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_tx_error(struct sk_buff *skb);
1028void consume_skb(struct sk_buff *skb);
1029void __consume_stateless_skb(struct sk_buff *skb);
1030void  __kfree_skb(struct sk_buff *skb);
1031extern struct kmem_cache *skbuff_head_cache;
1032
1033void kfree_skb_partial(struct sk_buff *skb, bool head_stolen);
1034bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
1035                      bool *fragstolen, int *delta_truesize);
1036
1037struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags,
1038                            int node);
1039struct sk_buff *__build_skb(void *data, unsigned int frag_size);
1040struct sk_buff *build_skb(void *data, unsigned int frag_size);
1041struct sk_buff *build_skb_around(struct sk_buff *skb,
1042                                 void *data, unsigned int frag_size);
1043
1044/**
1045 * alloc_skb - allocate a network buffer
1046 * @size: size to allocate
1047 * @priority: allocation mask
1048 *
1049 * This function is a convenient wrapper around __alloc_skb().
1050 */
1051static inline struct sk_buff *alloc_skb(unsigned int size,
1052                                        gfp_t priority)
1053{
1054        return __alloc_skb(size, priority, 0, NUMA_NO_NODE);
1055}
1056
1057struct sk_buff *alloc_skb_with_frags(unsigned long header_len,
1058                                     unsigned long data_len,
1059                                     int max_page_order,
1060                                     int *errcode,
1061                                     gfp_t gfp_mask);
1062
1063/* Layout of fast clones : [skb1][skb2][fclone_ref] */
1064struct sk_buff_fclones {
1065        struct sk_buff  skb1;
1066
1067        struct sk_buff  skb2;
1068
1069        refcount_t      fclone_ref;
1070};
1071
1072/**
1073 *      skb_fclone_busy - check if fclone is busy
1074 *      @sk: socket
1075 *      @skb: buffer
1076 *
1077 * Returns true if skb is a fast clone, and its clone is not freed.
1078 * Some drivers call skb_orphan() in their ndo_start_xmit(),
1079 * so we also check that this didnt happen.
1080 */
1081static inline bool skb_fclone_busy(const struct sock *sk,
1082                                   const struct sk_buff *skb)
1083{
1084        const struct sk_buff_fclones *fclones;
1085
1086        fclones = container_of(skb, struct sk_buff_fclones, skb1);
1087
1088        return skb->fclone == SKB_FCLONE_ORIG &&
1089               refcount_read(&fclones->fclone_ref) > 1 &&
1090               fclones->skb2.sk == sk;
1091}
1092
1093/**
1094 * alloc_skb_fclone - allocate a network buffer from fclone cache
1095 * @size: size to allocate
1096 * @priority: allocation mask
1097 *
1098 * This function is a convenient wrapper around __alloc_skb().
1099 */
1100static inline struct sk_buff *alloc_skb_fclone(unsigned int size,
1101                                               gfp_t priority)
1102{
1103        return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE);
1104}
1105
1106struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src);
1107void skb_headers_offset_update(struct sk_buff *skb, int off);
1108int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask);
1109struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority);
1110void skb_copy_header(struct sk_buff *new, const struct sk_buff *old);
1111struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority);
1112struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom,
1113                                   gfp_t gfp_mask, bool fclone);
1114static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom,
1115                                          gfp_t gfp_mask)
1116{
1117        return __pskb_copy_fclone(skb, headroom, gfp_mask, false);
1118}
1119
1120int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask);
1121struct sk_buff *skb_realloc_headroom(struct sk_buff *skb,
1122                                     unsigned int headroom);
1123struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom,
1124                                int newtailroom, gfp_t priority);
1125int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg,
1126                                     int offset, int len);
1127int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg,
1128                              int offset, int len);
1129int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer);
1130int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error);
1131
1132/**
1133 *      skb_pad                 -       zero pad the tail of an skb
1134 *      @skb: buffer to pad
1135 *      @pad: space to pad
1136 *
1137 *      Ensure that a buffer is followed by a padding area that is zero
1138 *      filled. Used by network drivers which may DMA or transfer data
1139 *      beyond the buffer end onto the wire.
1140 *
1141 *      May return error in out of memory cases. The skb is freed on error.
1142 */
1143static inline int skb_pad(struct sk_buff *skb, int pad)
1144{
1145        return __skb_pad(skb, pad, true);
1146}
1147#define dev_kfree_skb(a)        consume_skb(a)
1148
1149int skb_append_pagefrags(struct sk_buff *skb, struct page *page,
1150                         int offset, size_t size);
1151
1152struct skb_seq_state {
1153        __u32           lower_offset;
1154        __u32           upper_offset;
1155        __u32           frag_idx;
1156        __u32           stepped_offset;
1157        struct sk_buff  *root_skb;
1158        struct sk_buff  *cur_skb;
1159        __u8            *frag_data;
1160};
1161
1162void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
1163                          unsigned int to, struct skb_seq_state *st);
1164unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
1165                          struct skb_seq_state *st);
1166void skb_abort_seq_read(struct skb_seq_state *st);
1167
1168unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
1169                           unsigned int to, struct ts_config *config);
1170
1171/*
1172 * Packet hash types specify the type of hash in skb_set_hash.
1173 *
1174 * Hash types refer to the protocol layer addresses which are used to
1175 * construct a packet's hash. The hashes are used to differentiate or identify
1176 * flows of the protocol layer for the hash type. Hash types are either
1177 * layer-2 (L2), layer-3 (L3), or layer-4 (L4).
1178 *
1179 * Properties of hashes:
1180 *
1181 * 1) Two packets in different flows have different hash values
1182 * 2) Two packets in the same flow should have the same hash value
1183 *
1184 * A hash at a higher layer is considered to be more specific. A driver should
1185 * set the most specific hash possible.
1186 *
1187 * A driver cannot indicate a more specific hash than the layer at which a hash
1188 * was computed. For instance an L3 hash cannot be set as an L4 hash.
1189 *
1190 * A driver may indicate a hash level which is less specific than the
1191 * actual layer the hash was computed on. For instance, a hash computed
1192 * at L4 may be considered an L3 hash. This should only be done if the
1193 * driver can't unambiguously determine that the HW computed the hash at
1194 * the higher layer. Note that the "should" in the second property above
1195 * permits this.
1196 */
1197enum pkt_hash_types {
1198        PKT_HASH_TYPE_NONE,     /* Undefined type */
1199        PKT_HASH_TYPE_L2,       /* Input: src_MAC, dest_MAC */
1200        PKT_HASH_TYPE_L3,       /* Input: src_IP, dst_IP */
1201        PKT_HASH_TYPE_L4,       /* Input: src_IP, dst_IP, src_port, dst_port */
1202};
1203
1204static inline void skb_clear_hash(struct sk_buff *skb)
1205{
1206        skb->hash = 0;
1207        skb->sw_hash = 0;
1208        skb->l4_hash = 0;
1209}
1210
1211static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb)
1212{
1213        if (!skb->l4_hash)
1214                skb_clear_hash(skb);
1215}
1216
1217static inline void
1218__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4)
1219{
1220        skb->l4_hash = is_l4;
1221        skb->sw_hash = is_sw;
1222        skb->hash = hash;
1223}
1224
1225static inline void
1226skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type)
1227{
1228        /* Used by drivers to set hash from HW */
1229        __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4);
1230}
1231
1232static inline void
1233__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4)
1234{
1235        __skb_set_hash(skb, hash, true, is_l4);
1236}
1237
1238void __skb_get_hash(struct sk_buff *skb);
1239u32 __skb_get_hash_symmetric(const struct sk_buff *skb);
1240u32 skb_get_poff(const struct sk_buff *skb);
1241u32 __skb_get_poff(const struct sk_buff *skb, void *data,
1242                   const struct flow_keys_basic *keys, int hlen);
1243__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto,
1244                            void *data, int hlen_proto);
1245
1246static inline __be32 skb_flow_get_ports(const struct sk_buff *skb,
1247                                        int thoff, u8 ip_proto)
1248{
1249        return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0);
1250}
1251
1252void skb_flow_dissector_init(struct flow_dissector *flow_dissector,
1253                             const struct flow_dissector_key *key,
1254                             unsigned int key_count);
1255
1256#ifdef CONFIG_NET
1257int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1258                                  union bpf_attr __user *uattr);
1259int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1260                                       struct bpf_prog *prog);
1261
1262int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr);
1263#else
1264static inline int skb_flow_dissector_prog_query(const union bpf_attr *attr,
1265                                                union bpf_attr __user *uattr)
1266{
1267        return -EOPNOTSUPP;
1268}
1269
1270static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr,
1271                                                     struct bpf_prog *prog)
1272{
1273        return -EOPNOTSUPP;
1274}
1275
1276static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr)
1277{
1278        return -EOPNOTSUPP;
1279}
1280#endif
1281
1282struct bpf_flow_dissector;
1283bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx,
1284                      __be16 proto, int nhoff, int hlen);
1285
1286bool __skb_flow_dissect(const struct net *net,
1287                        const struct sk_buff *skb,
1288                        struct flow_dissector *flow_dissector,
1289                        void *target_container,
1290                        void *data, __be16 proto, int nhoff, int hlen,
1291                        unsigned int flags);
1292
1293static inline bool skb_flow_dissect(const struct sk_buff *skb,
1294                                    struct flow_dissector *flow_dissector,
1295                                    void *target_container, unsigned int flags)
1296{
1297        return __skb_flow_dissect(NULL, skb, flow_dissector,
1298                                  target_container, NULL, 0, 0, 0, flags);
1299}
1300
1301static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb,
1302                                              struct flow_keys *flow,
1303                                              unsigned int flags)
1304{
1305        memset(flow, 0, sizeof(*flow));
1306        return __skb_flow_dissect(NULL, skb, &flow_keys_dissector,
1307                                  flow, NULL, 0, 0, 0, flags);
1308}
1309
1310static inline bool
1311skb_flow_dissect_flow_keys_basic(const struct net *net,
1312                                 const struct sk_buff *skb,
1313                                 struct flow_keys_basic *flow, void *data,
1314                                 __be16 proto, int nhoff, int hlen,
1315                                 unsigned int flags)
1316{
1317        memset(flow, 0, sizeof(*flow));
1318        return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow,
1319                                  data, proto, nhoff, hlen, flags);
1320}
1321
1322void
1323skb_flow_dissect_tunnel_info(const struct sk_buff *skb,
1324                             struct flow_dissector *flow_dissector,
1325                             void *target_container);
1326
1327static inline __u32 skb_get_hash(struct sk_buff *skb)
1328{
1329        if (!skb->l4_hash && !skb->sw_hash)
1330                __skb_get_hash(skb);
1331
1332        return skb->hash;
1333}
1334
1335static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6)
1336{
1337        if (!skb->l4_hash && !skb->sw_hash) {
1338                struct flow_keys keys;
1339                __u32 hash = __get_hash_from_flowi6(fl6, &keys);
1340
1341                __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys));
1342        }
1343
1344        return skb->hash;
1345}
1346
1347__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb);
1348
1349static inline __u32 skb_get_hash_raw(const struct sk_buff *skb)
1350{
1351        return skb->hash;
1352}
1353
1354static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from)
1355{
1356        to->hash = from->hash;
1357        to->sw_hash = from->sw_hash;
1358        to->l4_hash = from->l4_hash;
1359};
1360
1361#ifdef NET_SKBUFF_DATA_USES_OFFSET
1362static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1363{
1364        return skb->head + skb->end;
1365}
1366
1367static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1368{
1369        return skb->end;
1370}
1371#else
1372static inline unsigned char *skb_end_pointer(const struct sk_buff *skb)
1373{
1374        return skb->end;
1375}
1376
1377static inline unsigned int skb_end_offset(const struct sk_buff *skb)
1378{
1379        return skb->end - skb->head;
1380}
1381#endif
1382
1383/* Internal */
1384#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB)))
1385
1386static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb)
1387{
1388        return &skb_shinfo(skb)->hwtstamps;
1389}
1390
1391static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb)
1392{
1393        bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY;
1394
1395        return is_zcopy ? skb_uarg(skb) : NULL;
1396}
1397
1398static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg,
1399                                 bool *have_ref)
1400{
1401        if (skb && uarg && !skb_zcopy(skb)) {
1402                if (unlikely(have_ref && *have_ref))
1403                        *have_ref = false;
1404                else
1405                        sock_zerocopy_get(uarg);
1406                skb_shinfo(skb)->destructor_arg = uarg;
1407                skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1408        }
1409}
1410
1411static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val)
1412{
1413        skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL);
1414        skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG;
1415}
1416
1417static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb)
1418{
1419        return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL;
1420}
1421
1422static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb)
1423{
1424        return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL);
1425}
1426
1427/* Release a reference on a zerocopy structure */
1428static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy)
1429{
1430        struct ubuf_info *uarg = skb_zcopy(skb);
1431
1432        if (uarg) {
1433                if (skb_zcopy_is_nouarg(skb)) {
1434                        /* no notification callback */
1435                } else if (uarg->callback == sock_zerocopy_callback) {
1436                        uarg->zerocopy = uarg->zerocopy && zerocopy;
1437                        sock_zerocopy_put(uarg);
1438                } else {
1439                        uarg->callback(uarg, zerocopy);
1440                }
1441
1442                skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1443        }
1444}
1445
1446/* Abort a zerocopy operation and revert zckey on error in send syscall */
1447static inline void skb_zcopy_abort(struct sk_buff *skb)
1448{
1449        struct ubuf_info *uarg = skb_zcopy(skb);
1450
1451        if (uarg) {
1452                sock_zerocopy_put_abort(uarg, false);
1453                skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG;
1454        }
1455}
1456
1457static inline void skb_mark_not_on_list(struct sk_buff *skb)
1458{
1459        skb->next = NULL;
1460}
1461
1462static inline void skb_list_del_init(struct sk_buff *skb)
1463{
1464        __list_del_entry(&skb->list);
1465        skb_mark_not_on_list(skb);
1466}
1467
1468/**
1469 *      skb_queue_empty - check if a queue is empty
1470 *      @list: queue head
1471 *
1472 *      Returns true if the queue is empty, false otherwise.
1473 */
1474static inline int skb_queue_empty(const struct sk_buff_head *list)
1475{
1476        return list->next == (const struct sk_buff *) list;
1477}
1478
1479/**
1480 *      skb_queue_is_last - check if skb is the last entry in the queue
1481 *      @list: queue head
1482 *      @skb: buffer
1483 *
1484 *      Returns true if @skb is the last buffer on the list.
1485 */
1486static inline bool skb_queue_is_last(const struct sk_buff_head *list,
1487                                     const struct sk_buff *skb)
1488{
1489        return skb->next == (const struct sk_buff *) list;
1490}
1491
1492/**
1493 *      skb_queue_is_first - check if skb is the first entry in the queue
1494 *      @list: queue head
1495 *      @skb: buffer
1496 *
1497 *      Returns true if @skb is the first buffer on the list.
1498 */
1499static inline bool skb_queue_is_first(const struct sk_buff_head *list,
1500                                      const struct sk_buff *skb)
1501{
1502        return skb->prev == (const struct sk_buff *) list;
1503}
1504
1505/**
1506 *      skb_queue_next - return the next packet in the queue
1507 *      @list: queue head
1508 *      @skb: current buffer
1509 *
1510 *      Return the next packet in @list after @skb.  It is only valid to
1511 *      call this if skb_queue_is_last() evaluates to false.
1512 */
1513static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list,
1514                                             const struct sk_buff *skb)
1515{
1516        /* This BUG_ON may seem severe, but if we just return then we
1517         * are going to dereference garbage.
1518         */
1519        BUG_ON(skb_queue_is_last(list, skb));
1520        return skb->next;
1521}
1522
1523/**
1524 *      skb_queue_prev - return the prev packet in the queue
1525 *      @list: queue head
1526 *      @skb: current buffer
1527 *
1528 *      Return the prev packet in @list before @skb.  It is only valid to
1529 *      call this if skb_queue_is_first() evaluates to false.
1530 */
1531static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list,
1532                                             const struct sk_buff *skb)
1533{
1534        /* This BUG_ON may seem severe, but if we just return then we
1535         * are going to dereference garbage.
1536         */
1537        BUG_ON(skb_queue_is_first(list, skb));
1538        return skb->prev;
1539}
1540
1541/**
1542 *      skb_get - reference buffer
1543 *      @skb: buffer to reference
1544 *
1545 *      Makes another reference to a socket buffer and returns a pointer
1546 *      to the buffer.
1547 */
1548static inline struct sk_buff *skb_get(struct sk_buff *skb)
1549{
1550        refcount_inc(&skb->users);
1551        return skb;
1552}
1553
1554/*
1555 * If users == 1, we are the only owner and can avoid redundant atomic changes.
1556 */
1557
1558/**
1559 *      skb_cloned - is the buffer a clone
1560 *      @skb: buffer to check
1561 *
1562 *      Returns true if the buffer was generated with skb_clone() and is
1563 *      one of multiple shared copies of the buffer. Cloned buffers are
1564 *      shared data so must not be written to under normal circumstances.
1565 */
1566static inline int skb_cloned(const struct sk_buff *skb)
1567{
1568        return skb->cloned &&
1569               (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1;
1570}
1571
1572static inline int skb_unclone(struct sk_buff *skb, gfp_t pri)
1573{
1574        might_sleep_if(gfpflags_allow_blocking(pri));
1575
1576        if (skb_cloned(skb))
1577                return pskb_expand_head(skb, 0, 0, pri);
1578
1579        return 0;
1580}
1581
1582/**
1583 *      skb_header_cloned - is the header a clone
1584 *      @skb: buffer to check
1585 *
1586 *      Returns true if modifying the header part of the buffer requires
1587 *      the data to be copied.
1588 */
1589static inline int skb_header_cloned(const struct sk_buff *skb)
1590{
1591        int dataref;
1592
1593        if (!skb->cloned)
1594                return 0;
1595
1596        dataref = atomic_read(&skb_shinfo(skb)->dataref);
1597        dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT);
1598        return dataref != 1;
1599}
1600
1601static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri)
1602{
1603        might_sleep_if(gfpflags_allow_blocking(pri));
1604
1605        if (skb_header_cloned(skb))
1606                return pskb_expand_head(skb, 0, 0, pri);
1607
1608        return 0;
1609}
1610
1611/**
1612 *      __skb_header_release - release reference to header
1613 *      @skb: buffer to operate on
1614 */
1615static inline void __skb_header_release(struct sk_buff *skb)
1616{
1617        skb->nohdr = 1;
1618        atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT));
1619}
1620
1621
1622/**
1623 *      skb_shared - is the buffer shared
1624 *      @skb: buffer to check
1625 *
1626 *      Returns true if more than one person has a reference to this
1627 *      buffer.
1628 */
1629static inline int skb_shared(const struct sk_buff *skb)
1630{
1631        return refcount_read(&skb->users) != 1;
1632}
1633
1634/**
1635 *      skb_share_check - check if buffer is shared and if so clone it
1636 *      @skb: buffer to check
1637 *      @pri: priority for memory allocation
1638 *
1639 *      If the buffer is shared the buffer is cloned and the old copy
1640 *      drops a reference. A new clone with a single reference is returned.
1641 *      If the buffer is not shared the original buffer is returned. When
1642 *      being called from interrupt status or with spinlocks held pri must
1643 *      be GFP_ATOMIC.
1644 *
1645 *      NULL is returned on a memory allocation failure.
1646 */
1647static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri)
1648{
1649        might_sleep_if(gfpflags_allow_blocking(pri));
1650        if (skb_shared(skb)) {
1651                struct sk_buff *nskb = skb_clone(skb, pri);
1652
1653                if (likely(nskb))
1654                        consume_skb(skb);
1655                else
1656                        kfree_skb(skb);
1657                skb = nskb;
1658        }
1659        return skb;
1660}
1661
1662/*
1663 *      Copy shared buffers into a new sk_buff. We effectively do COW on
1664 *      packets to handle cases where we have a local reader and forward
1665 *      and a couple of other messy ones. The normal one is tcpdumping
1666 *      a packet thats being forwarded.
1667 */
1668
1669/**
1670 *      skb_unshare - make a copy of a shared buffer
1671 *      @skb: buffer to check
1672 *      @pri: priority for memory allocation
1673 *
1674 *      If the socket buffer is a clone then this function creates a new
1675 *      copy of the data, drops a reference count on the old copy and returns
1676 *      the new copy with the reference count at 1. If the buffer is not a clone
1677 *      the original buffer is returned. When called with a spinlock held or
1678 *      from interrupt state @pri must be %GFP_ATOMIC
1679 *
1680 *      %NULL is returned on a memory allocation failure.
1681 */
1682static inline struct sk_buff *skb_unshare(struct sk_buff *skb,
1683                                          gfp_t pri)
1684{
1685        might_sleep_if(gfpflags_allow_blocking(pri));
1686        if (skb_cloned(skb)) {
1687                struct sk_buff *nskb = skb_copy(skb, pri);
1688
1689                /* Free our shared copy */
1690                if (likely(nskb))
1691                        consume_skb(skb);
1692                else
1693                        kfree_skb(skb);
1694                skb = nskb;
1695        }
1696        return skb;
1697}
1698
1699/**
1700 *      skb_peek - peek at the head of an &sk_buff_head
1701 *      @list_: list to peek at
1702 *
1703 *      Peek an &sk_buff. Unlike most other operations you _MUST_
1704 *      be careful with this one. A peek leaves the buffer on the
1705 *      list and someone else may run off with it. You must hold
1706 *      the appropriate locks or have a private queue to do this.
1707 *
1708 *      Returns %NULL for an empty list or a pointer to the head element.
1709 *      The reference count is not incremented and the reference is therefore
1710 *      volatile. Use with caution.
1711 */
1712static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_)
1713{
1714        struct sk_buff *skb = list_->next;
1715
1716        if (skb == (struct sk_buff *)list_)
1717                skb = NULL;
1718        return skb;
1719}
1720
1721/**
1722 *      __skb_peek - peek at the head of a non-empty &sk_buff_head
1723 *      @list_: list to peek at
1724 *
1725 *      Like skb_peek(), but the caller knows that the list is not empty.
1726 */
1727static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_)
1728{
1729        return list_->next;
1730}
1731
1732/**
1733 *      skb_peek_next - peek skb following the given one from a queue
1734 *      @skb: skb to start from
1735 *      @list_: list to peek at
1736 *
1737 *      Returns %NULL when the end of the list is met or a pointer to the
1738 *      next element. The reference count is not incremented and the
1739 *      reference is therefore volatile. Use with caution.
1740 */
1741static inline struct sk_buff *skb_peek_next(struct sk_buff *skb,
1742                const struct sk_buff_head *list_)
1743{
1744        struct sk_buff *next = skb->next;
1745
1746        if (next == (struct sk_buff *)list_)
1747                next = NULL;
1748        return next;
1749}
1750
1751/**
1752 *      skb_peek_tail - peek at the tail of an &sk_buff_head
1753 *      @list_: list to peek at
1754 *
1755 *      Peek an &sk_buff. Unlike most other operations you _MUST_
1756 *      be careful with this one. A peek leaves the buffer on the
1757 *      list and someone else may run off with it. You must hold
1758 *      the appropriate locks or have a private queue to do this.
1759 *
1760 *      Returns %NULL for an empty list or a pointer to the tail element.
1761 *      The reference count is not incremented and the reference is therefore
1762 *      volatile. Use with caution.
1763 */
1764static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_)
1765{
1766        struct sk_buff *skb = list_->prev;
1767
1768        if (skb == (struct sk_buff *)list_)
1769                skb = NULL;
1770        return skb;
1771
1772}
1773
1774/**
1775 *      skb_queue_len   - get queue length
1776 *      @list_: list to measure
1777 *
1778 *      Return the length of an &sk_buff queue.
1779 */
1780static inline __u32 skb_queue_len(const struct sk_buff_head *list_)
1781{
1782        return list_->qlen;
1783}
1784
1785/**
1786 *      __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head
1787 *      @list: queue to initialize
1788 *
1789 *      This initializes only the list and queue length aspects of
1790 *      an sk_buff_head object.  This allows to initialize the list
1791 *      aspects of an sk_buff_head without reinitializing things like
1792 *      the spinlock.  It can also be used for on-stack sk_buff_head
1793 *      objects where the spinlock is known to not be used.
1794 */
1795static inline void __skb_queue_head_init(struct sk_buff_head *list)
1796{
1797        list->prev = list->next = (struct sk_buff *)list;
1798        list->qlen = 0;
1799}
1800
1801/*
1802 * This function creates a split out lock class for each invocation;
1803 * this is needed for now since a whole lot of users of the skb-queue
1804 * infrastructure in drivers have different locking usage (in hardirq)
1805 * than the networking core (in softirq only). In the long run either the
1806 * network layer or drivers should need annotation to consolidate the
1807 * main types of usage into 3 classes.
1808 */
1809static inline void skb_queue_head_init(struct sk_buff_head *list)
1810{
1811        spin_lock_init(&list->lock);
1812        __skb_queue_head_init(list);
1813}
1814
1815static inline void skb_queue_head_init_class(struct sk_buff_head *list,
1816                struct lock_class_key *class)
1817{
1818        skb_queue_head_init(list);
1819        lockdep_set_class(&list->lock, class);
1820}
1821
1822/*
1823 *      Insert an sk_buff on a list.
1824 *
1825 *      The "__skb_xxxx()" functions are the non-atomic ones that
1826 *      can only be called with interrupts disabled.
1827 */
1828static inline void __skb_insert(struct sk_buff *newsk,
1829                                struct sk_buff *prev, struct sk_buff *next,
1830                                struct sk_buff_head *list)
1831{
1832        newsk->next = next;
1833        newsk->prev = prev;
1834        next->prev  = prev->next = newsk;
1835        list->qlen++;
1836}
1837
1838static inline void __skb_queue_splice(const struct sk_buff_head *list,
1839                                      struct sk_buff *prev,
1840                                      struct sk_buff *next)
1841{
1842        struct sk_buff *first = list->next;
1843        struct sk_buff *last = list->prev;
1844
1845        first->prev = prev;
1846        prev->next = first;
1847
1848        last->next = next;
1849        next->prev = last;
1850}
1851
1852/**
1853 *      skb_queue_splice - join two skb lists, this is designed for stacks
1854 *      @list: the new list to add
1855 *      @head: the place to add it in the first list
1856 */
1857static inline void skb_queue_splice(const struct sk_buff_head *list,
1858                                    struct sk_buff_head *head)
1859{
1860        if (!skb_queue_empty(list)) {
1861                __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1862                head->qlen += list->qlen;
1863        }
1864}
1865
1866/**
1867 *      skb_queue_splice_init - join two skb lists and reinitialise the emptied list
1868 *      @list: the new list to add
1869 *      @head: the place to add it in the first list
1870 *
1871 *      The list at @list is reinitialised
1872 */
1873static inline void skb_queue_splice_init(struct sk_buff_head *list,
1874                                         struct sk_buff_head *head)
1875{
1876        if (!skb_queue_empty(list)) {
1877                __skb_queue_splice(list, (struct sk_buff *) head, head->next);
1878                head->qlen += list->qlen;
1879                __skb_queue_head_init(list);
1880        }
1881}
1882
1883/**
1884 *      skb_queue_splice_tail - join two skb lists, each list being a queue
1885 *      @list: the new list to add
1886 *      @head: the place to add it in the first list
1887 */
1888static inline void skb_queue_splice_tail(const struct sk_buff_head *list,
1889                                         struct sk_buff_head *head)
1890{
1891        if (!skb_queue_empty(list)) {
1892                __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1893                head->qlen += list->qlen;
1894        }
1895}
1896
1897/**
1898 *      skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list
1899 *      @list: the new list to add
1900 *      @head: the place to add it in the first list
1901 *
1902 *      Each of the lists is a queue.
1903 *      The list at @list is reinitialised
1904 */
1905static inline void skb_queue_splice_tail_init(struct sk_buff_head *list,
1906                                              struct sk_buff_head *head)
1907{
1908        if (!skb_queue_empty(list)) {
1909                __skb_queue_splice(list, head->prev, (struct sk_buff *) head);
1910                head->qlen += list->qlen;
1911                __skb_queue_head_init(list);
1912        }
1913}
1914
1915/**
1916 *      __skb_queue_after - queue a buffer at the list head
1917 *      @list: list to use
1918 *      @prev: place after this buffer
1919 *      @newsk: buffer to queue
1920 *
1921 *      Queue a buffer int the middle of a list. This function takes no locks
1922 *      and you must therefore hold required locks before calling it.
1923 *
1924 *      A buffer cannot be placed on two lists at the same time.
1925 */
1926static inline void __skb_queue_after(struct sk_buff_head *list,
1927                                     struct sk_buff *prev,
1928                                     struct sk_buff *newsk)
1929{
1930        __skb_insert(newsk, prev, prev->next, list);
1931}
1932
1933void skb_append(struct sk_buff *old, struct sk_buff *newsk,
1934                struct sk_buff_head *list);
1935
1936static inline void __skb_queue_before(struct sk_buff_head *list,
1937                                      struct sk_buff *next,
1938                                      struct sk_buff *newsk)
1939{
1940        __skb_insert(newsk, next->prev, next, list);
1941}
1942
1943/**
1944 *      __skb_queue_head - queue a buffer at the list head
1945 *      @list: list to use
1946 *      @newsk: buffer to queue
1947 *
1948 *      Queue a buffer at the start of a list. This function takes no locks
1949 *      and you must therefore hold required locks before calling it.
1950 *
1951 *      A buffer cannot be placed on two lists at the same time.
1952 */
1953static inline void __skb_queue_head(struct sk_buff_head *list,
1954                                    struct sk_buff *newsk)
1955{
1956        __skb_queue_after(list, (struct sk_buff *)list, newsk);
1957}
1958void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk);
1959
1960/**
1961 *      __skb_queue_tail - queue a buffer at the list tail
1962 *      @list: list to use
1963 *      @newsk: buffer to queue
1964 *
1965 *      Queue a buffer at the end of a list. This function takes no locks
1966 *      and you must therefore hold required locks before calling it.
1967 *
1968 *      A buffer cannot be placed on two lists at the same time.
1969 */
1970static inline void __skb_queue_tail(struct sk_buff_head *list,
1971                                   struct sk_buff *newsk)
1972{
1973        __skb_queue_before(list, (struct sk_buff *)list, newsk);
1974}
1975void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk);
1976
1977/*
1978 * remove sk_buff from list. _Must_ be called atomically, and with
1979 * the list known..
1980 */
1981void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list);
1982static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
1983{
1984        struct sk_buff *next, *prev;
1985
1986        list->qlen--;
1987        next       = skb->next;
1988        prev       = skb->prev;
1989        skb->next  = skb->prev = NULL;
1990        next->prev = prev;
1991        prev->next = next;
1992}
1993
1994/**
1995 *      __skb_dequeue - remove from the head of the queue
1996 *      @list: list to dequeue from
1997 *
1998 *      Remove the head of the list. This function does not take any locks
1999 *      so must be used with appropriate locks held only. The head item is
2000 *      returned or %NULL if the list is empty.
2001 */
2002static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list)
2003{
2004        struct sk_buff *skb = skb_peek(list);
2005        if (skb)
2006                __skb_unlink(skb, list);
2007        return skb;
2008}
2009struct sk_buff *skb_dequeue(struct sk_buff_head *list);
2010
2011/**
2012 *      __skb_dequeue_tail - remove from the tail of the queue
2013 *      @list: list to dequeue from
2014 *
2015 *      Remove the tail of the list. This function does not take any locks
2016 *      so must be used with appropriate locks held only. The tail item is
2017 *      returned or %NULL if the list is empty.
2018 */
2019static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list)
2020{
2021        struct sk_buff *skb = skb_peek_tail(list);
2022        if (skb)
2023                __skb_unlink(skb, list);
2024        return skb;
2025}
2026struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list);
2027
2028
2029static inline bool skb_is_nonlinear(const struct sk_buff *skb)
2030{
2031        return skb->data_len;
2032}
2033
2034static inline unsigned int skb_headlen(const struct sk_buff *skb)
2035{
2036        return skb->len - skb->data_len;
2037}
2038
2039static inline unsigned int __skb_pagelen(const struct sk_buff *skb)
2040{
2041        unsigned int i, len = 0;
2042
2043        for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--)
2044                len += skb_frag_size(&skb_shinfo(skb)->frags[i]);
2045        return len;
2046}
2047
2048static inline unsigned int skb_pagelen(const struct sk_buff *skb)
2049{
2050        return skb_headlen(skb) + __skb_pagelen(skb);
2051}
2052
2053/**
2054 * __skb_fill_page_desc - initialise a paged fragment in an skb
2055 * @skb: buffer containing fragment to be initialised
2056 * @i: paged fragment index to initialise
2057 * @page: the page to use for this fragment
2058 * @off: the offset to the data with @page
2059 * @size: the length of the data
2060 *
2061 * Initialises the @i'th fragment of @skb to point to &size bytes at
2062 * offset @off within @page.
2063 *
2064 * Does not take any additional reference on the fragment.
2065 */
2066static inline void __skb_fill_page_desc(struct sk_buff *skb, int i,
2067                                        struct page *page, int off, int size)
2068{
2069        skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2070
2071        /*
2072         * Propagate page pfmemalloc to the skb if we can. The problem is
2073         * that not all callers have unique ownership of the page but rely
2074         * on page_is_pfmemalloc doing the right thing(tm).
2075         */
2076        frag->page.p              = page;
2077        frag->page_offset         = off;
2078        skb_frag_size_set(frag, size);
2079
2080        page = compound_head(page);
2081        if (page_is_pfmemalloc(page))
2082                skb->pfmemalloc = true;
2083}
2084
2085/**
2086 * skb_fill_page_desc - initialise a paged fragment in an skb
2087 * @skb: buffer containing fragment to be initialised
2088 * @i: paged fragment index to initialise
2089 * @page: the page to use for this fragment
2090 * @off: the offset to the data with @page
2091 * @size: the length of the data
2092 *
2093 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of
2094 * @skb to point to @size bytes at offset @off within @page. In
2095 * addition updates @skb such that @i is the last fragment.
2096 *
2097 * Does not take any additional reference on the fragment.
2098 */
2099static inline void skb_fill_page_desc(struct sk_buff *skb, int i,
2100                                      struct page *page, int off, int size)
2101{
2102        __skb_fill_page_desc(skb, i, page, off, size);
2103        skb_shinfo(skb)->nr_frags = i + 1;
2104}
2105
2106void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
2107                     int size, unsigned int truesize);
2108
2109void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size,
2110                          unsigned int truesize);
2111
2112#define SKB_LINEAR_ASSERT(skb)  BUG_ON(skb_is_nonlinear(skb))
2113
2114#ifdef NET_SKBUFF_DATA_USES_OFFSET
2115static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2116{
2117        return skb->head + skb->tail;
2118}
2119
2120static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2121{
2122        skb->tail = skb->data - skb->head;
2123}
2124
2125static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2126{
2127        skb_reset_tail_pointer(skb);
2128        skb->tail += offset;
2129}
2130
2131#else /* NET_SKBUFF_DATA_USES_OFFSET */
2132static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb)
2133{
2134        return skb->tail;
2135}
2136
2137static inline void skb_reset_tail_pointer(struct sk_buff *skb)
2138{
2139        skb->tail = skb->data;
2140}
2141
2142static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset)
2143{
2144        skb->tail = skb->data + offset;
2145}
2146
2147#endif /* NET_SKBUFF_DATA_USES_OFFSET */
2148
2149/*
2150 *      Add data to an sk_buff
2151 */
2152void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len);
2153void *skb_put(struct sk_buff *skb, unsigned int len);
2154static inline void *__skb_put(struct sk_buff *skb, unsigned int len)
2155{
2156        void *tmp = skb_tail_pointer(skb);
2157        SKB_LINEAR_ASSERT(skb);
2158        skb->tail += len;
2159        skb->len  += len;
2160        return tmp;
2161}
2162
2163static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len)
2164{
2165        void *tmp = __skb_put(skb, len);
2166
2167        memset(tmp, 0, len);
2168        return tmp;
2169}
2170
2171static inline void *__skb_put_data(struct sk_buff *skb, const void *data,
2172                                   unsigned int len)
2173{
2174        void *tmp = __skb_put(skb, len);
2175
2176        memcpy(tmp, data, len);
2177        return tmp;
2178}
2179
2180static inline void __skb_put_u8(struct sk_buff *skb, u8 val)
2181{
2182        *(u8 *)__skb_put(skb, 1) = val;
2183}
2184
2185static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len)
2186{
2187        void *tmp = skb_put(skb, len);
2188
2189        memset(tmp, 0, len);
2190
2191        return tmp;
2192}
2193
2194static inline void *skb_put_data(struct sk_buff *skb, const void *data,
2195                                 unsigned int len)
2196{
2197        void *tmp = skb_put(skb, len);
2198
2199        memcpy(tmp, data, len);
2200
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
2209void *skb_push(struct sk_buff *skb, unsigned int len);
2210static inline void *__skb_push(struct sk_buff *skb, unsigned int len)
2211{
2212        skb->data -= len;
2213        skb->len  += len;
2214        return skb->data;
2215}
2216
2217void *skb_pull(struct sk_buff *skb, unsigned int len);
2218static inline void *__skb_pull(struct sk_buff *skb, unsigned int len)
2219{
2220        skb->len -= len;
2221        BUG_ON(skb->len < skb->data_len);
2222        return skb->data += len;
2223}
2224
2225static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len)
2226{
2227        return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len);
2228}
2229
2230void *__pskb_pull_tail(struct sk_buff *skb, int delta);
2231
2232static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len)
2233{
2234        if (len > skb_headlen(skb) &&
2235            !__pskb_pull_tail(skb, len - skb_headlen(skb)))
2236                return NULL;
2237        skb->len -= len;
2238        return skb->data += len;
2239}
2240
2241static inline void *pskb_pull(struct sk_buff *skb, unsigned int len)
2242{
2243        return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len);
2244}
2245
2246static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len)
2247{
2248        if (likely(len <= skb_headlen(skb)))
2249                return 1;
2250        if (unlikely(len > skb->len))
2251                return 0;
2252        return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL;
2253}
2254
2255void skb_condense(struct sk_buff *skb);
2256
2257/**
2258 *      skb_headroom - bytes at buffer head
2259 *      @skb: buffer to check
2260 *
2261 *      Return the number of bytes of free space at the head of an &sk_buff.
2262 */
2263static inline unsigned int skb_headroom(const struct sk_buff *skb)
2264{
2265        return skb->data - skb->head;
2266}
2267
2268/**
2269 *      skb_tailroom - bytes at buffer end
2270 *      @skb: buffer to check
2271 *
2272 *      Return the number of bytes of free space at the tail of an sk_buff
2273 */
2274static inline int skb_tailroom(const struct sk_buff *skb)
2275{
2276        return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail;
2277}
2278
2279/**
2280 *      skb_availroom - bytes at buffer end
2281 *      @skb: buffer to check
2282 *
2283 *      Return the number of bytes of free space at the tail of an sk_buff
2284 *      allocated by sk_stream_alloc()
2285 */
2286static inline int skb_availroom(const struct sk_buff *skb)
2287{
2288        if (skb_is_nonlinear(skb))
2289                return 0;
2290
2291        return skb->end - skb->tail - skb->reserved_tailroom;
2292}
2293
2294/**
2295 *      skb_reserve - adjust headroom
2296 *      @skb: buffer to alter
2297 *      @len: bytes to move
2298 *
2299 *      Increase the headroom of an empty &sk_buff by reducing the tail
2300 *      room. This is only allowed for an empty buffer.
2301 */
2302static inline void skb_reserve(struct sk_buff *skb, int len)
2303{
2304        skb->data += len;
2305        skb->tail += len;
2306}
2307
2308/**
2309 *      skb_tailroom_reserve - adjust reserved_tailroom
2310 *      @skb: buffer to alter
2311 *      @mtu: maximum amount of headlen permitted
2312 *      @needed_tailroom: minimum amount of reserved_tailroom
2313 *
2314 *      Set reserved_tailroom so that headlen can be as large as possible but
2315 *      not larger than mtu and tailroom cannot be smaller than
2316 *      needed_tailroom.
2317 *      The required headroom should already have been reserved before using
2318 *      this function.
2319 */
2320static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu,
2321                                        unsigned int needed_tailroom)
2322{
2323        SKB_LINEAR_ASSERT(skb);
2324        if (mtu < skb_tailroom(skb) - needed_tailroom)
2325                /* use at most mtu */
2326                skb->reserved_tailroom = skb_tailroom(skb) - mtu;
2327        else
2328                /* use up to all available space */
2329                skb->reserved_tailroom = needed_tailroom;
2330}
2331
2332#define ENCAP_TYPE_ETHER        0
2333#define ENCAP_TYPE_IPPROTO      1
2334
2335static inline void skb_set_inner_protocol(struct sk_buff *skb,
2336                                          __be16 protocol)
2337{
2338        skb->inner_protocol = protocol;
2339        skb->inner_protocol_type = ENCAP_TYPE_ETHER;
2340}
2341
2342static inline void skb_set_inner_ipproto(struct sk_buff *skb,
2343                                         __u8 ipproto)
2344{
2345        skb->inner_ipproto = ipproto;
2346        skb->inner_protocol_type = ENCAP_TYPE_IPPROTO;
2347}
2348
2349static inline void skb_reset_inner_headers(struct sk_buff *skb)
2350{
2351        skb->inner_mac_header = skb->mac_header;
2352        skb->inner_network_header = skb->network_header;
2353        skb->inner_transport_header = skb->transport_header;
2354}
2355
2356static inline void skb_reset_mac_len(struct sk_buff *skb)
2357{
2358        skb->mac_len = skb->network_header - skb->mac_header;
2359}
2360
2361static inline unsigned char *skb_inner_transport_header(const struct sk_buff
2362                                                        *skb)
2363{
2364        return skb->head + skb->inner_transport_header;
2365}
2366
2367static inline int skb_inner_transport_offset(const struct sk_buff *skb)
2368{
2369        return skb_inner_transport_header(skb) - skb->data;
2370}
2371
2372static inline void skb_reset_inner_transport_header(struct sk_buff *skb)
2373{
2374        skb->inner_transport_header = skb->data - skb->head;
2375}
2376
2377static inline void skb_set_inner_transport_header(struct sk_buff *skb,
2378                                                   const int offset)
2379{
2380        skb_reset_inner_transport_header(skb);
2381        skb->inner_transport_header += offset;
2382}
2383
2384static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb)
2385{
2386        return skb->head + skb->inner_network_header;
2387}
2388
2389static inline void skb_reset_inner_network_header(struct sk_buff *skb)
2390{
2391        skb->inner_network_header = skb->data - skb->head;
2392}
2393
2394static inline void skb_set_inner_network_header(struct sk_buff *skb,
2395                                                const int offset)
2396{
2397        skb_reset_inner_network_header(skb);
2398        skb->inner_network_header += offset;
2399}
2400
2401static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb)
2402{
2403        return skb->head + skb->inner_mac_header;
2404}
2405
2406static inline void skb_reset_inner_mac_header(struct sk_buff *skb)
2407{
2408        skb->inner_mac_header = skb->data - skb->head;
2409}
2410
2411static inline void skb_set_inner_mac_header(struct sk_buff *skb,
2412                                            const int offset)
2413{
2414        skb_reset_inner_mac_header(skb);
2415        skb->inner_mac_header += offset;
2416}
2417static inline bool skb_transport_header_was_set(const struct sk_buff *skb)
2418{
2419        return skb->transport_header != (typeof(skb->transport_header))~0U;
2420}
2421
2422static inline unsigned char *skb_transport_header(const struct sk_buff *skb)
2423{
2424        return skb->head + skb->transport_header;
2425}
2426
2427static inline void skb_reset_transport_header(struct sk_buff *skb)
2428{
2429        skb->transport_header = skb->data - skb->head;
2430}
2431
2432static inline void skb_set_transport_header(struct sk_buff *skb,
2433                                            const int offset)
2434{
2435        skb_reset_transport_header(skb);
2436        skb->transport_header += offset;
2437}
2438
2439static inline unsigned char *skb_network_header(const struct sk_buff *skb)
2440{
2441        return skb->head + skb->network_header;
2442}
2443
2444static inline void skb_reset_network_header(struct sk_buff *skb)
2445{
2446        skb->network_header = skb->data - skb->head;
2447}
2448
2449static inline void skb_set_network_header(struct sk_buff *skb, const int offset)
2450{
2451        skb_reset_network_header(skb);
2452        skb->network_header += offset;
2453}
2454
2455static inline unsigned char *skb_mac_header(const struct sk_buff *skb)
2456{
2457        return skb->head + skb->mac_header;
2458}
2459
2460static inline int skb_mac_offset(const struct sk_buff *skb)
2461{
2462        return skb_mac_header(skb) - skb->data;
2463}
2464
2465static inline u32 skb_mac_header_len(const struct sk_buff *skb)
2466{
2467        return skb->network_header - skb->mac_header;
2468}
2469
2470static inline int skb_mac_header_was_set(const struct sk_buff *skb)
2471{
2472        return skb->mac_header != (typeof(skb->mac_header))~0U;
2473}
2474
2475static inline void skb_reset_mac_header(struct sk_buff *skb)
2476{
2477        skb->mac_header = skb->data - skb->head;
2478}
2479
2480static inline void skb_set_mac_header(struct sk_buff *skb, const int offset)
2481{
2482        skb_reset_mac_header(skb);
2483        skb->mac_header += offset;
2484}
2485
2486static inline void skb_pop_mac_header(struct sk_buff *skb)
2487{
2488        skb->mac_header = skb->network_header;
2489}
2490
2491static inline void skb_probe_transport_header(struct sk_buff *skb)
2492{
2493        struct flow_keys_basic keys;
2494
2495        if (skb_transport_header_was_set(skb))
2496                return;
2497
2498        if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys,
2499                                             NULL, 0, 0, 0, 0))
2500                skb_set_transport_header(skb, keys.control.thoff);
2501}
2502
2503static inline void skb_mac_header_rebuild(struct sk_buff *skb)
2504{
2505        if (skb_mac_header_was_set(skb)) {
2506                const unsigned char *old_mac = skb_mac_header(skb);
2507
2508                skb_set_mac_header(skb, -skb->mac_len);
2509                memmove(skb_mac_header(skb), old_mac, skb->mac_len);
2510        }
2511}
2512
2513static inline int skb_checksum_start_offset(const struct sk_buff *skb)
2514{
2515        return skb->csum_start - skb_headroom(skb);
2516}
2517
2518static inline unsigned char *skb_checksum_start(const struct sk_buff *skb)
2519{
2520        return skb->head + skb->csum_start;
2521}
2522
2523static inline int skb_transport_offset(const struct sk_buff *skb)
2524{
2525        return skb_transport_header(skb) - skb->data;
2526}
2527
2528static inline u32 skb_network_header_len(const struct sk_buff *skb)
2529{
2530        return skb->transport_header - skb->network_header;
2531}
2532
2533static inline u32 skb_inner_network_header_len(const struct sk_buff *skb)
2534{
2535        return skb->inner_transport_header - skb->inner_network_header;
2536}
2537
2538static inline int skb_network_offset(const struct sk_buff *skb)
2539{
2540        return skb_network_header(skb) - skb->data;
2541}
2542
2543static inline int skb_inner_network_offset(const struct sk_buff *skb)
2544{
2545        return skb_inner_network_header(skb) - skb->data;
2546}
2547
2548static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len)
2549{
2550        return pskb_may_pull(skb, skb_network_offset(skb) + len);
2551}
2552
2553/*
2554 * CPUs often take a performance hit when accessing unaligned memory
2555 * locations. The actual performance hit varies, it can be small if the
2556 * hardware handles it or large if we have to take an exception and fix it
2557 * in software.
2558 *
2559 * Since an ethernet header is 14 bytes network drivers often end up with
2560 * the IP header at an unaligned offset. The IP header can be aligned by
2561 * shifting the start of the packet by 2 bytes. Drivers should do this
2562 * with:
2563 *
2564 * skb_reserve(skb, NET_IP_ALIGN);
2565 *
2566 * The downside to this alignment of the IP header is that the DMA is now
2567 * unaligned. On some architectures the cost of an unaligned DMA is high
2568 * and this cost outweighs the gains made by aligning the IP header.
2569 *
2570 * Since this trade off varies between architectures, we allow NET_IP_ALIGN
2571 * to be overridden.
2572 */
2573#ifndef NET_IP_ALIGN
2574#define NET_IP_ALIGN    2
2575#endif
2576
2577/*
2578 * The networking layer reserves some headroom in skb data (via
2579 * dev_alloc_skb). This is used to avoid having to reallocate skb data when
2580 * the header has to grow. In the default case, if the header has to grow
2581 * 32 bytes or less we avoid the reallocation.
2582 *
2583 * Unfortunately this headroom changes the DMA alignment of the resulting
2584 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive
2585 * on some architectures. An architecture can override this value,
2586 * perhaps setting it to a cacheline in size (since that will maintain
2587 * cacheline alignment of the DMA). It must be a power of 2.
2588 *
2589 * Various parts of the networking layer expect at least 32 bytes of
2590 * headroom, you should not reduce this.
2591 *
2592 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS)
2593 * to reduce average number of cache lines per packet.
2594 * get_rps_cpus() for example only access one 64 bytes aligned block :
2595 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8)
2596 */
2597#ifndef NET_SKB_PAD
2598#define NET_SKB_PAD     max(32, L1_CACHE_BYTES)
2599#endif
2600
2601int ___pskb_trim(struct sk_buff *skb, unsigned int len);
2602
2603static inline void __skb_set_length(struct sk_buff *skb, unsigned int len)
2604{
2605        if (WARN_ON(skb_is_nonlinear(skb)))
2606                return;
2607        skb->len = len;
2608        skb_set_tail_pointer(skb, len);
2609}
2610
2611static inline void __skb_trim(struct sk_buff *skb, unsigned int len)
2612{
2613        __skb_set_length(skb, len);
2614}
2615
2616void skb_trim(struct sk_buff *skb, unsigned int len);
2617
2618static inline int __pskb_trim(struct sk_buff *skb, unsigned int len)
2619{
2620        if (skb->data_len)
2621                return ___pskb_trim(skb, len);
2622        __skb_trim(skb, len);
2623        return 0;
2624}
2625
2626static inline int pskb_trim(struct sk_buff *skb, unsigned int len)
2627{
2628        return (len < skb->len) ? __pskb_trim(skb, len) : 0;
2629}
2630
2631/**
2632 *      pskb_trim_unique - remove end from a paged unique (not cloned) buffer
2633 *      @skb: buffer to alter
2634 *      @len: new length
2635 *
2636 *      This is identical to pskb_trim except that the caller knows that
2637 *      the skb is not cloned so we should never get an error due to out-
2638 *      of-memory.
2639 */
2640static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len)
2641{
2642        int err = pskb_trim(skb, len);
2643        BUG_ON(err);
2644}
2645
2646static inline int __skb_grow(struct sk_buff *skb, unsigned int len)
2647{
2648        unsigned int diff = len - skb->len;
2649
2650        if (skb_tailroom(skb) < diff) {
2651                int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb),
2652                                           GFP_ATOMIC);
2653                if (ret)
2654                        return ret;
2655        }
2656        __skb_set_length(skb, len);
2657        return 0;
2658}
2659
2660/**
2661 *      skb_orphan - orphan a buffer
2662 *      @skb: buffer to orphan
2663 *
2664 *      If a buffer currently has an owner then we call the owner's
2665 *      destructor function and make the @skb unowned. The buffer continues
2666 *      to exist but is no longer charged to its former owner.
2667 */
2668static inline void skb_orphan(struct sk_buff *skb)
2669{
2670        if (skb->destructor) {
2671                skb->destructor(skb);
2672                skb->destructor = NULL;
2673                skb->sk         = NULL;
2674        } else {
2675                BUG_ON(skb->sk);
2676        }
2677}
2678
2679/**
2680 *      skb_orphan_frags - orphan the frags contained in a buffer
2681 *      @skb: buffer to orphan frags from
2682 *      @gfp_mask: allocation mask for replacement pages
2683 *
2684 *      For each frag in the SKB which needs a destructor (i.e. has an
2685 *      owner) create a copy of that frag and release the original
2686 *      page by calling the destructor.
2687 */
2688static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask)
2689{
2690        if (likely(!skb_zcopy(skb)))
2691                return 0;
2692        if (!skb_zcopy_is_nouarg(skb) &&
2693            skb_uarg(skb)->callback == sock_zerocopy_callback)
2694                return 0;
2695        return skb_copy_ubufs(skb, gfp_mask);
2696}
2697
2698/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */
2699static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask)
2700{
2701        if (likely(!skb_zcopy(skb)))
2702                return 0;
2703        return skb_copy_ubufs(skb, gfp_mask);
2704}
2705
2706/**
2707 *      __skb_queue_purge - empty a list
2708 *      @list: list to empty
2709 *
2710 *      Delete all buffers on an &sk_buff list. Each buffer is removed from
2711 *      the list and one reference dropped. This function does not take the
2712 *      list lock and the caller must hold the relevant locks to use it.
2713 */
2714static inline void __skb_queue_purge(struct sk_buff_head *list)
2715{
2716        struct sk_buff *skb;
2717        while ((skb = __skb_dequeue(list)) != NULL)
2718                kfree_skb(skb);
2719}
2720void skb_queue_purge(struct sk_buff_head *list);
2721
2722unsigned int skb_rbtree_purge(struct rb_root *root);
2723
2724void *netdev_alloc_frag(unsigned int fragsz);
2725
2726struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length,
2727                                   gfp_t gfp_mask);
2728
2729/**
2730 *      netdev_alloc_skb - allocate an skbuff for rx on a specific device
2731 *      @dev: network device to receive on
2732 *      @length: length to allocate
2733 *
2734 *      Allocate a new &sk_buff and assign it a usage count of one. The
2735 *      buffer has unspecified headroom built in. Users should allocate
2736 *      the headroom they think they need without accounting for the
2737 *      built in space. The built in space is used for optimisations.
2738 *
2739 *      %NULL is returned if there is no free memory. Although this function
2740 *      allocates memory it can be called from an interrupt.
2741 */
2742static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev,
2743                                               unsigned int length)
2744{
2745        return __netdev_alloc_skb(dev, length, GFP_ATOMIC);
2746}
2747
2748/* legacy helper around __netdev_alloc_skb() */
2749static inline struct sk_buff *__dev_alloc_skb(unsigned int length,
2750                                              gfp_t gfp_mask)
2751{
2752        return __netdev_alloc_skb(NULL, length, gfp_mask);
2753}
2754
2755/* legacy helper around netdev_alloc_skb() */
2756static inline struct sk_buff *dev_alloc_skb(unsigned int length)
2757{
2758        return netdev_alloc_skb(NULL, length);
2759}
2760
2761
2762static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev,
2763                unsigned int length, gfp_t gfp)
2764{
2765        struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp);
2766
2767        if (NET_IP_ALIGN && skb)
2768                skb_reserve(skb, NET_IP_ALIGN);
2769        return skb;
2770}
2771
2772static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev,
2773                unsigned int length)
2774{
2775        return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC);
2776}
2777
2778static inline void skb_free_frag(void *addr)
2779{
2780        page_frag_free(addr);
2781}
2782
2783void *napi_alloc_frag(unsigned int fragsz);
2784struct sk_buff *__napi_alloc_skb(struct napi_struct *napi,
2785                                 unsigned int length, gfp_t gfp_mask);
2786static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi,
2787                                             unsigned int length)
2788{
2789        return __napi_alloc_skb(napi, length, GFP_ATOMIC);
2790}
2791void napi_consume_skb(struct sk_buff *skb, int budget);
2792
2793void __kfree_skb_flush(void);
2794void __kfree_skb_defer(struct sk_buff *skb);
2795
2796/**
2797 * __dev_alloc_pages - allocate page for network Rx
2798 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2799 * @order: size of the allocation
2800 *
2801 * Allocate a new page.
2802 *
2803 * %NULL is returned if there is no free memory.
2804*/
2805static inline struct page *__dev_alloc_pages(gfp_t gfp_mask,
2806                                             unsigned int order)
2807{
2808        /* This piece of code contains several assumptions.
2809         * 1.  This is for device Rx, therefor a cold page is preferred.
2810         * 2.  The expectation is the user wants a compound page.
2811         * 3.  If requesting a order 0 page it will not be compound
2812         *     due to the check to see if order has a value in prep_new_page
2813         * 4.  __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to
2814         *     code in gfp_to_alloc_flags that should be enforcing this.
2815         */
2816        gfp_mask |= __GFP_COMP | __GFP_MEMALLOC;
2817
2818        return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order);
2819}
2820
2821static inline struct page *dev_alloc_pages(unsigned int order)
2822{
2823        return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order);
2824}
2825
2826/**
2827 * __dev_alloc_page - allocate a page for network Rx
2828 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx
2829 *
2830 * Allocate a new page.
2831 *
2832 * %NULL is returned if there is no free memory.
2833 */
2834static inline struct page *__dev_alloc_page(gfp_t gfp_mask)
2835{
2836        return __dev_alloc_pages(gfp_mask, 0);
2837}
2838
2839static inline struct page *dev_alloc_page(void)
2840{
2841        return dev_alloc_pages(0);
2842}
2843
2844/**
2845 *      skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page
2846 *      @page: The page that was allocated from skb_alloc_page
2847 *      @skb: The skb that may need pfmemalloc set
2848 */
2849static inline void skb_propagate_pfmemalloc(struct page *page,
2850                                             struct sk_buff *skb)
2851{
2852        if (page_is_pfmemalloc(page))
2853                skb->pfmemalloc = true;
2854}
2855
2856/**
2857 * skb_frag_page - retrieve the page referred to by a paged fragment
2858 * @frag: the paged fragment
2859 *
2860 * Returns the &struct page associated with @frag.
2861 */
2862static inline struct page *skb_frag_page(const skb_frag_t *frag)
2863{
2864        return frag->page.p;
2865}
2866
2867/**
2868 * __skb_frag_ref - take an addition reference on a paged fragment.
2869 * @frag: the paged fragment
2870 *
2871 * Takes an additional reference on the paged fragment @frag.
2872 */
2873static inline void __skb_frag_ref(skb_frag_t *frag)
2874{
2875        get_page(skb_frag_page(frag));
2876}
2877
2878/**
2879 * skb_frag_ref - take an addition reference on a paged fragment of an skb.
2880 * @skb: the buffer
2881 * @f: the fragment offset.
2882 *
2883 * Takes an additional reference on the @f'th paged fragment of @skb.
2884 */
2885static inline void skb_frag_ref(struct sk_buff *skb, int f)
2886{
2887        __skb_frag_ref(&skb_shinfo(skb)->frags[f]);
2888}
2889
2890/**
2891 * __skb_frag_unref - release a reference on a paged fragment.
2892 * @frag: the paged fragment
2893 *
2894 * Releases a reference on the paged fragment @frag.
2895 */
2896static inline void __skb_frag_unref(skb_frag_t *frag)
2897{
2898        put_page(skb_frag_page(frag));
2899}
2900
2901/**
2902 * skb_frag_unref - release a reference on a paged fragment of an skb.
2903 * @skb: the buffer
2904 * @f: the fragment offset
2905 *
2906 * Releases a reference on the @f'th paged fragment of @skb.
2907 */
2908static inline void skb_frag_unref(struct sk_buff *skb, int f)
2909{
2910        __skb_frag_unref(&skb_shinfo(skb)->frags[f]);
2911}
2912
2913/**
2914 * skb_frag_address - gets the address of the data contained in a paged fragment
2915 * @frag: the paged fragment buffer
2916 *
2917 * Returns the address of the data within @frag. The page must already
2918 * be mapped.
2919 */
2920static inline void *skb_frag_address(const skb_frag_t *frag)
2921{
2922        return page_address(skb_frag_page(frag)) + frag->page_offset;
2923}
2924
2925/**
2926 * skb_frag_address_safe - gets the address of the data contained in a paged fragment
2927 * @frag: the paged fragment buffer
2928 *
2929 * Returns the address of the data within @frag. Checks that the page
2930 * is mapped and returns %NULL otherwise.
2931 */
2932static inline void *skb_frag_address_safe(const skb_frag_t *frag)
2933{
2934        void *ptr = page_address(skb_frag_page(frag));
2935        if (unlikely(!ptr))
2936                return NULL;
2937
2938        return ptr + frag->page_offset;
2939}
2940
2941/**
2942 * __skb_frag_set_page - sets the page contained in a paged fragment
2943 * @frag: the paged fragment
2944 * @page: the page to set
2945 *
2946 * Sets the fragment @frag to contain @page.
2947 */
2948static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page)
2949{
2950        frag->page.p = page;
2951}
2952
2953/**
2954 * skb_frag_set_page - sets the page contained in a paged fragment of an skb
2955 * @skb: the buffer
2956 * @f: the fragment offset
2957 * @page: the page to set
2958 *
2959 * Sets the @f'th fragment of @skb to contain @page.
2960 */
2961static inline void skb_frag_set_page(struct sk_buff *skb, int f,
2962                                     struct page *page)
2963{
2964        __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page);
2965}
2966
2967bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio);
2968
2969/**
2970 * skb_frag_dma_map - maps a paged fragment via the DMA API
2971 * @dev: the device to map the fragment to
2972 * @frag: the paged fragment to map
2973 * @offset: the offset within the fragment (starting at the
2974 *          fragment's own offset)
2975 * @size: the number of bytes to map
2976 * @dir: the direction of the mapping (``PCI_DMA_*``)
2977 *
2978 * Maps the page associated with @frag to @device.
2979 */
2980static inline dma_addr_t skb_frag_dma_map(struct device *dev,
2981                                          const skb_frag_t *frag,
2982                                          size_t offset, size_t size,
2983                                          enum dma_data_direction dir)
2984{
2985        return dma_map_page(dev, skb_frag_page(frag),
2986                            frag->page_offset + offset, size, dir);
2987}
2988
2989static inline struct sk_buff *pskb_copy(struct sk_buff *skb,
2990                                        gfp_t gfp_mask)
2991{
2992        return __pskb_copy(skb, skb_headroom(skb), gfp_mask);
2993}
2994
2995
2996static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb,
2997                                                  gfp_t gfp_mask)
2998{
2999        return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true);
3000}
3001
3002
3003/**
3004 *      skb_clone_writable - is the header of a clone writable
3005 *      @skb: buffer to check
3006 *      @len: length up to which to write
3007 *
3008 *      Returns true if modifying the header part of the cloned buffer
3009 *      does not requires the data to be copied.
3010 */
3011static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len)
3012{
3013        return !skb_header_cloned(skb) &&
3014               skb_headroom(skb) + len <= skb->hdr_len;
3015}
3016
3017static inline int skb_try_make_writable(struct sk_buff *skb,
3018                                        unsigned int write_len)
3019{
3020        return skb_cloned(skb) && !skb_clone_writable(skb, write_len) &&
3021               pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
3022}
3023
3024static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom,
3025                            int cloned)
3026{
3027        int delta = 0;
3028
3029        if (headroom > skb_headroom(skb))
3030                delta = headroom - skb_headroom(skb);
3031
3032        if (delta || cloned)
3033                return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0,
3034                                        GFP_ATOMIC);
3035        return 0;
3036}
3037
3038/**
3039 *      skb_cow - copy header of skb when it is required
3040 *      @skb: buffer to cow
3041 *      @headroom: needed headroom
3042 *
3043 *      If the skb passed lacks sufficient headroom or its data part
3044 *      is shared, data is reallocated. If reallocation fails, an error
3045 *      is returned and original skb is not changed.
3046 *
3047 *      The result is skb with writable area skb->head...skb->tail
3048 *      and at least @headroom of space at head.
3049 */
3050static inline int skb_cow(struct sk_buff *skb, unsigned int headroom)
3051{
3052        return __skb_cow(skb, headroom, skb_cloned(skb));
3053}
3054
3055/**
3056 *      skb_cow_head - skb_cow but only making the head writable
3057 *      @skb: buffer to cow
3058 *      @headroom: needed headroom
3059 *
3060 *      This function is identical to skb_cow except that we replace the
3061 *      skb_cloned check by skb_header_cloned.  It should be used when
3062 *      you only need to push on some header and do not need to modify
3063 *      the data.
3064 */
3065static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom)
3066{
3067        return __skb_cow(skb, headroom, skb_header_cloned(skb));
3068}
3069
3070/**
3071 *      skb_padto       - pad an skbuff up to a minimal size
3072 *      @skb: buffer to pad
3073 *      @len: minimal length
3074 *
3075 *      Pads up a buffer to ensure the trailing bytes exist and are
3076 *      blanked. If the buffer already contains sufficient data it
3077 *      is untouched. Otherwise it is extended. Returns zero on
3078 *      success. The skb is freed on error.
3079 */
3080static inline int skb_padto(struct sk_buff *skb, unsigned int len)
3081{
3082        unsigned int size = skb->len;
3083        if (likely(size >= len))
3084                return 0;
3085        return skb_pad(skb, len - size);
3086}
3087
3088/**
3089 *      __skb_put_padto - increase size and pad an skbuff up to a minimal size
3090 *      @skb: buffer to pad
3091 *      @len: minimal length
3092 *      @free_on_error: free buffer on error
3093 *
3094 *      Pads up a buffer to ensure the trailing bytes exist and are
3095 *      blanked. If the buffer already contains sufficient data it
3096 *      is untouched. Otherwise it is extended. Returns zero on
3097 *      success. The skb is freed on error if @free_on_error is true.
3098 */
3099static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len,
3100                                  bool free_on_error)
3101{
3102        unsigned int size = skb->len;
3103
3104        if (unlikely(size < len)) {
3105                len -= size;
3106                if (__skb_pad(skb, len, free_on_error))
3107                        return -ENOMEM;
3108                __skb_put(skb, len);
3109        }
3110        return 0;
3111}
3112
3113/**
3114 *      skb_put_padto - increase size and pad an skbuff up to a minimal size
3115 *      @skb: buffer to pad
3116 *      @len: minimal length
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.
3122 */
3123static inline int skb_put_padto(struct sk_buff *skb, unsigned int len)
3124{
3125        return __skb_put_padto(skb, len, true);
3126}
3127
3128static inline int skb_add_data(struct sk_buff *skb,
3129                               struct iov_iter *from, int copy)
3130{
3131        const int off = skb->len;
3132
3133        if (skb->ip_summed == CHECKSUM_NONE) {
3134                __wsum csum = 0;
3135                if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy,
3136                                                 &csum, from)) {
3137                        skb->csum = csum_block_add(skb->csum, csum, off);
3138                        return 0;
3139                }
3140        } else if (copy_from_iter_full(skb_put(skb, copy), copy, from))
3141                return 0;
3142
3143        __skb_trim(skb, off);
3144        return -EFAULT;
3145}
3146
3147static inline bool skb_can_coalesce(struct sk_buff *skb, int i,
3148                                    const struct page *page, int off)
3149{
3150        if (skb_zcopy(skb))
3151                return false;
3152        if (i) {
3153                const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1];
3154
3155                return page == skb_frag_page(frag) &&
3156                       off == frag->page_offset + skb_frag_size(frag);
3157        }
3158        return false;
3159}
3160
3161static inline int __skb_linearize(struct sk_buff *skb)
3162{
3163        return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM;
3164}
3165
3166/**
3167 *      skb_linearize - convert paged skb to linear one
3168 *      @skb: buffer to linarize
3169 *
3170 *      If there is no free memory -ENOMEM is returned, otherwise zero
3171 *      is returned and the old skb data released.
3172 */
3173static inline int skb_linearize(struct sk_buff *skb)
3174{
3175        return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0;
3176}
3177
3178/**
3179 * skb_has_shared_frag - can any frag be overwritten
3180 * @skb: buffer to test
3181 *
3182 * Return true if the skb has at least one frag that might be modified
3183 * by an external entity (as in vmsplice()/sendfile())
3184 */
3185static inline bool skb_has_shared_frag(const struct sk_buff *skb)
3186{
3187        return skb_is_nonlinear(skb) &&
3188               skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG;
3189}
3190
3191/**
3192 *      skb_linearize_cow - make sure skb is linear and writable
3193 *      @skb: buffer to process
3194 *
3195 *      If there is no free memory -ENOMEM is returned, otherwise zero
3196 *      is returned and the old skb data released.
3197 */
3198static inline int skb_linearize_cow(struct sk_buff *skb)
3199{
3200        return skb_is_nonlinear(skb) || skb_cloned(skb) ?
3201               __skb_linearize(skb) : 0;
3202}
3203
3204static __always_inline void
3205__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3206                     unsigned int off)
3207{
3208        if (skb->ip_summed == CHECKSUM_COMPLETE)
3209                skb->csum = csum_block_sub(skb->csum,
3210                                           csum_partial(start, len, 0), off);
3211        else if (skb->ip_summed == CHECKSUM_PARTIAL &&
3212                 skb_checksum_start_offset(skb) < 0)
3213                skb->ip_summed = CHECKSUM_NONE;
3214}
3215
3216/**
3217 *      skb_postpull_rcsum - update checksum for received skb after pull
3218 *      @skb: buffer to update
3219 *      @start: start of data before pull
3220 *      @len: length of data pulled
3221 *
3222 *      After doing a pull on a received packet, you need to call this to
3223 *      update the CHECKSUM_COMPLETE checksum, or set ip_summed to
3224 *      CHECKSUM_NONE so that it can be recomputed from scratch.
3225 */
3226static inline void skb_postpull_rcsum(struct sk_buff *skb,
3227                                      const void *start, unsigned int len)
3228{
3229        __skb_postpull_rcsum(skb, start, len, 0);
3230}
3231
3232static __always_inline void
3233__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len,
3234                     unsigned int off)
3235{
3236        if (skb->ip_summed == CHECKSUM_COMPLETE)
3237                skb->csum = csum_block_add(skb->csum,
3238                                           csum_partial(start, len, 0), off);
3239}
3240
3241/**
3242 *      skb_postpush_rcsum - update checksum for received skb after push
3243 *      @skb: buffer to update
3244 *      @start: start of data after push
3245 *      @len: length of data pushed
3246 *
3247 *      After doing a push on a received packet, you need to call this to
3248 *      update the CHECKSUM_COMPLETE checksum.
3249 */
3250static inline void skb_postpush_rcsum(struct sk_buff *skb,
3251                                      const void *start, unsigned int len)
3252{
3253        __skb_postpush_rcsum(skb, start, len, 0);
3254}
3255
3256void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len);
3257
3258/**
3259 *      skb_push_rcsum - push skb and update receive checksum
3260 *      @skb: buffer to update
3261 *      @len: length of data pulled
3262 *
3263 *      This function performs an skb_push on the packet and updates
3264 *      the CHECKSUM_COMPLETE checksum.  It should be used on
3265 *      receive path processing instead of skb_push unless you know
3266 *      that the checksum difference is zero (e.g., a valid IP header)
3267 *      or you are setting ip_summed to CHECKSUM_NONE.
3268 */
3269static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len)
3270{
3271        skb_push(skb, len);
3272        skb_postpush_rcsum(skb, skb->data, len);
3273        return skb->data;
3274}
3275
3276int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len);
3277/**
3278 *      pskb_trim_rcsum - trim received skb and update checksum
3279 *      @skb: buffer to trim
3280 *      @len: new length
3281 *
3282 *      This is exactly the same as pskb_trim except that it ensures the
3283 *      checksum of received packets are still valid after the operation.
3284 *      It can change skb pointers.
3285 */
3286
3287static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3288{
3289        if (likely(len >= skb->len))
3290                return 0;
3291        return pskb_trim_rcsum_slow(skb, len);
3292}
3293
3294static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len)
3295{
3296        if (skb->ip_summed == CHECKSUM_COMPLETE)
3297                skb->ip_summed = CHECKSUM_NONE;
3298        __skb_trim(skb, len);
3299        return 0;
3300}
3301
3302static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len)
3303{
3304        if (skb->ip_summed == CHECKSUM_COMPLETE)
3305                skb->ip_summed = CHECKSUM_NONE;
3306        return __skb_grow(skb, len);
3307}
3308
3309#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode)
3310#define skb_rb_first(root) rb_to_skb(rb_first(root))
3311#define skb_rb_last(root)  rb_to_skb(rb_last(root))
3312#define skb_rb_next(skb)   rb_to_skb(rb_next(&(skb)->rbnode))
3313#define skb_rb_prev(skb)   rb_to_skb(rb_prev(&(skb)->rbnode))
3314
3315#define skb_queue_walk(queue, skb) \
3316                for (skb = (queue)->next;                                       \
3317                     skb != (struct sk_buff *)(queue);                          \
3318                     skb = skb->next)
3319
3320#define skb_queue_walk_safe(queue, skb, tmp)                                    \
3321                for (skb = (queue)->next, tmp = skb->next;                      \
3322                     skb != (struct sk_buff *)(queue);                          \
3323                     skb = tmp, tmp = skb->next)
3324
3325#define skb_queue_walk_from(queue, skb)                                         \
3326                for (; skb != (struct sk_buff *)(queue);                        \
3327                     skb = skb->next)
3328
3329#define skb_rbtree_walk(skb, root)                                              \
3330                for (skb = skb_rb_first(root); skb != NULL;                     \
3331                     skb = skb_rb_next(skb))
3332
3333#define skb_rbtree_walk_from(skb)                                               \
3334                for (; skb != NULL;                                             \
3335                     skb = skb_rb_next(skb))
3336
3337#define skb_rbtree_walk_from_safe(skb, tmp)                                     \
3338                for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL);      \
3339                     skb = tmp)
3340
3341#define skb_queue_walk_from_safe(queue, skb, tmp)                               \
3342                for (tmp = skb->next;                                           \
3343                     skb != (struct sk_buff *)(queue);                          \
3344                     skb = tmp, tmp = skb->next)
3345
3346#define skb_queue_reverse_walk(queue, skb) \
3347                for (skb = (queue)->prev;                                       \
3348                     skb != (struct sk_buff *)(queue);                          \
3349                     skb = skb->prev)
3350
3351#define skb_queue_reverse_walk_safe(queue, skb, tmp)                            \
3352                for (skb = (queue)->prev, tmp = skb->prev;                      \
3353                     skb != (struct sk_buff *)(queue);                          \
3354                     skb = tmp, tmp = skb->prev)
3355
3356#define skb_queue_reverse_walk_from_safe(queue, skb, tmp)                       \
3357                for (tmp = skb->prev;                                           \
3358                     skb != (struct sk_buff *)(queue);                          \
3359                     skb = tmp, tmp = skb->prev)
3360
3361static inline bool skb_has_frag_list(const struct sk_buff *skb)
3362{
3363        return skb_shinfo(skb)->frag_list != NULL;
3364}
3365
3366static inline void skb_frag_list_init(struct sk_buff *skb)
3367{
3368        skb_shinfo(skb)->frag_list = NULL;
3369}
3370
3371#define skb_walk_frags(skb, iter)       \
3372        for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next)
3373
3374
3375int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p,
3376                                const struct sk_buff *skb);
3377struct sk_buff *__skb_try_recv_from_queue(struct sock *sk,
3378                                          struct sk_buff_head *queue,
3379                                          unsigned int flags,
3380                                          void (*destructor)(struct sock *sk,
3381                                                           struct sk_buff *skb),
3382                                          int *off, int *err,
3383                                          struct sk_buff **last);
3384struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags,
3385                                        void (*destructor)(struct sock *sk,
3386                                                           struct sk_buff *skb),
3387                                        int *off, int *err,
3388                                        struct sk_buff **last);
3389struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags,
3390                                    void (*destructor)(struct sock *sk,
3391                                                       struct sk_buff *skb),
3392                                    int *off, int *err);
3393struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock,
3394                                  int *err);
3395__poll_t datagram_poll(struct file *file, struct socket *sock,
3396                           struct poll_table_struct *wait);
3397int skb_copy_datagram_iter(const struct sk_buff *from, int offset,
3398                           struct iov_iter *to, int size);
3399static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset,
3400                                        struct msghdr *msg, int size)
3401{
3402        return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size);
3403}
3404int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen,
3405                                   struct msghdr *msg);
3406int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset,
3407                           struct iov_iter *to, int len,
3408                           struct ahash_request *hash);
3409int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset,
3410                                 struct iov_iter *from, int len);
3411int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm);
3412void skb_free_datagram(struct sock *sk, struct sk_buff *skb);
3413void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len);
3414static inline void skb_free_datagram_locked(struct sock *sk,
3415                                            struct sk_buff *skb)
3416{
3417        __skb_free_datagram_locked(sk, skb, 0);
3418}
3419int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags);
3420int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len);
3421int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len);
3422__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to,
3423                              int len, __wsum csum);
3424int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset,
3425                    struct pipe_inode_info *pipe, unsigned int len,
3426                    unsigned int flags);
3427int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset,
3428                         int len);
3429void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to);
3430unsigned int skb_zerocopy_headlen(const struct sk_buff *from);
3431int skb_zerocopy(struct sk_buff *to, struct sk_buff *from,
3432                 int len, int hlen);
3433void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len);
3434int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen);
3435void skb_scrub_packet(struct sk_buff *skb, bool xnet);
3436bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu);
3437bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len);
3438struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features);
3439struct sk_buff *skb_vlan_untag(struct sk_buff *skb);
3440int skb_ensure_writable(struct sk_buff *skb, int write_len);
3441int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci);
3442int skb_vlan_pop(struct sk_buff *skb);
3443int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci);
3444struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy,
3445                             gfp_t gfp);
3446
3447static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len)
3448{
3449        return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT;
3450}
3451
3452static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len)
3453{
3454        return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT;
3455}
3456
3457struct skb_checksum_ops {
3458        __wsum (*update)(const void *mem, int len, __wsum wsum);
3459        __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len);
3460};
3461
3462extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly;
3463
3464__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len,
3465                      __wsum csum, const struct skb_checksum_ops *ops);
3466__wsum skb_checksum(const struct sk_buff *skb, int offset, int len,
3467                    __wsum csum);
3468
3469static inline void * __must_check
3470__skb_header_pointer(const struct sk_buff *skb, int offset,
3471                     int len, void *data, int hlen, void *buffer)
3472{
3473        if (hlen - offset >= len)
3474                return data + offset;
3475
3476        if (!skb ||
3477            skb_copy_bits(skb, offset, buffer, len) < 0)
3478                return NULL;
3479
3480        return buffer;
3481}
3482
3483static inline void * __must_check
3484skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer)
3485{
3486        return __skb_header_pointer(skb, offset, len, skb->data,
3487                                    skb_headlen(skb), buffer);
3488}
3489
3490/**
3491 *      skb_needs_linearize - check if we need to linearize a given skb
3492 *                            depending on the given device features.
3493 *      @skb: socket buffer to check
3494 *      @features: net device features
3495 *
3496 *      Returns true if either:
3497 *      1. skb has frag_list and the device doesn't support FRAGLIST, or
3498 *      2. skb is fragmented and the device does not support SG.
3499 */
3500static inline bool skb_needs_linearize(struct sk_buff *skb,
3501                                       netdev_features_t features)
3502{
3503        return skb_is_nonlinear(skb) &&
3504               ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) ||
3505                (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG)));
3506}
3507
3508static inline void skb_copy_from_linear_data(const struct sk_buff *skb,
3509                                             void *to,
3510                                             const unsigned int len)
3511{
3512        memcpy(to, skb->data, len);
3513}
3514
3515static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb,
3516                                                    const int offset, void *to,
3517                                                    const unsigned int len)
3518{
3519        memcpy(to, skb->data + offset, len);
3520}
3521
3522static inline void skb_copy_to_linear_data(struct sk_buff *skb,
3523                                           const void *from,
3524                                           const unsigned int len)
3525{
3526        memcpy(skb->data, from, len);
3527}
3528
3529static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb,
3530                                                  const int offset,
3531                                                  const void *from,
3532                                                  const unsigned int len)
3533{
3534        memcpy(skb->data + offset, from, len);
3535}
3536
3537void skb_init(void);
3538
3539static inline ktime_t skb_get_ktime(const struct sk_buff *skb)
3540{
3541        return skb->tstamp;
3542}
3543
3544/**
3545 *      skb_get_timestamp - get timestamp from a skb
3546 *      @skb: skb to get stamp from
3547 *      @stamp: pointer to struct __kernel_old_timeval to store stamp in
3548 *
3549 *      Timestamps are stored in the skb as offsets to a base timestamp.
3550 *      This function converts the offset back to a struct timeval and stores
3551 *      it in stamp.
3552 */
3553static inline void skb_get_timestamp(const struct sk_buff *skb,
3554                                     struct __kernel_old_timeval *stamp)
3555{
3556        *stamp = ns_to_kernel_old_timeval(skb->tstamp);
3557}
3558
3559static inline void skb_get_new_timestamp(const struct sk_buff *skb,
3560                                         struct __kernel_sock_timeval *stamp)
3561{
3562        struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3563
3564        stamp->tv_sec = ts.tv_sec;
3565        stamp->tv_usec = ts.tv_nsec / 1000;
3566}
3567
3568static inline void skb_get_timestampns(const struct sk_buff *skb,
3569                                       struct timespec *stamp)
3570{
3571        *stamp = ktime_to_timespec(skb->tstamp);
3572}
3573
3574static inline void skb_get_new_timestampns(const struct sk_buff *skb,
3575                                           struct __kernel_timespec *stamp)
3576{
3577        struct timespec64 ts = ktime_to_timespec64(skb->tstamp);
3578
3579        stamp->tv_sec = ts.tv_sec;
3580        stamp->tv_nsec = ts.tv_nsec;
3581}
3582
3583static inline void __net_timestamp(struct sk_buff *skb)
3584{
3585        skb->tstamp = ktime_get_real();
3586}
3587
3588static inline ktime_t net_timedelta(ktime_t t)
3589{
3590        return ktime_sub(ktime_get_real(), t);
3591}
3592
3593static inline ktime_t net_invalid_timestamp(void)
3594{
3595        return 0;
3596}
3597
3598static inline u8 skb_metadata_len(const struct sk_buff *skb)
3599{
3600        return skb_shinfo(skb)->meta_len;
3601}
3602
3603static inline void *skb_metadata_end(const struct sk_buff *skb)
3604{
3605        return skb_mac_header(skb);
3606}
3607
3608static inline bool __skb_metadata_differs(const struct sk_buff *skb_a,
3609                                          const struct sk_buff *skb_b,
3610                                          u8 meta_len)
3611{
3612        const void *a = skb_metadata_end(skb_a);
3613        const void *b = skb_metadata_end(skb_b);
3614        /* Using more efficient varaiant than plain call to memcmp(). */
3615#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64
3616        u64 diffs = 0;
3617
3618        switch (meta_len) {
3619#define __it(x, op) (x -= sizeof(u##op))
3620#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op))
3621        case 32: diffs |= __it_diff(a, b, 64);
3622                 /* fall through */
3623        case 24: diffs |= __it_diff(a, b, 64);
3624                 /* fall through */
3625        case 16: diffs |= __it_diff(a, b, 64);
3626                 /* fall through */
3627        case  8: diffs |= __it_diff(a, b, 64);
3628                break;
3629        case 28: diffs |= __it_diff(a, b, 64);
3630                 /* fall through */
3631        case 20: diffs |= __it_diff(a, b, 64);
3632                 /* fall through */
3633        case 12: diffs |= __it_diff(a, b, 64);
3634                 /* fall through */
3635        case  4: diffs |= __it_diff(a, b, 32);
3636                break;
3637        }
3638        return diffs;
3639#else
3640        return memcmp(a - meta_len, b - meta_len, meta_len);
3641#endif
3642}
3643
3644static inline bool skb_metadata_differs(const struct sk_buff *skb_a,
3645                                        const struct sk_buff *skb_b)
3646{
3647        u8 len_a = skb_metadata_len(skb_a);
3648        u8 len_b = skb_metadata_len(skb_b);
3649
3650        if (!(len_a | len_b))
3651                return false;
3652
3653        return len_a != len_b ?
3654               true : __skb_metadata_differs(skb_a, skb_b, len_a);
3655}
3656
3657static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len)
3658{
3659        skb_shinfo(skb)->meta_len = meta_len;
3660}
3661
3662static inline void skb_metadata_clear(struct sk_buff *skb)
3663{
3664        skb_metadata_set(skb, 0);
3665}
3666
3667struct sk_buff *skb_clone_sk(struct sk_buff *skb);
3668
3669#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING
3670
3671void skb_clone_tx_timestamp(struct sk_buff *skb);
3672bool skb_defer_rx_timestamp(struct sk_buff *skb);
3673
3674#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */
3675
3676static inline void skb_clone_tx_timestamp(struct sk_buff *skb)
3677{
3678}
3679
3680static inline bool skb_defer_rx_timestamp(struct sk_buff *skb)
3681{
3682        return false;
3683}
3684
3685#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */
3686
3687/**
3688 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps
3689 *
3690 * PHY drivers may accept clones of transmitted packets for
3691 * timestamping via their phy_driver.txtstamp method. These drivers
3692 * must call this function to return the skb back to the stack with a
3693 * timestamp.
3694 *
3695 * @skb: clone of the the original outgoing packet
3696 * @hwtstamps: hardware time stamps
3697 *
3698 */
3699void skb_complete_tx_timestamp(struct sk_buff *skb,
3700                               struct skb_shared_hwtstamps *hwtstamps);
3701
3702void __skb_tstamp_tx(struct sk_buff *orig_skb,
3703                     struct skb_shared_hwtstamps *hwtstamps,
3704                     struct sock *sk, int tstype);
3705
3706/**
3707 * skb_tstamp_tx - queue clone of skb with send time stamps
3708 * @orig_skb:   the original outgoing packet
3709 * @hwtstamps:  hardware time stamps, may be NULL if not available
3710 *
3711 * If the skb has a socket associated, then this function clones the
3712 * skb (thus sharing the actual data and optional structures), stores
3713 * the optional hardware time stamping information (if non NULL) or
3714 * generates a software time stamp (otherwise), then queues the clone
3715 * to the error queue of the socket.  Errors are silently ignored.
3716 */
3717void skb_tstamp_tx(struct sk_buff *orig_skb,
3718                   struct skb_shared_hwtstamps *hwtstamps);
3719
3720/**
3721 * skb_tx_timestamp() - Driver hook for transmit timestamping
3722 *
3723 * Ethernet MAC Drivers should call this function in their hard_xmit()
3724 * function immediately before giving the sk_buff to the MAC hardware.
3725 *
3726 * Specifically, one should make absolutely sure that this function is
3727 * called before TX completion of this packet can trigger.  Otherwise
3728 * the packet could potentially already be freed.
3729 *
3730 * @skb: A socket buffer.
3731 */
3732static inline void skb_tx_timestamp(struct sk_buff *skb)
3733{
3734        skb_clone_tx_timestamp(skb);
3735        if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP)
3736                skb_tstamp_tx(skb, NULL);
3737}
3738
3739/**
3740 * skb_complete_wifi_ack - deliver skb with wifi status
3741 *
3742 * @skb: the original outgoing packet
3743 * @acked: ack status
3744 *
3745 */
3746void skb_complete_wifi_ack(struct sk_buff *skb, bool acked);
3747
3748__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len);
3749__sum16 __skb_checksum_complete(struct sk_buff *skb);
3750
3751static inline int skb_csum_unnecessary(const struct sk_buff *skb)
3752{
3753        return ((skb->ip_summed == CHECKSUM_UNNECESSARY) ||
3754                skb->csum_valid ||
3755                (skb->ip_summed == CHECKSUM_PARTIAL &&
3756                 skb_checksum_start_offset(skb) >= 0));
3757}
3758
3759/**
3760 *      skb_checksum_complete - Calculate checksum of an entire packet
3761 *      @skb: packet to process
3762 *
3763 *      This function calculates the checksum over the entire packet plus
3764 *      the value of skb->csum.  The latter can be used to supply the
3765 *      checksum of a pseudo header as used by TCP/UDP.  It returns the
3766 *      checksum.
3767 *
3768 *      For protocols that contain complete checksums such as ICMP/TCP/UDP,
3769 *      this function can be used to verify that checksum on received
3770 *      packets.  In that case the function should return zero if the
3771 *      checksum is correct.  In particular, this function will return zero
3772 *      if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the
3773 *      hardware has already verified the correctness of the checksum.
3774 */
3775static inline __sum16 skb_checksum_complete(struct sk_buff *skb)
3776{
3777        return skb_csum_unnecessary(skb) ?
3778               0 : __skb_checksum_complete(skb);
3779}
3780
3781static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb)
3782{
3783        if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3784                if (skb->csum_level == 0)
3785                        skb->ip_summed = CHECKSUM_NONE;
3786                else
3787                        skb->csum_level--;
3788        }
3789}
3790
3791static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb)
3792{
3793        if (skb->ip_summed == CHECKSUM_UNNECESSARY) {
3794                if (skb->csum_level < SKB_MAX_CSUM_LEVEL)
3795                        skb->csum_level++;
3796        } else if (skb->ip_summed == CHECKSUM_NONE) {
3797                skb->ip_summed = CHECKSUM_UNNECESSARY;
3798                skb->csum_level = 0;
3799        }
3800}
3801
3802/* Check if we need to perform checksum complete validation.
3803 *
3804 * Returns true if checksum complete is needed, false otherwise
3805 * (either checksum is unnecessary or zero checksum is allowed).
3806 */
3807static inline bool __skb_checksum_validate_needed(struct sk_buff *skb,
3808                                                  bool zero_okay,
3809                                                  __sum16 check)
3810{
3811        if (skb_csum_unnecessary(skb) || (zero_okay && !check)) {
3812                skb->csum_valid = 1;
3813                __skb_decr_checksum_unnecessary(skb);
3814                return false;
3815        }
3816
3817        return true;
3818}
3819
3820/* For small packets <= CHECKSUM_BREAK perform checksum complete directly
3821 * in checksum_init.
3822 */
3823#define CHECKSUM_BREAK 76
3824
3825/* Unset checksum-complete
3826 *
3827 * Unset checksum complete can be done when packet is being modified
3828 * (uncompressed for instance) and checksum-complete value is
3829 * invalidated.
3830 */
3831static inline void skb_checksum_complete_unset(struct sk_buff *skb)
3832{
3833        if (skb->ip_summed == CHECKSUM_COMPLETE)
3834                skb->ip_summed = CHECKSUM_NONE;
3835}
3836
3837/* Validate (init) checksum based on checksum complete.
3838 *
3839 * Return values:
3840 *   0: checksum is validated or try to in skb_checksum_complete. In the latter
3841 *      case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo
3842 *      checksum is stored in skb->csum for use in __skb_checksum_complete
3843 *   non-zero: value of invalid checksum
3844 *
3845 */
3846static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb,
3847                                                       bool complete,
3848                                                       __wsum psum)
3849{
3850        if (skb->ip_summed == CHECKSUM_COMPLETE) {
3851                if (!csum_fold(csum_add(psum, skb->csum))) {
3852                        skb->csum_valid = 1;
3853                        return 0;
3854                }
3855        }
3856
3857        skb->csum = psum;
3858
3859        if (complete || skb->len <= CHECKSUM_BREAK) {
3860                __sum16 csum;
3861
3862                csum = __skb_checksum_complete(skb);
3863                skb->csum_valid = !csum;
3864                return csum;
3865        }
3866
3867        return 0;
3868}
3869
3870static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto)
3871{
3872        return 0;
3873}
3874
3875/* Perform checksum validate (init). Note that this is a macro since we only
3876 * want to calculate the pseudo header which is an input function if necessary.
3877 * First we try to validate without any computation (checksum unnecessary) and
3878 * then calculate based on checksum complete calling the function to compute
3879 * pseudo header.
3880 *
3881 * Return values:
3882 *   0: checksum is validated or try to in skb_checksum_complete
3883 *   non-zero: value of invalid checksum
3884 */
3885#define __skb_checksum_validate(skb, proto, complete,                   \
3886                                zero_okay, check, compute_pseudo)       \
3887({                                                                      \
3888        __sum16 __ret = 0;                                              \
3889        skb->csum_valid = 0;                                            \
3890        if (__skb_checksum_validate_needed(skb, zero_okay, check))      \
3891                __ret = __skb_checksum_validate_complete(skb,           \
3892                                complete, compute_pseudo(skb, proto));  \
3893        __ret;                                                          \
3894})
3895
3896#define skb_checksum_init(skb, proto, compute_pseudo)                   \
3897        __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo)
3898
3899#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \
3900        __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo)
3901
3902#define skb_checksum_validate(skb, proto, compute_pseudo)               \
3903        __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo)
3904
3905#define skb_checksum_validate_zero_check(skb, proto, check,             \
3906                                         compute_pseudo)                \
3907        __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo)
3908
3909#define skb_checksum_simple_validate(skb)                               \
3910        __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo)
3911
3912static inline bool __skb_checksum_convert_check(struct sk_buff *skb)
3913{
3914        return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid);
3915}
3916
3917static inline void __skb_checksum_convert(struct sk_buff *skb,
3918                                          __sum16 check, __wsum pseudo)
3919{
3920        skb->csum = ~pseudo;
3921        skb->ip_summed = CHECKSUM_COMPLETE;
3922}
3923
3924#define skb_checksum_try_convert(skb, proto, check, compute_pseudo)     \
3925do {                                                                    \
3926        if (__skb_checksum_convert_check(skb))                          \
3927                __skb_checksum_convert(skb, check,                      \
3928                                       compute_pseudo(skb, proto));     \
3929} while (0)
3930
3931static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr,
3932                                              u16 start, u16 offset)
3933{
3934        skb->ip_summed = CHECKSUM_PARTIAL;
3935        skb->csum_start = ((unsigned char *)ptr + start) - skb->head;
3936        skb->csum_offset = offset - start;
3937}
3938
3939/* Update skbuf and packet to reflect the remote checksum offload operation.
3940 * When called, ptr indicates the starting point for skb->csum when
3941 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete
3942 * here, skb_postpull_rcsum is done so skb->csum start is ptr.
3943 */
3944static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr,
3945                                       int start, int offset, bool nopartial)
3946{
3947        __wsum delta;
3948
3949        if (!nopartial) {
3950                skb_remcsum_adjust_partial(skb, ptr, start, offset);
3951                return;
3952        }
3953
3954         if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) {
3955                __skb_checksum_complete(skb);
3956                skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data);
3957        }
3958
3959        delta = remcsum_adjust(ptr, skb->csum, start, offset);
3960
3961        /* Adjust skb->csum since we changed the packet */
3962        skb->csum = csum_add(skb->csum, delta);
3963}
3964
3965static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb)
3966{
3967#if IS_ENABLED(CONFIG_NF_CONNTRACK)
3968        return (void *)(skb->_nfct & SKB_NFCT_PTRMASK);
3969#else
3970        return NULL;
3971#endif
3972}
3973
3974#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
3975void nf_conntrack_destroy(struct nf_conntrack *nfct);
3976static inline void nf_conntrack_put(struct nf_conntrack *nfct)
3977{
3978        if (nfct && atomic_dec_and_test(&nfct->use))
3979                nf_conntrack_destroy(nfct);
3980}
3981static inline void nf_conntrack_get(struct nf_conntrack *nfct)
3982{
3983        if (nfct)
3984                atomic_inc(&nfct->use);
3985}
3986#endif
3987
3988#ifdef CONFIG_SKB_EXTENSIONS
3989enum skb_ext_id {
3990#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
3991        SKB_EXT_BRIDGE_NF,
3992#endif
3993#ifdef CONFIG_XFRM
3994        SKB_EXT_SEC_PATH,
3995#endif
3996        SKB_EXT_NUM, /* must be last */
3997};
3998
3999/**
4000 *      struct skb_ext - sk_buff extensions
4001 *      @refcnt: 1 on allocation, deallocated on 0
4002 *      @offset: offset to add to @data to obtain extension address
4003 *      @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units
4004 *      @data: start of extension data, variable sized
4005 *
4006 *      Note: offsets/lengths are stored in chunks of 8 bytes, this allows
4007 *      to use 'u8' types while allowing up to 2kb worth of extension data.
4008 */
4009struct skb_ext {
4010        refcount_t refcnt;
4011        u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */
4012        u8 chunks;              /* same */
4013        char data[0] __aligned(8);
4014};
4015
4016void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id);
4017void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id);
4018void __skb_ext_put(struct skb_ext *ext);
4019
4020static inline void skb_ext_put(struct sk_buff *skb)
4021{
4022        if (skb->active_extensions)
4023                __skb_ext_put(skb->extensions);
4024}
4025
4026static inline void __skb_ext_copy(struct sk_buff *dst,
4027                                  const struct sk_buff *src)
4028{
4029        dst->active_extensions = src->active_extensions;
4030
4031        if (src->active_extensions) {
4032                struct skb_ext *ext = src->extensions;
4033
4034                refcount_inc(&ext->refcnt);
4035                dst->extensions = ext;
4036        }
4037}
4038
4039static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src)
4040{
4041        skb_ext_put(dst);
4042        __skb_ext_copy(dst, src);
4043}
4044
4045static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i)
4046{
4047        return !!ext->offset[i];
4048}
4049
4050static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id)
4051{
4052        return skb->active_extensions & (1 << id);
4053}
4054
4055static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id)
4056{
4057        if (skb_ext_exist(skb, id))
4058                __skb_ext_del(skb, id);
4059}
4060
4061static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id)
4062{
4063        if (skb_ext_exist(skb, id)) {
4064                struct skb_ext *ext = skb->extensions;
4065
4066                return (void *)ext + (ext->offset[id] << 3);
4067        }
4068
4069        return NULL;
4070}
4071#else
4072static inline void skb_ext_put(struct sk_buff *skb) {}
4073static inline void skb_ext_del(struct sk_buff *skb, int unused) {}
4074static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {}
4075static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {}
4076#endif /* CONFIG_SKB_EXTENSIONS */
4077
4078static inline void nf_reset(struct sk_buff *skb)
4079{
4080#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4081        nf_conntrack_put(skb_nfct(skb));
4082        skb->_nfct = 0;
4083#endif
4084#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER)
4085        skb_ext_del(skb, SKB_EXT_BRIDGE_NF);
4086#endif
4087}
4088
4089static inline void nf_reset_trace(struct sk_buff *skb)
4090{
4091#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4092        skb->nf_trace = 0;
4093#endif
4094}
4095
4096static inline void ipvs_reset(struct sk_buff *skb)
4097{
4098#if IS_ENABLED(CONFIG_IP_VS)
4099        skb->ipvs_property = 0;
4100#endif
4101}
4102
4103/* Note: This doesn't put any conntrack info in dst. */
4104static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src,
4105                             bool copy)
4106{
4107#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4108        dst->_nfct = src->_nfct;
4109        nf_conntrack_get(skb_nfct(src));
4110#endif
4111#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES)
4112        if (copy)
4113                dst->nf_trace = src->nf_trace;
4114#endif
4115}
4116
4117static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src)
4118{
4119#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE)
4120        nf_conntrack_put(skb_nfct(dst));
4121#endif
4122        __nf_copy(dst, src, true);
4123}
4124
4125#ifdef CONFIG_NETWORK_SECMARK
4126static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4127{
4128        to->secmark = from->secmark;
4129}
4130
4131static inline void skb_init_secmark(struct sk_buff *skb)
4132{
4133        skb->secmark = 0;
4134}
4135#else
4136static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from)
4137{ }
4138
4139static inline void skb_init_secmark(struct sk_buff *skb)
4140{ }
4141#endif
4142
4143static inline int secpath_exists(const struct sk_buff *skb)
4144{
4145#ifdef CONFIG_XFRM
4146        return skb_ext_exist(skb, SKB_EXT_SEC_PATH);
4147#else
4148        return 0;
4149#endif
4150}
4151
4152static inline bool skb_irq_freeable(const struct sk_buff *skb)
4153{
4154        return !skb->destructor &&
4155                !secpath_exists(skb) &&
4156                !skb_nfct(skb) &&
4157                !skb->_skb_refdst &&
4158                !skb_has_frag_list(skb);
4159}
4160
4161static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping)
4162{
4163        skb->queue_mapping = queue_mapping;
4164}
4165
4166static inline u16 skb_get_queue_mapping(const struct sk_buff *skb)
4167{
4168        return skb->queue_mapping;
4169}
4170
4171static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from)
4172{
4173        to->queue_mapping = from->queue_mapping;
4174}
4175
4176static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue)
4177{
4178        skb->queue_mapping = rx_queue + 1;
4179}
4180
4181static inline u16 skb_get_rx_queue(const struct sk_buff *skb)
4182{
4183        return skb->queue_mapping - 1;
4184}
4185
4186static inline bool skb_rx_queue_recorded(const struct sk_buff *skb)
4187{
4188        return skb->queue_mapping != 0;
4189}
4190
4191static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val)
4192{
4193        skb->dst_pending_confirm = val;
4194}
4195
4196static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb)
4197{
4198        return skb->dst_pending_confirm != 0;
4199}
4200
4201static inline struct sec_path *skb_sec_path(const struct sk_buff *skb)
4202{
4203#ifdef CONFIG_XFRM
4204        return skb_ext_find(skb, SKB_EXT_SEC_PATH);
4205#else
4206        return NULL;
4207#endif
4208}
4209
4210/* Keeps track of mac header offset relative to skb->head.
4211 * It is useful for TSO of Tunneling protocol. e.g. GRE.
4212 * For non-tunnel skb it points to skb_mac_header() and for
4213 * tunnel skb it points to outer mac header.
4214 * Keeps track of level of encapsulation of network headers.
4215 */
4216struct skb_gso_cb {
4217        union {
4218                int     mac_offset;
4219                int     data_offset;
4220        };
4221        int     encap_level;
4222        __wsum  csum;
4223        __u16   csum_start;
4224};
4225#define SKB_SGO_CB_OFFSET       32
4226#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET))
4227
4228static inline int skb_tnl_header_len(const struct sk_buff *inner_skb)
4229{
4230        return (skb_mac_header(inner_skb) - inner_skb->head) -
4231                SKB_GSO_CB(inner_skb)->mac_offset;
4232}
4233
4234static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra)
4235{
4236        int new_headroom, headroom;
4237        int ret;
4238
4239        headroom = skb_headroom(skb);
4240        ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC);
4241        if (ret)
4242                return ret;
4243
4244        new_headroom = skb_headroom(skb);
4245        SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom);
4246        return 0;
4247}
4248
4249static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res)
4250{
4251        /* Do not update partial checksums if remote checksum is enabled. */
4252        if (skb->remcsum_offload)
4253                return;
4254
4255        SKB_GSO_CB(skb)->csum = res;
4256        SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head;
4257}
4258
4259/* Compute the checksum for a gso segment. First compute the checksum value
4260 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and
4261 * then add in skb->csum (checksum from csum_start to end of packet).
4262 * skb->csum and csum_start are then updated to reflect the checksum of the
4263 * resultant packet starting from the transport header-- the resultant checksum
4264 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo
4265 * header.
4266 */
4267static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res)
4268{
4269        unsigned char *csum_start = skb_transport_header(skb);
4270        int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start;
4271        __wsum partial = SKB_GSO_CB(skb)->csum;
4272
4273        SKB_GSO_CB(skb)->csum = res;
4274        SKB_GSO_CB(skb)->csum_start = csum_start - skb->head;
4275
4276        return csum_fold(csum_partial(csum_start, plen, partial));
4277}
4278
4279static inline bool skb_is_gso(const struct sk_buff *skb)
4280{
4281        return skb_shinfo(skb)->gso_size;
4282}
4283
4284/* Note: Should be called only if skb_is_gso(skb) is true */
4285static inline bool skb_is_gso_v6(const struct sk_buff *skb)
4286{
4287        return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6;
4288}
4289
4290/* Note: Should be called only if skb_is_gso(skb) is true */
4291static inline bool skb_is_gso_sctp(const struct sk_buff *skb)
4292{
4293        return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP;
4294}
4295
4296/* Note: Should be called only if skb_is_gso(skb) is true */
4297static inline bool skb_is_gso_tcp(const struct sk_buff *skb)
4298{
4299        return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6);
4300}
4301
4302static inline void skb_gso_reset(struct sk_buff *skb)
4303{
4304        skb_shinfo(skb)->gso_size = 0;
4305        skb_shinfo(skb)->gso_segs = 0;
4306        skb_shinfo(skb)->gso_type = 0;
4307}
4308
4309static inline void skb_increase_gso_size(struct skb_shared_info *shinfo,
4310                                         u16 increment)
4311{
4312        if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4313                return;
4314        shinfo->gso_size += increment;
4315}
4316
4317static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo,
4318                                         u16 decrement)
4319{
4320        if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS))
4321                return;
4322        shinfo->gso_size -= decrement;
4323}
4324
4325void __skb_warn_lro_forwarding(const struct sk_buff *skb);
4326
4327static inline bool skb_warn_if_lro(const struct sk_buff *skb)
4328{
4329        /* LRO sets gso_size but not gso_type, whereas if GSO is really
4330         * wanted then gso_type will be set. */
4331        const struct skb_shared_info *shinfo = skb_shinfo(skb);
4332
4333        if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 &&
4334            unlikely(shinfo->gso_type == 0)) {
4335                __skb_warn_lro_forwarding(skb);
4336                return true;
4337        }
4338        return false;
4339}
4340
4341static inline void skb_forward_csum(struct sk_buff *skb)
4342{
4343        /* Unfortunately we don't support this one.  Any brave souls? */
4344        if (skb->ip_summed == CHECKSUM_COMPLETE)
4345                skb->ip_summed = CHECKSUM_NONE;
4346}
4347
4348/**
4349 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE
4350 * @skb: skb to check
4351 *
4352 * fresh skbs have their ip_summed set to CHECKSUM_NONE.
4353 * Instead of forcing ip_summed to CHECKSUM_NONE, we can
4354 * use this helper, to document places where we make this assertion.
4355 */
4356static inline void skb_checksum_none_assert(const struct sk_buff *skb)
4357{
4358#ifdef DEBUG
4359        BUG_ON(skb->ip_summed != CHECKSUM_NONE);
4360#endif
4361}
4362
4363bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off);
4364
4365int skb_checksum_setup(struct sk_buff *skb, bool recalculate);
4366struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb,
4367                                     unsigned int transport_len,
4368                                     __sum16(*skb_chkf)(struct sk_buff *skb));
4369
4370/**
4371 * skb_head_is_locked - Determine if the skb->head is locked down
4372 * @skb: skb to check
4373 *
4374 * The head on skbs build around a head frag can be removed if they are
4375 * not cloned.  This function returns true if the skb head is locked down
4376 * due to either being allocated via kmalloc, or by being a clone with
4377 * multiple references to the head.
4378 */
4379static inline bool skb_head_is_locked(const struct sk_buff *skb)
4380{
4381        return !skb->head_frag || skb_cloned(skb);
4382}
4383
4384/* Local Checksum Offload.
4385 * Compute outer checksum based on the assumption that the
4386 * inner checksum will be offloaded later.
4387 * See Documentation/networking/checksum-offloads.rst for
4388 * explanation of how this works.
4389 * Fill in outer checksum adjustment (e.g. with sum of outer
4390 * pseudo-header) before calling.
4391 * Also ensure that inner checksum is in linear data area.
4392 */
4393static inline __wsum lco_csum(struct sk_buff *skb)
4394{
4395        unsigned char *csum_start = skb_checksum_start(skb);
4396        unsigned char *l4_hdr = skb_transport_header(skb);
4397        __wsum partial;
4398
4399        /* Start with complement of inner checksum adjustment */
4400        partial = ~csum_unfold(*(__force __sum16 *)(csum_start +
4401                                                    skb->csum_offset));
4402
4403        /* Add in checksum of our headers (incl. outer checksum
4404         * adjustment filled in by caller) and return result.
4405         */
4406        return csum_partial(l4_hdr, csum_start - l4_hdr, partial);
4407}
4408
4409#endif  /* __KERNEL__ */
4410#endif  /* _LINUX_SKBUFF_H */
4411