiri Pirko1bd758e2015-05-12 14:56:07 +020039#include <net/flow_dissector.h>
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +020040#include <linux/splice.h>
Bernhard Thaler72b31f72015-05-30 15:27:40 +020041#include <linux/in6.h>
Jamal Hadi Salim8b10cab2016-07-02 06:43:14 -040042#include <linux/if_packet.h>
Tom Herbertf70ea012015-07-31 16:52:10 -070043#include <net/flow.h>
Linus Torvalds1da177e2005-04-16 15:20:36 -070044
Tom Herbert7a6ae712015-12-14 11:19:47 -080045/* The interface for checksum offload between the stack and networking drivers
46 * is as follows...
47 *
48 * A. IP checksum related features
49 *
50 * Drivers advertise checksum offload capabilities in the features of a device.
51 * From the stack's point of view these are capabilities offered by the driver,
52 * a driver typically only advertises features that it is capable of offloading
53 * to its device.
54 *
55 * The checksum related features are:
56 *
57 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
58 * IP (one's complement) checksum for any combination
59 * of protocols or protocol layering. The checksum is
60 * computed and set in a packet per the CHECKSUM_PARTIAL
61 * interface (see below).
62 *
63 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
64 * TCP or UDP packets over IPv4. These are specifically
65 * unencapsulated packets of the form IPv4|TCP or
66 * IPv4|UDP where the Protocol field in the IPv4 header
67 * is TCP or UDP. The IPv4 header may contain IP options
68 * This feature cannot be set in features for a device
69 * with NETIF_F_HW_CSUM also set. This feature is being
70 * DEPRECATED (see below).
71 *
72 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
73 * TCP or UDP packets over IPv6. These are specifically
74 * unencapsulated packets of the form IPv6|TCP or
75 * IPv4|UDP where the Next Header field in the IPv6
76 * header is either TCP or UDP. IPv6 extension headers
77 * are not supported with this feature. This feature
78 * cannot be set in features for a device with
79 * NETIF_F_HW_CSUM also set. This feature is being
80 * DEPRECATED (see below).
81 *
82 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
83 * This flag is used only used to disable the RX checksum
84 * feature for a device. The stack will accept receive
85 * checksum indication in packets received on a device
86 * regardless of whether NETIF_F_RXCSUM is set.
87 *
88 * B. Checksumming of received packets by device. Indication of checksum
89 * verification is in set skb->ip_summed. Possible values are:
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010090 *
91 * CHECKSUM_NONE:
92 *
Tom Herbert7a6ae712015-12-14 11:19:47 -080093 * Device did not checksum this packet e.g. due to lack of capabilities.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +010094 * The packet contains full (though not verified) checksum in packet but
95 * not in skb->csum. Thus, skb->csum is undefined in this case.
96 *
97 * CHECKSUM_UNNECESSARY:
98 *
99 * The hardware you're dealing with doesn't calculate the full checksum
100 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
Tom Herbert77cffe22014-08-27 21:26:46 -0700101 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
102 * if their checksums are okay. skb->csum is still undefined in this case
Tom Herbert7a6ae712015-12-14 11:19:47 -0800103 * though. A driver or device must never modify the checksum field in the
104 * packet even if checksum is verified.
Tom Herbert77cffe22014-08-27 21:26:46 -0700105 *
106 * CHECKSUM_UNNECESSARY is applicable to following protocols:
107 * TCP: IPv6 and IPv4.
108 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
109 * zero UDP checksum for either IPv4 or IPv6, the networking stack
110 * may perform further validation in this case.
111 * GRE: only if the checksum is present in the header.
112 * SCTP: indicates the CRC in SCTP header has been validated.
Davide Carattib4759dc2017-05-18 15:44:43 +0200113 * FCOE: indicates the CRC in FC frame has been validated.
Tom Herbert77cffe22014-08-27 21:26:46 -0700114 *
115 * skb->csum_level indicates the number of consecutive checksums found in
116 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
117 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
118 * and a device is able to verify the checksums for UDP (possibly zero),
119 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
120 * two. If the device were only able to verify the UDP checksum and not
121 * GRE, either because it doesn't support GRE checksum of because GRE
122 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is
123 * not considered in this case).
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100124 *
125 * CHECKSUM_COMPLETE:
126 *
127 * This is the most generic way. The device supplied checksum of the _whole_
128 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
129 * hardware doesn't need to parse L3/L4 headers to implement this.
130 *
Davide Carattib4759dc2017-05-18 15:44:43 +0200131 * Notes:
132 * - Even if device supports only some protocols, but is able to produce
133 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
134 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100135 *
136 * CHECKSUM_PARTIAL:
137 *
Tom Herbert6edec0e2015-02-10 16:30:28 -0800138 * A checksum is set up to be offloaded to a device as described in the
139 * output description for CHECKSUM_PARTIAL. This may occur on a packet
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100140 * received directly from another Linux OS, e.g., a virtualized Linux kernel
Tom Herbert6edec0e2015-02-10 16:30:28 -0800141 * on the same host, or it may be set in the input path in GRO or remote
142 * checksum offload. For the purposes of checksum verification, the checksum
143 * referred to by skb->csum_start + skb->csum_offset and any preceding
144 * checksums in the packet are considered verified. Any checksums in the
145 * packet that are after the checksum being offloaded are not considered to
146 * be verified.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100147 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800148 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
149 * in the skb->ip_summed for a packet. Values are:
150 *
151 * CHECKSUM_PARTIAL:
152 *
153 * The driver is required to checksum the packet as seen by hard_start_xmit()
154 * from skb->csum_start up to the end, and to record/write the checksum at
155 * offset skb->csum_start + skb->csum_offset. A driver may verify that the
156 * csum_start and csum_offset values are valid values given the length and
157 * offset of the packet, however they should not attempt to validate that the
158 * checksum refers to a legitimate transport layer checksum-- it is the
159 * purview of the stack to validate that csum_start and csum_offset are set
160 * correctly.
161 *
162 * When the stack requests checksum offload for a packet, the driver MUST
163 * ensure that the checksum is set correctly. A driver can either offload the
164 * checksum calculation to the device, or call skb_checksum_help (in the case
165 * that the device does not support offload for a particular checksum).
166 *
167 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
168 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
Davide Caratti43c26a12017-05-18 15:44:41 +0200169 * checksum offload capability.
170 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
171 * on network device checksumming capabilities: if a packet does not match
172 * them, skb_checksum_help or skb_crc32c_help (depending on the value of
173 * csum_not_inet, see item D.) is called to resolve the checksum.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100174 *
175 * CHECKSUM_NONE:
176 *
177 * The skb was already checksummed by the protocol, or a checksum is not
178 * required.
179 *
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100180 * CHECKSUM_UNNECESSARY:
181 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800182 * This has the same meaning on as CHECKSUM_NONE for checksum offload on
183 * output.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100184 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800185 * CHECKSUM_COMPLETE:
186 * Not used in checksum output. If a driver observes a packet with this value
187 * set in skbuff, if should treat as CHECKSUM_NONE being set.
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100188 *
Tom Herbert7a6ae712015-12-14 11:19:47 -0800189 * D. Non-IP checksum (CRC) offloads
190 *
191 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
192 * offloading the SCTP CRC in a packet. To perform this offload the stack
Davide Carattidba00302017-05-18 15:44:40 +0200193 * will set set csum_start and csum_offset accordingly, set ip_summed to
194 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
195 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
196 * A driver that supports both IP checksum offload and SCTP CRC32c offload
197 * must verify which offload is configured for a packet by testing the
198 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
199 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
Tom Herbert7a6ae712015-12-14 11:19:47 -0800200 *
201 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
202 * offloading the FCOE CRC in a packet. To perform this offload the stack
203 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
204 * accordingly. Note the there is no indication in the skbuff that the
205 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
206 * both IP checksum offload and FCOE CRC offload must verify which offload
207 * is configured for a packet presumably by inspecting packet headers.
208 *
209 * E. Checksumming on output with GSO.
210 *
211 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
212 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
213 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
214 * part of the GSO operation is implied. If a checksum is being offloaded
215 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
216 * are set to refer to the outermost checksum being offload (two offloaded
217 * checksums are possible with UDP encapsulation).
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100218 */
219
Herbert Xu60476372007-04-09 11:59:39 -0700220/* Don't change this without changing skb_csum_unnecessary! */
Daniel Borkmann78ea85f2013-12-16 23:27:09 +0100221#define CHECKSUM_NONE 0
222#define CHECKSUM_UNNECESSARY 1
223#define CHECKSUM_COMPLETE 2
224#define CHECKSUM_PARTIAL 3
Linus Torvalds1da177e2005-04-16 15:20:36 -0700225
Tom Herbert77cffe22014-08-27 21:26:46 -0700226/* Maximum value in skb->csum_level */
227#define SKB_MAX_CSUM_LEVEL 3
228
Tobias Klauser0bec8c82014-07-22 12:06:23 +0200229#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES)
David S. Millerfc910a22007-03-25 20:27:59 -0700230#define SKB_WITH_OVERHEAD(X) \
Herbert Xudeea84b2007-10-21 16:27:46 -0700231 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
David S. Millerfc910a22007-03-25 20:27:59 -0700232#define SKB_MAX_ORDER(X, ORDER) \
233 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
Linus Torvalds1da177e2005-04-16 15:20:36 -0700234#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0))
235#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2))
236
Eric Dumazet87fb4b72011-10-13 07:28:54 +0000237/* return minimum truesize of one skb containing X bytes of data */
238#define SKB_TRUESIZE(X) ((X) + \
239 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \
240 SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
241
Linus Torvalds1da177e2005-04-16 15:20:36 -0700242struct net_device;
David Howells716ea3a2007-04-02 20:19:53 -0700243struct scatterlist;
Jens Axboe9c55e012007-11-06 23:30:13 -0800244struct pipe_inode_info;
Herbert Xua8f820aa2014-11-07 21:22:22 +0800245struct iov_iter;
Alexander Duyckfd11a832014-12-09 19:40:49 -0800246struct napi_struct;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700247
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700248#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700249struct nf_conntrack {
250 atomic_t use;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700251};
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -0700252#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700253
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200254#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700255struct nf_bridge_info {
Reshetova, Elena53869ce2017-06-30 13:07:57 +0300256 refcount_t use;
Florian Westphal3eaf4022015-04-02 14:31:44 +0200257 enum {
258 BRNF_PROTO_UNCHANGED,
259 BRNF_PROTO_8021Q,
260 BRNF_PROTO_PPPOE
Florian Westphal7fb48c52015-05-03 22:05:28 +0200261 } orig_proto:8;
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200262 u8 pkt_otherhost:1;
263 u8 in_prerouting:1;
264 u8 bridged_dnat:1;
Bernhard Thaler411ffb42015-05-30 15:28:28 +0200265 __u16 frag_max_size;
Eric Dumazetbf1ac5c2012-04-18 23:19:25 +0000266 struct net_device *physindev;
Florian Westphal63cdbc02015-09-14 17:06:27 +0200267
268 /* always valid & non-NULL from FORWARD on, for physdev match */
269 struct net_device *physoutdev;
Florian Westphal7fb48c52015-05-03 22:05:28 +0200270 union {
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200271 /* prerouting: detect dnat in orig/reply direction */
Bernhard Thaler72b31f72015-05-30 15:27:40 +0200272 __be32 ipv4_daddr;
273 struct in6_addr ipv6_daddr;
Florian Westphal72b1e5e2015-07-23 16:21:30 +0200274
275 /* after prerouting + nat detected: store original source
276 * mac since neigh resolution overwrites it, only used while
277 * skb is out in neigh layer.
278 */
279 char neigh_header[8];
Bernhard Thaler72b31f72015-05-30 15:27:40 +0200280 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700281};
282#endif
283
Linus Torvalds1da177e2005-04-16 15:20:36 -0700284struct sk_buff_head {
285 /* These two members must be first. */
286 struct sk_buff *next;
287 struct sk_buff *prev;
288
289 __u32 qlen;
290 spinlock_t lock;
291};
292
293struct sk_buff;
294
Ian Campbell9d4dde52011-12-22 23:39:14 +0000295/* To allow 64K frame to be packed as single skb without frag_list we
296 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
297 * buffers which do not start on a page boundary.
298 *
299 * Since GRO uses frags we allocate at least 16 regardless of page
300 * size.
Anton Blancharda715dea2011-03-27 14:57:26 +0000301 */
Ian Campbell9d4dde52011-12-22 23:39:14 +0000302#if (65536/PAGE_SIZE + 1) < 16
David S. Millereec00952011-03-29 23:34:08 -0700303#define MAX_SKB_FRAGS 16UL
Anton Blancharda715dea2011-03-27 14:57:26 +0000304#else
Ian Campbell9d4dde52011-12-22 23:39:14 +0000305#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
Anton Blancharda715dea2011-03-27 14:57:26 +0000306#endif
Hans Westgaard Ry5f74f822016-02-03 09:26:57 +0100307extern int sysctl_max_skb_frags;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700308
Marcelo Ricardo Leitner3953c462016-06-02 15:05:40 -0300309/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
310 * segment using its current segmentation instead.
311 */
312#define GSO_BY_FRAGS 0xFFFF
313
Linus Torvalds1da177e2005-04-16 15:20:36 -0700314typedef struct skb_frag_struct skb_frag_t;
315
316struct skb_frag_struct {
Ian Campbella8605c62011-10-19 23:01:49 +0000317 struct {
318 struct page *p;
319 } page;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000320#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
David S. Millera309bb02007-07-30 18:47:03 -0700321 __u32 page_offset;
322 __u32 size;
Eric Dumazetcb4dfe52010-09-23 05:06:54 +0000323#else
324 __u16 page_offset;
325 __u16 size;
326#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700327};
328
Eric Dumazet9e903e02011-10-18 21:00:24 +0000329static inline unsigned int skb_frag_size(const skb_frag_t *frag)
330{
331 return frag->size;
332}
333
334static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
335{
336 frag->size = size;
337}
338
339static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
340{
341 frag->size += delta;
342}
343
344static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
345{
346 frag->size -= delta;
347}
348
Willem de Bruijnc613c202017-07-31 08:15:47 -0400349static inline bool skb_frag_must_loop(struct page *p)
350{
351#if defined(CONFIG_HIGHMEM)
352 if (PageHighMem(p))
353 return true;
354#endif
355 return false;
356}
357
358/**
359 * skb_frag_foreach_page - loop over pages in a fragment
360 *
361 * @f: skb frag to operate on
362 * @f_off: offset from start of f->page.p
363 * @f_len: length from f_off to loop over
364 * @p: (temp var) current page
365 * @p_off: (temp var) offset from start of current page,
366 * non-zero only on first page.
367 * @p_len: (temp var) length in current page,
368 * < PAGE_SIZE only on first and last page.
369 * @copied: (temp var) length so far, excluding current p_len.
370 *
371 * A fragment can hold a compound page, in which case per-page
372 * operations, notably kmap_atomic, must be called for each
373 * regular page.
374 */
375#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
376 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \
377 p_off = (f_off) & (PAGE_SIZE - 1), \
378 p_len = skb_frag_must_loop(p) ? \
379 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \
380 copied = 0; \
381 copied < f_len; \
382 copied += p_len, p++, p_off = 0, \
383 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \
384
Patrick Ohlyac45f602009-02-12 05:03:37 +0000385#define HAVE_HW_TIME_STAMP
386
387/**
Randy Dunlapd3a21be2009-03-02 03:15:58 -0800388 * struct skb_shared_hwtstamps - hardware time stamps
Patrick Ohlyac45f602009-02-12 05:03:37 +0000389 * @hwtstamp: hardware time stamp transformed into duration
390 * since arbitrary point in time
Patrick Ohlyac45f602009-02-12 05:03:37 +0000391 *
392 * Software time stamps generated by ktime_get_real() are stored in
Willem de Bruijn4d276eb2014-07-25 18:01:32 -0400393 * skb->tstamp.
Patrick Ohlyac45f602009-02-12 05:03:37 +0000394 *
395 * hwtstamps can only be compared against other hwtstamps from
396 * the same device.
397 *
398 * This structure is attached to packets as part of the
399 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
400 */
401struct skb_shared_hwtstamps {
402 ktime_t hwtstamp;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000403};
404
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000405/* Definitions for tx_flags in struct skb_shared_info */
406enum {
407 /* generate hardware time stamp */
408 SKBTX_HW_TSTAMP = 1 << 0,
409
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400410 /* generate software time stamp when queueing packet to NIC */
Oliver Hartkopp2244d072010-08-17 08:59:14 +0000411 SKBTX_SW_TSTAMP = 1 << 1,
412
413 /* device driver is going to provide hardware time stamp */
414 SKBTX_IN_PROGRESS = 1 << 2,
415
Shirley Maa6686f22011-07-06 12:22:12 +0000416 /* device driver supports TX zero-copy buffers */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000417 SKBTX_DEV_ZEROCOPY = 1 << 3,
Johannes Berg6e3e9392011-11-09 10:15:42 +0100418
419 /* generate wifi status information (where possible) */
Eric Dumazet62b1a8a2012-06-14 06:42:44 +0000420 SKBTX_WIFI_STATUS = 1 << 4,
Pravin B Shelarc9af6db2013-02-11 09:27:41 +0000421
422 /* This indicates at least one fragment might be overwritten
423 * (as in vmsplice(), sendfile() ...)
424 * If we need to compute a TX checksum, we'll need to copy
425 * all frags to avoid possible bad checksum
426 */
427 SKBTX_SHARED_FRAG = 1 << 5,
Willem de Bruijne7fd2882014-08-04 22:11:48 -0400428
429 /* generate software time stamp when entering packet scheduling */
430 SKBTX_SCHED_TSTAMP = 1 << 6,
Shirley Maa6686f22011-07-06 12:22:12 +0000431};
432
Willem de Bruijn52267792017-08-03 16:29:39 -0400433#define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
Willem de Bruijne1c8a602014-08-04 22:11:50 -0400434#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \
Soheil Hassas Yeganeh0a2cf202016-04-27 23:39:01 -0400435 SKBTX_SCHED_TSTAMP)
Willem de Bruijnf24b9be2014-08-04 22:11:45 -0400436#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
437
Shirley Maa6686f22011-07-06 12:22:12 +0000438/*
439 * The callback notifies userspace to release buffers when skb DMA is done in
440 * lower device, the skb last reference should be 0 when calling this.
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000441 * The zerocopy_success argument is true if zero copy transmit occurred,
442 * false on data copy or out of memory error caused by data copy attempt.
Michael S. Tsirkinca8f4fb2012-04-09 00:24:02 +0000443 * The ctx field is used to track device context.
444 * The desc field is used to track userspace buffer index.
Shirley Maa6686f22011-07-06 12:22:12 +0000445 */
446struct ubuf_info {
Michael S. Tsirkine19d6762012-11-01 09:16:22 +0000447 void (*callback)(struct ubuf_info *, bool zerocopy_success);
Willem de Bruijn4ab6c992017-08-03 16:29:42 -0400448 union {
449 struct {
450 unsigned long desc;
451 void *ctx;
452 };
453 struct {
454 u32 id;
455 u16 len;
456 u16 zerocopy:1;
457 u32 bytelen;
458 };
459 };
Eric Dumazetc1d1b432017-08-31 16:48:22 -0700460 refcount_t refcnt;
Willem de Bruijna91dbff2017-08-03 16:29:43 -0400461
462 struct mmpin {
463 struct user_struct *user;
464 unsigned int num_pg;
465 } mmp;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000466};
467
Willem de Bruijn52267792017-08-03 16:29:39 -0400468#define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
469
470struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
Willem de Bruijn4ab6c992017-08-03 16:29:42 -0400471struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
472 struct ubuf_info *uarg);
Willem de Bruijn52267792017-08-03 16:29:39 -0400473
474static inline void sock_zerocopy_get(struct ubuf_info *uarg)
475{
Eric Dumazetc1d1b432017-08-31 16:48:22 -0700476 refcount_inc(&uarg->refcnt);
Willem de Bruijn52267792017-08-03 16:29:39 -0400477}
478
479void sock_zerocopy_put(struct ubuf_info *uarg);
480void sock_zerocopy_put_abort(struct ubuf_info *uarg);
481
482void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
483
484int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
485 struct msghdr *msg, int len,
486 struct ubuf_info *uarg);
487
Linus Torvalds1da177e2005-04-16 15:20:36 -0700488/* This data is invariant across clones and lives at
489 * the end of the header data, ie. at skb->end.
490 */
491struct skb_shared_info {
Daniel Borkmannde8f3a82017-09-25 02:25:51 +0200492 __u8 __unused;
493 __u8 meta_len;
494 __u8 nr_frags;
Ian Campbell9f42f122012-01-05 07:13:39 +0000495 __u8 tx_flags;
Herbert Xu79671682006-06-22 02:40:14 -0700496 unsigned short gso_size;
497 /* Warning: this field is not always filled in (UFO)! */
498 unsigned short gso_segs;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700499 struct sk_buff *frag_list;
Patrick Ohlyac45f602009-02-12 05:03:37 +0000500 struct skb_shared_hwtstamps hwtstamps;
Steffen Klassert7f564522017-04-08 20:36:24 +0200501 unsigned int gso_type;
Willem de Bruijn09c2d252014-08-04 22:11:47 -0400502 u32 tskey;
Ian Campbell9f42f122012-01-05 07:13:39 +0000503 __be32 ip6_frag_id;
Eric Dumazetec7d2f22010-05-05 01:07:37 -0700504
505 /*
506 * Warning : all fields before dataref are cleared in __alloc_skb()
507 */
508 atomic_t dataref;
509
Johann Baudy69e3c752009-05-18 22:11:22 -0700510 /* Intermediate layers must ensure that destructor_arg
511 * remains valid until skb destructor */
512 void * destructor_arg;
Shirley Maa6686f22011-07-06 12:22:12 +0000513
Eric Dumazetfed66382010-07-22 19:09:08 +0000514 /* must be last field, see pskb_expand_head() */
515 skb_frag_t frags[MAX_SKB_FRAGS];
Linus Torvalds1da177e2005-04-16 15:20:36 -0700516};
517
518/* We divide dataref into two halves. The higher 16 bits hold references
519 * to the payload part of skb->data. The lower 16 bits hold references to
Patrick McHardy334a8132007-06-25 04:35:20 -0700520 * the entire skb->data. A clone of a headerless skb holds the length of
521 * the header in skb->hdr_len.
Linus Torvalds1da177e2005-04-16 15:20:36 -0700522 *
523 * All users must obey the rule that the skb->data reference count must be
524 * greater than or equal to the payload reference count.
525 *
526 * Holding a reference to the payload part means that the user does not
527 * care about modifications to the header part of skb->data.
528 */
529#define SKB_DATAREF_SHIFT 16
530#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
531
David S. Millerd179cd12005-08-17 14:57:30 -0700532
533enum {
Vijay Subramanianc8753d52014-10-02 10:00:43 -0700534 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
535 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */
536 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */
David S. Millerd179cd12005-08-17 14:57:30 -0700537};
538
Herbert Xu79671682006-06-22 02:40:14 -0700539enum {
540 SKB_GSO_TCPV4 = 1 << 0,
Herbert Xu576a30e2006-06-27 13:22:38 -0700541
542 /* This indicates the skb is from an untrusted source. */
David S. Millerd9d30ad2017-07-03 07:31:57 -0700543 SKB_GSO_DODGY = 1 << 1,
Michael Chanb0da85372006-06-29 12:30:00 -0700544
545 /* This indicates the tcp segment has CWR set. */
David S. Millerd9d30ad2017-07-03 07:31:57 -0700546 SKB_GSO_TCP_ECN = 1 << 2,
Herbert Xuf83ef8c2006-06-30 13:37:03 -0700547
David S. Millerd9d30ad2017-07-03 07:31:57 -0700548 SKB_GSO_TCP_FIXEDID = 1 << 3,
Chris Leech01d5b2f2009-02-27 14:06:49 -0800549
David S. Millerd9d30ad2017-07-03 07:31:57 -0700550 SKB_GSO_TCPV6 = 1 << 4,
Pravin B Shelar68c33162013-02-14 14:02:41 +0000551
David S. Millerd9d30ad2017-07-03 07:31:57 -0700552 SKB_GSO_FCOE = 1 << 5,
Pravin B Shelar73136262013-03-07 13:21:51 +0000553
David S. Millerd9d30ad2017-07-03 07:31:57 -0700554 SKB_GSO_GRE = 1 << 6,
Simon Horman0d89d202013-05-23 21:02:52 +0000555
David S. Millerd9d30ad2017-07-03 07:31:57 -0700556 SKB_GSO_GRE_CSUM = 1 << 7,
Eric Dumazetcb32f512013-10-19 11:42:57 -0700557
David S. Millerd9d30ad2017-07-03 07:31:57 -0700558 SKB_GSO_IPXIP4 = 1 << 8,
Eric Dumazet61c1db72013-10-20 20:47:30 -0700559
David S. Millerd9d30ad2017-07-03 07:31:57 -0700560 SKB_GSO_IPXIP6 = 1 << 9,
Tom Herbert0f4f4ff2014-06-04 17:20:16 -0700561
David S. Millerd9d30ad2017-07-03 07:31:57 -0700562 SKB_GSO_UDP_TUNNEL = 1 << 10,
Tom Herbert4749c092014-06-04 17:20:23 -0700563
David S. Millerd9d30ad2017-07-03 07:31:57 -0700564 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
Alexander Duyckcbc53e02016-04-10 21:44:51 -0400565
David S. Millerd9d30ad2017-07-03 07:31:57 -0700566 SKB_GSO_PARTIAL = 1 << 12,
Alexander Duyck802ab552016-04-10 21:45:03 -0400567
David S. Millerd9d30ad2017-07-03 07:31:57 -0700568 SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
Marcelo Ricardo Leitner90017ac2016-06-02 15:05:43 -0300569
David S. Millerd9d30ad2017-07-03 07:31:57 -0700570 SKB_GSO_SCTP = 1 << 14,
Steffen Klassertc7ef8f02017-04-14 10:05:36 +0200571
David S. Millerd9d30ad2017-07-03 07:31:57 -0700572 SKB_GSO_ESP = 1 << 15,
Herbert Xu79671682006-06-22 02:40:14 -0700573};
574
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -0700575#if BITS_PER_LONG > 32
576#define NET_SKBUFF_DATA_USES_OFFSET 1
577#endif
578
579#ifdef NET_SKBUFF_DATA_USES_OFFSET
580typedef unsigned int sk_buff_data_t;
581#else
582typedef unsigned char *sk_buff_data_t;
583#endif
584
Linus Torvalds1da177e2005-04-16 15:20:36 -0700585/**
586 * struct sk_buff - socket buffer
587 * @next: Next buffer in list
588 * @prev: Previous buffer in list
Eric Dumazet363ec39232014-02-26 14:02:11 -0800589 * @tstamp: Time we arrived/left
Eric Dumazet56b17422014-11-03 08:19:53 -0800590 * @rbnode: RB tree node, alternative to next/prev for netem/tcp
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700591 * @sk: Socket we are owned by
Linus Torvalds1da177e2005-04-16 15:20:36 -0700592 * @dev: Device we arrived on/are leaving by
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700593 * @cb: Control buffer. Free for use by every layer. Put private vars here
Eric Dumazet7fee2262010-05-11 23:19:48 +0000594 * @_skb_refdst: destination entry (with norefcount bit)
Martin Waitz67be2dd2005-05-01 08:59:26 -0700595 * @sp: the security path, used for xfrm
Linus Torvalds1da177e2005-04-16 15:20:36 -0700596 * @len: Length of actual data
597 * @data_len: Data length
598 * @mac_len: Length of link layer header
Patrick McHardy334a8132007-06-25 04:35:20 -0700599 * @hdr_len: writable header length of cloned skb
Herbert Xu663ead32007-04-09 11:59:07 -0700600 * @csum: Checksum (must include start/offset pair)
601 * @csum_start: Offset from skb->head where checksumming should start
602 * @csum_offset: Offset from csum_start where checksum should be stored
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700603 * @priority: Packet queueing priority
WANG Cong60ff7462014-05-04 16:39:18 -0700604 * @ignore_df: allow local fragmentation
Linus Torvalds1da177e2005-04-16 15:20:36 -0700605 * @cloned: Head may be cloned (check refcnt to be sure)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700606 * @ip_summed: Driver fed us an IP checksum
Linus Torvalds1da177e2005-04-16 15:20:36 -0700607 * @nohdr: Payload reference only, must not modify header
608 * @pkt_type: Packet class
Randy Dunlapc83c2482005-10-18 22:07:41 -0700609 * @fclone: skbuff clone status
Randy Dunlapc83c2482005-10-18 22:07:41 -0700610 * @ipvs_property: skbuff is owned by ipvs
Willem de Bruijne7246e12017-01-07 17:06:35 -0500611 * @tc_skip_classify: do not classify packet. set by IFB device
Willem de Bruijn8dc07fd2017-01-07 17:06:37 -0500612 * @tc_at_ingress: used within tc_classify to distinguish in/egress
Willem de Bruijnbc31c902017-01-07 17:06:38 -0500613 * @tc_redirected: packet was redirected by a tc action
614 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
Randy Dunlap31729362008-02-18 20:52:13 -0800615 * @peeked: this packet has been seen already, so stats have been
616 * done for it, don't do them again
Jozsef Kadlecsikba9dda32007-07-07 22:21:23 -0700617 * @nf_trace: netfilter packet trace flag
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700618 * @protocol: Packet protocol from driver
619 * @destructor: Destruct function
Eric Dumazete2080072017-10-04 12:59:58 -0700620 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
Florian Westphala9e419d2017-01-23 18:21:59 +0100621 * @_nfct: Associated connection, if any (with nfctinfo bits)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700622 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
Eric Dumazet8964be42009-11-20 15:35:04 -0800623 * @skb_iif: ifindex of device we arrived on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700624 * @tc_index: Traffic control index
Tom Herbert61b905d2014-03-24 15:34:47 -0700625 * @hash: the packet hash
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700626 * @queue_mapping: Queue mapping for multiqueue devices
David S. Miller0b725a22014-08-25 15:51:53 -0700627 * @xmit_more: More SKBs are pending for this queue
Randy Dunlap553a5672008-04-20 10:51:01 -0700628 * @ndisc_nodetype: router type (from link layer)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700629 * @ooo_okay: allow the mapping of a socket to a queue to be changed
Tom Herbert61b905d2014-03-24 15:34:47 -0700630 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
Changli Gao4ca24622011-08-19 07:26:44 -0700631 * ports.
Tom Herberta3b18dd2014-07-01 21:33:17 -0700632 * @sw_hash: indicates hash was computed in software stack
Johannes Berg6e3e9392011-11-09 10:15:42 +0100633 * @wifi_acked_valid: wifi_acked was set
634 * @wifi_acked: whether frame was acked on wifi or not
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000635 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
Davide Carattidba00302017-05-18 15:44:40 +0200636 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
Julian Anastasov4ff06202017-02-06 23:14:12 +0200637 * @dst_pending_confirm: need to confirm neighbour
Eliezer Tamir06021292013-06-10 11:39:50 +0300638 * @napi_id: id of the NAPI struct this skb came from
James Morris984bc162006-06-09 00:29:17 -0700639 * @secmark: security marking
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700640 * @mark: Generic packet mark
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000641 * @vlan_proto: vlan encapsulation protocol
Patrick McHardy6aa895b2008-07-14 22:49:06 -0700642 * @vlan_tci: vlan tag control information
Simon Horman0d89d202013-05-23 21:02:52 +0000643 * @inner_protocol: Protocol (encapsulation)
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000644 * @inner_transport_header: Inner transport layer header (encapsulation)
645 * @inner_network_header: Network layer header (encapsulation)
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000646 * @inner_mac_header: Link layer header (encapsulation)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700647 * @transport_header: Transport layer header
648 * @network_header: Network layer header
649 * @mac_header: Link layer header
650 * @tail: Tail pointer
651 * @end: End pointer
652 * @head: Head of buffer
653 * @data: Data head pointer
654 * @truesize: Buffer size
655 * @users: User count - see {datagram,tcp}.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700656 */
657
658struct sk_buff {
Eric Dumazet363ec39232014-02-26 14:02:11 -0800659 union {
Eric Dumazet56b17422014-11-03 08:19:53 -0800660 struct {
661 /* These two members must be first. */
662 struct sk_buff *next;
663 struct sk_buff *prev;
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000664
Eric Dumazet56b17422014-11-03 08:19:53 -0800665 union {
Eric Dumazetbffa72c2017-09-19 05:14:24 -0700666 struct net_device *dev;
667 /* Some protocols might use this space to store information,
668 * while device pointer would be NULL.
669 * UDP receive path is one user.
670 */
671 unsigned long dev_scratch;
Eric Dumazet56b17422014-11-03 08:19:53 -0800672 };
673 };
674 struct rb_node rbnode; /* used in netem & tcp stack */
675 };
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000676 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700677
Eric Dumazetc84d9492016-12-08 11:41:55 -0800678 union {
Eric Dumazetbffa72c2017-09-19 05:14:24 -0700679 ktime_t tstamp;
680 u64 skb_mstamp;
Eric Dumazetc84d9492016-12-08 11:41:55 -0800681 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700682 /*
683 * This is the control buffer. It is free to use for every
684 * layer. Please put your private variables there. If you
685 * want to keep them across layers you have to do a skb_clone()
686 * first. This is owned by whoever has the skb queued ATM.
687 */
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000688 char cb[48] __aligned(8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700689
Eric Dumazete2080072017-10-04 12:59:58 -0700690 union {
691 struct {
692 unsigned long _skb_refdst;
693 void (*destructor)(struct sk_buff *skb);
694 };
695 struct list_head tcp_tsorted_anchor;
696 };
697
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000698#ifdef CONFIG_XFRM
699 struct sec_path *sp;
700#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700701#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Florian Westphala9e419d2017-01-23 18:21:59 +0100702 unsigned long _nfct;
Eric Dumazetb1937222014-09-28 22:18:47 -0700703#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200704#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Eric Dumazetb1937222014-09-28 22:18:47 -0700705 struct nf_bridge_info *nf_bridge;
706#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700707 unsigned int len,
Patrick McHardy334a8132007-06-25 04:35:20 -0700708 data_len;
709 __u16 mac_len,
710 hdr_len;
Eric Dumazetb1937222014-09-28 22:18:47 -0700711
712 /* Following fields are _not_ copied in __copy_skb_header()
713 * Note that queue_mapping is here mostly to fill a hole.
714 */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200715 kmemcheck_bitfield_begin(flags1);
Eric Dumazetb1937222014-09-28 22:18:47 -0700716 __u16 queue_mapping;
Daniel Borkmann36bbef52016-09-20 00:26:13 +0200717
718/* if you move cloned around you also must adapt those constants */
719#ifdef __BIG_ENDIAN_BITFIELD
720#define CLONED_MASK (1 << 7)
721#else
722#define CLONED_MASK 1
723#endif
724#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
725
726 __u8 __cloned_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700727 __u8 cloned:1,
Harald Welte6869c4d2005-08-09 19:24:19 -0700728 nohdr:1,
Patrick McHardyb84f4cc2005-11-20 21:19:21 -0800729 fclone:2,
Herbert Xua59322b2007-12-05 01:53:40 -0800730 peeked:1,
Eric Dumazetb1937222014-09-28 22:18:47 -0700731 head_frag:1,
Daniel Borkmann36bbef52016-09-20 00:26:13 +0200732 xmit_more:1,
733 __unused:1; /* one bit hole */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200734 kmemcheck_bitfield_end(flags1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700735
Eric Dumazetb1937222014-09-28 22:18:47 -0700736 /* fields enclosed in headers_start/headers_end are copied
737 * using a single memcpy() in __copy_skb_header()
738 */
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700739 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700740 __u32 headers_start[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700741 /* public: */
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200742
743/* if you move pkt_type around you also must adapt those constants */
744#ifdef __BIG_ENDIAN_BITFIELD
745#define PKT_TYPE_MAX (7 << 5)
746#else
747#define PKT_TYPE_MAX 7
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100748#endif
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200749#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
750
751 __u8 __pkt_type_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700752 __u8 pkt_type:3;
753 __u8 pfmemalloc:1;
754 __u8 ignore_df:1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700755
Eric Dumazetb1937222014-09-28 22:18:47 -0700756 __u8 nf_trace:1;
757 __u8 ip_summed:2;
758 __u8 ooo_okay:1;
759 __u8 l4_hash:1;
760 __u8 sw_hash:1;
761 __u8 wifi_acked_valid:1;
762 __u8 wifi_acked:1;
763
764 __u8 no_fcs:1;
765 /* Indicates the inner headers are valid in the skbuff. */
766 __u8 encapsulation:1;
767 __u8 encap_hdr_csum:1;
768 __u8 csum_valid:1;
769 __u8 csum_complete_sw:1;
770 __u8 csum_level:2;
Davide Carattidba00302017-05-18 15:44:40 +0200771 __u8 csum_not_inet:1;
Eric Dumazetb1937222014-09-28 22:18:47 -0700772
Julian Anastasov4ff06202017-02-06 23:14:12 +0200773 __u8 dst_pending_confirm:1;
Eric Dumazetb1937222014-09-28 22:18:47 -0700774#ifdef CONFIG_IPV6_NDISC_NODETYPE
775 __u8 ndisc_nodetype:2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700776#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700777 __u8 ipvs_property:1;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700778 __u8 inner_protocol_type:1;
Tom Herberte585f232014-11-04 09:06:54 -0800779 __u8 remcsum_offload:1;
Ido Schimmel6bc506b2016-08-25 18:42:37 +0200780#ifdef CONFIG_NET_SWITCHDEV
781 __u8 offload_fwd_mark:1;
Yotam Gigiabf4bb62017-10-03 09:58:06 +0200782 __u8 offload_mr_fwd_mark:1;
Ido Schimmel6bc506b2016-08-25 18:42:37 +0200783#endif
Willem de Bruijne7246e12017-01-07 17:06:35 -0500784#ifdef CONFIG_NET_CLS_ACT
785 __u8 tc_skip_classify:1;
Willem de Bruijn8dc07fd2017-01-07 17:06:37 -0500786 __u8 tc_at_ingress:1;
Willem de Bruijnbc31c902017-01-07 17:06:38 -0500787 __u8 tc_redirected:1;
788 __u8 tc_from_ingress:1;
Willem de Bruijne7246e12017-01-07 17:06:35 -0500789#endif
Alexander Duyck4031ae62012-01-27 06:22:53 +0000790
Linus Torvalds1da177e2005-04-16 15:20:36 -0700791#ifdef CONFIG_NET_SCHED
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700792 __u16 tc_index; /* traffic control index */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700793#endif
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200794
Eric Dumazetb1937222014-09-28 22:18:47 -0700795 union {
796 __wsum csum;
797 struct {
798 __u16 csum_start;
799 __u16 csum_offset;
800 };
801 };
802 __u32 priority;
803 int skb_iif;
804 __u32 hash;
805 __be16 vlan_proto;
806 __u16 vlan_tci;
Eric Dumazet2bd82482015-02-03 23:48:24 -0800807#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
808 union {
809 unsigned int napi_id;
810 unsigned int sender_cpu;
811 };
Chris Leech97fc2f02006-05-23 17:55:33 -0700812#endif
James Morris984bc162006-06-09 00:29:17 -0700813#ifdef CONFIG_NETWORK_SECMARK
Ido Schimmel6bc506b2016-08-25 18:42:37 +0200814 __u32 secmark;
James Morris984bc162006-06-09 00:29:17 -0700815#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700816
Neil Horman3b885782009-10-12 13:26:31 -0700817 union {
818 __u32 mark;
Eric Dumazet16fad692013-03-14 05:40:32 +0000819 __u32 reserved_tailroom;
Neil Horman3b885782009-10-12 13:26:31 -0700820 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700821
Tom Herbert8bce6d72014-09-29 20:22:29 -0700822 union {
823 __be16 inner_protocol;
824 __u8 inner_ipproto;
825 };
826
Simon Horman1a37e412013-05-23 21:02:51 +0000827 __u16 inner_transport_header;
828 __u16 inner_network_header;
829 __u16 inner_mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700830
831 __be16 protocol;
Simon Horman1a37e412013-05-23 21:02:51 +0000832 __u16 transport_header;
833 __u16 network_header;
834 __u16 mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700835
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700836 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700837 __u32 headers_end[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700838 /* public: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700839
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840 /* These elements must be at the end, see alloc_skb() for details. */
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700841 sk_buff_data_t tail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700842 sk_buff_data_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700843 unsigned char *head,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700844 *data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700845 unsigned int truesize;
Reshetova, Elena63354792017-06-30 13:07:58 +0300846 refcount_t users;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700847};
848
849#ifdef __KERNEL__
850/*
851 * Handling routines are only of interest to the kernel
852 */
853#include <linux/slab.h>
854
Linus Torvalds1da177e2005-04-16 15:20:36 -0700855
Mel Gormanc93bdd02012-07-31 16:44:19 -0700856#define SKB_ALLOC_FCLONE 0x01
857#define SKB_ALLOC_RX 0x02
Alexander Duyckfd11a832014-12-09 19:40:49 -0800858#define SKB_ALLOC_NAPI 0x04
Mel Gormanc93bdd02012-07-31 16:44:19 -0700859
860/* Returns true if the skb was allocated from PFMEMALLOC reserves */
861static inline bool skb_pfmemalloc(const struct sk_buff *skb)
862{
863 return unlikely(skb->pfmemalloc);
864}
865
Eric Dumazet7fee2262010-05-11 23:19:48 +0000866/*
867 * skb might have a dst pointer attached, refcounted or not.
868 * _skb_refdst low order bit is set if refcount was _not_ taken
869 */
870#define SKB_DST_NOREF 1UL
871#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
872
Florian Westphala9e419d2017-01-23 18:21:59 +0100873#define SKB_NFCT_PTRMASK ~(7UL)
Eric Dumazet7fee2262010-05-11 23:19:48 +0000874/**
875 * skb_dst - returns skb dst_entry
876 * @skb: buffer
877 *
878 * Returns skb dst_entry, regardless of reference taken or not.
879 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000880static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
881{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000882 /* If refdst was not refcounted, check we still are in a
883 * rcu_read_lock section
884 */
885 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
886 !rcu_read_lock_held() &&
887 !rcu_read_lock_bh_held());
888 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
Eric Dumazetadf30902009-06-02 05:19:30 +0000889}
890
Eric Dumazet7fee2262010-05-11 23:19:48 +0000891/**
892 * skb_dst_set - sets skb dst
893 * @skb: buffer
894 * @dst: dst entry
895 *
896 * Sets skb dst, assuming a reference was taken on dst and should
897 * be released by skb_dst_drop()
898 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000899static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
900{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000901 skb->_skb_refdst = (unsigned long)dst;
902}
903
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200904/**
905 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
906 * @skb: buffer
907 * @dst: dst entry
908 *
909 * Sets skb dst, assuming a reference was not taken on dst.
910 * If dst entry is cached, we do not take reference and dst_release
911 * will be avoided by refdst_drop. If dst entry is not cached, we take
912 * reference, so that last dst_release can destroy the dst immediately.
913 */
914static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
915{
Hannes Frederic Sowadbfc4fb2014-12-06 19:19:42 +0100916 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
917 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200918}
Eric Dumazet7fee2262010-05-11 23:19:48 +0000919
920/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300921 * skb_dst_is_noref - Test if skb dst isn't refcounted
Eric Dumazet7fee2262010-05-11 23:19:48 +0000922 * @skb: buffer
923 */
924static inline bool skb_dst_is_noref(const struct sk_buff *skb)
925{
926 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
Eric Dumazetadf30902009-06-02 05:19:30 +0000927}
928
Eric Dumazet511c3f92009-06-02 05:14:27 +0000929static inline struct rtable *skb_rtable(const struct sk_buff *skb)
930{
Eric Dumazetadf30902009-06-02 05:19:30 +0000931 return (struct rtable *)skb_dst(skb);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000932}
933
Jamal Hadi Salim8b10cab2016-07-02 06:43:14 -0400934/* For mangling skb->pkt_type from user space side from applications
935 * such as nft, tc, etc, we only allow a conservative subset of
936 * possible pkt_types to be set.
937*/
938static inline bool skb_pkt_type_ok(u32 ptype)
939{
940 return ptype <= PACKET_OTHERHOST;
941}
942
Miroslav Lichvar90b602f2017-05-19 17:52:37 +0200943static inline unsigned int skb_napi_id(const struct sk_buff *skb)
944{
945#ifdef CONFIG_NET_RX_BUSY_POLL
946 return skb->napi_id;
947#else
948 return 0;
949#endif
950}
951
Paolo Abeni3889a8032017-06-12 11:23:41 +0200952/* decrement the reference count and return true if we can free the skb */
953static inline bool skb_unref(struct sk_buff *skb)
954{
955 if (unlikely(!skb))
956 return false;
Reshetova, Elena63354792017-06-30 13:07:58 +0300957 if (likely(refcount_read(&skb->users) == 1))
Paolo Abeni3889a8032017-06-12 11:23:41 +0200958 smp_rmb();
Reshetova, Elena63354792017-06-30 13:07:58 +0300959 else if (likely(!refcount_dec_and_test(&skb->users)))
Paolo Abeni3889a8032017-06-12 11:23:41 +0200960 return false;
961
962 return true;
963}
964
Paolo Abeni0a463c72017-06-12 11:23:42 +0200965void skb_release_head_state(struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -0700966void kfree_skb(struct sk_buff *skb);
967void kfree_skb_list(struct sk_buff *segs);
968void skb_tx_error(struct sk_buff *skb);
969void consume_skb(struct sk_buff *skb);
Paolo Abenica2c1412017-09-06 14:44:36 +0200970void __consume_stateless_skb(struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -0700971void __kfree_skb(struct sk_buff *skb);
Eric Dumazetd7e88832012-04-30 08:10:34 +0000972extern struct kmem_cache *skbuff_head_cache;
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000973
Joe Perches7965bd42013-09-26 14:48:15 -0700974void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
975bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
976 bool *fragstolen, int *delta_truesize);
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000977
Joe Perches7965bd42013-09-26 14:48:15 -0700978struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
979 int node);
Eric Dumazet2ea2f622015-04-24 16:05:01 -0700980struct sk_buff *__build_skb(void *data, unsigned int frag_size);
Joe Perches7965bd42013-09-26 14:48:15 -0700981struct sk_buff *build_skb(void *data, unsigned int frag_size);
David S. Millerd179cd12005-08-17 14:57:30 -0700982static inline struct sk_buff *alloc_skb(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100983 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700984{
Eric Dumazet564824b2010-10-11 19:05:25 +0000985 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700986}
987
Eric Dumazet2e4e4412014-09-17 04:49:49 -0700988struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
989 unsigned long data_len,
990 int max_page_order,
991 int *errcode,
992 gfp_t gfp_mask);
993
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700994/* Layout of fast clones : [skb1][skb2][fclone_ref] */
995struct sk_buff_fclones {
996 struct sk_buff skb1;
997
998 struct sk_buff skb2;
999
Reshetova, Elena26385952017-06-30 13:07:59 +03001000 refcount_t fclone_ref;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -07001001};
1002
1003/**
1004 * skb_fclone_busy - check if fclone is busy
Stephen Hemminger293de7d2016-10-23 09:28:29 -07001005 * @sk: socket
Eric Dumazetd0bf4a92014-09-29 13:29:15 -07001006 * @skb: buffer
1007 *
Masanari Iidabda13fe2015-12-13 16:53:02 +09001008 * Returns true if skb is a fast clone, and its clone is not freed.
Eric Dumazet39bb5e62014-10-30 10:32:34 -07001009 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1010 * so we also check that this didnt happen.
Eric Dumazetd0bf4a92014-09-29 13:29:15 -07001011 */
Eric Dumazet39bb5e62014-10-30 10:32:34 -07001012static inline bool skb_fclone_busy(const struct sock *sk,
1013 const struct sk_buff *skb)
Eric Dumazetd0bf4a92014-09-29 13:29:15 -07001014{
1015 const struct sk_buff_fclones *fclones;
1016
1017 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1018
1019 return skb->fclone == SKB_FCLONE_ORIG &&
Reshetova, Elena26385952017-06-30 13:07:59 +03001020 refcount_read(&fclones->fclone_ref) > 1 &&
Eric Dumazet39bb5e62014-10-30 10:32:34 -07001021 fclones->skb2.sk == sk;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -07001022}
1023
David S. Millerd179cd12005-08-17 14:57:30 -07001024static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +01001025 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -07001026{
Mel Gormanc93bdd02012-07-31 16:44:19 -07001027 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -07001028}
1029
Joe Perches7965bd42013-09-26 14:48:15 -07001030struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1031int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1032struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1033struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
Octavian Purdilabad93e92014-06-12 01:36:26 +03001034struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1035 gfp_t gfp_mask, bool fclone);
1036static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1037 gfp_t gfp_mask)
1038{
1039 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1040}
Eric Dumazet117632e2011-12-03 21:39:53 +00001041
Joe Perches7965bd42013-09-26 14:48:15 -07001042int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1043struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1044 unsigned int headroom);
1045struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1046 int newtailroom, gfp_t priority);
Jason A. Donenfeld48a1df62017-06-04 04:16:22 +02001047int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1048 int offset, int len);
1049int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1050 int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001051int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
Florian Fainellicd0a1372017-08-22 15:12:14 -07001052int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1053
1054/**
1055 * skb_pad - zero pad the tail of an skb
1056 * @skb: buffer to pad
1057 * @pad: space to pad
1058 *
1059 * Ensure that a buffer is followed by a padding area that is zero
1060 * filled. Used by network drivers which may DMA or transfer data
1061 * beyond the buffer end onto the wire.
1062 *
1063 * May return error in out of memory cases. The skb is freed on error.
1064 */
1065static inline int skb_pad(struct sk_buff *skb, int pad)
1066{
1067 return __skb_pad(skb, pad, true);
1068}
Neil Hormanead2ceb2009-03-11 09:49:55 +00001069#define dev_kfree_skb(a) consume_skb(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001070
Joe Perches7965bd42013-09-26 14:48:15 -07001071int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1072 int getfrag(void *from, char *to, int offset,
1073 int len, int odd, struct sk_buff *skb),
1074 void *from, int length);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001075
Hannes Frederic Sowabe12a1f2015-05-21 16:59:58 +02001076int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1077 int offset, size_t size);
1078
Eric Dumazetd94d9fe2009-11-04 09:50:58 -08001079struct skb_seq_state {
Thomas Graf677e90e2005-06-23 20:59:51 -07001080 __u32 lower_offset;
1081 __u32 upper_offset;
1082 __u32 frag_idx;
1083 __u32 stepped_offset;
1084 struct sk_buff *root_skb;
1085 struct sk_buff *cur_skb;
1086 __u8 *frag_data;
1087};
1088
Joe Perches7965bd42013-09-26 14:48:15 -07001089void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1090 unsigned int to, struct skb_seq_state *st);
1091unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1092 struct skb_seq_state *st);
1093void skb_abort_seq_read(struct skb_seq_state *st);
Thomas Graf677e90e2005-06-23 20:59:51 -07001094
Joe Perches7965bd42013-09-26 14:48:15 -07001095unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
Bojan Prtvar059a2442015-02-22 11:46:35 +01001096 unsigned int to, struct ts_config *config);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07001097
Tom Herbert09323cc2013-12-15 22:16:19 -08001098/*
1099 * Packet hash types specify the type of hash in skb_set_hash.
1100 *
1101 * Hash types refer to the protocol layer addresses which are used to
1102 * construct a packet's hash. The hashes are used to differentiate or identify
1103 * flows of the protocol layer for the hash type. Hash types are either
1104 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1105 *
1106 * Properties of hashes:
1107 *
1108 * 1) Two packets in different flows have different hash values
1109 * 2) Two packets in the same flow should have the same hash value
1110 *
1111 * A hash at a higher layer is considered to be more specific. A driver should
1112 * set the most specific hash possible.
1113 *
1114 * A driver cannot indicate a more specific hash than the layer at which a hash
1115 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1116 *
1117 * A driver may indicate a hash level which is less specific than the
1118 * actual layer the hash was computed on. For instance, a hash computed
1119 * at L4 may be considered an L3 hash. This should only be done if the
1120 * driver can't unambiguously determine that the HW computed the hash at
1121 * the higher layer. Note that the "should" in the second property above
1122 * permits this.
1123 */
1124enum pkt_hash_types {
1125 PKT_HASH_TYPE_NONE, /* Undefined type */
1126 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1127 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1128 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1129};
1130
Tom Herbertbcc83832015-09-01 09:24:24 -07001131static inline void skb_clear_hash(struct sk_buff *skb)
1132{
1133 skb->hash = 0;
1134 skb->sw_hash = 0;
1135 skb->l4_hash = 0;
1136}
1137
1138static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1139{
1140 if (!skb->l4_hash)
1141 skb_clear_hash(skb);
1142}
1143
1144static inline void
1145__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1146{
1147 skb->l4_hash = is_l4;
1148 skb->sw_hash = is_sw;
1149 skb->hash = hash;
1150}
1151
Tom Herbert09323cc2013-12-15 22:16:19 -08001152static inline void
1153skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1154{
Tom Herbertbcc83832015-09-01 09:24:24 -07001155 /* Used by drivers to set hash from HW */
1156 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1157}
1158
1159static inline void
1160__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1161{
1162 __skb_set_hash(skb, hash, true, is_l4);
Tom Herbert09323cc2013-12-15 22:16:19 -08001163}
1164
Tom Herberte5276932015-09-01 09:24:23 -07001165void __skb_get_hash(struct sk_buff *skb);
Florian Westphalb9177832016-10-26 18:49:46 +02001166u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
Tom Herberte5276932015-09-01 09:24:23 -07001167u32 skb_get_poff(const struct sk_buff *skb);
1168u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1169 const struct flow_keys *keys, int hlen);
1170__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1171 void *data, int hlen_proto);
1172
1173static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1174 int thoff, u8 ip_proto)
1175{
1176 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1177}
1178
1179void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1180 const struct flow_dissector_key *key,
1181 unsigned int key_count);
1182
1183bool __skb_flow_dissect(const struct sk_buff *skb,
1184 struct flow_dissector *flow_dissector,
1185 void *target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001186 void *data, __be16 proto, int nhoff, int hlen,
1187 unsigned int flags);
Tom Herberte5276932015-09-01 09:24:23 -07001188
1189static inline bool skb_flow_dissect(const struct sk_buff *skb,
1190 struct flow_dissector *flow_dissector,
Tom Herbertcd79a232015-09-01 09:24:27 -07001191 void *target_container, unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001192{
1193 return __skb_flow_dissect(skb, flow_dissector, target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001194 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001195}
1196
1197static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
Tom Herbertcd79a232015-09-01 09:24:27 -07001198 struct flow_keys *flow,
1199 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001200{
1201 memset(flow, 0, sizeof(*flow));
1202 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001203 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001204}
1205
1206static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1207 void *data, __be16 proto,
Tom Herbertcd79a232015-09-01 09:24:27 -07001208 int nhoff, int hlen,
1209 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001210{
1211 memset(flow, 0, sizeof(*flow));
1212 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001213 data, proto, nhoff, hlen, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001214}
1215
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001216static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001217{
Tom Herberta3b18dd2014-07-01 21:33:17 -07001218 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001219 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001220
Tom Herbert61b905d2014-03-24 15:34:47 -07001221 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001222}
1223
David S. Miller20a17bf2015-09-01 21:19:17 -07001224static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
Tom Herbertf70ea012015-07-31 16:52:10 -07001225{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001226 if (!skb->l4_hash && !skb->sw_hash) {
1227 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001228 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001229
Tom Herbertde4c1f82015-09-01 18:11:04 -07001230 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001231 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001232
1233 return skb->hash;
1234}
1235
Tom Herbert50fb7992015-05-01 11:30:12 -07001236__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1237
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001238static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1239{
Tom Herbert61b905d2014-03-24 15:34:47 -07001240 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001241}
1242
Tom Herbert3df7a742013-12-15 22:16:29 -08001243static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1244{
Tom Herbert61b905d2014-03-24 15:34:47 -07001245 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -07001246 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -07001247 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -08001248};
1249
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001250#ifdef NET_SKBUFF_DATA_USES_OFFSET
1251static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1252{
1253 return skb->head + skb->end;
1254}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001255
1256static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1257{
1258 return skb->end;
1259}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001260#else
1261static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1262{
1263 return skb->end;
1264}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001265
1266static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1267{
1268 return skb->end - skb->head;
1269}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001270#endif
1271
Linus Torvalds1da177e2005-04-16 15:20:36 -07001272/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001273#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001274
Patrick Ohlyac45f602009-02-12 05:03:37 +00001275static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1276{
1277 return &skb_shinfo(skb)->hwtstamps;
1278}
1279
Willem de Bruijn52267792017-08-03 16:29:39 -04001280static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1281{
1282 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1283
1284 return is_zcopy ? skb_uarg(skb) : NULL;
1285}
1286
1287static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg)
1288{
1289 if (skb && uarg && !skb_zcopy(skb)) {
1290 sock_zerocopy_get(uarg);
1291 skb_shinfo(skb)->destructor_arg = uarg;
1292 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1293 }
1294}
1295
1296/* Release a reference on a zerocopy structure */
1297static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1298{
1299 struct ubuf_info *uarg = skb_zcopy(skb);
1300
1301 if (uarg) {
Willem de Bruijn0a4a0602017-08-09 19:09:44 -04001302 if (uarg->callback == sock_zerocopy_callback) {
1303 uarg->zerocopy = uarg->zerocopy && zerocopy;
1304 sock_zerocopy_put(uarg);
1305 } else {
1306 uarg->callback(uarg, zerocopy);
1307 }
1308
Willem de Bruijn52267792017-08-03 16:29:39 -04001309 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1310 }
1311}
1312
1313/* Abort a zerocopy operation and revert zckey on error in send syscall */
1314static inline void skb_zcopy_abort(struct sk_buff *skb)
1315{
1316 struct ubuf_info *uarg = skb_zcopy(skb);
1317
1318 if (uarg) {
1319 sock_zerocopy_put_abort(uarg);
1320 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1321 }
1322}
1323
Linus Torvalds1da177e2005-04-16 15:20:36 -07001324/**
1325 * skb_queue_empty - check if a queue is empty
1326 * @list: queue head
1327 *
1328 * Returns true if the queue is empty, false otherwise.
1329 */
1330static inline int skb_queue_empty(const struct sk_buff_head *list)
1331{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001332 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001333}
1334
1335/**
David S. Millerfc7ebb22008-09-23 00:34:07 -07001336 * skb_queue_is_last - check if skb is the last entry in the queue
1337 * @list: queue head
1338 * @skb: buffer
1339 *
1340 * Returns true if @skb is the last buffer on the list.
1341 */
1342static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1343 const struct sk_buff *skb)
1344{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001345 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -07001346}
1347
1348/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001349 * skb_queue_is_first - check if skb is the first entry in the queue
1350 * @list: queue head
1351 * @skb: buffer
1352 *
1353 * Returns true if @skb is the first buffer on the list.
1354 */
1355static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1356 const struct sk_buff *skb)
1357{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001358 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001359}
1360
1361/**
David S. Miller249c8b42008-09-23 00:44:42 -07001362 * skb_queue_next - return the next packet in the queue
1363 * @list: queue head
1364 * @skb: current buffer
1365 *
1366 * Return the next packet in @list after @skb. It is only valid to
1367 * call this if skb_queue_is_last() evaluates to false.
1368 */
1369static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1370 const struct sk_buff *skb)
1371{
1372 /* This BUG_ON may seem severe, but if we just return then we
1373 * are going to dereference garbage.
1374 */
1375 BUG_ON(skb_queue_is_last(list, skb));
1376 return skb->next;
1377}
1378
1379/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001380 * skb_queue_prev - return the prev packet in the queue
1381 * @list: queue head
1382 * @skb: current buffer
1383 *
1384 * Return the prev packet in @list before @skb. It is only valid to
1385 * call this if skb_queue_is_first() evaluates to false.
1386 */
1387static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1388 const struct sk_buff *skb)
1389{
1390 /* This BUG_ON may seem severe, but if we just return then we
1391 * are going to dereference garbage.
1392 */
1393 BUG_ON(skb_queue_is_first(list, skb));
1394 return skb->prev;
1395}
1396
1397/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001398 * skb_get - reference buffer
1399 * @skb: buffer to reference
1400 *
1401 * Makes another reference to a socket buffer and returns a pointer
1402 * to the buffer.
1403 */
1404static inline struct sk_buff *skb_get(struct sk_buff *skb)
1405{
Reshetova, Elena63354792017-06-30 13:07:58 +03001406 refcount_inc(&skb->users);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001407 return skb;
1408}
1409
1410/*
1411 * If users == 1, we are the only owner and are can avoid redundant
1412 * atomic change.
1413 */
1414
1415/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001416 * skb_cloned - is the buffer a clone
1417 * @skb: buffer to check
1418 *
1419 * Returns true if the buffer was generated with skb_clone() and is
1420 * one of multiple shared copies of the buffer. Cloned buffers are
1421 * shared data so must not be written to under normal circumstances.
1422 */
1423static inline int skb_cloned(const struct sk_buff *skb)
1424{
1425 return skb->cloned &&
1426 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1427}
1428
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001429static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1430{
Mel Gormand0164ad2015-11-06 16:28:21 -08001431 might_sleep_if(gfpflags_allow_blocking(pri));
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001432
1433 if (skb_cloned(skb))
1434 return pskb_expand_head(skb, 0, 0, pri);
1435
1436 return 0;
1437}
1438
Linus Torvalds1da177e2005-04-16 15:20:36 -07001439/**
1440 * skb_header_cloned - is the header a clone
1441 * @skb: buffer to check
1442 *
1443 * Returns true if modifying the header part of the buffer requires
1444 * the data to be copied.
1445 */
1446static inline int skb_header_cloned(const struct sk_buff *skb)
1447{
1448 int dataref;
1449
1450 if (!skb->cloned)
1451 return 0;
1452
1453 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1454 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1455 return dataref != 1;
1456}
1457
Eric Dumazet9580bf22016-04-30 10:19:29 -07001458static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1459{
1460 might_sleep_if(gfpflags_allow_blocking(pri));
1461
1462 if (skb_header_cloned(skb))
1463 return pskb_expand_head(skb, 0, 0, pri);
1464
1465 return 0;
1466}
1467
Linus Torvalds1da177e2005-04-16 15:20:36 -07001468/**
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001469 * __skb_header_release - release reference to header
1470 * @skb: buffer to operate on
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001471 */
1472static inline void __skb_header_release(struct sk_buff *skb)
1473{
1474 skb->nohdr = 1;
1475 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1476}
1477
1478
1479/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001480 * skb_shared - is the buffer shared
1481 * @skb: buffer to check
1482 *
1483 * Returns true if more than one person has a reference to this
1484 * buffer.
1485 */
1486static inline int skb_shared(const struct sk_buff *skb)
1487{
Reshetova, Elena63354792017-06-30 13:07:58 +03001488 return refcount_read(&skb->users) != 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001489}
1490
1491/**
1492 * skb_share_check - check if buffer is shared and if so clone it
1493 * @skb: buffer to check
1494 * @pri: priority for memory allocation
1495 *
1496 * If the buffer is shared the buffer is cloned and the old copy
1497 * drops a reference. A new clone with a single reference is returned.
1498 * If the buffer is not shared the original buffer is returned. When
1499 * being called from interrupt status or with spinlocks held pri must
1500 * be GFP_ATOMIC.
1501 *
1502 * NULL is returned on a memory allocation failure.
1503 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001504static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001505{
Mel Gormand0164ad2015-11-06 16:28:21 -08001506 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 if (skb_shared(skb)) {
1508 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001509
1510 if (likely(nskb))
1511 consume_skb(skb);
1512 else
1513 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001514 skb = nskb;
1515 }
1516 return skb;
1517}
1518
1519/*
1520 * Copy shared buffers into a new sk_buff. We effectively do COW on
1521 * packets to handle cases where we have a local reader and forward
1522 * and a couple of other messy ones. The normal one is tcpdumping
1523 * a packet thats being forwarded.
1524 */
1525
1526/**
1527 * skb_unshare - make a copy of a shared buffer
1528 * @skb: buffer to check
1529 * @pri: priority for memory allocation
1530 *
1531 * If the socket buffer is a clone then this function creates a new
1532 * copy of the data, drops a reference count on the old copy and returns
1533 * the new copy with the reference count at 1. If the buffer is not a clone
1534 * the original buffer is returned. When called with a spinlock held or
1535 * from interrupt state @pri must be %GFP_ATOMIC
1536 *
1537 * %NULL is returned on a memory allocation failure.
1538 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001539static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001540 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001541{
Mel Gormand0164ad2015-11-06 16:28:21 -08001542 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001543 if (skb_cloned(skb)) {
1544 struct sk_buff *nskb = skb_copy(skb, pri);
Alexander Aring31eff812014-10-10 23:10:47 +02001545
1546 /* Free our shared copy */
1547 if (likely(nskb))
1548 consume_skb(skb);
1549 else
1550 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 skb = nskb;
1552 }
1553 return skb;
1554}
1555
1556/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001557 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001558 * @list_: list to peek at
1559 *
1560 * Peek an &sk_buff. Unlike most other operations you _MUST_
1561 * be careful with this one. A peek leaves the buffer on the
1562 * list and someone else may run off with it. You must hold
1563 * the appropriate locks or have a private queue to do this.
1564 *
1565 * Returns %NULL for an empty list or a pointer to the head element.
1566 * The reference count is not incremented and the reference is therefore
1567 * volatile. Use with caution.
1568 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001569static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001570{
Eric Dumazet18d07002012-04-30 16:31:46 +00001571 struct sk_buff *skb = list_->next;
1572
1573 if (skb == (struct sk_buff *)list_)
1574 skb = NULL;
1575 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001576}
1577
1578/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001579 * skb_peek_next - peek skb following the given one from a queue
1580 * @skb: skb to start from
1581 * @list_: list to peek at
1582 *
1583 * Returns %NULL when the end of the list is met or a pointer to the
1584 * next element. The reference count is not incremented and the
1585 * reference is therefore volatile. Use with caution.
1586 */
1587static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1588 const struct sk_buff_head *list_)
1589{
1590 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001591
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001592 if (next == (struct sk_buff *)list_)
1593 next = NULL;
1594 return next;
1595}
1596
1597/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001598 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001599 * @list_: list to peek at
1600 *
1601 * Peek an &sk_buff. Unlike most other operations you _MUST_
1602 * be careful with this one. A peek leaves the buffer on the
1603 * list and someone else may run off with it. You must hold
1604 * the appropriate locks or have a private queue to do this.
1605 *
1606 * Returns %NULL for an empty list or a pointer to the tail element.
1607 * The reference count is not incremented and the reference is therefore
1608 * volatile. Use with caution.
1609 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001610static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611{
Eric Dumazet18d07002012-04-30 16:31:46 +00001612 struct sk_buff *skb = list_->prev;
1613
1614 if (skb == (struct sk_buff *)list_)
1615 skb = NULL;
1616 return skb;
1617
Linus Torvalds1da177e2005-04-16 15:20:36 -07001618}
1619
1620/**
1621 * skb_queue_len - get queue length
1622 * @list_: list to measure
1623 *
1624 * Return the length of an &sk_buff queue.
1625 */
1626static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1627{
1628 return list_->qlen;
1629}
1630
David S. Miller67fed452008-09-21 22:36:24 -07001631/**
1632 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1633 * @list: queue to initialize
1634 *
1635 * This initializes only the list and queue length aspects of
1636 * an sk_buff_head object. This allows to initialize the list
1637 * aspects of an sk_buff_head without reinitializing things like
1638 * the spinlock. It can also be used for on-stack sk_buff_head
1639 * objects where the spinlock is known to not be used.
1640 */
1641static inline void __skb_queue_head_init(struct sk_buff_head *list)
1642{
1643 list->prev = list->next = (struct sk_buff *)list;
1644 list->qlen = 0;
1645}
1646
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001647/*
1648 * This function creates a split out lock class for each invocation;
1649 * this is needed for now since a whole lot of users of the skb-queue
1650 * infrastructure in drivers have different locking usage (in hardirq)
1651 * than the networking core (in softirq only). In the long run either the
1652 * network layer or drivers should need annotation to consolidate the
1653 * main types of usage into 3 classes.
1654 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001655static inline void skb_queue_head_init(struct sk_buff_head *list)
1656{
1657 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001658 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001659}
1660
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001661static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1662 struct lock_class_key *class)
1663{
1664 skb_queue_head_init(list);
1665 lockdep_set_class(&list->lock, class);
1666}
1667
Linus Torvalds1da177e2005-04-16 15:20:36 -07001668/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001669 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001670 *
1671 * The "__skb_xxxx()" functions are the non-atomic ones that
1672 * can only be called with interrupts disabled.
1673 */
Joe Perches7965bd42013-09-26 14:48:15 -07001674void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1675 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001676static inline void __skb_insert(struct sk_buff *newsk,
1677 struct sk_buff *prev, struct sk_buff *next,
1678 struct sk_buff_head *list)
1679{
1680 newsk->next = next;
1681 newsk->prev = prev;
1682 next->prev = prev->next = newsk;
1683 list->qlen++;
1684}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001685
David S. Miller67fed452008-09-21 22:36:24 -07001686static inline void __skb_queue_splice(const struct sk_buff_head *list,
1687 struct sk_buff *prev,
1688 struct sk_buff *next)
1689{
1690 struct sk_buff *first = list->next;
1691 struct sk_buff *last = list->prev;
1692
1693 first->prev = prev;
1694 prev->next = first;
1695
1696 last->next = next;
1697 next->prev = last;
1698}
1699
1700/**
1701 * skb_queue_splice - join two skb lists, this is designed for stacks
1702 * @list: the new list to add
1703 * @head: the place to add it in the first list
1704 */
1705static inline void skb_queue_splice(const struct sk_buff_head *list,
1706 struct sk_buff_head *head)
1707{
1708 if (!skb_queue_empty(list)) {
1709 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001710 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001711 }
1712}
1713
1714/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001715 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001716 * @list: the new list to add
1717 * @head: the place to add it in the first list
1718 *
1719 * The list at @list is reinitialised
1720 */
1721static inline void skb_queue_splice_init(struct sk_buff_head *list,
1722 struct sk_buff_head *head)
1723{
1724 if (!skb_queue_empty(list)) {
1725 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001726 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001727 __skb_queue_head_init(list);
1728 }
1729}
1730
1731/**
1732 * skb_queue_splice_tail - join two skb lists, each list being a queue
1733 * @list: the new list to add
1734 * @head: the place to add it in the first list
1735 */
1736static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1737 struct sk_buff_head *head)
1738{
1739 if (!skb_queue_empty(list)) {
1740 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001741 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001742 }
1743}
1744
1745/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001746 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001747 * @list: the new list to add
1748 * @head: the place to add it in the first list
1749 *
1750 * Each of the lists is a queue.
1751 * The list at @list is reinitialised
1752 */
1753static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1754 struct sk_buff_head *head)
1755{
1756 if (!skb_queue_empty(list)) {
1757 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001758 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001759 __skb_queue_head_init(list);
1760 }
1761}
1762
Linus Torvalds1da177e2005-04-16 15:20:36 -07001763/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001764 * __skb_queue_after - queue a buffer at the list head
1765 * @list: list to use
1766 * @prev: place after this buffer
1767 * @newsk: buffer to queue
1768 *
1769 * Queue a buffer int the middle of a list. This function takes no locks
1770 * and you must therefore hold required locks before calling it.
1771 *
1772 * A buffer cannot be placed on two lists at the same time.
1773 */
1774static inline void __skb_queue_after(struct sk_buff_head *list,
1775 struct sk_buff *prev,
1776 struct sk_buff *newsk)
1777{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001778 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001779}
1780
Joe Perches7965bd42013-09-26 14:48:15 -07001781void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1782 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001783
Gerrit Renkerf5572852008-04-14 00:05:28 -07001784static inline void __skb_queue_before(struct sk_buff_head *list,
1785 struct sk_buff *next,
1786 struct sk_buff *newsk)
1787{
1788 __skb_insert(newsk, next->prev, next, list);
1789}
1790
Stephen Hemminger300ce172005-10-30 13:47:34 -08001791/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001792 * __skb_queue_head - queue a buffer at the list head
1793 * @list: list to use
1794 * @newsk: buffer to queue
1795 *
1796 * Queue a buffer at the start of a list. This function takes no locks
1797 * and you must therefore hold required locks before calling it.
1798 *
1799 * A buffer cannot be placed on two lists at the same time.
1800 */
Joe Perches7965bd42013-09-26 14:48:15 -07001801void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001802static inline void __skb_queue_head(struct sk_buff_head *list,
1803 struct sk_buff *newsk)
1804{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001805 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001806}
1807
1808/**
1809 * __skb_queue_tail - queue a buffer at the list tail
1810 * @list: list to use
1811 * @newsk: buffer to queue
1812 *
1813 * Queue a buffer at the end of a list. This function takes no locks
1814 * and you must therefore hold required locks before calling it.
1815 *
1816 * A buffer cannot be placed on two lists at the same time.
1817 */
Joe Perches7965bd42013-09-26 14:48:15 -07001818void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819static inline void __skb_queue_tail(struct sk_buff_head *list,
1820 struct sk_buff *newsk)
1821{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001822 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823}
1824
Linus Torvalds1da177e2005-04-16 15:20:36 -07001825/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001826 * remove sk_buff from list. _Must_ be called atomically, and with
1827 * the list known..
1828 */
Joe Perches7965bd42013-09-26 14:48:15 -07001829void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001830static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1831{
1832 struct sk_buff *next, *prev;
1833
1834 list->qlen--;
1835 next = skb->next;
1836 prev = skb->prev;
1837 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001838 next->prev = prev;
1839 prev->next = next;
1840}
1841
Gerrit Renkerf525c062008-04-14 00:04:12 -07001842/**
1843 * __skb_dequeue - remove from the head of the queue
1844 * @list: list to dequeue from
1845 *
1846 * Remove the head of the list. This function does not take any locks
1847 * so must be used with appropriate locks held only. The head item is
1848 * returned or %NULL if the list is empty.
1849 */
Joe Perches7965bd42013-09-26 14:48:15 -07001850struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001851static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1852{
1853 struct sk_buff *skb = skb_peek(list);
1854 if (skb)
1855 __skb_unlink(skb, list);
1856 return skb;
1857}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001858
1859/**
1860 * __skb_dequeue_tail - remove from the tail of the queue
1861 * @list: list to dequeue from
1862 *
1863 * Remove the tail of the list. This function does not take any locks
1864 * so must be used with appropriate locks held only. The tail item is
1865 * returned or %NULL if the list is empty.
1866 */
Joe Perches7965bd42013-09-26 14:48:15 -07001867struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001868static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1869{
1870 struct sk_buff *skb = skb_peek_tail(list);
1871 if (skb)
1872 __skb_unlink(skb, list);
1873 return skb;
1874}
1875
1876
David S. Millerbdcc0922012-03-07 20:53:36 -05001877static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001878{
1879 return skb->data_len;
1880}
1881
1882static inline unsigned int skb_headlen(const struct sk_buff *skb)
1883{
1884 return skb->len - skb->data_len;
1885}
1886
Willem de Bruijn3ece7822017-08-03 16:29:38 -04001887static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001888{
Alexey Dobriyanc72d8cd2016-11-19 04:08:08 +03001889 unsigned int i, len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001890
Alexey Dobriyanc72d8cd2016-11-19 04:08:08 +03001891 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001892 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Willem de Bruijn3ece7822017-08-03 16:29:38 -04001893 return len;
1894}
1895
1896static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1897{
1898 return skb_headlen(skb) + __skb_pagelen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001899}
1900
Ian Campbell131ea662011-08-19 06:25:00 +00001901/**
1902 * __skb_fill_page_desc - initialise a paged fragment in an skb
1903 * @skb: buffer containing fragment to be initialised
1904 * @i: paged fragment index to initialise
1905 * @page: the page to use for this fragment
1906 * @off: the offset to the data with @page
1907 * @size: the length of the data
1908 *
1909 * Initialises the @i'th fragment of @skb to point to &size bytes at
1910 * offset @off within @page.
1911 *
1912 * Does not take any additional reference on the fragment.
1913 */
1914static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1915 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001916{
1917 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1918
Mel Gormanc48a11c2012-07-31 16:44:23 -07001919 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001920 * Propagate page pfmemalloc to the skb if we can. The problem is
1921 * that not all callers have unique ownership of the page but rely
1922 * on page_is_pfmemalloc doing the right thing(tm).
Mel Gormanc48a11c2012-07-31 16:44:23 -07001923 */
Ian Campbella8605c62011-10-19 23:01:49 +00001924 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001925 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001926 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001927
1928 page = compound_head(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001929 if (page_is_pfmemalloc(page))
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001930 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001931}
1932
1933/**
1934 * skb_fill_page_desc - initialise a paged fragment in an skb
1935 * @skb: buffer containing fragment to be initialised
1936 * @i: paged fragment index to initialise
1937 * @page: the page to use for this fragment
1938 * @off: the offset to the data with @page
1939 * @size: the length of the data
1940 *
1941 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001942 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001943 * addition updates @skb such that @i is the last fragment.
1944 *
1945 * Does not take any additional reference on the fragment.
1946 */
1947static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1948 struct page *page, int off, int size)
1949{
1950 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001951 skb_shinfo(skb)->nr_frags = i + 1;
1952}
1953
Joe Perches7965bd42013-09-26 14:48:15 -07001954void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1955 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001956
Jason Wangf8e617e2013-11-01 14:07:47 +08001957void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1958 unsigned int truesize);
1959
Linus Torvalds1da177e2005-04-16 15:20:36 -07001960#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001961#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001962#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1963
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001964#ifdef NET_SKBUFF_DATA_USES_OFFSET
1965static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1966{
1967 return skb->head + skb->tail;
1968}
1969
1970static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1971{
1972 skb->tail = skb->data - skb->head;
1973}
1974
1975static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1976{
1977 skb_reset_tail_pointer(skb);
1978 skb->tail += offset;
1979}
Simon Horman7cc46192013-05-28 20:34:29 +00001980
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001981#else /* NET_SKBUFF_DATA_USES_OFFSET */
1982static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1983{
1984 return skb->tail;
1985}
1986
1987static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1988{
1989 skb->tail = skb->data;
1990}
1991
1992static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1993{
1994 skb->tail = skb->data + offset;
1995}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001996
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001997#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1998
Linus Torvalds1da177e2005-04-16 15:20:36 -07001999/*
2000 * Add data to an sk_buff
2001 */
Johannes Berg4df864c2017-06-16 14:29:21 +02002002void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2003void *skb_put(struct sk_buff *skb, unsigned int len);
2004static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002005{
Johannes Berg4df864c2017-06-16 14:29:21 +02002006 void *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002007 SKB_LINEAR_ASSERT(skb);
2008 skb->tail += len;
2009 skb->len += len;
2010 return tmp;
2011}
2012
yuan linyude77b962017-06-18 22:48:17 +08002013static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2014{
2015 void *tmp = __skb_put(skb, len);
2016
2017 memset(tmp, 0, len);
2018 return tmp;
2019}
2020
2021static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2022 unsigned int len)
2023{
2024 void *tmp = __skb_put(skb, len);
2025
2026 memcpy(tmp, data, len);
2027 return tmp;
2028}
2029
2030static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2031{
2032 *(u8 *)__skb_put(skb, 1) = val;
2033}
2034
Johannes Berg83ad3572017-06-14 22:17:20 +02002035static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
Johannes Berge45a79d2017-05-24 09:07:47 +02002036{
Johannes Berg83ad3572017-06-14 22:17:20 +02002037 void *tmp = skb_put(skb, len);
Johannes Berge45a79d2017-05-24 09:07:47 +02002038
2039 memset(tmp, 0, len);
2040
2041 return tmp;
2042}
2043
Johannes Berg59ae1d12017-06-16 14:29:20 +02002044static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2045 unsigned int len)
2046{
2047 void *tmp = skb_put(skb, len);
2048
2049 memcpy(tmp, data, len);
2050
2051 return tmp;
2052}
2053
Johannes Berg634fef62017-06-16 14:29:24 +02002054static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2055{
2056 *(u8 *)skb_put(skb, 1) = val;
2057}
2058
Johannes Bergd58ff352017-06-16 14:29:23 +02002059void *skb_push(struct sk_buff *skb, unsigned int len);
2060static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002061{
2062 skb->data -= len;
2063 skb->len += len;
2064 return skb->data;
2065}
2066
Johannes Bergaf728682017-06-16 14:29:22 +02002067void *skb_pull(struct sk_buff *skb, unsigned int len);
2068static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002069{
2070 skb->len -= len;
2071 BUG_ON(skb->len < skb->data_len);
2072 return skb->data += len;
2073}
2074
Johannes Bergaf728682017-06-16 14:29:22 +02002075static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
David S. Miller47d29642010-05-02 02:21:44 -07002076{
2077 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2078}
2079
Johannes Bergaf728682017-06-16 14:29:22 +02002080void *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002081
Johannes Bergaf728682017-06-16 14:29:22 +02002082static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002083{
2084 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07002085 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002086 return NULL;
2087 skb->len -= len;
2088 return skb->data += len;
2089}
2090
Johannes Bergaf728682017-06-16 14:29:22 +02002091static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002092{
2093 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2094}
2095
2096static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2097{
2098 if (likely(len <= skb_headlen(skb)))
2099 return 1;
2100 if (unlikely(len > skb->len))
2101 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002102 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002103}
2104
Eric Dumazetc8c8b122016-12-07 09:19:33 -08002105void skb_condense(struct sk_buff *skb);
2106
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107/**
2108 * skb_headroom - bytes at buffer head
2109 * @skb: buffer to check
2110 *
2111 * Return the number of bytes of free space at the head of an &sk_buff.
2112 */
Chuck Leverc2636b42007-10-23 21:07:32 -07002113static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002114{
2115 return skb->data - skb->head;
2116}
2117
2118/**
2119 * skb_tailroom - bytes at buffer end
2120 * @skb: buffer to check
2121 *
2122 * Return the number of bytes of free space at the tail of an sk_buff
2123 */
2124static inline int skb_tailroom(const struct sk_buff *skb)
2125{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07002126 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002127}
2128
2129/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00002130 * skb_availroom - bytes at buffer end
2131 * @skb: buffer to check
2132 *
2133 * Return the number of bytes of free space at the tail of an sk_buff
2134 * allocated by sk_stream_alloc()
2135 */
2136static inline int skb_availroom(const struct sk_buff *skb)
2137{
Eric Dumazet16fad692013-03-14 05:40:32 +00002138 if (skb_is_nonlinear(skb))
2139 return 0;
2140
2141 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00002142}
2143
2144/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002145 * skb_reserve - adjust headroom
2146 * @skb: buffer to alter
2147 * @len: bytes to move
2148 *
2149 * Increase the headroom of an empty &sk_buff by reducing the tail
2150 * room. This is only allowed for an empty buffer.
2151 */
David S. Miller8243126c2006-01-17 02:54:21 -08002152static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002153{
2154 skb->data += len;
2155 skb->tail += len;
2156}
2157
Benjamin Poirier1837b2e2016-02-29 15:03:33 -08002158/**
2159 * skb_tailroom_reserve - adjust reserved_tailroom
2160 * @skb: buffer to alter
2161 * @mtu: maximum amount of headlen permitted
2162 * @needed_tailroom: minimum amount of reserved_tailroom
2163 *
2164 * Set reserved_tailroom so that headlen can be as large as possible but
2165 * not larger than mtu and tailroom cannot be smaller than
2166 * needed_tailroom.
2167 * The required headroom should already have been reserved before using
2168 * this function.
2169 */
2170static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2171 unsigned int needed_tailroom)
2172{
2173 SKB_LINEAR_ASSERT(skb);
2174 if (mtu < skb_tailroom(skb) - needed_tailroom)
2175 /* use at most mtu */
2176 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2177 else
2178 /* use up to all available space */
2179 skb->reserved_tailroom = needed_tailroom;
2180}
2181
Tom Herbert8bce6d72014-09-29 20:22:29 -07002182#define ENCAP_TYPE_ETHER 0
2183#define ENCAP_TYPE_IPPROTO 1
2184
2185static inline void skb_set_inner_protocol(struct sk_buff *skb,
2186 __be16 protocol)
2187{
2188 skb->inner_protocol = protocol;
2189 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2190}
2191
2192static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2193 __u8 ipproto)
2194{
2195 skb->inner_ipproto = ipproto;
2196 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2197}
2198
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002199static inline void skb_reset_inner_headers(struct sk_buff *skb)
2200{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002201 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002202 skb->inner_network_header = skb->network_header;
2203 skb->inner_transport_header = skb->transport_header;
2204}
2205
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00002206static inline void skb_reset_mac_len(struct sk_buff *skb)
2207{
2208 skb->mac_len = skb->network_header - skb->mac_header;
2209}
2210
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002211static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2212 *skb)
2213{
2214 return skb->head + skb->inner_transport_header;
2215}
2216
Tom Herbert55dc5a92015-12-14 11:19:40 -08002217static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2218{
2219 return skb_inner_transport_header(skb) - skb->data;
2220}
2221
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002222static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2223{
2224 skb->inner_transport_header = skb->data - skb->head;
2225}
2226
2227static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2228 const int offset)
2229{
2230 skb_reset_inner_transport_header(skb);
2231 skb->inner_transport_header += offset;
2232}
2233
2234static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2235{
2236 return skb->head + skb->inner_network_header;
2237}
2238
2239static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2240{
2241 skb->inner_network_header = skb->data - skb->head;
2242}
2243
2244static inline void skb_set_inner_network_header(struct sk_buff *skb,
2245 const int offset)
2246{
2247 skb_reset_inner_network_header(skb);
2248 skb->inner_network_header += offset;
2249}
2250
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002251static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2252{
2253 return skb->head + skb->inner_mac_header;
2254}
2255
2256static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2257{
2258 skb->inner_mac_header = skb->data - skb->head;
2259}
2260
2261static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2262 const int offset)
2263{
2264 skb_reset_inner_mac_header(skb);
2265 skb->inner_mac_header += offset;
2266}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002267static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2268{
Cong Wang35d04612013-05-29 15:16:05 +08002269 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002270}
2271
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002272static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2273{
2274 return skb->head + skb->transport_header;
2275}
2276
2277static inline void skb_reset_transport_header(struct sk_buff *skb)
2278{
2279 skb->transport_header = skb->data - skb->head;
2280}
2281
2282static inline void skb_set_transport_header(struct sk_buff *skb,
2283 const int offset)
2284{
2285 skb_reset_transport_header(skb);
2286 skb->transport_header += offset;
2287}
2288
2289static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2290{
2291 return skb->head + skb->network_header;
2292}
2293
2294static inline void skb_reset_network_header(struct sk_buff *skb)
2295{
2296 skb->network_header = skb->data - skb->head;
2297}
2298
2299static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2300{
2301 skb_reset_network_header(skb);
2302 skb->network_header += offset;
2303}
2304
2305static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2306{
2307 return skb->head + skb->mac_header;
2308}
2309
Amir Vadaiea6da4f2017-02-07 09:56:06 +02002310static inline int skb_mac_offset(const struct sk_buff *skb)
2311{
2312 return skb_mac_header(skb) - skb->data;
2313}
2314
Daniel Borkmann0daf4342017-07-02 02:13:25 +02002315static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2316{
2317 return skb->network_header - skb->mac_header;
2318}
2319
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002320static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2321{
Cong Wang35d04612013-05-29 15:16:05 +08002322 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002323}
2324
2325static inline void skb_reset_mac_header(struct sk_buff *skb)
2326{
2327 skb->mac_header = skb->data - skb->head;
2328}
2329
2330static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2331{
2332 skb_reset_mac_header(skb);
2333 skb->mac_header += offset;
2334}
2335
Timo Teräs0e3da5b2013-12-16 11:02:09 +02002336static inline void skb_pop_mac_header(struct sk_buff *skb)
2337{
2338 skb->mac_header = skb->network_header;
2339}
2340
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002341static inline void skb_probe_transport_header(struct sk_buff *skb,
2342 const int offset_hint)
2343{
2344 struct flow_keys keys;
2345
2346 if (skb_transport_header_was_set(skb))
2347 return;
Tom Herbertcd79a232015-09-01 09:24:27 -07002348 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
Tom Herbert42aecaa2015-06-04 09:16:39 -07002349 skb_set_transport_header(skb, keys.control.thoff);
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002350 else
2351 skb_set_transport_header(skb, offset_hint);
2352}
2353
Eric Dumazet03606892012-02-23 10:55:02 +00002354static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2355{
2356 if (skb_mac_header_was_set(skb)) {
2357 const unsigned char *old_mac = skb_mac_header(skb);
2358
2359 skb_set_mac_header(skb, -skb->mac_len);
2360 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2361 }
2362}
2363
Michał Mirosław04fb4512010-12-14 15:24:08 +00002364static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2365{
2366 return skb->csum_start - skb_headroom(skb);
2367}
2368
Alexander Duyck08b64fc2016-02-05 15:27:49 -08002369static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2370{
2371 return skb->head + skb->csum_start;
2372}
2373
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002374static inline int skb_transport_offset(const struct sk_buff *skb)
2375{
2376 return skb_transport_header(skb) - skb->data;
2377}
2378
2379static inline u32 skb_network_header_len(const struct sk_buff *skb)
2380{
2381 return skb->transport_header - skb->network_header;
2382}
2383
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002384static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2385{
2386 return skb->inner_transport_header - skb->inner_network_header;
2387}
2388
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002389static inline int skb_network_offset(const struct sk_buff *skb)
2390{
2391 return skb_network_header(skb) - skb->data;
2392}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03002393
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002394static inline int skb_inner_network_offset(const struct sk_buff *skb)
2395{
2396 return skb_inner_network_header(skb) - skb->data;
2397}
2398
Changli Gaof9599ce2010-08-04 04:43:44 +00002399static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2400{
2401 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2402}
2403
Linus Torvalds1da177e2005-04-16 15:20:36 -07002404/*
2405 * CPUs often take a performance hit when accessing unaligned memory
2406 * locations. The actual performance hit varies, it can be small if the
2407 * hardware handles it or large if we have to take an exception and fix it
2408 * in software.
2409 *
2410 * Since an ethernet header is 14 bytes network drivers often end up with
2411 * the IP header at an unaligned offset. The IP header can be aligned by
2412 * shifting the start of the packet by 2 bytes. Drivers should do this
2413 * with:
2414 *
Tobias Klauser8660c122009-07-13 22:48:16 +00002415 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002416 *
2417 * The downside to this alignment of the IP header is that the DMA is now
2418 * unaligned. On some architectures the cost of an unaligned DMA is high
2419 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00002420 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002421 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2422 * to be overridden.
2423 */
2424#ifndef NET_IP_ALIGN
2425#define NET_IP_ALIGN 2
2426#endif
2427
Anton Blanchard025be812006-03-31 02:27:06 -08002428/*
2429 * The networking layer reserves some headroom in skb data (via
2430 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2431 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08002432 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08002433 *
2434 * Unfortunately this headroom changes the DMA alignment of the resulting
2435 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2436 * on some architectures. An architecture can override this value,
2437 * perhaps setting it to a cacheline in size (since that will maintain
2438 * cacheline alignment of the DMA). It must be a power of 2.
2439 *
David S. Millerd6301d32009-02-08 19:24:13 -08002440 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08002441 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07002442 *
2443 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2444 * to reduce average number of cache lines per packet.
2445 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07002446 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08002447 */
2448#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07002449#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08002450#endif
2451
Joe Perches7965bd42013-09-26 14:48:15 -07002452int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002453
Daniel Borkmann5293efe2016-08-18 01:00:39 +02002454static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00002456 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07002457 WARN_ON(1);
2458 return;
2459 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002460 skb->len = len;
2461 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002462}
2463
Daniel Borkmann5293efe2016-08-18 01:00:39 +02002464static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2465{
2466 __skb_set_length(skb, len);
2467}
2468
Joe Perches7965bd42013-09-26 14:48:15 -07002469void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470
2471static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2472{
Herbert Xu3cc0e872006-06-09 16:13:38 -07002473 if (skb->data_len)
2474 return ___pskb_trim(skb, len);
2475 __skb_trim(skb, len);
2476 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002477}
2478
2479static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2480{
2481 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2482}
2483
2484/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07002485 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2486 * @skb: buffer to alter
2487 * @len: new length
2488 *
2489 * This is identical to pskb_trim except that the caller knows that
2490 * the skb is not cloned so we should never get an error due to out-
2491 * of-memory.
2492 */
2493static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2494{
2495 int err = pskb_trim(skb, len);
2496 BUG_ON(err);
2497}
2498
Daniel Borkmann5293efe2016-08-18 01:00:39 +02002499static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2500{
2501 unsigned int diff = len - skb->len;
2502
2503 if (skb_tailroom(skb) < diff) {
2504 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2505 GFP_ATOMIC);
2506 if (ret)
2507 return ret;
2508 }
2509 __skb_set_length(skb, len);
2510 return 0;
2511}
2512
Herbert Xue9fa4f72006-08-13 20:12:58 -07002513/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002514 * skb_orphan - orphan a buffer
2515 * @skb: buffer to orphan
2516 *
2517 * If a buffer currently has an owner then we call the owner's
2518 * destructor function and make the @skb unowned. The buffer continues
2519 * to exist but is no longer charged to its former owner.
2520 */
2521static inline void skb_orphan(struct sk_buff *skb)
2522{
Eric Dumazetc34a7612013-07-30 16:11:15 -07002523 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002524 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002525 skb->destructor = NULL;
2526 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07002527 } else {
2528 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002529 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002530}
2531
2532/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002533 * skb_orphan_frags - orphan the frags contained in a buffer
2534 * @skb: buffer to orphan frags from
2535 * @gfp_mask: allocation mask for replacement pages
2536 *
2537 * For each frag in the SKB which needs a destructor (i.e. has an
2538 * owner) create a copy of that frag and release the original
2539 * page by calling the destructor.
2540 */
2541static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2542{
Willem de Bruijn1f8b9772017-08-03 16:29:41 -04002543 if (likely(!skb_zcopy(skb)))
2544 return 0;
2545 if (skb_uarg(skb)->callback == sock_zerocopy_callback)
2546 return 0;
2547 return skb_copy_ubufs(skb, gfp_mask);
2548}
2549
2550/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2551static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2552{
2553 if (likely(!skb_zcopy(skb)))
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002554 return 0;
2555 return skb_copy_ubufs(skb, gfp_mask);
2556}
2557
2558/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002559 * __skb_queue_purge - empty a list
2560 * @list: list to empty
2561 *
2562 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2563 * the list and one reference dropped. This function does not take the
2564 * list lock and the caller must hold the relevant locks to use it.
2565 */
Joe Perches7965bd42013-09-26 14:48:15 -07002566void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002567static inline void __skb_queue_purge(struct sk_buff_head *list)
2568{
2569 struct sk_buff *skb;
2570 while ((skb = __skb_dequeue(list)) != NULL)
2571 kfree_skb(skb);
2572}
2573
Yaogong Wang9f5afea2016-09-07 14:49:28 -07002574void skb_rbtree_purge(struct rb_root *root);
2575
Joe Perches7965bd42013-09-26 14:48:15 -07002576void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002577
Joe Perches7965bd42013-09-26 14:48:15 -07002578struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2579 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002580
2581/**
2582 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2583 * @dev: network device to receive on
2584 * @length: length to allocate
2585 *
2586 * Allocate a new &sk_buff and assign it a usage count of one. The
2587 * buffer has unspecified headroom built in. Users should allocate
2588 * the headroom they think they need without accounting for the
2589 * built in space. The built in space is used for optimisations.
2590 *
2591 * %NULL is returned if there is no free memory. Although this function
2592 * allocates memory it can be called from an interrupt.
2593 */
2594static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002595 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002596{
2597 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2598}
2599
Eric Dumazet6f532612012-05-18 05:12:12 +00002600/* legacy helper around __netdev_alloc_skb() */
2601static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2602 gfp_t gfp_mask)
2603{
2604 return __netdev_alloc_skb(NULL, length, gfp_mask);
2605}
2606
2607/* legacy helper around netdev_alloc_skb() */
2608static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2609{
2610 return netdev_alloc_skb(NULL, length);
2611}
2612
2613
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002614static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2615 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002616{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002617 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002618
2619 if (NET_IP_ALIGN && skb)
2620 skb_reserve(skb, NET_IP_ALIGN);
2621 return skb;
2622}
2623
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002624static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2625 unsigned int length)
2626{
2627 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2628}
2629
Alexander Duyck181edb22015-05-06 21:12:03 -07002630static inline void skb_free_frag(void *addr)
2631{
Alexander Duyck8c2dd3e2017-01-10 16:58:06 -08002632 page_frag_free(addr);
Alexander Duyck181edb22015-05-06 21:12:03 -07002633}
2634
Alexander Duyckffde7322014-12-09 19:40:42 -08002635void *napi_alloc_frag(unsigned int fragsz);
Alexander Duyckfd11a832014-12-09 19:40:49 -08002636struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2637 unsigned int length, gfp_t gfp_mask);
2638static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2639 unsigned int length)
2640{
2641 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2642}
Jesper Dangaard Brouer795bb1c2016-02-08 13:14:59 +01002643void napi_consume_skb(struct sk_buff *skb, int budget);
2644
2645void __kfree_skb_flush(void);
Jesper Dangaard Brouer15fad712016-02-08 13:15:04 +01002646void __kfree_skb_defer(struct sk_buff *skb);
Alexander Duyckffde7322014-12-09 19:40:42 -08002647
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002648/**
Alexander Duyck71dfda52014-11-11 09:26:34 -08002649 * __dev_alloc_pages - allocate page for network Rx
2650 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2651 * @order: size of the allocation
2652 *
2653 * Allocate a new page.
2654 *
2655 * %NULL is returned if there is no free memory.
2656*/
2657static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2658 unsigned int order)
2659{
2660 /* This piece of code contains several assumptions.
2661 * 1. This is for device Rx, therefor a cold page is preferred.
2662 * 2. The expectation is the user wants a compound page.
2663 * 3. If requesting a order 0 page it will not be compound
2664 * due to the check to see if order has a value in prep_new_page
2665 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2666 * code in gfp_to_alloc_flags that should be enforcing this.
2667 */
2668 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2669
2670 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2671}
2672
2673static inline struct page *dev_alloc_pages(unsigned int order)
2674{
Neil Horman95829b32016-05-19 11:30:54 -04002675 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
Alexander Duyck71dfda52014-11-11 09:26:34 -08002676}
2677
2678/**
2679 * __dev_alloc_page - allocate a page for network Rx
2680 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2681 *
2682 * Allocate a new page.
2683 *
2684 * %NULL is returned if there is no free memory.
2685 */
2686static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2687{
2688 return __dev_alloc_pages(gfp_mask, 0);
2689}
2690
2691static inline struct page *dev_alloc_page(void)
2692{
Neil Horman95829b32016-05-19 11:30:54 -04002693 return dev_alloc_pages(0);
Alexander Duyck71dfda52014-11-11 09:26:34 -08002694}
2695
2696/**
Mel Gorman06140022012-07-31 16:44:24 -07002697 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2698 * @page: The page that was allocated from skb_alloc_page
2699 * @skb: The skb that may need pfmemalloc set
2700 */
2701static inline void skb_propagate_pfmemalloc(struct page *page,
2702 struct sk_buff *skb)
2703{
Michal Hocko2f064f32015-08-21 14:11:51 -07002704 if (page_is_pfmemalloc(page))
Mel Gorman06140022012-07-31 16:44:24 -07002705 skb->pfmemalloc = true;
2706}
2707
Eric Dumazet564824b2010-10-11 19:05:25 +00002708/**
Masanari Iidae2278672014-02-18 22:54:36 +09002709 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002710 * @frag: the paged fragment
2711 *
2712 * Returns the &struct page associated with @frag.
2713 */
2714static inline struct page *skb_frag_page(const skb_frag_t *frag)
2715{
Ian Campbella8605c62011-10-19 23:01:49 +00002716 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002717}
2718
2719/**
2720 * __skb_frag_ref - take an addition reference on a paged fragment.
2721 * @frag: the paged fragment
2722 *
2723 * Takes an additional reference on the paged fragment @frag.
2724 */
2725static inline void __skb_frag_ref(skb_frag_t *frag)
2726{
2727 get_page(skb_frag_page(frag));
2728}
2729
2730/**
2731 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2732 * @skb: the buffer
2733 * @f: the fragment offset.
2734 *
2735 * Takes an additional reference on the @f'th paged fragment of @skb.
2736 */
2737static inline void skb_frag_ref(struct sk_buff *skb, int f)
2738{
2739 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2740}
2741
2742/**
2743 * __skb_frag_unref - release a reference on a paged fragment.
2744 * @frag: the paged fragment
2745 *
2746 * Releases a reference on the paged fragment @frag.
2747 */
2748static inline void __skb_frag_unref(skb_frag_t *frag)
2749{
2750 put_page(skb_frag_page(frag));
2751}
2752
2753/**
2754 * skb_frag_unref - release a reference on a paged fragment of an skb.
2755 * @skb: the buffer
2756 * @f: the fragment offset
2757 *
2758 * Releases a reference on the @f'th paged fragment of @skb.
2759 */
2760static inline void skb_frag_unref(struct sk_buff *skb, int f)
2761{
2762 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2763}
2764
2765/**
2766 * skb_frag_address - gets the address of the data contained in a paged fragment
2767 * @frag: the paged fragment buffer
2768 *
2769 * Returns the address of the data within @frag. The page must already
2770 * be mapped.
2771 */
2772static inline void *skb_frag_address(const skb_frag_t *frag)
2773{
2774 return page_address(skb_frag_page(frag)) + frag->page_offset;
2775}
2776
2777/**
2778 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2779 * @frag: the paged fragment buffer
2780 *
2781 * Returns the address of the data within @frag. Checks that the page
2782 * is mapped and returns %NULL otherwise.
2783 */
2784static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2785{
2786 void *ptr = page_address(skb_frag_page(frag));
2787 if (unlikely(!ptr))
2788 return NULL;
2789
2790 return ptr + frag->page_offset;
2791}
2792
2793/**
2794 * __skb_frag_set_page - sets the page contained in a paged fragment
2795 * @frag: the paged fragment
2796 * @page: the page to set
2797 *
2798 * Sets the fragment @frag to contain @page.
2799 */
2800static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2801{
Ian Campbella8605c62011-10-19 23:01:49 +00002802 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002803}
2804
2805/**
2806 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2807 * @skb: the buffer
2808 * @f: the fragment offset
2809 * @page: the page to set
2810 *
2811 * Sets the @f'th fragment of @skb to contain @page.
2812 */
2813static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2814 struct page *page)
2815{
2816 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2817}
2818
Eric Dumazet400dfd32013-10-17 16:27:07 -07002819bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2820
Ian Campbell131ea662011-08-19 06:25:00 +00002821/**
2822 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002823 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002824 * @frag: the paged fragment to map
2825 * @offset: the offset within the fragment (starting at the
2826 * fragment's own offset)
2827 * @size: the number of bytes to map
Mauro Carvalho Chehab771b00a2017-05-12 09:19:29 -03002828 * @dir: the direction of the mapping (``PCI_DMA_*``)
Ian Campbell131ea662011-08-19 06:25:00 +00002829 *
2830 * Maps the page associated with @frag to @device.
2831 */
2832static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2833 const skb_frag_t *frag,
2834 size_t offset, size_t size,
2835 enum dma_data_direction dir)
2836{
2837 return dma_map_page(dev, skb_frag_page(frag),
2838 frag->page_offset + offset, size, dir);
2839}
2840
Eric Dumazet117632e2011-12-03 21:39:53 +00002841static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2842 gfp_t gfp_mask)
2843{
2844 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2845}
2846
Octavian Purdilabad93e92014-06-12 01:36:26 +03002847
2848static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2849 gfp_t gfp_mask)
2850{
2851 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2852}
2853
2854
Ian Campbell131ea662011-08-19 06:25:00 +00002855/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002856 * skb_clone_writable - is the header of a clone writable
2857 * @skb: buffer to check
2858 * @len: length up to which to write
2859 *
2860 * Returns true if modifying the header part of the cloned buffer
2861 * does not requires the data to be copied.
2862 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002863static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002864{
2865 return !skb_header_cloned(skb) &&
2866 skb_headroom(skb) + len <= skb->hdr_len;
2867}
2868
Daniel Borkmann36976492016-02-19 23:05:25 +01002869static inline int skb_try_make_writable(struct sk_buff *skb,
2870 unsigned int write_len)
2871{
2872 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2873 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2874}
2875
Herbert Xud9cc2042007-09-16 16:21:16 -07002876static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2877 int cloned)
2878{
2879 int delta = 0;
2880
Herbert Xud9cc2042007-09-16 16:21:16 -07002881 if (headroom > skb_headroom(skb))
2882 delta = headroom - skb_headroom(skb);
2883
2884 if (delta || cloned)
2885 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2886 GFP_ATOMIC);
2887 return 0;
2888}
2889
Patrick McHardy334a8132007-06-25 04:35:20 -07002890/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002891 * skb_cow - copy header of skb when it is required
2892 * @skb: buffer to cow
2893 * @headroom: needed headroom
2894 *
2895 * If the skb passed lacks sufficient headroom or its data part
2896 * is shared, data is reallocated. If reallocation fails, an error
2897 * is returned and original skb is not changed.
2898 *
2899 * The result is skb with writable area skb->head...skb->tail
2900 * and at least @headroom of space at head.
2901 */
2902static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2903{
Herbert Xud9cc2042007-09-16 16:21:16 -07002904 return __skb_cow(skb, headroom, skb_cloned(skb));
2905}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002906
Herbert Xud9cc2042007-09-16 16:21:16 -07002907/**
2908 * skb_cow_head - skb_cow but only making the head writable
2909 * @skb: buffer to cow
2910 * @headroom: needed headroom
2911 *
2912 * This function is identical to skb_cow except that we replace the
2913 * skb_cloned check by skb_header_cloned. It should be used when
2914 * you only need to push on some header and do not need to modify
2915 * the data.
2916 */
2917static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2918{
2919 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002920}
2921
2922/**
2923 * skb_padto - pad an skbuff up to a minimal size
2924 * @skb: buffer to pad
2925 * @len: minimal length
2926 *
2927 * Pads up a buffer to ensure the trailing bytes exist and are
2928 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002929 * is untouched. Otherwise it is extended. Returns zero on
2930 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931 */
Herbert Xu5b057c62006-06-23 02:06:41 -07002932static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002933{
2934 unsigned int size = skb->len;
2935 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002936 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002937 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002938}
2939
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002940/**
2941 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2942 * @skb: buffer to pad
2943 * @len: minimal length
Florian Fainellicd0a1372017-08-22 15:12:14 -07002944 * @free_on_error: free buffer on error
2945 *
2946 * Pads up a buffer to ensure the trailing bytes exist and are
2947 * blanked. If the buffer already contains sufficient data it
2948 * is untouched. Otherwise it is extended. Returns zero on
2949 * success. The skb is freed on error if @free_on_error is true.
2950 */
2951static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
2952 bool free_on_error)
2953{
2954 unsigned int size = skb->len;
2955
2956 if (unlikely(size < len)) {
2957 len -= size;
2958 if (__skb_pad(skb, len, free_on_error))
2959 return -ENOMEM;
2960 __skb_put(skb, len);
2961 }
2962 return 0;
2963}
2964
2965/**
2966 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2967 * @skb: buffer to pad
2968 * @len: minimal length
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002969 *
2970 * Pads up a buffer to ensure the trailing bytes exist and are
2971 * blanked. If the buffer already contains sufficient data it
2972 * is untouched. Otherwise it is extended. Returns zero on
2973 * success. The skb is freed on error.
2974 */
2975static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2976{
Florian Fainellicd0a1372017-08-22 15:12:14 -07002977 return __skb_put_padto(skb, len, true);
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002978}
2979
Linus Torvalds1da177e2005-04-16 15:20:36 -07002980static inline int skb_add_data(struct sk_buff *skb,
Al Viroaf2b0402014-11-27 21:44:24 -05002981 struct iov_iter *from, int copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982{
2983 const int off = skb->len;
2984
2985 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viroaf2b0402014-11-27 21:44:24 -05002986 __wsum csum = 0;
Al Viro15e6cb42016-11-01 22:42:45 -04002987 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2988 &csum, from)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002989 skb->csum = csum_block_add(skb->csum, csum, off);
2990 return 0;
2991 }
Al Viro15e6cb42016-11-01 22:42:45 -04002992 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002993 return 0;
2994
2995 __skb_trim(skb, off);
2996 return -EFAULT;
2997}
2998
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002999static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3000 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003001{
Willem de Bruijn1f8b9772017-08-03 16:29:41 -04003002 if (skb_zcopy(skb))
3003 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00003005 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07003006
Ian Campbellea2ab692011-08-22 23:44:58 +00003007 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00003008 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003009 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00003010 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011}
3012
Herbert Xu364c6ba2006-06-09 16:10:40 -07003013static inline int __skb_linearize(struct sk_buff *skb)
3014{
3015 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3016}
3017
Linus Torvalds1da177e2005-04-16 15:20:36 -07003018/**
3019 * skb_linearize - convert paged skb to linear one
3020 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07003021 *
3022 * If there is no free memory -ENOMEM is returned, otherwise zero
3023 * is returned and the old skb data released.
3024 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07003025static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003026{
Herbert Xu364c6ba2006-06-09 16:10:40 -07003027 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3028}
3029
3030/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00003031 * skb_has_shared_frag - can any frag be overwritten
3032 * @skb: buffer to test
3033 *
3034 * Return true if the skb has at least one frag that might be modified
3035 * by an external entity (as in vmsplice()/sendfile())
3036 */
3037static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3038{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00003039 return skb_is_nonlinear(skb) &&
3040 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00003041}
3042
3043/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07003044 * skb_linearize_cow - make sure skb is linear and writable
3045 * @skb: buffer to process
3046 *
3047 * If there is no free memory -ENOMEM is returned, otherwise zero
3048 * is returned and the old skb data released.
3049 */
3050static inline int skb_linearize_cow(struct sk_buff *skb)
3051{
3052 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3053 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003054}
3055
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003056static __always_inline void
3057__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3058 unsigned int off)
3059{
3060 if (skb->ip_summed == CHECKSUM_COMPLETE)
3061 skb->csum = csum_block_sub(skb->csum,
3062 csum_partial(start, len, 0), off);
3063 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3064 skb_checksum_start_offset(skb) < 0)
3065 skb->ip_summed = CHECKSUM_NONE;
3066}
3067
Linus Torvalds1da177e2005-04-16 15:20:36 -07003068/**
3069 * skb_postpull_rcsum - update checksum for received skb after pull
3070 * @skb: buffer to update
3071 * @start: start of data before pull
3072 * @len: length of data pulled
3073 *
3074 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07003075 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3076 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003077 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003078static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f2006-03-20 22:43:56 -08003079 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003080{
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003081 __skb_postpull_rcsum(skb, start, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003082}
3083
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003084static __always_inline void
3085__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3086 unsigned int off)
3087{
3088 if (skb->ip_summed == CHECKSUM_COMPLETE)
3089 skb->csum = csum_block_add(skb->csum,
3090 csum_partial(start, len, 0), off);
3091}
Herbert Xucbb042f2006-03-20 22:43:56 -08003092
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003093/**
3094 * skb_postpush_rcsum - update checksum for received skb after push
3095 * @skb: buffer to update
3096 * @start: start of data after push
3097 * @len: length of data pushed
3098 *
3099 * After doing a push on a received packet, you need to call this to
3100 * update the CHECKSUM_COMPLETE checksum.
3101 */
Daniel Borkmannf8ffad62016-01-07 15:50:23 +01003102static inline void skb_postpush_rcsum(struct sk_buff *skb,
3103 const void *start, unsigned int len)
3104{
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003105 __skb_postpush_rcsum(skb, start, len, 0);
Daniel Borkmannf8ffad62016-01-07 15:50:23 +01003106}
3107
Johannes Bergaf728682017-06-16 14:29:22 +02003108void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003109
David S. Miller7ce5a272013-12-02 17:26:05 -05003110/**
WANG Cong82a31b92016-06-30 10:15:22 -07003111 * skb_push_rcsum - push skb and update receive checksum
3112 * @skb: buffer to update
3113 * @len: length of data pulled
3114 *
3115 * This function performs an skb_push on the packet and updates
3116 * the CHECKSUM_COMPLETE checksum. It should be used on
3117 * receive path processing instead of skb_push unless you know
3118 * that the checksum difference is zero (e.g., a valid IP header)
3119 * or you are setting ip_summed to CHECKSUM_NONE.
3120 */
Johannes Bergd58ff352017-06-16 14:29:23 +02003121static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
WANG Cong82a31b92016-06-30 10:15:22 -07003122{
3123 skb_push(skb, len);
3124 skb_postpush_rcsum(skb, skb->data, len);
3125 return skb->data;
3126}
3127
3128/**
David S. Miller7ce5a272013-12-02 17:26:05 -05003129 * pskb_trim_rcsum - trim received skb and update checksum
3130 * @skb: buffer to trim
3131 * @len: new length
3132 *
3133 * This is exactly the same as pskb_trim except that it ensures the
3134 * checksum of received packets are still valid after the operation.
3135 */
3136
3137static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3138{
3139 if (likely(len >= skb->len))
3140 return 0;
3141 if (skb->ip_summed == CHECKSUM_COMPLETE)
3142 skb->ip_summed = CHECKSUM_NONE;
3143 return __pskb_trim(skb, len);
3144}
3145
Daniel Borkmann5293efe2016-08-18 01:00:39 +02003146static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3147{
3148 if (skb->ip_summed == CHECKSUM_COMPLETE)
3149 skb->ip_summed = CHECKSUM_NONE;
3150 __skb_trim(skb, len);
3151 return 0;
3152}
3153
3154static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3155{
3156 if (skb->ip_summed == CHECKSUM_COMPLETE)
3157 skb->ip_summed = CHECKSUM_NONE;
3158 return __skb_grow(skb, len);
3159}
3160
Linus Torvalds1da177e2005-04-16 15:20:36 -07003161#define skb_queue_walk(queue, skb) \
3162 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07003163 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07003164 skb = skb->next)
3165
James Chapman46f89142007-04-30 00:07:31 -07003166#define skb_queue_walk_safe(queue, skb, tmp) \
3167 for (skb = (queue)->next, tmp = skb->next; \
3168 skb != (struct sk_buff *)(queue); \
3169 skb = tmp, tmp = skb->next)
3170
David S. Miller1164f522008-09-23 00:49:44 -07003171#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07003172 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07003173 skb = skb->next)
3174
3175#define skb_queue_walk_from_safe(queue, skb, tmp) \
3176 for (tmp = skb->next; \
3177 skb != (struct sk_buff *)(queue); \
3178 skb = tmp, tmp = skb->next)
3179
Stephen Hemminger300ce172005-10-30 13:47:34 -08003180#define skb_queue_reverse_walk(queue, skb) \
3181 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07003182 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08003183 skb = skb->prev)
3184
David S. Miller686a2952011-01-20 22:47:32 -08003185#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3186 for (skb = (queue)->prev, tmp = skb->prev; \
3187 skb != (struct sk_buff *)(queue); \
3188 skb = tmp, tmp = skb->prev)
3189
3190#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3191 for (tmp = skb->prev; \
3192 skb != (struct sk_buff *)(queue); \
3193 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003194
David S. Miller21dc3302010-08-23 00:13:46 -07003195static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07003196{
3197 return skb_shinfo(skb)->frag_list != NULL;
3198}
3199
3200static inline void skb_frag_list_init(struct sk_buff *skb)
3201{
3202 skb_shinfo(skb)->frag_list = NULL;
3203}
3204
David S. Milleree039872009-06-09 00:17:13 -07003205#define skb_walk_frags(skb, iter) \
3206 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3207
Rainer Weikusatea3793e2015-12-06 21:11:34 +00003208
3209int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3210 const struct sk_buff *skb);
Paolo Abeni65101ae2017-05-16 11:20:13 +02003211struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3212 struct sk_buff_head *queue,
3213 unsigned int flags,
3214 void (*destructor)(struct sock *sk,
3215 struct sk_buff *skb),
3216 int *peeked, int *off, int *err,
3217 struct sk_buff **last);
Rainer Weikusatea3793e2015-12-06 21:11:34 +00003218struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
Paolo Abeni7c13f972016-11-04 11:28:59 +01003219 void (*destructor)(struct sock *sk,
3220 struct sk_buff *skb),
Rainer Weikusatea3793e2015-12-06 21:11:34 +00003221 int *peeked, int *off, int *err,
3222 struct sk_buff **last);
Joe Perches7965bd42013-09-26 14:48:15 -07003223struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
Paolo Abeni7c13f972016-11-04 11:28:59 +01003224 void (*destructor)(struct sock *sk,
3225 struct sk_buff *skb),
Joe Perches7965bd42013-09-26 14:48:15 -07003226 int *peeked, int *off, int *err);
3227struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3228 int *err);
3229unsigned int datagram_poll(struct file *file, struct socket *sock,
3230 struct poll_table_struct *wait);
Al Viroc0371da2014-11-24 10:42:55 -05003231int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3232 struct iov_iter *to, int size);
David S. Miller51f3d022014-11-05 16:46:40 -05003233static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3234 struct msghdr *msg, int size)
3235{
Al Viroe5a4b0b2014-11-24 18:17:55 -05003236 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
David S. Miller51f3d022014-11-05 16:46:40 -05003237}
Al Viroe5a4b0b2014-11-24 18:17:55 -05003238int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3239 struct msghdr *msg);
Al Viro3a654f92014-06-19 14:15:22 -04003240int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3241 struct iov_iter *from, int len);
Al Viro3a654f92014-06-19 14:15:22 -04003242int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
Joe Perches7965bd42013-09-26 14:48:15 -07003243void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
samanthakumar627d2d62016-04-05 12:41:16 -04003244void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3245static inline void skb_free_datagram_locked(struct sock *sk,
3246 struct sk_buff *skb)
3247{
3248 __skb_free_datagram_locked(sk, skb, 0);
3249}
Joe Perches7965bd42013-09-26 14:48:15 -07003250int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07003251int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3252int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3253__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3254 int len, __wsum csum);
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02003255int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
Joe Perches7965bd42013-09-26 14:48:15 -07003256 struct pipe_inode_info *pipe, unsigned int len,
Al Viro25869262016-09-17 21:02:10 -04003257 unsigned int flags);
Tom Herbert20bf50d2017-07-28 16:22:42 -07003258int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3259 int len);
3260int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07003261void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01003262unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00003263int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3264 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07003265void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3266int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3267void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01003268unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Marcelo Ricardo Leitnerae7ef812016-06-02 15:05:41 -03003269bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
Joe Perches7965bd42013-09-26 14:48:15 -07003270struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Vlad Yasevich0d5501c2014-08-08 14:42:13 -04003271struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
Jiri Pirkoe2195122014-11-19 14:05:01 +01003272int skb_ensure_writable(struct sk_buff *skb, int write_len);
Shmulik Ladkanibfca4c52016-09-19 19:11:09 +03003273int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
Jiri Pirko93515d52014-11-19 14:05:02 +01003274int skb_vlan_pop(struct sk_buff *skb);
3275int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
Sowmini Varadhan6fa01cc2016-04-22 18:36:35 -07003276struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3277 gfp_t gfp);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03003278
Al Viro6ce8e9c2014-04-06 21:25:44 -04003279static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3280{
Al Viro3073f072017-02-17 23:13:25 -05003281 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
Al Viro6ce8e9c2014-04-06 21:25:44 -04003282}
3283
Al Viro7eab8d92014-04-06 21:51:23 -04003284static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3285{
Al Viroe5a4b0b2014-11-24 18:17:55 -05003286 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro7eab8d92014-04-06 21:51:23 -04003287}
3288
Daniel Borkmann2817a332013-10-30 11:50:51 +01003289struct skb_checksum_ops {
3290 __wsum (*update)(const void *mem, int len, __wsum wsum);
3291 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3292};
3293
Davide Caratti96178132017-05-18 15:44:37 +02003294extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3295
Daniel Borkmann2817a332013-10-30 11:50:51 +01003296__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3297 __wsum csum, const struct skb_checksum_ops *ops);
3298__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3299 __wsum csum);
3300
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003301static inline void * __must_check
3302__skb_header_pointer(const struct sk_buff *skb, int offset,
3303 int len, void *data, int hlen, void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003304{
Patrick McHardy55820ee2005-07-05 14:08:10 -07003305 if (hlen - offset >= len)
David S. Miller690e36e2014-08-23 12:13:41 -07003306 return data + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003307
David S. Miller690e36e2014-08-23 12:13:41 -07003308 if (!skb ||
3309 skb_copy_bits(skb, offset, buffer, len) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003310 return NULL;
3311
3312 return buffer;
3313}
3314
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003315static inline void * __must_check
3316skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
David S. Miller690e36e2014-08-23 12:13:41 -07003317{
3318 return __skb_header_pointer(skb, offset, len, skb->data,
3319 skb_headlen(skb), buffer);
3320}
3321
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01003322/**
3323 * skb_needs_linearize - check if we need to linearize a given skb
3324 * depending on the given device features.
3325 * @skb: socket buffer to check
3326 * @features: net device features
3327 *
3328 * Returns true if either:
3329 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3330 * 2. skb is fragmented and the device does not support SG.
3331 */
3332static inline bool skb_needs_linearize(struct sk_buff *skb,
3333 netdev_features_t features)
3334{
3335 return skb_is_nonlinear(skb) &&
3336 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3337 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3338}
3339
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03003340static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3341 void *to,
3342 const unsigned int len)
3343{
3344 memcpy(to, skb->data, len);
3345}
3346
3347static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3348 const int offset, void *to,
3349 const unsigned int len)
3350{
3351 memcpy(to, skb->data + offset, len);
3352}
3353
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03003354static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3355 const void *from,
3356 const unsigned int len)
3357{
3358 memcpy(skb->data, from, len);
3359}
3360
3361static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3362 const int offset,
3363 const void *from,
3364 const unsigned int len)
3365{
3366 memcpy(skb->data + offset, from, len);
3367}
3368
Joe Perches7965bd42013-09-26 14:48:15 -07003369void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003370
Patrick Ohlyac45f602009-02-12 05:03:37 +00003371static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3372{
3373 return skb->tstamp;
3374}
3375
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003376/**
3377 * skb_get_timestamp - get timestamp from a skb
3378 * @skb: skb to get stamp from
3379 * @stamp: pointer to struct timeval to store stamp in
3380 *
3381 * Timestamps are stored in the skb as offsets to a base timestamp.
3382 * This function converts the offset back to a struct timeval and stores
3383 * it in stamp.
3384 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00003385static inline void skb_get_timestamp(const struct sk_buff *skb,
3386 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003387{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003388 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003389}
3390
Patrick Ohlyac45f602009-02-12 05:03:37 +00003391static inline void skb_get_timestampns(const struct sk_buff *skb,
3392 struct timespec *stamp)
3393{
3394 *stamp = ktime_to_timespec(skb->tstamp);
3395}
3396
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003397static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003398{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003399 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003400}
3401
Stephen Hemminger164891a2007-04-23 22:26:16 -07003402static inline ktime_t net_timedelta(ktime_t t)
3403{
3404 return ktime_sub(ktime_get_real(), t);
3405}
3406
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07003407static inline ktime_t net_invalid_timestamp(void)
3408{
Thomas Gleixner8b0e1952016-12-25 12:30:41 +01003409 return 0;
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07003410}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003411
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02003412static inline u8 skb_metadata_len(const struct sk_buff *skb)
3413{
3414 return skb_shinfo(skb)->meta_len;
3415}
3416
3417static inline void *skb_metadata_end(const struct sk_buff *skb)
3418{
3419 return skb_mac_header(skb);
3420}
3421
3422static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3423 const struct sk_buff *skb_b,
3424 u8 meta_len)
3425{
3426 const void *a = skb_metadata_end(skb_a);
3427 const void *b = skb_metadata_end(skb_b);
3428 /* Using more efficient varaiant than plain call to memcmp(). */
3429#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3430 u64 diffs = 0;
3431
3432 switch (meta_len) {
3433#define __it(x, op) (x -= sizeof(u##op))
3434#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3435 case 32: diffs |= __it_diff(a, b, 64);
3436 case 24: diffs |= __it_diff(a, b, 64);
3437 case 16: diffs |= __it_diff(a, b, 64);
3438 case 8: diffs |= __it_diff(a, b, 64);
3439 break;
3440 case 28: diffs |= __it_diff(a, b, 64);
3441 case 20: diffs |= __it_diff(a, b, 64);
3442 case 12: diffs |= __it_diff(a, b, 64);
3443 case 4: diffs |= __it_diff(a, b, 32);
3444 break;
3445 }
3446 return diffs;
3447#else
3448 return memcmp(a - meta_len, b - meta_len, meta_len);
3449#endif
3450}
3451
3452static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3453 const struct sk_buff *skb_b)
3454{
3455 u8 len_a = skb_metadata_len(skb_a);
3456 u8 len_b = skb_metadata_len(skb_b);
3457
3458 if (!(len_a | len_b))
3459 return false;
3460
3461 return len_a != len_b ?
3462 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3463}
3464
3465static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3466{
3467 skb_shinfo(skb)->meta_len = meta_len;
3468}
3469
3470static inline void skb_metadata_clear(struct sk_buff *skb)
3471{
3472 skb_metadata_set(skb, 0);
3473}
3474
Alexander Duyck62bccb82014-09-04 13:31:35 -04003475struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3476
Richard Cochranc1f19b52010-07-17 08:49:36 +00003477#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3478
Joe Perches7965bd42013-09-26 14:48:15 -07003479void skb_clone_tx_timestamp(struct sk_buff *skb);
3480bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00003481
3482#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3483
3484static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3485{
3486}
3487
3488static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3489{
3490 return false;
3491}
3492
3493#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3494
3495/**
3496 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3497 *
Richard Cochranda92b192011-10-21 00:49:15 +00003498 * PHY drivers may accept clones of transmitted packets for
3499 * timestamping via their phy_driver.txtstamp method. These drivers
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003500 * must call this function to return the skb back to the stack with a
3501 * timestamp.
Richard Cochranda92b192011-10-21 00:49:15 +00003502 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00003503 * @skb: clone of the the original outgoing packet
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003504 * @hwtstamps: hardware time stamps
Richard Cochranc1f19b52010-07-17 08:49:36 +00003505 *
3506 */
3507void skb_complete_tx_timestamp(struct sk_buff *skb,
3508 struct skb_shared_hwtstamps *hwtstamps);
3509
Willem de Bruijne7fd2882014-08-04 22:11:48 -04003510void __skb_tstamp_tx(struct sk_buff *orig_skb,
3511 struct skb_shared_hwtstamps *hwtstamps,
3512 struct sock *sk, int tstype);
3513
Patrick Ohlyac45f602009-02-12 05:03:37 +00003514/**
3515 * skb_tstamp_tx - queue clone of skb with send time stamps
3516 * @orig_skb: the original outgoing packet
3517 * @hwtstamps: hardware time stamps, may be NULL if not available
3518 *
3519 * If the skb has a socket associated, then this function clones the
3520 * skb (thus sharing the actual data and optional structures), stores
3521 * the optional hardware time stamping information (if non NULL) or
3522 * generates a software time stamp (otherwise), then queues the clone
3523 * to the error queue of the socket. Errors are silently ignored.
3524 */
Joe Perches7965bd42013-09-26 14:48:15 -07003525void skb_tstamp_tx(struct sk_buff *orig_skb,
3526 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00003527
Richard Cochran4507a712010-07-17 08:48:28 +00003528/**
3529 * skb_tx_timestamp() - Driver hook for transmit timestamping
3530 *
3531 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00003532 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00003533 *
David S. Miller73409f32013-12-27 13:04:33 -05003534 * Specifically, one should make absolutely sure that this function is
3535 * called before TX completion of this packet can trigger. Otherwise
3536 * the packet could potentially already be freed.
3537 *
Richard Cochran4507a712010-07-17 08:48:28 +00003538 * @skb: A socket buffer.
3539 */
3540static inline void skb_tx_timestamp(struct sk_buff *skb)
3541{
Richard Cochranc1f19b52010-07-17 08:49:36 +00003542 skb_clone_tx_timestamp(skb);
Miroslav Lichvarb50a5c72017-05-19 17:52:40 +02003543 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3544 skb_tstamp_tx(skb, NULL);
Richard Cochran4507a712010-07-17 08:48:28 +00003545}
3546
Johannes Berg6e3e9392011-11-09 10:15:42 +01003547/**
3548 * skb_complete_wifi_ack - deliver skb with wifi status
3549 *
3550 * @skb: the original outgoing packet
3551 * @acked: ack status
3552 *
3553 */
3554void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3555
Joe Perches7965bd42013-09-26 14:48:15 -07003556__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3557__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003558
Herbert Xu60476372007-04-09 11:59:39 -07003559static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3560{
Tom Herbert6edec0e2015-02-10 16:30:28 -08003561 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3562 skb->csum_valid ||
3563 (skb->ip_summed == CHECKSUM_PARTIAL &&
3564 skb_checksum_start_offset(skb) >= 0));
Herbert Xu60476372007-04-09 11:59:39 -07003565}
3566
Herbert Xufb286bb2005-11-10 13:01:24 -08003567/**
3568 * skb_checksum_complete - Calculate checksum of an entire packet
3569 * @skb: packet to process
3570 *
3571 * This function calculates the checksum over the entire packet plus
3572 * the value of skb->csum. The latter can be used to supply the
3573 * checksum of a pseudo header as used by TCP/UDP. It returns the
3574 * checksum.
3575 *
3576 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3577 * this function can be used to verify that checksum on received
3578 * packets. In that case the function should return zero if the
3579 * checksum is correct. In particular, this function will return zero
3580 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3581 * hardware has already verified the correctness of the checksum.
3582 */
Al Viro4381ca32007-07-15 21:00:11 +01003583static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08003584{
Herbert Xu60476372007-04-09 11:59:39 -07003585 return skb_csum_unnecessary(skb) ?
3586 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003587}
3588
Tom Herbert77cffe22014-08-27 21:26:46 -07003589static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3590{
3591 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3592 if (skb->csum_level == 0)
3593 skb->ip_summed = CHECKSUM_NONE;
3594 else
3595 skb->csum_level--;
3596 }
3597}
3598
3599static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3600{
3601 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3602 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3603 skb->csum_level++;
3604 } else if (skb->ip_summed == CHECKSUM_NONE) {
3605 skb->ip_summed = CHECKSUM_UNNECESSARY;
3606 skb->csum_level = 0;
3607 }
3608}
3609
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003610/* Check if we need to perform checksum complete validation.
3611 *
3612 * Returns true if checksum complete is needed, false otherwise
3613 * (either checksum is unnecessary or zero checksum is allowed).
3614 */
3615static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3616 bool zero_okay,
3617 __sum16 check)
3618{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003619 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3620 skb->csum_valid = 1;
Tom Herbert77cffe22014-08-27 21:26:46 -07003621 __skb_decr_checksum_unnecessary(skb);
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003622 return false;
3623 }
3624
3625 return true;
3626}
3627
3628/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3629 * in checksum_init.
3630 */
3631#define CHECKSUM_BREAK 76
3632
Tom Herbert4e18b9a2015-04-20 14:10:04 -07003633/* Unset checksum-complete
3634 *
3635 * Unset checksum complete can be done when packet is being modified
3636 * (uncompressed for instance) and checksum-complete value is
3637 * invalidated.
3638 */
3639static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3640{
3641 if (skb->ip_summed == CHECKSUM_COMPLETE)
3642 skb->ip_summed = CHECKSUM_NONE;
3643}
3644
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003645/* Validate (init) checksum based on checksum complete.
3646 *
3647 * Return values:
3648 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3649 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3650 * checksum is stored in skb->csum for use in __skb_checksum_complete
3651 * non-zero: value of invalid checksum
3652 *
3653 */
3654static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3655 bool complete,
3656 __wsum psum)
3657{
3658 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3659 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003660 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003661 return 0;
3662 }
3663 }
3664
3665 skb->csum = psum;
3666
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003667 if (complete || skb->len <= CHECKSUM_BREAK) {
3668 __sum16 csum;
3669
3670 csum = __skb_checksum_complete(skb);
3671 skb->csum_valid = !csum;
3672 return csum;
3673 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003674
3675 return 0;
3676}
3677
3678static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3679{
3680 return 0;
3681}
3682
3683/* Perform checksum validate (init). Note that this is a macro since we only
3684 * want to calculate the pseudo header which is an input function if necessary.
3685 * First we try to validate without any computation (checksum unnecessary) and
3686 * then calculate based on checksum complete calling the function to compute
3687 * pseudo header.
3688 *
3689 * Return values:
3690 * 0: checksum is validated or try to in skb_checksum_complete
3691 * non-zero: value of invalid checksum
3692 */
3693#define __skb_checksum_validate(skb, proto, complete, \
3694 zero_okay, check, compute_pseudo) \
3695({ \
3696 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003697 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003698 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3699 __ret = __skb_checksum_validate_complete(skb, \
3700 complete, compute_pseudo(skb, proto)); \
3701 __ret; \
3702})
3703
3704#define skb_checksum_init(skb, proto, compute_pseudo) \
3705 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3706
3707#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3708 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3709
3710#define skb_checksum_validate(skb, proto, compute_pseudo) \
3711 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3712
3713#define skb_checksum_validate_zero_check(skb, proto, check, \
3714 compute_pseudo) \
Sabrina Dubroca096a4cf2015-02-06 18:54:19 +01003715 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003716
3717#define skb_checksum_simple_validate(skb) \
3718 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3719
Tom Herbertd96535a2014-08-31 15:12:42 -07003720static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3721{
Davide Caratti219f1d792017-05-18 15:44:39 +02003722 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
Tom Herbertd96535a2014-08-31 15:12:42 -07003723}
3724
3725static inline void __skb_checksum_convert(struct sk_buff *skb,
3726 __sum16 check, __wsum pseudo)
3727{
3728 skb->csum = ~pseudo;
3729 skb->ip_summed = CHECKSUM_COMPLETE;
3730}
3731
3732#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3733do { \
3734 if (__skb_checksum_convert_check(skb)) \
3735 __skb_checksum_convert(skb, check, \
3736 compute_pseudo(skb, proto)); \
3737} while (0)
3738
Tom Herbert15e23962015-02-10 16:30:31 -08003739static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3740 u16 start, u16 offset)
3741{
3742 skb->ip_summed = CHECKSUM_PARTIAL;
3743 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3744 skb->csum_offset = offset - start;
3745}
3746
Tom Herbertdcdc8992015-02-02 16:07:34 -08003747/* Update skbuf and packet to reflect the remote checksum offload operation.
3748 * When called, ptr indicates the starting point for skb->csum when
3749 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3750 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3751 */
3752static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
Tom Herbert15e23962015-02-10 16:30:31 -08003753 int start, int offset, bool nopartial)
Tom Herbertdcdc8992015-02-02 16:07:34 -08003754{
3755 __wsum delta;
3756
Tom Herbert15e23962015-02-10 16:30:31 -08003757 if (!nopartial) {
3758 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3759 return;
3760 }
3761
Tom Herbertdcdc8992015-02-02 16:07:34 -08003762 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3763 __skb_checksum_complete(skb);
3764 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3765 }
3766
3767 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3768
3769 /* Adjust skb->csum since we changed the packet */
3770 skb->csum = csum_add(skb->csum, delta);
3771}
3772
Florian Westphalcb9c6832017-01-23 18:21:56 +01003773static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3774{
3775#if IS_ENABLED(CONFIG_NF_CONNTRACK)
Florian Westphala9e419d2017-01-23 18:21:59 +01003776 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
Florian Westphalcb9c6832017-01-23 18:21:56 +01003777#else
3778 return NULL;
3779#endif
3780}
3781
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003782#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07003783void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003784static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3785{
3786 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07003787 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003788}
3789static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3790{
3791 if (nfct)
3792 atomic_inc(&nfct->use);
3793}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003794#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003795#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003796static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3797{
Reshetova, Elena53869ce2017-06-30 13:07:57 +03003798 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003799 kfree(nf_bridge);
3800}
3801static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3802{
3803 if (nf_bridge)
Reshetova, Elena53869ce2017-06-30 13:07:57 +03003804 refcount_inc(&nf_bridge->use);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003805}
3806#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08003807static inline void nf_reset(struct sk_buff *skb)
3808{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003809#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Florian Westphala9e419d2017-01-23 18:21:59 +01003810 nf_conntrack_put(skb_nfct(skb));
3811 skb->_nfct = 0;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003812#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003813#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003814 nf_bridge_put(skb->nf_bridge);
3815 skb->nf_bridge = NULL;
3816#endif
3817}
3818
Patrick McHardy124dff02013-04-05 20:42:05 +02003819static inline void nf_reset_trace(struct sk_buff *skb)
3820{
Florian Westphal478b3602014-02-15 23:48:45 +01003821#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00003822 skb->nf_trace = 0;
3823#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003824}
3825
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003826/* Note: This doesn't put any conntrack and bridge info in dst. */
Eric Dumazetb1937222014-09-28 22:18:47 -07003827static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3828 bool copy)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003829{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003830#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Florian Westphala9e419d2017-01-23 18:21:59 +01003831 dst->_nfct = src->_nfct;
3832 nf_conntrack_get(skb_nfct(src));
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003833#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003834#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003835 dst->nf_bridge = src->nf_bridge;
3836 nf_bridge_get(src->nf_bridge);
3837#endif
Florian Westphal478b3602014-02-15 23:48:45 +01003838#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Eric Dumazetb1937222014-09-28 22:18:47 -07003839 if (copy)
3840 dst->nf_trace = src->nf_trace;
Florian Westphal478b3602014-02-15 23:48:45 +01003841#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003842}
3843
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003844static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3845{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003846#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Florian Westphala9e419d2017-01-23 18:21:59 +01003847 nf_conntrack_put(skb_nfct(dst));
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003848#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003849#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003850 nf_bridge_put(dst->nf_bridge);
3851#endif
Eric Dumazetb1937222014-09-28 22:18:47 -07003852 __nf_copy(dst, src, true);
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003853}
3854
James Morris984bc162006-06-09 00:29:17 -07003855#ifdef CONFIG_NETWORK_SECMARK
3856static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3857{
3858 to->secmark = from->secmark;
3859}
3860
3861static inline void skb_init_secmark(struct sk_buff *skb)
3862{
3863 skb->secmark = 0;
3864}
3865#else
3866static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3867{ }
3868
3869static inline void skb_init_secmark(struct sk_buff *skb)
3870{ }
3871#endif
3872
Eric W. Biederman574f7192014-04-01 12:20:24 -07003873static inline bool skb_irq_freeable(const struct sk_buff *skb)
3874{
3875 return !skb->destructor &&
3876#if IS_ENABLED(CONFIG_XFRM)
3877 !skb->sp &&
3878#endif
Florian Westphalcb9c6832017-01-23 18:21:56 +01003879 !skb_nfct(skb) &&
Eric W. Biederman574f7192014-04-01 12:20:24 -07003880 !skb->_skb_refdst &&
3881 !skb_has_frag_list(skb);
3882}
3883
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003884static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3885{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003886 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003887}
3888
Stephen Hemminger92477442009-03-21 13:39:26 -07003889static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003890{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003891 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003892}
3893
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003894static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3895{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003896 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003897}
3898
David S. Millerd5a9e242009-01-27 16:22:11 -08003899static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3900{
3901 skb->queue_mapping = rx_queue + 1;
3902}
3903
Stephen Hemminger92477442009-03-21 13:39:26 -07003904static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003905{
3906 return skb->queue_mapping - 1;
3907}
3908
Stephen Hemminger92477442009-03-21 13:39:26 -07003909static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003910{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003911 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003912}
3913
Julian Anastasov4ff06202017-02-06 23:14:12 +02003914static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3915{
3916 skb->dst_pending_confirm = val;
3917}
3918
3919static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3920{
3921 return skb->dst_pending_confirm != 0;
3922}
3923
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003924static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3925{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003926#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003927 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003928#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003929 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003930#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003931}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003932
Pravin B Shelar68c33162013-02-14 14:02:41 +00003933/* Keeps track of mac header offset relative to skb->head.
3934 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3935 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003936 * tunnel skb it points to outer mac header.
3937 * Keeps track of level of encapsulation of network headers.
3938 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003939struct skb_gso_cb {
Alexander Duyck802ab552016-04-10 21:45:03 -04003940 union {
3941 int mac_offset;
3942 int data_offset;
3943 };
Eric Dumazet3347c962013-10-19 11:42:56 -07003944 int encap_level;
Alexander Duyck76443452016-02-05 15:27:37 -08003945 __wsum csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003946 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003947};
Konstantin Khlebnikov9207f9d2016-01-08 15:21:46 +03003948#define SKB_SGO_CB_OFFSET 32
3949#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
Pravin B Shelar68c33162013-02-14 14:02:41 +00003950
3951static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3952{
3953 return (skb_mac_header(inner_skb) - inner_skb->head) -
3954 SKB_GSO_CB(inner_skb)->mac_offset;
3955}
3956
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003957static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3958{
3959 int new_headroom, headroom;
3960 int ret;
3961
3962 headroom = skb_headroom(skb);
3963 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3964 if (ret)
3965 return ret;
3966
3967 new_headroom = skb_headroom(skb);
3968 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3969 return 0;
3970}
3971
Alexander Duyck08b64fc2016-02-05 15:27:49 -08003972static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3973{
3974 /* Do not update partial checksums if remote checksum is enabled. */
3975 if (skb->remcsum_offload)
3976 return;
3977
3978 SKB_GSO_CB(skb)->csum = res;
3979 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3980}
3981
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003982/* Compute the checksum for a gso segment. First compute the checksum value
3983 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3984 * then add in skb->csum (checksum from csum_start to end of packet).
3985 * skb->csum and csum_start are then updated to reflect the checksum of the
3986 * resultant packet starting from the transport header-- the resultant checksum
3987 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3988 * header.
3989 */
3990static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3991{
Alexander Duyck76443452016-02-05 15:27:37 -08003992 unsigned char *csum_start = skb_transport_header(skb);
3993 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3994 __wsum partial = SKB_GSO_CB(skb)->csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003995
Alexander Duyck76443452016-02-05 15:27:37 -08003996 SKB_GSO_CB(skb)->csum = res;
3997 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003998
Alexander Duyck76443452016-02-05 15:27:37 -08003999 return csum_fold(csum_partial(csum_start, plen, partial));
Tom Herbert7e2b10c2014-06-04 17:20:02 -07004000}
4001
David S. Millerbdcc0922012-03-07 20:53:36 -05004002static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07004003{
4004 return skb_shinfo(skb)->gso_size;
4005}
4006
Eric Dumazet36a8f392013-09-29 01:21:32 -07004007/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05004008static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02004009{
4010 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4011}
4012
Daniel Borkmann5293efe2016-08-18 01:00:39 +02004013static inline void skb_gso_reset(struct sk_buff *skb)
4014{
4015 skb_shinfo(skb)->gso_size = 0;
4016 skb_shinfo(skb)->gso_segs = 0;
4017 skb_shinfo(skb)->gso_type = 0;
4018}
4019
Joe Perches7965bd42013-09-26 14:48:15 -07004020void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07004021
4022static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4023{
4024 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4025 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04004026 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4027
Alexander Duyckb78462e2010-06-02 12:24:37 +00004028 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4029 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07004030 __skb_warn_lro_forwarding(skb);
4031 return true;
4032 }
4033 return false;
4034}
4035
Herbert Xu35fc92a2007-03-26 23:22:20 -07004036static inline void skb_forward_csum(struct sk_buff *skb)
4037{
4038 /* Unfortunately we don't support this one. Any brave souls? */
4039 if (skb->ip_summed == CHECKSUM_COMPLETE)
4040 skb->ip_summed = CHECKSUM_NONE;
4041}
4042
Eric Dumazetbc8acf22010-09-02 13:07:41 -07004043/**
4044 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4045 * @skb: skb to check
4046 *
4047 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4048 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4049 * use this helper, to document places where we make this assertion.
4050 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04004051static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07004052{
4053#ifdef DEBUG
4054 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4055#endif
4056}
4057
Rusty Russellf35d9d82008-02-04 23:49:54 -05004058bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00004059
Paul Durranted1f50c2014-01-09 10:02:46 +00004060int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
Linus Lüssing9afd85c2015-05-02 14:01:07 +02004061struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4062 unsigned int transport_len,
4063 __sum16(*skb_chkf)(struct sk_buff *skb));
Paul Durranted1f50c2014-01-09 10:02:46 +00004064
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00004065/**
4066 * skb_head_is_locked - Determine if the skb->head is locked down
4067 * @skb: skb to check
4068 *
4069 * The head on skbs build around a head frag can be removed if they are
4070 * not cloned. This function returns true if the skb head is locked down
4071 * due to either being allocated via kmalloc, or by being a clone with
4072 * multiple references to the head.
4073 */
4074static inline bool skb_head_is_locked(const struct sk_buff *skb)
4075{
4076 return !skb->head_frag || skb_cloned(skb);
4077}
Florian Westphalfe6cc552014-02-13 23:09:12 +01004078
4079/**
4080 * skb_gso_network_seglen - Return length of individual segments of a gso packet
4081 *
4082 * @skb: GSO skb
4083 *
4084 * skb_gso_network_seglen is used to determine the real size of the
4085 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
4086 *
4087 * The MAC/L2 header is not accounted for.
4088 */
4089static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
4090{
4091 unsigned int hdr_len = skb_transport_header(skb) -
4092 skb_network_header(skb);
4093 return hdr_len + skb_gso_transport_seglen(skb);
4094}
Thomas Grafee122c72015-07-21 10:43:58 +02004095
Edward Cree179bc672016-02-11 20:48:04 +00004096/* Local Checksum Offload.
4097 * Compute outer checksum based on the assumption that the
4098 * inner checksum will be offloaded later.
Edward Creee8ae7b02016-02-11 21:03:37 +00004099 * See Documentation/networking/checksum-offloads.txt for
4100 * explanation of how this works.
Edward Cree179bc672016-02-11 20:48:04 +00004101 * Fill in outer checksum adjustment (e.g. with sum of outer
4102 * pseudo-header) before calling.
4103 * Also ensure that inner checksum is in linear data area.
4104 */
4105static inline __wsum lco_csum(struct sk_buff *skb)
4106{
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08004107 unsigned char *csum_start = skb_checksum_start(skb);
4108 unsigned char *l4_hdr = skb_transport_header(skb);
4109 __wsum partial;
Edward Cree179bc672016-02-11 20:48:04 +00004110
4111 /* Start with complement of inner checksum adjustment */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08004112 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4113 skb->csum_offset));
4114
Edward Cree179bc672016-02-11 20:48:04 +00004115 /* Add in checksum of our headers (incl. outer checksum
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08004116 * adjustment filled in by caller) and return result.
Edward Cree179bc672016-02-11 20:48:04 +00004117 */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08004118 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
Edward Cree179bc672016-02-11 20:48:04 +00004119}
4120
Linus Torvalds1da177e2005-04-16 15:20:36 -07004121#endif /* __KERNEL__ */
4122#endif /* _LINUX_SKBUFF_H */