linux/include/linux/skbuff.h
<<
>>
Prefs
   1/* SPDX-License-Identifier: GPL-2.0-or-later */
   2/*
   3 *      Definitions for the 'struct sk_buff' memory handlers.
   4 *
   5 *      Authors:
   6 *              Alan Cox, <gw4pts@gw4pts.ampr.org>
   7 *              Florian La Roche, <rzsfl@rz.uni-sb.de>
   8 */
   9
  10#ifndef _LINUX_SKBUFF_H
  11#define _LINUX_SKBUFF_H
  12
  13#include <linux/kernel.h>
  14#include <linux/compiler.h>
  15#include <linux/time.h>
  16#include <linux/bug.h>
  17#include <linux/cache.h>
  18#include <linux/rbtree.h>
  19#include <linux/socket.h>
  20#include <linux/refcount.h>
  21
  22#include <linux/atomic.h>
  23#include <asm/types.h>
  24#include <linux/spinlock.h>
  25#include <linux/net.h>
  26#include <linux/textsearch.h>
  27#include <net/checksum.h>
  28#include <linux/rcupdate.h>
  29#include <linux/hrtimer.h>
  30#include <linux/dma-mapping.h>
  31#include <linux/netdev_features.h>
  32#include <linux/sched.h>
  33#include <linux/sched/clock.h>
  34#include <net/flow_dissector.h>
  35#include <linux/splice.h>
  36#include <linux/in6.h>
  37#include <linux/if_packet.h>
  38#include <net/flow.h>
  39
  40/* The interface for checksum offload between the stack and networking drivers
  41 * is as follows...
  42 *
  43 * A. IP checksum related features
  44 *
  45 * Drivers advertise checksum offload capabilities in the features of a device.
  46 * From the stack's point of view these are capabilities offered by the driver,
  47 * a driver typically only advertises features that it is capable of offloading
  48 * to its device.
  49 *
  50 * The checksum related features are:
  51 *
  52 *      NETIF_F_HW_CSUM - The driver (or its device) is able to compute one
  53 *                        IP (one's complement) checksum for any combination
  54 *                        of protocols or protocol layering. The checksum is
  55 *                        computed and set in a packet per the CHECKSUM_PARTIAL
  56 *                        interface (see below).
  57 *
  58 *      NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain
  59 *                        TCP or UDP packets over IPv4. These are specifically
  60 *                        unencapsulated packets of the form IPv4|TCP or
  61 *                        IPv4|UDP where the Protocol field in the IPv4 header
  62 *                        is TCP or UDP. The IPv4 header may contain IP options
  63 *                        This feature cannot be set in features for a device
  64 *                        with NETIF_F_HW_CSUM also set. This feature is being
  65 *                        DEPRECATED (see below).
  66 *
  67 *      NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain
  68 *                        TCP or UDP packets over IPv6. These are specifically
  69 *                        unencapsulated packets of the form IPv6|TCP or
  70 *                        IPv4|UDP where the Next Header field in the IPv6
  71 *                        header is either TCP or UDP. IPv6 extension headers
  72 *                        are not supported with this feature. This feature
  73 *                        cannot be set in features for a device with
  74 *                        NETIF_F_HW_CSUM also set. This feature is being
  75 *                        DEPRECATED (see below).
  76 *
  77 *      NETIF_F_RXCSUM - Driver (device) performs receive checksum offload.
  78 *                       This flag is used only used to disable the RX checksum
  79 *                       feature for a device. The stack will accept receive
  80 *                       checksum indication in packets received on a device
  81 *                       regardless of whether NETIF_F_RXCSUM is set.
  82 *
  83 * B. Checksumming of received packets by device. Indication of checksum
  84 *    verification is in set skb->ip_summed. Possible values are:
  85 *
  86 * CHECKSUM_NONE:
  87 *
  88 *   Device did not checksum this packet e.g. due to lack of capabilities.
  89 *   The packet contains full (though not verified) checksum in packet but
  90 *   not in skb->csum. Thus, skb->csum is undefined in this case.
  91 *
  92 * CHECKSUM_UNNECESSARY:
  93 *
  94 *   The hardware you're dealing with doesn't calculate the full checksum
  95 *   (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums
  96 *   for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY
  97 *   if their checksums are okay. skb->csum is still undefined in this case
  98 *   though. A driver or device must never modify the checksum field in the
  99 *   packet even if checksum is verified.
 100 *
 101 *   CHECKSUM_UNNECESSARY is applicable to following protocols:
 102 *     TCP: IPv6 and IPv4.
 103 *     UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a
 104 *       zero UDP checksum for either IPv4 or IPv6, the networking stack
 105 *       may perform further validation in this case.
 106 *     GRE: only if the checksum is present in the header.
 107 *     SCTP: indicates the CRC in SCTP header has been validated.
 108 *     FCOE: indicates the CRC in FC frame has been validated.
 109 *
 110 *   skb->csum_level indicates the number of consecutive checksums found in
 111 *   the packet minus one that have been verified as CHECKSUM_UNNECESSARY.
 112 *   For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet
 113 *   and a device is able to verify the checksums for UDP (possibly zero),
 114 *   GRE (checksum flag is set), and TCP-- skb->csum_level would be set to
 115 *   two. If the device were only able to verify the UDP checksum and not
 116 *   GRE, either because it doesn't support GRE checksum of because GRE
 117 *   checksum is bad, skb->csum_level would be set to zero (TCP checksum is
 118 *   not considered in this case).
 119 *
 120 * CHECKSUM_COMPLETE:
 121 *
 122 *   This is the most generic way. The device supplied checksum of the _whole_
 123 *   packet as seen by netif_rx() and fills out in skb->csum. Meaning, the
 124 *   hardware doesn't need to parse L3/L4 headers to implement this.
 125 *
 126 *   Notes:
 127 *   - Even if device supports only some protocols, but is able to produce
 128 *     skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY.
 129 *   - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols.
 130 *
 131 * CHECKSUM_PARTIAL:
 132 *
 133 *   A checksum is set up to be offloaded to a device as described in the
 134 *   output description for CHECKSUM_PARTIAL. This may occur on a packet
 135 *   received directly from another Linux OS, e.g., a virtualized Linux kernel
 136 *   on the same host, or it may be set in the input path in GRO or remote
 137 *   checksum offload. For the purposes of checksum verification, the checksum
 138 *   referred to by skb->csum_start + skb->csum_offset and any preceding
 139 *   checksums in the packet are considered verified. Any checksums in the
 140 *   packet that are after the checksum being offloaded are not considered to
 141 *   be verified.
 142 *
 143 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload
 144 *    in the skb->ip_summed for a packet. Values are:
 145 *
 146 * CHECKSUM_PARTIAL:
 147 *
 148 *   The driver is required to checksum the packet as seen by hard_start_xmit()
 149 *   from skb->csum_start up to the end, and to record/write the checksum at
 150 *   offset skb->csum_start + skb->csum_offset. A driver may verify that the
 151 *   csum_start and csum_offset values are valid values given the length and
 152 *   offset of the packet, however they should not attempt to validate that the
 153 *   checksum refers to a legitimate transport layer checksum-- it is the
 154 *   purview of the stack to validate that csum_start and csum_offset are set
 155 *   correctly.
 156 *
 157 *   When the stack requests checksum offload for a packet, the driver MUST
 158 *   ensure that the checksum is set correctly. A driver can either offload the
 159 *   checksum calculation to the device, or call skb_checksum_help (in the case
 160 *   that the device does not support offload for a particular checksum).
 161 *
 162 *   NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of
 163 *   NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate
 164 *   checksum offload capability.
 165 *   skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based
 166 *   on network device checksumming capabilities: if a packet does not match
 167 *   them, skb_checksum_help or skb_crc32c_help (depending on the value of
 168 *   csum_not_inet, see item D.) is called to resolve the checksum.
 169 *
 170 * CHECKSUM_NONE:
 171 *
 172 *   The skb was already checksummed by the protocol, or a checksum is not
 173 *   required.
 174 *
 175 * CHECKSUM_UNNECESSARY:
 176 *
 177 *   This has the same meaning on as CHECKSUM_NONE for checksum offload on
 178 *   output.
 179 *
 180 * CHECKSUM_COMPLETE:
 181 *   Not used in checksum output. If a driver observes a packet with this value
 182 *   set in skbuff, if should treat as CHECKSUM_NONE being set.
 183 *
 184 * D. Non-IP checksum (CRC) offloads
 185 *
 186 *   NETIF_F_SCTP_CRC - This feature indicates that a device is capable of
 187 *     offloading the SCTP CRC in a packet. To perform this offload the stack
 188 *     will set set csum_start and csum_offset accordingly, set ip_summed to
 189 *     CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in
 190 *     the skbuff that the CHECKSUM_PARTIAL refers to CRC32c.
 191 *     A driver that supports both IP checksum offload and SCTP CRC32c offload
 192 *     must verify which offload is configured for a packet by testing the
 193 *     value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve
 194 *     CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1.
 195 *
 196 *   NETIF_F_FCOE_CRC - This feature indicates that a device is capable of
 197 *     offloading the FCOE CRC in a packet. To perform this offload the stack
 198 *     will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset
 199 *     accordingly. Note the there is no indication in the skbuff that the
 200 *     CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports
 201 *     both IP checksum offload and FCOE CRC offload must verify which offload
 202 *     is configured for a packet presumably by inspecting packet headers.
 203 *
 204 * E. Checksumming on output with GSO.
 205 *
 206 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload
 207 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the
 208 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as
 209 * part of the GSO operation is implied. If a checksum is being offloaded
 210 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset
 211 * are set to refer to the outermost checksum being offload (two offloaded
 212 * checksums are possible with UDP encapsulation).
 213 */
 214
 215/* Don't change this without changing skb_csum_unnecessary! */
 216#define CHECKSUM_NONE           0
 217#define CHECKSUM_UNNECESSARY    1
 218#define CHECKSUM_COMPLETE       2
 219#define CHECKSUM_PARTIAL        3
 220
 221/* Maximum value in skb->csum_level */
 222#define SKB_MAX_CSUM_LEVEL      3
 223
 224#define SKB_DATA_ALIGN(X)       ALIGN(X, SMP_CACHE_BYTES)
 225#define SKB_WITH_OVERHEAD(X)    \
 226        ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
 227#define SKB_MAX_ORDER(X, ORDER) \
 228        SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X))
 229#define SKB_MAX_HEAD(X)         (SKB_MAX_ORDER((X), 0))
 230#define SKB_MAX_ALLOC           (SKB_MAX_ORDER(0, 2))
 231
 232/* return minimum truesize of one skb containing X bytes of data */
 233#define SKB_TRUESIZE(X) ((X) +                                          \
 234                         SKB_DATA_ALIGN(sizeof(struct sk_buff)) +       \
 235                         SKB_DATA_ALIGN(sizeof(struct skb_shared_info)))
 236
 237struct net_device;
 238struct scatterlist;
 239struct pipe_inode_info;
 240struct iov_iter;
 241struct napi_struct;
 242struct bpf_prog;
 243union bpf_attr;
 244struct skb_ext;
 245
 246#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 247struct nf_conntrack {
 248        atomic_t use;
 249};
 250#endif
 251
 252#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
 253struct nf_bridge_info {
 254        enum {
 255                BRNF_PROTO_UNCHANGED,
 256                BRNF_PROTO_8021Q,
 257                BRNF_PROTO_PPPOE
 258        } orig_proto:8;
 259        u8                      pkt_otherhost:1;
 260        u8                      in_prerouting:1;
 261        u8                      bridged_dnat:1;
 262        __u16                   frag_max_size;
 263        struct net_device       *physindev;
 264
 265        /* always valid & non-NULL from FORWARD on, for physdev match */
 266        struct net_device       *physoutdev;
 267        union {
 268                /* prerouting: detect dnat in orig/reply direction */
 269                __be32          ipv4_daddr;
 270                struct in6_addr ipv6_daddr;
 271
 272                /* after prerouting + nat detected: store original source
 273                 * mac since neigh resolution overwrites it, only used while
 274                 * skb is out in neigh layer.
 275                 */
 276                char neigh_header[8];
 277        };
 278};
 279#endif
 280
 281struct sk_buff_head {
 282        /* These two members must be first. */
 283        struct sk_buff  *next;
 284        struct sk_buff  *prev;
 285
 286        __u32           qlen;
 287        spinlock_t      lock;
 288};
 289
 290struct sk_buff;
 291
 292/* To allow 64K frame to be packed as single skb without frag_list we
 293 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for
 294 * buffers which do not start on a page boundary.
 295 *
 296 * Since GRO uses frags we allocate at least 16 regardless of page
 297 * size.
 298 */
 299#if (65536/PAGE_SIZE + 1) < 16
 300#define MAX_SKB_FRAGS 16UL
 301#else
 302#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1)
 303#endif
 304extern int sysctl_max_skb_frags;
 305
 306/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to
 307 * segment using its current segmentation instead.
 308 */
 309#define GSO_BY_FRAGS    0xFFFF
 310
 311typedef struct skb_frag_struct skb_frag_t;
 312
 313struct skb_frag_struct {
 314        struct {
 315                struct page *p;
 316        } page;
 317#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536)
 318        __u32 page_offset;
 319        __u32 size;
 320#else
 321        __u16 page_offset;
 322        __u16 size;
 323#endif
 324};
 325
 326/**
 327 * skb_frag_size - Returns the size of a skb fragment
 328 * @frag: skb fragment
 329 */
 330static inline unsigned int skb_frag_size(const skb_frag_t *frag)
 331{
 332        return frag->size;
 333}
 334
 335/**
 336 * skb_frag_size_set - Sets the size of a skb fragment
 337 * @frag: skb fragment
 338 * @size: size of fragment
 339 */
 340static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size)
 341{
 342        frag->size = size;
 343}
 344
 345/**
 346 * skb_frag_size_add - Incrementes the size of a skb fragment by %delta
 347 * @frag: skb fragment
 348 * @delta: value to add
 349 */
 350static inline void skb_frag_size_add(skb_frag_t *frag, int delta)
 351{
 352        frag->size += delta;
 353}
 354
 355/**
 356 * skb_frag_size_sub - Decrements the size of a skb fragment by %delta
 357 * @frag: skb fragment
 358 * @delta: value to subtract
 359 */
 360static inline void skb_frag_size_sub(skb_frag_t *frag, int delta)
 361{
 362        frag->size -= delta;
 363}
 364
 365/**
 366 * skb_frag_must_loop - Test if %p is a high memory page
 367 * @p: fragment's page
 368 */
 369static inline bool skb_frag_must_loop(struct page *p)
 370{
 371#if defined(CONFIG_HIGHMEM)
 372        if (PageHighMem(p))
 373                return true;
 374#endif
 375        return false;
 376}
 377
 378/**
 379 *      skb_frag_foreach_page - loop over pages in a fragment
 380 *
 381 *      @f:             skb frag to operate on
 382 *      @f_off:         offset from start of f->page.p
 383 *      @f_len:         length from f_off to loop over
 384 *      @p:             (temp var) current page
 385 *      @p_off:         (temp var) offset from start of current page,
 386 *                                 non-zero only on first page.
 387 *      @p_len:         (temp var) length in current page,
 388 *                                 < PAGE_SIZE only on first and last page.
 389 *      @copied:        (temp var) length so far, excluding current p_len.
 390 *
 391 *      A fragment can hold a compound page, in which case per-page
 392 *      operations, notably kmap_atomic, must be called for each
 393 *      regular page.
 394 */
 395#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \
 396        for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT),            \
 397             p_off = (f_off) & (PAGE_SIZE - 1),                         \
 398             p_len = skb_frag_must_loop(p) ?                            \
 399             min_t(u32, f_len, PAGE_SIZE - p_off) : f_len,              \
 400             copied = 0;                                                \
 401             copied < f_len;                                            \
 402             copied += p_len, p++, p_off = 0,                           \
 403             p_len = min_t(u32, f_len - copied, PAGE_SIZE))             \
 404
 405#define HAVE_HW_TIME_STAMP
 406
 407/**
 408 * struct skb_shared_hwtstamps - hardware time stamps
 409 * @hwtstamp:   hardware time stamp transformed into duration
 410 *              since arbitrary point in time
 411 *
 412 * Software time stamps generated by ktime_get_real() are stored in
 413 * skb->tstamp.
 414 *
 415 * hwtstamps can only be compared against other hwtstamps from
 416 * the same device.
 417 *
 418 * This structure is attached to packets as part of the
 419 * &skb_shared_info. Use skb_hwtstamps() to get a pointer.
 420 */
 421struct skb_shared_hwtstamps {
 422        ktime_t hwtstamp;
 423};
 424
 425/* Definitions for tx_flags in struct skb_shared_info */
 426enum {
 427        /* generate hardware time stamp */
 428        SKBTX_HW_TSTAMP = 1 << 0,
 429
 430        /* generate software time stamp when queueing packet to NIC */
 431        SKBTX_SW_TSTAMP = 1 << 1,
 432
 433        /* device driver is going to provide hardware time stamp */
 434        SKBTX_IN_PROGRESS = 1 << 2,
 435
 436        /* device driver supports TX zero-copy buffers */
 437        SKBTX_DEV_ZEROCOPY = 1 << 3,
 438
 439        /* generate wifi status information (where possible) */
 440        SKBTX_WIFI_STATUS = 1 << 4,
 441
 442        /* This indicates at least one fragment might be overwritten
 443         * (as in vmsplice(), sendfile() ...)
 444         * If we need to compute a TX checksum, we'll need to copy
 445         * all frags to avoid possible bad checksum
 446         */
 447        SKBTX_SHARED_FRAG = 1 << 5,
 448
 449        /* generate software time stamp when entering packet scheduling */
 450        SKBTX_SCHED_TSTAMP = 1 << 6,
 451};
 452
 453#define SKBTX_ZEROCOPY_FRAG     (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG)
 454#define SKBTX_ANY_SW_TSTAMP     (SKBTX_SW_TSTAMP    | \
 455                                 SKBTX_SCHED_TSTAMP)
 456#define SKBTX_ANY_TSTAMP        (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP)
 457
 458/*
 459 * The callback notifies userspace to release buffers when skb DMA is done in
 460 * lower device, the skb last reference should be 0 when calling this.
 461 * The zerocopy_success argument is true if zero copy transmit occurred,
 462 * false on data copy or out of memory error caused by data copy attempt.
 463 * The ctx field is used to track device context.
 464 * The desc field is used to track userspace buffer index.
 465 */
 466struct ubuf_info {
 467        void (*callback)(struct ubuf_info *, bool zerocopy_success);
 468        union {
 469                struct {
 470                        unsigned long desc;
 471                        void *ctx;
 472                };
 473                struct {
 474                        u32 id;
 475                        u16 len;
 476                        u16 zerocopy:1;
 477                        u32 bytelen;
 478                };
 479        };
 480        refcount_t refcnt;
 481
 482        struct mmpin {
 483                struct user_struct *user;
 484                unsigned int num_pg;
 485        } mmp;
 486};
 487
 488#define skb_uarg(SKB)   ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg))
 489
 490int mm_account_pinned_pages(struct mmpin *mmp, size_t size);
 491void mm_unaccount_pinned_pages(struct mmpin *mmp);
 492
 493struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size);
 494struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size,
 495                                        struct ubuf_info *uarg);
 496
 497static inline void sock_zerocopy_get(struct ubuf_info *uarg)
 498{
 499        refcount_inc(&uarg->refcnt);
 500}
 501
 502void sock_zerocopy_put(struct ubuf_info *uarg);
 503void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref);
 504
 505void sock_zerocopy_callback(struct ubuf_info *uarg, bool success);
 506
 507int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len);
 508int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb,
 509                             struct msghdr *msg, int len,
 510                             struct ubuf_info *uarg);
 511
 512/* This data is invariant across clones and lives at
 513 * the end of the header data, ie. at skb->end.
 514 */
 515struct skb_shared_info {
 516        __u8            __unused;
 517        __u8            meta_len;
 518        __u8            nr_frags;
 519        __u8            tx_flags;
 520        unsigned short  gso_size;
 521        /* Warning: this field is not always filled in (UFO)! */
 522        unsigned short  gso_segs;
 523        struct sk_buff  *frag_list;
 524        struct skb_shared_hwtstamps hwtstamps;
 525        unsigned int    gso_type;
 526        u32             tskey;
 527
 528        /*
 529         * Warning : all fields before dataref are cleared in __alloc_skb()
 530         */
 531        atomic_t        dataref;
 532
 533        /* Intermediate layers must ensure that destructor_arg
 534         * remains valid until skb destructor */
 535        void *          destructor_arg;
 536
 537        /* must be last field, see pskb_expand_head() */
 538        skb_frag_t      frags[MAX_SKB_FRAGS];
 539};
 540
 541/* We divide dataref into two halves.  The higher 16 bits hold references
 542 * to the payload part of skb->data.  The lower 16 bits hold references to
 543 * the entire skb->data.  A clone of a headerless skb holds the length of
 544 * the header in skb->hdr_len.
 545 *
 546 * All users must obey the rule that the skb->data reference count must be
 547 * greater than or equal to the payload reference count.
 548 *
 549 * Holding a reference to the payload part means that the user does not
 550 * care about modifications to the header part of skb->data.
 551 */
 552#define SKB_DATAREF_SHIFT 16
 553#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1)
 554
 555
 556enum {
 557        SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */
 558        SKB_FCLONE_ORIG,        /* orig skb (from fclone_cache) */
 559        SKB_FCLONE_CLONE,       /* companion fclone skb (from fclone_cache) */
 560};
 561
 562enum {
 563        SKB_GSO_TCPV4 = 1 << 0,
 564
 565        /* This indicates the skb is from an untrusted source. */
 566        SKB_GSO_DODGY = 1 << 1,
 567
 568        /* This indicates the tcp segment has CWR set. */
 569        SKB_GSO_TCP_ECN = 1 << 2,
 570
 571        SKB_GSO_TCP_FIXEDID = 1 << 3,
 572
 573        SKB_GSO_TCPV6 = 1 << 4,
 574
 575        SKB_GSO_FCOE = 1 << 5,
 576
 577        SKB_GSO_GRE = 1 << 6,
 578
 579        SKB_GSO_GRE_CSUM = 1 << 7,
 580
 581        SKB_GSO_IPXIP4 = 1 << 8,
 582
 583        SKB_GSO_IPXIP6 = 1 << 9,
 584
 585        SKB_GSO_UDP_TUNNEL = 1 << 10,
 586
 587        SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11,
 588
 589        SKB_GSO_PARTIAL = 1 << 12,
 590
 591        SKB_GSO_TUNNEL_REMCSUM = 1 << 13,
 592
 593        SKB_GSO_SCTP = 1 << 14,
 594
 595        SKB_GSO_ESP = 1 << 15,
 596
 597        SKB_GSO_UDP = 1 << 16,
 598
 599        SKB_GSO_UDP_L4 = 1 << 17,
 600};
 601
 602#if BITS_PER_LONG > 32
 603#define NET_SKBUFF_DATA_USES_OFFSET 1
 604#endif
 605
 606#ifdef NET_SKBUFF_DATA_USES_OFFSET
 607typedef unsigned int sk_buff_data_t;
 608#else
 609typedef unsigned char *sk_buff_data_t;
 610#endif
 611
 612/**
 613 *      struct sk_buff - socket buffer
 614 *      @next: Next buffer in list
 615 *      @prev: Previous buffer in list
 616 *      @tstamp: Time we arrived/left
 617 *      @rbnode: RB tree node, alternative to next/prev for netem/tcp
 618 *      @sk: Socket we are owned by
 619 *      @dev: Device we arrived on/are leaving by
 620 *      @cb: Control buffer. Free for use by every layer. Put private vars here
 621 *      @_skb_refdst: destination entry (with norefcount bit)
 622 *      @sp: the security path, used for xfrm
 623 *      @len: Length of actual data
 624 *      @data_len: Data length
 625 *      @mac_len: Length of link layer header
 626 *      @hdr_len: writable header length of cloned skb
 627 *      @csum: Checksum (must include start/offset pair)
 628 *      @csum_start: Offset from skb->head where checksumming should start
 629 *      @csum_offset: Offset from csum_start where checksum should be stored
 630 *      @priority: Packet queueing priority
 631 *      @ignore_df: allow local fragmentation
 632 *      @cloned: Head may be cloned (check refcnt to be sure)
 633 *      @ip_summed: Driver fed us an IP checksum
 634 *      @nohdr: Payload reference only, must not modify header
 635 *      @pkt_type: Packet class
 636 *      @fclone: skbuff clone status
 637 *      @ipvs_property: skbuff is owned by ipvs
 638 *      @offload_fwd_mark: Packet was L2-forwarded in hardware
 639 *      @offload_l3_fwd_mark: Packet was L3-forwarded in hardware
 640 *      @tc_skip_classify: do not classify packet. set by IFB device
 641 *      @tc_at_ingress: used within tc_classify to distinguish in/egress
 642 *      @tc_redirected: packet was redirected by a tc action
 643 *      @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect
 644 *      @peeked: this packet has been seen already, so stats have been
 645 *              done for it, don't do them again
 646 *      @nf_trace: netfilter packet trace flag
 647 *      @protocol: Packet protocol from driver
 648 *      @destructor: Destruct function
 649 *      @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue)
 650 *      @_nfct: Associated connection, if any (with nfctinfo bits)
 651 *      @nf_bridge: Saved data about a bridged frame - see br_netfilter.c
 652 *      @skb_iif: ifindex of device we arrived on
 653 *      @tc_index: Traffic control index
 654 *      @hash: the packet hash
 655 *      @queue_mapping: Queue mapping for multiqueue devices
 656 *      @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves
 657 *      @active_extensions: active extensions (skb_ext_id types)
 658 *      @ndisc_nodetype: router type (from link layer)
 659 *      @ooo_okay: allow the mapping of a socket to a queue to be changed
 660 *      @l4_hash: indicate hash is a canonical 4-tuple hash over transport
 661 *              ports.
 662 *      @sw_hash: indicates hash was computed in software stack
 663 *      @wifi_acked_valid: wifi_acked was set
 664 *      @wifi_acked: whether frame was acked on wifi or not
 665 *      @no_fcs:  Request NIC to treat last 4 bytes as Ethernet FCS
 666 *      @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL
 667 *      @dst_pending_confirm: need to confirm neighbour
 668 *      @decrypted: Decrypted SKB
 669 *      @napi_id: id of the NAPI struct this skb came from
 670 *      @secmark: security marking
 671 *      @mark: Generic packet mark
 672 *      @vlan_proto: vlan encapsulation protocol
 673 *      @vlan_tci: vlan tag control information
 674 *      @inner_protocol: Protocol (encapsulation)
 675 *      @inner_transport_header: Inner transport layer header (encapsulation)
 676 *      @inner_network_header: Network layer header (encapsulation)
 677 *      @inner_mac_header: Link layer header (encapsulation)
 678 *      @transport_header: Transport layer header
 679 *      @network_header: Network layer header
 680 *      @mac_header: Link layer header
 681 *      @tail: Tail pointer
 682 *      @end: End pointer
 683 *      @head: Head of buffer
 684 *      @data: Data head pointer
 685 *      @truesize: Buffer size
 686 *      @users: User count - see {datagram,tcp}.c
 687 *      @extensions: allocated extensions, valid if active_extensions is nonzero
 688 */
 689
 690struct sk_buff {
 691        union {
 692                struct {
 693                        /* These two members must be first. */
 694                        struct sk_buff          *next;
 695                        struct sk_buff          *prev;
 696
 697                        union {
 698                                struct net_device       *dev;
 699                                /* Some protocols might use this space to store information,
 700                                 * while device pointer would be NULL.
 701                                 * UDP receive path is one user.
 702                                 */
 703                                unsigned long           dev_scratch;
 704                        };
 705                };
 706                struct rb_node          rbnode; /* used in netem, ip4 defrag, and tcp stack */
 707                struct list_head        list;
 708        };
 709
 710        union {
 711                struct sock             *sk;
 712                int                     ip_defrag_offset;
 713        };
 714
 715        union {
 716                ktime_t         tstamp;
 717                u64             skb_mstamp_ns; /* earliest departure time */
 718        };
 719        /*
 720         * This is the control buffer. It is free to use for every
 721         * layer. Please put your private variables there. If you
 722         * want to keep them across layers you have to do a skb_clone()
 723         * first. This is owned by whoever has the skb queued ATM.
 724         */
 725        char                    cb[48] __aligned(8);
 726
 727        union {
 728                struct {
 729                        unsigned long   _skb_refdst;
 730                        void            (*destructor)(struct sk_buff *skb);
 731                };
 732                struct list_head        tcp_tsorted_anchor;
 733        };
 734
 735#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
 736        unsigned long            _nfct;
 737#endif
 738        unsigned int            len,
 739                                data_len;
 740        __u16                   mac_len,
 741                                hdr_len;
 742
 743        /* Following fields are _not_ copied in __copy_skb_header()
 744         * Note that queue_mapping is here mostly to fill a hole.
 745         */
 746        __u16                   queue_mapping;
 747
 748/* if you move cloned around you also must adapt those constants */
 749#ifdef __BIG_ENDIAN_BITFIELD
 750#define CLONED_MASK     (1 << 7)
 751#else
 752#define CLONED_MASK     1
 753#endif
 754#define CLONED_OFFSET()         offsetof(struct sk_buff, __cloned_offset)
 755
 756        __u8                    __cloned_offset[0];
 757        __u8                    cloned:1,
 758                                nohdr:1,
 759                                fclone:2,
 760                                peeked:1,
 761                                head_frag:1,
 762                                pfmemalloc:1;
 763#ifdef CONFIG_SKB_EXTENSIONS
 764        __u8                    active_extensions;
 765#endif
 766        /* fields enclosed in headers_start/headers_end are copied
 767         * using a single memcpy() in __copy_skb_header()
 768         */
 769        /* private: */
 770        __u32                   headers_start[0];
 771        /* public: */
 772
 773/* if you move pkt_type around you also must adapt those constants */
 774#ifdef __BIG_ENDIAN_BITFIELD
 775#define PKT_TYPE_MAX    (7 << 5)
 776#else
 777#define PKT_TYPE_MAX    7
 778#endif
 779#define PKT_TYPE_OFFSET()       offsetof(struct sk_buff, __pkt_type_offset)
 780
 781        __u8                    __pkt_type_offset[0];
 782        __u8                    pkt_type:3;
 783        __u8                    ignore_df:1;
 784        __u8                    nf_trace:1;
 785        __u8                    ip_summed:2;
 786        __u8                    ooo_okay:1;
 787
 788        __u8                    l4_hash:1;
 789        __u8                    sw_hash:1;
 790        __u8                    wifi_acked_valid:1;
 791        __u8                    wifi_acked:1;
 792        __u8                    no_fcs:1;
 793        /* Indicates the inner headers are valid in the skbuff. */
 794        __u8                    encapsulation:1;
 795        __u8                    encap_hdr_csum:1;
 796        __u8                    csum_valid:1;
 797
 798#ifdef __BIG_ENDIAN_BITFIELD
 799#define PKT_VLAN_PRESENT_BIT    7
 800#else
 801#define PKT_VLAN_PRESENT_BIT    0
 802#endif
 803#define PKT_VLAN_PRESENT_OFFSET()       offsetof(struct sk_buff, __pkt_vlan_present_offset)
 804        __u8                    __pkt_vlan_present_offset[0];
 805        __u8                    vlan_present:1;
 806        __u8                    csum_complete_sw:1;
 807        __u8                    csum_level:2;
 808        __u8                    csum_not_inet:1;
 809        __u8                    dst_pending_confirm:1;
 810#ifdef CONFIG_IPV6_NDISC_NODETYPE
 811        __u8                    ndisc_nodetype:2;
 812#endif
 813
 814        __u8                    ipvs_property:1;
 815        __u8                    inner_protocol_type:1;
 816        __u8                    remcsum_offload:1;
 817#ifdef CONFIG_NET_SWITCHDEV
 818        __u8                    offload_fwd_mark:1;
 819        __u8                    offload_l3_fwd_mark:1;
 820#endif
 821#ifdef CONFIG_NET_CLS_ACT
 822        __u8                    tc_skip_classify:1;
 823        __u8                    tc_at_ingress:1;
 824        __u8                    tc_redirected:1;
 825        __u8                    tc_from_ingress:1;
 826#endif
 827#ifdef CONFIG_TLS_DEVICE
 828        __u8                    decrypted:1;
 829#endif
 830
 831#ifdef CONFIG_NET_SCHED
 832        __u16                   tc_index;       /* traffic control index */
 833#endif
 834
 835        union {
 836                __wsum          csum;
 837                struct {
 838                        __u16   csum_start;
 839                        __u16   csum_offset;
 840                };
 841        };
 842        __u32                   priority;
 843        int                     skb_iif;
 844        __u32                   hash;
 845        __be16                  vlan_proto;
 846        __u16                   vlan_tci;
 847#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS)
 848        union {
 849                unsigned int    napi_id;
 850                unsigned int    sender_cpu;
 851        };
 852#endif
 853#ifdef CONFIG_NETWORK_SECMARK
 854        __u32           secmark;
 855#endif
 856
 857        union {
 858                __u32           mark;
 859                __u32           reserved_tailroom;
 860        };
 861
 862        union {
 863                __be16          inner_protocol;
 864                __u8            inner_ipproto;
 865        };
 866
 867        __u16                   inner_transport_header;
 868        __u16                   inner_network_header;
 869        __u16                   inner_mac_header;
 870
 871        __be16                  protocol;
 872        __u16                   transport_header;
 873        __u16                   network_header;
 874        __u16                   mac_header;
 875
 876        /* private: */
 877        __u32                   headers_end[0];
 878        /* public: */
 879
 880        /* These elements must be at the end, see alloc_skb() for details.  */
 881        sk_buff_data_t          tail;
 882        sk_buff_data_t          end;
 883        unsigned char           *head,
 884                                *data;
 885        unsigned int            truesize;
 886        refcount_t              users;
 887
 888#ifdef CONFIG_SKB_EXTENSIONS
 889        /* only useable after checking ->active_extensions != 0 */
 890        struct skb_ext          *extensions;
 891#endif
 892};
 893
 894#ifdef __KERNEL__
 895/*
 896 *      Handling routines are only of interest to the kernel
 897 */
 898
 899#define SKB_ALLOC_FCLONE        0x01
 900#define SKB_ALLOC_RX            0x02
 901#define SKB_ALLOC_NAPI          0x04
 902
 903/**
 904 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves
 905 * @skb: buffer
 906 */
 907static inline bool skb_pfmemalloc(const struct sk_buff *skb)
 908{
 909        return unlikely(skb->pfmemalloc);
 910}
 911
 912/*
 913 * skb might have a dst pointer attached, refcounted or not.
 914 * _skb_refdst low order bit is set if refcount was _not_ taken
 915 */
 916#define SKB_DST_NOREF   1UL
 917#define SKB_DST_PTRMASK ~(SKB_DST_NOREF)
 918
 919#define SKB_NFCT_PTRMASK        ~(7UL)
 920/**
 921 * skb_dst - returns skb dst_entry
 922 * @skb: buffer
 923 *
 924 * Returns skb dst_entry, regardless of reference taken or not.
 925 */
 926static inline struct dst_entry *skb_dst(const struct sk_buff *skb)
 927{
 928        /* If refdst was not refcounted, check we still are in a
 929         * rcu_read_lock section
 930         */
 931        WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) &&
 932                !rcu_read_lock_held() &&
 933                !rcu_read_lock_bh_held());
 934        return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK);
 935}
 936
 937/**
 938 * skb_dst_set - sets skb dst
 939 * @skb: buffer
 940 * @dst: dst entry
 941 *
 942 * Sets skb dst, assuming a reference was taken on dst and should
 943 * be released by skb_dst_drop()
 944 */
 945static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst)
 946{
 947        skb->_skb_refdst = (unsigned long)dst;
 948}
 949
 950/**
 951 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference
 952 * @skb: buffer
 953 * @dst: dst entry
 954 *
 955 * Sets skb dst, assuming a reference was not taken on dst.
 956 * If dst entry is cached, we do not take reference and dst_release
 957 * will be avoided by refdst_drop. If dst entry is not cached, we take
 958 * reference, so that last dst_release can destroy the dst immediately.
 959 */
 960static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst)
 961{
 962        WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held());
 963        skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF;
 964}
 965
 966/**
 967 * skb_dst_is_noref - Test if skb dst isn't refcounted
 968 * @skb: buffer
 969 */
 970static inline bool skb_dst_is_noref(const struct sk_buff *skb)
 971{
 972        return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb);
 973}
 974
 975/**
 976 * skb_rtable - Returns the skb &rtable
 977 * @skb: buffer
 978 */
 979static inline struct rtable *skb_rtable(const struct sk_buff *skb)
 980{
 981        return (struct rtable *)skb_dst(skb);
 982}
 983
 984/* For mangling skb->pkt_type from user space side from applications
 985 * such as nft, tc, etc, we only allow a conservative subset of
 986 * possible pkt_types to be set.
 987*/
 988static inline bool skb_pkt_type_ok(u32 ptype)
 989{
 990        return ptype <= PACKET_OTHERHOST;
 991}
 992
 993/**
 994 * skb_napi_id - Returns the skb's NAPI id
 995 * @skb: buffer
 996 */
 997static inline unsigned int skb_napi_id(const struct sk_buff *skb)
 998{
 999#ifdef CONFIG_NET_RX_BUSY_POLL
1000        return skb->napi_id;
1001#else
1002        return 0;
1003#endif
1004}
1005
1006/**
1007 * skb_unref - decrement the skb's reference count
1008 * @skb: buffer
1009 *
1010 * Returns true if we can free the skb.
1011 */
1012static inline bool skb_unref(struct sk_buff *skb)
1013{
1014        if (unlikely(!skb))
1015                return false;
1016        if (likely(refcount_read(&skb->users) == 1))
1017                smp_rmb();
1018        else if (likely(!refcount_dec_and_test(&skb->users)))
1019                return false;
1020
1021        return true;
1022}
1023
1024void skb_release_head_state(struct sk_buff *skb);
1025void kfree_skb(struct sk_buff *skb);
1026void kfree_skb_list(struct sk_buff *segs);
1027void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt);
1028void skb_tx_error(struct sk_buff *skb);
1029void consume_skb(struct sk_buff *skb);
1030void __consume_stateless_skb(struct sk_buff *skb);
1031void  __kfree_skb(struct sk_buff *skb);
1032extern struct kmem_cache *skbuff_head_cache;
1033
1034void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1035bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1036                      bool *fragstolen, int *delta_truesize);
1037
1038struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1039                            int node);
1040struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1041struct sk_buff *build_skb(void *data, unsigned int frag_size);
1042struct sk_buff *build_skb_around(struct sk_buff *skb,
1043                                 void *data, unsigned int frag_size);
1044
1045/**
1046 * alloc_skb - allocate a network buffer
1047 * @size: size to allocate
1048 * @priority: allocation mask
1049 *
1050 * This function is a convenient wrapper around __alloc_skb().
1051 */
1052static inline struct sk_buff *alloc_skb(unsigned int size,
1053                                        gfp_t priority)
1054{
1055        return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1056}
1057
1058struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1059                                     unsigned long data_len,
1060                                     int max_page_order,
1061                                     int *errcode,
1062                                     gfp_t gfp_mask);
1063struct sk_buff *alloc_skb_for_msg(struct sk_buff *first);
1064
1065/* Layout of fast clones : [skb1][skb2][fclone_ref] */
1066struct sk_buff_fclones {
1067        struct sk_buff  skb1;
1068
1069        struct sk_buff  skb2;
1070
1071        refcount_t      fclone_ref;
1072};
1073
1074/**
1075 *      skb_fclone_busy - check if fclone is busy
1076 *      @sk: socket
1077 *      @skb: buffer
1078 *
1079 * Returns true if skb is a fast clone, and its clone is not freed.
1080 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1081 * so we also check that this didnt happen.
1082 */
1083static inline bool skb_fclone_busy(const struct sock *sk,
1084                                   const struct sk_buff *skb)
1085{
1086        const struct sk_buff_fclones *fclones;
1087
1088        fclones = container_of(skb, struct sk_buff_fclones, skb1);
1089
1090        return skb->fclone == SKB_FCLONE_ORIG &&
1091               refcount_read(&fclones->fclone_ref) > 1 &&
1092               fclones->skb2.sk == sk;
1093}
1094
1095/**
1096 * alloc_skb_fclone - allocate a network buffer from fclone cache
1097 * @size: size to allocate
1098 * @priority: allocation mask
1099 *
1100 * This function is a convenient wrapper around __alloc_skb().
1101 */
1102static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1103                                               gfp_t priority)
1104{
1105        return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1106}
1107
1108struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1109void skb_headers_offset_update(struct sk_buff *skb, int off);
1110int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1111struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1112void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1113struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1114struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1115                                   gfp_t gfp_mask, bool fclone);
1116static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1117                                          gfp_t gfp_mask)
1118{
1119        return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1120}
1121
1122int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1123struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1124                                     unsigned int headroom);
1125struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1126                                int newtailroom, gfp_t priority);
1127int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1128                                     int offset, int len);
1129int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1130                              int offset, int len);
1131int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1132int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1133
1134/**
1135 *      skb_pad                 -       zero pad the tail of an skb
1136 *      @skb: buffer to pad
1137 *      @pad: space to pad
1138 *
1139 *      Ensure that a buffer is followed by a padding area that is zero
1140 *      filled. Used by network drivers which may DMA or transfer data
1141 *      beyond the buffer end onto the wire.
1142 *
1143 *      May return error in out of memory cases. The skb is freed on error.
1144 */
1145static inline int skb_pad(struct sk_buff *skb, int pad)
1146{
1147        return __skb_pad(skb, pad, true);
1148}
1149#define dev_kfree_skb(a)        consume_skb(a)
1150
1151int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1152                         int offset, size_t size);
1153
1154struct skb_seq_state {
1155        __u32           lower_offset;
1156        __u32           upper_offset;
1157        __u32           frag_idx;
1158        __u32           stepped_offset;
1159        struct sk_buff  *root_skb;
1160        struct sk_buff  *cur_skb;
1161        __u8            *frag_data;
1162};
1163
1164void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1165                          unsigned int to, struct skb_seq_state *st);
1166unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1167                          struct skb_seq_state *st);
1168void skb_abort_seq_read(struct skb_seq_state *st);
1169
1170unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1171                           unsigned int to, struct ts_config *config);
1172
1173/*
1174 * Packet hash types specify the type of hash in skb_set_hash.
1175 *
1176 * Hash types refer to the protocol layer addresses which are used to
1177 * construct a packet's hash. The hashes are used to differentiate or identify
1178 * flows of the protocol layer for the hash type. Hash types are either
1179 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1180 *
1181 * Properties of hashes:
1182 *
1183 * 1) Two packets in different flows have different hash values
1184 * 2) Two packets in the same flow should have the same hash value
1185 *
1186 * A hash at a higher layer is considered to be more specific. A driver should
1187 * set the most specific hash possible.
1188 *
1189 * A driver cannot indicate a more specific hash than the layer at which a hash
1190 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1191 *
1192 * A driver may indicate a hash level which is less specific than the
1193 * actual layer the hash was computed on. For instance, a hash computed
1194 * at L4 may be considered an L3 hash. This should only be done if the
1195 * driver can't unambiguously determine that the HW computed the hash at
1196 * the higher layer. Note that the "should" in the second property above
1197 * permits this.
1198 */
1199enum pkt_hash_types {
1200        PKT_HASH_TYPE_NONE,     /* Undefined type */
1201        PKT_HASH_TYPE_L2,       /* Input: src_MAC, dest_MAC */
1202        PKT_HASH_TYPE_L3,       /* Input: src_IP, dst_IP */
1203        PKT_HASH_TYPE_L4,       /* Input: src_IP, dst_IP, src_port, dst_port */
1204};
1205
1206static inline void skb_clear_hash(struct sk_buff *skb)
1207{
1208        skb->hash = 0;
1209        skb->sw_hash = 0;
1210        skb->l4_hash = 0;
1211}
1212
1213static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1214{
1215        if (!skb->l4_hash)
1216                skb_clear_hash(skb);
1217}
1218
1219static inline void
1220__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1221{
1222        skb->l4_hash = is_l4;
1223        skb->sw_hash = is_sw;
1224        skb->hash = hash;
1225}
1226
1227static inline void
1228skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1229{
1230        /* Used by drivers to set hash from HW */
1231        __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1232}
1233
1234static inline void
1235__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1236{
1237        __skb_set_hash(skb, hash, true, is_l4);
1238}
1239
1240void __skb_get_hash(struct sk_buff *skb);
1241u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1242u32 skb_get_poff(const struct sk_buff *skb);
1243u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1244                   const struct flow_keys_basic *keys, int hlen);
1245__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1246                            void *data, int hlen_proto);
1247
1248static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1249                                        int thoff, u8 ip_proto)
1250{
1251        return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1252}
1253
1254void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1255                             const struct flow_dissector_key *key,
1256                             unsigned int key_count);
1257
1258#ifdef CONFIG_NET
1259int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1260                                  union bpf_attr __user *uattr);
1261int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1262                                       struct bpf_prog *prog);
1263
1264int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr);
1265#else
1266static inline int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1267                                                union bpf_attr __user *uattr)
1268{
1269        return -EOPNOTSUPP;
1270}
1271
1272static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1273                                                     struct bpf_prog *prog)
1274{
1275        return -EOPNOTSUPP;
1276}
1277
1278static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr)
1279{
1280        return -EOPNOTSUPP;
1281}
1282#endif
1283
1284struct bpf_flow_dissector;
1285bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1286                      __be16 proto, int nhoff, int hlen);
1287
1288bool __skb_flow_dissect(const struct net *net,
1289                        const struct sk_buff *skb,
1290                        struct flow_dissector *flow_dissector,
1291                        void *target_container,
1292                        void *data, __be16 proto, int nhoff, int hlen,
1293                        unsigned int flags);
1294
1295static inline bool skb_flow_dissect(const struct sk_buff *skb,
1296                                    struct flow_dissector *flow_dissector,
1297                                    void *target_container, unsigned int flags)
1298{
1299        return __skb_flow_dissect(NULL, skb, flow_dissector,
1300                                  target_container, NULL, 0, 0, 0, flags);
1301}
1302
1303static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1304                                              struct flow_keys *flow,
1305                                              unsigned int flags)
1306{
1307        memset(flow, 0, sizeof(*flow));
1308        return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1309                                  flow, NULL, 0, 0, 0, flags);
1310}
1311
1312static inline bool
1313skb_flow_dissect_flow_keys_basic(const struct net *net,
1314                                 const struct sk_buff *skb,
1315                                 struct flow_keys_basic *flow, void *data,
1316                                 __be16 proto, int nhoff, int hlen,
1317                                 unsigned int flags)
1318{
1319        memset(flow, 0, sizeof(*flow));
1320        return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1321                                  data, proto, nhoff, hlen, flags);
1322}
1323
1324void skb_flow_dissect_meta(const struct sk_buff *skb,
1325                           struct flow_dissector *flow_dissector,
1326                           void *target_container);
1327
1328/* Gets a skb connection tracking info, ctinfo map should be a
1329 * a map of mapsize to translate enum ip_conntrack_info states
1330 * to user states.
1331 */
1332void
1333skb_flow_dissect_ct(const struct sk_buff *skb,
1334                    struct flow_dissector *flow_dissector,
1335                    void *target_container,
1336                    u16 *ctinfo_map,
1337                    size_t mapsize);
1338void
1339skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1340                             struct flow_dissector *flow_dissector,
1341                             void *target_container);
1342
1343static inline __u32 skb_get_hash(struct sk_buff *skb)
1344{
1345        if (!skb->l4_hash && !skb->sw_hash)
1346                __skb_get_hash(skb);
1347
1348        return skb->hash;
1349}
1350
1351static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1352{
1353        if (!skb->l4_hash && !skb->sw_hash) {
1354                struct flow_keys keys;
1355                __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1356
1357                __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1358        }
1359
1360        return skb->hash;
1361}
1362
1363__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1364
1365static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1366{
1367        return skb->hash;
1368}
1369
1370static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1371{
1372        to->hash = from->hash;
1373        to->sw_hash = from->sw_hash;
1374        to->l4_hash = from->l4_hash;
1375};
1376
1377static inline void skb_copy_decrypted(struct sk_buff *to,
1378                                      const struct sk_buff *from)
1379{
1380#ifdef CONFIG_TLS_DEVICE
1381        to->decrypted = from->decrypted;
1382#endif
1383}
1384
1385#ifdef NET_SKBUFF_DATA_USES_OFFSET
1386static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1387{
1388        return skb->head + skb->end;
1389}
1390
1391static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1392{
1393        return skb->end;
1394}
1395#else
1396static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1397{
1398        return skb->end;
1399}
1400
1401static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1402{
1403        return skb->end - skb->head;
1404}
1405#endif
1406
1407/* Internal */
1408#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1409
1410static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1411{
1412        return &skb_shinfo(skb)->hwtstamps;
1413}
1414
1415static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1416{
1417        bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1418
1419        return is_zcopy ? skb_uarg(skb) : NULL;
1420}
1421
1422static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1423                                 bool *have_ref)
1424{
1425        if (skb && uarg && !skb_zcopy(skb)) {
1426                if (unlikely(have_ref && *have_ref))
1427                        *have_ref = false;
1428                else
1429                        sock_zerocopy_get(uarg);
1430                skb_shinfo(skb)->destructor_arg = uarg;
1431                skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1432        }
1433}
1434
1435static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1436{
1437        skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1438        skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1439}
1440
1441static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1442{
1443        return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1444}
1445
1446static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1447{
1448        return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1449}
1450
1451/* Release a reference on a zerocopy structure */
1452static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1453{
1454        struct ubuf_info *uarg = skb_zcopy(skb);
1455
1456        if (uarg) {
1457                if (skb_zcopy_is_nouarg(skb)) {
1458                        /* no notification callback */
1459                } else if (uarg->callback == sock_zerocopy_callback) {
1460                        uarg->zerocopy = uarg->zerocopy && zerocopy;
1461                        sock_zerocopy_put(uarg);
1462                } else {
1463                        uarg->callback(uarg, zerocopy);
1464                }
1465
1466                skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1467        }
1468}
1469
1470/* Abort a zerocopy operation and revert zckey on error in send syscall */
1471static inline void skb_zcopy_abort(struct sk_buff *skb)
1472{
1473        struct ubuf_info *uarg = skb_zcopy(skb);
1474
1475        if (uarg) {
1476                sock_zerocopy_put_abort(uarg, false);
1477                skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1478        }
1479}
1480
1481static inline void skb_mark_not_on_list(struct sk_buff *skb)
1482{
1483        skb->next = NULL;
1484}
1485
1486static inline void skb_list_del_init(struct sk_buff *skb)
1487{
1488        __list_del_entry(&skb->list);
1489        skb_mark_not_on_list(skb);
1490}
1491
1492/**
1493 *      skb_queue_empty - check if a queue is empty
1494 *      @list: queue head
1495 *
1496 *      Returns true if the queue is empty, false otherwise.
1497 */
1498static inline int skb_queue_empty(const struct sk_buff_head *list)
1499{
1500        return list->next == (const struct sk_buff *) list;
1501}
1502
1503/**
1504 *      skb_queue_is_last - check if skb is the last entry in the queue
1505 *      @list: queue head
1506 *      @skb: buffer
1507 *
1508 *      Returns true if @skb is the last buffer on the list.
1509 */
1510static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1511                                     const struct sk_buff *skb)
1512{
1513        return skb->next == (const struct sk_buff *) list;
1514}
1515
1516/**
1517 *      skb_queue_is_first - check if skb is the first entry in the queue
1518 *      @list: queue head
1519 *      @skb: buffer
1520 *
1521 *      Returns true if @skb is the first buffer on the list.
1522 */
1523static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1524                                      const struct sk_buff *skb)
1525{
1526        return skb->prev == (const struct sk_buff *) list;
1527}
1528
1529/**
1530 *      skb_queue_next - return the next packet in the queue
1531 *      @list: queue head
1532 *      @skb: current buffer
1533 *
1534 *      Return the next packet in @list after @skb.  It is only valid to
1535 *      call this if skb_queue_is_last() evaluates to false.
1536 */
1537static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1538                                             const struct sk_buff *skb)
1539{
1540        /* This BUG_ON may seem severe, but if we just return then we
1541         * are going to dereference garbage.
1542         */
1543        BUG_ON(skb_queue_is_last(list, skb));
1544        return skb->next;
1545}
1546
1547/**
1548 *      skb_queue_prev - return the prev packet in the queue
1549 *      @list: queue head
1550 *      @skb: current buffer
1551 *
1552 *      Return the prev packet in @list before @skb.  It is only valid to
1553 *      call this if skb_queue_is_first() evaluates to false.
1554 */
1555static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1556                                             const struct sk_buff *skb)
1557{
1558        /* This BUG_ON may seem severe, but if we just return then we
1559         * are going to dereference garbage.
1560         */
1561        BUG_ON(skb_queue_is_first(list, skb));
1562        return skb->prev;
1563}
1564
1565/**
1566 *      skb_get - reference buffer
1567 *      @skb: buffer to reference
1568 *
1569 *      Makes another reference to a socket buffer and returns a pointer
1570 *      to the buffer.
1571 */
1572static inline struct sk_buff *skb_get(struct sk_buff *skb)
1573{
1574        refcount_inc(&skb->users);
1575        return skb;
1576}
1577
1578/*
1579 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1580 */
1581
1582/**
1583 *      skb_cloned - is the buffer a clone
1584 *      @skb: buffer to check
1585 *
1586 *      Returns true if the buffer was generated with skb_clone() and is
1587 *      one of multiple shared copies of the buffer. Cloned buffers are
1588 *      shared data so must not be written to under normal circumstances.
1589 */
1590static inline int skb_cloned(const struct sk_buff *skb)
1591{
1592        return skb->cloned &&
1593               (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1594}
1595
1596static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1597{
1598        might_sleep_if(gfpflags_allow_blocking(pri));
1599
1600        if (skb_cloned(skb))
1601                return pskb_expand_head(skb, 0, 0, pri);
1602
1603        return 0;
1604}
1605
1606/**
1607 *      skb_header_cloned - is the header a clone
1608 *      @skb: buffer to check
1609 *
1610 *      Returns true if modifying the header part of the buffer requires
1611 *      the data to be copied.
1612 */
1613static inline int skb_header_cloned(const struct sk_buff *skb)
1614{
1615        int dataref;
1616
1617        if (!skb->cloned)
1618                return 0;
1619
1620        dataref = atomic_read(&skb_shinfo(skb)->dataref);
1621        dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1622        return dataref != 1;
1623}
1624
1625static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1626{
1627        might_sleep_if(gfpflags_allow_blocking(pri));
1628
1629        if (skb_header_cloned(skb))
1630                return pskb_expand_head(skb, 0, 0, pri);
1631
1632        return 0;
1633}
1634
1635/**
1636 *      __skb_header_release - release reference to header
1637 *      @skb: buffer to operate on
1638 */
1639static inline void __skb_header_release(struct sk_buff *skb)
1640{
1641        skb->nohdr = 1;
1642        atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1643}
1644
1645
1646/**
1647 *      skb_shared - is the buffer shared
1648 *      @skb: buffer to check
1649 *
1650 *      Returns true if more than one person has a reference to this
1651 *      buffer.
1652 */
1653static inline int skb_shared(const struct sk_buff *skb)
1654{
1655        return refcount_read(&skb->users) != 1;
1656}
1657
1658/**
1659 *      skb_share_check - check if buffer is shared and if so clone it
1660 *      @skb: buffer to check
1661 *      @pri: priority for memory allocation
1662 *
1663 *      If the buffer is shared the buffer is cloned and the old copy
1664 *      drops a reference. A new clone with a single reference is returned.
1665 *      If the buffer is not shared the original buffer is returned. When
1666 *      being called from interrupt status or with spinlocks held pri must
1667 *      be GFP_ATOMIC.
1668 *
1669 *      NULL is returned on a memory allocation failure.
1670 */
1671static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1672{
1673        might_sleep_if(gfpflags_allow_blocking(pri));
1674        if (skb_shared(skb)) {
1675                struct sk_buff *nskb = skb_clone(skb, pri);
1676
1677                if (likely(nskb))
1678                        consume_skb(skb);
1679                else
1680                        kfree_skb(skb);
1681                skb = nskb;
1682        }
1683        return skb;
1684}
1685
1686/*
1687 *      Copy shared buffers into a new sk_buff. We effectively do COW on
1688 *      packets to handle cases where we have a local reader and forward
1689 *      and a couple of other messy ones. The normal one is tcpdumping
1690 *      a packet thats being forwarded.
1691 */
1692
1693/**
1694 *      skb_unshare - make a copy of a shared buffer
1695 *      @skb: buffer to check
1696 *      @pri: priority for memory allocation
1697 *
1698 *      If the socket buffer is a clone then this function creates a new
1699 *      copy of the data, drops a reference count on the old copy and returns
1700 *      the new copy with the reference count at 1. If the buffer is not a clone
1701 *      the original buffer is returned. When called with a spinlock held or
1702 *      from interrupt state @pri must be %GFP_ATOMIC
1703 *
1704 *      %NULL is returned on a memory allocation failure.
1705 */
1706static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1707                                          gfp_t pri)
1708{
1709        might_sleep_if(gfpflags_allow_blocking(pri));
1710        if (skb_cloned(skb)) {
1711                struct sk_buff *nskb = skb_copy(skb, pri);
1712
1713                /* Free our shared copy */
1714                if (likely(nskb))
1715                        consume_skb(skb);
1716                else
1717                        kfree_skb(skb);
1718                skb = nskb;
1719        }
1720        return skb;
1721}
1722
1723/**
1724 *      skb_peek - peek at the head of an &sk_buff_head
1725 *      @list_: list to peek at
1726 *
1727 *      Peek an &sk_buff. Unlike most other operations you _MUST_
1728 *      be careful with this one. A peek leaves the buffer on the
1729 *      list and someone else may run off with it. You must hold
1730 *      the appropriate locks or have a private queue to do this.
1731 *
1732 *      Returns %NULL for an empty list or a pointer to the head element.
1733 *      The reference count is not incremented and the reference is therefore
1734 *      volatile. Use with caution.
1735 */
1736static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1737{
1738        struct sk_buff *skb = list_->next;
1739
1740        if (skb == (struct sk_buff *)list_)
1741                skb = NULL;
1742        return skb;
1743}
1744
1745/**
1746 *      __skb_peek - peek at the head of a non-empty &sk_buff_head
1747 *      @list_: list to peek at
1748 *
1749 *      Like skb_peek(), but the caller knows that the list is not empty.
1750 */
1751static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1752{
1753        return list_->next;
1754}
1755
1756/**
1757 *      skb_peek_next - peek skb following the given one from a queue
1758 *      @skb: skb to start from
1759 *      @list_: list to peek at
1760 *
1761 *      Returns %NULL when the end of the list is met or a pointer to the
1762 *      next element. The reference count is not incremented and the
1763 *      reference is therefore volatile. Use with caution.
1764 */
1765static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1766                const struct sk_buff_head *list_)
1767{
1768        struct sk_buff *next = skb->next;
1769
1770        if (next == (struct sk_buff *)list_)
1771                next = NULL;
1772        return next;
1773}
1774
1775/**
1776 *      skb_peek_tail - peek at the tail of an &sk_buff_head
1777 *      @list_: list to peek at
1778 *
1779 *      Peek an &sk_buff. Unlike most other operations you _MUST_
1780 *      be careful with this one. A peek leaves the buffer on the
1781 *      list and someone else may run off with it. You must hold
1782 *      the appropriate locks or have a private queue to do this.
1783 *
1784 *      Returns %NULL for an empty list or a pointer to the tail element.
1785 *      The reference count is not incremented and the reference is therefore
1786 *      volatile. Use with caution.
1787 */
1788static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1789{
1790        struct sk_buff *skb = list_->prev;
1791
1792        if (skb == (struct sk_buff *)list_)
1793                skb = NULL;
1794        return skb;
1795
1796}
1797
1798/**
1799 *      skb_queue_len   - get queue length
1800 *      @list_: list to measure
1801 *
1802 *      Return the length of an &sk_buff queue.
1803 */
1804static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1805{
1806        return list_->qlen;
1807}
1808
1809/**
1810 *      __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1811 *      @list: queue to initialize
1812 *
1813 *      This initializes only the list and queue length aspects of
1814 *      an sk_buff_head object.  This allows to initialize the list
1815 *      aspects of an sk_buff_head without reinitializing things like
1816 *      the spinlock.  It can also be used for on-stack sk_buff_head
1817 *      objects where the spinlock is known to not be used.
1818 */
1819static inline void __skb_queue_head_init(struct sk_buff_head *list)
1820{
1821        list->prev = list->next = (struct sk_buff *)list;
1822        list->qlen = 0;
1823}
1824
1825/*
1826 * This function creates a split out lock class for each invocation;
1827 * this is needed for now since a whole lot of users of the skb-queue
1828 * infrastructure in drivers have different locking usage (in hardirq)
1829 * than the networking core (in softirq only). In the long run either the
1830 * network layer or drivers should need annotation to consolidate the
1831 * main types of usage into 3 classes.
1832 */
1833static inline void skb_queue_head_init(struct sk_buff_head *list)
1834{
1835        spin_lock_init(&list->lock);
1836        __skb_queue_head_init(list);
1837}
1838
1839static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1840                struct lock_class_key *class)
1841{
1842        skb_queue_head_init(list);
1843        lockdep_set_class(&list->lock, class);
1844}
1845
1846/*
1847 *      Insert an sk_buff on a list.
1848 *
1849 *      The "__skb_xxxx()" functions are the non-atomic ones that
1850 *      can only be called with interrupts disabled.
1851 */
1852static inline void __skb_insert(struct sk_buff *newsk,
1853                                struct sk_buff *prev, struct sk_buff *next,
1854                                struct sk_buff_head *list)
1855{
1856        newsk->next = next;
1857        newsk->prev = prev;
1858        next->prev  = prev->next = newsk;
1859        list->qlen++;
1860}
1861
1862static inline void __skb_queue_splice(const struct sk_buff_head *list,
1863                                      struct sk_buff *prev,
1864                                      struct sk_buff *next)
1865{
1866        struct sk_buff *first = list->next;
1867        struct sk_buff *last = list->prev;
1868
1869        first->prev = prev;
1870        prev->next = first;
1871
1872        last->next = next;
1873        next->prev = last;
1874}
1875
1876/**
1877 *      skb_queue_splice - join two skb lists, this is designed for stacks
1878 *      @list: the new list to add
1879 *      @head: the place to add it in the first list
1880 */
1881static inline void skb_queue_splice(const struct sk_buff_head *list,
1882                                    struct sk_buff_head *head)
1883{
1884        if (!skb_queue_empty(list)) {
1885                __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1886                head->qlen += list->qlen;
1887        }
1888}
1889
1890/**
1891 *      skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1892 *      @list: the new list to add
1893 *      @head: the place to add it in the first list
1894 *
1895 *      The list at @list is reinitialised
1896 */
1897static inline void skb_queue_splice_init(struct sk_buff_head *list,
1898                                         struct sk_buff_head *head)
1899{
1900        if (!skb_queue_empty(list)) {
1901                __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1902                head->qlen += list->qlen;
1903                __skb_queue_head_init(list);
1904        }
1905}
1906
1907/**
1908 *      skb_queue_splice_tail - join two skb lists, each list being a queue
1909 *      @list: the new list to add
1910 *      @head: the place to add it in the first list
1911 */
1912static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1913                                         struct sk_buff_head *head)
1914{
1915        if (!skb_queue_empty(list)) {
1916                __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1917                head->qlen += list->qlen;
1918        }
1919}
1920
1921/**
1922 *      skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1923 *      @list: the new list to add
1924 *      @head: the place to add it in the first list
1925 *
1926 *      Each of the lists is a queue.
1927 *      The list at @list is reinitialised
1928 */
1929static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1930                                              struct sk_buff_head *head)
1931{
1932        if (!skb_queue_empty(list)) {
1933                __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1934                head->qlen += list->qlen;
1935                __skb_queue_head_init(list);
1936        }
1937}
1938
1939/**
1940 *      __skb_queue_after - queue a buffer at the list head
1941 *      @list: list to use
1942 *      @prev: place after this buffer
1943 *      @newsk: buffer to queue
1944 *
1945 *      Queue a buffer int the middle of a list. This function takes no locks
1946 *      and you must therefore hold required locks before calling it.
1947 *
1948 *      A buffer cannot be placed on two lists at the same time.
1949 */
1950static inline void __skb_queue_after(struct sk_buff_head *list,
1951                                     struct sk_buff *prev,
1952                                     struct sk_buff *newsk)
1953{
1954        __skb_insert(newsk, prev, prev->next, list);
1955}
1956
1957void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1958                struct sk_buff_head *list);
1959
1960static inline void __skb_queue_before(struct sk_buff_head *list,
1961                                      struct sk_buff *next,
1962                                      struct sk_buff *newsk)
1963{
1964        __skb_insert(newsk, next->prev, next, list);
1965}
1966
1967/**
1968 *      __skb_queue_head - queue a buffer at the list head
1969 *      @list: list to use
1970 *      @newsk: buffer to queue
1971 *
1972 *      Queue a buffer at the start of a list. This function takes no locks
1973 *      and you must therefore hold required locks before calling it.
1974 *
1975 *      A buffer cannot be placed on two lists at the same time.
1976 */
1977static inline void __skb_queue_head(struct sk_buff_head *list,
1978                                    struct sk_buff *newsk)
1979{
1980        __skb_queue_after(list, (struct sk_buff *)list, newsk);
1981}
1982void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1983
1984/**
1985 *      __skb_queue_tail - queue a buffer at the list tail
1986 *      @list: list to use
1987 *      @newsk: buffer to queue
1988 *
1989 *      Queue a buffer at the end of a list. This function takes no locks
1990 *      and you must therefore hold required locks before calling it.
1991 *
1992 *      A buffer cannot be placed on two lists at the same time.
1993 */
1994static inline void __skb_queue_tail(struct sk_buff_head *list,
1995                                   struct sk_buff *newsk)
1996{
1997        __skb_queue_before(list, (struct sk_buff *)list, newsk);
1998}
1999void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
2000
2001/*
2002 * remove sk_buff from list. _Must_ be called atomically, and with
2003 * the list known..
2004 */
2005void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
2006static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2007{
2008        struct sk_buff *next, *prev;
2009
2010        list->qlen--;
2011        next       = skb->next;
2012        prev       = skb->prev;
2013        skb->next  = skb->prev = NULL;
2014        next->prev = prev;
2015        prev->next = next;
2016}
2017
2018/**
2019 *      __skb_dequeue - remove from the head of the queue
2020 *      @list: list to dequeue from
2021 *
2022 *      Remove the head of the list. This function does not take any locks
2023 *      so must be used with appropriate locks held only. The head item is
2024 *      returned or %NULL if the list is empty.
2025 */
2026static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2027{
2028        struct sk_buff *skb = skb_peek(list);
2029        if (skb)
2030                __skb_unlink(skb, list);
2031        return skb;
2032}
2033struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2034
2035/**
2036 *      __skb_dequeue_tail - remove from the tail of the queue
2037 *      @list: list to dequeue from
2038 *
2039 *      Remove the tail of the list. This function does not take any locks
2040 *      so must be used with appropriate locks held only. The tail item is
2041 *      returned or %NULL if the list is empty.
2042 */
2043static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2044{
2045        struct sk_buff *skb = skb_peek_tail(list);
2046        if (skb)
2047                __skb_unlink(skb, list);
2048        return skb;
2049}
2050struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2051
2052
2053static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2054{
2055        return skb->data_len;
2056}
2057
2058static inline unsigned int skb_headlen(const struct sk_buff *skb)
2059{
2060        return skb->len - skb->data_len;
2061}
2062
2063static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2064{
2065        unsigned int i, len = 0;
2066
2067        for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2068                len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2069        return len;
2070}
2071
2072static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2073{
2074        return skb_headlen(skb) + __skb_pagelen(skb);
2075}
2076
2077/**
2078 * __skb_fill_page_desc - initialise a paged fragment in an skb
2079 * @skb: buffer containing fragment to be initialised
2080 * @i: paged fragment index to initialise
2081 * @page: the page to use for this fragment
2082 * @off: the offset to the data with @page
2083 * @size: the length of the data
2084 *
2085 * Initialises the @i'th fragment of @skb to point to &size bytes at
2086 * offset @off within @page.
2087 *
2088 * Does not take any additional reference on the fragment.
2089 */
2090static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2091                                        struct page *page, int off, int size)
2092{
2093        skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2094
2095        /*
2096         * Propagate page pfmemalloc to the skb if we can. The problem is
2097         * that not all callers have unique ownership of the page but rely
2098         * on page_is_pfmemalloc doing the right thing(tm).
2099         */
2100        frag->page.p              = page;
2101        frag->page_offset         = off;
2102        skb_frag_size_set(frag, size);
2103
2104        page = compound_head(page);
2105        if (page_is_pfmemalloc(page))
2106                skb->pfmemalloc = true;
2107}
2108
2109/**
2110 * skb_fill_page_desc - initialise a paged fragment in an skb
2111 * @skb: buffer containing fragment to be initialised
2112 * @i: paged fragment index to initialise
2113 * @page: the page to use for this fragment
2114 * @off: the offset to the data with @page
2115 * @size: the length of the data
2116 *
2117 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2118 * @skb to point to @size bytes at offset @off within @page. In
2119 * addition updates @skb such that @i is the last fragment.
2120 *
2121 * Does not take any additional reference on the fragment.
2122 */
2123static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2124                                      struct page *page, int off, int size)
2125{
2126        __skb_fill_page_desc(skb, i, page, off, size);
2127        skb_shinfo(skb)->nr_frags = i + 1;
2128}
2129
2130void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2131                     int size, unsigned int truesize);
2132
2133void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2134                          unsigned int truesize);
2135
2136#define SKB_LINEAR_ASSERT(skb)  BUG_ON(skb_is_nonlinear(skb))
2137
2138#ifdef NET_SKBUFF_DATA_USES_OFFSET
2139static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2140{
2141        return skb->head + skb->tail;
2142}
2143
2144static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2145{
2146        skb->tail = skb->data - skb->head;
2147}
2148
2149static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2150{
2151        skb_reset_tail_pointer(skb);
2152        skb->tail += offset;
2153}
2154
2155#else /* NET_SKBUFF_DATA_USES_OFFSET */
2156static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2157{
2158        return skb->tail;
2159}
2160
2161static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2162{
2163        skb->tail = skb->data;
2164}
2165
2166static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2167{
2168        skb->tail = skb->data + offset;
2169}
2170
2171#endif /* NET_SKBUFF_DATA_USES_OFFSET */
2172
2173/*
2174 *      Add data to an sk_buff
2175 */
2176void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2177void *skb_put(struct sk_buff *skb, unsigned int len);
2178static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2179{
2180        void *tmp = skb_tail_pointer(skb);
2181        SKB_LINEAR_ASSERT(skb);
2182        skb->tail += len;
2183        skb->len  += len;
2184        return tmp;
2185}
2186
2187static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2188{
2189        void *tmp = __skb_put(skb, len);
2190
2191        memset(tmp, 0, len);
2192        return tmp;
2193}
2194
2195static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2196                                   unsigned int len)
2197{
2198        void *tmp = __skb_put(skb, len);
2199
2200        memcpy(tmp, data, len);
2201        return tmp;
2202}
2203
2204static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2205{
2206        *(u8 *)__skb_put(skb, 1) = val;
2207}
2208
2209static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2210{
2211        void *tmp = skb_put(skb, len);
2212
2213        memset(tmp, 0, len);
2214
2215        return tmp;
2216}
2217
2218static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2219                                 unsigned int len)
2220{
2221        void *tmp = skb_put(skb, len);
2222
2223        memcpy(tmp, data, len);
2224
2225        return tmp;
2226}
2227
2228static inline void skb_put_u8(struct sk_buff *skb, u8 val)
2229{
2230        *(u8 *)skb_put(skb, 1) = val;
2231}
2232
2233void *skb_push(struct sk_buff *skb, unsigned int len);
2234static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2235{
2236        skb->data -= len;
2237        skb->len  += len;
2238        return skb->data;
2239}
2240
2241void *skb_pull(struct sk_buff *skb, unsigned int len);
2242static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2243{
2244        skb->len -= len;
2245        BUG_ON(skb->len < skb->data_len);
2246        return skb->data += len;
2247}
2248
2249static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2250{
2251        return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2252}
2253
2254void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2255
2256static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2257{
2258        if (len > skb_headlen(skb) &&
2259            !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2260                return NULL;
2261        skb->len -= len;
2262        return skb->data += len;
2263}
2264
2265static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2266{
2267        return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2268}
2269
2270static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2271{
2272        if (likely(len <= skb_headlen(skb)))
2273                return 1;
2274        if (unlikely(len > skb->len))
2275                return 0;
2276        return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2277}
2278
2279void skb_condense(struct sk_buff *skb);
2280
2281/**
2282 *      skb_headroom - bytes at buffer head
2283 *      @skb: buffer to check
2284 *
2285 *      Return the number of bytes of free space at the head of an &sk_buff.
2286 */
2287static inline unsigned int skb_headroom(const struct sk_buff *skb)
2288{
2289        return skb->data - skb->head;
2290}
2291
2292/**
2293 *      skb_tailroom - bytes at buffer end
2294 *      @skb: buffer to check
2295 *
2296 *      Return the number of bytes of free space at the tail of an sk_buff
2297 */
2298static inline int skb_tailroom(const struct sk_buff *skb)
2299{
2300        return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2301}
2302
2303/**
2304 *      skb_availroom - bytes at buffer end
2305 *      @skb: buffer to check
2306 *
2307 *      Return the number of bytes of free space at the tail of an sk_buff
2308 *      allocated by sk_stream_alloc()
2309 */
2310static inline int skb_availroom(const struct sk_buff *skb)
2311{
2312        if (skb_is_nonlinear(skb))
2313                return 0;
2314
2315        return skb->end - skb->tail - skb->reserved_tailroom;
2316}
2317
2318/**
2319 *      skb_reserve - adjust headroom
2320 *      @skb: buffer to alter
2321 *      @len: bytes to move
2322 *
2323 *      Increase the headroom of an empty &sk_buff by reducing the tail
2324 *      room. This is only allowed for an empty buffer.
2325 */
2326static inline void skb_reserve(struct sk_buff *skb, int len)
2327{
2328        skb->data += len;
2329        skb->tail += len;
2330}
2331
2332/**
2333 *      skb_tailroom_reserve - adjust reserved_tailroom
2334 *      @skb: buffer to alter
2335 *      @mtu: maximum amount of headlen permitted
2336 *      @needed_tailroom: minimum amount of reserved_tailroom
2337 *
2338 *      Set reserved_tailroom so that headlen can be as large as possible but
2339 *      not larger than mtu and tailroom cannot be smaller than
2340 *      needed_tailroom.
2341 *      The required headroom should already have been reserved before using
2342 *      this function.
2343 */
2344static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2345                                        unsigned int needed_tailroom)
2346{
2347        SKB_LINEAR_ASSERT(skb);
2348        if (mtu < skb_tailroom(skb) - needed_tailroom)
2349                /* use at most mtu */
2350                skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2351        else
2352                /* use up to all available space */
2353                skb->reserved_tailroom = needed_tailroom;
2354}
2355
2356#define ENCAP_TYPE_ETHER        0
2357#define ENCAP_TYPE_IPPROTO      1
2358
2359static inline void skb_set_inner_protocol(struct sk_buff *skb,
2360                                          __be16 protocol)
2361{
2362        skb->inner_protocol = protocol;
2363        skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2364}
2365
2366static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2367                                         __u8 ipproto)
2368{
2369        skb->inner_ipproto = ipproto;
2370        skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2371}
2372
2373static inline void skb_reset_inner_headers(struct sk_buff *skb)
2374{
2375        skb->inner_mac_header = skb->mac_header;
2376        skb->inner_network_header = skb->network_header;
2377        skb->inner_transport_header = skb->transport_header;
2378}
2379
2380static inline void skb_reset_mac_len(struct sk_buff *skb)
2381{
2382        skb->mac_len = skb->network_header - skb->mac_header;
2383}
2384
2385static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2386                                                        *skb)
2387{
2388        return skb->head + skb->inner_transport_header;
2389}
2390
2391static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2392{
2393        return skb_inner_transport_header(skb) - skb->data;
2394}
2395
2396static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2397{
2398        skb->inner_transport_header = skb->data - skb->head;
2399}
2400
2401static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2402                                                   const int offset)
2403{
2404        skb_reset_inner_transport_header(skb);
2405        skb->inner_transport_header += offset;
2406}
2407
2408static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2409{
2410        return skb->head + skb->inner_network_header;
2411}
2412
2413static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2414{
2415        skb->inner_network_header = skb->data - skb->head;
2416}
2417
2418static inline void skb_set_inner_network_header(struct sk_buff *skb,
2419                                                const int offset)
2420{
2421        skb_reset_inner_network_header(skb);
2422        skb->inner_network_header += offset;
2423}
2424
2425static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2426{
2427        return skb->head + skb->inner_mac_header;
2428}
2429
2430static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2431{
2432        skb->inner_mac_header = skb->data - skb->head;
2433}
2434
2435static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2436                                            const int offset)
2437{
2438        skb_reset_inner_mac_header(skb);
2439        skb->inner_mac_header += offset;
2440}
2441static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2442{
2443        return skb->transport_header != (typeof(skb->transport_header))~0U;
2444}
2445
2446static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2447{
2448        return skb->head + skb->transport_header;
2449}
2450
2451static inline void skb_reset_transport_header(struct sk_buff *skb)
2452{
2453        skb->transport_header = skb->data - skb->head;
2454}
2455
2456static inline void skb_set_transport_header(struct sk_buff *skb,
2457                                            const int offset)
2458{
2459        skb_reset_transport_header(skb);
2460        skb->transport_header += offset;
2461}
2462
2463static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2464{
2465        return skb->head + skb->network_header;
2466}
2467
2468static inline void skb_reset_network_header(struct sk_buff *skb)
2469{
2470        skb->network_header = skb->data - skb->head;
2471}
2472
2473static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2474{
2475        skb_reset_network_header(skb);
2476        skb->network_header += offset;
2477}
2478
2479static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2480{
2481        return skb->head + skb->mac_header;
2482}
2483
2484static inline int skb_mac_offset(const struct sk_buff *skb)
2485{
2486        return skb_mac_header(skb) - skb->data;
2487}
2488
2489static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2490{
2491        return skb->network_header - skb->mac_header;
2492}
2493
2494static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2495{
2496        return skb->mac_header != (typeof(skb->mac_header))~0U;
2497}
2498
2499static inline void skb_reset_mac_header(struct sk_buff *skb)
2500{
2501        skb->mac_header = skb->data - skb->head;
2502}
2503
2504static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2505{
2506        skb_reset_mac_header(skb);
2507        skb->mac_header += offset;
2508}
2509
2510static inline void skb_pop_mac_header(struct sk_buff *skb)
2511{
2512        skb->mac_header = skb->network_header;
2513}
2514
2515static inline void skb_probe_transport_header(struct sk_buff *skb)
2516{
2517        struct flow_keys_basic keys;
2518
2519        if (skb_transport_header_was_set(skb))
2520                return;
2521
2522        if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2523                                             NULL, 0, 0, 0, 0))
2524                skb_set_transport_header(skb, keys.control.thoff);
2525}
2526
2527static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2528{
2529        if (skb_mac_header_was_set(skb)) {
2530                const unsigned char *old_mac = skb_mac_header(skb);
2531
2532                skb_set_mac_header(skb, -skb->mac_len);
2533                memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2534        }
2535}
2536
2537static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2538{
2539        return skb->csum_start - skb_headroom(skb);
2540}
2541
2542static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2543{
2544        return skb->head + skb->csum_start;
2545}
2546
2547static inline int skb_transport_offset(const struct sk_buff *skb)
2548{
2549        return skb_transport_header(skb) - skb->data;
2550}
2551
2552static inline u32 skb_network_header_len(const struct sk_buff *skb)
2553{
2554        return skb->transport_header - skb->network_header;
2555}
2556
2557static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2558{
2559        return skb->inner_transport_header - skb->inner_network_header;
2560}
2561
2562static inline int skb_network_offset(const struct sk_buff *skb)
2563{
2564        return skb_network_header(skb) - skb->data;
2565}
2566
2567static inline int skb_inner_network_offset(const struct sk_buff *skb)
2568{
2569        return skb_inner_network_header(skb) - skb->data;
2570}
2571
2572static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2573{
2574        return pskb_may_pull(skb, skb_network_offset(skb) + len);
2575}
2576
2577/*
2578 * CPUs often take a performance hit when accessing unaligned memory
2579 * locations. The actual performance hit varies, it can be small if the
2580 * hardware handles it or large if we have to take an exception and fix it
2581 * in software.
2582 *
2583 * Since an ethernet header is 14 bytes network drivers often end up with
2584 * the IP header at an unaligned offset. The IP header can be aligned by
2585 * shifting the start of the packet by 2 bytes. Drivers should do this
2586 * with:
2587 *
2588 * skb_reserve(skb, NET_IP_ALIGN);
2589 *
2590 * The downside to this alignment of the IP header is that the DMA is now
2591 * unaligned. On some architectures the cost of an unaligned DMA is high
2592 * and this cost outweighs the gains made by aligning the IP header.
2593 *
2594 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2595 * to be overridden.
2596 */
2597#ifndef NET_IP_ALIGN
2598#define NET_IP_ALIGN    2
2599#endif
2600
2601/*
2602 * The networking layer reserves some headroom in skb data (via
2603 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2604 * the header has to grow. In the default case, if the header has to grow
2605 * 32 bytes or less we avoid the reallocation.
2606 *
2607 * Unfortunately this headroom changes the DMA alignment of the resulting
2608 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2609 * on some architectures. An architecture can override this value,
2610 * perhaps setting it to a cacheline in size (since that will maintain
2611 * cacheline alignment of the DMA). It must be a power of 2.
2612 *
2613 * Various parts of the networking layer expect at least 32 bytes of
2614 * headroom, you should not reduce this.
2615 *
2616 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2617 * to reduce average number of cache lines per packet.
2618 * get_rps_cpus() for example only access one 64 bytes aligned block :
2619 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2620 */
2621#ifndef NET_SKB_PAD
2622#define NET_SKB_PAD     max(32, L1_CACHE_BYTES)
2623#endif
2624
2625int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2626
2627static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2628{
2629        if (WARN_ON(skb_is_nonlinear(skb)))
2630                return;
2631        skb->len = len;
2632        skb_set_tail_pointer(skb, len);
2633}
2634
2635static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2636{
2637        __skb_set_length(skb, len);
2638}
2639
2640void skb_trim(struct sk_buff *skb, unsigned int len);
2641
2642static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2643{
2644        if (skb->data_len)
2645                return ___pskb_trim(skb, len);
2646        __skb_trim(skb, len);
2647        return 0;
2648}
2649
2650static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2651{
2652        return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2653}
2654
2655/**
2656 *      pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2657 *      @skb: buffer to alter
2658 *      @len: new length
2659 *
2660 *      This is identical to pskb_trim except that the caller knows that
2661 *      the skb is not cloned so we should never get an error due to out-
2662 *      of-memory.
2663 */
2664static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2665{
2666        int err = pskb_trim(skb, len);
2667        BUG_ON(err);
2668}
2669
2670static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2671{
2672        unsigned int diff = len - skb->len;
2673
2674        if (skb_tailroom(skb) < diff) {
2675                int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2676                                           GFP_ATOMIC);
2677                if (ret)
2678                        return ret;
2679        }
2680        __skb_set_length(skb, len);
2681        return 0;
2682}
2683
2684/**
2685 *      skb_orphan - orphan a buffer
2686 *      @skb: buffer to orphan
2687 *
2688 *      If a buffer currently has an owner then we call the owner's
2689 *      destructor function and make the @skb unowned. The buffer continues
2690 *      to exist but is no longer charged to its former owner.
2691 */
2692static inline void skb_orphan(struct sk_buff *skb)
2693{
2694        if (skb->destructor) {
2695                skb->destructor(skb);
2696                skb->destructor = NULL;
2697                skb->sk         = NULL;
2698        } else {
2699                BUG_ON(skb->sk);
2700        }
2701}
2702
2703/**
2704 *      skb_orphan_frags - orphan the frags contained in a buffer
2705 *      @skb: buffer to orphan frags from
2706 *      @gfp_mask: allocation mask for replacement pages
2707 *
2708 *      For each frag in the SKB which needs a destructor (i.e. has an
2709 *      owner) create a copy of that frag and release the original
2710 *      page by calling the destructor.
2711 */
2712static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2713{
2714        if (likely(!skb_zcopy(skb)))
2715                return 0;
2716        if (!skb_zcopy_is_nouarg(skb) &&
2717            skb_uarg(skb)->callback == sock_zerocopy_callback)
2718                return 0;
2719        return skb_copy_ubufs(skb, gfp_mask);
2720}
2721
2722/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2723static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2724{
2725        if (likely(!skb_zcopy(skb)))
2726                return 0;
2727        return skb_copy_ubufs(skb, gfp_mask);
2728}
2729
2730/**
2731 *      __skb_queue_purge - empty a list
2732 *      @list: list to empty
2733 *
2734 *      Delete all buffers on an &sk_buff list. Each buffer is removed from
2735 *      the list and one reference dropped. This function does not take the
2736 *      list lock and the caller must hold the relevant locks to use it.
2737 */
2738static inline void __skb_queue_purge(struct sk_buff_head *list)
2739{
2740        struct sk_buff *skb;
2741        while ((skb = __skb_dequeue(list)) != NULL)
2742                kfree_skb(skb);
2743}
2744void skb_queue_purge(struct sk_buff_head *list);
2745
2746unsigned int skb_rbtree_purge(struct rb_root *root);
2747
2748void *netdev_alloc_frag(unsigned int fragsz);
2749
2750struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2751                                   gfp_t gfp_mask);
2752
2753/**
2754 *      netdev_alloc_skb - allocate an skbuff for rx on a specific device
2755 *      @dev: network device to receive on
2756 *      @length: length to allocate
2757 *
2758 *      Allocate a new &sk_buff and assign it a usage count of one. The
2759 *      buffer has unspecified headroom built in. Users should allocate
2760 *      the headroom they think they need without accounting for the
2761 *      built in space. The built in space is used for optimisations.
2762 *
2763 *      %NULL is returned if there is no free memory. Although this function
2764 *      allocates memory it can be called from an interrupt.
2765 */
2766static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2767                                               unsigned int length)
2768{
2769        return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2770}
2771
2772/* legacy helper around __netdev_alloc_skb() */
2773static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2774                                              gfp_t gfp_mask)
2775{
2776        return __netdev_alloc_skb(NULL, length, gfp_mask);
2777}
2778
2779/* legacy helper around netdev_alloc_skb() */
2780static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2781{
2782        return netdev_alloc_skb(NULL, length);
2783}
2784
2785
2786static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2787                unsigned int length, gfp_t gfp)
2788{
2789        struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2790
2791        if (NET_IP_ALIGN && skb)
2792                skb_reserve(skb, NET_IP_ALIGN);
2793        return skb;
2794}
2795
2796static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2797                unsigned int length)
2798{
2799        return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2800}
2801
2802static inline void skb_free_frag(void *addr)
2803{
2804        page_frag_free(addr);
2805}
2806
2807void *napi_alloc_frag(unsigned int fragsz);
2808struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2809                                 unsigned int length, gfp_t gfp_mask);
2810static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2811                                             unsigned int length)
2812{
2813        return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2814}
2815void napi_consume_skb(struct sk_buff *skb, int budget);
2816
2817void __kfree_skb_flush(void);
2818void __kfree_skb_defer(struct sk_buff *skb);
2819
2820/**
2821 * __dev_alloc_pages - allocate page for network Rx
2822 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2823 * @order: size of the allocation
2824 *
2825 * Allocate a new page.
2826 *
2827 * %NULL is returned if there is no free memory.
2828*/
2829static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2830                                             unsigned int order)
2831{
2832        /* This piece of code contains several assumptions.
2833         * 1.  This is for device Rx, therefor a cold page is preferred.
2834         * 2.  The expectation is the user wants a compound page.
2835         * 3.  If requesting a order 0 page it will not be compound
2836         *     due to the check to see if order has a value in prep_new_page
2837         * 4.  __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2838         *     code in gfp_to_alloc_flags that should be enforcing this.
2839         */
2840        gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2841
2842        return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2843}
2844
2845static inline struct page *dev_alloc_pages(unsigned int order)
2846{
2847        return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2848}
2849
2850/**
2851 * __dev_alloc_page - allocate a page for network Rx
2852 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2853 *
2854 * Allocate a new page.
2855 *
2856 * %NULL is returned if there is no free memory.
2857 */
2858static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2859{
2860        return __dev_alloc_pages(gfp_mask, 0);
2861}
2862
2863static inline struct page *dev_alloc_page(void)
2864{
2865        return dev_alloc_pages(0);
2866}
2867
2868/**
2869 *      skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2870 *      @page: The page that was allocated from skb_alloc_page
2871 *      @skb: The skb that may need pfmemalloc set
2872 */
2873static inline void skb_propagate_pfmemalloc(struct page *page,
2874                                             struct sk_buff *skb)
2875{
2876        if (page_is_pfmemalloc(page))
2877                skb->pfmemalloc = true;
2878}
2879
2880/**
2881 * skb_frag_page - retrieve the page referred to by a paged fragment
2882 * @frag: the paged fragment
2883 *
2884 * Returns the &struct page associated with @frag.
2885 */
2886static inline struct page *skb_frag_page(const skb_frag_t *frag)
2887{
2888        return frag->page.p;
2889}
2890
2891/**
2892 * __skb_frag_ref - take an addition reference on a paged fragment.
2893 * @frag: the paged fragment
2894 *
2895 * Takes an additional reference on the paged fragment @frag.
2896 */
2897static inline void __skb_frag_ref(skb_frag_t *frag)
2898{
2899        get_page(skb_frag_page(frag));
2900}
2901
2902/**
2903 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2904 * @skb: the buffer
2905 * @f: the fragment offset.
2906 *
2907 * Takes an additional reference on the @f'th paged fragment of @skb.
2908 */
2909static inline void skb_frag_ref(struct sk_buff *skb, int f)
2910{
2911        __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2912}
2913
2914/**
2915 * __skb_frag_unref - release a reference on a paged fragment.
2916 * @frag: the paged fragment
2917 *
2918 * Releases a reference on the paged fragment @frag.
2919 */
2920static inline void __skb_frag_unref(skb_frag_t *frag)
2921{
2922        put_page(skb_frag_page(frag));
2923}
2924
2925/**
2926 * skb_frag_unref - release a reference on a paged fragment of an skb.
2927 * @skb: the buffer
2928 * @f: the fragment offset
2929 *
2930 * Releases a reference on the @f'th paged fragment of @skb.
2931 */
2932static inline void skb_frag_unref(struct sk_buff *skb, int f)
2933{
2934        __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2935}
2936
2937/**
2938 * skb_frag_address - gets the address of the data contained in a paged fragment
2939 * @frag: the paged fragment buffer
2940 *
2941 * Returns the address of the data within @frag. The page must already
2942 * be mapped.
2943 */
2944static inline void *skb_frag_address(const skb_frag_t *frag)
2945{
2946        return page_address(skb_frag_page(frag)) + frag->page_offset;
2947}
2948
2949/**
2950 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2951 * @frag: the paged fragment buffer
2952 *
2953 * Returns the address of the data within @frag. Checks that the page
2954 * is mapped and returns %NULL otherwise.
2955 */
2956static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2957{
2958        void *ptr = page_address(skb_frag_page(frag));
2959        if (unlikely(!ptr))
2960                return NULL;
2961
2962        return ptr + frag->page_offset;
2963}
2964
2965/**
2966 * __skb_frag_set_page - sets the page contained in a paged fragment
2967 * @frag: the paged fragment
2968 * @page: the page to set
2969 *
2970 * Sets the fragment @frag to contain @page.
2971 */
2972static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2973{
2974        frag->page.p = page;
2975}
2976
2977/**
2978 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2979 * @skb: the buffer
2980 * @f: the fragment offset
2981 * @page: the page to set
2982 *
2983 * Sets the @f'th fragment of @skb to contain @page.
2984 */
2985static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2986                                     struct page *page)
2987{
2988        __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2989}
2990
2991bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2992
2993/**
2994 * skb_frag_dma_map - maps a paged fragment via the DMA API
2995 * @dev: the device to map the fragment to
2996 * @frag: the paged fragment to map
2997 * @offset: the offset within the fragment (starting at the
2998 *          fragment's own offset)
2999 * @size: the number of bytes to map
3000 * @dir: the direction of the mapping (``PCI_DMA_*``)
3001 *
3002 * Maps the page associated with @frag to @device.
3003 */
3004static inline dma_addr_t skb_frag_dma_map(struct device *dev,
3005                                          const skb_frag_t *frag,
3006                                          size_t offset, size_t size,
3007                                          enum dma_data_direction dir)
3008{
3009        return dma_map_page(dev, skb_frag_page(frag),
3010                            frag->page_offset + offset, size, dir);
3011}
3012
3013static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
3014                                        gfp_t gfp_mask)
3015{
3016        return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
3017}
3018
3019
3020static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
3021                                                  gfp_t gfp_mask)
3022{
3023        return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3024}
3025
3026
3027/**
3028 *      skb_clone_writable - is the header of a clone writable
3029 *      @skb: buffer to check
3030 *      @len: length up to which to write
3031 *
3032 *      Returns true if modifying the header part of the cloned buffer
3033 *      does not requires the data to be copied.
3034 */
3035static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3036{
3037        return !skb_header_cloned(skb) &&
3038               skb_headroom(skb) + len <= skb->hdr_len;
3039}
3040
3041static inline int skb_try_make_writable(struct sk_buff *skb,
3042                                        unsigned int write_len)
3043{
3044        return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3045               pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3046}
3047
3048static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3049                            int cloned)
3050{
3051        int delta = 0;
3052
3053        if (headroom > skb_headroom(skb))
3054                delta = headroom - skb_headroom(skb);
3055
3056        if (delta || cloned)
3057                return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3058                                        GFP_ATOMIC);
3059        return 0;
3060}
3061
3062/**
3063 *      skb_cow - copy header of skb when it is required
3064 *      @skb: buffer to cow
3065 *      @headroom: needed headroom
3066 *
3067 *      If the skb passed lacks sufficient headroom or its data part
3068 *      is shared, data is reallocated. If reallocation fails, an error
3069 *      is returned and original skb is not changed.
3070 *
3071 *      The result is skb with writable area skb->head...skb->tail
3072 *      and at least @headroom of space at head.
3073 */
3074static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3075{
3076        return __skb_cow(skb, headroom, skb_cloned(skb));
3077}
3078
3079/**
3080 *      skb_cow_head - skb_cow but only making the head writable
3081 *      @skb: buffer to cow
3082 *      @headroom: needed headroom
3083 *
3084 *      This function is identical to skb_cow except that we replace the
3085 *      skb_cloned check by skb_header_cloned.  It should be used when
3086 *      you only need to push on some header and do not need to modify
3087 *      the data.
3088 */
3089static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3090{
3091        return __skb_cow(skb, headroom, skb_header_cloned(skb));
3092}
3093
3094/**
3095 *      skb_padto       - pad an skbuff up to a minimal size
3096 *      @skb: buffer to pad
3097 *      @len: minimal length
3098 *
3099 *      Pads up a buffer to ensure the trailing bytes exist and are
3100 *      blanked. If the buffer already contains sufficient data it
3101 *      is untouched. Otherwise it is extended. Returns zero on
3102 *      success. The skb is freed on error.
3103 */
3104static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3105{
3106        unsigned int size = skb->len;
3107        if (likely(size >= len))
3108                return 0;
3109        return skb_pad(skb, len - size);
3110}
3111
3112/**
3113 *      __skb_put_padto - increase size and pad an skbuff up to a minimal size
3114 *      @skb: buffer to pad
3115 *      @len: minimal length
3116 *      @free_on_error: free buffer on error
3117 *
3118 *      Pads up a buffer to ensure the trailing bytes exist and are
3119 *      blanked. If the buffer already contains sufficient data it
3120 *      is untouched. Otherwise it is extended. Returns zero on
3121 *      success. The skb is freed on error if @free_on_error is true.
3122 */
3123static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
3124                                  bool free_on_error)
3125{
3126        unsigned int size = skb->len;
3127
3128        if (unlikely(size < len)) {
3129                len -= size;
3130                if (__skb_pad(skb, len, free_on_error))
3131                        return -ENOMEM;
3132                __skb_put(skb, len);
3133        }
3134        return 0;
3135}
3136
3137/**
3138 *      skb_put_padto - increase size and pad an skbuff up to a minimal size
3139 *      @skb: buffer to pad
3140 *      @len: minimal length
3141 *
3142 *      Pads up a buffer to ensure the trailing bytes exist and are
3143 *      blanked. If the buffer already contains sufficient data it
3144 *      is untouched. Otherwise it is extended. Returns zero on
3145 *      success. The skb is freed on error.
3146 */
3147static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
3148{
3149        return __skb_put_padto(skb, len, true);
3150}
3151
3152static inline int skb_add_data(struct sk_buff *skb,
3153                               struct iov_iter *from, int copy)
3154{
3155        const int off = skb->len;
3156
3157        if (skb->ip_summed == CHECKSUM_NONE) {
3158                __wsum csum = 0;
3159                if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3160                                                 &csum, from)) {
3161                        skb->csum = csum_block_add(skb->csum, csum, off);
3162                        return 0;
3163                }
3164        } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3165                return 0;
3166
3167        __skb_trim(skb, off);
3168        return -EFAULT;
3169}
3170
3171static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3172                                    const struct page *page, int off)
3173{
3174        if (skb_zcopy(skb))
3175                return false;
3176        if (i) {
3177                const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
3178
3179                return page == skb_frag_page(frag) &&
3180                       off == frag->page_offset + skb_frag_size(frag);
3181        }
3182        return false;
3183}
3184
3185static inline int __skb_linearize(struct sk_buff *skb)
3186{
3187        return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3188}
3189
3190/**
3191 *      skb_linearize - convert paged skb to linear one
3192 *      @skb: buffer to linarize
3193 *
3194 *      If there is no free memory -ENOMEM is returned, otherwise zero
3195 *      is returned and the old skb data released.
3196 */
3197static inline int skb_linearize(struct sk_buff *skb)
3198{
3199        return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3200}
3201
3202/**
3203 * skb_has_shared_frag - can any frag be overwritten
3204 * @skb: buffer to test
3205 *
3206 * Return true if the skb has at least one frag that might be modified
3207 * by an external entity (as in vmsplice()/sendfile())
3208 */
3209static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3210{
3211        return skb_is_nonlinear(skb) &&
3212               skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3213}
3214
3215/**
3216 *      skb_linearize_cow - make sure skb is linear and writable
3217 *      @skb: buffer to process
3218 *
3219 *      If there is no free memory -ENOMEM is returned, otherwise zero
3220 *      is returned and the old skb data released.
3221 */
3222static inline int skb_linearize_cow(struct sk_buff *skb)
3223{
3224        return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3225               __skb_linearize(skb) : 0;
3226}
3227
3228static __always_inline void
3229__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3230                     unsigned int off)
3231{
3232        if (skb->ip_summed == CHECKSUM_COMPLETE)
3233                skb->csum = csum_block_sub(skb->csum,
3234                                           csum_partial(start, len, 0), off);
3235        else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3236                 skb_checksum_start_offset(skb) < 0)
3237                skb->ip_summed = CHECKSUM_NONE;
3238}
3239
3240/**
3241 *      skb_postpull_rcsum - update checksum for received skb after pull
3242 *      @skb: buffer to update
3243 *      @start: start of data before pull
3244 *      @len: length of data pulled
3245 *
3246 *      After doing a pull on a received packet, you need to call this to
3247 *      update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3248 *      CHECKSUM_NONE so that it can be recomputed from scratch.
3249 */
3250static inline void skb_postpull_rcsum(struct sk_buff *skb,
3251                                      const void *start, unsigned int len)
3252{
3253        __skb_postpull_rcsum(skb, start, len, 0);
3254}
3255
3256static __always_inline void
3257__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3258                     unsigned int off)
3259{
3260        if (skb->ip_summed == CHECKSUM_COMPLETE)
3261                skb->csum = csum_block_add(skb->csum,
3262                                           csum_partial(start, len, 0), off);
3263}
3264
3265/**
3266 *      skb_postpush_rcsum - update checksum for received skb after push
3267 *      @skb: buffer to update
3268 *      @start: start of data after push
3269 *      @len: length of data pushed
3270 *
3271 *      After doing a push on a received packet, you need to call this to
3272 *      update the CHECKSUM_COMPLETE checksum.
3273 */
3274static inline void skb_postpush_rcsum(struct sk_buff *skb,
3275                                      const void *start, unsigned int len)
3276{
3277        __skb_postpush_rcsum(skb, start, len, 0);
3278}
3279
3280void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3281
3282/**
3283 *      skb_push_rcsum - push skb and update receive checksum
3284 *      @skb: buffer to update
3285 *      @len: length of data pulled
3286 *
3287 *      This function performs an skb_push on the packet and updates
3288 *      the CHECKSUM_COMPLETE checksum.  It should be used on
3289 *      receive path processing instead of skb_push unless you know
3290 *      that the checksum difference is zero (e.g., a valid IP header)
3291 *      or you are setting ip_summed to CHECKSUM_NONE.
3292 */
3293static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3294{
3295        skb_push(skb, len);
3296        skb_postpush_rcsum(skb, skb->data, len);
3297        return skb->data;
3298}
3299
3300int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3301/**
3302 *      pskb_trim_rcsum - trim received skb and update checksum
3303 *      @skb: buffer to trim
3304 *      @len: new length
3305 *
3306 *      This is exactly the same as pskb_trim except that it ensures the
3307 *      checksum of received packets are still valid after the operation.
3308 *      It can change skb pointers.
3309 */
3310
3311static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3312{
3313        if (likely(len >= skb->len))
3314                return 0;
3315        return pskb_trim_rcsum_slow(skb, len);
3316}
3317
3318static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3319{
3320        if (skb->ip_summed == CHECKSUM_COMPLETE)
3321                skb->ip_summed = CHECKSUM_NONE;
3322        __skb_trim(skb, len);
3323        return 0;
3324}
3325
3326static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3327{
3328        if (skb->ip_summed == CHECKSUM_COMPLETE)
3329                skb->ip_summed = CHECKSUM_NONE;
3330        return __skb_grow(skb, len);
3331}
3332
3333#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3334#define skb_rb_first(root) rb_to_skb(rb_first(root))
3335#define skb_rb_last(root)  rb_to_skb(rb_last(root))
3336#define skb_rb_next(skb)   rb_to_skb(rb_next(&(skb)->rbnode))
3337#define skb_rb_prev(skb)   rb_to_skb(rb_prev(&(skb)->rbnode))
3338
3339#define skb_queue_walk(queue, skb) \
3340                for (skb = (queue)->next;                                       \
3341                     skb != (struct sk_buff *)(queue);                          \
3342                     skb = skb->next)
3343
3344#define skb_queue_walk_safe(queue, skb, tmp)                                    \
3345                for (skb = (queue)->next, tmp = skb->next;                      \
3346                     skb != (struct sk_buff *)(queue);                          \
3347                     skb = tmp, tmp = skb->next)
3348
3349#define skb_queue_walk_from(queue, skb)                                         \
3350                for (; skb != (struct sk_buff *)(queue);                        \
3351                     skb = skb->next)
3352
3353#define skb_rbtree_walk(skb, root)                                              \
3354                for (skb = skb_rb_first(root); skb != NULL;                     \
3355                     skb = skb_rb_next(skb))
3356
3357#define skb_rbtree_walk_from(skb)                                               \
3358                for (; skb != NULL;                                             \
3359                     skb = skb_rb_next(skb))
3360
3361#define skb_rbtree_walk_from_safe(skb, tmp)                                     \
3362                for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL);      \
3363                     skb = tmp)
3364
3365#define skb_queue_walk_from_safe(queue, skb, tmp)                               \
3366                for (tmp = skb->next;                                           \
3367                     skb != (struct sk_buff *)(queue);                          \
3368                     skb = tmp, tmp = skb->next)
3369
3370#define skb_queue_reverse_walk(queue, skb) \
3371                for (skb = (queue)->prev;                                       \
3372                     skb != (struct sk_buff *)(queue);                          \
3373                     skb = skb->prev)
3374
3375#define skb_queue_reverse_walk_safe(queue, skb, tmp)                            \
3376                for (skb = (queue)->prev, tmp = skb->prev;                      \
3377                     skb != (struct sk_buff *)(queue);                          \
3378                     skb = tmp, tmp = skb->prev)
3379
3380#define skb_queue_reverse_walk_from_safe(queue, skb, tmp)                       \
3381                for (tmp = skb->prev;                                           \
3382                     skb != (struct sk_buff *)(queue);                          \
3383                     skb = tmp, tmp = skb->prev)
3384
3385static inline bool skb_has_frag_list(const struct sk_buff *skb)
3386{
3387        return skb_shinfo(skb)->frag_list != NULL;
3388}
3389
3390static inline void skb_frag_list_init(struct sk_buff *skb)
3391{
3392        skb_shinfo(skb)->frag_list = NULL;
3393}
3394
3395#define skb_walk_frags(skb, iter)       \
3396        for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3397
3398
3399int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3400                                const struct sk_buff *skb);
3401struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3402                                          struct sk_buff_head *queue,
3403                                          unsigned int flags,
3404                                          void (*destructor)(struct sock *sk,
3405                                                           struct sk_buff *skb),
3406                                          int *off, int *err,
3407                                          struct sk_buff **last);
3408struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3409                                        void (*destructor)(struct sock *sk,
3410                                                           struct sk_buff *skb),
3411                                        int *off, int *err,
3412                                        struct sk_buff **last);
3413struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3414                                    void (*destructor)(struct sock *sk,
3415                                                       struct sk_buff *skb),
3416                                    int *off, int *err);
3417struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3418                                  int *err);
3419__poll_t datagram_poll(struct file *file, struct socket *sock,
3420                           struct poll_table_struct *wait);
3421int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3422                           struct iov_iter *to, int size);
3423static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3424                                        struct msghdr *msg, int size)
3425{
3426        return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3427}
3428int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3429                                   struct msghdr *msg);
3430int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3431                           struct iov_iter *to, int len,
3432                           struct ahash_request *hash);
3433int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3434                                 struct iov_iter *from, int len);
3435int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3436void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3437void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3438static inline void skb_free_datagram_locked(struct sock *sk,
3439                                            struct sk_buff *skb)
3440{
3441        __skb_free_datagram_locked(sk, skb, 0);
3442}
3443int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3444int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3445int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3446__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3447                              int len, __wsum csum);
3448int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3449                    struct pipe_inode_info *pipe, unsigned int len,
3450                    unsigned int flags);
3451int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3452                         int len);
3453void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3454unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3455int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3456                 int len, int hlen);
3457void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3458int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3459void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3460bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3461bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3462struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3463struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3464int skb_ensure_writable(struct sk_buff *skb, int write_len);
3465int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3466int skb_vlan_pop(struct sk_buff *skb);
3467int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3468int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto);
3469int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto);
3470int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse);
3471int skb_mpls_dec_ttl(struct sk_buff *skb);
3472struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3473                             gfp_t gfp);
3474
3475static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3476{
3477        return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3478}
3479
3480static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3481{
3482        return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3483}
3484
3485struct skb_checksum_ops {
3486        __wsum (*update)(const void *mem, int len, __wsum wsum);
3487        __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3488};
3489
3490extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3491
3492__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3493                      __wsum csum, const struct skb_checksum_ops *ops);
3494__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3495                    __wsum csum);
3496
3497static inline void * __must_check
3498__skb_header_pointer(const struct sk_buff *skb, int offset,
3499                     int len, void *data, int hlen, void *buffer)
3500{
3501        if (hlen - offset >= len)
3502                return data + offset;
3503
3504        if (!skb ||
3505            skb_copy_bits(skb, offset, buffer, len) < 0)
3506                return NULL;
3507
3508        return buffer;
3509}
3510
3511static inline void * __must_check
3512skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3513{
3514        return __skb_header_pointer(skb, offset, len, skb->data,
3515                                    skb_headlen(skb), buffer);
3516}
3517
3518/**
3519 *      skb_needs_linearize - check if we need to linearize a given skb
3520 *                            depending on the given device features.
3521 *      @skb: socket buffer to check
3522 *      @features: net device features
3523 *
3524 *      Returns true if either:
3525 *      1. skb has frag_list and the device doesn't support FRAGLIST, or
3526 *      2. skb is fragmented and the device does not support SG.
3527 */
3528static inline bool skb_needs_linearize(struct sk_buff *skb,
3529                                       netdev_features_t features)
3530{
3531        return skb_is_nonlinear(skb) &&
3532               ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3533                (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3534}
3535
3536static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3537                                             void *to,
3538                                             const unsigned int len)
3539{
3540        memcpy(to, skb->data, len);
3541}
3542
3543static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3544                                                    const int offset, void *to,
3545                                                    const unsigned int len)
3546{
3547        memcpy(to, skb->data + offset, len);
3548}
3549
3550static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3551                                           const void *from,
3552                                           const unsigned int len)
3553{
3554        memcpy(skb->data, from, len);
3555}
3556
3557static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3558                                                  const int offset,
3559                                                  const void *from,
3560                                                  const unsigned int len)
3561{
3562        memcpy(skb->data + offset, from, len);
3563}
3564
3565void skb_init(void);
3566
3567static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3568{
3569        return skb->tstamp;
3570}
3571
3572/**
3573 *      skb_get_timestamp - get timestamp from a skb
3574 *      @skb: skb to get stamp from
3575 *      @stamp: pointer to struct __kernel_old_timeval to store stamp in
3576 *
3577 *      Timestamps are stored in the skb as offsets to a base timestamp.
3578 *      This function converts the offset back to a struct timeval and stores
3579 *      it in stamp.
3580 */
3581static inline void skb_get_timestamp(const struct sk_buff *skb,
3582                                     struct __kernel_old_timeval *stamp)
3583{
3584        *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3585}
3586
3587static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3588                                         struct __kernel_sock_timeval *stamp)
3589{
3590        struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3591
3592        stamp->tv_sec = ts.tv_sec;
3593        stamp->tv_usec = ts.tv_nsec / 1000;
3594}
3595
3596static inline void skb_get_timestampns(const struct sk_buff *skb,
3597                                       struct timespec *stamp)
3598{
3599        *stamp = ktime_to_timespec(skb->tstamp);
3600}
3601
3602static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3603                                           struct __kernel_timespec *stamp)
3604{
3605        struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3606
3607        stamp->tv_sec = ts.tv_sec;
3608        stamp->tv_nsec = ts.tv_nsec;
3609}
3610
3611static inline void __net_timestamp(struct sk_buff *skb)
3612{
3613        skb->tstamp = ktime_get_real();
3614}
3615
3616static inline ktime_t net_timedelta(ktime_t t)
3617{
3618        return ktime_sub(ktime_get_real(), t);
3619}
3620
3621static inline ktime_t net_invalid_timestamp(void)
3622{
3623        return 0;
3624}
3625
3626static inline u8 skb_metadata_len(const struct sk_buff *skb)
3627{
3628        return skb_shinfo(skb)->meta_len;
3629}
3630
3631static inline void *skb_metadata_end(const struct sk_buff *skb)
3632{
3633        return skb_mac_header(skb);
3634}
3635
3636static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3637                                          const struct sk_buff *skb_b,
3638                                          u8 meta_len)
3639{
3640        const void *a = skb_metadata_end(skb_a);
3641        const void *b = skb_metadata_end(skb_b);
3642        /* Using more efficient varaiant than plain call to memcmp(). */
3643#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3644        u64 diffs = 0;
3645
3646        switch (meta_len) {
3647#define __it(x, op) (x -= sizeof(u##op))
3648#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3649        case 32: diffs |= __it_diff(a, b, 64);
3650                 /* fall through */
3651        case 24: diffs |= __it_diff(a, b, 64);
3652                 /* fall through */
3653        case 16: diffs |= __it_diff(a, b, 64);
3654                 /* fall through */
3655        case  8: diffs |= __it_diff(a, b, 64);
3656                break;
3657        case 28: diffs |= __it_diff(a, b, 64);
3658                 /* fall through */
3659        case 20: diffs |= __it_diff(a, b, 64);
3660                 /* fall through */
3661        case 12: diffs |= __it_diff(a, b, 64);
3662                 /* fall through */
3663        case  4: diffs |= __it_diff(a, b, 32);
3664                break;
3665        }
3666        return diffs;
3667#else
3668        return memcmp(a - meta_len, b - meta_len, meta_len);
3669#endif
3670}
3671
3672static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3673                                        const struct sk_buff *skb_b)
3674{
3675        u8 len_a = skb_metadata_len(skb_a);
3676        u8 len_b = skb_metadata_len(skb_b);
3677
3678        if (!(len_a | len_b))
3679                return false;
3680
3681        return len_a != len_b ?
3682               true : __skb_metadata_differs(skb_a, skb_b, len_a);
3683}
3684
3685static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3686{
3687        skb_shinfo(skb)->meta_len = meta_len;
3688}
3689
3690static inline void skb_metadata_clear(struct sk_buff *skb)
3691{
3692        skb_metadata_set(skb, 0);
3693}
3694
3695struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3696
3697#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3698
3699void skb_clone_tx_timestamp(struct sk_buff *skb);
3700bool skb_defer_rx_timestamp(struct sk_buff *skb);
3701
3702#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3703
3704static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3705{
3706}
3707
3708static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3709{
3710        return false;
3711}
3712
3713#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3714
3715/**
3716 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3717 *
3718 * PHY drivers may accept clones of transmitted packets for
3719 * timestamping via their phy_driver.txtstamp method. These drivers
3720 * must call this function to return the skb back to the stack with a
3721 * timestamp.
3722 *
3723 * @skb: clone of the the original outgoing packet
3724 * @hwtstamps: hardware time stamps
3725 *
3726 */
3727void skb_complete_tx_timestamp(struct sk_buff *skb,
3728                               struct skb_shared_hwtstamps *hwtstamps);
3729
3730void __skb_tstamp_tx(struct sk_buff *orig_skb,
3731                     struct skb_shared_hwtstamps *hwtstamps,
3732                     struct sock *sk, int tstype);
3733
3734/**
3735 * skb_tstamp_tx - queue clone of skb with send time stamps
3736 * @orig_skb:   the original outgoing packet
3737 * @hwtstamps:  hardware time stamps, may be NULL if not available
3738 *
3739 * If the skb has a socket associated, then this function clones the
3740 * skb (thus sharing the actual data and optional structures), stores
3741 * the optional hardware time stamping information (if non NULL) or
3742 * generates a software time stamp (otherwise), then queues the clone
3743 * to the error queue of the socket.  Errors are silently ignored.
3744 */
3745void skb_tstamp_tx(struct sk_buff *orig_skb,
3746                   struct skb_shared_hwtstamps *hwtstamps);
3747
3748/**
3749 * skb_tx_timestamp() - Driver hook for transmit timestamping
3750 *
3751 * Ethernet MAC Drivers should call this function in their hard_xmit()
3752 * function immediately before giving the sk_buff to the MAC hardware.
3753 *
3754 * Specifically, one should make absolutely sure that this function is
3755 * called before TX completion of this packet can trigger.  Otherwise
3756 * the packet could potentially already be freed.
3757 *
3758 * @skb: A socket buffer.
3759 */
3760static inline void skb_tx_timestamp(struct sk_buff *skb)
3761{
3762        skb_clone_tx_timestamp(skb);
3763        if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3764                skb_tstamp_tx(skb, NULL);
3765}
3766
3767/**
3768 * skb_complete_wifi_ack - deliver skb with wifi status
3769 *
3770 * @skb: the original outgoing packet
3771 * @acked: ack status
3772 *
3773 */
3774void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3775
3776__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3777__sum16 __skb_checksum_complete(struct sk_buff *skb);
3778
3779static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3780{
3781        return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3782                skb->csum_valid ||
3783                (skb->ip_summed == CHECKSUM_PARTIAL &&
3784                 skb_checksum_start_offset(skb) >= 0));
3785}
3786
3787/**
3788 *      skb_checksum_complete - Calculate checksum of an entire packet
3789 *      @skb: packet to process
3790 *
3791 *      This function calculates the checksum over the entire packet plus
3792 *      the value of skb->csum.  The latter can be used to supply the
3793 *      checksum of a pseudo header as used by TCP/UDP.  It returns the
3794 *      checksum.
3795 *
3796 *      For protocols that contain complete checksums such as ICMP/TCP/UDP,
3797 *      this function can be used to verify that checksum on received
3798 *      packets.  In that case the function should return zero if the
3799 *      checksum is correct.  In particular, this function will return zero
3800 *      if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3801 *      hardware has already verified the correctness of the checksum.
3802 */
3803static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3804{
3805        return skb_csum_unnecessary(skb) ?
3806               0 : __skb_checksum_complete(skb);
3807}
3808
3809static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3810{
3811        if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3812                if (skb->csum_level == 0)
3813                        skb->ip_summed = CHECKSUM_NONE;
3814                else
3815                        skb->csum_level--;
3816        }
3817}
3818
3819static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3820{
3821        if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3822                if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3823                        skb->csum_level++;
3824        } else if (skb->ip_summed == CHECKSUM_NONE) {
3825                skb->ip_summed = CHECKSUM_UNNECESSARY;
3826                skb->csum_level = 0;
3827        }
3828}
3829
3830/* Check if we need to perform checksum complete validation.
3831 *
3832 * Returns true if checksum complete is needed, false otherwise
3833 * (either checksum is unnecessary or zero checksum is allowed).
3834 */
3835static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3836                                                  bool zero_okay,
3837                                                  __sum16 check)
3838{
3839        if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3840                skb->csum_valid = 1;
3841                __skb_decr_checksum_unnecessary(skb);
3842                return false;
3843        }
3844
3845        return true;
3846}
3847
3848/* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3849 * in checksum_init.
3850 */
3851#define CHECKSUM_BREAK 76
3852
3853/* Unset checksum-complete
3854 *
3855 * Unset checksum complete can be done when packet is being modified
3856 * (uncompressed for instance) and checksum-complete value is
3857 * invalidated.
3858 */
3859static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3860{
3861        if (skb->ip_summed == CHECKSUM_COMPLETE)
3862                skb->ip_summed = CHECKSUM_NONE;
3863}
3864
3865/* Validate (init) checksum based on checksum complete.
3866 *
3867 * Return values:
3868 *   0: checksum is validated or try to in skb_checksum_complete. In the latter
3869 *      case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3870 *      checksum is stored in skb->csum for use in __skb_checksum_complete
3871 *   non-zero: value of invalid checksum
3872 *
3873 */
3874static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3875                                                       bool complete,
3876                                                       __wsum psum)
3877{
3878        if (skb->ip_summed == CHECKSUM_COMPLETE) {
3879                if (!csum_fold(csum_add(psum, skb->csum))) {
3880                        skb->csum_valid = 1;
3881                        return 0;
3882                }
3883        }
3884
3885        skb->csum = psum;
3886
3887        if (complete || skb->len <= CHECKSUM_BREAK) {
3888                __sum16 csum;
3889
3890                csum = __skb_checksum_complete(skb);
3891                skb->csum_valid = !csum;
3892                return csum;
3893        }
3894
3895        return 0;
3896}
3897
3898static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3899{
3900        return 0;
3901}
3902
3903/* Perform checksum validate (init). Note that this is a macro since we only
3904 * want to calculate the pseudo header which is an input function if necessary.
3905 * First we try to validate without any computation (checksum unnecessary) and
3906 * then calculate based on checksum complete calling the function to compute
3907 * pseudo header.
3908 *
3909 * Return values:
3910 *   0: checksum is validated or try to in skb_checksum_complete
3911 *   non-zero: value of invalid checksum
3912 */
3913#define __skb_checksum_validate(skb, proto, complete,                   \
3914                                zero_okay, check, compute_pseudo)       \
3915({                                                                      \
3916        __sum16 __ret = 0;                                              \
3917        skb->csum_valid = 0;                                            \
3918        if (__skb_checksum_validate_needed(skb, zero_okay, check))      \
3919                __ret = __skb_checksum_validate_complete(skb,           \
3920                                complete, compute_pseudo(skb, proto));  \
3921        __ret;                                                          \
3922})
3923
3924#define skb_checksum_init(skb, proto, compute_pseudo)                   \
3925        __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3926
3927#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3928        __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3929
3930#define skb_checksum_validate(skb, proto, compute_pseudo)               \
3931        __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3932
3933#define skb_checksum_validate_zero_check(skb, proto, check,             \
3934                                         compute_pseudo)                \
3935        __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3936
3937#define skb_checksum_simple_validate(skb)                               \
3938        __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3939
3940static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3941{
3942        return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3943}
3944
3945static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo)
3946{
3947        skb->csum = ~pseudo;
3948        skb->ip_summed = CHECKSUM_COMPLETE;
3949}
3950
3951#define skb_checksum_try_convert(skb, proto, compute_pseudo)    \
3952do {                                                                    \
3953        if (__skb_checksum_convert_check(skb))                          \
3954                __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \
3955} while (0)
3956
3957static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3958                                              u16 start, u16 offset)
3959{
3960        skb->ip_summed = CHECKSUM_PARTIAL;
3961        skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3962        skb->csum_offset = offset - start;
3963}
3964
3965/* Update skbuf and packet to reflect the remote checksum offload operation.
3966 * When called, ptr indicates the starting point for skb->csum when
3967 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3968 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3969 */
3970static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3971                                       int start, int offset, bool nopartial)
3972{
3973        __wsum delta;
3974
3975        if (!nopartial) {
3976                skb_remcsum_adjust_partial(skb, ptr, start, offset);
3977                return;
3978        }
3979
3980         if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3981                __skb_checksum_complete(skb);
3982                skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3983        }
3984
3985        delta = remcsum_adjust(ptr, skb->csum, start, offset);
3986
3987        /* Adjust skb->csum since we changed the packet */
3988        skb->csum = csum_add(skb->csum, delta);
3989}
3990
3991static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3992{
3993#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3994        return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3995#else
3996        return NULL;
3997#endif
3998}
3999
4000#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4001void nf_conntrack_destroy(struct nf_conntrack *nfct);
4002static inline void nf_conntrack_put(struct nf_conntrack *nfct)
4003{
4004        if (nfct && atomic_dec_and_test(&nfct->use))
4005                nf_conntrack_destroy(nfct);
4006}
4007static inline void nf_conntrack_get(struct nf_conntrack *nfct)
4008{
4009        if (nfct)
4010                atomic_inc(&nfct->use);
4011}
4012#endif
4013
4014#ifdef CONFIG_SKB_EXTENSIONS
4015enum skb_ext_id {
4016#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4017        SKB_EXT_BRIDGE_NF,
4018#endif
4019#ifdef CONFIG_XFRM
4020        SKB_EXT_SEC_PATH,
4021#endif
4022        SKB_EXT_NUM, /* must be last */
4023};
4024
4025/**
4026 *      struct skb_ext - sk_buff extensions
4027 *      @refcnt: 1 on allocation, deallocated on 0
4028 *      @offset: offset to add to @data to obtain extension address
4029 *      @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4030 *      @data: start of extension data, variable sized
4031 *
4032 *      Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4033 *      to use 'u8' types while allowing up to 2kb worth of extension data.
4034 */
4035struct skb_ext {
4036        refcount_t refcnt;
4037        u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4038        u8 chunks;              /* same */
4039        char data[0] __aligned(8);
4040};
4041
4042void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4043void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4044void __skb_ext_put(struct skb_ext *ext);
4045
4046static inline void skb_ext_put(struct sk_buff *skb)
4047{
4048        if (skb->active_extensions)
4049                __skb_ext_put(skb->extensions);
4050}
4051
4052static inline void __skb_ext_copy(struct sk_buff *dst,
4053                                  const struct sk_buff *src)
4054{
4055        dst->active_extensions = src->active_extensions;
4056
4057        if (src->active_extensions) {
4058                struct skb_ext *ext = src->extensions;
4059
4060                refcount_inc(&ext->refcnt);
4061                dst->extensions = ext;
4062        }
4063}
4064
4065static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4066{
4067        skb_ext_put(dst);
4068        __skb_ext_copy(dst, src);
4069}
4070
4071static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4072{
4073        return !!ext->offset[i];
4074}
4075
4076static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4077{
4078        return skb->active_extensions & (1 << id);
4079}
4080
4081static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4082{
4083        if (skb_ext_exist(skb, id))
4084                __skb_ext_del(skb, id);
4085}
4086
4087static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4088{
4089        if (skb_ext_exist(skb, id)) {
4090                struct skb_ext *ext = skb->extensions;
4091
4092                return (void *)ext + (ext->offset[id] << 3);
4093        }
4094
4095        return NULL;
4096}
4097#else
4098static inline void skb_ext_put(struct sk_buff *skb) {}
4099static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4100static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4101static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4102#endif /* CONFIG_SKB_EXTENSIONS */
4103
4104static inline void nf_reset(struct sk_buff *skb)
4105{
4106#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4107        nf_conntrack_put(skb_nfct(skb));
4108        skb->_nfct = 0;
4109#endif
4110#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4111        skb_ext_del(skb, SKB_EXT_BRIDGE_NF);
4112#endif
4113}
4114
4115static inline void nf_reset_trace(struct sk_buff *skb)
4116{
4117#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4118        skb->nf_trace = 0;
4119#endif
4120}
4121
4122static inline void ipvs_reset(struct sk_buff *skb)
4123{
4124#if IS_ENABLED(CONFIG_IP_VS)
4125        skb->ipvs_property = 0;
4126#endif
4127}
4128
4129/* Note: This doesn't put any conntrack info in dst. */
4130static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4131                             bool copy)
4132{
4133#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4134        dst->_nfct = src->_nfct;
4135        nf_conntrack_get(skb_nfct(src));
4136#endif
4137#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4138        if (copy)
4139                dst->nf_trace = src->nf_trace;
4140#endif
4141}
4142
4143static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4144{
4145#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4146        nf_conntrack_put(skb_nfct(dst));
4147#endif
4148        __nf_copy(dst, src, true);
4149}
4150
4151#ifdef CONFIG_NETWORK_SECMARK
4152static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4153{
4154        to->secmark = from->secmark;
4155}
4156
4157static inline void skb_init_secmark(struct sk_buff *skb)
4158{
4159        skb->secmark = 0;
4160}
4161#else
4162static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4163{ }
4164
4165static inline void skb_init_secmark(struct sk_buff *skb)
4166{ }
4167#endif
4168
4169static inline int secpath_exists(const struct sk_buff *skb)
4170{
4171#ifdef CONFIG_XFRM
4172        return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4173#else
4174        return 0;
4175#endif
4176}
4177
4178static inline bool skb_irq_freeable(const struct sk_buff *skb)
4179{
4180        return !skb->destructor &&
4181                !secpath_exists(skb) &&
4182                !skb_nfct(skb) &&
4183                !skb->_skb_refdst &&
4184                !skb_has_frag_list(skb);
4185}
4186
4187static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4188{
4189        skb->queue_mapping = queue_mapping;
4190}
4191
4192static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4193{
4194        return skb->queue_mapping;
4195}
4196
4197static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4198{
4199        to->queue_mapping = from->queue_mapping;
4200}
4201
4202static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4203{
4204        skb->queue_mapping = rx_queue + 1;
4205}
4206
4207static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4208{
4209        return skb->queue_mapping - 1;
4210}
4211
4212static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4213{
4214        return skb->queue_mapping != 0;
4215}
4216
4217static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4218{
4219        skb->dst_pending_confirm = val;
4220}
4221
4222static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4223{
4224        return skb->dst_pending_confirm != 0;
4225}
4226
4227static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4228{
4229#ifdef CONFIG_XFRM
4230        return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4231#else
4232        return NULL;
4233#endif
4234}
4235
4236/* Keeps track of mac header offset relative to skb->head.
4237 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4238 * For non-tunnel skb it points to skb_mac_header() and for
4239 * tunnel skb it points to outer mac header.
4240 * Keeps track of level of encapsulation of network headers.
4241 */
4242struct skb_gso_cb {
4243        union {
4244                int     mac_offset;
4245                int     data_offset;
4246        };
4247        int     encap_level;
4248        __wsum  csum;
4249        __u16   csum_start;
4250};
4251#define SKB_SGO_CB_OFFSET       32
4252#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
4253
4254static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4255{
4256        return (skb_mac_header(inner_skb) - inner_skb->head) -
4257                SKB_GSO_CB(inner_skb)->mac_offset;
4258}
4259
4260static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4261{
4262        int new_headroom, headroom;
4263        int ret;
4264
4265        headroom = skb_headroom(skb);
4266        ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4267        if (ret)
4268                return ret;
4269
4270        new_headroom = skb_headroom(skb);
4271        SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4272        return 0;
4273}
4274
4275static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4276{
4277        /* Do not update partial checksums if remote checksum is enabled. */
4278        if (skb->remcsum_offload)
4279                return;
4280
4281        SKB_GSO_CB(skb)->csum = res;
4282        SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4283}
4284
4285/* Compute the checksum for a gso segment. First compute the checksum value
4286 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4287 * then add in skb->csum (checksum from csum_start to end of packet).
4288 * skb->csum and csum_start are then updated to reflect the checksum of the
4289 * resultant packet starting from the transport header-- the resultant checksum
4290 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4291 * header.
4292 */
4293static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4294{
4295        unsigned char *csum_start = skb_transport_header(skb);
4296        int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4297        __wsum partial = SKB_GSO_CB(skb)->csum;
4298
4299        SKB_GSO_CB(skb)->csum = res;
4300        SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4301
4302        return csum_fold(csum_partial(csum_start, plen, partial));
4303}
4304
4305static inline bool skb_is_gso(const struct sk_buff *skb)
4306{
4307        return skb_shinfo(skb)->gso_size;
4308}
4309
4310/* Note: Should be called only if skb_is_gso(skb) is true */
4311static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4312{
4313        return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4314}
4315
4316/* Note: Should be called only if skb_is_gso(skb) is true */
4317static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4318{
4319        return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4320}
4321
4322/* Note: Should be called only if skb_is_gso(skb) is true */
4323static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4324{
4325        return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4326}
4327
4328static inline void skb_gso_reset(struct sk_buff *skb)
4329{
4330        skb_shinfo(skb)->gso_size = 0;
4331        skb_shinfo(skb)->gso_segs = 0;
4332        skb_shinfo(skb)->gso_type = 0;
4333}
4334
4335static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4336                                         u16 increment)
4337{
4338        if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4339                return;
4340        shinfo->gso_size += increment;
4341}
4342
4343static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4344                                         u16 decrement)
4345{
4346        if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4347                return;
4348        shinfo->gso_size -= decrement;
4349}
4350
4351void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4352
4353static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4354{
4355        /* LRO sets gso_size but not gso_type, whereas if GSO is really
4356         * wanted then gso_type will be set. */
4357        const struct skb_shared_info *shinfo = skb_shinfo(skb);
4358
4359        if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4360            unlikely(shinfo->gso_type == 0)) {
4361                __skb_warn_lro_forwarding(skb);
4362                return true;
4363        }
4364        return false;
4365}
4366
4367static inline void skb_forward_csum(struct sk_buff *skb)
4368{
4369        /* Unfortunately we don't support this one.  Any brave souls? */
4370        if (skb->ip_summed == CHECKSUM_COMPLETE)
4371                skb->ip_summed = CHECKSUM_NONE;
4372}
4373
4374/**
4375 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4376 * @skb: skb to check
4377 *
4378 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4379 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4380 * use this helper, to document places where we make this assertion.
4381 */
4382static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4383{
4384#ifdef DEBUG
4385        BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4386#endif
4387}
4388
4389bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4390
4391int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4392struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4393                                     unsigned int transport_len,
4394                                     __sum16(*skb_chkf)(struct sk_buff *skb));
4395
4396/**
4397 * skb_head_is_locked - Determine if the skb->head is locked down
4398 * @skb: skb to check
4399 *
4400 * The head on skbs build around a head frag can be removed if they are
4401 * not cloned.  This function returns true if the skb head is locked down
4402 * due to either being allocated via kmalloc, or by being a clone with
4403 * multiple references to the head.
4404 */
4405static inline bool skb_head_is_locked(const struct sk_buff *skb)
4406{
4407        return !skb->head_frag || skb_cloned(skb);
4408}
4409
4410/* Local Checksum Offload.
4411 * Compute outer checksum based on the assumption that the
4412 * inner checksum will be offloaded later.
4413 * See Documentation/networking/checksum-offloads.rst for
4414 * explanation of how this works.
4415 * Fill in outer checksum adjustment (e.g. with sum of outer
4416 * pseudo-header) before calling.
4417 * Also ensure that inner checksum is in linear data area.
4418 */
4419static inline __wsum lco_csum(struct sk_buff *skb)
4420{
4421        unsigned char *csum_start = skb_checksum_start(skb);
4422        unsigned char *l4_hdr = skb_transport_header(skb);
4423        __wsum partial;
4424
4425        /* Start with complement of inner checksum adjustment */
4426        partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4427                                                    skb->csum_offset));
4428
4429        /* Add in checksum of our headers (incl. outer checksum
4430         * adjustment filled in by caller) and return result.
4431         */
4432        return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4433}
4434
4435#endif  /* __KERNEL__ */
4436#endif  /* _LINUX_SKBUFF_H */
4437