linux/net/core/skbuff.c
<<
>>
Prefs
   1/*
   2 *      Routines having to do with the 'struct sk_buff' memory handlers.
   3 *
   4 *      Authors:        Alan Cox <alan@lxorguk.ukuu.org.uk>
   5 *                      Florian La Roche <rzsfl@rz.uni-sb.de>
   6 *
   7 *      Fixes:
   8 *              Alan Cox        :       Fixed the worst of the load
   9 *                                      balancer bugs.
  10 *              Dave Platt      :       Interrupt stacking fix.
  11 *      Richard Kooijman        :       Timestamp fixes.
  12 *              Alan Cox        :       Changed buffer format.
  13 *              Alan Cox        :       destructor hook for AF_UNIX etc.
  14 *              Linus Torvalds  :       Better skb_clone.
  15 *              Alan Cox        :       Added skb_copy.
  16 *              Alan Cox        :       Added all the changed routines Linus
  17 *                                      only put in the headers
  18 *              Ray VanTassle   :       Fixed --skb->lock in free
  19 *              Alan Cox        :       skb_copy copy arp field
  20 *              Andi Kleen      :       slabified it.
  21 *              Robert Olsson   :       Removed skb_head_pool
  22 *
  23 *      NOTE:
  24 *              The __skb_ routines should be called with interrupts
  25 *      disabled, or you better be *real* sure that the operation is atomic
  26 *      with respect to whatever list is being frobbed (e.g. via lock_sock()
  27 *      or via disabling bottom half handlers, etc).
  28 *
  29 *      This program is free software; you can redistribute it and/or
  30 *      modify it under the terms of the GNU General Public License
  31 *      as published by the Free Software Foundation; either version
  32 *      2 of the License, or (at your option) any later version.
  33 */
  34
  35/*
  36 *      The functions in this file will not compile correctly with gcc 2.4.x
  37 */
  38
  39#define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
  40
  41#include <linux/module.h>
  42#include <linux/types.h>
  43#include <linux/kernel.h>
  44#include <linux/kmemcheck.h>
  45#include <linux/mm.h>
  46#include <linux/interrupt.h>
  47#include <linux/in.h>
  48#include <linux/inet.h>
  49#include <linux/slab.h>
  50#include <linux/netdevice.h>
  51#ifdef CONFIG_NET_CLS_ACT
  52#include <net/pkt_sched.h>
  53#endif
  54#include <linux/string.h>
  55#include <linux/skbuff.h>
  56#include <linux/splice.h>
  57#include <linux/cache.h>
  58#include <linux/rtnetlink.h>
  59#include <linux/init.h>
  60#include <linux/scatterlist.h>
  61#include <linux/errqueue.h>
  62#include <linux/prefetch.h>
  63
  64#include <net/protocol.h>
  65#include <net/dst.h>
  66#include <net/sock.h>
  67#include <net/checksum.h>
  68#include <net/xfrm.h>
  69
  70#include <asm/uaccess.h>
  71#include <trace/events/skb.h>
  72#include <linux/highmem.h>
  73
  74struct kmem_cache *skbuff_head_cache __read_mostly;
  75static struct kmem_cache *skbuff_fclone_cache __read_mostly;
  76
  77static void sock_pipe_buf_release(struct pipe_inode_info *pipe,
  78                                  struct pipe_buffer *buf)
  79{
  80        put_page(buf->page);
  81}
  82
  83static void sock_pipe_buf_get(struct pipe_inode_info *pipe,
  84                                struct pipe_buffer *buf)
  85{
  86        get_page(buf->page);
  87}
  88
  89static int sock_pipe_buf_steal(struct pipe_inode_info *pipe,
  90                               struct pipe_buffer *buf)
  91{
  92        return 1;
  93}
  94
  95
  96/* Pipe buffer operations for a socket. */
  97static const struct pipe_buf_operations sock_pipe_buf_ops = {
  98        .can_merge = 0,
  99        .map = generic_pipe_buf_map,
 100        .unmap = generic_pipe_buf_unmap,
 101        .confirm = generic_pipe_buf_confirm,
 102        .release = sock_pipe_buf_release,
 103        .steal = sock_pipe_buf_steal,
 104        .get = sock_pipe_buf_get,
 105};
 106
 107/*
 108 *      Keep out-of-line to prevent kernel bloat.
 109 *      __builtin_return_address is not used because it is not always
 110 *      reliable.
 111 */
 112
 113/**
 114 *      skb_over_panic  -       private function
 115 *      @skb: buffer
 116 *      @sz: size
 117 *      @here: address
 118 *
 119 *      Out of line support code for skb_put(). Not user callable.
 120 */
 121static void skb_over_panic(struct sk_buff *skb, int sz, void *here)
 122{
 123        pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
 124                 __func__, here, skb->len, sz, skb->head, skb->data,
 125                 (unsigned long)skb->tail, (unsigned long)skb->end,
 126                 skb->dev ? skb->dev->name : "<NULL>");
 127        BUG();
 128}
 129
 130/**
 131 *      skb_under_panic -       private function
 132 *      @skb: buffer
 133 *      @sz: size
 134 *      @here: address
 135 *
 136 *      Out of line support code for skb_push(). Not user callable.
 137 */
 138
 139static void skb_under_panic(struct sk_buff *skb, int sz, void *here)
 140{
 141        pr_emerg("%s: text:%p len:%d put:%d head:%p data:%p tail:%#lx end:%#lx dev:%s\n",
 142                 __func__, here, skb->len, sz, skb->head, skb->data,
 143                 (unsigned long)skb->tail, (unsigned long)skb->end,
 144                 skb->dev ? skb->dev->name : "<NULL>");
 145        BUG();
 146}
 147
 148
 149/*
 150 * kmalloc_reserve is a wrapper around kmalloc_node_track_caller that tells
 151 * the caller if emergency pfmemalloc reserves are being used. If it is and
 152 * the socket is later found to be SOCK_MEMALLOC then PFMEMALLOC reserves
 153 * may be used. Otherwise, the packet data may be discarded until enough
 154 * memory is free
 155 */
 156#define kmalloc_reserve(size, gfp, node, pfmemalloc) \
 157         __kmalloc_reserve(size, gfp, node, _RET_IP_, pfmemalloc)
 158void *__kmalloc_reserve(size_t size, gfp_t flags, int node, unsigned long ip,
 159                         bool *pfmemalloc)
 160{
 161        void *obj;
 162        bool ret_pfmemalloc = false;
 163
 164        /*
 165         * Try a regular allocation, when that fails and we're not entitled
 166         * to the reserves, fail.
 167         */
 168        obj = kmalloc_node_track_caller(size,
 169                                        flags | __GFP_NOMEMALLOC | __GFP_NOWARN,
 170                                        node);
 171        if (obj || !(gfp_pfmemalloc_allowed(flags)))
 172                goto out;
 173
 174        /* Try again but now we are using pfmemalloc reserves */
 175        ret_pfmemalloc = true;
 176        obj = kmalloc_node_track_caller(size, flags, node);
 177
 178out:
 179        if (pfmemalloc)
 180                *pfmemalloc = ret_pfmemalloc;
 181
 182        return obj;
 183}
 184
 185/*      Allocate a new skbuff. We do this ourselves so we can fill in a few
 186 *      'private' fields and also do memory statistics to find all the
 187 *      [BEEP] leaks.
 188 *
 189 */
 190
 191/**
 192 *      __alloc_skb     -       allocate a network buffer
 193 *      @size: size to allocate
 194 *      @gfp_mask: allocation mask
 195 *      @flags: If SKB_ALLOC_FCLONE is set, allocate from fclone cache
 196 *              instead of head cache and allocate a cloned (child) skb.
 197 *              If SKB_ALLOC_RX is set, __GFP_MEMALLOC will be used for
 198 *              allocations in case the data is required for writeback
 199 *      @node: numa node to allocate memory on
 200 *
 201 *      Allocate a new &sk_buff. The returned buffer has no headroom and a
 202 *      tail room of at least size bytes. The object has a reference count
 203 *      of one. The return is the buffer. On a failure the return is %NULL.
 204 *
 205 *      Buffers may only be allocated from interrupts using a @gfp_mask of
 206 *      %GFP_ATOMIC.
 207 */
 208struct sk_buff *__alloc_skb(unsigned int size, gfp_t gfp_mask,
 209                            int flags, int node)
 210{
 211        struct kmem_cache *cache;
 212        struct skb_shared_info *shinfo;
 213        struct sk_buff *skb;
 214        u8 *data;
 215        bool pfmemalloc;
 216
 217        cache = (flags & SKB_ALLOC_FCLONE)
 218                ? skbuff_fclone_cache : skbuff_head_cache;
 219
 220        if (sk_memalloc_socks() && (flags & SKB_ALLOC_RX))
 221                gfp_mask |= __GFP_MEMALLOC;
 222
 223        /* Get the HEAD */
 224        skb = kmem_cache_alloc_node(cache, gfp_mask & ~__GFP_DMA, node);
 225        if (!skb)
 226                goto out;
 227        prefetchw(skb);
 228
 229        /* We do our best to align skb_shared_info on a separate cache
 230         * line. It usually works because kmalloc(X > SMP_CACHE_BYTES) gives
 231         * aligned memory blocks, unless SLUB/SLAB debug is enabled.
 232         * Both skb->head and skb_shared_info are cache line aligned.
 233         */
 234        size = SKB_DATA_ALIGN(size);
 235        size += SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 236        data = kmalloc_reserve(size, gfp_mask, node, &pfmemalloc);
 237        if (!data)
 238                goto nodata;
 239        /* kmalloc(size) might give us more room than requested.
 240         * Put skb_shared_info exactly at the end of allocated zone,
 241         * to allow max possible filling before reallocation.
 242         */
 243        size = SKB_WITH_OVERHEAD(ksize(data));
 244        prefetchw(data + size);
 245
 246        /*
 247         * Only clear those fields we need to clear, not those that we will
 248         * actually initialise below. Hence, don't put any more fields after
 249         * the tail pointer in struct sk_buff!
 250         */
 251        memset(skb, 0, offsetof(struct sk_buff, tail));
 252        /* Account for allocated memory : skb + skb->head */
 253        skb->truesize = SKB_TRUESIZE(size);
 254        skb->pfmemalloc = pfmemalloc;
 255        atomic_set(&skb->users, 1);
 256        skb->head = data;
 257        skb->data = data;
 258        skb_reset_tail_pointer(skb);
 259        skb->end = skb->tail + size;
 260#ifdef NET_SKBUFF_DATA_USES_OFFSET
 261        skb->mac_header = ~0U;
 262#endif
 263
 264        /* make sure we initialize shinfo sequentially */
 265        shinfo = skb_shinfo(skb);
 266        memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 267        atomic_set(&shinfo->dataref, 1);
 268        kmemcheck_annotate_variable(shinfo->destructor_arg);
 269
 270        if (flags & SKB_ALLOC_FCLONE) {
 271                struct sk_buff *child = skb + 1;
 272                atomic_t *fclone_ref = (atomic_t *) (child + 1);
 273
 274                kmemcheck_annotate_bitfield(child, flags1);
 275                kmemcheck_annotate_bitfield(child, flags2);
 276                skb->fclone = SKB_FCLONE_ORIG;
 277                atomic_set(fclone_ref, 1);
 278
 279                child->fclone = SKB_FCLONE_UNAVAILABLE;
 280                child->pfmemalloc = pfmemalloc;
 281        }
 282out:
 283        return skb;
 284nodata:
 285        kmem_cache_free(cache, skb);
 286        skb = NULL;
 287        goto out;
 288}
 289EXPORT_SYMBOL(__alloc_skb);
 290
 291/**
 292 * build_skb - build a network buffer
 293 * @data: data buffer provided by caller
 294 * @frag_size: size of fragment, or 0 if head was kmalloced
 295 *
 296 * Allocate a new &sk_buff. Caller provides space holding head and
 297 * skb_shared_info. @data must have been allocated by kmalloc()
 298 * The return is the new skb buffer.
 299 * On a failure the return is %NULL, and @data is not freed.
 300 * Notes :
 301 *  Before IO, driver allocates only data buffer where NIC put incoming frame
 302 *  Driver should add room at head (NET_SKB_PAD) and
 303 *  MUST add room at tail (SKB_DATA_ALIGN(skb_shared_info))
 304 *  After IO, driver calls build_skb(), to allocate sk_buff and populate it
 305 *  before giving packet to stack.
 306 *  RX rings only contains data buffers, not full skbs.
 307 */
 308struct sk_buff *build_skb(void *data, unsigned int frag_size)
 309{
 310        struct skb_shared_info *shinfo;
 311        struct sk_buff *skb;
 312        unsigned int size = frag_size ? : ksize(data);
 313
 314        skb = kmem_cache_alloc(skbuff_head_cache, GFP_ATOMIC);
 315        if (!skb)
 316                return NULL;
 317
 318        size -= SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 319
 320        memset(skb, 0, offsetof(struct sk_buff, tail));
 321        skb->truesize = SKB_TRUESIZE(size);
 322        skb->head_frag = frag_size != 0;
 323        atomic_set(&skb->users, 1);
 324        skb->head = data;
 325        skb->data = data;
 326        skb_reset_tail_pointer(skb);
 327        skb->end = skb->tail + size;
 328#ifdef NET_SKBUFF_DATA_USES_OFFSET
 329        skb->mac_header = ~0U;
 330#endif
 331
 332        /* make sure we initialize shinfo sequentially */
 333        shinfo = skb_shinfo(skb);
 334        memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 335        atomic_set(&shinfo->dataref, 1);
 336        kmemcheck_annotate_variable(shinfo->destructor_arg);
 337
 338        return skb;
 339}
 340EXPORT_SYMBOL(build_skb);
 341
 342struct netdev_alloc_cache {
 343        struct page *page;
 344        unsigned int offset;
 345        unsigned int pagecnt_bias;
 346};
 347static DEFINE_PER_CPU(struct netdev_alloc_cache, netdev_alloc_cache);
 348
 349#define NETDEV_PAGECNT_BIAS (PAGE_SIZE / SMP_CACHE_BYTES)
 350
 351static void *__netdev_alloc_frag(unsigned int fragsz, gfp_t gfp_mask)
 352{
 353        struct netdev_alloc_cache *nc;
 354        void *data = NULL;
 355        unsigned long flags;
 356
 357        local_irq_save(flags);
 358        nc = &__get_cpu_var(netdev_alloc_cache);
 359        if (unlikely(!nc->page)) {
 360refill:
 361                nc->page = alloc_page(gfp_mask);
 362                if (unlikely(!nc->page))
 363                        goto end;
 364recycle:
 365                atomic_set(&nc->page->_count, NETDEV_PAGECNT_BIAS);
 366                nc->pagecnt_bias = NETDEV_PAGECNT_BIAS;
 367                nc->offset = 0;
 368        }
 369
 370        if (nc->offset + fragsz > PAGE_SIZE) {
 371                /* avoid unnecessary locked operations if possible */
 372                if ((atomic_read(&nc->page->_count) == nc->pagecnt_bias) ||
 373                    atomic_sub_and_test(nc->pagecnt_bias, &nc->page->_count))
 374                        goto recycle;
 375                goto refill;
 376        }
 377
 378        data = page_address(nc->page) + nc->offset;
 379        nc->offset += fragsz;
 380        nc->pagecnt_bias--;
 381end:
 382        local_irq_restore(flags);
 383        return data;
 384}
 385
 386/**
 387 * netdev_alloc_frag - allocate a page fragment
 388 * @fragsz: fragment size
 389 *
 390 * Allocates a frag from a page for receive buffer.
 391 * Uses GFP_ATOMIC allocations.
 392 */
 393void *netdev_alloc_frag(unsigned int fragsz)
 394{
 395        return __netdev_alloc_frag(fragsz, GFP_ATOMIC | __GFP_COLD);
 396}
 397EXPORT_SYMBOL(netdev_alloc_frag);
 398
 399/**
 400 *      __netdev_alloc_skb - allocate an skbuff for rx on a specific device
 401 *      @dev: network device to receive on
 402 *      @length: length to allocate
 403 *      @gfp_mask: get_free_pages mask, passed to alloc_skb
 404 *
 405 *      Allocate a new &sk_buff and assign it a usage count of one. The
 406 *      buffer has unspecified headroom built in. Users should allocate
 407 *      the headroom they think they need without accounting for the
 408 *      built in space. The built in space is used for optimisations.
 409 *
 410 *      %NULL is returned if there is no free memory.
 411 */
 412struct sk_buff *__netdev_alloc_skb(struct net_device *dev,
 413                                   unsigned int length, gfp_t gfp_mask)
 414{
 415        struct sk_buff *skb = NULL;
 416        unsigned int fragsz = SKB_DATA_ALIGN(length + NET_SKB_PAD) +
 417                              SKB_DATA_ALIGN(sizeof(struct skb_shared_info));
 418
 419        if (fragsz <= PAGE_SIZE && !(gfp_mask & (__GFP_WAIT | GFP_DMA))) {
 420                void *data;
 421
 422                if (sk_memalloc_socks())
 423                        gfp_mask |= __GFP_MEMALLOC;
 424
 425                data = __netdev_alloc_frag(fragsz, gfp_mask);
 426
 427                if (likely(data)) {
 428                        skb = build_skb(data, fragsz);
 429                        if (unlikely(!skb))
 430                                put_page(virt_to_head_page(data));
 431                }
 432        } else {
 433                skb = __alloc_skb(length + NET_SKB_PAD, gfp_mask,
 434                                  SKB_ALLOC_RX, NUMA_NO_NODE);
 435        }
 436        if (likely(skb)) {
 437                skb_reserve(skb, NET_SKB_PAD);
 438                skb->dev = dev;
 439        }
 440        return skb;
 441}
 442EXPORT_SYMBOL(__netdev_alloc_skb);
 443
 444void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off,
 445                     int size, unsigned int truesize)
 446{
 447        skb_fill_page_desc(skb, i, page, off, size);
 448        skb->len += size;
 449        skb->data_len += size;
 450        skb->truesize += truesize;
 451}
 452EXPORT_SYMBOL(skb_add_rx_frag);
 453
 454static void skb_drop_list(struct sk_buff **listp)
 455{
 456        struct sk_buff *list = *listp;
 457
 458        *listp = NULL;
 459
 460        do {
 461                struct sk_buff *this = list;
 462                list = list->next;
 463                kfree_skb(this);
 464        } while (list);
 465}
 466
 467static inline void skb_drop_fraglist(struct sk_buff *skb)
 468{
 469        skb_drop_list(&skb_shinfo(skb)->frag_list);
 470}
 471
 472static void skb_clone_fraglist(struct sk_buff *skb)
 473{
 474        struct sk_buff *list;
 475
 476        skb_walk_frags(skb, list)
 477                skb_get(list);
 478}
 479
 480static void skb_free_head(struct sk_buff *skb)
 481{
 482        if (skb->head_frag)
 483                put_page(virt_to_head_page(skb->head));
 484        else
 485                kfree(skb->head);
 486}
 487
 488static void skb_release_data(struct sk_buff *skb)
 489{
 490        if (!skb->cloned ||
 491            !atomic_sub_return(skb->nohdr ? (1 << SKB_DATAREF_SHIFT) + 1 : 1,
 492                               &skb_shinfo(skb)->dataref)) {
 493                if (skb_shinfo(skb)->nr_frags) {
 494                        int i;
 495                        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
 496                                skb_frag_unref(skb, i);
 497                }
 498
 499                /*
 500                 * If skb buf is from userspace, we need to notify the caller
 501                 * the lower device DMA has done;
 502                 */
 503                if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) {
 504                        struct ubuf_info *uarg;
 505
 506                        uarg = skb_shinfo(skb)->destructor_arg;
 507                        if (uarg->callback)
 508                                uarg->callback(uarg);
 509                }
 510
 511                if (skb_has_frag_list(skb))
 512                        skb_drop_fraglist(skb);
 513
 514                skb_free_head(skb);
 515        }
 516}
 517
 518/*
 519 *      Free an skbuff by memory without cleaning the state.
 520 */
 521static void kfree_skbmem(struct sk_buff *skb)
 522{
 523        struct sk_buff *other;
 524        atomic_t *fclone_ref;
 525
 526        switch (skb->fclone) {
 527        case SKB_FCLONE_UNAVAILABLE:
 528                kmem_cache_free(skbuff_head_cache, skb);
 529                break;
 530
 531        case SKB_FCLONE_ORIG:
 532                fclone_ref = (atomic_t *) (skb + 2);
 533                if (atomic_dec_and_test(fclone_ref))
 534                        kmem_cache_free(skbuff_fclone_cache, skb);
 535                break;
 536
 537        case SKB_FCLONE_CLONE:
 538                fclone_ref = (atomic_t *) (skb + 1);
 539                other = skb - 1;
 540
 541                /* The clone portion is available for
 542                 * fast-cloning again.
 543                 */
 544                skb->fclone = SKB_FCLONE_UNAVAILABLE;
 545
 546                if (atomic_dec_and_test(fclone_ref))
 547                        kmem_cache_free(skbuff_fclone_cache, other);
 548                break;
 549        }
 550}
 551
 552static void skb_release_head_state(struct sk_buff *skb)
 553{
 554        skb_dst_drop(skb);
 555#ifdef CONFIG_XFRM
 556        secpath_put(skb->sp);
 557#endif
 558        if (skb->destructor) {
 559                WARN_ON(in_irq());
 560                skb->destructor(skb);
 561        }
 562#if IS_ENABLED(CONFIG_NF_CONNTRACK)
 563        nf_conntrack_put(skb->nfct);
 564#endif
 565#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED
 566        nf_conntrack_put_reasm(skb->nfct_reasm);
 567#endif
 568#ifdef CONFIG_BRIDGE_NETFILTER
 569        nf_bridge_put(skb->nf_bridge);
 570#endif
 571/* XXX: IS this still necessary? - JHS */
 572#ifdef CONFIG_NET_SCHED
 573        skb->tc_index = 0;
 574#ifdef CONFIG_NET_CLS_ACT
 575        skb->tc_verd = 0;
 576#endif
 577#endif
 578}
 579
 580/* Free everything but the sk_buff shell. */
 581static void skb_release_all(struct sk_buff *skb)
 582{
 583        skb_release_head_state(skb);
 584        skb_release_data(skb);
 585}
 586
 587/**
 588 *      __kfree_skb - private function
 589 *      @skb: buffer
 590 *
 591 *      Free an sk_buff. Release anything attached to the buffer.
 592 *      Clean the state. This is an internal helper function. Users should
 593 *      always call kfree_skb
 594 */
 595
 596void __kfree_skb(struct sk_buff *skb)
 597{
 598        skb_release_all(skb);
 599        kfree_skbmem(skb);
 600}
 601EXPORT_SYMBOL(__kfree_skb);
 602
 603/**
 604 *      kfree_skb - free an sk_buff
 605 *      @skb: buffer to free
 606 *
 607 *      Drop a reference to the buffer and free it if the usage count has
 608 *      hit zero.
 609 */
 610void kfree_skb(struct sk_buff *skb)
 611{
 612        if (unlikely(!skb))
 613                return;
 614        if (likely(atomic_read(&skb->users) == 1))
 615                smp_rmb();
 616        else if (likely(!atomic_dec_and_test(&skb->users)))
 617                return;
 618        trace_kfree_skb(skb, __builtin_return_address(0));
 619        __kfree_skb(skb);
 620}
 621EXPORT_SYMBOL(kfree_skb);
 622
 623/**
 624 *      consume_skb - free an skbuff
 625 *      @skb: buffer to free
 626 *
 627 *      Drop a ref to the buffer and free it if the usage count has hit zero
 628 *      Functions identically to kfree_skb, but kfree_skb assumes that the frame
 629 *      is being dropped after a failure and notes that
 630 */
 631void consume_skb(struct sk_buff *skb)
 632{
 633        if (unlikely(!skb))
 634                return;
 635        if (likely(atomic_read(&skb->users) == 1))
 636                smp_rmb();
 637        else if (likely(!atomic_dec_and_test(&skb->users)))
 638                return;
 639        trace_consume_skb(skb);
 640        __kfree_skb(skb);
 641}
 642EXPORT_SYMBOL(consume_skb);
 643
 644/**
 645 *      skb_recycle - clean up an skb for reuse
 646 *      @skb: buffer
 647 *
 648 *      Recycles the skb to be reused as a receive buffer. This
 649 *      function does any necessary reference count dropping, and
 650 *      cleans up the skbuff as if it just came from __alloc_skb().
 651 */
 652void skb_recycle(struct sk_buff *skb)
 653{
 654        struct skb_shared_info *shinfo;
 655
 656        skb_release_head_state(skb);
 657
 658        shinfo = skb_shinfo(skb);
 659        memset(shinfo, 0, offsetof(struct skb_shared_info, dataref));
 660        atomic_set(&shinfo->dataref, 1);
 661
 662        memset(skb, 0, offsetof(struct sk_buff, tail));
 663        skb->data = skb->head + NET_SKB_PAD;
 664        skb_reset_tail_pointer(skb);
 665}
 666EXPORT_SYMBOL(skb_recycle);
 667
 668/**
 669 *      skb_recycle_check - check if skb can be reused for receive
 670 *      @skb: buffer
 671 *      @skb_size: minimum receive buffer size
 672 *
 673 *      Checks that the skb passed in is not shared or cloned, and
 674 *      that it is linear and its head portion at least as large as
 675 *      skb_size so that it can be recycled as a receive buffer.
 676 *      If these conditions are met, this function does any necessary
 677 *      reference count dropping and cleans up the skbuff as if it
 678 *      just came from __alloc_skb().
 679 */
 680bool skb_recycle_check(struct sk_buff *skb, int skb_size)
 681{
 682        if (!skb_is_recycleable(skb, skb_size))
 683                return false;
 684
 685        skb_recycle(skb);
 686
 687        return true;
 688}
 689EXPORT_SYMBOL(skb_recycle_check);
 690
 691static void __copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
 692{
 693        new->tstamp             = old->tstamp;
 694        new->dev                = old->dev;
 695        new->transport_header   = old->transport_header;
 696        new->network_header     = old->network_header;
 697        new->mac_header         = old->mac_header;
 698        skb_dst_copy(new, old);
 699        new->rxhash             = old->rxhash;
 700        new->ooo_okay           = old->ooo_okay;
 701        new->l4_rxhash          = old->l4_rxhash;
 702        new->no_fcs             = old->no_fcs;
 703#ifdef CONFIG_XFRM
 704        new->sp                 = secpath_get(old->sp);
 705#endif
 706        memcpy(new->cb, old->cb, sizeof(old->cb));
 707        new->csum               = old->csum;
 708        new->local_df           = old->local_df;
 709        new->pkt_type           = old->pkt_type;
 710        new->ip_summed          = old->ip_summed;
 711        skb_copy_queue_mapping(new, old);
 712        new->priority           = old->priority;
 713#if IS_ENABLED(CONFIG_IP_VS)
 714        new->ipvs_property      = old->ipvs_property;
 715#endif
 716        new->pfmemalloc         = old->pfmemalloc;
 717        new->protocol           = old->protocol;
 718        new->mark               = old->mark;
 719        new->skb_iif            = old->skb_iif;
 720        __nf_copy(new, old);
 721#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE)
 722        new->nf_trace           = old->nf_trace;
 723#endif
 724#ifdef CONFIG_NET_SCHED
 725        new->tc_index           = old->tc_index;
 726#ifdef CONFIG_NET_CLS_ACT
 727        new->tc_verd            = old->tc_verd;
 728#endif
 729#endif
 730        new->vlan_tci           = old->vlan_tci;
 731
 732        skb_copy_secmark(new, old);
 733}
 734
 735/*
 736 * You should not add any new code to this function.  Add it to
 737 * __copy_skb_header above instead.
 738 */
 739static struct sk_buff *__skb_clone(struct sk_buff *n, struct sk_buff *skb)
 740{
 741#define C(x) n->x = skb->x
 742
 743        n->next = n->prev = NULL;
 744        n->sk = NULL;
 745        __copy_skb_header(n, skb);
 746
 747        C(len);
 748        C(data_len);
 749        C(mac_len);
 750        n->hdr_len = skb->nohdr ? skb_headroom(skb) : skb->hdr_len;
 751        n->cloned = 1;
 752        n->nohdr = 0;
 753        n->destructor = NULL;
 754        C(tail);
 755        C(end);
 756        C(head);
 757        C(head_frag);
 758        C(data);
 759        C(truesize);
 760        atomic_set(&n->users, 1);
 761
 762        atomic_inc(&(skb_shinfo(skb)->dataref));
 763        skb->cloned = 1;
 764
 765        return n;
 766#undef C
 767}
 768
 769/**
 770 *      skb_morph       -       morph one skb into another
 771 *      @dst: the skb to receive the contents
 772 *      @src: the skb to supply the contents
 773 *
 774 *      This is identical to skb_clone except that the target skb is
 775 *      supplied by the user.
 776 *
 777 *      The target skb is returned upon exit.
 778 */
 779struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src)
 780{
 781        skb_release_all(dst);
 782        return __skb_clone(dst, src);
 783}
 784EXPORT_SYMBOL_GPL(skb_morph);
 785
 786/**
 787 *      skb_copy_ubufs  -       copy userspace skb frags buffers to kernel
 788 *      @skb: the skb to modify
 789 *      @gfp_mask: allocation priority
 790 *
 791 *      This must be called on SKBTX_DEV_ZEROCOPY skb.
 792 *      It will copy all frags into kernel and drop the reference
 793 *      to userspace pages.
 794 *
 795 *      If this function is called from an interrupt gfp_mask() must be
 796 *      %GFP_ATOMIC.
 797 *
 798 *      Returns 0 on success or a negative error code on failure
 799 *      to allocate kernel memory to copy to.
 800 */
 801int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask)
 802{
 803        int i;
 804        int num_frags = skb_shinfo(skb)->nr_frags;
 805        struct page *page, *head = NULL;
 806        struct ubuf_info *uarg = skb_shinfo(skb)->destructor_arg;
 807
 808        for (i = 0; i < num_frags; i++) {
 809                u8 *vaddr;
 810                skb_frag_t *f = &skb_shinfo(skb)->frags[i];
 811
 812                page = alloc_page(gfp_mask);
 813                if (!page) {
 814                        while (head) {
 815                                struct page *next = (struct page *)head->private;
 816                                put_page(head);
 817                                head = next;
 818                        }
 819                        return -ENOMEM;
 820                }
 821                vaddr = kmap_atomic(skb_frag_page(f));
 822                memcpy(page_address(page),
 823                       vaddr + f->page_offset, skb_frag_size(f));
 824                kunmap_atomic(vaddr);
 825                page->private = (unsigned long)head;
 826                head = page;
 827        }
 828
 829        /* skb frags release userspace buffers */
 830        for (i = 0; i < num_frags; i++)
 831                skb_frag_unref(skb, i);
 832
 833        uarg->callback(uarg);
 834
 835        /* skb frags point to kernel buffers */
 836        for (i = num_frags - 1; i >= 0; i--) {
 837                __skb_fill_page_desc(skb, i, head, 0,
 838                                     skb_shinfo(skb)->frags[i].size);
 839                head = (struct page *)head->private;
 840        }
 841
 842        skb_shinfo(skb)->tx_flags &= ~SKBTX_DEV_ZEROCOPY;
 843        return 0;
 844}
 845EXPORT_SYMBOL_GPL(skb_copy_ubufs);
 846
 847/**
 848 *      skb_clone       -       duplicate an sk_buff
 849 *      @skb: buffer to clone
 850 *      @gfp_mask: allocation priority
 851 *
 852 *      Duplicate an &sk_buff. The new one is not owned by a socket. Both
 853 *      copies share the same packet data but not structure. The new
 854 *      buffer has a reference count of 1. If the allocation fails the
 855 *      function returns %NULL otherwise the new buffer is returned.
 856 *
 857 *      If this function is called from an interrupt gfp_mask() must be
 858 *      %GFP_ATOMIC.
 859 */
 860
 861struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t gfp_mask)
 862{
 863        struct sk_buff *n;
 864
 865        if (skb_orphan_frags(skb, gfp_mask))
 866                return NULL;
 867
 868        n = skb + 1;
 869        if (skb->fclone == SKB_FCLONE_ORIG &&
 870            n->fclone == SKB_FCLONE_UNAVAILABLE) {
 871                atomic_t *fclone_ref = (atomic_t *) (n + 1);
 872                n->fclone = SKB_FCLONE_CLONE;
 873                atomic_inc(fclone_ref);
 874        } else {
 875                if (skb_pfmemalloc(skb))
 876                        gfp_mask |= __GFP_MEMALLOC;
 877
 878                n = kmem_cache_alloc(skbuff_head_cache, gfp_mask);
 879                if (!n)
 880                        return NULL;
 881
 882                kmemcheck_annotate_bitfield(n, flags1);
 883                kmemcheck_annotate_bitfield(n, flags2);
 884                n->fclone = SKB_FCLONE_UNAVAILABLE;
 885        }
 886
 887        return __skb_clone(n, skb);
 888}
 889EXPORT_SYMBOL(skb_clone);
 890
 891static void copy_skb_header(struct sk_buff *new, const struct sk_buff *old)
 892{
 893#ifndef NET_SKBUFF_DATA_USES_OFFSET
 894        /*
 895         *      Shift between the two data areas in bytes
 896         */
 897        unsigned long offset = new->data - old->data;
 898#endif
 899
 900        __copy_skb_header(new, old);
 901
 902#ifndef NET_SKBUFF_DATA_USES_OFFSET
 903        /* {transport,network,mac}_header are relative to skb->head */
 904        new->transport_header += offset;
 905        new->network_header   += offset;
 906        if (skb_mac_header_was_set(new))
 907                new->mac_header       += offset;
 908#endif
 909        skb_shinfo(new)->gso_size = skb_shinfo(old)->gso_size;
 910        skb_shinfo(new)->gso_segs = skb_shinfo(old)->gso_segs;
 911        skb_shinfo(new)->gso_type = skb_shinfo(old)->gso_type;
 912}
 913
 914static inline int skb_alloc_rx_flag(const struct sk_buff *skb)
 915{
 916        if (skb_pfmemalloc(skb))
 917                return SKB_ALLOC_RX;
 918        return 0;
 919}
 920
 921/**
 922 *      skb_copy        -       create private copy of an sk_buff
 923 *      @skb: buffer to copy
 924 *      @gfp_mask: allocation priority
 925 *
 926 *      Make a copy of both an &sk_buff and its data. This is used when the
 927 *      caller wishes to modify the data and needs a private copy of the
 928 *      data to alter. Returns %NULL on failure or the pointer to the buffer
 929 *      on success. The returned buffer has a reference count of 1.
 930 *
 931 *      As by-product this function converts non-linear &sk_buff to linear
 932 *      one, so that &sk_buff becomes completely private and caller is allowed
 933 *      to modify all the data of returned buffer. This means that this
 934 *      function is not recommended for use in circumstances when only
 935 *      header is going to be modified. Use pskb_copy() instead.
 936 */
 937
 938struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t gfp_mask)
 939{
 940        int headerlen = skb_headroom(skb);
 941        unsigned int size = skb_end_offset(skb) + skb->data_len;
 942        struct sk_buff *n = __alloc_skb(size, gfp_mask,
 943                                        skb_alloc_rx_flag(skb), NUMA_NO_NODE);
 944
 945        if (!n)
 946                return NULL;
 947
 948        /* Set the data pointer */
 949        skb_reserve(n, headerlen);
 950        /* Set the tail pointer and length */
 951        skb_put(n, skb->len);
 952
 953        if (skb_copy_bits(skb, -headerlen, n->head, headerlen + skb->len))
 954                BUG();
 955
 956        copy_skb_header(n, skb);
 957        return n;
 958}
 959EXPORT_SYMBOL(skb_copy);
 960
 961/**
 962 *      __pskb_copy     -       create copy of an sk_buff with private head.
 963 *      @skb: buffer to copy
 964 *      @headroom: headroom of new skb
 965 *      @gfp_mask: allocation priority
 966 *
 967 *      Make a copy of both an &sk_buff and part of its data, located
 968 *      in header. Fragmented data remain shared. This is used when
 969 *      the caller wishes to modify only header of &sk_buff and needs
 970 *      private copy of the header to alter. Returns %NULL on failure
 971 *      or the pointer to the buffer on success.
 972 *      The returned buffer has a reference count of 1.
 973 */
 974
 975struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, gfp_t gfp_mask)
 976{
 977        unsigned int size = skb_headlen(skb) + headroom;
 978        struct sk_buff *n = __alloc_skb(size, gfp_mask,
 979                                        skb_alloc_rx_flag(skb), NUMA_NO_NODE);
 980
 981        if (!n)
 982                goto out;
 983
 984        /* Set the data pointer */
 985        skb_reserve(n, headroom);
 986        /* Set the tail pointer and length */
 987        skb_put(n, skb_headlen(skb));
 988        /* Copy the bytes */
 989        skb_copy_from_linear_data(skb, n->data, n->len);
 990
 991        n->truesize += skb->data_len;
 992        n->data_len  = skb->data_len;
 993        n->len       = skb->len;
 994
 995        if (skb_shinfo(skb)->nr_frags) {
 996                int i;
 997
 998                if (skb_orphan_frags(skb, gfp_mask)) {
 999                        kfree_skb(n);
1000                        n = NULL;
1001                        goto out;
1002                }
1003                for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1004                        skb_shinfo(n)->frags[i] = skb_shinfo(skb)->frags[i];
1005                        skb_frag_ref(skb, i);
1006                }
1007                skb_shinfo(n)->nr_frags = i;
1008        }
1009
1010        if (skb_has_frag_list(skb)) {
1011                skb_shinfo(n)->frag_list = skb_shinfo(skb)->frag_list;
1012                skb_clone_fraglist(n);
1013        }
1014
1015        copy_skb_header(n, skb);
1016out:
1017        return n;
1018}
1019EXPORT_SYMBOL(__pskb_copy);
1020
1021/**
1022 *      pskb_expand_head - reallocate header of &sk_buff
1023 *      @skb: buffer to reallocate
1024 *      @nhead: room to add at head
1025 *      @ntail: room to add at tail
1026 *      @gfp_mask: allocation priority
1027 *
1028 *      Expands (or creates identical copy, if &nhead and &ntail are zero)
1029 *      header of skb. &sk_buff itself is not changed. &sk_buff MUST have
1030 *      reference count of 1. Returns zero in the case of success or error,
1031 *      if expansion failed. In the last case, &sk_buff is not changed.
1032 *
1033 *      All the pointers pointing into skb header may change and must be
1034 *      reloaded after call to this function.
1035 */
1036
1037int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail,
1038                     gfp_t gfp_mask)
1039{
1040        int i;
1041        u8 *data;
1042        int size = nhead + skb_end_offset(skb) + ntail;
1043        long off;
1044
1045        BUG_ON(nhead < 0);
1046
1047        if (skb_shared(skb))
1048                BUG();
1049
1050        size = SKB_DATA_ALIGN(size);
1051
1052        if (skb_pfmemalloc(skb))
1053                gfp_mask |= __GFP_MEMALLOC;
1054        data = kmalloc_reserve(size + SKB_DATA_ALIGN(sizeof(struct skb_shared_info)),
1055                               gfp_mask, NUMA_NO_NODE, NULL);
1056        if (!data)
1057                goto nodata;
1058        size = SKB_WITH_OVERHEAD(ksize(data));
1059
1060        /* Copy only real data... and, alas, header. This should be
1061         * optimized for the cases when header is void.
1062         */
1063        memcpy(data + nhead, skb->head, skb_tail_pointer(skb) - skb->head);
1064
1065        memcpy((struct skb_shared_info *)(data + size),
1066               skb_shinfo(skb),
1067               offsetof(struct skb_shared_info, frags[skb_shinfo(skb)->nr_frags]));
1068
1069        /*
1070         * if shinfo is shared we must drop the old head gracefully, but if it
1071         * is not we can just drop the old head and let the existing refcount
1072         * be since all we did is relocate the values
1073         */
1074        if (skb_cloned(skb)) {
1075                /* copy this zero copy skb frags */
1076                if (skb_orphan_frags(skb, gfp_mask))
1077                        goto nofrags;
1078                for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
1079                        skb_frag_ref(skb, i);
1080
1081                if (skb_has_frag_list(skb))
1082                        skb_clone_fraglist(skb);
1083
1084                skb_release_data(skb);
1085        } else {
1086                skb_free_head(skb);
1087        }
1088        off = (data + nhead) - skb->head;
1089
1090        skb->head     = data;
1091        skb->head_frag = 0;
1092        skb->data    += off;
1093#ifdef NET_SKBUFF_DATA_USES_OFFSET
1094        skb->end      = size;
1095        off           = nhead;
1096#else
1097        skb->end      = skb->head + size;
1098#endif
1099        /* {transport,network,mac}_header and tail are relative to skb->head */
1100        skb->tail             += off;
1101        skb->transport_header += off;
1102        skb->network_header   += off;
1103        if (skb_mac_header_was_set(skb))
1104                skb->mac_header += off;
1105        /* Only adjust this if it actually is csum_start rather than csum */
1106        if (skb->ip_summed == CHECKSUM_PARTIAL)
1107                skb->csum_start += nhead;
1108        skb->cloned   = 0;
1109        skb->hdr_len  = 0;
1110        skb->nohdr    = 0;
1111        atomic_set(&skb_shinfo(skb)->dataref, 1);
1112        return 0;
1113
1114nofrags:
1115        kfree(data);
1116nodata:
1117        return -ENOMEM;
1118}
1119EXPORT_SYMBOL(pskb_expand_head);
1120
1121/* Make private copy of skb with writable head and some headroom */
1122
1123struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, unsigned int headroom)
1124{
1125        struct sk_buff *skb2;
1126        int delta = headroom - skb_headroom(skb);
1127
1128        if (delta <= 0)
1129                skb2 = pskb_copy(skb, GFP_ATOMIC);
1130        else {
1131                skb2 = skb_clone(skb, GFP_ATOMIC);
1132                if (skb2 && pskb_expand_head(skb2, SKB_DATA_ALIGN(delta), 0,
1133                                             GFP_ATOMIC)) {
1134                        kfree_skb(skb2);
1135                        skb2 = NULL;
1136                }
1137        }
1138        return skb2;
1139}
1140EXPORT_SYMBOL(skb_realloc_headroom);
1141
1142/**
1143 *      skb_copy_expand -       copy and expand sk_buff
1144 *      @skb: buffer to copy
1145 *      @newheadroom: new free bytes at head
1146 *      @newtailroom: new free bytes at tail
1147 *      @gfp_mask: allocation priority
1148 *
1149 *      Make a copy of both an &sk_buff and its data and while doing so
1150 *      allocate additional space.
1151 *
1152 *      This is used when the caller wishes to modify the data and needs a
1153 *      private copy of the data to alter as well as more space for new fields.
1154 *      Returns %NULL on failure or the pointer to the buffer
1155 *      on success. The returned buffer has a reference count of 1.
1156 *
1157 *      You must pass %GFP_ATOMIC as the allocation priority if this function
1158 *      is called from an interrupt.
1159 */
1160struct sk_buff *skb_copy_expand(const struct sk_buff *skb,
1161                                int newheadroom, int newtailroom,
1162                                gfp_t gfp_mask)
1163{
1164        /*
1165         *      Allocate the copy buffer
1166         */
1167        struct sk_buff *n = __alloc_skb(newheadroom + skb->len + newtailroom,
1168                                        gfp_mask, skb_alloc_rx_flag(skb),
1169                                        NUMA_NO_NODE);
1170        int oldheadroom = skb_headroom(skb);
1171        int head_copy_len, head_copy_off;
1172        int off;
1173
1174        if (!n)
1175                return NULL;
1176
1177        skb_reserve(n, newheadroom);
1178
1179        /* Set the tail pointer and length */
1180        skb_put(n, skb->len);
1181
1182        head_copy_len = oldheadroom;
1183        head_copy_off = 0;
1184        if (newheadroom <= head_copy_len)
1185                head_copy_len = newheadroom;
1186        else
1187                head_copy_off = newheadroom - head_copy_len;
1188
1189        /* Copy the linear header and data. */
1190        if (skb_copy_bits(skb, -head_copy_len, n->head + head_copy_off,
1191                          skb->len + head_copy_len))
1192                BUG();
1193
1194        copy_skb_header(n, skb);
1195
1196        off                  = newheadroom - oldheadroom;
1197        if (n->ip_summed == CHECKSUM_PARTIAL)
1198                n->csum_start += off;
1199#ifdef NET_SKBUFF_DATA_USES_OFFSET
1200        n->transport_header += off;
1201        n->network_header   += off;
1202        if (skb_mac_header_was_set(skb))
1203                n->mac_header += off;
1204#endif
1205
1206        return n;
1207}
1208EXPORT_SYMBOL(skb_copy_expand);
1209
1210/**
1211 *      skb_pad                 -       zero pad the tail of an skb
1212 *      @skb: buffer to pad
1213 *      @pad: space to pad
1214 *
1215 *      Ensure that a buffer is followed by a padding area that is zero
1216 *      filled. Used by network drivers which may DMA or transfer data
1217 *      beyond the buffer end onto the wire.
1218 *
1219 *      May return error in out of memory cases. The skb is freed on error.
1220 */
1221
1222int skb_pad(struct sk_buff *skb, int pad)
1223{
1224        int err;
1225        int ntail;
1226
1227        /* If the skbuff is non linear tailroom is always zero.. */
1228        if (!skb_cloned(skb) && skb_tailroom(skb) >= pad) {
1229                memset(skb->data+skb->len, 0, pad);
1230                return 0;
1231        }
1232
1233        ntail = skb->data_len + pad - (skb->end - skb->tail);
1234        if (likely(skb_cloned(skb) || ntail > 0)) {
1235                err = pskb_expand_head(skb, 0, ntail, GFP_ATOMIC);
1236                if (unlikely(err))
1237                        goto free_skb;
1238        }
1239
1240        /* FIXME: The use of this function with non-linear skb's really needs
1241         * to be audited.
1242         */
1243        err = skb_linearize(skb);
1244        if (unlikely(err))
1245                goto free_skb;
1246
1247        memset(skb->data + skb->len, 0, pad);
1248        return 0;
1249
1250free_skb:
1251        kfree_skb(skb);
1252        return err;
1253}
1254EXPORT_SYMBOL(skb_pad);
1255
1256/**
1257 *      skb_put - add data to a buffer
1258 *      @skb: buffer to use
1259 *      @len: amount of data to add
1260 *
1261 *      This function extends the used data area of the buffer. If this would
1262 *      exceed the total buffer size the kernel will panic. A pointer to the
1263 *      first byte of the extra data is returned.
1264 */
1265unsigned char *skb_put(struct sk_buff *skb, unsigned int len)
1266{
1267        unsigned char *tmp = skb_tail_pointer(skb);
1268        SKB_LINEAR_ASSERT(skb);
1269        skb->tail += len;
1270        skb->len  += len;
1271        if (unlikely(skb->tail > skb->end))
1272                skb_over_panic(skb, len, __builtin_return_address(0));
1273        return tmp;
1274}
1275EXPORT_SYMBOL(skb_put);
1276
1277/**
1278 *      skb_push - add data to the start of a buffer
1279 *      @skb: buffer to use
1280 *      @len: amount of data to add
1281 *
1282 *      This function extends the used data area of the buffer at the buffer
1283 *      start. If this would exceed the total buffer headroom the kernel will
1284 *      panic. A pointer to the first byte of the extra data is returned.
1285 */
1286unsigned char *skb_push(struct sk_buff *skb, unsigned int len)
1287{
1288        skb->data -= len;
1289        skb->len  += len;
1290        if (unlikely(skb->data<skb->head))
1291                skb_under_panic(skb, len, __builtin_return_address(0));
1292        return skb->data;
1293}
1294EXPORT_SYMBOL(skb_push);
1295
1296/**
1297 *      skb_pull - remove data from the start of a buffer
1298 *      @skb: buffer to use
1299 *      @len: amount of data to remove
1300 *
1301 *      This function removes data from the start of a buffer, returning
1302 *      the memory to the headroom. A pointer to the next data in the buffer
1303 *      is returned. Once the data has been pulled future pushes will overwrite
1304 *      the old data.
1305 */
1306unsigned char *skb_pull(struct sk_buff *skb, unsigned int len)
1307{
1308        return skb_pull_inline(skb, len);
1309}
1310EXPORT_SYMBOL(skb_pull);
1311
1312/**
1313 *      skb_trim - remove end from a buffer
1314 *      @skb: buffer to alter
1315 *      @len: new length
1316 *
1317 *      Cut the length of a buffer down by removing data from the tail. If
1318 *      the buffer is already under the length specified it is not modified.
1319 *      The skb must be linear.
1320 */
1321void skb_trim(struct sk_buff *skb, unsigned int len)
1322{
1323        if (skb->len > len)
1324                __skb_trim(skb, len);
1325}
1326EXPORT_SYMBOL(skb_trim);
1327
1328/* Trims skb to length len. It can change skb pointers.
1329 */
1330
1331int ___pskb_trim(struct sk_buff *skb, unsigned int len)
1332{
1333        struct sk_buff **fragp;
1334        struct sk_buff *frag;
1335        int offset = skb_headlen(skb);
1336        int nfrags = skb_shinfo(skb)->nr_frags;
1337        int i;
1338        int err;
1339
1340        if (skb_cloned(skb) &&
1341            unlikely((err = pskb_expand_head(skb, 0, 0, GFP_ATOMIC))))
1342                return err;
1343
1344        i = 0;
1345        if (offset >= len)
1346                goto drop_pages;
1347
1348        for (; i < nfrags; i++) {
1349                int end = offset + skb_frag_size(&skb_shinfo(skb)->frags[i]);
1350
1351                if (end < len) {
1352                        offset = end;
1353                        continue;
1354                }
1355
1356                skb_frag_size_set(&skb_shinfo(skb)->frags[i++], len - offset);
1357
1358drop_pages:
1359                skb_shinfo(skb)->nr_frags = i;
1360
1361                for (; i < nfrags; i++)
1362                        skb_frag_unref(skb, i);
1363
1364                if (skb_has_frag_list(skb))
1365                        skb_drop_fraglist(skb);
1366                goto done;
1367        }
1368
1369        for (fragp = &skb_shinfo(skb)->frag_list; (frag = *fragp);
1370             fragp = &frag->next) {
1371                int end = offset + frag->len;
1372
1373                if (skb_shared(frag)) {
1374                        struct sk_buff *nfrag;
1375
1376                        nfrag = skb_clone(frag, GFP_ATOMIC);
1377                        if (unlikely(!nfrag))
1378                                return -ENOMEM;
1379
1380                        nfrag->next = frag->next;
1381                        consume_skb(frag);
1382                        frag = nfrag;
1383                        *fragp = frag;
1384                }
1385
1386                if (end < len) {
1387                        offset = end;
1388                        continue;
1389                }
1390
1391                if (end > len &&
1392                    unlikely((err = pskb_trim(frag, len - offset))))
1393                        return err;
1394
1395                if (frag->next)
1396                        skb_drop_list(&frag->next);
1397                break;
1398        }
1399
1400done:
1401        if (len > skb_headlen(skb)) {
1402                skb->data_len -= skb->len - len;
1403                skb->len       = len;
1404        } else {
1405                skb->len       = len;
1406                skb->data_len  = 0;
1407                skb_set_tail_pointer(skb, len);
1408        }
1409
1410        return 0;
1411}
1412EXPORT_SYMBOL(___pskb_trim);
1413
1414/**
1415 *      __pskb_pull_tail - advance tail of skb header
1416 *      @skb: buffer to reallocate
1417 *      @delta: number of bytes to advance tail
1418 *
1419 *      The function makes a sense only on a fragmented &sk_buff,
1420 *      it expands header moving its tail forward and copying necessary
1421 *      data from fragmented part.
1422 *
1423 *      &sk_buff MUST have reference count of 1.
1424 *
1425 *      Returns %NULL (and &sk_buff does not change) if pull failed
1426 *      or value of new tail of skb in the case of success.
1427 *
1428 *      All the pointers pointing into skb header may change and must be
1429 *      reloaded after call to this function.
1430 */
1431
1432/* Moves tail of skb head forward, copying data from fragmented part,
1433 * when it is necessary.
1434 * 1. It may fail due to malloc failure.
1435 * 2. It may change skb pointers.
1436 *
1437 * It is pretty complicated. Luckily, it is called only in exceptional cases.
1438 */
1439unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta)
1440{
1441        /* If skb has not enough free space at tail, get new one
1442         * plus 128 bytes for future expansions. If we have enough
1443         * room at tail, reallocate without expansion only if skb is cloned.
1444         */
1445        int i, k, eat = (skb->tail + delta) - skb->end;
1446
1447        if (eat > 0 || skb_cloned(skb)) {
1448                if (pskb_expand_head(skb, 0, eat > 0 ? eat + 128 : 0,
1449                                     GFP_ATOMIC))
1450                        return NULL;
1451        }
1452
1453        if (skb_copy_bits(skb, skb_headlen(skb), skb_tail_pointer(skb), delta))
1454                BUG();
1455
1456        /* Optimization: no fragments, no reasons to preestimate
1457         * size of pulled pages. Superb.
1458         */
1459        if (!skb_has_frag_list(skb))
1460                goto pull_pages;
1461
1462        /* Estimate size of pulled pages. */
1463        eat = delta;
1464        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1465                int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1466
1467                if (size >= eat)
1468                        goto pull_pages;
1469                eat -= size;
1470        }
1471
1472        /* If we need update frag list, we are in troubles.
1473         * Certainly, it possible to add an offset to skb data,
1474         * but taking into account that pulling is expected to
1475         * be very rare operation, it is worth to fight against
1476         * further bloating skb head and crucify ourselves here instead.
1477         * Pure masohism, indeed. 8)8)
1478         */
1479        if (eat) {
1480                struct sk_buff *list = skb_shinfo(skb)->frag_list;
1481                struct sk_buff *clone = NULL;
1482                struct sk_buff *insp = NULL;
1483
1484                do {
1485                        BUG_ON(!list);
1486
1487                        if (list->len <= eat) {
1488                                /* Eaten as whole. */
1489                                eat -= list->len;
1490                                list = list->next;
1491                                insp = list;
1492                        } else {
1493                                /* Eaten partially. */
1494
1495                                if (skb_shared(list)) {
1496                                        /* Sucks! We need to fork list. :-( */
1497                                        clone = skb_clone(list, GFP_ATOMIC);
1498                                        if (!clone)
1499                                                return NULL;
1500                                        insp = list->next;
1501                                        list = clone;
1502                                } else {
1503                                        /* This may be pulled without
1504                                         * problems. */
1505                                        insp = list;
1506                                }
1507                                if (!pskb_pull(list, eat)) {
1508                                        kfree_skb(clone);
1509                                        return NULL;
1510                                }
1511                                break;
1512                        }
1513                } while (eat);
1514
1515                /* Free pulled out fragments. */
1516                while ((list = skb_shinfo(skb)->frag_list) != insp) {
1517                        skb_shinfo(skb)->frag_list = list->next;
1518                        kfree_skb(list);
1519                }
1520                /* And insert new clone at head. */
1521                if (clone) {
1522                        clone->next = list;
1523                        skb_shinfo(skb)->frag_list = clone;
1524                }
1525        }
1526        /* Success! Now we may commit changes to skb data. */
1527
1528pull_pages:
1529        eat = delta;
1530        k = 0;
1531        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1532                int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
1533
1534                if (size <= eat) {
1535                        skb_frag_unref(skb, i);
1536                        eat -= size;
1537                } else {
1538                        skb_shinfo(skb)->frags[k] = skb_shinfo(skb)->frags[i];
1539                        if (eat) {
1540                                skb_shinfo(skb)->frags[k].page_offset += eat;
1541                                skb_frag_size_sub(&skb_shinfo(skb)->frags[k], eat);
1542                                eat = 0;
1543                        }
1544                        k++;
1545                }
1546        }
1547        skb_shinfo(skb)->nr_frags = k;
1548
1549        skb->tail     += delta;
1550        skb->data_len -= delta;
1551
1552        return skb_tail_pointer(skb);
1553}
1554EXPORT_SYMBOL(__pskb_pull_tail);
1555
1556/**
1557 *      skb_copy_bits - copy bits from skb to kernel buffer
1558 *      @skb: source skb
1559 *      @offset: offset in source
1560 *      @to: destination buffer
1561 *      @len: number of bytes to copy
1562 *
1563 *      Copy the specified number of bytes from the source skb to the
1564 *      destination buffer.
1565 *
1566 *      CAUTION ! :
1567 *              If its prototype is ever changed,
1568 *              check arch/{*}/net/{*}.S files,
1569 *              since it is called from BPF assembly code.
1570 */
1571int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len)
1572{
1573        int start = skb_headlen(skb);
1574        struct sk_buff *frag_iter;
1575        int i, copy;
1576
1577        if (offset > (int)skb->len - len)
1578                goto fault;
1579
1580        /* Copy header. */
1581        if ((copy = start - offset) > 0) {
1582                if (copy > len)
1583                        copy = len;
1584                skb_copy_from_linear_data_offset(skb, offset, to, copy);
1585                if ((len -= copy) == 0)
1586                        return 0;
1587                offset += copy;
1588                to     += copy;
1589        }
1590
1591        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1592                int end;
1593                skb_frag_t *f = &skb_shinfo(skb)->frags[i];
1594
1595                WARN_ON(start > offset + len);
1596
1597                end = start + skb_frag_size(f);
1598                if ((copy = end - offset) > 0) {
1599                        u8 *vaddr;
1600
1601                        if (copy > len)
1602                                copy = len;
1603
1604                        vaddr = kmap_atomic(skb_frag_page(f));
1605                        memcpy(to,
1606                               vaddr + f->page_offset + offset - start,
1607                               copy);
1608                        kunmap_atomic(vaddr);
1609
1610                        if ((len -= copy) == 0)
1611                                return 0;
1612                        offset += copy;
1613                        to     += copy;
1614                }
1615                start = end;
1616        }
1617
1618        skb_walk_frags(skb, frag_iter) {
1619                int end;
1620
1621                WARN_ON(start > offset + len);
1622
1623                end = start + frag_iter->len;
1624                if ((copy = end - offset) > 0) {
1625                        if (copy > len)
1626                                copy = len;
1627                        if (skb_copy_bits(frag_iter, offset - start, to, copy))
1628                                goto fault;
1629                        if ((len -= copy) == 0)
1630                                return 0;
1631                        offset += copy;
1632                        to     += copy;
1633                }
1634                start = end;
1635        }
1636
1637        if (!len)
1638                return 0;
1639
1640fault:
1641        return -EFAULT;
1642}
1643EXPORT_SYMBOL(skb_copy_bits);
1644
1645/*
1646 * Callback from splice_to_pipe(), if we need to release some pages
1647 * at the end of the spd in case we error'ed out in filling the pipe.
1648 */
1649static void sock_spd_release(struct splice_pipe_desc *spd, unsigned int i)
1650{
1651        put_page(spd->pages[i]);
1652}
1653
1654static struct page *linear_to_page(struct page *page, unsigned int *len,
1655                                   unsigned int *offset,
1656                                   struct sk_buff *skb, struct sock *sk)
1657{
1658        struct page *p = sk->sk_sndmsg_page;
1659        unsigned int off;
1660
1661        if (!p) {
1662new_page:
1663                p = sk->sk_sndmsg_page = alloc_pages(sk->sk_allocation, 0);
1664                if (!p)
1665                        return NULL;
1666
1667                off = sk->sk_sndmsg_off = 0;
1668                /* hold one ref to this page until it's full */
1669        } else {
1670                unsigned int mlen;
1671
1672                /* If we are the only user of the page, we can reset offset */
1673                if (page_count(p) == 1)
1674                        sk->sk_sndmsg_off = 0;
1675                off = sk->sk_sndmsg_off;
1676                mlen = PAGE_SIZE - off;
1677                if (mlen < 64 && mlen < *len) {
1678                        put_page(p);
1679                        goto new_page;
1680                }
1681
1682                *len = min_t(unsigned int, *len, mlen);
1683        }
1684
1685        memcpy(page_address(p) + off, page_address(page) + *offset, *len);
1686        sk->sk_sndmsg_off += *len;
1687        *offset = off;
1688
1689        return p;
1690}
1691
1692static bool spd_can_coalesce(const struct splice_pipe_desc *spd,
1693                             struct page *page,
1694                             unsigned int offset)
1695{
1696        return  spd->nr_pages &&
1697                spd->pages[spd->nr_pages - 1] == page &&
1698                (spd->partial[spd->nr_pages - 1].offset +
1699                 spd->partial[spd->nr_pages - 1].len == offset);
1700}
1701
1702/*
1703 * Fill page/offset/length into spd, if it can hold more pages.
1704 */
1705static bool spd_fill_page(struct splice_pipe_desc *spd,
1706                          struct pipe_inode_info *pipe, struct page *page,
1707                          unsigned int *len, unsigned int offset,
1708                          struct sk_buff *skb, bool linear,
1709                          struct sock *sk)
1710{
1711        if (unlikely(spd->nr_pages == MAX_SKB_FRAGS))
1712                return true;
1713
1714        if (linear) {
1715                page = linear_to_page(page, len, &offset, skb, sk);
1716                if (!page)
1717                        return true;
1718        }
1719        if (spd_can_coalesce(spd, page, offset)) {
1720                spd->partial[spd->nr_pages - 1].len += *len;
1721                return false;
1722        }
1723        get_page(page);
1724        spd->pages[spd->nr_pages] = page;
1725        spd->partial[spd->nr_pages].len = *len;
1726        spd->partial[spd->nr_pages].offset = offset;
1727        spd->nr_pages++;
1728
1729        return false;
1730}
1731
1732static inline void __segment_seek(struct page **page, unsigned int *poff,
1733                                  unsigned int *plen, unsigned int off)
1734{
1735        unsigned long n;
1736
1737        *poff += off;
1738        n = *poff / PAGE_SIZE;
1739        if (n)
1740                *page = nth_page(*page, n);
1741
1742        *poff = *poff % PAGE_SIZE;
1743        *plen -= off;
1744}
1745
1746static bool __splice_segment(struct page *page, unsigned int poff,
1747                             unsigned int plen, unsigned int *off,
1748                             unsigned int *len, struct sk_buff *skb,
1749                             struct splice_pipe_desc *spd, bool linear,
1750                             struct sock *sk,
1751                             struct pipe_inode_info *pipe)
1752{
1753        if (!*len)
1754                return true;
1755
1756        /* skip this segment if already processed */
1757        if (*off >= plen) {
1758                *off -= plen;
1759                return false;
1760        }
1761
1762        /* ignore any bits we already processed */
1763        if (*off) {
1764                __segment_seek(&page, &poff, &plen, *off);
1765                *off = 0;
1766        }
1767
1768        do {
1769                unsigned int flen = min(*len, plen);
1770
1771                /* the linear region may spread across several pages  */
1772                flen = min_t(unsigned int, flen, PAGE_SIZE - poff);
1773
1774                if (spd_fill_page(spd, pipe, page, &flen, poff, skb, linear, sk))
1775                        return true;
1776
1777                __segment_seek(&page, &poff, &plen, flen);
1778                *len -= flen;
1779
1780        } while (*len && plen);
1781
1782        return false;
1783}
1784
1785/*
1786 * Map linear and fragment data from the skb to spd. It reports true if the
1787 * pipe is full or if we already spliced the requested length.
1788 */
1789static bool __skb_splice_bits(struct sk_buff *skb, struct pipe_inode_info *pipe,
1790                              unsigned int *offset, unsigned int *len,
1791                              struct splice_pipe_desc *spd, struct sock *sk)
1792{
1793        int seg;
1794
1795        /* map the linear part :
1796         * If skb->head_frag is set, this 'linear' part is backed by a
1797         * fragment, and if the head is not shared with any clones then
1798         * we can avoid a copy since we own the head portion of this page.
1799         */
1800        if (__splice_segment(virt_to_page(skb->data),
1801                             (unsigned long) skb->data & (PAGE_SIZE - 1),
1802                             skb_headlen(skb),
1803                             offset, len, skb, spd,
1804                             skb_head_is_locked(skb),
1805                             sk, pipe))
1806                return true;
1807
1808        /*
1809         * then map the fragments
1810         */
1811        for (seg = 0; seg < skb_shinfo(skb)->nr_frags; seg++) {
1812                const skb_frag_t *f = &skb_shinfo(skb)->frags[seg];
1813
1814                if (__splice_segment(skb_frag_page(f),
1815                                     f->page_offset, skb_frag_size(f),
1816                                     offset, len, skb, spd, false, sk, pipe))
1817                        return true;
1818        }
1819
1820        return false;
1821}
1822
1823/*
1824 * Map data from the skb to a pipe. Should handle both the linear part,
1825 * the fragments, and the frag list. It does NOT handle frag lists within
1826 * the frag list, if such a thing exists. We'd probably need to recurse to
1827 * handle that cleanly.
1828 */
1829int skb_splice_bits(struct sk_buff *skb, unsigned int offset,
1830                    struct pipe_inode_info *pipe, unsigned int tlen,
1831                    unsigned int flags)
1832{
1833        struct partial_page partial[MAX_SKB_FRAGS];
1834        struct page *pages[MAX_SKB_FRAGS];
1835        struct splice_pipe_desc spd = {
1836                .pages = pages,
1837                .partial = partial,
1838                .nr_pages_max = MAX_SKB_FRAGS,
1839                .flags = flags,
1840                .ops = &sock_pipe_buf_ops,
1841                .spd_release = sock_spd_release,
1842        };
1843        struct sk_buff *frag_iter;
1844        struct sock *sk = skb->sk;
1845        int ret = 0;
1846
1847        /*
1848         * __skb_splice_bits() only fails if the output has no room left,
1849         * so no point in going over the frag_list for the error case.
1850         */
1851        if (__skb_splice_bits(skb, pipe, &offset, &tlen, &spd, sk))
1852                goto done;
1853        else if (!tlen)
1854                goto done;
1855
1856        /*
1857         * now see if we have a frag_list to map
1858         */
1859        skb_walk_frags(skb, frag_iter) {
1860                if (!tlen)
1861                        break;
1862                if (__skb_splice_bits(frag_iter, pipe, &offset, &tlen, &spd, sk))
1863                        break;
1864        }
1865
1866done:
1867        if (spd.nr_pages) {
1868                /*
1869                 * Drop the socket lock, otherwise we have reverse
1870                 * locking dependencies between sk_lock and i_mutex
1871                 * here as compared to sendfile(). We enter here
1872                 * with the socket lock held, and splice_to_pipe() will
1873                 * grab the pipe inode lock. For sendfile() emulation,
1874                 * we call into ->sendpage() with the i_mutex lock held
1875                 * and networking will grab the socket lock.
1876                 */
1877                release_sock(sk);
1878                ret = splice_to_pipe(pipe, &spd);
1879                lock_sock(sk);
1880        }
1881
1882        return ret;
1883}
1884
1885/**
1886 *      skb_store_bits - store bits from kernel buffer to skb
1887 *      @skb: destination buffer
1888 *      @offset: offset in destination
1889 *      @from: source buffer
1890 *      @len: number of bytes to copy
1891 *
1892 *      Copy the specified number of bytes from the source buffer to the
1893 *      destination skb.  This function handles all the messy bits of
1894 *      traversing fragment lists and such.
1895 */
1896
1897int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len)
1898{
1899        int start = skb_headlen(skb);
1900        struct sk_buff *frag_iter;
1901        int i, copy;
1902
1903        if (offset > (int)skb->len - len)
1904                goto fault;
1905
1906        if ((copy = start - offset) > 0) {
1907                if (copy > len)
1908                        copy = len;
1909                skb_copy_to_linear_data_offset(skb, offset, from, copy);
1910                if ((len -= copy) == 0)
1911                        return 0;
1912                offset += copy;
1913                from += copy;
1914        }
1915
1916        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1917                skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1918                int end;
1919
1920                WARN_ON(start > offset + len);
1921
1922                end = start + skb_frag_size(frag);
1923                if ((copy = end - offset) > 0) {
1924                        u8 *vaddr;
1925
1926                        if (copy > len)
1927                                copy = len;
1928
1929                        vaddr = kmap_atomic(skb_frag_page(frag));
1930                        memcpy(vaddr + frag->page_offset + offset - start,
1931                               from, copy);
1932                        kunmap_atomic(vaddr);
1933
1934                        if ((len -= copy) == 0)
1935                                return 0;
1936                        offset += copy;
1937                        from += copy;
1938                }
1939                start = end;
1940        }
1941
1942        skb_walk_frags(skb, frag_iter) {
1943                int end;
1944
1945                WARN_ON(start > offset + len);
1946
1947                end = start + frag_iter->len;
1948                if ((copy = end - offset) > 0) {
1949                        if (copy > len)
1950                                copy = len;
1951                        if (skb_store_bits(frag_iter, offset - start,
1952                                           from, copy))
1953                                goto fault;
1954                        if ((len -= copy) == 0)
1955                                return 0;
1956                        offset += copy;
1957                        from += copy;
1958                }
1959                start = end;
1960        }
1961        if (!len)
1962                return 0;
1963
1964fault:
1965        return -EFAULT;
1966}
1967EXPORT_SYMBOL(skb_store_bits);
1968
1969/* Checksum skb data. */
1970
1971__wsum skb_checksum(const struct sk_buff *skb, int offset,
1972                          int len, __wsum csum)
1973{
1974        int start = skb_headlen(skb);
1975        int i, copy = start - offset;
1976        struct sk_buff *frag_iter;
1977        int pos = 0;
1978
1979        /* Checksum header. */
1980        if (copy > 0) {
1981                if (copy > len)
1982                        copy = len;
1983                csum = csum_partial(skb->data + offset, copy, csum);
1984                if ((len -= copy) == 0)
1985                        return csum;
1986                offset += copy;
1987                pos     = copy;
1988        }
1989
1990        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
1991                int end;
1992                skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
1993
1994                WARN_ON(start > offset + len);
1995
1996                end = start + skb_frag_size(frag);
1997                if ((copy = end - offset) > 0) {
1998                        __wsum csum2;
1999                        u8 *vaddr;
2000
2001                        if (copy > len)
2002                                copy = len;
2003                        vaddr = kmap_atomic(skb_frag_page(frag));
2004                        csum2 = csum_partial(vaddr + frag->page_offset +
2005                                             offset - start, copy, 0);
2006                        kunmap_atomic(vaddr);
2007                        csum = csum_block_add(csum, csum2, pos);
2008                        if (!(len -= copy))
2009                                return csum;
2010                        offset += copy;
2011                        pos    += copy;
2012                }
2013                start = end;
2014        }
2015
2016        skb_walk_frags(skb, frag_iter) {
2017                int end;
2018
2019                WARN_ON(start > offset + len);
2020
2021                end = start + frag_iter->len;
2022                if ((copy = end - offset) > 0) {
2023                        __wsum csum2;
2024                        if (copy > len)
2025                                copy = len;
2026                        csum2 = skb_checksum(frag_iter, offset - start,
2027                                             copy, 0);
2028                        csum = csum_block_add(csum, csum2, pos);
2029                        if ((len -= copy) == 0)
2030                                return csum;
2031                        offset += copy;
2032                        pos    += copy;
2033                }
2034                start = end;
2035        }
2036        BUG_ON(len);
2037
2038        return csum;
2039}
2040EXPORT_SYMBOL(skb_checksum);
2041
2042/* Both of above in one bottle. */
2043
2044__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset,
2045                                    u8 *to, int len, __wsum csum)
2046{
2047        int start = skb_headlen(skb);
2048        int i, copy = start - offset;
2049        struct sk_buff *frag_iter;
2050        int pos = 0;
2051
2052        /* Copy header. */
2053        if (copy > 0) {
2054                if (copy > len)
2055                        copy = len;
2056                csum = csum_partial_copy_nocheck(skb->data + offset, to,
2057                                                 copy, csum);
2058                if ((len -= copy) == 0)
2059                        return csum;
2060                offset += copy;
2061                to     += copy;
2062                pos     = copy;
2063        }
2064
2065        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
2066                int end;
2067
2068                WARN_ON(start > offset + len);
2069
2070                end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
2071                if ((copy = end - offset) > 0) {
2072                        __wsum csum2;
2073                        u8 *vaddr;
2074                        skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
2075
2076                        if (copy > len)
2077                                copy = len;
2078                        vaddr = kmap_atomic(skb_frag_page(frag));
2079                        csum2 = csum_partial_copy_nocheck(vaddr +
2080                                                          frag->page_offset +
2081                                                          offset - start, to,
2082                                                          copy, 0);
2083                        kunmap_atomic(vaddr);
2084                        csum = csum_block_add(csum, csum2, pos);
2085                        if (!(len -= copy))
2086                                return csum;
2087                        offset += copy;
2088                        to     += copy;
2089                        pos    += copy;
2090                }
2091                start = end;
2092        }
2093
2094        skb_walk_frags(skb, frag_iter) {
2095                __wsum csum2;
2096                int end;
2097
2098                WARN_ON(start > offset + len);
2099
2100                end = start + frag_iter->len;
2101                if ((copy = end - offset) > 0) {
2102                        if (copy > len)
2103                                copy = len;
2104                        csum2 = skb_copy_and_csum_bits(frag_iter,
2105                                                       offset - start,
2106                                                       to, copy, 0);
2107                        csum = csum_block_add(csum, csum2, pos);
2108                        if ((len -= copy) == 0)
2109                                return csum;
2110                        offset += copy;
2111                        to     += copy;
2112                        pos    += copy;
2113                }
2114                start = end;
2115        }
2116        BUG_ON(len);
2117        return csum;
2118}
2119EXPORT_SYMBOL(skb_copy_and_csum_bits);
2120
2121void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to)
2122{
2123        __wsum csum;
2124        long csstart;
2125
2126        if (skb->ip_summed == CHECKSUM_PARTIAL)
2127                csstart = skb_checksum_start_offset(skb);
2128        else
2129                csstart = skb_headlen(skb);
2130
2131        BUG_ON(csstart > skb_headlen(skb));
2132
2133        skb_copy_from_linear_data(skb, to, csstart);
2134
2135        csum = 0;
2136        if (csstart != skb->len)
2137                csum = skb_copy_and_csum_bits(skb, csstart, to + csstart,
2138                                              skb->len - csstart, 0);
2139
2140        if (skb->ip_summed == CHECKSUM_PARTIAL) {
2141                long csstuff = csstart + skb->csum_offset;
2142
2143                *((__sum16 *)(to + csstuff)) = csum_fold(csum);
2144        }
2145}
2146EXPORT_SYMBOL(skb_copy_and_csum_dev);
2147
2148/**
2149 *      skb_dequeue - remove from the head of the queue
2150 *      @list: list to dequeue from
2151 *
2152 *      Remove the head of the list. The list lock is taken so the function
2153 *      may be used safely with other locking list functions. The head item is
2154 *      returned or %NULL if the list is empty.
2155 */
2156
2157struct sk_buff *skb_dequeue(struct sk_buff_head *list)
2158{
2159        unsigned long flags;
2160        struct sk_buff *result;
2161
2162        spin_lock_irqsave(&list->lock, flags);
2163        result = __skb_dequeue(list);
2164        spin_unlock_irqrestore(&list->lock, flags);
2165        return result;
2166}
2167EXPORT_SYMBOL(skb_dequeue);
2168
2169/**
2170 *      skb_dequeue_tail - remove from the tail of the queue
2171 *      @list: list to dequeue from
2172 *
2173 *      Remove the tail of the list. The list lock is taken so the function
2174 *      may be used safely with other locking list functions. The tail item is
2175 *      returned or %NULL if the list is empty.
2176 */
2177struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list)
2178{
2179        unsigned long flags;
2180        struct sk_buff *result;
2181
2182        spin_lock_irqsave(&list->lock, flags);
2183        result = __skb_dequeue_tail(list);
2184        spin_unlock_irqrestore(&list->lock, flags);
2185        return result;
2186}
2187EXPORT_SYMBOL(skb_dequeue_tail);
2188
2189/**
2190 *      skb_queue_purge - empty a list
2191 *      @list: list to empty
2192 *
2193 *      Delete all buffers on an &sk_buff list. Each buffer is removed from
2194 *      the list and one reference dropped. This function takes the list
2195 *      lock and is atomic with respect to other list locking functions.
2196 */
2197void skb_queue_purge(struct sk_buff_head *list)
2198{
2199        struct sk_buff *skb;
2200        while ((skb = skb_dequeue(list)) != NULL)
2201                kfree_skb(skb);
2202}
2203EXPORT_SYMBOL(skb_queue_purge);
2204
2205/**
2206 *      skb_queue_head - queue a buffer at the list head
2207 *      @list: list to use
2208 *      @newsk: buffer to queue
2209 *
2210 *      Queue a buffer at the start of the list. This function takes the
2211 *      list lock and can be used safely with other locking &sk_buff functions
2212 *      safely.
2213 *
2214 *      A buffer cannot be placed on two lists at the same time.
2215 */
2216void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk)
2217{
2218        unsigned long flags;
2219
2220        spin_lock_irqsave(&list->lock, flags);
2221        __skb_queue_head(list, newsk);
2222        spin_unlock_irqrestore(&list->lock, flags);
2223}
2224EXPORT_SYMBOL(skb_queue_head);
2225
2226/**
2227 *      skb_queue_tail - queue a buffer at the list tail
2228 *      @list: list to use
2229 *      @newsk: buffer to queue
2230 *
2231 *      Queue a buffer at the tail of the list. This function takes the
2232 *      list lock and can be used safely with other locking &sk_buff functions
2233 *      safely.
2234 *
2235 *      A buffer cannot be placed on two lists at the same time.
2236 */
2237void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk)
2238{
2239        unsigned long flags;
2240
2241        spin_lock_irqsave(&list->lock, flags);
2242        __skb_queue_tail(list, newsk);
2243        spin_unlock_irqrestore(&list->lock, flags);
2244}
2245EXPORT_SYMBOL(skb_queue_tail);
2246
2247/**
2248 *      skb_unlink      -       remove a buffer from a list
2249 *      @skb: buffer to remove
2250 *      @list: list to use
2251 *
2252 *      Remove a packet from a list. The list locks are taken and this
2253 *      function is atomic with respect to other list locked calls
2254 *
2255 *      You must know what list the SKB is on.
2256 */
2257void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list)
2258{
2259        unsigned long flags;
2260
2261        spin_lock_irqsave(&list->lock, flags);
2262        __skb_unlink(skb, list);
2263        spin_unlock_irqrestore(&list->lock, flags);
2264}
2265EXPORT_SYMBOL(skb_unlink);
2266
2267/**
2268 *      skb_append      -       append a buffer
2269 *      @old: buffer to insert after
2270 *      @newsk: buffer to insert
2271 *      @list: list to use
2272 *
2273 *      Place a packet after a given packet in a list. The list locks are taken
2274 *      and this function is atomic with respect to other list locked calls.
2275 *      A buffer cannot be placed on two lists at the same time.
2276 */
2277void skb_append(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2278{
2279        unsigned long flags;
2280
2281        spin_lock_irqsave(&list->lock, flags);
2282        __skb_queue_after(list, old, newsk);
2283        spin_unlock_irqrestore(&list->lock, flags);
2284}
2285EXPORT_SYMBOL(skb_append);
2286
2287/**
2288 *      skb_insert      -       insert a buffer
2289 *      @old: buffer to insert before
2290 *      @newsk: buffer to insert
2291 *      @list: list to use
2292 *
2293 *      Place a packet before a given packet in a list. The list locks are
2294 *      taken and this function is atomic with respect to other list locked
2295 *      calls.
2296 *
2297 *      A buffer cannot be placed on two lists at the same time.
2298 */
2299void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list)
2300{
2301        unsigned long flags;
2302
2303        spin_lock_irqsave(&list->lock, flags);
2304        __skb_insert(newsk, old->prev, old, list);
2305        spin_unlock_irqrestore(&list->lock, flags);
2306}
2307EXPORT_SYMBOL(skb_insert);
2308
2309static inline void skb_split_inside_header(struct sk_buff *skb,
2310                                           struct sk_buff* skb1,
2311                                           const u32 len, const int pos)
2312{
2313        int i;
2314
2315        skb_copy_from_linear_data_offset(skb, len, skb_put(skb1, pos - len),
2316                                         pos - len);
2317        /* And move data appendix as is. */
2318        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++)
2319                skb_shinfo(skb1)->frags[i] = skb_shinfo(skb)->frags[i];
2320
2321        skb_shinfo(skb1)->nr_frags = skb_shinfo(skb)->nr_frags;
2322        skb_shinfo(skb)->nr_frags  = 0;
2323        skb1->data_len             = skb->data_len;
2324        skb1->len                  += skb1->data_len;
2325        skb->data_len              = 0;
2326        skb->len                   = len;
2327        skb_set_tail_pointer(skb, len);
2328}
2329
2330static inline void skb_split_no_header(struct sk_buff *skb,
2331                                       struct sk_buff* skb1,
2332                                       const u32 len, int pos)
2333{
2334        int i, k = 0;
2335        const int nfrags = skb_shinfo(skb)->nr_frags;
2336
2337        skb_shinfo(skb)->nr_frags = 0;
2338        skb1->len                 = skb1->data_len = skb->len - len;
2339        skb->len                  = len;
2340        skb->data_len             = len - pos;
2341
2342        for (i = 0; i < nfrags; i++) {
2343                int size = skb_frag_size(&skb_shinfo(skb)->frags[i]);
2344
2345                if (pos + size > len) {
2346                        skb_shinfo(skb1)->frags[k] = skb_shinfo(skb)->frags[i];
2347
2348                        if (pos < len) {
2349                                /* Split frag.
2350                                 * We have two variants in this case:
2351                                 * 1. Move all the frag to the second
2352                                 *    part, if it is possible. F.e.
2353                                 *    this approach is mandatory for TUX,
2354                                 *    where splitting is expensive.
2355                                 * 2. Split is accurately. We make this.
2356                                 */
2357                                skb_frag_ref(skb, i);
2358                                skb_shinfo(skb1)->frags[0].page_offset += len - pos;
2359                                skb_frag_size_sub(&skb_shinfo(skb1)->frags[0], len - pos);
2360                                skb_frag_size_set(&skb_shinfo(skb)->frags[i], len - pos);
2361                                skb_shinfo(skb)->nr_frags++;
2362                        }
2363                        k++;
2364                } else
2365                        skb_shinfo(skb)->nr_frags++;
2366                pos += size;
2367        }
2368        skb_shinfo(skb1)->nr_frags = k;
2369}
2370
2371/**
2372 * skb_split - Split fragmented skb to two parts at length len.
2373 * @skb: the buffer to split
2374 * @skb1: the buffer to receive the second part
2375 * @len: new length for skb
2376 */
2377void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len)
2378{
2379        int pos = skb_headlen(skb);
2380
2381        if (len < pos)  /* Split line is inside header. */
2382                skb_split_inside_header(skb, skb1, len, pos);
2383        else            /* Second chunk has no header, nothing to copy. */
2384                skb_split_no_header(skb, skb1, len, pos);
2385}
2386EXPORT_SYMBOL(skb_split);
2387
2388/* Shifting from/to a cloned skb is a no-go.
2389 *
2390 * Caller cannot keep skb_shinfo related pointers past calling here!
2391 */
2392static int skb_prepare_for_shift(struct sk_buff *skb)
2393{
2394        return skb_cloned(skb) && pskb_expand_head(skb, 0, 0, GFP_ATOMIC);
2395}
2396
2397/**
2398 * skb_shift - Shifts paged data partially from skb to another
2399 * @tgt: buffer into which tail data gets added
2400 * @skb: buffer from which the paged data comes from
2401 * @shiftlen: shift up to this many bytes
2402 *
2403 * Attempts to shift up to shiftlen worth of bytes, which may be less than
2404 * the length of the skb, from skb to tgt. Returns number bytes shifted.
2405 * It's up to caller to free skb if everything was shifted.
2406 *
2407 * If @tgt runs out of frags, the whole operation is aborted.
2408 *
2409 * Skb cannot include anything else but paged data while tgt is allowed
2410 * to have non-paged data as well.
2411 *
2412 * TODO: full sized shift could be optimized but that would need
2413 * specialized skb free'er to handle frags without up-to-date nr_frags.
2414 */
2415int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen)
2416{
2417        int from, to, merge, todo;
2418        struct skb_frag_struct *fragfrom, *fragto;
2419
2420        BUG_ON(shiftlen > skb->len);
2421        BUG_ON(skb_headlen(skb));       /* Would corrupt stream */
2422
2423        todo = shiftlen;
2424        from = 0;
2425        to = skb_shinfo(tgt)->nr_frags;
2426        fragfrom = &skb_shinfo(skb)->frags[from];
2427
2428        /* Actual merge is delayed until the point when we know we can
2429         * commit all, so that we don't have to undo partial changes
2430         */
2431        if (!to ||
2432            !skb_can_coalesce(tgt, to, skb_frag_page(fragfrom),
2433                              fragfrom->page_offset)) {
2434                merge = -1;
2435        } else {
2436                merge = to - 1;
2437
2438                todo -= skb_frag_size(fragfrom);
2439                if (todo < 0) {
2440                        if (skb_prepare_for_shift(skb) ||
2441                            skb_prepare_for_shift(tgt))
2442                                return 0;
2443
2444                        /* All previous frag pointers might be stale! */
2445                        fragfrom = &skb_shinfo(skb)->frags[from];
2446                        fragto = &skb_shinfo(tgt)->frags[merge];
2447
2448                        skb_frag_size_add(fragto, shiftlen);
2449                        skb_frag_size_sub(fragfrom, shiftlen);
2450                        fragfrom->page_offset += shiftlen;
2451
2452                        goto onlymerged;
2453                }
2454
2455                from++;
2456        }
2457
2458        /* Skip full, not-fitting skb to avoid expensive operations */
2459        if ((shiftlen == skb->len) &&
2460            (skb_shinfo(skb)->nr_frags - from) > (MAX_SKB_FRAGS - to))
2461                return 0;
2462
2463        if (skb_prepare_for_shift(skb) || skb_prepare_for_shift(tgt))
2464                return 0;
2465
2466        while ((todo > 0) && (from < skb_shinfo(skb)->nr_frags)) {
2467                if (to == MAX_SKB_FRAGS)
2468                        return 0;
2469
2470                fragfrom = &skb_shinfo(skb)->frags[from];
2471                fragto = &skb_shinfo(tgt)->frags[to];
2472
2473                if (todo >= skb_frag_size(fragfrom)) {
2474                        *fragto = *fragfrom;
2475                        todo -= skb_frag_size(fragfrom);
2476                        from++;
2477                        to++;
2478
2479                } else {
2480                        __skb_frag_ref(fragfrom);
2481                        fragto->page = fragfrom->page;
2482                        fragto->page_offset = fragfrom->page_offset;
2483                        skb_frag_size_set(fragto, todo);
2484
2485                        fragfrom->page_offset += todo;
2486                        skb_frag_size_sub(fragfrom, todo);
2487                        todo = 0;
2488
2489                        to++;
2490                        break;
2491                }
2492        }
2493
2494        /* Ready to "commit" this state change to tgt */
2495        skb_shinfo(tgt)->nr_frags = to;
2496
2497        if (merge >= 0) {
2498                fragfrom = &skb_shinfo(skb)->frags[0];
2499                fragto = &skb_shinfo(tgt)->frags[merge];
2500
2501                skb_frag_size_add(fragto, skb_frag_size(fragfrom));
2502                __skb_frag_unref(fragfrom);
2503        }
2504
2505        /* Reposition in the original skb */
2506        to = 0;
2507        while (from < skb_shinfo(skb)->nr_frags)
2508                skb_shinfo(skb)->frags[to++] = skb_shinfo(skb)->frags[from++];
2509        skb_shinfo(skb)->nr_frags = to;
2510
2511        BUG_ON(todo > 0 && !skb_shinfo(skb)->nr_frags);
2512
2513onlymerged:
2514        /* Most likely the tgt won't ever need its checksum anymore, skb on
2515         * the other hand might need it if it needs to be resent
2516         */
2517        tgt->ip_summed = CHECKSUM_PARTIAL;
2518        skb->ip_summed = CHECKSUM_PARTIAL;
2519
2520        /* Yak, is it really working this way? Some helper please? */
2521        skb->len -= shiftlen;
2522        skb->data_len -= shiftlen;
2523        skb->truesize -= shiftlen;
2524        tgt->len += shiftlen;
2525        tgt->data_len += shiftlen;
2526        tgt->truesize += shiftlen;
2527
2528        return shiftlen;
2529}
2530
2531/**
2532 * skb_prepare_seq_read - Prepare a sequential read of skb data
2533 * @skb: the buffer to read
2534 * @from: lower offset of data to be read
2535 * @to: upper offset of data to be read
2536 * @st: state variable
2537 *
2538 * Initializes the specified state variable. Must be called before
2539 * invoking skb_seq_read() for the first time.
2540 */
2541void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from,
2542                          unsigned int to, struct skb_seq_state *st)
2543{
2544        st->lower_offset = from;
2545        st->upper_offset = to;
2546        st->root_skb = st->cur_skb = skb;
2547        st->frag_idx = st->stepped_offset = 0;
2548        st->frag_data = NULL;
2549}
2550EXPORT_SYMBOL(skb_prepare_seq_read);
2551
2552/**
2553 * skb_seq_read - Sequentially read skb data
2554 * @consumed: number of bytes consumed by the caller so far
2555 * @data: destination pointer for data to be returned
2556 * @st: state variable
2557 *
2558 * Reads a block of skb data at &consumed relative to the
2559 * lower offset specified to skb_prepare_seq_read(). Assigns
2560 * the head of the data block to &data and returns the length
2561 * of the block or 0 if the end of the skb data or the upper
2562 * offset has been reached.
2563 *
2564 * The caller is not required to consume all of the data
2565 * returned, i.e. &consumed is typically set to the number
2566 * of bytes already consumed and the next call to
2567 * skb_seq_read() will return the remaining part of the block.
2568 *
2569 * Note 1: The size of each block of data returned can be arbitrary,
2570 *       this limitation is the cost for zerocopy seqeuental
2571 *       reads of potentially non linear data.
2572 *
2573 * Note 2: Fragment lists within fragments are not implemented
2574 *       at the moment, state->root_skb could be replaced with
2575 *       a stack for this purpose.
2576 */
2577unsigned int skb_seq_read(unsigned int consumed, const u8 **data,
2578                          struct skb_seq_state *st)
2579{
2580        unsigned int block_limit, abs_offset = consumed + st->lower_offset;
2581        skb_frag_t *frag;
2582
2583        if (unlikely(abs_offset >= st->upper_offset))
2584                return 0;
2585
2586next_skb:
2587        block_limit = skb_headlen(st->cur_skb) + st->stepped_offset;
2588
2589        if (abs_offset < block_limit && !st->frag_data) {
2590                *data = st->cur_skb->data + (abs_offset - st->stepped_offset);
2591                return block_limit - abs_offset;
2592        }
2593
2594        if (st->frag_idx == 0 && !st->frag_data)
2595                st->stepped_offset += skb_headlen(st->cur_skb);
2596
2597        while (st->frag_idx < skb_shinfo(st->cur_skb)->nr_frags) {
2598                frag = &skb_shinfo(st->cur_skb)->frags[st->frag_idx];
2599                block_limit = skb_frag_size(frag) + st->stepped_offset;
2600
2601                if (abs_offset < block_limit) {
2602                        if (!st->frag_data)
2603                                st->frag_data = kmap_atomic(skb_frag_page(frag));
2604
2605                        *data = (u8 *) st->frag_data + frag->page_offset +
2606                                (abs_offset - st->stepped_offset);
2607
2608                        return block_limit - abs_offset;
2609                }
2610
2611                if (st->frag_data) {
2612                        kunmap_atomic(st->frag_data);
2613                        st->frag_data = NULL;
2614                }
2615
2616                st->frag_idx++;
2617                st->stepped_offset += skb_frag_size(frag);
2618        }
2619
2620        if (st->frag_data) {
2621                kunmap_atomic(st->frag_data);
2622                st->frag_data = NULL;
2623        }
2624
2625        if (st->root_skb == st->cur_skb && skb_has_frag_list(st->root_skb)) {
2626                st->cur_skb = skb_shinfo(st->root_skb)->frag_list;
2627                st->frag_idx = 0;
2628                goto next_skb;
2629        } else if (st->cur_skb->next) {
2630                st->cur_skb = st->cur_skb->next;
2631                st->frag_idx = 0;
2632                goto next_skb;
2633        }
2634
2635        return 0;
2636}
2637EXPORT_SYMBOL(skb_seq_read);
2638
2639/**
2640 * skb_abort_seq_read - Abort a sequential read of skb data
2641 * @st: state variable
2642 *
2643 * Must be called if skb_seq_read() was not called until it
2644 * returned 0.
2645 */
2646void skb_abort_seq_read(struct skb_seq_state *st)
2647{
2648        if (st->frag_data)
2649                kunmap_atomic(st->frag_data);
2650}
2651EXPORT_SYMBOL(skb_abort_seq_read);
2652
2653#define TS_SKB_CB(state)        ((struct skb_seq_state *) &((state)->cb))
2654
2655static unsigned int skb_ts_get_next_block(unsigned int offset, const u8 **text,
2656                                          struct ts_config *conf,
2657                                          struct ts_state *state)
2658{
2659        return skb_seq_read(offset, text, TS_SKB_CB(state));
2660}
2661
2662static void skb_ts_finish(struct ts_config *conf, struct ts_state *state)
2663{
2664        skb_abort_seq_read(TS_SKB_CB(state));
2665}
2666
2667/**
2668 * skb_find_text - Find a text pattern in skb data
2669 * @skb: the buffer to look in
2670 * @from: search offset
2671 * @to: search limit
2672 * @config: textsearch configuration
2673 * @state: uninitialized textsearch state variable
2674 *
2675 * Finds a pattern in the skb data according to the specified
2676 * textsearch configuration. Use textsearch_next() to retrieve
2677 * subsequent occurrences of the pattern. Returns the offset
2678 * to the first occurrence or UINT_MAX if no match was found.
2679 */
2680unsigned int skb_find_text(struct sk_buff *skb, unsigned int from,
2681                           unsigned int to, struct ts_config *config,
2682                           struct ts_state *state)
2683{
2684        unsigned int ret;
2685
2686        config->get_next_block = skb_ts_get_next_block;
2687        config->finish = skb_ts_finish;
2688
2689        skb_prepare_seq_read(skb, from, to, TS_SKB_CB(state));
2690
2691        ret = textsearch_find(config, state);
2692        return (ret <= to - from ? ret : UINT_MAX);
2693}
2694EXPORT_SYMBOL(skb_find_text);
2695
2696/**
2697 * skb_append_datato_frags - append the user data to a skb
2698 * @sk: sock  structure
2699 * @skb: skb structure to be appened with user data.
2700 * @getfrag: call back function to be used for getting the user data
2701 * @from: pointer to user message iov
2702 * @length: length of the iov message
2703 *
2704 * Description: This procedure append the user data in the fragment part
2705 * of the skb if any page alloc fails user this procedure returns  -ENOMEM
2706 */
2707int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb,
2708                        int (*getfrag)(void *from, char *to, int offset,
2709                                        int len, int odd, struct sk_buff *skb),
2710                        void *from, int length)
2711{
2712        int frg_cnt = 0;
2713        skb_frag_t *frag = NULL;
2714        struct page *page = NULL;
2715        int copy, left;
2716        int offset = 0;
2717        int ret;
2718
2719        do {
2720                /* Return error if we don't have space for new frag */
2721                frg_cnt = skb_shinfo(skb)->nr_frags;
2722                if (frg_cnt >= MAX_SKB_FRAGS)
2723                        return -EFAULT;
2724
2725                /* allocate a new page for next frag */
2726                page = alloc_pages(sk->sk_allocation, 0);
2727
2728                /* If alloc_page fails just return failure and caller will
2729                 * free previous allocated pages by doing kfree_skb()
2730                 */
2731                if (page == NULL)
2732                        return -ENOMEM;
2733
2734                /* initialize the next frag */
2735                skb_fill_page_desc(skb, frg_cnt, page, 0, 0);
2736                skb->truesize += PAGE_SIZE;
2737                atomic_add(PAGE_SIZE, &sk->sk_wmem_alloc);
2738
2739                /* get the new initialized frag */
2740                frg_cnt = skb_shinfo(skb)->nr_frags;
2741                frag = &skb_shinfo(skb)->frags[frg_cnt - 1];
2742
2743                /* copy the user data to page */
2744                left = PAGE_SIZE - frag->page_offset;
2745                copy = (length > left)? left : length;
2746
2747                ret = getfrag(from, skb_frag_address(frag) + skb_frag_size(frag),
2748                            offset, copy, 0, skb);
2749                if (ret < 0)
2750                        return -EFAULT;
2751
2752                /* copy was successful so update the size parameters */
2753                skb_frag_size_add(frag, copy);
2754                skb->len += copy;
2755                skb->data_len += copy;
2756                offset += copy;
2757                length -= copy;
2758
2759        } while (length > 0);
2760
2761        return 0;
2762}
2763EXPORT_SYMBOL(skb_append_datato_frags);
2764
2765/**
2766 *      skb_pull_rcsum - pull skb and update receive checksum
2767 *      @skb: buffer to update
2768 *      @len: length of data pulled
2769 *
2770 *      This function performs an skb_pull on the packet and updates
2771 *      the CHECKSUM_COMPLETE checksum.  It should be used on
2772 *      receive path processing instead of skb_pull unless you know
2773 *      that the checksum difference is zero (e.g., a valid IP header)
2774 *      or you are setting ip_summed to CHECKSUM_NONE.
2775 */
2776unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len)
2777{
2778        BUG_ON(len > skb->len);
2779        skb->len -= len;
2780        BUG_ON(skb->len < skb->data_len);
2781        skb_postpull_rcsum(skb, skb->data, len);
2782        return skb->data += len;
2783}
2784EXPORT_SYMBOL_GPL(skb_pull_rcsum);
2785
2786/**
2787 *      skb_segment - Perform protocol segmentation on skb.
2788 *      @skb: buffer to segment
2789 *      @features: features for the output path (see dev->features)
2790 *
2791 *      This function performs segmentation on the given skb.  It returns
2792 *      a pointer to the first in a list of new skbs for the segments.
2793 *      In case of error it returns ERR_PTR(err).
2794 */
2795struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features)
2796{
2797        struct sk_buff *segs = NULL;
2798        struct sk_buff *tail = NULL;
2799        struct sk_buff *fskb = skb_shinfo(skb)->frag_list;
2800        unsigned int mss = skb_shinfo(skb)->gso_size;
2801        unsigned int doffset = skb->data - skb_mac_header(skb);
2802        unsigned int offset = doffset;
2803        unsigned int headroom;
2804        unsigned int len;
2805        int sg = !!(features & NETIF_F_SG);
2806        int nfrags = skb_shinfo(skb)->nr_frags;
2807        int err = -ENOMEM;
2808        int i = 0;
2809        int pos;
2810
2811        __skb_push(skb, doffset);
2812        headroom = skb_headroom(skb);
2813        pos = skb_headlen(skb);
2814
2815        do {
2816                struct sk_buff *nskb;
2817                skb_frag_t *frag;
2818                int hsize;
2819                int size;
2820
2821                len = skb->len - offset;
2822                if (len > mss)
2823                        len = mss;
2824
2825                hsize = skb_headlen(skb) - offset;
2826                if (hsize < 0)
2827                        hsize = 0;
2828                if (hsize > len || !sg)
2829                        hsize = len;
2830
2831                if (!hsize && i >= nfrags) {
2832                        BUG_ON(fskb->len != len);
2833
2834                        pos += len;
2835                        nskb = skb_clone(fskb, GFP_ATOMIC);
2836                        fskb = fskb->next;
2837
2838                        if (unlikely(!nskb))
2839                                goto err;
2840
2841                        hsize = skb_end_offset(nskb);
2842                        if (skb_cow_head(nskb, doffset + headroom)) {
2843                                kfree_skb(nskb);
2844                                goto err;
2845                        }
2846
2847                        nskb->truesize += skb_end_offset(nskb) - hsize;
2848                        skb_release_head_state(nskb);
2849                        __skb_push(nskb, doffset);
2850                } else {
2851                        nskb = __alloc_skb(hsize + doffset + headroom,
2852                                           GFP_ATOMIC, skb_alloc_rx_flag(skb),
2853                                           NUMA_NO_NODE);
2854
2855                        if (unlikely(!nskb))
2856                                goto err;
2857
2858                        skb_reserve(nskb, headroom);
2859                        __skb_put(nskb, doffset);
2860                }
2861
2862                if (segs)
2863                        tail->next = nskb;
2864                else
2865                        segs = nskb;
2866                tail = nskb;
2867
2868                __copy_skb_header(nskb, skb);
2869                nskb->mac_len = skb->mac_len;
2870
2871                /* nskb and skb might have different headroom */
2872                if (nskb->ip_summed == CHECKSUM_PARTIAL)
2873                        nskb->csum_start += skb_headroom(nskb) - headroom;
2874
2875                skb_reset_mac_header(nskb);
2876                skb_set_network_header(nskb, skb->mac_len);
2877                nskb->transport_header = (nskb->network_header +
2878                                          skb_network_header_len(skb));
2879                skb_copy_from_linear_data(skb, nskb->data, doffset);
2880
2881                if (fskb != skb_shinfo(skb)->frag_list)
2882                        continue;
2883
2884                if (!sg) {
2885                        nskb->ip_summed = CHECKSUM_NONE;
2886                        nskb->csum = skb_copy_and_csum_bits(skb, offset,
2887                                                            skb_put(nskb, len),
2888                                                            len, 0);
2889                        continue;
2890                }
2891
2892                frag = skb_shinfo(nskb)->frags;
2893
2894                skb_copy_from_linear_data_offset(skb, offset,
2895                                                 skb_put(nskb, hsize), hsize);
2896
2897                while (pos < offset + len && i < nfrags) {
2898                        *frag = skb_shinfo(skb)->frags[i];
2899                        __skb_frag_ref(frag);
2900                        size = skb_frag_size(frag);
2901
2902                        if (pos < offset) {
2903                                frag->page_offset += offset - pos;
2904                                skb_frag_size_sub(frag, offset - pos);
2905                        }
2906
2907                        skb_shinfo(nskb)->nr_frags++;
2908
2909                        if (pos + size <= offset + len) {
2910                                i++;
2911                                pos += size;
2912                        } else {
2913                                skb_frag_size_sub(frag, pos + size - (offset + len));
2914                                goto skip_fraglist;
2915                        }
2916
2917                        frag++;
2918                }
2919
2920                if (pos < offset + len) {
2921                        struct sk_buff *fskb2 = fskb;
2922
2923                        BUG_ON(pos + fskb->len != offset + len);
2924
2925                        pos += fskb->len;
2926                        fskb = fskb->next;
2927
2928                        if (fskb2->next) {
2929                                fskb2 = skb_clone(fskb2, GFP_ATOMIC);
2930                                if (!fskb2)
2931                                        goto err;
2932                        } else
2933                                skb_get(fskb2);
2934
2935                        SKB_FRAG_ASSERT(nskb);
2936                        skb_shinfo(nskb)->frag_list = fskb2;
2937                }
2938
2939skip_fraglist:
2940                nskb->data_len = len - hsize;
2941                nskb->len += nskb->data_len;
2942                nskb->truesize += nskb->data_len;
2943        } while ((offset += len) < skb->len);
2944
2945        return segs;
2946
2947err:
2948        while ((skb = segs)) {
2949                segs = skb->next;
2950                kfree_skb(skb);
2951        }
2952        return ERR_PTR(err);
2953}
2954EXPORT_SYMBOL_GPL(skb_segment);
2955
2956int skb_gro_receive(struct sk_buff **head, struct sk_buff *skb)
2957{
2958        struct sk_buff *p = *head;
2959        struct sk_buff *nskb;
2960        struct skb_shared_info *skbinfo = skb_shinfo(skb);
2961        struct skb_shared_info *pinfo = skb_shinfo(p);
2962        unsigned int headroom;
2963        unsigned int len = skb_gro_len(skb);
2964        unsigned int offset = skb_gro_offset(skb);
2965        unsigned int headlen = skb_headlen(skb);
2966        unsigned int delta_truesize;
2967
2968        if (p->len + len >= 65536)
2969                return -E2BIG;
2970
2971        if (pinfo->frag_list)
2972                goto merge;
2973        else if (headlen <= offset) {
2974                skb_frag_t *frag;
2975                skb_frag_t *frag2;
2976                int i = skbinfo->nr_frags;
2977                int nr_frags = pinfo->nr_frags + i;
2978
2979                offset -= headlen;
2980
2981                if (nr_frags > MAX_SKB_FRAGS)
2982                        return -E2BIG;
2983
2984                pinfo->nr_frags = nr_frags;
2985                skbinfo->nr_frags = 0;
2986
2987                frag = pinfo->frags + nr_frags;
2988                frag2 = skbinfo->frags + i;
2989                do {
2990                        *--frag = *--frag2;
2991                } while (--i);
2992
2993                frag->page_offset += offset;
2994                skb_frag_size_sub(frag, offset);
2995
2996                /* all fragments truesize : remove (head size + sk_buff) */
2997                delta_truesize = skb->truesize -
2998                                 SKB_TRUESIZE(skb_end_offset(skb));
2999
3000                skb->truesize -= skb->data_len;
3001                skb->len -= skb->data_len;
3002                skb->data_len = 0;
3003
3004                NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE;
3005                goto done;
3006        } else if (skb->head_frag) {
3007                int nr_frags = pinfo->nr_frags;
3008                skb_frag_t *frag = pinfo->frags + nr_frags;
3009                struct page *page = virt_to_head_page(skb->head);
3010                unsigned int first_size = headlen - offset;
3011                unsigned int first_offset;
3012
3013                if (nr_frags + 1 + skbinfo->nr_frags > MAX_SKB_FRAGS)
3014                        return -E2BIG;
3015
3016                first_offset = skb->data -
3017                               (unsigned char *)page_address(page) +
3018                               offset;
3019
3020                pinfo->nr_frags = nr_frags + 1 + skbinfo->nr_frags;
3021
3022                frag->page.p      = page;
3023                frag->page_offset = first_offset;
3024                skb_frag_size_set(frag, first_size);
3025
3026                memcpy(frag + 1, skbinfo->frags, sizeof(*frag) * skbinfo->nr_frags);
3027                /* We dont need to clear skbinfo->nr_frags here */
3028
3029                delta_truesize = skb->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3030                NAPI_GRO_CB(skb)->free = NAPI_GRO_FREE_STOLEN_HEAD;
3031                goto done;
3032        } else if (skb_gro_len(p) != pinfo->gso_size)
3033                return -E2BIG;
3034
3035        headroom = skb_headroom(p);
3036        nskb = alloc_skb(headroom + skb_gro_offset(p), GFP_ATOMIC);
3037        if (unlikely(!nskb))
3038                return -ENOMEM;
3039
3040        __copy_skb_header(nskb, p);
3041        nskb->mac_len = p->mac_len;
3042
3043        skb_reserve(nskb, headroom);
3044        __skb_put(nskb, skb_gro_offset(p));
3045
3046        skb_set_mac_header(nskb, skb_mac_header(p) - p->data);
3047        skb_set_network_header(nskb, skb_network_offset(p));
3048        skb_set_transport_header(nskb, skb_transport_offset(p));
3049
3050        __skb_pull(p, skb_gro_offset(p));
3051        memcpy(skb_mac_header(nskb), skb_mac_header(p),
3052               p->data - skb_mac_header(p));
3053
3054        *NAPI_GRO_CB(nskb) = *NAPI_GRO_CB(p);
3055        skb_shinfo(nskb)->frag_list = p;
3056        skb_shinfo(nskb)->gso_size = pinfo->gso_size;
3057        pinfo->gso_size = 0;
3058        skb_header_release(p);
3059        nskb->prev = p;
3060
3061        nskb->data_len += p->len;
3062        nskb->truesize += p->truesize;
3063        nskb->len += p->len;
3064
3065        *head = nskb;
3066        nskb->next = p->next;
3067        p->next = NULL;
3068
3069        p = nskb;
3070
3071merge:
3072        delta_truesize = skb->truesize;
3073        if (offset > headlen) {
3074                unsigned int eat = offset - headlen;
3075
3076                skbinfo->frags[0].page_offset += eat;
3077                skb_frag_size_sub(&skbinfo->frags[0], eat);
3078                skb->data_len -= eat;
3079                skb->len -= eat;
3080                offset = headlen;
3081        }
3082
3083        __skb_pull(skb, offset);
3084
3085        p->prev->next = skb;
3086        p->prev = skb;
3087        skb_header_release(skb);
3088
3089done:
3090        NAPI_GRO_CB(p)->count++;
3091        p->data_len += len;
3092        p->truesize += delta_truesize;
3093        p->len += len;
3094
3095        NAPI_GRO_CB(skb)->same_flow = 1;
3096        return 0;
3097}
3098EXPORT_SYMBOL_GPL(skb_gro_receive);
3099
3100void __init skb_init(void)
3101{
3102        skbuff_head_cache = kmem_cache_create("skbuff_head_cache",
3103                                              sizeof(struct sk_buff),
3104                                              0,
3105                                              SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3106                                              NULL);
3107        skbuff_fclone_cache = kmem_cache_create("skbuff_fclone_cache",
3108                                                (2*sizeof(struct sk_buff)) +
3109                                                sizeof(atomic_t),
3110                                                0,
3111                                                SLAB_HWCACHE_ALIGN|SLAB_PANIC,
3112                                                NULL);
3113}
3114
3115/**
3116 *      skb_to_sgvec - Fill a scatter-gather list from a socket buffer
3117 *      @skb: Socket buffer containing the buffers to be mapped
3118 *      @sg: The scatter-gather list to map into
3119 *      @offset: The offset into the buffer's contents to start mapping
3120 *      @len: Length of buffer space to be mapped
3121 *
3122 *      Fill the specified scatter-gather list with mappings/pointers into a
3123 *      region of the buffer space attached to a socket buffer.
3124 */
3125static int
3126__skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3127{
3128        int start = skb_headlen(skb);
3129        int i, copy = start - offset;
3130        struct sk_buff *frag_iter;
3131        int elt = 0;
3132
3133        if (copy > 0) {
3134                if (copy > len)
3135                        copy = len;
3136                sg_set_buf(sg, skb->data + offset, copy);
3137                elt++;
3138                if ((len -= copy) == 0)
3139                        return elt;
3140                offset += copy;
3141        }
3142
3143        for (i = 0; i < skb_shinfo(skb)->nr_frags; i++) {
3144                int end;
3145
3146                WARN_ON(start > offset + len);
3147
3148                end = start + skb_frag_size(&skb_shinfo(skb)->frags[i]);
3149                if ((copy = end - offset) > 0) {
3150                        skb_frag_t *frag = &skb_shinfo(skb)->frags[i];
3151
3152                        if (copy > len)
3153                                copy = len;
3154                        sg_set_page(&sg[elt], skb_frag_page(frag), copy,
3155                                        frag->page_offset+offset-start);
3156                        elt++;
3157                        if (!(len -= copy))
3158                                return elt;
3159                        offset += copy;
3160                }
3161                start = end;
3162        }
3163
3164        skb_walk_frags(skb, frag_iter) {
3165                int end;
3166
3167                WARN_ON(start > offset + len);
3168
3169                end = start + frag_iter->len;
3170                if ((copy = end - offset) > 0) {
3171                        if (copy > len)
3172                                copy = len;
3173                        elt += __skb_to_sgvec(frag_iter, sg+elt, offset - start,
3174                                              copy);
3175                        if ((len -= copy) == 0)
3176                                return elt;
3177                        offset += copy;
3178                }
3179                start = end;
3180        }
3181        BUG_ON(len);
3182        return elt;
3183}
3184
3185int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, int len)
3186{
3187        int nsg = __skb_to_sgvec(skb, sg, offset, len);
3188
3189        sg_mark_end(&sg[nsg - 1]);
3190
3191        return nsg;
3192}
3193EXPORT_SYMBOL_GPL(skb_to_sgvec);
3194
3195/**
3196 *      skb_cow_data - Check that a socket buffer's data buffers are writable
3197 *      @skb: The socket buffer to check.
3198 *      @tailbits: Amount of trailing space to be added
3199 *      @trailer: Returned pointer to the skb where the @tailbits space begins
3200 *
3201 *      Make sure that the data buffers attached to a socket buffer are
3202 *      writable. If they are not, private copies are made of the data buffers
3203 *      and the socket buffer is set to use these instead.
3204 *
3205 *      If @tailbits is given, make sure that there is space to write @tailbits
3206 *      bytes of data beyond current end of socket buffer.  @trailer will be
3207 *      set to point to the skb in which this space begins.
3208 *
3209 *      The number of scatterlist elements required to completely map the
3210 *      COW'd and extended socket buffer will be returned.
3211 */
3212int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer)
3213{
3214        int copyflag;
3215        int elt;
3216        struct sk_buff *skb1, **skb_p;
3217
3218        /* If skb is cloned or its head is paged, reallocate
3219         * head pulling out all the pages (pages are considered not writable
3220         * at the moment even if they are anonymous).
3221         */
3222        if ((skb_cloned(skb) || skb_shinfo(skb)->nr_frags) &&
3223            __pskb_pull_tail(skb, skb_pagelen(skb)-skb_headlen(skb)) == NULL)
3224                return -ENOMEM;
3225
3226        /* Easy case. Most of packets will go this way. */
3227        if (!skb_has_frag_list(skb)) {
3228                /* A little of trouble, not enough of space for trailer.
3229                 * This should not happen, when stack is tuned to generate
3230                 * good frames. OK, on miss we reallocate and reserve even more
3231                 * space, 128 bytes is fair. */
3232
3233                if (skb_tailroom(skb) < tailbits &&
3234                    pskb_expand_head(skb, 0, tailbits-skb_tailroom(skb)+128, GFP_ATOMIC))
3235                        return -ENOMEM;
3236
3237                /* Voila! */
3238                *trailer = skb;
3239                return 1;
3240        }
3241
3242        /* Misery. We are in troubles, going to mincer fragments... */
3243
3244        elt = 1;
3245        skb_p = &skb_shinfo(skb)->frag_list;
3246        copyflag = 0;
3247
3248        while ((skb1 = *skb_p) != NULL) {
3249                int ntail = 0;
3250
3251                /* The fragment is partially pulled by someone,
3252                 * this can happen on input. Copy it and everything
3253                 * after it. */
3254
3255                if (skb_shared(skb1))
3256                        copyflag = 1;
3257
3258                /* If the skb is the last, worry about trailer. */
3259
3260                if (skb1->next == NULL && tailbits) {
3261                        if (skb_shinfo(skb1)->nr_frags ||
3262                            skb_has_frag_list(skb1) ||
3263                            skb_tailroom(skb1) < tailbits)
3264                                ntail = tailbits + 128;
3265                }
3266
3267                if (copyflag ||
3268                    skb_cloned(skb1) ||
3269                    ntail ||
3270                    skb_shinfo(skb1)->nr_frags ||
3271                    skb_has_frag_list(skb1)) {
3272                        struct sk_buff *skb2;
3273
3274                        /* Fuck, we are miserable poor guys... */
3275                        if (ntail == 0)
3276                                skb2 = skb_copy(skb1, GFP_ATOMIC);
3277                        else
3278                                skb2 = skb_copy_expand(skb1,
3279                                                       skb_headroom(skb1),
3280                                                       ntail,
3281                                                       GFP_ATOMIC);
3282                        if (unlikely(skb2 == NULL))
3283                                return -ENOMEM;
3284
3285                        if (skb1->sk)
3286                                skb_set_owner_w(skb2, skb1->sk);
3287
3288                        /* Looking around. Are we still alive?
3289                         * OK, link new skb, drop old one */
3290
3291                        skb2->next = skb1->next;
3292                        *skb_p = skb2;
3293                        kfree_skb(skb1);
3294                        skb1 = skb2;
3295                }
3296                elt++;
3297                *trailer = skb1;
3298                skb_p = &skb1->next;
3299        }
3300
3301        return elt;
3302}
3303EXPORT_SYMBOL_GPL(skb_cow_data);
3304
3305static void sock_rmem_free(struct sk_buff *skb)
3306{
3307        struct sock *sk = skb->sk;
3308
3309        atomic_sub(skb->truesize, &sk->sk_rmem_alloc);
3310}
3311
3312/*
3313 * Note: We dont mem charge error packets (no sk_forward_alloc changes)
3314 */
3315int sock_queue_err_skb(struct sock *sk, struct sk_buff *skb)
3316{
3317        int len = skb->len;
3318
3319        if (atomic_read(&sk->sk_rmem_alloc) + skb->truesize >=
3320            (unsigned int)sk->sk_rcvbuf)
3321                return -ENOMEM;
3322
3323        skb_orphan(skb);
3324        skb->sk = sk;
3325        skb->destructor = sock_rmem_free;
3326        atomic_add(skb->truesize, &sk->sk_rmem_alloc);
3327
3328        /* before exiting rcu section, make sure dst is refcounted */
3329        skb_dst_force(skb);
3330
3331        skb_queue_tail(&sk->sk_error_queue, skb);
3332        if (!sock_flag(sk, SOCK_DEAD))
3333                sk->sk_data_ready(sk, len);
3334        return 0;
3335}
3336EXPORT_SYMBOL(sock_queue_err_skb);
3337
3338void skb_tstamp_tx(struct sk_buff *orig_skb,
3339                struct skb_shared_hwtstamps *hwtstamps)
3340{
3341        struct sock *sk = orig_skb->sk;
3342        struct sock_exterr_skb *serr;
3343        struct sk_buff *skb;
3344        int err;
3345
3346        if (!sk)
3347                return;
3348
3349        skb = skb_clone(orig_skb, GFP_ATOMIC);
3350        if (!skb)
3351                return;
3352
3353        if (hwtstamps) {
3354                *skb_hwtstamps(skb) =
3355                        *hwtstamps;
3356        } else {
3357                /*
3358                 * no hardware time stamps available,
3359                 * so keep the shared tx_flags and only
3360                 * store software time stamp
3361                 */
3362                skb->tstamp = ktime_get_real();
3363        }
3364
3365        serr = SKB_EXT_ERR(skb);
3366        memset(serr, 0, sizeof(*serr));
3367        serr->ee.ee_errno = ENOMSG;
3368        serr->ee.ee_origin = SO_EE_ORIGIN_TIMESTAMPING;
3369
3370        err = sock_queue_err_skb(sk, skb);
3371
3372        if (err)
3373                kfree_skb(skb);
3374}
3375EXPORT_SYMBOL_GPL(skb_tstamp_tx);
3376
3377void skb_complete_wifi_ack(struct sk_buff *skb, bool acked)
3378{
3379        struct sock *sk = skb->sk;
3380        struct sock_exterr_skb *serr;
3381        int err;
3382
3383        skb->wifi_acked_valid = 1;
3384        skb->wifi_acked = acked;
3385
3386        serr = SKB_EXT_ERR(skb);
3387        memset(serr, 0, sizeof(*serr));
3388        serr->ee.ee_errno = ENOMSG;
3389        serr->ee.ee_origin = SO_EE_ORIGIN_TXSTATUS;
3390
3391        err = sock_queue_err_skb(sk, skb);
3392        if (err)
3393                kfree_skb(skb);
3394}
3395EXPORT_SYMBOL_GPL(skb_complete_wifi_ack);
3396
3397
3398/**
3399 * skb_partial_csum_set - set up and verify partial csum values for packet
3400 * @skb: the skb to set
3401 * @start: the number of bytes after skb->data to start checksumming.
3402 * @off: the offset from start to place the checksum.
3403 *
3404 * For untrusted partially-checksummed packets, we need to make sure the values
3405 * for skb->csum_start and skb->csum_offset are valid so we don't oops.
3406 *
3407 * This function checks and sets those values and skb->ip_summed: if this
3408 * returns false you should drop the packet.
3409 */
3410bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off)
3411{
3412        if (unlikely(start > skb_headlen(skb)) ||
3413            unlikely((int)start + off > skb_headlen(skb) - 2)) {
3414                net_warn_ratelimited("bad partial csum: csum=%u/%u len=%u\n",
3415                                     start, off, skb_headlen(skb));
3416                return false;
3417        }
3418        skb->ip_summed = CHECKSUM_PARTIAL;
3419        skb->csum_start = skb_headroom(skb) + start;
3420        skb->csum_offset = off;
3421        return true;
3422}
3423EXPORT_SYMBOL_GPL(skb_partial_csum_set);
3424
3425void __skb_warn_lro_forwarding(const struct sk_buff *skb)
3426{
3427        net_warn_ratelimited("%s: received packets cannot be forwarded while LRO is enabled\n",
3428                             skb->dev->name);
3429}
3430EXPORT_SYMBOL(__skb_warn_lro_forwarding);
3431
3432void kfree_skb_partial(struct sk_buff *skb, bool head_stolen)
3433{
3434        if (head_stolen)
3435                kmem_cache_free(skbuff_head_cache, skb);
3436        else
3437                __kfree_skb(skb);
3438}
3439EXPORT_SYMBOL(kfree_skb_partial);
3440
3441/**
3442 * skb_try_coalesce - try to merge skb to prior one
3443 * @to: prior buffer
3444 * @from: buffer to add
3445 * @fragstolen: pointer to boolean
3446 * @delta_truesize: how much more was allocated than was requested
3447 */
3448bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from,
3449                      bool *fragstolen, int *delta_truesize)
3450{
3451        int i, delta, len = from->len;
3452
3453        *fragstolen = false;
3454
3455        if (skb_cloned(to))
3456                return false;
3457
3458        if (len <= skb_tailroom(to)) {
3459                BUG_ON(skb_copy_bits(from, 0, skb_put(to, len), len));
3460                *delta_truesize = 0;
3461                return true;
3462        }
3463
3464        if (skb_has_frag_list(to) || skb_has_frag_list(from))
3465                return false;
3466
3467        if (skb_headlen(from) != 0) {
3468                struct page *page;
3469                unsigned int offset;
3470
3471                if (skb_shinfo(to)->nr_frags +
3472                    skb_shinfo(from)->nr_frags >= MAX_SKB_FRAGS)
3473                        return false;
3474
3475                if (skb_head_is_locked(from))
3476                        return false;
3477
3478                delta = from->truesize - SKB_DATA_ALIGN(sizeof(struct sk_buff));
3479
3480                page = virt_to_head_page(from->head);
3481                offset = from->data - (unsigned char *)page_address(page);
3482
3483                skb_fill_page_desc(to, skb_shinfo(to)->nr_frags,
3484                                   page, offset, skb_headlen(from));
3485                *fragstolen = true;
3486        } else {
3487                if (skb_shinfo(to)->nr_frags +
3488                    skb_shinfo(from)->nr_frags > MAX_SKB_FRAGS)
3489                        return false;
3490
3491                delta = from->truesize -
3492                        SKB_TRUESIZE(skb_end_pointer(from) - from->head);
3493        }
3494
3495        WARN_ON_ONCE(delta < len);
3496
3497        memcpy(skb_shinfo(to)->frags + skb_shinfo(to)->nr_frags,
3498               skb_shinfo(from)->frags,
3499               skb_shinfo(from)->nr_frags * sizeof(skb_frag_t));
3500        skb_shinfo(to)->nr_frags += skb_shinfo(from)->nr_frags;
3501
3502        if (!skb_cloned(from))
3503                skb_shinfo(from)->nr_frags = 0;
3504
3505        /* if the skb is not cloned this does nothing
3506         * since we set nr_frags to 0.
3507         */
3508        for (i = 0; i < skb_shinfo(from)->nr_frags; i++)
3509                skb_frag_ref(from, i);
3510
3511        to->truesize += delta;
3512        to->len += len;
3513        to->data_len += len;
3514
3515        *delta_truesize = delta;
3516        return true;
3517}
3518EXPORT_SYMBOL(skb_try_coalesce);
3519
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.