at v3.6 76 kB view raw
1/* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <gw4pts@gw4pts.ampr.org> 6 * Florian La Roche, <rzsfl@rz.uni-sb.de> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14#ifndef _LINUX_SKBUFF_H 15#define _LINUX_SKBUFF_H 16 17#include <linux/kernel.h> 18#include <linux/kmemcheck.h> 19#include <linux/compiler.h> 20#include <linux/time.h> 21#include <linux/bug.h> 22#include <linux/cache.h> 23 24#include <linux/atomic.h> 25#include <asm/types.h> 26#include <linux/spinlock.h> 27#include <linux/net.h> 28#include <linux/textsearch.h> 29#include <net/checksum.h> 30#include <linux/rcupdate.h> 31#include <linux/dmaengine.h> 32#include <linux/hrtimer.h> 33#include <linux/dma-mapping.h> 34#include <linux/netdev_features.h> 35 36/* Don't change this without changing skb_csum_unnecessary! */ 37#define CHECKSUM_NONE 0 38#define CHECKSUM_UNNECESSARY 1 39#define CHECKSUM_COMPLETE 2 40#define CHECKSUM_PARTIAL 3 41 42#define SKB_DATA_ALIGN(X) (((X) + (SMP_CACHE_BYTES - 1)) & \ 43 ~(SMP_CACHE_BYTES - 1)) 44#define SKB_WITH_OVERHEAD(X) \ 45 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 46#define SKB_MAX_ORDER(X, ORDER) \ 47 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 48#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 49#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 50 51/* return minimum truesize of one skb containing X bytes of data */ 52#define SKB_TRUESIZE(X) ((X) + \ 53 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 54 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 55 56/* A. Checksumming of received packets by device. 57 * 58 * NONE: device failed to checksum this packet. 59 * skb->csum is undefined. 60 * 61 * UNNECESSARY: device parsed packet and wouldbe verified checksum. 62 * skb->csum is undefined. 63 * It is bad option, but, unfortunately, many of vendors do this. 64 * Apparently with secret goal to sell you new device, when you 65 * will add new protocol to your host. F.e. IPv6. 8) 66 * 67 * COMPLETE: the most generic way. Device supplied checksum of _all_ 68 * the packet as seen by netif_rx in skb->csum. 69 * NOTE: Even if device supports only some protocols, but 70 * is able to produce some skb->csum, it MUST use COMPLETE, 71 * not UNNECESSARY. 72 * 73 * PARTIAL: identical to the case for output below. This may occur 74 * on a packet received directly from another Linux OS, e.g., 75 * a virtualised Linux kernel on the same host. The packet can 76 * be treated in the same way as UNNECESSARY except that on 77 * output (i.e., forwarding) the checksum must be filled in 78 * by the OS or the hardware. 79 * 80 * B. Checksumming on output. 81 * 82 * NONE: skb is checksummed by protocol or csum is not required. 83 * 84 * PARTIAL: device is required to csum packet as seen by hard_start_xmit 85 * from skb->csum_start to the end and to record the checksum 86 * at skb->csum_start + skb->csum_offset. 87 * 88 * Device must show its capabilities in dev->features, set 89 * at device setup time. 90 * NETIF_F_HW_CSUM - it is clever device, it is able to checksum 91 * everything. 92 * NETIF_F_IP_CSUM - device is dumb. It is able to csum only 93 * TCP/UDP over IPv4. Sigh. Vendors like this 94 * way by an unknown reason. Though, see comment above 95 * about CHECKSUM_UNNECESSARY. 8) 96 * NETIF_F_IPV6_CSUM about as dumb as the last one but does IPv6 instead. 97 * 98 * UNNECESSARY: device will do per protocol specific csum. Protocol drivers 99 * that do not want net to perform the checksum calculation should use 100 * this flag in their outgoing skbs. 101 * NETIF_F_FCOE_CRC this indicates the device can do FCoE FC CRC 102 * offload. Correspondingly, the FCoE protocol driver 103 * stack should use CHECKSUM_UNNECESSARY. 104 * 105 * Any questions? No questions, good. --ANK 106 */ 107 108struct net_device; 109struct scatterlist; 110struct pipe_inode_info; 111 112#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 113struct nf_conntrack { 114 atomic_t use; 115}; 116#endif 117 118#ifdef CONFIG_BRIDGE_NETFILTER 119struct nf_bridge_info { 120 atomic_t use; 121 unsigned int mask; 122 struct net_device *physindev; 123 struct net_device *physoutdev; 124 unsigned long data[32 / sizeof(unsigned long)]; 125}; 126#endif 127 128struct sk_buff_head { 129 /* These two members must be first. */ 130 struct sk_buff *next; 131 struct sk_buff *prev; 132 133 __u32 qlen; 134 spinlock_t lock; 135}; 136 137struct sk_buff; 138 139/* To allow 64K frame to be packed as single skb without frag_list we 140 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 141 * buffers which do not start on a page boundary. 142 * 143 * Since GRO uses frags we allocate at least 16 regardless of page 144 * size. 145 */ 146#if (65536/PAGE_SIZE + 1) < 16 147#define MAX_SKB_FRAGS 16UL 148#else 149#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 150#endif 151 152typedef struct skb_frag_struct skb_frag_t; 153 154struct skb_frag_struct { 155 struct { 156 struct page *p; 157 } page; 158#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 159 __u32 page_offset; 160 __u32 size; 161#else 162 __u16 page_offset; 163 __u16 size; 164#endif 165}; 166 167static inline unsigned int skb_frag_size(const skb_frag_t *frag) 168{ 169 return frag->size; 170} 171 172static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 173{ 174 frag->size = size; 175} 176 177static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 178{ 179 frag->size += delta; 180} 181 182static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 183{ 184 frag->size -= delta; 185} 186 187#define HAVE_HW_TIME_STAMP 188 189/** 190 * struct skb_shared_hwtstamps - hardware time stamps 191 * @hwtstamp: hardware time stamp transformed into duration 192 * since arbitrary point in time 193 * @syststamp: hwtstamp transformed to system time base 194 * 195 * Software time stamps generated by ktime_get_real() are stored in 196 * skb->tstamp. The relation between the different kinds of time 197 * stamps is as follows: 198 * 199 * syststamp and tstamp can be compared against each other in 200 * arbitrary combinations. The accuracy of a 201 * syststamp/tstamp/"syststamp from other device" comparison is 202 * limited by the accuracy of the transformation into system time 203 * base. This depends on the device driver and its underlying 204 * hardware. 205 * 206 * hwtstamps can only be compared against other hwtstamps from 207 * the same device. 208 * 209 * This structure is attached to packets as part of the 210 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 211 */ 212struct skb_shared_hwtstamps { 213 ktime_t hwtstamp; 214 ktime_t syststamp; 215}; 216 217/* Definitions for tx_flags in struct skb_shared_info */ 218enum { 219 /* generate hardware time stamp */ 220 SKBTX_HW_TSTAMP = 1 << 0, 221 222 /* generate software time stamp */ 223 SKBTX_SW_TSTAMP = 1 << 1, 224 225 /* device driver is going to provide hardware time stamp */ 226 SKBTX_IN_PROGRESS = 1 << 2, 227 228 /* device driver supports TX zero-copy buffers */ 229 SKBTX_DEV_ZEROCOPY = 1 << 3, 230 231 /* generate wifi status information (where possible) */ 232 SKBTX_WIFI_STATUS = 1 << 4, 233}; 234 235/* 236 * The callback notifies userspace to release buffers when skb DMA is done in 237 * lower device, the skb last reference should be 0 when calling this. 238 * The ctx field is used to track device context. 239 * The desc field is used to track userspace buffer index. 240 */ 241struct ubuf_info { 242 void (*callback)(struct ubuf_info *); 243 void *ctx; 244 unsigned long desc; 245}; 246 247/* This data is invariant across clones and lives at 248 * the end of the header data, ie. at skb->end. 249 */ 250struct skb_shared_info { 251 unsigned char nr_frags; 252 __u8 tx_flags; 253 unsigned short gso_size; 254 /* Warning: this field is not always filled in (UFO)! */ 255 unsigned short gso_segs; 256 unsigned short gso_type; 257 struct sk_buff *frag_list; 258 struct skb_shared_hwtstamps hwtstamps; 259 __be32 ip6_frag_id; 260 261 /* 262 * Warning : all fields before dataref are cleared in __alloc_skb() 263 */ 264 atomic_t dataref; 265 266 /* Intermediate layers must ensure that destructor_arg 267 * remains valid until skb destructor */ 268 void * destructor_arg; 269 270 /* must be last field, see pskb_expand_head() */ 271 skb_frag_t frags[MAX_SKB_FRAGS]; 272}; 273 274/* We divide dataref into two halves. The higher 16 bits hold references 275 * to the payload part of skb->data. The lower 16 bits hold references to 276 * the entire skb->data. A clone of a headerless skb holds the length of 277 * the header in skb->hdr_len. 278 * 279 * All users must obey the rule that the skb->data reference count must be 280 * greater than or equal to the payload reference count. 281 * 282 * Holding a reference to the payload part means that the user does not 283 * care about modifications to the header part of skb->data. 284 */ 285#define SKB_DATAREF_SHIFT 16 286#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 287 288 289enum { 290 SKB_FCLONE_UNAVAILABLE, 291 SKB_FCLONE_ORIG, 292 SKB_FCLONE_CLONE, 293}; 294 295enum { 296 SKB_GSO_TCPV4 = 1 << 0, 297 SKB_GSO_UDP = 1 << 1, 298 299 /* This indicates the skb is from an untrusted source. */ 300 SKB_GSO_DODGY = 1 << 2, 301 302 /* This indicates the tcp segment has CWR set. */ 303 SKB_GSO_TCP_ECN = 1 << 3, 304 305 SKB_GSO_TCPV6 = 1 << 4, 306 307 SKB_GSO_FCOE = 1 << 5, 308}; 309 310#if BITS_PER_LONG > 32 311#define NET_SKBUFF_DATA_USES_OFFSET 1 312#endif 313 314#ifdef NET_SKBUFF_DATA_USES_OFFSET 315typedef unsigned int sk_buff_data_t; 316#else 317typedef unsigned char *sk_buff_data_t; 318#endif 319 320#if defined(CONFIG_NF_DEFRAG_IPV4) || defined(CONFIG_NF_DEFRAG_IPV4_MODULE) || \ 321 defined(CONFIG_NF_DEFRAG_IPV6) || defined(CONFIG_NF_DEFRAG_IPV6_MODULE) 322#define NET_SKBUFF_NF_DEFRAG_NEEDED 1 323#endif 324 325/** 326 * struct sk_buff - socket buffer 327 * @next: Next buffer in list 328 * @prev: Previous buffer in list 329 * @tstamp: Time we arrived 330 * @sk: Socket we are owned by 331 * @dev: Device we arrived on/are leaving by 332 * @cb: Control buffer. Free for use by every layer. Put private vars here 333 * @_skb_refdst: destination entry (with norefcount bit) 334 * @sp: the security path, used for xfrm 335 * @len: Length of actual data 336 * @data_len: Data length 337 * @mac_len: Length of link layer header 338 * @hdr_len: writable header length of cloned skb 339 * @csum: Checksum (must include start/offset pair) 340 * @csum_start: Offset from skb->head where checksumming should start 341 * @csum_offset: Offset from csum_start where checksum should be stored 342 * @priority: Packet queueing priority 343 * @local_df: allow local fragmentation 344 * @cloned: Head may be cloned (check refcnt to be sure) 345 * @ip_summed: Driver fed us an IP checksum 346 * @nohdr: Payload reference only, must not modify header 347 * @nfctinfo: Relationship of this skb to the connection 348 * @pkt_type: Packet class 349 * @fclone: skbuff clone status 350 * @ipvs_property: skbuff is owned by ipvs 351 * @peeked: this packet has been seen already, so stats have been 352 * done for it, don't do them again 353 * @nf_trace: netfilter packet trace flag 354 * @protocol: Packet protocol from driver 355 * @destructor: Destruct function 356 * @nfct: Associated connection, if any 357 * @nfct_reasm: netfilter conntrack re-assembly pointer 358 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 359 * @skb_iif: ifindex of device we arrived on 360 * @tc_index: Traffic control index 361 * @tc_verd: traffic control verdict 362 * @rxhash: the packet hash computed on receive 363 * @queue_mapping: Queue mapping for multiqueue devices 364 * @ndisc_nodetype: router type (from link layer) 365 * @ooo_okay: allow the mapping of a socket to a queue to be changed 366 * @l4_rxhash: indicate rxhash is a canonical 4-tuple hash over transport 367 * ports. 368 * @wifi_acked_valid: wifi_acked was set 369 * @wifi_acked: whether frame was acked on wifi or not 370 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 371 * @dma_cookie: a cookie to one of several possible DMA operations 372 * done by skb DMA functions 373 * @secmark: security marking 374 * @mark: Generic packet mark 375 * @dropcount: total number of sk_receive_queue overflows 376 * @vlan_tci: vlan tag control information 377 * @transport_header: Transport layer header 378 * @network_header: Network layer header 379 * @mac_header: Link layer header 380 * @tail: Tail pointer 381 * @end: End pointer 382 * @head: Head of buffer 383 * @data: Data head pointer 384 * @truesize: Buffer size 385 * @users: User count - see {datagram,tcp}.c 386 */ 387 388struct sk_buff { 389 /* These two members must be first. */ 390 struct sk_buff *next; 391 struct sk_buff *prev; 392 393 ktime_t tstamp; 394 395 struct sock *sk; 396 struct net_device *dev; 397 398 /* 399 * This is the control buffer. It is free to use for every 400 * layer. Please put your private variables there. If you 401 * want to keep them across layers you have to do a skb_clone() 402 * first. This is owned by whoever has the skb queued ATM. 403 */ 404 char cb[48] __aligned(8); 405 406 unsigned long _skb_refdst; 407#ifdef CONFIG_XFRM 408 struct sec_path *sp; 409#endif 410 unsigned int len, 411 data_len; 412 __u16 mac_len, 413 hdr_len; 414 union { 415 __wsum csum; 416 struct { 417 __u16 csum_start; 418 __u16 csum_offset; 419 }; 420 }; 421 __u32 priority; 422 kmemcheck_bitfield_begin(flags1); 423 __u8 local_df:1, 424 cloned:1, 425 ip_summed:2, 426 nohdr:1, 427 nfctinfo:3; 428 __u8 pkt_type:3, 429 fclone:2, 430 ipvs_property:1, 431 peeked:1, 432 nf_trace:1; 433 kmemcheck_bitfield_end(flags1); 434 __be16 protocol; 435 436 void (*destructor)(struct sk_buff *skb); 437#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 438 struct nf_conntrack *nfct; 439#endif 440#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED 441 struct sk_buff *nfct_reasm; 442#endif 443#ifdef CONFIG_BRIDGE_NETFILTER 444 struct nf_bridge_info *nf_bridge; 445#endif 446 447 int skb_iif; 448 449 __u32 rxhash; 450 451 __u16 vlan_tci; 452 453#ifdef CONFIG_NET_SCHED 454 __u16 tc_index; /* traffic control index */ 455#ifdef CONFIG_NET_CLS_ACT 456 __u16 tc_verd; /* traffic control verdict */ 457#endif 458#endif 459 460 __u16 queue_mapping; 461 kmemcheck_bitfield_begin(flags2); 462#ifdef CONFIG_IPV6_NDISC_NODETYPE 463 __u8 ndisc_nodetype:2; 464#endif 465 __u8 pfmemalloc:1; 466 __u8 ooo_okay:1; 467 __u8 l4_rxhash:1; 468 __u8 wifi_acked_valid:1; 469 __u8 wifi_acked:1; 470 __u8 no_fcs:1; 471 __u8 head_frag:1; 472 /* 8/10 bit hole (depending on ndisc_nodetype presence) */ 473 kmemcheck_bitfield_end(flags2); 474 475#ifdef CONFIG_NET_DMA 476 dma_cookie_t dma_cookie; 477#endif 478#ifdef CONFIG_NETWORK_SECMARK 479 __u32 secmark; 480#endif 481 union { 482 __u32 mark; 483 __u32 dropcount; 484 __u32 avail_size; 485 }; 486 487 sk_buff_data_t transport_header; 488 sk_buff_data_t network_header; 489 sk_buff_data_t mac_header; 490 /* These elements must be at the end, see alloc_skb() for details. */ 491 sk_buff_data_t tail; 492 sk_buff_data_t end; 493 unsigned char *head, 494 *data; 495 unsigned int truesize; 496 atomic_t users; 497}; 498 499#ifdef __KERNEL__ 500/* 501 * Handling routines are only of interest to the kernel 502 */ 503#include <linux/slab.h> 504 505 506#define SKB_ALLOC_FCLONE 0x01 507#define SKB_ALLOC_RX 0x02 508 509/* Returns true if the skb was allocated from PFMEMALLOC reserves */ 510static inline bool skb_pfmemalloc(const struct sk_buff *skb) 511{ 512 return unlikely(skb->pfmemalloc); 513} 514 515/* 516 * skb might have a dst pointer attached, refcounted or not. 517 * _skb_refdst low order bit is set if refcount was _not_ taken 518 */ 519#define SKB_DST_NOREF 1UL 520#define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 521 522/** 523 * skb_dst - returns skb dst_entry 524 * @skb: buffer 525 * 526 * Returns skb dst_entry, regardless of reference taken or not. 527 */ 528static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 529{ 530 /* If refdst was not refcounted, check we still are in a 531 * rcu_read_lock section 532 */ 533 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 534 !rcu_read_lock_held() && 535 !rcu_read_lock_bh_held()); 536 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 537} 538 539/** 540 * skb_dst_set - sets skb dst 541 * @skb: buffer 542 * @dst: dst entry 543 * 544 * Sets skb dst, assuming a reference was taken on dst and should 545 * be released by skb_dst_drop() 546 */ 547static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 548{ 549 skb->_skb_refdst = (unsigned long)dst; 550} 551 552extern void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst); 553 554/** 555 * skb_dst_is_noref - Test if skb dst isn't refcounted 556 * @skb: buffer 557 */ 558static inline bool skb_dst_is_noref(const struct sk_buff *skb) 559{ 560 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 561} 562 563static inline struct rtable *skb_rtable(const struct sk_buff *skb) 564{ 565 return (struct rtable *)skb_dst(skb); 566} 567 568extern void kfree_skb(struct sk_buff *skb); 569extern void consume_skb(struct sk_buff *skb); 570extern void __kfree_skb(struct sk_buff *skb); 571extern struct kmem_cache *skbuff_head_cache; 572 573extern void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 574extern bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 575 bool *fragstolen, int *delta_truesize); 576 577extern struct sk_buff *__alloc_skb(unsigned int size, 578 gfp_t priority, int flags, int node); 579extern struct sk_buff *build_skb(void *data, unsigned int frag_size); 580static inline struct sk_buff *alloc_skb(unsigned int size, 581 gfp_t priority) 582{ 583 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 584} 585 586static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 587 gfp_t priority) 588{ 589 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 590} 591 592extern void skb_recycle(struct sk_buff *skb); 593extern bool skb_recycle_check(struct sk_buff *skb, int skb_size); 594 595extern struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 596extern int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 597extern struct sk_buff *skb_clone(struct sk_buff *skb, 598 gfp_t priority); 599extern struct sk_buff *skb_copy(const struct sk_buff *skb, 600 gfp_t priority); 601extern struct sk_buff *__pskb_copy(struct sk_buff *skb, 602 int headroom, gfp_t gfp_mask); 603 604extern int pskb_expand_head(struct sk_buff *skb, 605 int nhead, int ntail, 606 gfp_t gfp_mask); 607extern struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 608 unsigned int headroom); 609extern struct sk_buff *skb_copy_expand(const struct sk_buff *skb, 610 int newheadroom, int newtailroom, 611 gfp_t priority); 612extern int skb_to_sgvec(struct sk_buff *skb, 613 struct scatterlist *sg, int offset, 614 int len); 615extern int skb_cow_data(struct sk_buff *skb, int tailbits, 616 struct sk_buff **trailer); 617extern int skb_pad(struct sk_buff *skb, int pad); 618#define dev_kfree_skb(a) consume_skb(a) 619 620extern int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 621 int getfrag(void *from, char *to, int offset, 622 int len,int odd, struct sk_buff *skb), 623 void *from, int length); 624 625struct skb_seq_state { 626 __u32 lower_offset; 627 __u32 upper_offset; 628 __u32 frag_idx; 629 __u32 stepped_offset; 630 struct sk_buff *root_skb; 631 struct sk_buff *cur_skb; 632 __u8 *frag_data; 633}; 634 635extern void skb_prepare_seq_read(struct sk_buff *skb, 636 unsigned int from, unsigned int to, 637 struct skb_seq_state *st); 638extern unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 639 struct skb_seq_state *st); 640extern void skb_abort_seq_read(struct skb_seq_state *st); 641 642extern unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 643 unsigned int to, struct ts_config *config, 644 struct ts_state *state); 645 646extern void __skb_get_rxhash(struct sk_buff *skb); 647static inline __u32 skb_get_rxhash(struct sk_buff *skb) 648{ 649 if (!skb->rxhash) 650 __skb_get_rxhash(skb); 651 652 return skb->rxhash; 653} 654 655#ifdef NET_SKBUFF_DATA_USES_OFFSET 656static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 657{ 658 return skb->head + skb->end; 659} 660 661static inline unsigned int skb_end_offset(const struct sk_buff *skb) 662{ 663 return skb->end; 664} 665#else 666static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 667{ 668 return skb->end; 669} 670 671static inline unsigned int skb_end_offset(const struct sk_buff *skb) 672{ 673 return skb->end - skb->head; 674} 675#endif 676 677/* Internal */ 678#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 679 680static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 681{ 682 return &skb_shinfo(skb)->hwtstamps; 683} 684 685/** 686 * skb_queue_empty - check if a queue is empty 687 * @list: queue head 688 * 689 * Returns true if the queue is empty, false otherwise. 690 */ 691static inline int skb_queue_empty(const struct sk_buff_head *list) 692{ 693 return list->next == (struct sk_buff *)list; 694} 695 696/** 697 * skb_queue_is_last - check if skb is the last entry in the queue 698 * @list: queue head 699 * @skb: buffer 700 * 701 * Returns true if @skb is the last buffer on the list. 702 */ 703static inline bool skb_queue_is_last(const struct sk_buff_head *list, 704 const struct sk_buff *skb) 705{ 706 return skb->next == (struct sk_buff *)list; 707} 708 709/** 710 * skb_queue_is_first - check if skb is the first entry in the queue 711 * @list: queue head 712 * @skb: buffer 713 * 714 * Returns true if @skb is the first buffer on the list. 715 */ 716static inline bool skb_queue_is_first(const struct sk_buff_head *list, 717 const struct sk_buff *skb) 718{ 719 return skb->prev == (struct sk_buff *)list; 720} 721 722/** 723 * skb_queue_next - return the next packet in the queue 724 * @list: queue head 725 * @skb: current buffer 726 * 727 * Return the next packet in @list after @skb. It is only valid to 728 * call this if skb_queue_is_last() evaluates to false. 729 */ 730static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 731 const struct sk_buff *skb) 732{ 733 /* This BUG_ON may seem severe, but if we just return then we 734 * are going to dereference garbage. 735 */ 736 BUG_ON(skb_queue_is_last(list, skb)); 737 return skb->next; 738} 739 740/** 741 * skb_queue_prev - return the prev packet in the queue 742 * @list: queue head 743 * @skb: current buffer 744 * 745 * Return the prev packet in @list before @skb. It is only valid to 746 * call this if skb_queue_is_first() evaluates to false. 747 */ 748static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 749 const struct sk_buff *skb) 750{ 751 /* This BUG_ON may seem severe, but if we just return then we 752 * are going to dereference garbage. 753 */ 754 BUG_ON(skb_queue_is_first(list, skb)); 755 return skb->prev; 756} 757 758/** 759 * skb_get - reference buffer 760 * @skb: buffer to reference 761 * 762 * Makes another reference to a socket buffer and returns a pointer 763 * to the buffer. 764 */ 765static inline struct sk_buff *skb_get(struct sk_buff *skb) 766{ 767 atomic_inc(&skb->users); 768 return skb; 769} 770 771/* 772 * If users == 1, we are the only owner and are can avoid redundant 773 * atomic change. 774 */ 775 776/** 777 * skb_cloned - is the buffer a clone 778 * @skb: buffer to check 779 * 780 * Returns true if the buffer was generated with skb_clone() and is 781 * one of multiple shared copies of the buffer. Cloned buffers are 782 * shared data so must not be written to under normal circumstances. 783 */ 784static inline int skb_cloned(const struct sk_buff *skb) 785{ 786 return skb->cloned && 787 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 788} 789 790/** 791 * skb_header_cloned - is the header a clone 792 * @skb: buffer to check 793 * 794 * Returns true if modifying the header part of the buffer requires 795 * the data to be copied. 796 */ 797static inline int skb_header_cloned(const struct sk_buff *skb) 798{ 799 int dataref; 800 801 if (!skb->cloned) 802 return 0; 803 804 dataref = atomic_read(&skb_shinfo(skb)->dataref); 805 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 806 return dataref != 1; 807} 808 809/** 810 * skb_header_release - release reference to header 811 * @skb: buffer to operate on 812 * 813 * Drop a reference to the header part of the buffer. This is done 814 * by acquiring a payload reference. You must not read from the header 815 * part of skb->data after this. 816 */ 817static inline void skb_header_release(struct sk_buff *skb) 818{ 819 BUG_ON(skb->nohdr); 820 skb->nohdr = 1; 821 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 822} 823 824/** 825 * skb_shared - is the buffer shared 826 * @skb: buffer to check 827 * 828 * Returns true if more than one person has a reference to this 829 * buffer. 830 */ 831static inline int skb_shared(const struct sk_buff *skb) 832{ 833 return atomic_read(&skb->users) != 1; 834} 835 836/** 837 * skb_share_check - check if buffer is shared and if so clone it 838 * @skb: buffer to check 839 * @pri: priority for memory allocation 840 * 841 * If the buffer is shared the buffer is cloned and the old copy 842 * drops a reference. A new clone with a single reference is returned. 843 * If the buffer is not shared the original buffer is returned. When 844 * being called from interrupt status or with spinlocks held pri must 845 * be GFP_ATOMIC. 846 * 847 * NULL is returned on a memory allocation failure. 848 */ 849static inline struct sk_buff *skb_share_check(struct sk_buff *skb, 850 gfp_t pri) 851{ 852 might_sleep_if(pri & __GFP_WAIT); 853 if (skb_shared(skb)) { 854 struct sk_buff *nskb = skb_clone(skb, pri); 855 kfree_skb(skb); 856 skb = nskb; 857 } 858 return skb; 859} 860 861/* 862 * Copy shared buffers into a new sk_buff. We effectively do COW on 863 * packets to handle cases where we have a local reader and forward 864 * and a couple of other messy ones. The normal one is tcpdumping 865 * a packet thats being forwarded. 866 */ 867 868/** 869 * skb_unshare - make a copy of a shared buffer 870 * @skb: buffer to check 871 * @pri: priority for memory allocation 872 * 873 * If the socket buffer is a clone then this function creates a new 874 * copy of the data, drops a reference count on the old copy and returns 875 * the new copy with the reference count at 1. If the buffer is not a clone 876 * the original buffer is returned. When called with a spinlock held or 877 * from interrupt state @pri must be %GFP_ATOMIC 878 * 879 * %NULL is returned on a memory allocation failure. 880 */ 881static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 882 gfp_t pri) 883{ 884 might_sleep_if(pri & __GFP_WAIT); 885 if (skb_cloned(skb)) { 886 struct sk_buff *nskb = skb_copy(skb, pri); 887 kfree_skb(skb); /* Free our shared copy */ 888 skb = nskb; 889 } 890 return skb; 891} 892 893/** 894 * skb_peek - peek at the head of an &sk_buff_head 895 * @list_: list to peek at 896 * 897 * Peek an &sk_buff. Unlike most other operations you _MUST_ 898 * be careful with this one. A peek leaves the buffer on the 899 * list and someone else may run off with it. You must hold 900 * the appropriate locks or have a private queue to do this. 901 * 902 * Returns %NULL for an empty list or a pointer to the head element. 903 * The reference count is not incremented and the reference is therefore 904 * volatile. Use with caution. 905 */ 906static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 907{ 908 struct sk_buff *skb = list_->next; 909 910 if (skb == (struct sk_buff *)list_) 911 skb = NULL; 912 return skb; 913} 914 915/** 916 * skb_peek_next - peek skb following the given one from a queue 917 * @skb: skb to start from 918 * @list_: list to peek at 919 * 920 * Returns %NULL when the end of the list is met or a pointer to the 921 * next element. The reference count is not incremented and the 922 * reference is therefore volatile. Use with caution. 923 */ 924static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 925 const struct sk_buff_head *list_) 926{ 927 struct sk_buff *next = skb->next; 928 929 if (next == (struct sk_buff *)list_) 930 next = NULL; 931 return next; 932} 933 934/** 935 * skb_peek_tail - peek at the tail of an &sk_buff_head 936 * @list_: list to peek at 937 * 938 * Peek an &sk_buff. Unlike most other operations you _MUST_ 939 * be careful with this one. A peek leaves the buffer on the 940 * list and someone else may run off with it. You must hold 941 * the appropriate locks or have a private queue to do this. 942 * 943 * Returns %NULL for an empty list or a pointer to the tail element. 944 * The reference count is not incremented and the reference is therefore 945 * volatile. Use with caution. 946 */ 947static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 948{ 949 struct sk_buff *skb = list_->prev; 950 951 if (skb == (struct sk_buff *)list_) 952 skb = NULL; 953 return skb; 954 955} 956 957/** 958 * skb_queue_len - get queue length 959 * @list_: list to measure 960 * 961 * Return the length of an &sk_buff queue. 962 */ 963static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 964{ 965 return list_->qlen; 966} 967 968/** 969 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 970 * @list: queue to initialize 971 * 972 * This initializes only the list and queue length aspects of 973 * an sk_buff_head object. This allows to initialize the list 974 * aspects of an sk_buff_head without reinitializing things like 975 * the spinlock. It can also be used for on-stack sk_buff_head 976 * objects where the spinlock is known to not be used. 977 */ 978static inline void __skb_queue_head_init(struct sk_buff_head *list) 979{ 980 list->prev = list->next = (struct sk_buff *)list; 981 list->qlen = 0; 982} 983 984/* 985 * This function creates a split out lock class for each invocation; 986 * this is needed for now since a whole lot of users of the skb-queue 987 * infrastructure in drivers have different locking usage (in hardirq) 988 * than the networking core (in softirq only). In the long run either the 989 * network layer or drivers should need annotation to consolidate the 990 * main types of usage into 3 classes. 991 */ 992static inline void skb_queue_head_init(struct sk_buff_head *list) 993{ 994 spin_lock_init(&list->lock); 995 __skb_queue_head_init(list); 996} 997 998static inline void skb_queue_head_init_class(struct sk_buff_head *list, 999 struct lock_class_key *class) 1000{ 1001 skb_queue_head_init(list); 1002 lockdep_set_class(&list->lock, class); 1003} 1004 1005/* 1006 * Insert an sk_buff on a list. 1007 * 1008 * The "__skb_xxxx()" functions are the non-atomic ones that 1009 * can only be called with interrupts disabled. 1010 */ 1011extern void skb_insert(struct sk_buff *old, struct sk_buff *newsk, struct sk_buff_head *list); 1012static inline void __skb_insert(struct sk_buff *newsk, 1013 struct sk_buff *prev, struct sk_buff *next, 1014 struct sk_buff_head *list) 1015{ 1016 newsk->next = next; 1017 newsk->prev = prev; 1018 next->prev = prev->next = newsk; 1019 list->qlen++; 1020} 1021 1022static inline void __skb_queue_splice(const struct sk_buff_head *list, 1023 struct sk_buff *prev, 1024 struct sk_buff *next) 1025{ 1026 struct sk_buff *first = list->next; 1027 struct sk_buff *last = list->prev; 1028 1029 first->prev = prev; 1030 prev->next = first; 1031 1032 last->next = next; 1033 next->prev = last; 1034} 1035 1036/** 1037 * skb_queue_splice - join two skb lists, this is designed for stacks 1038 * @list: the new list to add 1039 * @head: the place to add it in the first list 1040 */ 1041static inline void skb_queue_splice(const struct sk_buff_head *list, 1042 struct sk_buff_head *head) 1043{ 1044 if (!skb_queue_empty(list)) { 1045 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1046 head->qlen += list->qlen; 1047 } 1048} 1049 1050/** 1051 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1052 * @list: the new list to add 1053 * @head: the place to add it in the first list 1054 * 1055 * The list at @list is reinitialised 1056 */ 1057static inline void skb_queue_splice_init(struct sk_buff_head *list, 1058 struct sk_buff_head *head) 1059{ 1060 if (!skb_queue_empty(list)) { 1061 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1062 head->qlen += list->qlen; 1063 __skb_queue_head_init(list); 1064 } 1065} 1066 1067/** 1068 * skb_queue_splice_tail - join two skb lists, each list being a queue 1069 * @list: the new list to add 1070 * @head: the place to add it in the first list 1071 */ 1072static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1073 struct sk_buff_head *head) 1074{ 1075 if (!skb_queue_empty(list)) { 1076 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1077 head->qlen += list->qlen; 1078 } 1079} 1080 1081/** 1082 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1083 * @list: the new list to add 1084 * @head: the place to add it in the first list 1085 * 1086 * Each of the lists is a queue. 1087 * The list at @list is reinitialised 1088 */ 1089static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1090 struct sk_buff_head *head) 1091{ 1092 if (!skb_queue_empty(list)) { 1093 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1094 head->qlen += list->qlen; 1095 __skb_queue_head_init(list); 1096 } 1097} 1098 1099/** 1100 * __skb_queue_after - queue a buffer at the list head 1101 * @list: list to use 1102 * @prev: place after this buffer 1103 * @newsk: buffer to queue 1104 * 1105 * Queue a buffer int the middle of a list. This function takes no locks 1106 * and you must therefore hold required locks before calling it. 1107 * 1108 * A buffer cannot be placed on two lists at the same time. 1109 */ 1110static inline void __skb_queue_after(struct sk_buff_head *list, 1111 struct sk_buff *prev, 1112 struct sk_buff *newsk) 1113{ 1114 __skb_insert(newsk, prev, prev->next, list); 1115} 1116 1117extern void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1118 struct sk_buff_head *list); 1119 1120static inline void __skb_queue_before(struct sk_buff_head *list, 1121 struct sk_buff *next, 1122 struct sk_buff *newsk) 1123{ 1124 __skb_insert(newsk, next->prev, next, list); 1125} 1126 1127/** 1128 * __skb_queue_head - queue a buffer at the list head 1129 * @list: list to use 1130 * @newsk: buffer to queue 1131 * 1132 * Queue a buffer at the start of a list. This function takes no locks 1133 * and you must therefore hold required locks before calling it. 1134 * 1135 * A buffer cannot be placed on two lists at the same time. 1136 */ 1137extern void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1138static inline void __skb_queue_head(struct sk_buff_head *list, 1139 struct sk_buff *newsk) 1140{ 1141 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1142} 1143 1144/** 1145 * __skb_queue_tail - queue a buffer at the list tail 1146 * @list: list to use 1147 * @newsk: buffer to queue 1148 * 1149 * Queue a buffer at the end of a list. This function takes no locks 1150 * and you must therefore hold required locks before calling it. 1151 * 1152 * A buffer cannot be placed on two lists at the same time. 1153 */ 1154extern void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1155static inline void __skb_queue_tail(struct sk_buff_head *list, 1156 struct sk_buff *newsk) 1157{ 1158 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1159} 1160 1161/* 1162 * remove sk_buff from list. _Must_ be called atomically, and with 1163 * the list known.. 1164 */ 1165extern void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1166static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1167{ 1168 struct sk_buff *next, *prev; 1169 1170 list->qlen--; 1171 next = skb->next; 1172 prev = skb->prev; 1173 skb->next = skb->prev = NULL; 1174 next->prev = prev; 1175 prev->next = next; 1176} 1177 1178/** 1179 * __skb_dequeue - remove from the head of the queue 1180 * @list: list to dequeue from 1181 * 1182 * Remove the head of the list. This function does not take any locks 1183 * so must be used with appropriate locks held only. The head item is 1184 * returned or %NULL if the list is empty. 1185 */ 1186extern struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1187static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1188{ 1189 struct sk_buff *skb = skb_peek(list); 1190 if (skb) 1191 __skb_unlink(skb, list); 1192 return skb; 1193} 1194 1195/** 1196 * __skb_dequeue_tail - remove from the tail of the queue 1197 * @list: list to dequeue from 1198 * 1199 * Remove the tail of the list. This function does not take any locks 1200 * so must be used with appropriate locks held only. The tail item is 1201 * returned or %NULL if the list is empty. 1202 */ 1203extern struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1204static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1205{ 1206 struct sk_buff *skb = skb_peek_tail(list); 1207 if (skb) 1208 __skb_unlink(skb, list); 1209 return skb; 1210} 1211 1212 1213static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1214{ 1215 return skb->data_len; 1216} 1217 1218static inline unsigned int skb_headlen(const struct sk_buff *skb) 1219{ 1220 return skb->len - skb->data_len; 1221} 1222 1223static inline int skb_pagelen(const struct sk_buff *skb) 1224{ 1225 int i, len = 0; 1226 1227 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1228 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1229 return len + skb_headlen(skb); 1230} 1231 1232/** 1233 * __skb_fill_page_desc - initialise a paged fragment in an skb 1234 * @skb: buffer containing fragment to be initialised 1235 * @i: paged fragment index to initialise 1236 * @page: the page to use for this fragment 1237 * @off: the offset to the data with @page 1238 * @size: the length of the data 1239 * 1240 * Initialises the @i'th fragment of @skb to point to &size bytes at 1241 * offset @off within @page. 1242 * 1243 * Does not take any additional reference on the fragment. 1244 */ 1245static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1246 struct page *page, int off, int size) 1247{ 1248 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1249 1250 /* 1251 * Propagate page->pfmemalloc to the skb if we can. The problem is 1252 * that not all callers have unique ownership of the page. If 1253 * pfmemalloc is set, we check the mapping as a mapping implies 1254 * page->index is set (index and pfmemalloc share space). 1255 * If it's a valid mapping, we cannot use page->pfmemalloc but we 1256 * do not lose pfmemalloc information as the pages would not be 1257 * allocated using __GFP_MEMALLOC. 1258 */ 1259 if (page->pfmemalloc && !page->mapping) 1260 skb->pfmemalloc = true; 1261 frag->page.p = page; 1262 frag->page_offset = off; 1263 skb_frag_size_set(frag, size); 1264} 1265 1266/** 1267 * skb_fill_page_desc - initialise a paged fragment in an skb 1268 * @skb: buffer containing fragment to be initialised 1269 * @i: paged fragment index to initialise 1270 * @page: the page to use for this fragment 1271 * @off: the offset to the data with @page 1272 * @size: the length of the data 1273 * 1274 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1275 * @skb to point to &size bytes at offset @off within @page. In 1276 * addition updates @skb such that @i is the last fragment. 1277 * 1278 * Does not take any additional reference on the fragment. 1279 */ 1280static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1281 struct page *page, int off, int size) 1282{ 1283 __skb_fill_page_desc(skb, i, page, off, size); 1284 skb_shinfo(skb)->nr_frags = i + 1; 1285} 1286 1287extern void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, 1288 int off, int size, unsigned int truesize); 1289 1290#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1291#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1292#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1293 1294#ifdef NET_SKBUFF_DATA_USES_OFFSET 1295static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1296{ 1297 return skb->head + skb->tail; 1298} 1299 1300static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1301{ 1302 skb->tail = skb->data - skb->head; 1303} 1304 1305static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1306{ 1307 skb_reset_tail_pointer(skb); 1308 skb->tail += offset; 1309} 1310#else /* NET_SKBUFF_DATA_USES_OFFSET */ 1311static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1312{ 1313 return skb->tail; 1314} 1315 1316static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1317{ 1318 skb->tail = skb->data; 1319} 1320 1321static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1322{ 1323 skb->tail = skb->data + offset; 1324} 1325 1326#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1327 1328/* 1329 * Add data to an sk_buff 1330 */ 1331extern unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1332static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1333{ 1334 unsigned char *tmp = skb_tail_pointer(skb); 1335 SKB_LINEAR_ASSERT(skb); 1336 skb->tail += len; 1337 skb->len += len; 1338 return tmp; 1339} 1340 1341extern unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1342static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1343{ 1344 skb->data -= len; 1345 skb->len += len; 1346 return skb->data; 1347} 1348 1349extern unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1350static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1351{ 1352 skb->len -= len; 1353 BUG_ON(skb->len < skb->data_len); 1354 return skb->data += len; 1355} 1356 1357static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1358{ 1359 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1360} 1361 1362extern unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1363 1364static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1365{ 1366 if (len > skb_headlen(skb) && 1367 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1368 return NULL; 1369 skb->len -= len; 1370 return skb->data += len; 1371} 1372 1373static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1374{ 1375 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1376} 1377 1378static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1379{ 1380 if (likely(len <= skb_headlen(skb))) 1381 return 1; 1382 if (unlikely(len > skb->len)) 1383 return 0; 1384 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1385} 1386 1387/** 1388 * skb_headroom - bytes at buffer head 1389 * @skb: buffer to check 1390 * 1391 * Return the number of bytes of free space at the head of an &sk_buff. 1392 */ 1393static inline unsigned int skb_headroom(const struct sk_buff *skb) 1394{ 1395 return skb->data - skb->head; 1396} 1397 1398/** 1399 * skb_tailroom - bytes at buffer end 1400 * @skb: buffer to check 1401 * 1402 * Return the number of bytes of free space at the tail of an sk_buff 1403 */ 1404static inline int skb_tailroom(const struct sk_buff *skb) 1405{ 1406 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1407} 1408 1409/** 1410 * skb_availroom - bytes at buffer end 1411 * @skb: buffer to check 1412 * 1413 * Return the number of bytes of free space at the tail of an sk_buff 1414 * allocated by sk_stream_alloc() 1415 */ 1416static inline int skb_availroom(const struct sk_buff *skb) 1417{ 1418 return skb_is_nonlinear(skb) ? 0 : skb->avail_size - skb->len; 1419} 1420 1421/** 1422 * skb_reserve - adjust headroom 1423 * @skb: buffer to alter 1424 * @len: bytes to move 1425 * 1426 * Increase the headroom of an empty &sk_buff by reducing the tail 1427 * room. This is only allowed for an empty buffer. 1428 */ 1429static inline void skb_reserve(struct sk_buff *skb, int len) 1430{ 1431 skb->data += len; 1432 skb->tail += len; 1433} 1434 1435static inline void skb_reset_mac_len(struct sk_buff *skb) 1436{ 1437 skb->mac_len = skb->network_header - skb->mac_header; 1438} 1439 1440#ifdef NET_SKBUFF_DATA_USES_OFFSET 1441static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 1442{ 1443 return skb->head + skb->transport_header; 1444} 1445 1446static inline void skb_reset_transport_header(struct sk_buff *skb) 1447{ 1448 skb->transport_header = skb->data - skb->head; 1449} 1450 1451static inline void skb_set_transport_header(struct sk_buff *skb, 1452 const int offset) 1453{ 1454 skb_reset_transport_header(skb); 1455 skb->transport_header += offset; 1456} 1457 1458static inline unsigned char *skb_network_header(const struct sk_buff *skb) 1459{ 1460 return skb->head + skb->network_header; 1461} 1462 1463static inline void skb_reset_network_header(struct sk_buff *skb) 1464{ 1465 skb->network_header = skb->data - skb->head; 1466} 1467 1468static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 1469{ 1470 skb_reset_network_header(skb); 1471 skb->network_header += offset; 1472} 1473 1474static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 1475{ 1476 return skb->head + skb->mac_header; 1477} 1478 1479static inline int skb_mac_header_was_set(const struct sk_buff *skb) 1480{ 1481 return skb->mac_header != ~0U; 1482} 1483 1484static inline void skb_reset_mac_header(struct sk_buff *skb) 1485{ 1486 skb->mac_header = skb->data - skb->head; 1487} 1488 1489static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 1490{ 1491 skb_reset_mac_header(skb); 1492 skb->mac_header += offset; 1493} 1494 1495#else /* NET_SKBUFF_DATA_USES_OFFSET */ 1496 1497static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 1498{ 1499 return skb->transport_header; 1500} 1501 1502static inline void skb_reset_transport_header(struct sk_buff *skb) 1503{ 1504 skb->transport_header = skb->data; 1505} 1506 1507static inline void skb_set_transport_header(struct sk_buff *skb, 1508 const int offset) 1509{ 1510 skb->transport_header = skb->data + offset; 1511} 1512 1513static inline unsigned char *skb_network_header(const struct sk_buff *skb) 1514{ 1515 return skb->network_header; 1516} 1517 1518static inline void skb_reset_network_header(struct sk_buff *skb) 1519{ 1520 skb->network_header = skb->data; 1521} 1522 1523static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 1524{ 1525 skb->network_header = skb->data + offset; 1526} 1527 1528static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 1529{ 1530 return skb->mac_header; 1531} 1532 1533static inline int skb_mac_header_was_set(const struct sk_buff *skb) 1534{ 1535 return skb->mac_header != NULL; 1536} 1537 1538static inline void skb_reset_mac_header(struct sk_buff *skb) 1539{ 1540 skb->mac_header = skb->data; 1541} 1542 1543static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 1544{ 1545 skb->mac_header = skb->data + offset; 1546} 1547#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1548 1549static inline void skb_mac_header_rebuild(struct sk_buff *skb) 1550{ 1551 if (skb_mac_header_was_set(skb)) { 1552 const unsigned char *old_mac = skb_mac_header(skb); 1553 1554 skb_set_mac_header(skb, -skb->mac_len); 1555 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 1556 } 1557} 1558 1559static inline int skb_checksum_start_offset(const struct sk_buff *skb) 1560{ 1561 return skb->csum_start - skb_headroom(skb); 1562} 1563 1564static inline int skb_transport_offset(const struct sk_buff *skb) 1565{ 1566 return skb_transport_header(skb) - skb->data; 1567} 1568 1569static inline u32 skb_network_header_len(const struct sk_buff *skb) 1570{ 1571 return skb->transport_header - skb->network_header; 1572} 1573 1574static inline int skb_network_offset(const struct sk_buff *skb) 1575{ 1576 return skb_network_header(skb) - skb->data; 1577} 1578 1579static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 1580{ 1581 return pskb_may_pull(skb, skb_network_offset(skb) + len); 1582} 1583 1584/* 1585 * CPUs often take a performance hit when accessing unaligned memory 1586 * locations. The actual performance hit varies, it can be small if the 1587 * hardware handles it or large if we have to take an exception and fix it 1588 * in software. 1589 * 1590 * Since an ethernet header is 14 bytes network drivers often end up with 1591 * the IP header at an unaligned offset. The IP header can be aligned by 1592 * shifting the start of the packet by 2 bytes. Drivers should do this 1593 * with: 1594 * 1595 * skb_reserve(skb, NET_IP_ALIGN); 1596 * 1597 * The downside to this alignment of the IP header is that the DMA is now 1598 * unaligned. On some architectures the cost of an unaligned DMA is high 1599 * and this cost outweighs the gains made by aligning the IP header. 1600 * 1601 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 1602 * to be overridden. 1603 */ 1604#ifndef NET_IP_ALIGN 1605#define NET_IP_ALIGN 2 1606#endif 1607 1608/* 1609 * The networking layer reserves some headroom in skb data (via 1610 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 1611 * the header has to grow. In the default case, if the header has to grow 1612 * 32 bytes or less we avoid the reallocation. 1613 * 1614 * Unfortunately this headroom changes the DMA alignment of the resulting 1615 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 1616 * on some architectures. An architecture can override this value, 1617 * perhaps setting it to a cacheline in size (since that will maintain 1618 * cacheline alignment of the DMA). It must be a power of 2. 1619 * 1620 * Various parts of the networking layer expect at least 32 bytes of 1621 * headroom, you should not reduce this. 1622 * 1623 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 1624 * to reduce average number of cache lines per packet. 1625 * get_rps_cpus() for example only access one 64 bytes aligned block : 1626 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 1627 */ 1628#ifndef NET_SKB_PAD 1629#define NET_SKB_PAD max(32, L1_CACHE_BYTES) 1630#endif 1631 1632extern int ___pskb_trim(struct sk_buff *skb, unsigned int len); 1633 1634static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 1635{ 1636 if (unlikely(skb_is_nonlinear(skb))) { 1637 WARN_ON(1); 1638 return; 1639 } 1640 skb->len = len; 1641 skb_set_tail_pointer(skb, len); 1642} 1643 1644extern void skb_trim(struct sk_buff *skb, unsigned int len); 1645 1646static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 1647{ 1648 if (skb->data_len) 1649 return ___pskb_trim(skb, len); 1650 __skb_trim(skb, len); 1651 return 0; 1652} 1653 1654static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 1655{ 1656 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 1657} 1658 1659/** 1660 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 1661 * @skb: buffer to alter 1662 * @len: new length 1663 * 1664 * This is identical to pskb_trim except that the caller knows that 1665 * the skb is not cloned so we should never get an error due to out- 1666 * of-memory. 1667 */ 1668static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 1669{ 1670 int err = pskb_trim(skb, len); 1671 BUG_ON(err); 1672} 1673 1674/** 1675 * skb_orphan - orphan a buffer 1676 * @skb: buffer to orphan 1677 * 1678 * If a buffer currently has an owner then we call the owner's 1679 * destructor function and make the @skb unowned. The buffer continues 1680 * to exist but is no longer charged to its former owner. 1681 */ 1682static inline void skb_orphan(struct sk_buff *skb) 1683{ 1684 if (skb->destructor) 1685 skb->destructor(skb); 1686 skb->destructor = NULL; 1687 skb->sk = NULL; 1688} 1689 1690/** 1691 * skb_orphan_frags - orphan the frags contained in a buffer 1692 * @skb: buffer to orphan frags from 1693 * @gfp_mask: allocation mask for replacement pages 1694 * 1695 * For each frag in the SKB which needs a destructor (i.e. has an 1696 * owner) create a copy of that frag and release the original 1697 * page by calling the destructor. 1698 */ 1699static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 1700{ 1701 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 1702 return 0; 1703 return skb_copy_ubufs(skb, gfp_mask); 1704} 1705 1706/** 1707 * __skb_queue_purge - empty a list 1708 * @list: list to empty 1709 * 1710 * Delete all buffers on an &sk_buff list. Each buffer is removed from 1711 * the list and one reference dropped. This function does not take the 1712 * list lock and the caller must hold the relevant locks to use it. 1713 */ 1714extern void skb_queue_purge(struct sk_buff_head *list); 1715static inline void __skb_queue_purge(struct sk_buff_head *list) 1716{ 1717 struct sk_buff *skb; 1718 while ((skb = __skb_dequeue(list)) != NULL) 1719 kfree_skb(skb); 1720} 1721 1722extern void *netdev_alloc_frag(unsigned int fragsz); 1723 1724extern struct sk_buff *__netdev_alloc_skb(struct net_device *dev, 1725 unsigned int length, 1726 gfp_t gfp_mask); 1727 1728/** 1729 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 1730 * @dev: network device to receive on 1731 * @length: length to allocate 1732 * 1733 * Allocate a new &sk_buff and assign it a usage count of one. The 1734 * buffer has unspecified headroom built in. Users should allocate 1735 * the headroom they think they need without accounting for the 1736 * built in space. The built in space is used for optimisations. 1737 * 1738 * %NULL is returned if there is no free memory. Although this function 1739 * allocates memory it can be called from an interrupt. 1740 */ 1741static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 1742 unsigned int length) 1743{ 1744 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 1745} 1746 1747/* legacy helper around __netdev_alloc_skb() */ 1748static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 1749 gfp_t gfp_mask) 1750{ 1751 return __netdev_alloc_skb(NULL, length, gfp_mask); 1752} 1753 1754/* legacy helper around netdev_alloc_skb() */ 1755static inline struct sk_buff *dev_alloc_skb(unsigned int length) 1756{ 1757 return netdev_alloc_skb(NULL, length); 1758} 1759 1760 1761static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 1762 unsigned int length, gfp_t gfp) 1763{ 1764 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 1765 1766 if (NET_IP_ALIGN && skb) 1767 skb_reserve(skb, NET_IP_ALIGN); 1768 return skb; 1769} 1770 1771static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 1772 unsigned int length) 1773{ 1774 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 1775} 1776 1777/* 1778 * __skb_alloc_page - allocate pages for ps-rx on a skb and preserve pfmemalloc data 1779 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX 1780 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used 1781 * @order: size of the allocation 1782 * 1783 * Allocate a new page. 1784 * 1785 * %NULL is returned if there is no free memory. 1786*/ 1787static inline struct page *__skb_alloc_pages(gfp_t gfp_mask, 1788 struct sk_buff *skb, 1789 unsigned int order) 1790{ 1791 struct page *page; 1792 1793 gfp_mask |= __GFP_COLD; 1794 1795 if (!(gfp_mask & __GFP_NOMEMALLOC)) 1796 gfp_mask |= __GFP_MEMALLOC; 1797 1798 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 1799 if (skb && page && page->pfmemalloc) 1800 skb->pfmemalloc = true; 1801 1802 return page; 1803} 1804 1805/** 1806 * __skb_alloc_page - allocate a page for ps-rx for a given skb and preserve pfmemalloc data 1807 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX 1808 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used 1809 * 1810 * Allocate a new page. 1811 * 1812 * %NULL is returned if there is no free memory. 1813 */ 1814static inline struct page *__skb_alloc_page(gfp_t gfp_mask, 1815 struct sk_buff *skb) 1816{ 1817 return __skb_alloc_pages(gfp_mask, skb, 0); 1818} 1819 1820/** 1821 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 1822 * @page: The page that was allocated from skb_alloc_page 1823 * @skb: The skb that may need pfmemalloc set 1824 */ 1825static inline void skb_propagate_pfmemalloc(struct page *page, 1826 struct sk_buff *skb) 1827{ 1828 if (page && page->pfmemalloc) 1829 skb->pfmemalloc = true; 1830} 1831 1832/** 1833 * skb_frag_page - retrieve the page refered to by a paged fragment 1834 * @frag: the paged fragment 1835 * 1836 * Returns the &struct page associated with @frag. 1837 */ 1838static inline struct page *skb_frag_page(const skb_frag_t *frag) 1839{ 1840 return frag->page.p; 1841} 1842 1843/** 1844 * __skb_frag_ref - take an addition reference on a paged fragment. 1845 * @frag: the paged fragment 1846 * 1847 * Takes an additional reference on the paged fragment @frag. 1848 */ 1849static inline void __skb_frag_ref(skb_frag_t *frag) 1850{ 1851 get_page(skb_frag_page(frag)); 1852} 1853 1854/** 1855 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 1856 * @skb: the buffer 1857 * @f: the fragment offset. 1858 * 1859 * Takes an additional reference on the @f'th paged fragment of @skb. 1860 */ 1861static inline void skb_frag_ref(struct sk_buff *skb, int f) 1862{ 1863 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 1864} 1865 1866/** 1867 * __skb_frag_unref - release a reference on a paged fragment. 1868 * @frag: the paged fragment 1869 * 1870 * Releases a reference on the paged fragment @frag. 1871 */ 1872static inline void __skb_frag_unref(skb_frag_t *frag) 1873{ 1874 put_page(skb_frag_page(frag)); 1875} 1876 1877/** 1878 * skb_frag_unref - release a reference on a paged fragment of an skb. 1879 * @skb: the buffer 1880 * @f: the fragment offset 1881 * 1882 * Releases a reference on the @f'th paged fragment of @skb. 1883 */ 1884static inline void skb_frag_unref(struct sk_buff *skb, int f) 1885{ 1886 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 1887} 1888 1889/** 1890 * skb_frag_address - gets the address of the data contained in a paged fragment 1891 * @frag: the paged fragment buffer 1892 * 1893 * Returns the address of the data within @frag. The page must already 1894 * be mapped. 1895 */ 1896static inline void *skb_frag_address(const skb_frag_t *frag) 1897{ 1898 return page_address(skb_frag_page(frag)) + frag->page_offset; 1899} 1900 1901/** 1902 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 1903 * @frag: the paged fragment buffer 1904 * 1905 * Returns the address of the data within @frag. Checks that the page 1906 * is mapped and returns %NULL otherwise. 1907 */ 1908static inline void *skb_frag_address_safe(const skb_frag_t *frag) 1909{ 1910 void *ptr = page_address(skb_frag_page(frag)); 1911 if (unlikely(!ptr)) 1912 return NULL; 1913 1914 return ptr + frag->page_offset; 1915} 1916 1917/** 1918 * __skb_frag_set_page - sets the page contained in a paged fragment 1919 * @frag: the paged fragment 1920 * @page: the page to set 1921 * 1922 * Sets the fragment @frag to contain @page. 1923 */ 1924static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 1925{ 1926 frag->page.p = page; 1927} 1928 1929/** 1930 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 1931 * @skb: the buffer 1932 * @f: the fragment offset 1933 * @page: the page to set 1934 * 1935 * Sets the @f'th fragment of @skb to contain @page. 1936 */ 1937static inline void skb_frag_set_page(struct sk_buff *skb, int f, 1938 struct page *page) 1939{ 1940 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 1941} 1942 1943/** 1944 * skb_frag_dma_map - maps a paged fragment via the DMA API 1945 * @dev: the device to map the fragment to 1946 * @frag: the paged fragment to map 1947 * @offset: the offset within the fragment (starting at the 1948 * fragment's own offset) 1949 * @size: the number of bytes to map 1950 * @dir: the direction of the mapping (%PCI_DMA_*) 1951 * 1952 * Maps the page associated with @frag to @device. 1953 */ 1954static inline dma_addr_t skb_frag_dma_map(struct device *dev, 1955 const skb_frag_t *frag, 1956 size_t offset, size_t size, 1957 enum dma_data_direction dir) 1958{ 1959 return dma_map_page(dev, skb_frag_page(frag), 1960 frag->page_offset + offset, size, dir); 1961} 1962 1963static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 1964 gfp_t gfp_mask) 1965{ 1966 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 1967} 1968 1969/** 1970 * skb_clone_writable - is the header of a clone writable 1971 * @skb: buffer to check 1972 * @len: length up to which to write 1973 * 1974 * Returns true if modifying the header part of the cloned buffer 1975 * does not requires the data to be copied. 1976 */ 1977static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 1978{ 1979 return !skb_header_cloned(skb) && 1980 skb_headroom(skb) + len <= skb->hdr_len; 1981} 1982 1983static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 1984 int cloned) 1985{ 1986 int delta = 0; 1987 1988 if (headroom > skb_headroom(skb)) 1989 delta = headroom - skb_headroom(skb); 1990 1991 if (delta || cloned) 1992 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 1993 GFP_ATOMIC); 1994 return 0; 1995} 1996 1997/** 1998 * skb_cow - copy header of skb when it is required 1999 * @skb: buffer to cow 2000 * @headroom: needed headroom 2001 * 2002 * If the skb passed lacks sufficient headroom or its data part 2003 * is shared, data is reallocated. If reallocation fails, an error 2004 * is returned and original skb is not changed. 2005 * 2006 * The result is skb with writable area skb->head...skb->tail 2007 * and at least @headroom of space at head. 2008 */ 2009static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2010{ 2011 return __skb_cow(skb, headroom, skb_cloned(skb)); 2012} 2013 2014/** 2015 * skb_cow_head - skb_cow but only making the head writable 2016 * @skb: buffer to cow 2017 * @headroom: needed headroom 2018 * 2019 * This function is identical to skb_cow except that we replace the 2020 * skb_cloned check by skb_header_cloned. It should be used when 2021 * you only need to push on some header and do not need to modify 2022 * the data. 2023 */ 2024static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2025{ 2026 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2027} 2028 2029/** 2030 * skb_padto - pad an skbuff up to a minimal size 2031 * @skb: buffer to pad 2032 * @len: minimal length 2033 * 2034 * Pads up a buffer to ensure the trailing bytes exist and are 2035 * blanked. If the buffer already contains sufficient data it 2036 * is untouched. Otherwise it is extended. Returns zero on 2037 * success. The skb is freed on error. 2038 */ 2039 2040static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2041{ 2042 unsigned int size = skb->len; 2043 if (likely(size >= len)) 2044 return 0; 2045 return skb_pad(skb, len - size); 2046} 2047 2048static inline int skb_add_data(struct sk_buff *skb, 2049 char __user *from, int copy) 2050{ 2051 const int off = skb->len; 2052 2053 if (skb->ip_summed == CHECKSUM_NONE) { 2054 int err = 0; 2055 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy), 2056 copy, 0, &err); 2057 if (!err) { 2058 skb->csum = csum_block_add(skb->csum, csum, off); 2059 return 0; 2060 } 2061 } else if (!copy_from_user(skb_put(skb, copy), from, copy)) 2062 return 0; 2063 2064 __skb_trim(skb, off); 2065 return -EFAULT; 2066} 2067 2068static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2069 const struct page *page, int off) 2070{ 2071 if (i) { 2072 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2073 2074 return page == skb_frag_page(frag) && 2075 off == frag->page_offset + skb_frag_size(frag); 2076 } 2077 return false; 2078} 2079 2080static inline int __skb_linearize(struct sk_buff *skb) 2081{ 2082 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2083} 2084 2085/** 2086 * skb_linearize - convert paged skb to linear one 2087 * @skb: buffer to linarize 2088 * 2089 * If there is no free memory -ENOMEM is returned, otherwise zero 2090 * is returned and the old skb data released. 2091 */ 2092static inline int skb_linearize(struct sk_buff *skb) 2093{ 2094 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2095} 2096 2097/** 2098 * skb_linearize_cow - make sure skb is linear and writable 2099 * @skb: buffer to process 2100 * 2101 * If there is no free memory -ENOMEM is returned, otherwise zero 2102 * is returned and the old skb data released. 2103 */ 2104static inline int skb_linearize_cow(struct sk_buff *skb) 2105{ 2106 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2107 __skb_linearize(skb) : 0; 2108} 2109 2110/** 2111 * skb_postpull_rcsum - update checksum for received skb after pull 2112 * @skb: buffer to update 2113 * @start: start of data before pull 2114 * @len: length of data pulled 2115 * 2116 * After doing a pull on a received packet, you need to call this to 2117 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2118 * CHECKSUM_NONE so that it can be recomputed from scratch. 2119 */ 2120 2121static inline void skb_postpull_rcsum(struct sk_buff *skb, 2122 const void *start, unsigned int len) 2123{ 2124 if (skb->ip_summed == CHECKSUM_COMPLETE) 2125 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0)); 2126} 2127 2128unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2129 2130/** 2131 * pskb_trim_rcsum - trim received skb and update checksum 2132 * @skb: buffer to trim 2133 * @len: new length 2134 * 2135 * This is exactly the same as pskb_trim except that it ensures the 2136 * checksum of received packets are still valid after the operation. 2137 */ 2138 2139static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2140{ 2141 if (likely(len >= skb->len)) 2142 return 0; 2143 if (skb->ip_summed == CHECKSUM_COMPLETE) 2144 skb->ip_summed = CHECKSUM_NONE; 2145 return __pskb_trim(skb, len); 2146} 2147 2148#define skb_queue_walk(queue, skb) \ 2149 for (skb = (queue)->next; \ 2150 skb != (struct sk_buff *)(queue); \ 2151 skb = skb->next) 2152 2153#define skb_queue_walk_safe(queue, skb, tmp) \ 2154 for (skb = (queue)->next, tmp = skb->next; \ 2155 skb != (struct sk_buff *)(queue); \ 2156 skb = tmp, tmp = skb->next) 2157 2158#define skb_queue_walk_from(queue, skb) \ 2159 for (; skb != (struct sk_buff *)(queue); \ 2160 skb = skb->next) 2161 2162#define skb_queue_walk_from_safe(queue, skb, tmp) \ 2163 for (tmp = skb->next; \ 2164 skb != (struct sk_buff *)(queue); \ 2165 skb = tmp, tmp = skb->next) 2166 2167#define skb_queue_reverse_walk(queue, skb) \ 2168 for (skb = (queue)->prev; \ 2169 skb != (struct sk_buff *)(queue); \ 2170 skb = skb->prev) 2171 2172#define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 2173 for (skb = (queue)->prev, tmp = skb->prev; \ 2174 skb != (struct sk_buff *)(queue); \ 2175 skb = tmp, tmp = skb->prev) 2176 2177#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 2178 for (tmp = skb->prev; \ 2179 skb != (struct sk_buff *)(queue); \ 2180 skb = tmp, tmp = skb->prev) 2181 2182static inline bool skb_has_frag_list(const struct sk_buff *skb) 2183{ 2184 return skb_shinfo(skb)->frag_list != NULL; 2185} 2186 2187static inline void skb_frag_list_init(struct sk_buff *skb) 2188{ 2189 skb_shinfo(skb)->frag_list = NULL; 2190} 2191 2192static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag) 2193{ 2194 frag->next = skb_shinfo(skb)->frag_list; 2195 skb_shinfo(skb)->frag_list = frag; 2196} 2197 2198#define skb_walk_frags(skb, iter) \ 2199 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 2200 2201extern struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 2202 int *peeked, int *off, int *err); 2203extern struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, 2204 int noblock, int *err); 2205extern unsigned int datagram_poll(struct file *file, struct socket *sock, 2206 struct poll_table_struct *wait); 2207extern int skb_copy_datagram_iovec(const struct sk_buff *from, 2208 int offset, struct iovec *to, 2209 int size); 2210extern int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, 2211 int hlen, 2212 struct iovec *iov); 2213extern int skb_copy_datagram_from_iovec(struct sk_buff *skb, 2214 int offset, 2215 const struct iovec *from, 2216 int from_offset, 2217 int len); 2218extern int skb_copy_datagram_const_iovec(const struct sk_buff *from, 2219 int offset, 2220 const struct iovec *to, 2221 int to_offset, 2222 int size); 2223extern void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 2224extern void skb_free_datagram_locked(struct sock *sk, 2225 struct sk_buff *skb); 2226extern int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, 2227 unsigned int flags); 2228extern __wsum skb_checksum(const struct sk_buff *skb, int offset, 2229 int len, __wsum csum); 2230extern int skb_copy_bits(const struct sk_buff *skb, int offset, 2231 void *to, int len); 2232extern int skb_store_bits(struct sk_buff *skb, int offset, 2233 const void *from, int len); 2234extern __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, 2235 int offset, u8 *to, int len, 2236 __wsum csum); 2237extern int skb_splice_bits(struct sk_buff *skb, 2238 unsigned int offset, 2239 struct pipe_inode_info *pipe, 2240 unsigned int len, 2241 unsigned int flags); 2242extern void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 2243extern void skb_split(struct sk_buff *skb, 2244 struct sk_buff *skb1, const u32 len); 2245extern int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, 2246 int shiftlen); 2247 2248extern struct sk_buff *skb_segment(struct sk_buff *skb, 2249 netdev_features_t features); 2250 2251static inline void *skb_header_pointer(const struct sk_buff *skb, int offset, 2252 int len, void *buffer) 2253{ 2254 int hlen = skb_headlen(skb); 2255 2256 if (hlen - offset >= len) 2257 return skb->data + offset; 2258 2259 if (skb_copy_bits(skb, offset, buffer, len) < 0) 2260 return NULL; 2261 2262 return buffer; 2263} 2264 2265static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 2266 void *to, 2267 const unsigned int len) 2268{ 2269 memcpy(to, skb->data, len); 2270} 2271 2272static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 2273 const int offset, void *to, 2274 const unsigned int len) 2275{ 2276 memcpy(to, skb->data + offset, len); 2277} 2278 2279static inline void skb_copy_to_linear_data(struct sk_buff *skb, 2280 const void *from, 2281 const unsigned int len) 2282{ 2283 memcpy(skb->data, from, len); 2284} 2285 2286static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 2287 const int offset, 2288 const void *from, 2289 const unsigned int len) 2290{ 2291 memcpy(skb->data + offset, from, len); 2292} 2293 2294extern void skb_init(void); 2295 2296static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 2297{ 2298 return skb->tstamp; 2299} 2300 2301/** 2302 * skb_get_timestamp - get timestamp from a skb 2303 * @skb: skb to get stamp from 2304 * @stamp: pointer to struct timeval to store stamp in 2305 * 2306 * Timestamps are stored in the skb as offsets to a base timestamp. 2307 * This function converts the offset back to a struct timeval and stores 2308 * it in stamp. 2309 */ 2310static inline void skb_get_timestamp(const struct sk_buff *skb, 2311 struct timeval *stamp) 2312{ 2313 *stamp = ktime_to_timeval(skb->tstamp); 2314} 2315 2316static inline void skb_get_timestampns(const struct sk_buff *skb, 2317 struct timespec *stamp) 2318{ 2319 *stamp = ktime_to_timespec(skb->tstamp); 2320} 2321 2322static inline void __net_timestamp(struct sk_buff *skb) 2323{ 2324 skb->tstamp = ktime_get_real(); 2325} 2326 2327static inline ktime_t net_timedelta(ktime_t t) 2328{ 2329 return ktime_sub(ktime_get_real(), t); 2330} 2331 2332static inline ktime_t net_invalid_timestamp(void) 2333{ 2334 return ktime_set(0, 0); 2335} 2336 2337extern void skb_timestamping_init(void); 2338 2339#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 2340 2341extern void skb_clone_tx_timestamp(struct sk_buff *skb); 2342extern bool skb_defer_rx_timestamp(struct sk_buff *skb); 2343 2344#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 2345 2346static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 2347{ 2348} 2349 2350static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 2351{ 2352 return false; 2353} 2354 2355#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 2356 2357/** 2358 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 2359 * 2360 * PHY drivers may accept clones of transmitted packets for 2361 * timestamping via their phy_driver.txtstamp method. These drivers 2362 * must call this function to return the skb back to the stack, with 2363 * or without a timestamp. 2364 * 2365 * @skb: clone of the the original outgoing packet 2366 * @hwtstamps: hardware time stamps, may be NULL if not available 2367 * 2368 */ 2369void skb_complete_tx_timestamp(struct sk_buff *skb, 2370 struct skb_shared_hwtstamps *hwtstamps); 2371 2372/** 2373 * skb_tstamp_tx - queue clone of skb with send time stamps 2374 * @orig_skb: the original outgoing packet 2375 * @hwtstamps: hardware time stamps, may be NULL if not available 2376 * 2377 * If the skb has a socket associated, then this function clones the 2378 * skb (thus sharing the actual data and optional structures), stores 2379 * the optional hardware time stamping information (if non NULL) or 2380 * generates a software time stamp (otherwise), then queues the clone 2381 * to the error queue of the socket. Errors are silently ignored. 2382 */ 2383extern void skb_tstamp_tx(struct sk_buff *orig_skb, 2384 struct skb_shared_hwtstamps *hwtstamps); 2385 2386static inline void sw_tx_timestamp(struct sk_buff *skb) 2387{ 2388 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 2389 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 2390 skb_tstamp_tx(skb, NULL); 2391} 2392 2393/** 2394 * skb_tx_timestamp() - Driver hook for transmit timestamping 2395 * 2396 * Ethernet MAC Drivers should call this function in their hard_xmit() 2397 * function immediately before giving the sk_buff to the MAC hardware. 2398 * 2399 * @skb: A socket buffer. 2400 */ 2401static inline void skb_tx_timestamp(struct sk_buff *skb) 2402{ 2403 skb_clone_tx_timestamp(skb); 2404 sw_tx_timestamp(skb); 2405} 2406 2407/** 2408 * skb_complete_wifi_ack - deliver skb with wifi status 2409 * 2410 * @skb: the original outgoing packet 2411 * @acked: ack status 2412 * 2413 */ 2414void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 2415 2416extern __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 2417extern __sum16 __skb_checksum_complete(struct sk_buff *skb); 2418 2419static inline int skb_csum_unnecessary(const struct sk_buff *skb) 2420{ 2421 return skb->ip_summed & CHECKSUM_UNNECESSARY; 2422} 2423 2424/** 2425 * skb_checksum_complete - Calculate checksum of an entire packet 2426 * @skb: packet to process 2427 * 2428 * This function calculates the checksum over the entire packet plus 2429 * the value of skb->csum. The latter can be used to supply the 2430 * checksum of a pseudo header as used by TCP/UDP. It returns the 2431 * checksum. 2432 * 2433 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 2434 * this function can be used to verify that checksum on received 2435 * packets. In that case the function should return zero if the 2436 * checksum is correct. In particular, this function will return zero 2437 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 2438 * hardware has already verified the correctness of the checksum. 2439 */ 2440static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 2441{ 2442 return skb_csum_unnecessary(skb) ? 2443 0 : __skb_checksum_complete(skb); 2444} 2445 2446#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 2447extern void nf_conntrack_destroy(struct nf_conntrack *nfct); 2448static inline void nf_conntrack_put(struct nf_conntrack *nfct) 2449{ 2450 if (nfct && atomic_dec_and_test(&nfct->use)) 2451 nf_conntrack_destroy(nfct); 2452} 2453static inline void nf_conntrack_get(struct nf_conntrack *nfct) 2454{ 2455 if (nfct) 2456 atomic_inc(&nfct->use); 2457} 2458#endif 2459#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED 2460static inline void nf_conntrack_get_reasm(struct sk_buff *skb) 2461{ 2462 if (skb) 2463 atomic_inc(&skb->users); 2464} 2465static inline void nf_conntrack_put_reasm(struct sk_buff *skb) 2466{ 2467 if (skb) 2468 kfree_skb(skb); 2469} 2470#endif 2471#ifdef CONFIG_BRIDGE_NETFILTER 2472static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 2473{ 2474 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 2475 kfree(nf_bridge); 2476} 2477static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 2478{ 2479 if (nf_bridge) 2480 atomic_inc(&nf_bridge->use); 2481} 2482#endif /* CONFIG_BRIDGE_NETFILTER */ 2483static inline void nf_reset(struct sk_buff *skb) 2484{ 2485#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 2486 nf_conntrack_put(skb->nfct); 2487 skb->nfct = NULL; 2488#endif 2489#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED 2490 nf_conntrack_put_reasm(skb->nfct_reasm); 2491 skb->nfct_reasm = NULL; 2492#endif 2493#ifdef CONFIG_BRIDGE_NETFILTER 2494 nf_bridge_put(skb->nf_bridge); 2495 skb->nf_bridge = NULL; 2496#endif 2497} 2498 2499/* Note: This doesn't put any conntrack and bridge info in dst. */ 2500static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src) 2501{ 2502#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 2503 dst->nfct = src->nfct; 2504 nf_conntrack_get(src->nfct); 2505 dst->nfctinfo = src->nfctinfo; 2506#endif 2507#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED 2508 dst->nfct_reasm = src->nfct_reasm; 2509 nf_conntrack_get_reasm(src->nfct_reasm); 2510#endif 2511#ifdef CONFIG_BRIDGE_NETFILTER 2512 dst->nf_bridge = src->nf_bridge; 2513 nf_bridge_get(src->nf_bridge); 2514#endif 2515} 2516 2517static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 2518{ 2519#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 2520 nf_conntrack_put(dst->nfct); 2521#endif 2522#ifdef NET_SKBUFF_NF_DEFRAG_NEEDED 2523 nf_conntrack_put_reasm(dst->nfct_reasm); 2524#endif 2525#ifdef CONFIG_BRIDGE_NETFILTER 2526 nf_bridge_put(dst->nf_bridge); 2527#endif 2528 __nf_copy(dst, src); 2529} 2530 2531#ifdef CONFIG_NETWORK_SECMARK 2532static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 2533{ 2534 to->secmark = from->secmark; 2535} 2536 2537static inline void skb_init_secmark(struct sk_buff *skb) 2538{ 2539 skb->secmark = 0; 2540} 2541#else 2542static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 2543{ } 2544 2545static inline void skb_init_secmark(struct sk_buff *skb) 2546{ } 2547#endif 2548 2549static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 2550{ 2551 skb->queue_mapping = queue_mapping; 2552} 2553 2554static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 2555{ 2556 return skb->queue_mapping; 2557} 2558 2559static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 2560{ 2561 to->queue_mapping = from->queue_mapping; 2562} 2563 2564static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 2565{ 2566 skb->queue_mapping = rx_queue + 1; 2567} 2568 2569static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 2570{ 2571 return skb->queue_mapping - 1; 2572} 2573 2574static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 2575{ 2576 return skb->queue_mapping != 0; 2577} 2578 2579extern u16 __skb_tx_hash(const struct net_device *dev, 2580 const struct sk_buff *skb, 2581 unsigned int num_tx_queues); 2582 2583#ifdef CONFIG_XFRM 2584static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 2585{ 2586 return skb->sp; 2587} 2588#else 2589static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 2590{ 2591 return NULL; 2592} 2593#endif 2594 2595static inline bool skb_is_gso(const struct sk_buff *skb) 2596{ 2597 return skb_shinfo(skb)->gso_size; 2598} 2599 2600static inline bool skb_is_gso_v6(const struct sk_buff *skb) 2601{ 2602 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 2603} 2604 2605extern void __skb_warn_lro_forwarding(const struct sk_buff *skb); 2606 2607static inline bool skb_warn_if_lro(const struct sk_buff *skb) 2608{ 2609 /* LRO sets gso_size but not gso_type, whereas if GSO is really 2610 * wanted then gso_type will be set. */ 2611 const struct skb_shared_info *shinfo = skb_shinfo(skb); 2612 2613 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 2614 unlikely(shinfo->gso_type == 0)) { 2615 __skb_warn_lro_forwarding(skb); 2616 return true; 2617 } 2618 return false; 2619} 2620 2621static inline void skb_forward_csum(struct sk_buff *skb) 2622{ 2623 /* Unfortunately we don't support this one. Any brave souls? */ 2624 if (skb->ip_summed == CHECKSUM_COMPLETE) 2625 skb->ip_summed = CHECKSUM_NONE; 2626} 2627 2628/** 2629 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 2630 * @skb: skb to check 2631 * 2632 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 2633 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 2634 * use this helper, to document places where we make this assertion. 2635 */ 2636static inline void skb_checksum_none_assert(const struct sk_buff *skb) 2637{ 2638#ifdef DEBUG 2639 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 2640#endif 2641} 2642 2643bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 2644 2645static inline bool skb_is_recycleable(const struct sk_buff *skb, int skb_size) 2646{ 2647 if (irqs_disabled()) 2648 return false; 2649 2650 if (skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY) 2651 return false; 2652 2653 if (skb_is_nonlinear(skb) || skb->fclone != SKB_FCLONE_UNAVAILABLE) 2654 return false; 2655 2656 skb_size = SKB_DATA_ALIGN(skb_size + NET_SKB_PAD); 2657 if (skb_end_offset(skb) < skb_size) 2658 return false; 2659 2660 if (skb_shared(skb) || skb_cloned(skb)) 2661 return false; 2662 2663 return true; 2664} 2665 2666/** 2667 * skb_head_is_locked - Determine if the skb->head is locked down 2668 * @skb: skb to check 2669 * 2670 * The head on skbs build around a head frag can be removed if they are 2671 * not cloned. This function returns true if the skb head is locked down 2672 * due to either being allocated via kmalloc, or by being a clone with 2673 * multiple references to the head. 2674 */ 2675static inline bool skb_head_is_locked(const struct sk_buff *skb) 2676{ 2677 return !skb->head_frag || skb_cloned(skb); 2678} 2679#endif /* __KERNEL__ */ 2680#endif /* _LINUX_SKBUFF_H */