blob: ada821466e883b443c7ae7c0c698e1a29869d170 [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 Ry5f74f82e2016-02-03 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
Florian Westphala9e419d2017-01-23 18:21:59 +0100620 * @_nfct: Associated connection, if any (with nfctinfo bits)
Linus Torvalds1da177e2005-04-16 15:20:36 -0700621 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
Eric Dumazet8964be42009-11-20 15:35:04 -0800622 * @skb_iif: ifindex of device we arrived on
Linus Torvalds1da177e2005-04-16 15:20:36 -0700623 * @tc_index: Traffic control index
Tom Herbert61b905d2014-03-24 15:34:47 -0700624 * @hash: the packet hash
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700625 * @queue_mapping: Queue mapping for multiqueue devices
David S. Miller0b725a22014-08-25 15:51:53 -0700626 * @xmit_more: More SKBs are pending for this queue
Randy Dunlap553a5672008-04-20 10:51:01 -0700627 * @ndisc_nodetype: router type (from link layer)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700628 * @ooo_okay: allow the mapping of a socket to a queue to be changed
Tom Herbert61b905d2014-03-24 15:34:47 -0700629 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport
Changli Gao4ca24622011-08-19 07:26:44 -0700630 * ports.
Tom Herberta3b18dd2014-07-01 21:33:17 -0700631 * @sw_hash: indicates hash was computed in software stack
Johannes Berg6e3e9392011-11-09 10:15:42 +0100632 * @wifi_acked_valid: wifi_acked was set
633 * @wifi_acked: whether frame was acked on wifi or not
Ben Greear3bdc0eb2012-02-11 15:39:30 +0000634 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS
Davide Carattidba00302017-05-18 15:44:40 +0200635 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
Julian Anastasov4ff06202017-02-06 23:14:12 +0200636 * @dst_pending_confirm: need to confirm neighbour
Eliezer Tamir06021292013-06-10 11:39:50 +0300637 * @napi_id: id of the NAPI struct this skb came from
James Morris984bc162006-06-09 00:29:17 -0700638 * @secmark: security marking
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700639 * @mark: Generic packet mark
Patrick McHardy86a9bad2013-04-19 02:04:30 +0000640 * @vlan_proto: vlan encapsulation protocol
Patrick McHardy6aa895b2008-07-14 22:49:06 -0700641 * @vlan_tci: vlan tag control information
Simon Horman0d89d202013-05-23 21:02:52 +0000642 * @inner_protocol: Protocol (encapsulation)
Joseph Gasparakis6a674e92012-12-07 14:14:14 +0000643 * @inner_transport_header: Inner transport layer header (encapsulation)
644 * @inner_network_header: Network layer header (encapsulation)
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +0000645 * @inner_mac_header: Link layer header (encapsulation)
Daniel Balutad84e0bd2011-07-10 07:04:04 -0700646 * @transport_header: Transport layer header
647 * @network_header: Network layer header
648 * @mac_header: Link layer header
649 * @tail: Tail pointer
650 * @end: End pointer
651 * @head: Head of buffer
652 * @data: Data head pointer
653 * @truesize: Buffer size
654 * @users: User count - see {datagram,tcp}.c
Linus Torvalds1da177e2005-04-16 15:20:36 -0700655 */
656
657struct sk_buff {
Eric Dumazet363ec39232014-02-26 14:02:11 -0800658 union {
Eric Dumazet56b17422014-11-03 08:19:53 -0800659 struct {
660 /* These two members must be first. */
661 struct sk_buff *next;
662 struct sk_buff *prev;
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000663
Eric Dumazet56b17422014-11-03 08:19:53 -0800664 union {
Eric Dumazetbffa72c2017-09-19 05:14:24 -0700665 struct net_device *dev;
666 /* Some protocols might use this space to store information,
667 * while device pointer would be NULL.
668 * UDP receive path is one user.
669 */
670 unsigned long dev_scratch;
Eric Dumazet56b17422014-11-03 08:19:53 -0800671 };
672 };
673 struct rb_node rbnode; /* used in netem & tcp stack */
674 };
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000675 struct sock *sk;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700676
Eric Dumazetc84d9492016-12-08 11:41:55 -0800677 union {
Eric Dumazetbffa72c2017-09-19 05:14:24 -0700678 ktime_t tstamp;
679 u64 skb_mstamp;
Eric Dumazetc84d9492016-12-08 11:41:55 -0800680 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700681 /*
682 * This is the control buffer. It is free to use for every
683 * layer. Please put your private variables there. If you
684 * want to keep them across layers you have to do a skb_clone()
685 * first. This is owned by whoever has the skb queued ATM.
686 */
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000687 char cb[48] __aligned(8);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700688
Eric Dumazet7fee2262010-05-11 23:19:48 +0000689 unsigned long _skb_refdst;
Eric Dumazetb1937222014-09-28 22:18:47 -0700690 void (*destructor)(struct sk_buff *skb);
Felix Fietkauda3f5cf2010-02-23 11:45:51 +0000691#ifdef CONFIG_XFRM
692 struct sec_path *sp;
693#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700694#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Florian Westphala9e419d2017-01-23 18:21:59 +0100695 unsigned long _nfct;
Eric Dumazetb1937222014-09-28 22:18:47 -0700696#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +0200697#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Eric Dumazetb1937222014-09-28 22:18:47 -0700698 struct nf_bridge_info *nf_bridge;
699#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -0700700 unsigned int len,
Patrick McHardy334a8132007-06-25 04:35:20 -0700701 data_len;
702 __u16 mac_len,
703 hdr_len;
Eric Dumazetb1937222014-09-28 22:18:47 -0700704
705 /* Following fields are _not_ copied in __copy_skb_header()
706 * Note that queue_mapping is here mostly to fill a hole.
707 */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200708 kmemcheck_bitfield_begin(flags1);
Eric Dumazetb1937222014-09-28 22:18:47 -0700709 __u16 queue_mapping;
Daniel Borkmann36bbef52016-09-20 00:26:13 +0200710
711/* if you move cloned around you also must adapt those constants */
712#ifdef __BIG_ENDIAN_BITFIELD
713#define CLONED_MASK (1 << 7)
714#else
715#define CLONED_MASK 1
716#endif
717#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset)
718
719 __u8 __cloned_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700720 __u8 cloned:1,
Harald Welte6869c4d2005-08-09 19:24:19 -0700721 nohdr:1,
Patrick McHardyb84f4cc2005-11-20 21:19:21 -0800722 fclone:2,
Herbert Xua59322b2007-12-05 01:53:40 -0800723 peeked:1,
Eric Dumazetb1937222014-09-28 22:18:47 -0700724 head_frag:1,
Daniel Borkmann36bbef52016-09-20 00:26:13 +0200725 xmit_more:1,
726 __unused:1; /* one bit hole */
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200727 kmemcheck_bitfield_end(flags1);
Linus Torvalds1da177e2005-04-16 15:20:36 -0700728
Eric Dumazetb1937222014-09-28 22:18:47 -0700729 /* fields enclosed in headers_start/headers_end are copied
730 * using a single memcpy() in __copy_skb_header()
731 */
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700732 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700733 __u32 headers_start[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700734 /* public: */
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200735
736/* if you move pkt_type around you also must adapt those constants */
737#ifdef __BIG_ENDIAN_BITFIELD
738#define PKT_TYPE_MAX (7 << 5)
739#else
740#define PKT_TYPE_MAX 7
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +0100741#endif
Hannes Frederic Sowa233577a2014-09-12 14:04:43 +0200742#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset)
743
744 __u8 __pkt_type_offset[0];
Eric Dumazetb1937222014-09-28 22:18:47 -0700745 __u8 pkt_type:3;
746 __u8 pfmemalloc:1;
747 __u8 ignore_df:1;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700748
Eric Dumazetb1937222014-09-28 22:18:47 -0700749 __u8 nf_trace:1;
750 __u8 ip_summed:2;
751 __u8 ooo_okay:1;
752 __u8 l4_hash:1;
753 __u8 sw_hash:1;
754 __u8 wifi_acked_valid:1;
755 __u8 wifi_acked:1;
756
757 __u8 no_fcs:1;
758 /* Indicates the inner headers are valid in the skbuff. */
759 __u8 encapsulation:1;
760 __u8 encap_hdr_csum:1;
761 __u8 csum_valid:1;
762 __u8 csum_complete_sw:1;
763 __u8 csum_level:2;
Davide Carattidba00302017-05-18 15:44:40 +0200764 __u8 csum_not_inet:1;
Eric Dumazetb1937222014-09-28 22:18:47 -0700765
Julian Anastasov4ff06202017-02-06 23:14:12 +0200766 __u8 dst_pending_confirm:1;
Eric Dumazetb1937222014-09-28 22:18:47 -0700767#ifdef CONFIG_IPV6_NDISC_NODETYPE
768 __u8 ndisc_nodetype:2;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700769#endif
Eric Dumazetb1937222014-09-28 22:18:47 -0700770 __u8 ipvs_property:1;
Tom Herbert8bce6d72014-09-29 20:22:29 -0700771 __u8 inner_protocol_type:1;
Tom Herberte585f232014-11-04 09:06:54 -0800772 __u8 remcsum_offload:1;
Ido Schimmel6bc506b2016-08-25 18:42:37 +0200773#ifdef CONFIG_NET_SWITCHDEV
774 __u8 offload_fwd_mark:1;
Yotam Gigiabf4bb62017-10-03 09:58:06 +0200775 __u8 offload_mr_fwd_mark:1;
Ido Schimmel6bc506b2016-08-25 18:42:37 +0200776#endif
Willem de Bruijne7246e12017-01-07 17:06:35 -0500777#ifdef CONFIG_NET_CLS_ACT
778 __u8 tc_skip_classify:1;
Willem de Bruijn8dc07fd2017-01-07 17:06:37 -0500779 __u8 tc_at_ingress:1;
Willem de Bruijnbc31c902017-01-07 17:06:38 -0500780 __u8 tc_redirected:1;
781 __u8 tc_from_ingress:1;
Willem de Bruijne7246e12017-01-07 17:06:35 -0500782#endif
Alexander Duyck4031ae62012-01-27 06:22:53 +0000783
Linus Torvalds1da177e2005-04-16 15:20:36 -0700784#ifdef CONFIG_NET_SCHED
Patrick McHardyb6b99eb2005-08-09 19:33:51 -0700785 __u16 tc_index; /* traffic control index */
Linus Torvalds1da177e2005-04-16 15:20:36 -0700786#endif
Vegard Nossumfe55f6d2008-08-30 12:16:35 +0200787
Eric Dumazetb1937222014-09-28 22:18:47 -0700788 union {
789 __wsum csum;
790 struct {
791 __u16 csum_start;
792 __u16 csum_offset;
793 };
794 };
795 __u32 priority;
796 int skb_iif;
797 __u32 hash;
798 __be16 vlan_proto;
799 __u16 vlan_tci;
Eric Dumazet2bd82482015-02-03 23:48:24 -0800800#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
801 union {
802 unsigned int napi_id;
803 unsigned int sender_cpu;
804 };
Chris Leech97fc2f02006-05-23 17:55:33 -0700805#endif
James Morris984bc162006-06-09 00:29:17 -0700806#ifdef CONFIG_NETWORK_SECMARK
Ido Schimmel6bc506b2016-08-25 18:42:37 +0200807 __u32 secmark;
James Morris984bc162006-06-09 00:29:17 -0700808#endif
Scott Feldman0c4f6912015-07-18 18:24:48 -0700809
Neil Horman3b885782009-10-12 13:26:31 -0700810 union {
811 __u32 mark;
Eric Dumazet16fad692013-03-14 05:40:32 +0000812 __u32 reserved_tailroom;
Neil Horman3b885782009-10-12 13:26:31 -0700813 };
Linus Torvalds1da177e2005-04-16 15:20:36 -0700814
Tom Herbert8bce6d72014-09-29 20:22:29 -0700815 union {
816 __be16 inner_protocol;
817 __u8 inner_ipproto;
818 };
819
Simon Horman1a37e412013-05-23 21:02:51 +0000820 __u16 inner_transport_header;
821 __u16 inner_network_header;
822 __u16 inner_mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700823
824 __be16 protocol;
Simon Horman1a37e412013-05-23 21:02:51 +0000825 __u16 transport_header;
826 __u16 network_header;
827 __u16 mac_header;
Eric Dumazetb1937222014-09-28 22:18:47 -0700828
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700829 /* private: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700830 __u32 headers_end[0];
Randy Dunlapebcf34f2014-10-26 19:14:06 -0700831 /* public: */
Eric Dumazetb1937222014-09-28 22:18:47 -0700832
Linus Torvalds1da177e2005-04-16 15:20:36 -0700833 /* These elements must be at the end, see alloc_skb() for details. */
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700834 sk_buff_data_t tail;
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700835 sk_buff_data_t end;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700836 unsigned char *head,
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -0700837 *data;
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -0700838 unsigned int truesize;
Reshetova, Elena63354792017-06-30 13:07:58 +0300839 refcount_t users;
Linus Torvalds1da177e2005-04-16 15:20:36 -0700840};
841
842#ifdef __KERNEL__
843/*
844 * Handling routines are only of interest to the kernel
845 */
846#include <linux/slab.h>
847
Linus Torvalds1da177e2005-04-16 15:20:36 -0700848
Mel Gormanc93bdd02012-07-31 16:44:19 -0700849#define SKB_ALLOC_FCLONE 0x01
850#define SKB_ALLOC_RX 0x02
Alexander Duyckfd11a832014-12-09 19:40:49 -0800851#define SKB_ALLOC_NAPI 0x04
Mel Gormanc93bdd02012-07-31 16:44:19 -0700852
853/* Returns true if the skb was allocated from PFMEMALLOC reserves */
854static inline bool skb_pfmemalloc(const struct sk_buff *skb)
855{
856 return unlikely(skb->pfmemalloc);
857}
858
Eric Dumazet7fee2262010-05-11 23:19:48 +0000859/*
860 * skb might have a dst pointer attached, refcounted or not.
861 * _skb_refdst low order bit is set if refcount was _not_ taken
862 */
863#define SKB_DST_NOREF 1UL
864#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
865
Florian Westphala9e419d2017-01-23 18:21:59 +0100866#define SKB_NFCT_PTRMASK ~(7UL)
Eric Dumazet7fee2262010-05-11 23:19:48 +0000867/**
868 * skb_dst - returns skb dst_entry
869 * @skb: buffer
870 *
871 * Returns skb dst_entry, regardless of reference taken or not.
872 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000873static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
874{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000875 /* If refdst was not refcounted, check we still are in a
876 * rcu_read_lock section
877 */
878 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
879 !rcu_read_lock_held() &&
880 !rcu_read_lock_bh_held());
881 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
Eric Dumazetadf30902009-06-02 05:19:30 +0000882}
883
Eric Dumazet7fee2262010-05-11 23:19:48 +0000884/**
885 * skb_dst_set - sets skb dst
886 * @skb: buffer
887 * @dst: dst entry
888 *
889 * Sets skb dst, assuming a reference was taken on dst and should
890 * be released by skb_dst_drop()
891 */
Eric Dumazetadf30902009-06-02 05:19:30 +0000892static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
893{
Eric Dumazet7fee2262010-05-11 23:19:48 +0000894 skb->_skb_refdst = (unsigned long)dst;
895}
896
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200897/**
898 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
899 * @skb: buffer
900 * @dst: dst entry
901 *
902 * Sets skb dst, assuming a reference was not taken on dst.
903 * If dst entry is cached, we do not take reference and dst_release
904 * will be avoided by refdst_drop. If dst entry is not cached, we take
905 * reference, so that last dst_release can destroy the dst immediately.
906 */
907static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
908{
Hannes Frederic Sowadbfc4fb2014-12-06 19:19:42 +0100909 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
910 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
Julian Anastasov932bc4d2013-03-21 11:57:58 +0200911}
Eric Dumazet7fee2262010-05-11 23:19:48 +0000912
913/**
Lucas De Marchi25985ed2011-03-30 22:57:33 -0300914 * skb_dst_is_noref - Test if skb dst isn't refcounted
Eric Dumazet7fee2262010-05-11 23:19:48 +0000915 * @skb: buffer
916 */
917static inline bool skb_dst_is_noref(const struct sk_buff *skb)
918{
919 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
Eric Dumazetadf30902009-06-02 05:19:30 +0000920}
921
Eric Dumazet511c3f92009-06-02 05:14:27 +0000922static inline struct rtable *skb_rtable(const struct sk_buff *skb)
923{
Eric Dumazetadf30902009-06-02 05:19:30 +0000924 return (struct rtable *)skb_dst(skb);
Eric Dumazet511c3f92009-06-02 05:14:27 +0000925}
926
Jamal Hadi Salim8b10cab2016-07-02 06:43:14 -0400927/* For mangling skb->pkt_type from user space side from applications
928 * such as nft, tc, etc, we only allow a conservative subset of
929 * possible pkt_types to be set.
930*/
931static inline bool skb_pkt_type_ok(u32 ptype)
932{
933 return ptype <= PACKET_OTHERHOST;
934}
935
Miroslav Lichvar90b602f2017-05-19 17:52:37 +0200936static inline unsigned int skb_napi_id(const struct sk_buff *skb)
937{
938#ifdef CONFIG_NET_RX_BUSY_POLL
939 return skb->napi_id;
940#else
941 return 0;
942#endif
943}
944
Paolo Abeni3889a8032017-06-12 11:23:41 +0200945/* decrement the reference count and return true if we can free the skb */
946static inline bool skb_unref(struct sk_buff *skb)
947{
948 if (unlikely(!skb))
949 return false;
Reshetova, Elena63354792017-06-30 13:07:58 +0300950 if (likely(refcount_read(&skb->users) == 1))
Paolo Abeni3889a8032017-06-12 11:23:41 +0200951 smp_rmb();
Reshetova, Elena63354792017-06-30 13:07:58 +0300952 else if (likely(!refcount_dec_and_test(&skb->users)))
Paolo Abeni3889a8032017-06-12 11:23:41 +0200953 return false;
954
955 return true;
956}
957
Paolo Abeni0a463c72017-06-12 11:23:42 +0200958void skb_release_head_state(struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -0700959void kfree_skb(struct sk_buff *skb);
960void kfree_skb_list(struct sk_buff *segs);
961void skb_tx_error(struct sk_buff *skb);
962void consume_skb(struct sk_buff *skb);
Paolo Abenica2c1412017-09-06 14:44:36 +0200963void __consume_stateless_skb(struct sk_buff *skb);
Joe Perches7965bd42013-09-26 14:48:15 -0700964void __kfree_skb(struct sk_buff *skb);
Eric Dumazetd7e88832012-04-30 08:10:34 +0000965extern struct kmem_cache *skbuff_head_cache;
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000966
Joe Perches7965bd42013-09-26 14:48:15 -0700967void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
968bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
969 bool *fragstolen, int *delta_truesize);
Eric Dumazetbad43ca2012-05-19 03:02:02 +0000970
Joe Perches7965bd42013-09-26 14:48:15 -0700971struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
972 int node);
Eric Dumazet2ea2f622015-04-24 16:05:01 -0700973struct sk_buff *__build_skb(void *data, unsigned int frag_size);
Joe Perches7965bd42013-09-26 14:48:15 -0700974struct sk_buff *build_skb(void *data, unsigned int frag_size);
David S. Millerd179cd12005-08-17 14:57:30 -0700975static inline struct sk_buff *alloc_skb(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +0100976 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -0700977{
Eric Dumazet564824b2010-10-11 19:05:25 +0000978 return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -0700979}
980
Eric Dumazet2e4e4412014-09-17 04:49:49 -0700981struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
982 unsigned long data_len,
983 int max_page_order,
984 int *errcode,
985 gfp_t gfp_mask);
986
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700987/* Layout of fast clones : [skb1][skb2][fclone_ref] */
988struct sk_buff_fclones {
989 struct sk_buff skb1;
990
991 struct sk_buff skb2;
992
Reshetova, Elena26385952017-06-30 13:07:59 +0300993 refcount_t fclone_ref;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700994};
995
996/**
997 * skb_fclone_busy - check if fclone is busy
Stephen Hemminger293de7d2016-10-23 09:28:29 -0700998 * @sk: socket
Eric Dumazetd0bf4a92014-09-29 13:29:15 -0700999 * @skb: buffer
1000 *
Masanari Iidabda13fe2015-12-13 16:53:02 +09001001 * Returns true if skb is a fast clone, and its clone is not freed.
Eric Dumazet39bb5e62014-10-30 10:32:34 -07001002 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1003 * so we also check that this didnt happen.
Eric Dumazetd0bf4a92014-09-29 13:29:15 -07001004 */
Eric Dumazet39bb5e62014-10-30 10:32:34 -07001005static inline bool skb_fclone_busy(const struct sock *sk,
1006 const struct sk_buff *skb)
Eric Dumazetd0bf4a92014-09-29 13:29:15 -07001007{
1008 const struct sk_buff_fclones *fclones;
1009
1010 fclones = container_of(skb, struct sk_buff_fclones, skb1);
1011
1012 return skb->fclone == SKB_FCLONE_ORIG &&
Reshetova, Elena26385952017-06-30 13:07:59 +03001013 refcount_read(&fclones->fclone_ref) > 1 &&
Eric Dumazet39bb5e62014-10-30 10:32:34 -07001014 fclones->skb2.sk == sk;
Eric Dumazetd0bf4a92014-09-29 13:29:15 -07001015}
1016
David S. Millerd179cd12005-08-17 14:57:30 -07001017static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
Al Virodd0fc662005-10-07 07:46:04 +01001018 gfp_t priority)
David S. Millerd179cd12005-08-17 14:57:30 -07001019{
Mel Gormanc93bdd02012-07-31 16:44:19 -07001020 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
David S. Millerd179cd12005-08-17 14:57:30 -07001021}
1022
Joe Perches7965bd42013-09-26 14:48:15 -07001023struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1024int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1025struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1026struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
Octavian Purdilabad93e92014-06-12 01:36:26 +03001027struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1028 gfp_t gfp_mask, bool fclone);
1029static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1030 gfp_t gfp_mask)
1031{
1032 return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1033}
Eric Dumazet117632e2011-12-03 21:39:53 +00001034
Joe Perches7965bd42013-09-26 14:48:15 -07001035int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1036struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1037 unsigned int headroom);
1038struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1039 int newtailroom, gfp_t priority);
Jason A. Donenfeld48a1df62017-06-04 04:16:22 +02001040int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1041 int offset, int len);
1042int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1043 int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07001044int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
Florian Fainellicd0a1372017-08-22 15:12:14 -07001045int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1046
1047/**
1048 * skb_pad - zero pad the tail of an skb
1049 * @skb: buffer to pad
1050 * @pad: space to pad
1051 *
1052 * Ensure that a buffer is followed by a padding area that is zero
1053 * filled. Used by network drivers which may DMA or transfer data
1054 * beyond the buffer end onto the wire.
1055 *
1056 * May return error in out of memory cases. The skb is freed on error.
1057 */
1058static inline int skb_pad(struct sk_buff *skb, int pad)
1059{
1060 return __skb_pad(skb, pad, true);
1061}
Neil Hormanead2ceb2009-03-11 09:49:55 +00001062#define dev_kfree_skb(a) consume_skb(a)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001063
Joe Perches7965bd42013-09-26 14:48:15 -07001064int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
1065 int getfrag(void *from, char *to, int offset,
1066 int len, int odd, struct sk_buff *skb),
1067 void *from, int length);
Ananda Rajue89e9cf2005-10-18 15:46:41 -07001068
Hannes Frederic Sowabe12a1f2015-05-21 16:59:58 +02001069int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1070 int offset, size_t size);
1071
Eric Dumazetd94d9fe2009-11-04 09:50:58 -08001072struct skb_seq_state {
Thomas Graf677e90e2005-06-23 20:59:51 -07001073 __u32 lower_offset;
1074 __u32 upper_offset;
1075 __u32 frag_idx;
1076 __u32 stepped_offset;
1077 struct sk_buff *root_skb;
1078 struct sk_buff *cur_skb;
1079 __u8 *frag_data;
1080};
1081
Joe Perches7965bd42013-09-26 14:48:15 -07001082void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1083 unsigned int to, struct skb_seq_state *st);
1084unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1085 struct skb_seq_state *st);
1086void skb_abort_seq_read(struct skb_seq_state *st);
Thomas Graf677e90e2005-06-23 20:59:51 -07001087
Joe Perches7965bd42013-09-26 14:48:15 -07001088unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
Bojan Prtvar059a2442015-02-22 11:46:35 +01001089 unsigned int to, struct ts_config *config);
Thomas Graf3fc7e8a2005-06-23 21:00:17 -07001090
Tom Herbert09323cc2013-12-15 22:16:19 -08001091/*
1092 * Packet hash types specify the type of hash in skb_set_hash.
1093 *
1094 * Hash types refer to the protocol layer addresses which are used to
1095 * construct a packet's hash. The hashes are used to differentiate or identify
1096 * flows of the protocol layer for the hash type. Hash types are either
1097 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1098 *
1099 * Properties of hashes:
1100 *
1101 * 1) Two packets in different flows have different hash values
1102 * 2) Two packets in the same flow should have the same hash value
1103 *
1104 * A hash at a higher layer is considered to be more specific. A driver should
1105 * set the most specific hash possible.
1106 *
1107 * A driver cannot indicate a more specific hash than the layer at which a hash
1108 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1109 *
1110 * A driver may indicate a hash level which is less specific than the
1111 * actual layer the hash was computed on. For instance, a hash computed
1112 * at L4 may be considered an L3 hash. This should only be done if the
1113 * driver can't unambiguously determine that the HW computed the hash at
1114 * the higher layer. Note that the "should" in the second property above
1115 * permits this.
1116 */
1117enum pkt_hash_types {
1118 PKT_HASH_TYPE_NONE, /* Undefined type */
1119 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */
1120 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */
1121 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */
1122};
1123
Tom Herbertbcc83832015-09-01 09:24:24 -07001124static inline void skb_clear_hash(struct sk_buff *skb)
1125{
1126 skb->hash = 0;
1127 skb->sw_hash = 0;
1128 skb->l4_hash = 0;
1129}
1130
1131static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1132{
1133 if (!skb->l4_hash)
1134 skb_clear_hash(skb);
1135}
1136
1137static inline void
1138__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1139{
1140 skb->l4_hash = is_l4;
1141 skb->sw_hash = is_sw;
1142 skb->hash = hash;
1143}
1144
Tom Herbert09323cc2013-12-15 22:16:19 -08001145static inline void
1146skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1147{
Tom Herbertbcc83832015-09-01 09:24:24 -07001148 /* Used by drivers to set hash from HW */
1149 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1150}
1151
1152static inline void
1153__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1154{
1155 __skb_set_hash(skb, hash, true, is_l4);
Tom Herbert09323cc2013-12-15 22:16:19 -08001156}
1157
Tom Herberte5276932015-09-01 09:24:23 -07001158void __skb_get_hash(struct sk_buff *skb);
Florian Westphalb9177832016-10-26 18:49:46 +02001159u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
Tom Herberte5276932015-09-01 09:24:23 -07001160u32 skb_get_poff(const struct sk_buff *skb);
1161u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1162 const struct flow_keys *keys, int hlen);
1163__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1164 void *data, int hlen_proto);
1165
1166static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1167 int thoff, u8 ip_proto)
1168{
1169 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1170}
1171
1172void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1173 const struct flow_dissector_key *key,
1174 unsigned int key_count);
1175
1176bool __skb_flow_dissect(const struct sk_buff *skb,
1177 struct flow_dissector *flow_dissector,
1178 void *target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001179 void *data, __be16 proto, int nhoff, int hlen,
1180 unsigned int flags);
Tom Herberte5276932015-09-01 09:24:23 -07001181
1182static inline bool skb_flow_dissect(const struct sk_buff *skb,
1183 struct flow_dissector *flow_dissector,
Tom Herbertcd79a232015-09-01 09:24:27 -07001184 void *target_container, unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001185{
1186 return __skb_flow_dissect(skb, flow_dissector, target_container,
Tom Herbertcd79a232015-09-01 09:24:27 -07001187 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001188}
1189
1190static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
Tom Herbertcd79a232015-09-01 09:24:27 -07001191 struct flow_keys *flow,
1192 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001193{
1194 memset(flow, 0, sizeof(*flow));
1195 return __skb_flow_dissect(skb, &flow_keys_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001196 NULL, 0, 0, 0, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001197}
1198
1199static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow,
1200 void *data, __be16 proto,
Tom Herbertcd79a232015-09-01 09:24:27 -07001201 int nhoff, int hlen,
1202 unsigned int flags)
Tom Herberte5276932015-09-01 09:24:23 -07001203{
1204 memset(flow, 0, sizeof(*flow));
1205 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow,
Tom Herbertcd79a232015-09-01 09:24:27 -07001206 data, proto, nhoff, hlen, flags);
Tom Herberte5276932015-09-01 09:24:23 -07001207}
1208
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001209static inline __u32 skb_get_hash(struct sk_buff *skb)
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001210{
Tom Herberta3b18dd2014-07-01 21:33:17 -07001211 if (!skb->l4_hash && !skb->sw_hash)
Tom Herbert3958afa1b2013-12-15 22:12:06 -08001212 __skb_get_hash(skb);
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001213
Tom Herbert61b905d2014-03-24 15:34:47 -07001214 return skb->hash;
Krishna Kumarbfb564e2010-08-04 06:15:52 +00001215}
1216
David S. Miller20a17bf2015-09-01 21:19:17 -07001217static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
Tom Herbertf70ea012015-07-31 16:52:10 -07001218{
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001219 if (!skb->l4_hash && !skb->sw_hash) {
1220 struct flow_keys keys;
Tom Herbertde4c1f82015-09-01 18:11:04 -07001221 __u32 hash = __get_hash_from_flowi6(fl6, &keys);
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001222
Tom Herbertde4c1f82015-09-01 18:11:04 -07001223 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
Tom Herbertc6cc1ca2015-09-01 09:24:25 -07001224 }
Tom Herbertf70ea012015-07-31 16:52:10 -07001225
1226 return skb->hash;
1227}
1228
Tom Herbert50fb7992015-05-01 11:30:12 -07001229__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1230
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001231static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1232{
Tom Herbert61b905d2014-03-24 15:34:47 -07001233 return skb->hash;
Tom Herbert57bdf7f42014-01-15 08:57:54 -08001234}
1235
Tom Herbert3df7a742013-12-15 22:16:29 -08001236static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1237{
Tom Herbert61b905d2014-03-24 15:34:47 -07001238 to->hash = from->hash;
Tom Herberta3b18dd2014-07-01 21:33:17 -07001239 to->sw_hash = from->sw_hash;
Tom Herbert61b905d2014-03-24 15:34:47 -07001240 to->l4_hash = from->l4_hash;
Tom Herbert3df7a742013-12-15 22:16:29 -08001241};
1242
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001243#ifdef NET_SKBUFF_DATA_USES_OFFSET
1244static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1245{
1246 return skb->head + skb->end;
1247}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001248
1249static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1250{
1251 return skb->end;
1252}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001253#else
1254static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1255{
1256 return skb->end;
1257}
Alexander Duyckec47ea82012-05-04 14:26:56 +00001258
1259static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1260{
1261 return skb->end - skb->head;
1262}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001263#endif
1264
Linus Torvalds1da177e2005-04-16 15:20:36 -07001265/* Internal */
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001266#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001267
Patrick Ohlyac45f602009-02-12 05:03:37 +00001268static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1269{
1270 return &skb_shinfo(skb)->hwtstamps;
1271}
1272
Willem de Bruijn52267792017-08-03 16:29:39 -04001273static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1274{
1275 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1276
1277 return is_zcopy ? skb_uarg(skb) : NULL;
1278}
1279
1280static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg)
1281{
1282 if (skb && uarg && !skb_zcopy(skb)) {
1283 sock_zerocopy_get(uarg);
1284 skb_shinfo(skb)->destructor_arg = uarg;
1285 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1286 }
1287}
1288
1289/* Release a reference on a zerocopy structure */
1290static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1291{
1292 struct ubuf_info *uarg = skb_zcopy(skb);
1293
1294 if (uarg) {
Willem de Bruijn0a4a0602017-08-09 19:09:44 -04001295 if (uarg->callback == sock_zerocopy_callback) {
1296 uarg->zerocopy = uarg->zerocopy && zerocopy;
1297 sock_zerocopy_put(uarg);
1298 } else {
1299 uarg->callback(uarg, zerocopy);
1300 }
1301
Willem de Bruijn52267792017-08-03 16:29:39 -04001302 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1303 }
1304}
1305
1306/* Abort a zerocopy operation and revert zckey on error in send syscall */
1307static inline void skb_zcopy_abort(struct sk_buff *skb)
1308{
1309 struct ubuf_info *uarg = skb_zcopy(skb);
1310
1311 if (uarg) {
1312 sock_zerocopy_put_abort(uarg);
1313 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1314 }
1315}
1316
Linus Torvalds1da177e2005-04-16 15:20:36 -07001317/**
1318 * skb_queue_empty - check if a queue is empty
1319 * @list: queue head
1320 *
1321 * Returns true if the queue is empty, false otherwise.
1322 */
1323static inline int skb_queue_empty(const struct sk_buff_head *list)
1324{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001325 return list->next == (const struct sk_buff *) list;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001326}
1327
1328/**
David S. Millerfc7ebb22008-09-23 00:34:07 -07001329 * skb_queue_is_last - check if skb is the last entry in the queue
1330 * @list: queue head
1331 * @skb: buffer
1332 *
1333 * Returns true if @skb is the last buffer on the list.
1334 */
1335static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1336 const struct sk_buff *skb)
1337{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001338 return skb->next == (const struct sk_buff *) list;
David S. Millerfc7ebb22008-09-23 00:34:07 -07001339}
1340
1341/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001342 * skb_queue_is_first - check if skb is the first entry in the queue
1343 * @list: queue head
1344 * @skb: buffer
1345 *
1346 * Returns true if @skb is the first buffer on the list.
1347 */
1348static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1349 const struct sk_buff *skb)
1350{
Daniel Borkmannfd44b932014-01-07 23:23:44 +01001351 return skb->prev == (const struct sk_buff *) list;
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001352}
1353
1354/**
David S. Miller249c8b42008-09-23 00:44:42 -07001355 * skb_queue_next - return the next packet in the queue
1356 * @list: queue head
1357 * @skb: current buffer
1358 *
1359 * Return the next packet in @list after @skb. It is only valid to
1360 * call this if skb_queue_is_last() evaluates to false.
1361 */
1362static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1363 const struct sk_buff *skb)
1364{
1365 /* This BUG_ON may seem severe, but if we just return then we
1366 * are going to dereference garbage.
1367 */
1368 BUG_ON(skb_queue_is_last(list, skb));
1369 return skb->next;
1370}
1371
1372/**
Ilpo Järvinen832d11c2008-11-24 21:20:15 -08001373 * skb_queue_prev - return the prev packet in the queue
1374 * @list: queue head
1375 * @skb: current buffer
1376 *
1377 * Return the prev packet in @list before @skb. It is only valid to
1378 * call this if skb_queue_is_first() evaluates to false.
1379 */
1380static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1381 const struct sk_buff *skb)
1382{
1383 /* This BUG_ON may seem severe, but if we just return then we
1384 * are going to dereference garbage.
1385 */
1386 BUG_ON(skb_queue_is_first(list, skb));
1387 return skb->prev;
1388}
1389
1390/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001391 * skb_get - reference buffer
1392 * @skb: buffer to reference
1393 *
1394 * Makes another reference to a socket buffer and returns a pointer
1395 * to the buffer.
1396 */
1397static inline struct sk_buff *skb_get(struct sk_buff *skb)
1398{
Reshetova, Elena63354792017-06-30 13:07:58 +03001399 refcount_inc(&skb->users);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001400 return skb;
1401}
1402
1403/*
1404 * If users == 1, we are the only owner and are can avoid redundant
1405 * atomic change.
1406 */
1407
1408/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001409 * skb_cloned - is the buffer a clone
1410 * @skb: buffer to check
1411 *
1412 * Returns true if the buffer was generated with skb_clone() and is
1413 * one of multiple shared copies of the buffer. Cloned buffers are
1414 * shared data so must not be written to under normal circumstances.
1415 */
1416static inline int skb_cloned(const struct sk_buff *skb)
1417{
1418 return skb->cloned &&
1419 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1420}
1421
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001422static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1423{
Mel Gormand0164ad2015-11-06 16:28:21 -08001424 might_sleep_if(gfpflags_allow_blocking(pri));
Pravin B Shelar14bbd6a2013-02-14 09:44:49 +00001425
1426 if (skb_cloned(skb))
1427 return pskb_expand_head(skb, 0, 0, pri);
1428
1429 return 0;
1430}
1431
Linus Torvalds1da177e2005-04-16 15:20:36 -07001432/**
1433 * skb_header_cloned - is the header a clone
1434 * @skb: buffer to check
1435 *
1436 * Returns true if modifying the header part of the buffer requires
1437 * the data to be copied.
1438 */
1439static inline int skb_header_cloned(const struct sk_buff *skb)
1440{
1441 int dataref;
1442
1443 if (!skb->cloned)
1444 return 0;
1445
1446 dataref = atomic_read(&skb_shinfo(skb)->dataref);
1447 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1448 return dataref != 1;
1449}
1450
Eric Dumazet9580bf22016-04-30 10:19:29 -07001451static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1452{
1453 might_sleep_if(gfpflags_allow_blocking(pri));
1454
1455 if (skb_header_cloned(skb))
1456 return pskb_expand_head(skb, 0, 0, pri);
1457
1458 return 0;
1459}
1460
Linus Torvalds1da177e2005-04-16 15:20:36 -07001461/**
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001462 * __skb_header_release - release reference to header
1463 * @skb: buffer to operate on
Eric Dumazetf4a775d2014-09-22 16:29:32 -07001464 */
1465static inline void __skb_header_release(struct sk_buff *skb)
1466{
1467 skb->nohdr = 1;
1468 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1469}
1470
1471
1472/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001473 * skb_shared - is the buffer shared
1474 * @skb: buffer to check
1475 *
1476 * Returns true if more than one person has a reference to this
1477 * buffer.
1478 */
1479static inline int skb_shared(const struct sk_buff *skb)
1480{
Reshetova, Elena63354792017-06-30 13:07:58 +03001481 return refcount_read(&skb->users) != 1;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001482}
1483
1484/**
1485 * skb_share_check - check if buffer is shared and if so clone it
1486 * @skb: buffer to check
1487 * @pri: priority for memory allocation
1488 *
1489 * If the buffer is shared the buffer is cloned and the old copy
1490 * drops a reference. A new clone with a single reference is returned.
1491 * If the buffer is not shared the original buffer is returned. When
1492 * being called from interrupt status or with spinlocks held pri must
1493 * be GFP_ATOMIC.
1494 *
1495 * NULL is returned on a memory allocation failure.
1496 */
Eric Dumazet47061bc2012-08-03 20:54:15 +00001497static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001498{
Mel Gormand0164ad2015-11-06 16:28:21 -08001499 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001500 if (skb_shared(skb)) {
1501 struct sk_buff *nskb = skb_clone(skb, pri);
Eric Dumazet47061bc2012-08-03 20:54:15 +00001502
1503 if (likely(nskb))
1504 consume_skb(skb);
1505 else
1506 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001507 skb = nskb;
1508 }
1509 return skb;
1510}
1511
1512/*
1513 * Copy shared buffers into a new sk_buff. We effectively do COW on
1514 * packets to handle cases where we have a local reader and forward
1515 * and a couple of other messy ones. The normal one is tcpdumping
1516 * a packet thats being forwarded.
1517 */
1518
1519/**
1520 * skb_unshare - make a copy of a shared buffer
1521 * @skb: buffer to check
1522 * @pri: priority for memory allocation
1523 *
1524 * If the socket buffer is a clone then this function creates a new
1525 * copy of the data, drops a reference count on the old copy and returns
1526 * the new copy with the reference count at 1. If the buffer is not a clone
1527 * the original buffer is returned. When called with a spinlock held or
1528 * from interrupt state @pri must be %GFP_ATOMIC
1529 *
1530 * %NULL is returned on a memory allocation failure.
1531 */
Victor Fuscoe2bf5212005-07-18 13:36:38 -07001532static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
Al Virodd0fc662005-10-07 07:46:04 +01001533 gfp_t pri)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001534{
Mel Gormand0164ad2015-11-06 16:28:21 -08001535 might_sleep_if(gfpflags_allow_blocking(pri));
Linus Torvalds1da177e2005-04-16 15:20:36 -07001536 if (skb_cloned(skb)) {
1537 struct sk_buff *nskb = skb_copy(skb, pri);
Alexander Aring31eff812014-10-10 23:10:47 +02001538
1539 /* Free our shared copy */
1540 if (likely(nskb))
1541 consume_skb(skb);
1542 else
1543 kfree_skb(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001544 skb = nskb;
1545 }
1546 return skb;
1547}
1548
1549/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001550 * skb_peek - peek at the head of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001551 * @list_: list to peek at
1552 *
1553 * Peek an &sk_buff. Unlike most other operations you _MUST_
1554 * be careful with this one. A peek leaves the buffer on the
1555 * list and someone else may run off with it. You must hold
1556 * the appropriate locks or have a private queue to do this.
1557 *
1558 * Returns %NULL for an empty list or a pointer to the head element.
1559 * The reference count is not incremented and the reference is therefore
1560 * volatile. Use with caution.
1561 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001562static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001563{
Eric Dumazet18d07002012-04-30 16:31:46 +00001564 struct sk_buff *skb = list_->next;
1565
1566 if (skb == (struct sk_buff *)list_)
1567 skb = NULL;
1568 return skb;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001569}
1570
1571/**
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001572 * skb_peek_next - peek skb following the given one from a queue
1573 * @skb: skb to start from
1574 * @list_: list to peek at
1575 *
1576 * Returns %NULL when the end of the list is met or a pointer to the
1577 * next element. The reference count is not incremented and the
1578 * reference is therefore volatile. Use with caution.
1579 */
1580static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1581 const struct sk_buff_head *list_)
1582{
1583 struct sk_buff *next = skb->next;
Eric Dumazet18d07002012-04-30 16:31:46 +00001584
Pavel Emelyanovda5ef6e2012-02-21 07:31:18 +00001585 if (next == (struct sk_buff *)list_)
1586 next = NULL;
1587 return next;
1588}
1589
1590/**
Ben Hutchings1a5778a2010-02-14 22:35:47 -08001591 * skb_peek_tail - peek at the tail of an &sk_buff_head
Linus Torvalds1da177e2005-04-16 15:20:36 -07001592 * @list_: list to peek at
1593 *
1594 * Peek an &sk_buff. Unlike most other operations you _MUST_
1595 * be careful with this one. A peek leaves the buffer on the
1596 * list and someone else may run off with it. You must hold
1597 * the appropriate locks or have a private queue to do this.
1598 *
1599 * Returns %NULL for an empty list or a pointer to the tail element.
1600 * The reference count is not incremented and the reference is therefore
1601 * volatile. Use with caution.
1602 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04001603static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001604{
Eric Dumazet18d07002012-04-30 16:31:46 +00001605 struct sk_buff *skb = list_->prev;
1606
1607 if (skb == (struct sk_buff *)list_)
1608 skb = NULL;
1609 return skb;
1610
Linus Torvalds1da177e2005-04-16 15:20:36 -07001611}
1612
1613/**
1614 * skb_queue_len - get queue length
1615 * @list_: list to measure
1616 *
1617 * Return the length of an &sk_buff queue.
1618 */
1619static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1620{
1621 return list_->qlen;
1622}
1623
David S. Miller67fed452008-09-21 22:36:24 -07001624/**
1625 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1626 * @list: queue to initialize
1627 *
1628 * This initializes only the list and queue length aspects of
1629 * an sk_buff_head object. This allows to initialize the list
1630 * aspects of an sk_buff_head without reinitializing things like
1631 * the spinlock. It can also be used for on-stack sk_buff_head
1632 * objects where the spinlock is known to not be used.
1633 */
1634static inline void __skb_queue_head_init(struct sk_buff_head *list)
1635{
1636 list->prev = list->next = (struct sk_buff *)list;
1637 list->qlen = 0;
1638}
1639
Arjan van de Ven76f10ad2006-08-02 14:06:55 -07001640/*
1641 * This function creates a split out lock class for each invocation;
1642 * this is needed for now since a whole lot of users of the skb-queue
1643 * infrastructure in drivers have different locking usage (in hardirq)
1644 * than the networking core (in softirq only). In the long run either the
1645 * network layer or drivers should need annotation to consolidate the
1646 * main types of usage into 3 classes.
1647 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07001648static inline void skb_queue_head_init(struct sk_buff_head *list)
1649{
1650 spin_lock_init(&list->lock);
David S. Miller67fed452008-09-21 22:36:24 -07001651 __skb_queue_head_init(list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001652}
1653
Pavel Emelianovc2ecba72007-04-17 12:45:31 -07001654static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1655 struct lock_class_key *class)
1656{
1657 skb_queue_head_init(list);
1658 lockdep_set_class(&list->lock, class);
1659}
1660
Linus Torvalds1da177e2005-04-16 15:20:36 -07001661/*
Gerrit Renkerbf299272008-04-14 00:04:51 -07001662 * Insert an sk_buff on a list.
Linus Torvalds1da177e2005-04-16 15:20:36 -07001663 *
1664 * The "__skb_xxxx()" functions are the non-atomic ones that
1665 * can only be called with interrupts disabled.
1666 */
Joe Perches7965bd42013-09-26 14:48:15 -07001667void skb_insert(struct sk_buff *old, struct sk_buff *newsk,
1668 struct sk_buff_head *list);
Gerrit Renkerbf299272008-04-14 00:04:51 -07001669static inline void __skb_insert(struct sk_buff *newsk,
1670 struct sk_buff *prev, struct sk_buff *next,
1671 struct sk_buff_head *list)
1672{
1673 newsk->next = next;
1674 newsk->prev = prev;
1675 next->prev = prev->next = newsk;
1676 list->qlen++;
1677}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001678
David S. Miller67fed452008-09-21 22:36:24 -07001679static inline void __skb_queue_splice(const struct sk_buff_head *list,
1680 struct sk_buff *prev,
1681 struct sk_buff *next)
1682{
1683 struct sk_buff *first = list->next;
1684 struct sk_buff *last = list->prev;
1685
1686 first->prev = prev;
1687 prev->next = first;
1688
1689 last->next = next;
1690 next->prev = last;
1691}
1692
1693/**
1694 * skb_queue_splice - join two skb lists, this is designed for stacks
1695 * @list: the new list to add
1696 * @head: the place to add it in the first list
1697 */
1698static inline void skb_queue_splice(const struct sk_buff_head *list,
1699 struct sk_buff_head *head)
1700{
1701 if (!skb_queue_empty(list)) {
1702 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001703 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001704 }
1705}
1706
1707/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001708 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001709 * @list: the new list to add
1710 * @head: the place to add it in the first list
1711 *
1712 * The list at @list is reinitialised
1713 */
1714static inline void skb_queue_splice_init(struct sk_buff_head *list,
1715 struct sk_buff_head *head)
1716{
1717 if (!skb_queue_empty(list)) {
1718 __skb_queue_splice(list, (struct sk_buff *) head, head->next);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001719 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001720 __skb_queue_head_init(list);
1721 }
1722}
1723
1724/**
1725 * skb_queue_splice_tail - join two skb lists, each list being a queue
1726 * @list: the new list to add
1727 * @head: the place to add it in the first list
1728 */
1729static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1730 struct sk_buff_head *head)
1731{
1732 if (!skb_queue_empty(list)) {
1733 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001734 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001735 }
1736}
1737
1738/**
Eric Dumazetd961949662012-04-30 21:29:16 +00001739 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
David S. Miller67fed452008-09-21 22:36:24 -07001740 * @list: the new list to add
1741 * @head: the place to add it in the first list
1742 *
1743 * Each of the lists is a queue.
1744 * The list at @list is reinitialised
1745 */
1746static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1747 struct sk_buff_head *head)
1748{
1749 if (!skb_queue_empty(list)) {
1750 __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
David S. Miller1d4a31d2008-09-22 21:57:21 -07001751 head->qlen += list->qlen;
David S. Miller67fed452008-09-21 22:36:24 -07001752 __skb_queue_head_init(list);
1753 }
1754}
1755
Linus Torvalds1da177e2005-04-16 15:20:36 -07001756/**
Stephen Hemminger300ce172005-10-30 13:47:34 -08001757 * __skb_queue_after - queue a buffer at the list head
1758 * @list: list to use
1759 * @prev: place after this buffer
1760 * @newsk: buffer to queue
1761 *
1762 * Queue a buffer int the middle of a list. This function takes no locks
1763 * and you must therefore hold required locks before calling it.
1764 *
1765 * A buffer cannot be placed on two lists at the same time.
1766 */
1767static inline void __skb_queue_after(struct sk_buff_head *list,
1768 struct sk_buff *prev,
1769 struct sk_buff *newsk)
1770{
Gerrit Renkerbf299272008-04-14 00:04:51 -07001771 __skb_insert(newsk, prev, prev->next, list);
Stephen Hemminger300ce172005-10-30 13:47:34 -08001772}
1773
Joe Perches7965bd42013-09-26 14:48:15 -07001774void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1775 struct sk_buff_head *list);
Gerrit Renker7de6c032008-04-14 00:05:09 -07001776
Gerrit Renkerf5572852008-04-14 00:05:28 -07001777static inline void __skb_queue_before(struct sk_buff_head *list,
1778 struct sk_buff *next,
1779 struct sk_buff *newsk)
1780{
1781 __skb_insert(newsk, next->prev, next, list);
1782}
1783
Stephen Hemminger300ce172005-10-30 13:47:34 -08001784/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07001785 * __skb_queue_head - queue a buffer at the list head
1786 * @list: list to use
1787 * @newsk: buffer to queue
1788 *
1789 * Queue a buffer at the start of a list. This function takes no locks
1790 * and you must therefore hold required locks before calling it.
1791 *
1792 * A buffer cannot be placed on two lists at the same time.
1793 */
Joe Perches7965bd42013-09-26 14:48:15 -07001794void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001795static inline void __skb_queue_head(struct sk_buff_head *list,
1796 struct sk_buff *newsk)
1797{
Stephen Hemminger300ce172005-10-30 13:47:34 -08001798 __skb_queue_after(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001799}
1800
1801/**
1802 * __skb_queue_tail - queue a buffer at the list tail
1803 * @list: list to use
1804 * @newsk: buffer to queue
1805 *
1806 * Queue a buffer at the end of a list. This function takes no locks
1807 * and you must therefore hold required locks before calling it.
1808 *
1809 * A buffer cannot be placed on two lists at the same time.
1810 */
Joe Perches7965bd42013-09-26 14:48:15 -07001811void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001812static inline void __skb_queue_tail(struct sk_buff_head *list,
1813 struct sk_buff *newsk)
1814{
Gerrit Renkerf5572852008-04-14 00:05:28 -07001815 __skb_queue_before(list, (struct sk_buff *)list, newsk);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001816}
1817
Linus Torvalds1da177e2005-04-16 15:20:36 -07001818/*
Linus Torvalds1da177e2005-04-16 15:20:36 -07001819 * remove sk_buff from list. _Must_ be called atomically, and with
1820 * the list known..
1821 */
Joe Perches7965bd42013-09-26 14:48:15 -07001822void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001823static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1824{
1825 struct sk_buff *next, *prev;
1826
1827 list->qlen--;
1828 next = skb->next;
1829 prev = skb->prev;
1830 skb->next = skb->prev = NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001831 next->prev = prev;
1832 prev->next = next;
1833}
1834
Gerrit Renkerf525c062008-04-14 00:04:12 -07001835/**
1836 * __skb_dequeue - remove from the head of the queue
1837 * @list: list to dequeue from
1838 *
1839 * Remove the head of the list. This function does not take any locks
1840 * so must be used with appropriate locks held only. The head item is
1841 * returned or %NULL if the list is empty.
1842 */
Joe Perches7965bd42013-09-26 14:48:15 -07001843struct sk_buff *skb_dequeue(struct sk_buff_head *list);
Gerrit Renkerf525c062008-04-14 00:04:12 -07001844static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
1845{
1846 struct sk_buff *skb = skb_peek(list);
1847 if (skb)
1848 __skb_unlink(skb, list);
1849 return skb;
1850}
Linus Torvalds1da177e2005-04-16 15:20:36 -07001851
1852/**
1853 * __skb_dequeue_tail - remove from the tail of the queue
1854 * @list: list to dequeue from
1855 *
1856 * Remove the tail of the list. This function does not take any locks
1857 * so must be used with appropriate locks held only. The tail item is
1858 * returned or %NULL if the list is empty.
1859 */
Joe Perches7965bd42013-09-26 14:48:15 -07001860struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001861static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
1862{
1863 struct sk_buff *skb = skb_peek_tail(list);
1864 if (skb)
1865 __skb_unlink(skb, list);
1866 return skb;
1867}
1868
1869
David S. Millerbdcc0922012-03-07 20:53:36 -05001870static inline bool skb_is_nonlinear(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001871{
1872 return skb->data_len;
1873}
1874
1875static inline unsigned int skb_headlen(const struct sk_buff *skb)
1876{
1877 return skb->len - skb->data_len;
1878}
1879
Willem de Bruijn3ece7822017-08-03 16:29:38 -04001880static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001881{
Alexey Dobriyanc72d8cd2016-11-19 04:08:08 +03001882 unsigned int i, len = 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001883
Alexey Dobriyanc72d8cd2016-11-19 04:08:08 +03001884 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
Eric Dumazet9e903e02011-10-18 21:00:24 +00001885 len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
Willem de Bruijn3ece7822017-08-03 16:29:38 -04001886 return len;
1887}
1888
1889static inline unsigned int skb_pagelen(const struct sk_buff *skb)
1890{
1891 return skb_headlen(skb) + __skb_pagelen(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001892}
1893
Ian Campbell131ea662011-08-19 06:25:00 +00001894/**
1895 * __skb_fill_page_desc - initialise a paged fragment in an skb
1896 * @skb: buffer containing fragment to be initialised
1897 * @i: paged fragment index to initialise
1898 * @page: the page to use for this fragment
1899 * @off: the offset to the data with @page
1900 * @size: the length of the data
1901 *
1902 * Initialises the @i'th fragment of @skb to point to &size bytes at
1903 * offset @off within @page.
1904 *
1905 * Does not take any additional reference on the fragment.
1906 */
1907static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
1908 struct page *page, int off, int size)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001909{
1910 skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1911
Mel Gormanc48a11c2012-07-31 16:44:23 -07001912 /*
Michal Hocko2f064f32015-08-21 14:11:51 -07001913 * Propagate page pfmemalloc to the skb if we can. The problem is
1914 * that not all callers have unique ownership of the page but rely
1915 * on page_is_pfmemalloc doing the right thing(tm).
Mel Gormanc48a11c2012-07-31 16:44:23 -07001916 */
Ian Campbella8605c62011-10-19 23:01:49 +00001917 frag->page.p = page;
Linus Torvalds1da177e2005-04-16 15:20:36 -07001918 frag->page_offset = off;
Eric Dumazet9e903e02011-10-18 21:00:24 +00001919 skb_frag_size_set(frag, size);
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001920
1921 page = compound_head(page);
Michal Hocko2f064f32015-08-21 14:11:51 -07001922 if (page_is_pfmemalloc(page))
Pavel Emelyanovcca7af32013-03-14 03:29:40 +00001923 skb->pfmemalloc = true;
Ian Campbell131ea662011-08-19 06:25:00 +00001924}
1925
1926/**
1927 * skb_fill_page_desc - initialise a paged fragment in an skb
1928 * @skb: buffer containing fragment to be initialised
1929 * @i: paged fragment index to initialise
1930 * @page: the page to use for this fragment
1931 * @off: the offset to the data with @page
1932 * @size: the length of the data
1933 *
1934 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
Mathias Krausebc323832013-11-07 14:18:26 +01001935 * @skb to point to @size bytes at offset @off within @page. In
Ian Campbell131ea662011-08-19 06:25:00 +00001936 * addition updates @skb such that @i is the last fragment.
1937 *
1938 * Does not take any additional reference on the fragment.
1939 */
1940static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
1941 struct page *page, int off, int size)
1942{
1943 __skb_fill_page_desc(skb, i, page, off, size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07001944 skb_shinfo(skb)->nr_frags = i + 1;
1945}
1946
Joe Perches7965bd42013-09-26 14:48:15 -07001947void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
1948 int size, unsigned int truesize);
Peter Zijlstra654bed12008-10-07 14:22:33 -07001949
Jason Wangf8e617e2013-11-01 14:07:47 +08001950void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
1951 unsigned int truesize);
1952
Linus Torvalds1da177e2005-04-16 15:20:36 -07001953#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags)
David S. Miller21dc3302010-08-23 00:13:46 -07001954#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb))
Linus Torvalds1da177e2005-04-16 15:20:36 -07001955#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb))
1956
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001957#ifdef NET_SKBUFF_DATA_USES_OFFSET
1958static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1959{
1960 return skb->head + skb->tail;
1961}
1962
1963static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1964{
1965 skb->tail = skb->data - skb->head;
1966}
1967
1968static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1969{
1970 skb_reset_tail_pointer(skb);
1971 skb->tail += offset;
1972}
Simon Horman7cc46192013-05-28 20:34:29 +00001973
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001974#else /* NET_SKBUFF_DATA_USES_OFFSET */
1975static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
1976{
1977 return skb->tail;
1978}
1979
1980static inline void skb_reset_tail_pointer(struct sk_buff *skb)
1981{
1982 skb->tail = skb->data;
1983}
1984
1985static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
1986{
1987 skb->tail = skb->data + offset;
1988}
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07001989
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07001990#endif /* NET_SKBUFF_DATA_USES_OFFSET */
1991
Linus Torvalds1da177e2005-04-16 15:20:36 -07001992/*
1993 * Add data to an sk_buff
1994 */
Johannes Berg4df864c2017-06-16 14:29:21 +02001995void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
1996void *skb_put(struct sk_buff *skb, unsigned int len);
1997static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07001998{
Johannes Berg4df864c2017-06-16 14:29:21 +02001999 void *tmp = skb_tail_pointer(skb);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002000 SKB_LINEAR_ASSERT(skb);
2001 skb->tail += len;
2002 skb->len += len;
2003 return tmp;
2004}
2005
yuan linyude77b962017-06-18 22:48:17 +08002006static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2007{
2008 void *tmp = __skb_put(skb, len);
2009
2010 memset(tmp, 0, len);
2011 return tmp;
2012}
2013
2014static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2015 unsigned int len)
2016{
2017 void *tmp = __skb_put(skb, len);
2018
2019 memcpy(tmp, data, len);
2020 return tmp;
2021}
2022
2023static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2024{
2025 *(u8 *)__skb_put(skb, 1) = val;
2026}
2027
Johannes Berg83ad3572017-06-14 22:17:20 +02002028static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
Johannes Berge45a79d2017-05-24 09:07:47 +02002029{
Johannes Berg83ad3572017-06-14 22:17:20 +02002030 void *tmp = skb_put(skb, len);
Johannes Berge45a79d2017-05-24 09:07:47 +02002031
2032 memset(tmp, 0, len);
2033
2034 return tmp;
2035}
2036
Johannes Berg59ae1d12017-06-16 14:29:20 +02002037static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2038 unsigned int len)
2039{
2040 void *tmp = skb_put(skb, len);
2041
2042 memcpy(tmp, data, len);
2043
2044 return tmp;
2045}
2046
Johannes Berg634fef62017-06-16 14:29:24 +02002047static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2048{
2049 *(u8 *)skb_put(skb, 1) = val;
2050}
2051
Johannes Bergd58ff352017-06-16 14:29:23 +02002052void *skb_push(struct sk_buff *skb, unsigned int len);
2053static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002054{
2055 skb->data -= len;
2056 skb->len += len;
2057 return skb->data;
2058}
2059
Johannes Bergaf728682017-06-16 14:29:22 +02002060void *skb_pull(struct sk_buff *skb, unsigned int len);
2061static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002062{
2063 skb->len -= len;
2064 BUG_ON(skb->len < skb->data_len);
2065 return skb->data += len;
2066}
2067
Johannes Bergaf728682017-06-16 14:29:22 +02002068static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
David S. Miller47d29642010-05-02 02:21:44 -07002069{
2070 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2071}
2072
Johannes Bergaf728682017-06-16 14:29:22 +02002073void *__pskb_pull_tail(struct sk_buff *skb, int delta);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002074
Johannes Bergaf728682017-06-16 14:29:22 +02002075static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002076{
2077 if (len > skb_headlen(skb) &&
Gerrit Renker987c4022008-08-11 18:17:17 -07002078 !__pskb_pull_tail(skb, len - skb_headlen(skb)))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002079 return NULL;
2080 skb->len -= len;
2081 return skb->data += len;
2082}
2083
Johannes Bergaf728682017-06-16 14:29:22 +02002084static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002085{
2086 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2087}
2088
2089static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2090{
2091 if (likely(len <= skb_headlen(skb)))
2092 return 1;
2093 if (unlikely(len > skb->len))
2094 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002095 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002096}
2097
Eric Dumazetc8c8b122016-12-07 09:19:33 -08002098void skb_condense(struct sk_buff *skb);
2099
Linus Torvalds1da177e2005-04-16 15:20:36 -07002100/**
2101 * skb_headroom - bytes at buffer head
2102 * @skb: buffer to check
2103 *
2104 * Return the number of bytes of free space at the head of an &sk_buff.
2105 */
Chuck Leverc2636b42007-10-23 21:07:32 -07002106static inline unsigned int skb_headroom(const struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002107{
2108 return skb->data - skb->head;
2109}
2110
2111/**
2112 * skb_tailroom - bytes at buffer end
2113 * @skb: buffer to check
2114 *
2115 * Return the number of bytes of free space at the tail of an sk_buff
2116 */
2117static inline int skb_tailroom(const struct sk_buff *skb)
2118{
Arnaldo Carvalho de Melo4305b542007-04-19 20:43:29 -07002119 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002120}
2121
2122/**
Eric Dumazeta21d4572012-04-10 20:30:48 +00002123 * skb_availroom - bytes at buffer end
2124 * @skb: buffer to check
2125 *
2126 * Return the number of bytes of free space at the tail of an sk_buff
2127 * allocated by sk_stream_alloc()
2128 */
2129static inline int skb_availroom(const struct sk_buff *skb)
2130{
Eric Dumazet16fad692013-03-14 05:40:32 +00002131 if (skb_is_nonlinear(skb))
2132 return 0;
2133
2134 return skb->end - skb->tail - skb->reserved_tailroom;
Eric Dumazeta21d4572012-04-10 20:30:48 +00002135}
2136
2137/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002138 * skb_reserve - adjust headroom
2139 * @skb: buffer to alter
2140 * @len: bytes to move
2141 *
2142 * Increase the headroom of an empty &sk_buff by reducing the tail
2143 * room. This is only allowed for an empty buffer.
2144 */
David S. Miller8243126c2006-01-17 02:54:21 -08002145static inline void skb_reserve(struct sk_buff *skb, int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002146{
2147 skb->data += len;
2148 skb->tail += len;
2149}
2150
Benjamin Poirier1837b2e2016-02-29 15:03:33 -08002151/**
2152 * skb_tailroom_reserve - adjust reserved_tailroom
2153 * @skb: buffer to alter
2154 * @mtu: maximum amount of headlen permitted
2155 * @needed_tailroom: minimum amount of reserved_tailroom
2156 *
2157 * Set reserved_tailroom so that headlen can be as large as possible but
2158 * not larger than mtu and tailroom cannot be smaller than
2159 * needed_tailroom.
2160 * The required headroom should already have been reserved before using
2161 * this function.
2162 */
2163static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2164 unsigned int needed_tailroom)
2165{
2166 SKB_LINEAR_ASSERT(skb);
2167 if (mtu < skb_tailroom(skb) - needed_tailroom)
2168 /* use at most mtu */
2169 skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2170 else
2171 /* use up to all available space */
2172 skb->reserved_tailroom = needed_tailroom;
2173}
2174
Tom Herbert8bce6d72014-09-29 20:22:29 -07002175#define ENCAP_TYPE_ETHER 0
2176#define ENCAP_TYPE_IPPROTO 1
2177
2178static inline void skb_set_inner_protocol(struct sk_buff *skb,
2179 __be16 protocol)
2180{
2181 skb->inner_protocol = protocol;
2182 skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2183}
2184
2185static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2186 __u8 ipproto)
2187{
2188 skb->inner_ipproto = ipproto;
2189 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2190}
2191
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002192static inline void skb_reset_inner_headers(struct sk_buff *skb)
2193{
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002194 skb->inner_mac_header = skb->mac_header;
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002195 skb->inner_network_header = skb->network_header;
2196 skb->inner_transport_header = skb->transport_header;
2197}
2198
Jiri Pirko0b5c9db2011-06-10 06:56:58 +00002199static inline void skb_reset_mac_len(struct sk_buff *skb)
2200{
2201 skb->mac_len = skb->network_header - skb->mac_header;
2202}
2203
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002204static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2205 *skb)
2206{
2207 return skb->head + skb->inner_transport_header;
2208}
2209
Tom Herbert55dc5a92015-12-14 11:19:40 -08002210static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2211{
2212 return skb_inner_transport_header(skb) - skb->data;
2213}
2214
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002215static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2216{
2217 skb->inner_transport_header = skb->data - skb->head;
2218}
2219
2220static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2221 const int offset)
2222{
2223 skb_reset_inner_transport_header(skb);
2224 skb->inner_transport_header += offset;
2225}
2226
2227static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2228{
2229 return skb->head + skb->inner_network_header;
2230}
2231
2232static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2233{
2234 skb->inner_network_header = skb->data - skb->head;
2235}
2236
2237static inline void skb_set_inner_network_header(struct sk_buff *skb,
2238 const int offset)
2239{
2240 skb_reset_inner_network_header(skb);
2241 skb->inner_network_header += offset;
2242}
2243
Pravin B Shelaraefbd2b2013-03-07 13:21:46 +00002244static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2245{
2246 return skb->head + skb->inner_mac_header;
2247}
2248
2249static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2250{
2251 skb->inner_mac_header = skb->data - skb->head;
2252}
2253
2254static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2255 const int offset)
2256{
2257 skb_reset_inner_mac_header(skb);
2258 skb->inner_mac_header += offset;
2259}
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002260static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2261{
Cong Wang35d04612013-05-29 15:16:05 +08002262 return skb->transport_header != (typeof(skb->transport_header))~0U;
Eric Dumazetfda55ec2013-01-07 09:28:21 +00002263}
2264
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002265static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2266{
2267 return skb->head + skb->transport_header;
2268}
2269
2270static inline void skb_reset_transport_header(struct sk_buff *skb)
2271{
2272 skb->transport_header = skb->data - skb->head;
2273}
2274
2275static inline void skb_set_transport_header(struct sk_buff *skb,
2276 const int offset)
2277{
2278 skb_reset_transport_header(skb);
2279 skb->transport_header += offset;
2280}
2281
2282static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2283{
2284 return skb->head + skb->network_header;
2285}
2286
2287static inline void skb_reset_network_header(struct sk_buff *skb)
2288{
2289 skb->network_header = skb->data - skb->head;
2290}
2291
2292static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2293{
2294 skb_reset_network_header(skb);
2295 skb->network_header += offset;
2296}
2297
2298static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2299{
2300 return skb->head + skb->mac_header;
2301}
2302
Amir Vadaiea6da4f2017-02-07 09:56:06 +02002303static inline int skb_mac_offset(const struct sk_buff *skb)
2304{
2305 return skb_mac_header(skb) - skb->data;
2306}
2307
Daniel Borkmann0daf4342017-07-02 02:13:25 +02002308static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2309{
2310 return skb->network_header - skb->mac_header;
2311}
2312
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002313static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2314{
Cong Wang35d04612013-05-29 15:16:05 +08002315 return skb->mac_header != (typeof(skb->mac_header))~0U;
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002316}
2317
2318static inline void skb_reset_mac_header(struct sk_buff *skb)
2319{
2320 skb->mac_header = skb->data - skb->head;
2321}
2322
2323static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2324{
2325 skb_reset_mac_header(skb);
2326 skb->mac_header += offset;
2327}
2328
Timo Teräs0e3da5b2013-12-16 11:02:09 +02002329static inline void skb_pop_mac_header(struct sk_buff *skb)
2330{
2331 skb->mac_header = skb->network_header;
2332}
2333
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002334static inline void skb_probe_transport_header(struct sk_buff *skb,
2335 const int offset_hint)
2336{
2337 struct flow_keys keys;
2338
2339 if (skb_transport_header_was_set(skb))
2340 return;
Tom Herbertcd79a232015-09-01 09:24:27 -07002341 else if (skb_flow_dissect_flow_keys(skb, &keys, 0))
Tom Herbert42aecaa2015-06-04 09:16:39 -07002342 skb_set_transport_header(skb, keys.control.thoff);
Ying Xuefbbdb8f2013-03-27 16:46:06 +00002343 else
2344 skb_set_transport_header(skb, offset_hint);
2345}
2346
Eric Dumazet03606892012-02-23 10:55:02 +00002347static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2348{
2349 if (skb_mac_header_was_set(skb)) {
2350 const unsigned char *old_mac = skb_mac_header(skb);
2351
2352 skb_set_mac_header(skb, -skb->mac_len);
2353 memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2354 }
2355}
2356
Michał Mirosław04fb4512010-12-14 15:24:08 +00002357static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2358{
2359 return skb->csum_start - skb_headroom(skb);
2360}
2361
Alexander Duyck08b64fc2016-02-05 15:27:49 -08002362static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2363{
2364 return skb->head + skb->csum_start;
2365}
2366
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002367static inline int skb_transport_offset(const struct sk_buff *skb)
2368{
2369 return skb_transport_header(skb) - skb->data;
2370}
2371
2372static inline u32 skb_network_header_len(const struct sk_buff *skb)
2373{
2374 return skb->transport_header - skb->network_header;
2375}
2376
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002377static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2378{
2379 return skb->inner_transport_header - skb->inner_network_header;
2380}
2381
Arnaldo Carvalho de Melo2e07fa92007-04-10 21:22:35 -07002382static inline int skb_network_offset(const struct sk_buff *skb)
2383{
2384 return skb_network_header(skb) - skb->data;
2385}
Arnaldo Carvalho de Melo48d49d0c2007-03-10 12:30:58 -03002386
Joseph Gasparakis6a674e92012-12-07 14:14:14 +00002387static inline int skb_inner_network_offset(const struct sk_buff *skb)
2388{
2389 return skb_inner_network_header(skb) - skb->data;
2390}
2391
Changli Gaof9599ce2010-08-04 04:43:44 +00002392static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2393{
2394 return pskb_may_pull(skb, skb_network_offset(skb) + len);
2395}
2396
Linus Torvalds1da177e2005-04-16 15:20:36 -07002397/*
2398 * CPUs often take a performance hit when accessing unaligned memory
2399 * locations. The actual performance hit varies, it can be small if the
2400 * hardware handles it or large if we have to take an exception and fix it
2401 * in software.
2402 *
2403 * Since an ethernet header is 14 bytes network drivers often end up with
2404 * the IP header at an unaligned offset. The IP header can be aligned by
2405 * shifting the start of the packet by 2 bytes. Drivers should do this
2406 * with:
2407 *
Tobias Klauser8660c122009-07-13 22:48:16 +00002408 * skb_reserve(skb, NET_IP_ALIGN);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002409 *
2410 * The downside to this alignment of the IP header is that the DMA is now
2411 * unaligned. On some architectures the cost of an unaligned DMA is high
2412 * and this cost outweighs the gains made by aligning the IP header.
Tobias Klauser8660c122009-07-13 22:48:16 +00002413 *
Linus Torvalds1da177e2005-04-16 15:20:36 -07002414 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2415 * to be overridden.
2416 */
2417#ifndef NET_IP_ALIGN
2418#define NET_IP_ALIGN 2
2419#endif
2420
Anton Blanchard025be812006-03-31 02:27:06 -08002421/*
2422 * The networking layer reserves some headroom in skb data (via
2423 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2424 * the header has to grow. In the default case, if the header has to grow
David S. Millerd6301d32009-02-08 19:24:13 -08002425 * 32 bytes or less we avoid the reallocation.
Anton Blanchard025be812006-03-31 02:27:06 -08002426 *
2427 * Unfortunately this headroom changes the DMA alignment of the resulting
2428 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2429 * on some architectures. An architecture can override this value,
2430 * perhaps setting it to a cacheline in size (since that will maintain
2431 * cacheline alignment of the DMA). It must be a power of 2.
2432 *
David S. Millerd6301d32009-02-08 19:24:13 -08002433 * Various parts of the networking layer expect at least 32 bytes of
Anton Blanchard025be812006-03-31 02:27:06 -08002434 * headroom, you should not reduce this.
Eric Dumazet5933dd22010-06-15 18:16:43 -07002435 *
2436 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2437 * to reduce average number of cache lines per packet.
2438 * get_rps_cpus() for example only access one 64 bytes aligned block :
Eric Dumazet18e8c132010-05-06 21:58:51 -07002439 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
Anton Blanchard025be812006-03-31 02:27:06 -08002440 */
2441#ifndef NET_SKB_PAD
Eric Dumazet5933dd22010-06-15 18:16:43 -07002442#define NET_SKB_PAD max(32, L1_CACHE_BYTES)
Anton Blanchard025be812006-03-31 02:27:06 -08002443#endif
2444
Joe Perches7965bd42013-09-26 14:48:15 -07002445int ___pskb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002446
Daniel Borkmann5293efe2016-08-18 01:00:39 +02002447static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002448{
Emmanuel Grumbachc4264f22011-05-21 19:46:09 +00002449 if (unlikely(skb_is_nonlinear(skb))) {
Herbert Xu3cc0e872006-06-09 16:13:38 -07002450 WARN_ON(1);
2451 return;
2452 }
Arnaldo Carvalho de Melo27a884d2007-04-19 20:29:13 -07002453 skb->len = len;
2454 skb_set_tail_pointer(skb, len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002455}
2456
Daniel Borkmann5293efe2016-08-18 01:00:39 +02002457static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2458{
2459 __skb_set_length(skb, len);
2460}
2461
Joe Perches7965bd42013-09-26 14:48:15 -07002462void skb_trim(struct sk_buff *skb, unsigned int len);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002463
2464static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2465{
Herbert Xu3cc0e872006-06-09 16:13:38 -07002466 if (skb->data_len)
2467 return ___pskb_trim(skb, len);
2468 __skb_trim(skb, len);
2469 return 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002470}
2471
2472static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2473{
2474 return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2475}
2476
2477/**
Herbert Xue9fa4f72006-08-13 20:12:58 -07002478 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2479 * @skb: buffer to alter
2480 * @len: new length
2481 *
2482 * This is identical to pskb_trim except that the caller knows that
2483 * the skb is not cloned so we should never get an error due to out-
2484 * of-memory.
2485 */
2486static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2487{
2488 int err = pskb_trim(skb, len);
2489 BUG_ON(err);
2490}
2491
Daniel Borkmann5293efe2016-08-18 01:00:39 +02002492static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2493{
2494 unsigned int diff = len - skb->len;
2495
2496 if (skb_tailroom(skb) < diff) {
2497 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2498 GFP_ATOMIC);
2499 if (ret)
2500 return ret;
2501 }
2502 __skb_set_length(skb, len);
2503 return 0;
2504}
2505
Herbert Xue9fa4f72006-08-13 20:12:58 -07002506/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002507 * skb_orphan - orphan a buffer
2508 * @skb: buffer to orphan
2509 *
2510 * If a buffer currently has an owner then we call the owner's
2511 * destructor function and make the @skb unowned. The buffer continues
2512 * to exist but is no longer charged to its former owner.
2513 */
2514static inline void skb_orphan(struct sk_buff *skb)
2515{
Eric Dumazetc34a7612013-07-30 16:11:15 -07002516 if (skb->destructor) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002517 skb->destructor(skb);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002518 skb->destructor = NULL;
2519 skb->sk = NULL;
Eric Dumazet376c7312013-08-01 11:43:08 -07002520 } else {
2521 BUG_ON(skb->sk);
Eric Dumazetc34a7612013-07-30 16:11:15 -07002522 }
Linus Torvalds1da177e2005-04-16 15:20:36 -07002523}
2524
2525/**
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002526 * skb_orphan_frags - orphan the frags contained in a buffer
2527 * @skb: buffer to orphan frags from
2528 * @gfp_mask: allocation mask for replacement pages
2529 *
2530 * For each frag in the SKB which needs a destructor (i.e. has an
2531 * owner) create a copy of that frag and release the original
2532 * page by calling the destructor.
2533 */
2534static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2535{
Willem de Bruijn1f8b9772017-08-03 16:29:41 -04002536 if (likely(!skb_zcopy(skb)))
2537 return 0;
2538 if (skb_uarg(skb)->callback == sock_zerocopy_callback)
2539 return 0;
2540 return skb_copy_ubufs(skb, gfp_mask);
2541}
2542
2543/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2544static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2545{
2546 if (likely(!skb_zcopy(skb)))
Michael S. Tsirkina353e0c2012-07-20 09:23:07 +00002547 return 0;
2548 return skb_copy_ubufs(skb, gfp_mask);
2549}
2550
2551/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002552 * __skb_queue_purge - empty a list
2553 * @list: list to empty
2554 *
2555 * Delete all buffers on an &sk_buff list. Each buffer is removed from
2556 * the list and one reference dropped. This function does not take the
2557 * list lock and the caller must hold the relevant locks to use it.
2558 */
Joe Perches7965bd42013-09-26 14:48:15 -07002559void skb_queue_purge(struct sk_buff_head *list);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002560static inline void __skb_queue_purge(struct sk_buff_head *list)
2561{
2562 struct sk_buff *skb;
2563 while ((skb = __skb_dequeue(list)) != NULL)
2564 kfree_skb(skb);
2565}
2566
Yaogong Wang9f5afea2016-09-07 14:49:28 -07002567void skb_rbtree_purge(struct rb_root *root);
2568
Joe Perches7965bd42013-09-26 14:48:15 -07002569void *netdev_alloc_frag(unsigned int fragsz);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002570
Joe Perches7965bd42013-09-26 14:48:15 -07002571struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2572 gfp_t gfp_mask);
Christoph Hellwig8af27452006-07-31 22:35:23 -07002573
2574/**
2575 * netdev_alloc_skb - allocate an skbuff for rx on a specific device
2576 * @dev: network device to receive on
2577 * @length: length to allocate
2578 *
2579 * Allocate a new &sk_buff and assign it a usage count of one. The
2580 * buffer has unspecified headroom built in. Users should allocate
2581 * the headroom they think they need without accounting for the
2582 * built in space. The built in space is used for optimisations.
2583 *
2584 * %NULL is returned if there is no free memory. Although this function
2585 * allocates memory it can be called from an interrupt.
2586 */
2587static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
Eric Dumazet6f532612012-05-18 05:12:12 +00002588 unsigned int length)
Christoph Hellwig8af27452006-07-31 22:35:23 -07002589{
2590 return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2591}
2592
Eric Dumazet6f532612012-05-18 05:12:12 +00002593/* legacy helper around __netdev_alloc_skb() */
2594static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2595 gfp_t gfp_mask)
2596{
2597 return __netdev_alloc_skb(NULL, length, gfp_mask);
2598}
2599
2600/* legacy helper around netdev_alloc_skb() */
2601static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2602{
2603 return netdev_alloc_skb(NULL, length);
2604}
2605
2606
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002607static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2608 unsigned int length, gfp_t gfp)
Eric Dumazet61321bb2009-10-07 17:11:23 +00002609{
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002610 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
Eric Dumazet61321bb2009-10-07 17:11:23 +00002611
2612 if (NET_IP_ALIGN && skb)
2613 skb_reserve(skb, NET_IP_ALIGN);
2614 return skb;
2615}
2616
Eric Dumazet4915a0d2011-07-11 20:08:34 -07002617static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2618 unsigned int length)
2619{
2620 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2621}
2622
Alexander Duyck181edb22015-05-06 21:12:03 -07002623static inline void skb_free_frag(void *addr)
2624{
Alexander Duyck8c2dd3e2017-01-10 16:58:06 -08002625 page_frag_free(addr);
Alexander Duyck181edb22015-05-06 21:12:03 -07002626}
2627
Alexander Duyckffde7322014-12-09 19:40:42 -08002628void *napi_alloc_frag(unsigned int fragsz);
Alexander Duyckfd11a832014-12-09 19:40:49 -08002629struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2630 unsigned int length, gfp_t gfp_mask);
2631static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2632 unsigned int length)
2633{
2634 return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2635}
Jesper Dangaard Brouer795bb1c2016-02-08 13:14:59 +01002636void napi_consume_skb(struct sk_buff *skb, int budget);
2637
2638void __kfree_skb_flush(void);
Jesper Dangaard Brouer15fad712016-02-08 13:15:04 +01002639void __kfree_skb_defer(struct sk_buff *skb);
Alexander Duyckffde7322014-12-09 19:40:42 -08002640
Florian Fainellibc6fc9f2013-08-30 15:36:14 +01002641/**
Alexander Duyck71dfda52014-11-11 09:26:34 -08002642 * __dev_alloc_pages - allocate page for network Rx
2643 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2644 * @order: size of the allocation
2645 *
2646 * Allocate a new page.
2647 *
2648 * %NULL is returned if there is no free memory.
2649*/
2650static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2651 unsigned int order)
2652{
2653 /* This piece of code contains several assumptions.
2654 * 1. This is for device Rx, therefor a cold page is preferred.
2655 * 2. The expectation is the user wants a compound page.
2656 * 3. If requesting a order 0 page it will not be compound
2657 * due to the check to see if order has a value in prep_new_page
2658 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2659 * code in gfp_to_alloc_flags that should be enforcing this.
2660 */
2661 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC;
2662
2663 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2664}
2665
2666static inline struct page *dev_alloc_pages(unsigned int order)
2667{
Neil Horman95829b32016-05-19 11:30:54 -04002668 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
Alexander Duyck71dfda52014-11-11 09:26:34 -08002669}
2670
2671/**
2672 * __dev_alloc_page - allocate a page for network Rx
2673 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2674 *
2675 * Allocate a new page.
2676 *
2677 * %NULL is returned if there is no free memory.
2678 */
2679static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2680{
2681 return __dev_alloc_pages(gfp_mask, 0);
2682}
2683
2684static inline struct page *dev_alloc_page(void)
2685{
Neil Horman95829b32016-05-19 11:30:54 -04002686 return dev_alloc_pages(0);
Alexander Duyck71dfda52014-11-11 09:26:34 -08002687}
2688
2689/**
Mel Gorman06140022012-07-31 16:44:24 -07002690 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2691 * @page: The page that was allocated from skb_alloc_page
2692 * @skb: The skb that may need pfmemalloc set
2693 */
2694static inline void skb_propagate_pfmemalloc(struct page *page,
2695 struct sk_buff *skb)
2696{
Michal Hocko2f064f32015-08-21 14:11:51 -07002697 if (page_is_pfmemalloc(page))
Mel Gorman06140022012-07-31 16:44:24 -07002698 skb->pfmemalloc = true;
2699}
2700
Eric Dumazet564824b2010-10-11 19:05:25 +00002701/**
Masanari Iidae2278672014-02-18 22:54:36 +09002702 * skb_frag_page - retrieve the page referred to by a paged fragment
Ian Campbell131ea662011-08-19 06:25:00 +00002703 * @frag: the paged fragment
2704 *
2705 * Returns the &struct page associated with @frag.
2706 */
2707static inline struct page *skb_frag_page(const skb_frag_t *frag)
2708{
Ian Campbella8605c62011-10-19 23:01:49 +00002709 return frag->page.p;
Ian Campbell131ea662011-08-19 06:25:00 +00002710}
2711
2712/**
2713 * __skb_frag_ref - take an addition reference on a paged fragment.
2714 * @frag: the paged fragment
2715 *
2716 * Takes an additional reference on the paged fragment @frag.
2717 */
2718static inline void __skb_frag_ref(skb_frag_t *frag)
2719{
2720 get_page(skb_frag_page(frag));
2721}
2722
2723/**
2724 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2725 * @skb: the buffer
2726 * @f: the fragment offset.
2727 *
2728 * Takes an additional reference on the @f'th paged fragment of @skb.
2729 */
2730static inline void skb_frag_ref(struct sk_buff *skb, int f)
2731{
2732 __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2733}
2734
2735/**
2736 * __skb_frag_unref - release a reference on a paged fragment.
2737 * @frag: the paged fragment
2738 *
2739 * Releases a reference on the paged fragment @frag.
2740 */
2741static inline void __skb_frag_unref(skb_frag_t *frag)
2742{
2743 put_page(skb_frag_page(frag));
2744}
2745
2746/**
2747 * skb_frag_unref - release a reference on a paged fragment of an skb.
2748 * @skb: the buffer
2749 * @f: the fragment offset
2750 *
2751 * Releases a reference on the @f'th paged fragment of @skb.
2752 */
2753static inline void skb_frag_unref(struct sk_buff *skb, int f)
2754{
2755 __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2756}
2757
2758/**
2759 * skb_frag_address - gets the address of the data contained in a paged fragment
2760 * @frag: the paged fragment buffer
2761 *
2762 * Returns the address of the data within @frag. The page must already
2763 * be mapped.
2764 */
2765static inline void *skb_frag_address(const skb_frag_t *frag)
2766{
2767 return page_address(skb_frag_page(frag)) + frag->page_offset;
2768}
2769
2770/**
2771 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2772 * @frag: the paged fragment buffer
2773 *
2774 * Returns the address of the data within @frag. Checks that the page
2775 * is mapped and returns %NULL otherwise.
2776 */
2777static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2778{
2779 void *ptr = page_address(skb_frag_page(frag));
2780 if (unlikely(!ptr))
2781 return NULL;
2782
2783 return ptr + frag->page_offset;
2784}
2785
2786/**
2787 * __skb_frag_set_page - sets the page contained in a paged fragment
2788 * @frag: the paged fragment
2789 * @page: the page to set
2790 *
2791 * Sets the fragment @frag to contain @page.
2792 */
2793static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2794{
Ian Campbella8605c62011-10-19 23:01:49 +00002795 frag->page.p = page;
Ian Campbell131ea662011-08-19 06:25:00 +00002796}
2797
2798/**
2799 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2800 * @skb: the buffer
2801 * @f: the fragment offset
2802 * @page: the page to set
2803 *
2804 * Sets the @f'th fragment of @skb to contain @page.
2805 */
2806static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2807 struct page *page)
2808{
2809 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2810}
2811
Eric Dumazet400dfd32013-10-17 16:27:07 -07002812bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2813
Ian Campbell131ea662011-08-19 06:25:00 +00002814/**
2815 * skb_frag_dma_map - maps a paged fragment via the DMA API
Marcos Paulo de Souzaf83347d2011-10-31 15:11:45 +00002816 * @dev: the device to map the fragment to
Ian Campbell131ea662011-08-19 06:25:00 +00002817 * @frag: the paged fragment to map
2818 * @offset: the offset within the fragment (starting at the
2819 * fragment's own offset)
2820 * @size: the number of bytes to map
Mauro Carvalho Chehab771b00a2017-05-12 09:19:29 -03002821 * @dir: the direction of the mapping (``PCI_DMA_*``)
Ian Campbell131ea662011-08-19 06:25:00 +00002822 *
2823 * Maps the page associated with @frag to @device.
2824 */
2825static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2826 const skb_frag_t *frag,
2827 size_t offset, size_t size,
2828 enum dma_data_direction dir)
2829{
2830 return dma_map_page(dev, skb_frag_page(frag),
2831 frag->page_offset + offset, size, dir);
2832}
2833
Eric Dumazet117632e2011-12-03 21:39:53 +00002834static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2835 gfp_t gfp_mask)
2836{
2837 return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2838}
2839
Octavian Purdilabad93e92014-06-12 01:36:26 +03002840
2841static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2842 gfp_t gfp_mask)
2843{
2844 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
2845}
2846
2847
Ian Campbell131ea662011-08-19 06:25:00 +00002848/**
Patrick McHardy334a8132007-06-25 04:35:20 -07002849 * skb_clone_writable - is the header of a clone writable
2850 * @skb: buffer to check
2851 * @len: length up to which to write
2852 *
2853 * Returns true if modifying the header part of the cloned buffer
2854 * does not requires the data to be copied.
2855 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04002856static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
Patrick McHardy334a8132007-06-25 04:35:20 -07002857{
2858 return !skb_header_cloned(skb) &&
2859 skb_headroom(skb) + len <= skb->hdr_len;
2860}
2861
Daniel Borkmann36976492016-02-19 23:05:25 +01002862static inline int skb_try_make_writable(struct sk_buff *skb,
2863 unsigned int write_len)
2864{
2865 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
2866 pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2867}
2868
Herbert Xud9cc2042007-09-16 16:21:16 -07002869static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
2870 int cloned)
2871{
2872 int delta = 0;
2873
Herbert Xud9cc2042007-09-16 16:21:16 -07002874 if (headroom > skb_headroom(skb))
2875 delta = headroom - skb_headroom(skb);
2876
2877 if (delta || cloned)
2878 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
2879 GFP_ATOMIC);
2880 return 0;
2881}
2882
Patrick McHardy334a8132007-06-25 04:35:20 -07002883/**
Linus Torvalds1da177e2005-04-16 15:20:36 -07002884 * skb_cow - copy header of skb when it is required
2885 * @skb: buffer to cow
2886 * @headroom: needed headroom
2887 *
2888 * If the skb passed lacks sufficient headroom or its data part
2889 * is shared, data is reallocated. If reallocation fails, an error
2890 * is returned and original skb is not changed.
2891 *
2892 * The result is skb with writable area skb->head...skb->tail
2893 * and at least @headroom of space at head.
2894 */
2895static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
2896{
Herbert Xud9cc2042007-09-16 16:21:16 -07002897 return __skb_cow(skb, headroom, skb_cloned(skb));
2898}
Linus Torvalds1da177e2005-04-16 15:20:36 -07002899
Herbert Xud9cc2042007-09-16 16:21:16 -07002900/**
2901 * skb_cow_head - skb_cow but only making the head writable
2902 * @skb: buffer to cow
2903 * @headroom: needed headroom
2904 *
2905 * This function is identical to skb_cow except that we replace the
2906 * skb_cloned check by skb_header_cloned. It should be used when
2907 * you only need to push on some header and do not need to modify
2908 * the data.
2909 */
2910static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
2911{
2912 return __skb_cow(skb, headroom, skb_header_cloned(skb));
Linus Torvalds1da177e2005-04-16 15:20:36 -07002913}
2914
2915/**
2916 * skb_padto - pad an skbuff up to a minimal size
2917 * @skb: buffer to pad
2918 * @len: minimal length
2919 *
2920 * Pads up a buffer to ensure the trailing bytes exist and are
2921 * blanked. If the buffer already contains sufficient data it
Herbert Xu5b057c62006-06-23 02:06:41 -07002922 * is untouched. Otherwise it is extended. Returns zero on
2923 * success. The skb is freed on error.
Linus Torvalds1da177e2005-04-16 15:20:36 -07002924 */
Herbert Xu5b057c62006-06-23 02:06:41 -07002925static inline int skb_padto(struct sk_buff *skb, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002926{
2927 unsigned int size = skb->len;
2928 if (likely(size >= len))
Herbert Xu5b057c62006-06-23 02:06:41 -07002929 return 0;
Gerrit Renker987c4022008-08-11 18:17:17 -07002930 return skb_pad(skb, len - size);
Linus Torvalds1da177e2005-04-16 15:20:36 -07002931}
2932
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002933/**
2934 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2935 * @skb: buffer to pad
2936 * @len: minimal length
Florian Fainellicd0a1372017-08-22 15:12:14 -07002937 * @free_on_error: free buffer on error
2938 *
2939 * Pads up a buffer to ensure the trailing bytes exist and are
2940 * blanked. If the buffer already contains sufficient data it
2941 * is untouched. Otherwise it is extended. Returns zero on
2942 * success. The skb is freed on error if @free_on_error is true.
2943 */
2944static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
2945 bool free_on_error)
2946{
2947 unsigned int size = skb->len;
2948
2949 if (unlikely(size < len)) {
2950 len -= size;
2951 if (__skb_pad(skb, len, free_on_error))
2952 return -ENOMEM;
2953 __skb_put(skb, len);
2954 }
2955 return 0;
2956}
2957
2958/**
2959 * skb_put_padto - increase size and pad an skbuff up to a minimal size
2960 * @skb: buffer to pad
2961 * @len: minimal length
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002962 *
2963 * Pads up a buffer to ensure the trailing bytes exist and are
2964 * blanked. If the buffer already contains sufficient data it
2965 * is untouched. Otherwise it is extended. Returns zero on
2966 * success. The skb is freed on error.
2967 */
2968static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
2969{
Florian Fainellicd0a1372017-08-22 15:12:14 -07002970 return __skb_put_padto(skb, len, true);
Alexander Duyck9c0c1122014-12-03 08:17:33 -08002971}
2972
Linus Torvalds1da177e2005-04-16 15:20:36 -07002973static inline int skb_add_data(struct sk_buff *skb,
Al Viroaf2b0402014-11-27 21:44:24 -05002974 struct iov_iter *from, int copy)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002975{
2976 const int off = skb->len;
2977
2978 if (skb->ip_summed == CHECKSUM_NONE) {
Al Viroaf2b0402014-11-27 21:44:24 -05002979 __wsum csum = 0;
Al Viro15e6cb42016-11-01 22:42:45 -04002980 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
2981 &csum, from)) {
Linus Torvalds1da177e2005-04-16 15:20:36 -07002982 skb->csum = csum_block_add(skb->csum, csum, off);
2983 return 0;
2984 }
Al Viro15e6cb42016-11-01 22:42:45 -04002985 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
Linus Torvalds1da177e2005-04-16 15:20:36 -07002986 return 0;
2987
2988 __skb_trim(skb, off);
2989 return -EFAULT;
2990}
2991
Eric Dumazet38ba0a62012-04-23 17:48:27 +00002992static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
2993 const struct page *page, int off)
Linus Torvalds1da177e2005-04-16 15:20:36 -07002994{
Willem de Bruijn1f8b9772017-08-03 16:29:41 -04002995 if (skb_zcopy(skb))
2996 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07002997 if (i) {
Eric Dumazet9e903e02011-10-18 21:00:24 +00002998 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
Linus Torvalds1da177e2005-04-16 15:20:36 -07002999
Ian Campbellea2ab692011-08-22 23:44:58 +00003000 return page == skb_frag_page(frag) &&
Eric Dumazet9e903e02011-10-18 21:00:24 +00003001 off == frag->page_offset + skb_frag_size(frag);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003002 }
Eric Dumazet38ba0a62012-04-23 17:48:27 +00003003 return false;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003004}
3005
Herbert Xu364c6ba2006-06-09 16:10:40 -07003006static inline int __skb_linearize(struct sk_buff *skb)
3007{
3008 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3009}
3010
Linus Torvalds1da177e2005-04-16 15:20:36 -07003011/**
3012 * skb_linearize - convert paged skb to linear one
3013 * @skb: buffer to linarize
Linus Torvalds1da177e2005-04-16 15:20:36 -07003014 *
3015 * If there is no free memory -ENOMEM is returned, otherwise zero
3016 * is returned and the old skb data released.
3017 */
Herbert Xu364c6ba2006-06-09 16:10:40 -07003018static inline int skb_linearize(struct sk_buff *skb)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003019{
Herbert Xu364c6ba2006-06-09 16:10:40 -07003020 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3021}
3022
3023/**
Eric Dumazetcef401d2013-01-25 20:34:37 +00003024 * skb_has_shared_frag - can any frag be overwritten
3025 * @skb: buffer to test
3026 *
3027 * Return true if the skb has at least one frag that might be modified
3028 * by an external entity (as in vmsplice()/sendfile())
3029 */
3030static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3031{
Pravin B Shelarc9af6db2013-02-11 09:27:41 +00003032 return skb_is_nonlinear(skb) &&
3033 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
Eric Dumazetcef401d2013-01-25 20:34:37 +00003034}
3035
3036/**
Herbert Xu364c6ba2006-06-09 16:10:40 -07003037 * skb_linearize_cow - make sure skb is linear and writable
3038 * @skb: buffer to process
3039 *
3040 * If there is no free memory -ENOMEM is returned, otherwise zero
3041 * is returned and the old skb data released.
3042 */
3043static inline int skb_linearize_cow(struct sk_buff *skb)
3044{
3045 return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3046 __skb_linearize(skb) : 0;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003047}
3048
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003049static __always_inline void
3050__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3051 unsigned int off)
3052{
3053 if (skb->ip_summed == CHECKSUM_COMPLETE)
3054 skb->csum = csum_block_sub(skb->csum,
3055 csum_partial(start, len, 0), off);
3056 else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3057 skb_checksum_start_offset(skb) < 0)
3058 skb->ip_summed = CHECKSUM_NONE;
3059}
3060
Linus Torvalds1da177e2005-04-16 15:20:36 -07003061/**
3062 * skb_postpull_rcsum - update checksum for received skb after pull
3063 * @skb: buffer to update
3064 * @start: start of data before pull
3065 * @len: length of data pulled
3066 *
3067 * After doing a pull on a received packet, you need to call this to
Patrick McHardy84fa7932006-08-29 16:44:56 -07003068 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3069 * CHECKSUM_NONE so that it can be recomputed from scratch.
Linus Torvalds1da177e2005-04-16 15:20:36 -07003070 */
Linus Torvalds1da177e2005-04-16 15:20:36 -07003071static inline void skb_postpull_rcsum(struct sk_buff *skb,
Herbert Xucbb042f2006-03-20 22:43:56 -08003072 const void *start, unsigned int len)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003073{
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003074 __skb_postpull_rcsum(skb, start, len, 0);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003075}
3076
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003077static __always_inline void
3078__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3079 unsigned int off)
3080{
3081 if (skb->ip_summed == CHECKSUM_COMPLETE)
3082 skb->csum = csum_block_add(skb->csum,
3083 csum_partial(start, len, 0), off);
3084}
Herbert Xucbb042f2006-03-20 22:43:56 -08003085
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003086/**
3087 * skb_postpush_rcsum - update checksum for received skb after push
3088 * @skb: buffer to update
3089 * @start: start of data after push
3090 * @len: length of data pushed
3091 *
3092 * After doing a push on a received packet, you need to call this to
3093 * update the CHECKSUM_COMPLETE checksum.
3094 */
Daniel Borkmannf8ffad62016-01-07 15:50:23 +01003095static inline void skb_postpush_rcsum(struct sk_buff *skb,
3096 const void *start, unsigned int len)
3097{
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003098 __skb_postpush_rcsum(skb, start, len, 0);
Daniel Borkmannf8ffad62016-01-07 15:50:23 +01003099}
3100
Johannes Bergaf728682017-06-16 14:29:22 +02003101void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
Daniel Borkmann479ffccc2016-08-05 00:11:12 +02003102
David S. Miller7ce5a272013-12-02 17:26:05 -05003103/**
WANG Cong82a31b92016-06-30 10:15:22 -07003104 * skb_push_rcsum - push skb and update receive checksum
3105 * @skb: buffer to update
3106 * @len: length of data pulled
3107 *
3108 * This function performs an skb_push on the packet and updates
3109 * the CHECKSUM_COMPLETE checksum. It should be used on
3110 * receive path processing instead of skb_push unless you know
3111 * that the checksum difference is zero (e.g., a valid IP header)
3112 * or you are setting ip_summed to CHECKSUM_NONE.
3113 */
Johannes Bergd58ff352017-06-16 14:29:23 +02003114static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
WANG Cong82a31b92016-06-30 10:15:22 -07003115{
3116 skb_push(skb, len);
3117 skb_postpush_rcsum(skb, skb->data, len);
3118 return skb->data;
3119}
3120
3121/**
David S. Miller7ce5a272013-12-02 17:26:05 -05003122 * pskb_trim_rcsum - trim received skb and update checksum
3123 * @skb: buffer to trim
3124 * @len: new length
3125 *
3126 * This is exactly the same as pskb_trim except that it ensures the
3127 * checksum of received packets are still valid after the operation.
3128 */
3129
3130static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3131{
3132 if (likely(len >= skb->len))
3133 return 0;
3134 if (skb->ip_summed == CHECKSUM_COMPLETE)
3135 skb->ip_summed = CHECKSUM_NONE;
3136 return __pskb_trim(skb, len);
3137}
3138
Daniel Borkmann5293efe2016-08-18 01:00:39 +02003139static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3140{
3141 if (skb->ip_summed == CHECKSUM_COMPLETE)
3142 skb->ip_summed = CHECKSUM_NONE;
3143 __skb_trim(skb, len);
3144 return 0;
3145}
3146
3147static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3148{
3149 if (skb->ip_summed == CHECKSUM_COMPLETE)
3150 skb->ip_summed = CHECKSUM_NONE;
3151 return __skb_grow(skb, len);
3152}
3153
Linus Torvalds1da177e2005-04-16 15:20:36 -07003154#define skb_queue_walk(queue, skb) \
3155 for (skb = (queue)->next; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07003156 skb != (struct sk_buff *)(queue); \
Linus Torvalds1da177e2005-04-16 15:20:36 -07003157 skb = skb->next)
3158
James Chapman46f89142007-04-30 00:07:31 -07003159#define skb_queue_walk_safe(queue, skb, tmp) \
3160 for (skb = (queue)->next, tmp = skb->next; \
3161 skb != (struct sk_buff *)(queue); \
3162 skb = tmp, tmp = skb->next)
3163
David S. Miller1164f522008-09-23 00:49:44 -07003164#define skb_queue_walk_from(queue, skb) \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07003165 for (; skb != (struct sk_buff *)(queue); \
David S. Miller1164f522008-09-23 00:49:44 -07003166 skb = skb->next)
3167
3168#define skb_queue_walk_from_safe(queue, skb, tmp) \
3169 for (tmp = skb->next; \
3170 skb != (struct sk_buff *)(queue); \
3171 skb = tmp, tmp = skb->next)
3172
Stephen Hemminger300ce172005-10-30 13:47:34 -08003173#define skb_queue_reverse_walk(queue, skb) \
3174 for (skb = (queue)->prev; \
Linus Torvaldsa1e48912011-05-22 16:51:43 -07003175 skb != (struct sk_buff *)(queue); \
Stephen Hemminger300ce172005-10-30 13:47:34 -08003176 skb = skb->prev)
3177
David S. Miller686a2952011-01-20 22:47:32 -08003178#define skb_queue_reverse_walk_safe(queue, skb, tmp) \
3179 for (skb = (queue)->prev, tmp = skb->prev; \
3180 skb != (struct sk_buff *)(queue); \
3181 skb = tmp, tmp = skb->prev)
3182
3183#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \
3184 for (tmp = skb->prev; \
3185 skb != (struct sk_buff *)(queue); \
3186 skb = tmp, tmp = skb->prev)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003187
David S. Miller21dc3302010-08-23 00:13:46 -07003188static inline bool skb_has_frag_list(const struct sk_buff *skb)
David S. Milleree039872009-06-09 00:17:13 -07003189{
3190 return skb_shinfo(skb)->frag_list != NULL;
3191}
3192
3193static inline void skb_frag_list_init(struct sk_buff *skb)
3194{
3195 skb_shinfo(skb)->frag_list = NULL;
3196}
3197
David S. Milleree039872009-06-09 00:17:13 -07003198#define skb_walk_frags(skb, iter) \
3199 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3200
Rainer Weikusatea3793e2015-12-06 21:11:34 +00003201
3202int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3203 const struct sk_buff *skb);
Paolo Abeni65101ae2017-05-16 11:20:13 +02003204struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3205 struct sk_buff_head *queue,
3206 unsigned int flags,
3207 void (*destructor)(struct sock *sk,
3208 struct sk_buff *skb),
3209 int *peeked, int *off, int *err,
3210 struct sk_buff **last);
Rainer Weikusatea3793e2015-12-06 21:11:34 +00003211struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
Paolo Abeni7c13f972016-11-04 11:28:59 +01003212 void (*destructor)(struct sock *sk,
3213 struct sk_buff *skb),
Rainer Weikusatea3793e2015-12-06 21:11:34 +00003214 int *peeked, int *off, int *err,
3215 struct sk_buff **last);
Joe Perches7965bd42013-09-26 14:48:15 -07003216struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
Paolo Abeni7c13f972016-11-04 11:28:59 +01003217 void (*destructor)(struct sock *sk,
3218 struct sk_buff *skb),
Joe Perches7965bd42013-09-26 14:48:15 -07003219 int *peeked, int *off, int *err);
3220struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3221 int *err);
3222unsigned int datagram_poll(struct file *file, struct socket *sock,
3223 struct poll_table_struct *wait);
Al Viroc0371da2014-11-24 10:42:55 -05003224int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3225 struct iov_iter *to, int size);
David S. Miller51f3d022014-11-05 16:46:40 -05003226static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3227 struct msghdr *msg, int size)
3228{
Al Viroe5a4b0b2014-11-24 18:17:55 -05003229 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
David S. Miller51f3d022014-11-05 16:46:40 -05003230}
Al Viroe5a4b0b2014-11-24 18:17:55 -05003231int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3232 struct msghdr *msg);
Al Viro3a654f92014-06-19 14:15:22 -04003233int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3234 struct iov_iter *from, int len);
Al Viro3a654f92014-06-19 14:15:22 -04003235int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
Joe Perches7965bd42013-09-26 14:48:15 -07003236void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
samanthakumar627d2d62016-04-05 12:41:16 -04003237void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3238static inline void skb_free_datagram_locked(struct sock *sk,
3239 struct sk_buff *skb)
3240{
3241 __skb_free_datagram_locked(sk, skb, 0);
3242}
Joe Perches7965bd42013-09-26 14:48:15 -07003243int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
Joe Perches7965bd42013-09-26 14:48:15 -07003244int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3245int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3246__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3247 int len, __wsum csum);
Hannes Frederic Sowaa60e3cc2015-05-21 17:00:00 +02003248int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
Joe Perches7965bd42013-09-26 14:48:15 -07003249 struct pipe_inode_info *pipe, unsigned int len,
Al Viro25869262016-09-17 21:02:10 -04003250 unsigned int flags);
Tom Herbert20bf50d2017-07-28 16:22:42 -07003251int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3252 int len);
3253int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len);
Joe Perches7965bd42013-09-26 14:48:15 -07003254void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
Thomas Grafaf2806f2013-12-13 15:22:17 +01003255unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
Zoltan Kiss36d5fe62014-03-26 22:37:45 +00003256int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3257 int len, int hlen);
Joe Perches7965bd42013-09-26 14:48:15 -07003258void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3259int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3260void skb_scrub_packet(struct sk_buff *skb, bool xnet);
Florian Westphalde960aa2014-01-26 10:58:16 +01003261unsigned int skb_gso_transport_seglen(const struct sk_buff *skb);
Marcelo Ricardo Leitnerae7ef812016-06-02 15:05:41 -03003262bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu);
Joe Perches7965bd42013-09-26 14:48:15 -07003263struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
Vlad Yasevich0d5501c2014-08-08 14:42:13 -04003264struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
Jiri Pirkoe2195122014-11-19 14:05:01 +01003265int skb_ensure_writable(struct sk_buff *skb, int write_len);
Shmulik Ladkanibfca4c52016-09-19 19:11:09 +03003266int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
Jiri Pirko93515d52014-11-19 14:05:02 +01003267int skb_vlan_pop(struct sk_buff *skb);
3268int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
Sowmini Varadhan6fa01cc2016-04-22 18:36:35 -07003269struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3270 gfp_t gfp);
Arnaldo Carvalho de Melo20380732005-08-16 02:18:02 -03003271
Al Viro6ce8e9c2014-04-06 21:25:44 -04003272static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3273{
Al Viro3073f072017-02-17 23:13:25 -05003274 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
Al Viro6ce8e9c2014-04-06 21:25:44 -04003275}
3276
Al Viro7eab8d92014-04-06 21:51:23 -04003277static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3278{
Al Viroe5a4b0b2014-11-24 18:17:55 -05003279 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
Al Viro7eab8d92014-04-06 21:51:23 -04003280}
3281
Daniel Borkmann2817a332013-10-30 11:50:51 +01003282struct skb_checksum_ops {
3283 __wsum (*update)(const void *mem, int len, __wsum wsum);
3284 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3285};
3286
Davide Caratti96178132017-05-18 15:44:37 +02003287extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3288
Daniel Borkmann2817a332013-10-30 11:50:51 +01003289__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3290 __wsum csum, const struct skb_checksum_ops *ops);
3291__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3292 __wsum csum);
3293
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003294static inline void * __must_check
3295__skb_header_pointer(const struct sk_buff *skb, int offset,
3296 int len, void *data, int hlen, void *buffer)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003297{
Patrick McHardy55820ee2005-07-05 14:08:10 -07003298 if (hlen - offset >= len)
David S. Miller690e36e2014-08-23 12:13:41 -07003299 return data + offset;
Linus Torvalds1da177e2005-04-16 15:20:36 -07003300
David S. Miller690e36e2014-08-23 12:13:41 -07003301 if (!skb ||
3302 skb_copy_bits(skb, offset, buffer, len) < 0)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003303 return NULL;
3304
3305 return buffer;
3306}
3307
Eric Dumazet1e98a0f2015-06-12 19:31:32 -07003308static inline void * __must_check
3309skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
David S. Miller690e36e2014-08-23 12:13:41 -07003310{
3311 return __skb_header_pointer(skb, offset, len, skb->data,
3312 skb_headlen(skb), buffer);
3313}
3314
Daniel Borkmann4262e5c2013-12-06 11:36:16 +01003315/**
3316 * skb_needs_linearize - check if we need to linearize a given skb
3317 * depending on the given device features.
3318 * @skb: socket buffer to check
3319 * @features: net device features
3320 *
3321 * Returns true if either:
3322 * 1. skb has frag_list and the device doesn't support FRAGLIST, or
3323 * 2. skb is fragmented and the device does not support SG.
3324 */
3325static inline bool skb_needs_linearize(struct sk_buff *skb,
3326 netdev_features_t features)
3327{
3328 return skb_is_nonlinear(skb) &&
3329 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3330 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3331}
3332
Arnaldo Carvalho de Melod626f622007-03-27 18:55:52 -03003333static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3334 void *to,
3335 const unsigned int len)
3336{
3337 memcpy(to, skb->data, len);
3338}
3339
3340static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3341 const int offset, void *to,
3342 const unsigned int len)
3343{
3344 memcpy(to, skb->data + offset, len);
3345}
3346
Arnaldo Carvalho de Melo27d7ff42007-03-31 11:55:19 -03003347static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3348 const void *from,
3349 const unsigned int len)
3350{
3351 memcpy(skb->data, from, len);
3352}
3353
3354static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3355 const int offset,
3356 const void *from,
3357 const unsigned int len)
3358{
3359 memcpy(skb->data + offset, from, len);
3360}
3361
Joe Perches7965bd42013-09-26 14:48:15 -07003362void skb_init(void);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003363
Patrick Ohlyac45f602009-02-12 05:03:37 +00003364static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3365{
3366 return skb->tstamp;
3367}
3368
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003369/**
3370 * skb_get_timestamp - get timestamp from a skb
3371 * @skb: skb to get stamp from
3372 * @stamp: pointer to struct timeval to store stamp in
3373 *
3374 * Timestamps are stored in the skb as offsets to a base timestamp.
3375 * This function converts the offset back to a struct timeval and stores
3376 * it in stamp.
3377 */
Patrick Ohlyac45f602009-02-12 05:03:37 +00003378static inline void skb_get_timestamp(const struct sk_buff *skb,
3379 struct timeval *stamp)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003380{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003381 *stamp = ktime_to_timeval(skb->tstamp);
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003382}
3383
Patrick Ohlyac45f602009-02-12 05:03:37 +00003384static inline void skb_get_timestampns(const struct sk_buff *skb,
3385 struct timespec *stamp)
3386{
3387 *stamp = ktime_to_timespec(skb->tstamp);
3388}
3389
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003390static inline void __net_timestamp(struct sk_buff *skb)
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003391{
Eric Dumazetb7aa0bf2007-04-19 16:16:32 -07003392 skb->tstamp = ktime_get_real();
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003393}
3394
Stephen Hemminger164891a2007-04-23 22:26:16 -07003395static inline ktime_t net_timedelta(ktime_t t)
3396{
3397 return ktime_sub(ktime_get_real(), t);
3398}
3399
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07003400static inline ktime_t net_invalid_timestamp(void)
3401{
Thomas Gleixner8b0e1952016-12-25 12:30:41 +01003402 return 0;
Ilpo Järvinenb9ce2042007-06-15 15:08:43 -07003403}
Patrick McHardya61bbcf2005-08-14 17:24:31 -07003404
Daniel Borkmannde8f3a82017-09-25 02:25:51 +02003405static inline u8 skb_metadata_len(const struct sk_buff *skb)
3406{
3407 return skb_shinfo(skb)->meta_len;
3408}
3409
3410static inline void *skb_metadata_end(const struct sk_buff *skb)
3411{
3412 return skb_mac_header(skb);
3413}
3414
3415static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3416 const struct sk_buff *skb_b,
3417 u8 meta_len)
3418{
3419 const void *a = skb_metadata_end(skb_a);
3420 const void *b = skb_metadata_end(skb_b);
3421 /* Using more efficient varaiant than plain call to memcmp(). */
3422#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3423 u64 diffs = 0;
3424
3425 switch (meta_len) {
3426#define __it(x, op) (x -= sizeof(u##op))
3427#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3428 case 32: diffs |= __it_diff(a, b, 64);
3429 case 24: diffs |= __it_diff(a, b, 64);
3430 case 16: diffs |= __it_diff(a, b, 64);
3431 case 8: diffs |= __it_diff(a, b, 64);
3432 break;
3433 case 28: diffs |= __it_diff(a, b, 64);
3434 case 20: diffs |= __it_diff(a, b, 64);
3435 case 12: diffs |= __it_diff(a, b, 64);
3436 case 4: diffs |= __it_diff(a, b, 32);
3437 break;
3438 }
3439 return diffs;
3440#else
3441 return memcmp(a - meta_len, b - meta_len, meta_len);
3442#endif
3443}
3444
3445static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3446 const struct sk_buff *skb_b)
3447{
3448 u8 len_a = skb_metadata_len(skb_a);
3449 u8 len_b = skb_metadata_len(skb_b);
3450
3451 if (!(len_a | len_b))
3452 return false;
3453
3454 return len_a != len_b ?
3455 true : __skb_metadata_differs(skb_a, skb_b, len_a);
3456}
3457
3458static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3459{
3460 skb_shinfo(skb)->meta_len = meta_len;
3461}
3462
3463static inline void skb_metadata_clear(struct sk_buff *skb)
3464{
3465 skb_metadata_set(skb, 0);
3466}
3467
Alexander Duyck62bccb82014-09-04 13:31:35 -04003468struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3469
Richard Cochranc1f19b52010-07-17 08:49:36 +00003470#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3471
Joe Perches7965bd42013-09-26 14:48:15 -07003472void skb_clone_tx_timestamp(struct sk_buff *skb);
3473bool skb_defer_rx_timestamp(struct sk_buff *skb);
Richard Cochranc1f19b52010-07-17 08:49:36 +00003474
3475#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3476
3477static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3478{
3479}
3480
3481static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3482{
3483 return false;
3484}
3485
3486#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3487
3488/**
3489 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3490 *
Richard Cochranda92b192011-10-21 00:49:15 +00003491 * PHY drivers may accept clones of transmitted packets for
3492 * timestamping via their phy_driver.txtstamp method. These drivers
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003493 * must call this function to return the skb back to the stack with a
3494 * timestamp.
Richard Cochranda92b192011-10-21 00:49:15 +00003495 *
Richard Cochranc1f19b52010-07-17 08:49:36 +00003496 * @skb: clone of the the original outgoing packet
Benjamin Poirier7a76a022015-08-07 09:32:21 -07003497 * @hwtstamps: hardware time stamps
Richard Cochranc1f19b52010-07-17 08:49:36 +00003498 *
3499 */
3500void skb_complete_tx_timestamp(struct sk_buff *skb,
3501 struct skb_shared_hwtstamps *hwtstamps);
3502
Willem de Bruijne7fd2882014-08-04 22:11:48 -04003503void __skb_tstamp_tx(struct sk_buff *orig_skb,
3504 struct skb_shared_hwtstamps *hwtstamps,
3505 struct sock *sk, int tstype);
3506
Patrick Ohlyac45f602009-02-12 05:03:37 +00003507/**
3508 * skb_tstamp_tx - queue clone of skb with send time stamps
3509 * @orig_skb: the original outgoing packet
3510 * @hwtstamps: hardware time stamps, may be NULL if not available
3511 *
3512 * If the skb has a socket associated, then this function clones the
3513 * skb (thus sharing the actual data and optional structures), stores
3514 * the optional hardware time stamping information (if non NULL) or
3515 * generates a software time stamp (otherwise), then queues the clone
3516 * to the error queue of the socket. Errors are silently ignored.
3517 */
Joe Perches7965bd42013-09-26 14:48:15 -07003518void skb_tstamp_tx(struct sk_buff *orig_skb,
3519 struct skb_shared_hwtstamps *hwtstamps);
Patrick Ohlyac45f602009-02-12 05:03:37 +00003520
Richard Cochran4507a712010-07-17 08:48:28 +00003521/**
3522 * skb_tx_timestamp() - Driver hook for transmit timestamping
3523 *
3524 * Ethernet MAC Drivers should call this function in their hard_xmit()
Richard Cochran4ff75b72011-06-19 03:31:39 +00003525 * function immediately before giving the sk_buff to the MAC hardware.
Richard Cochran4507a712010-07-17 08:48:28 +00003526 *
David S. Miller73409f32013-12-27 13:04:33 -05003527 * Specifically, one should make absolutely sure that this function is
3528 * called before TX completion of this packet can trigger. Otherwise
3529 * the packet could potentially already be freed.
3530 *
Richard Cochran4507a712010-07-17 08:48:28 +00003531 * @skb: A socket buffer.
3532 */
3533static inline void skb_tx_timestamp(struct sk_buff *skb)
3534{
Richard Cochranc1f19b52010-07-17 08:49:36 +00003535 skb_clone_tx_timestamp(skb);
Miroslav Lichvarb50a5c72017-05-19 17:52:40 +02003536 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3537 skb_tstamp_tx(skb, NULL);
Richard Cochran4507a712010-07-17 08:48:28 +00003538}
3539
Johannes Berg6e3e9392011-11-09 10:15:42 +01003540/**
3541 * skb_complete_wifi_ack - deliver skb with wifi status
3542 *
3543 * @skb: the original outgoing packet
3544 * @acked: ack status
3545 *
3546 */
3547void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3548
Joe Perches7965bd42013-09-26 14:48:15 -07003549__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3550__sum16 __skb_checksum_complete(struct sk_buff *skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003551
Herbert Xu60476372007-04-09 11:59:39 -07003552static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3553{
Tom Herbert6edec0e2015-02-10 16:30:28 -08003554 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3555 skb->csum_valid ||
3556 (skb->ip_summed == CHECKSUM_PARTIAL &&
3557 skb_checksum_start_offset(skb) >= 0));
Herbert Xu60476372007-04-09 11:59:39 -07003558}
3559
Herbert Xufb286bb2005-11-10 13:01:24 -08003560/**
3561 * skb_checksum_complete - Calculate checksum of an entire packet
3562 * @skb: packet to process
3563 *
3564 * This function calculates the checksum over the entire packet plus
3565 * the value of skb->csum. The latter can be used to supply the
3566 * checksum of a pseudo header as used by TCP/UDP. It returns the
3567 * checksum.
3568 *
3569 * For protocols that contain complete checksums such as ICMP/TCP/UDP,
3570 * this function can be used to verify that checksum on received
3571 * packets. In that case the function should return zero if the
3572 * checksum is correct. In particular, this function will return zero
3573 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3574 * hardware has already verified the correctness of the checksum.
3575 */
Al Viro4381ca32007-07-15 21:00:11 +01003576static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
Herbert Xufb286bb2005-11-10 13:01:24 -08003577{
Herbert Xu60476372007-04-09 11:59:39 -07003578 return skb_csum_unnecessary(skb) ?
3579 0 : __skb_checksum_complete(skb);
Herbert Xufb286bb2005-11-10 13:01:24 -08003580}
3581
Tom Herbert77cffe22014-08-27 21:26:46 -07003582static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3583{
3584 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3585 if (skb->csum_level == 0)
3586 skb->ip_summed = CHECKSUM_NONE;
3587 else
3588 skb->csum_level--;
3589 }
3590}
3591
3592static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3593{
3594 if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3595 if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3596 skb->csum_level++;
3597 } else if (skb->ip_summed == CHECKSUM_NONE) {
3598 skb->ip_summed = CHECKSUM_UNNECESSARY;
3599 skb->csum_level = 0;
3600 }
3601}
3602
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003603/* Check if we need to perform checksum complete validation.
3604 *
3605 * Returns true if checksum complete is needed, false otherwise
3606 * (either checksum is unnecessary or zero checksum is allowed).
3607 */
3608static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3609 bool zero_okay,
3610 __sum16 check)
3611{
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003612 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3613 skb->csum_valid = 1;
Tom Herbert77cffe22014-08-27 21:26:46 -07003614 __skb_decr_checksum_unnecessary(skb);
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003615 return false;
3616 }
3617
3618 return true;
3619}
3620
3621/* For small packets <= CHECKSUM_BREAK peform checksum complete directly
3622 * in checksum_init.
3623 */
3624#define CHECKSUM_BREAK 76
3625
Tom Herbert4e18b9a2015-04-20 14:10:04 -07003626/* Unset checksum-complete
3627 *
3628 * Unset checksum complete can be done when packet is being modified
3629 * (uncompressed for instance) and checksum-complete value is
3630 * invalidated.
3631 */
3632static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3633{
3634 if (skb->ip_summed == CHECKSUM_COMPLETE)
3635 skb->ip_summed = CHECKSUM_NONE;
3636}
3637
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003638/* Validate (init) checksum based on checksum complete.
3639 *
3640 * Return values:
3641 * 0: checksum is validated or try to in skb_checksum_complete. In the latter
3642 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3643 * checksum is stored in skb->csum for use in __skb_checksum_complete
3644 * non-zero: value of invalid checksum
3645 *
3646 */
3647static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3648 bool complete,
3649 __wsum psum)
3650{
3651 if (skb->ip_summed == CHECKSUM_COMPLETE) {
3652 if (!csum_fold(csum_add(psum, skb->csum))) {
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003653 skb->csum_valid = 1;
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003654 return 0;
3655 }
3656 }
3657
3658 skb->csum = psum;
3659
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003660 if (complete || skb->len <= CHECKSUM_BREAK) {
3661 __sum16 csum;
3662
3663 csum = __skb_checksum_complete(skb);
3664 skb->csum_valid = !csum;
3665 return csum;
3666 }
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003667
3668 return 0;
3669}
3670
3671static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3672{
3673 return 0;
3674}
3675
3676/* Perform checksum validate (init). Note that this is a macro since we only
3677 * want to calculate the pseudo header which is an input function if necessary.
3678 * First we try to validate without any computation (checksum unnecessary) and
3679 * then calculate based on checksum complete calling the function to compute
3680 * pseudo header.
3681 *
3682 * Return values:
3683 * 0: checksum is validated or try to in skb_checksum_complete
3684 * non-zero: value of invalid checksum
3685 */
3686#define __skb_checksum_validate(skb, proto, complete, \
3687 zero_okay, check, compute_pseudo) \
3688({ \
3689 __sum16 __ret = 0; \
Tom Herbert5d0c2b92014-06-10 18:54:13 -07003690 skb->csum_valid = 0; \
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003691 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \
3692 __ret = __skb_checksum_validate_complete(skb, \
3693 complete, compute_pseudo(skb, proto)); \
3694 __ret; \
3695})
3696
3697#define skb_checksum_init(skb, proto, compute_pseudo) \
3698 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3699
3700#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3701 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3702
3703#define skb_checksum_validate(skb, proto, compute_pseudo) \
3704 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3705
3706#define skb_checksum_validate_zero_check(skb, proto, check, \
3707 compute_pseudo) \
Sabrina Dubroca096a4cf2015-02-06 18:54:19 +01003708 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
Tom Herbert76ba0aa2014-05-02 16:29:18 -07003709
3710#define skb_checksum_simple_validate(skb) \
3711 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3712
Tom Herbertd96535a2014-08-31 15:12:42 -07003713static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3714{
Davide Caratti219f1d792017-05-18 15:44:39 +02003715 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
Tom Herbertd96535a2014-08-31 15:12:42 -07003716}
3717
3718static inline void __skb_checksum_convert(struct sk_buff *skb,
3719 __sum16 check, __wsum pseudo)
3720{
3721 skb->csum = ~pseudo;
3722 skb->ip_summed = CHECKSUM_COMPLETE;
3723}
3724
3725#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \
3726do { \
3727 if (__skb_checksum_convert_check(skb)) \
3728 __skb_checksum_convert(skb, check, \
3729 compute_pseudo(skb, proto)); \
3730} while (0)
3731
Tom Herbert15e23962015-02-10 16:30:31 -08003732static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3733 u16 start, u16 offset)
3734{
3735 skb->ip_summed = CHECKSUM_PARTIAL;
3736 skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3737 skb->csum_offset = offset - start;
3738}
3739
Tom Herbertdcdc8992015-02-02 16:07:34 -08003740/* Update skbuf and packet to reflect the remote checksum offload operation.
3741 * When called, ptr indicates the starting point for skb->csum when
3742 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3743 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3744 */
3745static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
Tom Herbert15e23962015-02-10 16:30:31 -08003746 int start, int offset, bool nopartial)
Tom Herbertdcdc8992015-02-02 16:07:34 -08003747{
3748 __wsum delta;
3749
Tom Herbert15e23962015-02-10 16:30:31 -08003750 if (!nopartial) {
3751 skb_remcsum_adjust_partial(skb, ptr, start, offset);
3752 return;
3753 }
3754
Tom Herbertdcdc8992015-02-02 16:07:34 -08003755 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3756 __skb_checksum_complete(skb);
3757 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3758 }
3759
3760 delta = remcsum_adjust(ptr, skb->csum, start, offset);
3761
3762 /* Adjust skb->csum since we changed the packet */
3763 skb->csum = csum_add(skb->csum, delta);
3764}
3765
Florian Westphalcb9c6832017-01-23 18:21:56 +01003766static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3767{
3768#if IS_ENABLED(CONFIG_NF_CONNTRACK)
Florian Westphala9e419d2017-01-23 18:21:59 +01003769 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
Florian Westphalcb9c6832017-01-23 18:21:56 +01003770#else
3771 return NULL;
3772#endif
3773}
3774
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003775#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Joe Perches7965bd42013-09-26 14:48:15 -07003776void nf_conntrack_destroy(struct nf_conntrack *nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003777static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3778{
3779 if (nfct && atomic_dec_and_test(&nfct->use))
Yasuyuki Kozakaide6e05c2007-03-23 11:17:27 -07003780 nf_conntrack_destroy(nfct);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003781}
3782static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3783{
3784 if (nfct)
3785 atomic_inc(&nfct->use);
3786}
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003787#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003788#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Linus Torvalds1da177e2005-04-16 15:20:36 -07003789static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge)
3790{
Reshetova, Elena53869ce2017-06-30 13:07:57 +03003791 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use))
Linus Torvalds1da177e2005-04-16 15:20:36 -07003792 kfree(nf_bridge);
3793}
3794static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge)
3795{
3796 if (nf_bridge)
Reshetova, Elena53869ce2017-06-30 13:07:57 +03003797 refcount_inc(&nf_bridge->use);
Linus Torvalds1da177e2005-04-16 15:20:36 -07003798}
3799#endif /* CONFIG_BRIDGE_NETFILTER */
Patrick McHardya193a4a2006-03-20 19:23:05 -08003800static inline void nf_reset(struct sk_buff *skb)
3801{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003802#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Florian Westphala9e419d2017-01-23 18:21:59 +01003803 nf_conntrack_put(skb_nfct(skb));
3804 skb->_nfct = 0;
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003805#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003806#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Patrick McHardya193a4a2006-03-20 19:23:05 -08003807 nf_bridge_put(skb->nf_bridge);
3808 skb->nf_bridge = NULL;
3809#endif
3810}
3811
Patrick McHardy124dff02013-04-05 20:42:05 +02003812static inline void nf_reset_trace(struct sk_buff *skb)
3813{
Florian Westphal478b3602014-02-15 23:48:45 +01003814#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Gao feng130549fe2013-03-21 19:48:41 +00003815 skb->nf_trace = 0;
3816#endif
Linus Torvalds1da177e2005-04-16 15:20:36 -07003817}
3818
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003819/* Note: This doesn't put any conntrack and bridge info in dst. */
Eric Dumazetb1937222014-09-28 22:18:47 -07003820static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
3821 bool copy)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003822{
Yasuyuki Kozakai5f79e0f2007-03-23 11:17:07 -07003823#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Florian Westphala9e419d2017-01-23 18:21:59 +01003824 dst->_nfct = src->_nfct;
3825 nf_conntrack_get(skb_nfct(src));
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003826#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003827#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003828 dst->nf_bridge = src->nf_bridge;
3829 nf_bridge_get(src->nf_bridge);
3830#endif
Florian Westphal478b3602014-02-15 23:48:45 +01003831#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
Eric Dumazetb1937222014-09-28 22:18:47 -07003832 if (copy)
3833 dst->nf_trace = src->nf_trace;
Florian Westphal478b3602014-02-15 23:48:45 +01003834#endif
Yasuyuki Kozakaiedda5532007-03-14 16:43:37 -07003835}
3836
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003837static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
3838{
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003839#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
Florian Westphala9e419d2017-01-23 18:21:59 +01003840 nf_conntrack_put(skb_nfct(dst));
KOVACS Krisztian2fc72c72011-01-12 20:25:08 +01003841#endif
Pablo Neira Ayuso34666d42014-09-18 11:29:03 +02003842#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003843 nf_bridge_put(dst->nf_bridge);
3844#endif
Eric Dumazetb1937222014-09-28 22:18:47 -07003845 __nf_copy(dst, src, true);
Yasuyuki Kozakaie7ac05f2007-03-14 16:44:01 -07003846}
3847
James Morris984bc162006-06-09 00:29:17 -07003848#ifdef CONFIG_NETWORK_SECMARK
3849static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3850{
3851 to->secmark = from->secmark;
3852}
3853
3854static inline void skb_init_secmark(struct sk_buff *skb)
3855{
3856 skb->secmark = 0;
3857}
3858#else
3859static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
3860{ }
3861
3862static inline void skb_init_secmark(struct sk_buff *skb)
3863{ }
3864#endif
3865
Eric W. Biederman574f7192014-04-01 12:20:24 -07003866static inline bool skb_irq_freeable(const struct sk_buff *skb)
3867{
3868 return !skb->destructor &&
3869#if IS_ENABLED(CONFIG_XFRM)
3870 !skb->sp &&
3871#endif
Florian Westphalcb9c6832017-01-23 18:21:56 +01003872 !skb_nfct(skb) &&
Eric W. Biederman574f7192014-04-01 12:20:24 -07003873 !skb->_skb_refdst &&
3874 !skb_has_frag_list(skb);
3875}
3876
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003877static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
3878{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003879 skb->queue_mapping = queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003880}
3881
Stephen Hemminger92477442009-03-21 13:39:26 -07003882static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003883{
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003884 return skb->queue_mapping;
Pavel Emelyanov4e3ab472007-10-21 17:01:29 -07003885}
3886
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003887static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
3888{
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003889 to->queue_mapping = from->queue_mapping;
Peter P Waskiewicz Jrf25f4e42007-07-06 13:36:20 -07003890}
3891
David S. Millerd5a9e242009-01-27 16:22:11 -08003892static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
3893{
3894 skb->queue_mapping = rx_queue + 1;
3895}
3896
Stephen Hemminger92477442009-03-21 13:39:26 -07003897static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003898{
3899 return skb->queue_mapping - 1;
3900}
3901
Stephen Hemminger92477442009-03-21 13:39:26 -07003902static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
David S. Millerd5a9e242009-01-27 16:22:11 -08003903{
Eric Dumazeta02cec22010-09-22 20:43:57 +00003904 return skb->queue_mapping != 0;
David S. Millerd5a9e242009-01-27 16:22:11 -08003905}
3906
Julian Anastasov4ff06202017-02-06 23:14:12 +02003907static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
3908{
3909 skb->dst_pending_confirm = val;
3910}
3911
3912static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
3913{
3914 return skb->dst_pending_confirm != 0;
3915}
3916
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003917static inline struct sec_path *skb_sec_path(struct sk_buff *skb)
3918{
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003919#ifdef CONFIG_XFRM
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003920 return skb->sp;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003921#else
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003922 return NULL;
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003923#endif
Denis Kirjanov0b3d8e02013-10-02 05:58:32 +04003924}
Alexey Dobriyandef8b4f2008-10-28 13:24:06 -07003925
Pravin B Shelar68c33162013-02-14 14:02:41 +00003926/* Keeps track of mac header offset relative to skb->head.
3927 * It is useful for TSO of Tunneling protocol. e.g. GRE.
3928 * For non-tunnel skb it points to skb_mac_header() and for
Eric Dumazet3347c962013-10-19 11:42:56 -07003929 * tunnel skb it points to outer mac header.
3930 * Keeps track of level of encapsulation of network headers.
3931 */
Pravin B Shelar68c33162013-02-14 14:02:41 +00003932struct skb_gso_cb {
Alexander Duyck802ab552016-04-10 21:45:03 -04003933 union {
3934 int mac_offset;
3935 int data_offset;
3936 };
Eric Dumazet3347c962013-10-19 11:42:56 -07003937 int encap_level;
Alexander Duyck76443452016-02-05 15:27:37 -08003938 __wsum csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003939 __u16 csum_start;
Pravin B Shelar68c33162013-02-14 14:02:41 +00003940};
Konstantin Khlebnikov9207f9d2016-01-08 15:21:46 +03003941#define SKB_SGO_CB_OFFSET 32
3942#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
Pravin B Shelar68c33162013-02-14 14:02:41 +00003943
3944static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
3945{
3946 return (skb_mac_header(inner_skb) - inner_skb->head) -
3947 SKB_GSO_CB(inner_skb)->mac_offset;
3948}
3949
Pravin B Shelar1e2bd512013-05-30 06:45:27 +00003950static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
3951{
3952 int new_headroom, headroom;
3953 int ret;
3954
3955 headroom = skb_headroom(skb);
3956 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
3957 if (ret)
3958 return ret;
3959
3960 new_headroom = skb_headroom(skb);
3961 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
3962 return 0;
3963}
3964
Alexander Duyck08b64fc2016-02-05 15:27:49 -08003965static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
3966{
3967 /* Do not update partial checksums if remote checksum is enabled. */
3968 if (skb->remcsum_offload)
3969 return;
3970
3971 SKB_GSO_CB(skb)->csum = res;
3972 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
3973}
3974
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003975/* Compute the checksum for a gso segment. First compute the checksum value
3976 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
3977 * then add in skb->csum (checksum from csum_start to end of packet).
3978 * skb->csum and csum_start are then updated to reflect the checksum of the
3979 * resultant packet starting from the transport header-- the resultant checksum
3980 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
3981 * header.
3982 */
3983static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
3984{
Alexander Duyck76443452016-02-05 15:27:37 -08003985 unsigned char *csum_start = skb_transport_header(skb);
3986 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
3987 __wsum partial = SKB_GSO_CB(skb)->csum;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003988
Alexander Duyck76443452016-02-05 15:27:37 -08003989 SKB_GSO_CB(skb)->csum = res;
3990 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003991
Alexander Duyck76443452016-02-05 15:27:37 -08003992 return csum_fold(csum_partial(csum_start, plen, partial));
Tom Herbert7e2b10c2014-06-04 17:20:02 -07003993}
3994
David S. Millerbdcc0922012-03-07 20:53:36 -05003995static inline bool skb_is_gso(const struct sk_buff *skb)
Herbert Xu89114af2006-07-08 13:34:32 -07003996{
3997 return skb_shinfo(skb)->gso_size;
3998}
3999
Eric Dumazet36a8f392013-09-29 01:21:32 -07004000/* Note: Should be called only if skb_is_gso(skb) is true */
David S. Millerbdcc0922012-03-07 20:53:36 -05004001static inline bool skb_is_gso_v6(const struct sk_buff *skb)
Brice Goglineabd7e32007-10-13 12:33:32 +02004002{
4003 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4004}
4005
Daniel Borkmann5293efe2016-08-18 01:00:39 +02004006static inline void skb_gso_reset(struct sk_buff *skb)
4007{
4008 skb_shinfo(skb)->gso_size = 0;
4009 skb_shinfo(skb)->gso_segs = 0;
4010 skb_shinfo(skb)->gso_type = 0;
4011}
4012
Joe Perches7965bd42013-09-26 14:48:15 -07004013void __skb_warn_lro_forwarding(const struct sk_buff *skb);
Ben Hutchings4497b072008-06-19 16:22:28 -07004014
4015static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4016{
4017 /* LRO sets gso_size but not gso_type, whereas if GSO is really
4018 * wanted then gso_type will be set. */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04004019 const struct skb_shared_info *shinfo = skb_shinfo(skb);
4020
Alexander Duyckb78462e2010-06-02 12:24:37 +00004021 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4022 unlikely(shinfo->gso_type == 0)) {
Ben Hutchings4497b072008-06-19 16:22:28 -07004023 __skb_warn_lro_forwarding(skb);
4024 return true;
4025 }
4026 return false;
4027}
4028
Herbert Xu35fc92a2007-03-26 23:22:20 -07004029static inline void skb_forward_csum(struct sk_buff *skb)
4030{
4031 /* Unfortunately we don't support this one. Any brave souls? */
4032 if (skb->ip_summed == CHECKSUM_COMPLETE)
4033 skb->ip_summed = CHECKSUM_NONE;
4034}
4035
Eric Dumazetbc8acf22010-09-02 13:07:41 -07004036/**
4037 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4038 * @skb: skb to check
4039 *
4040 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4041 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4042 * use this helper, to document places where we make this assertion.
4043 */
Eric Dumazet05bdd2f2011-10-20 17:45:43 -04004044static inline void skb_checksum_none_assert(const struct sk_buff *skb)
Eric Dumazetbc8acf22010-09-02 13:07:41 -07004045{
4046#ifdef DEBUG
4047 BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4048#endif
4049}
4050
Rusty Russellf35d9d82008-02-04 23:49:54 -05004051bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
Shirley Maa6686f22011-07-06 12:22:12 +00004052
Paul Durranted1f50c2014-01-09 10:02:46 +00004053int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
Linus Lüssing9afd85c2015-05-02 14:01:07 +02004054struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4055 unsigned int transport_len,
4056 __sum16(*skb_chkf)(struct sk_buff *skb));
Paul Durranted1f50c2014-01-09 10:02:46 +00004057
Alexander Duyck3a7c1ee42012-05-03 01:09:42 +00004058/**
4059 * skb_head_is_locked - Determine if the skb->head is locked down
4060 * @skb: skb to check
4061 *
4062 * The head on skbs build around a head frag can be removed if they are
4063 * not cloned. This function returns true if the skb head is locked down
4064 * due to either being allocated via kmalloc, or by being a clone with
4065 * multiple references to the head.
4066 */
4067static inline bool skb_head_is_locked(const struct sk_buff *skb)
4068{
4069 return !skb->head_frag || skb_cloned(skb);
4070}
Florian Westphalfe6cc552014-02-13 23:09:12 +01004071
4072/**
4073 * skb_gso_network_seglen - Return length of individual segments of a gso packet
4074 *
4075 * @skb: GSO skb
4076 *
4077 * skb_gso_network_seglen is used to determine the real size of the
4078 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP).
4079 *
4080 * The MAC/L2 header is not accounted for.
4081 */
4082static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb)
4083{
4084 unsigned int hdr_len = skb_transport_header(skb) -
4085 skb_network_header(skb);
4086 return hdr_len + skb_gso_transport_seglen(skb);
4087}
Thomas Grafee122c72015-07-21 10:43:58 +02004088
Edward Cree179bc672016-02-11 20:48:04 +00004089/* Local Checksum Offload.
4090 * Compute outer checksum based on the assumption that the
4091 * inner checksum will be offloaded later.
Edward Creee8ae7b02016-02-11 21:03:37 +00004092 * See Documentation/networking/checksum-offloads.txt for
4093 * explanation of how this works.
Edward Cree179bc672016-02-11 20:48:04 +00004094 * Fill in outer checksum adjustment (e.g. with sum of outer
4095 * pseudo-header) before calling.
4096 * Also ensure that inner checksum is in linear data area.
4097 */
4098static inline __wsum lco_csum(struct sk_buff *skb)
4099{
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08004100 unsigned char *csum_start = skb_checksum_start(skb);
4101 unsigned char *l4_hdr = skb_transport_header(skb);
4102 __wsum partial;
Edward Cree179bc672016-02-11 20:48:04 +00004103
4104 /* Start with complement of inner checksum adjustment */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08004105 partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4106 skb->csum_offset));
4107
Edward Cree179bc672016-02-11 20:48:04 +00004108 /* Add in checksum of our headers (incl. outer checksum
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08004109 * adjustment filled in by caller) and return result.
Edward Cree179bc672016-02-11 20:48:04 +00004110 */
Alexander Duyck9e74a6d2016-02-17 11:23:55 -08004111 return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
Edward Cree179bc672016-02-11 20:48:04 +00004112}
4113
Linus Torvalds1da177e2005-04-16 15:20:36 -07004114#endif /* __KERNEL__ */
4115#endif /* _LINUX_SKBUFF_H */