at v3.16 90 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#include <linux/sched.h> 36#include <net/flow_keys.h> 37 38/* A. Checksumming of received packets by device. 39 * 40 * CHECKSUM_NONE: 41 * 42 * Device failed to checksum this packet e.g. due to lack of capabilities. 43 * The packet contains full (though not verified) checksum in packet but 44 * not in skb->csum. Thus, skb->csum is undefined in this case. 45 * 46 * CHECKSUM_UNNECESSARY: 47 * 48 * The hardware you're dealing with doesn't calculate the full checksum 49 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 50 * for specific protocols e.g. TCP/UDP/SCTP, then, for such packets it will 51 * set CHECKSUM_UNNECESSARY if their checksums are okay. skb->csum is still 52 * undefined in this case though. It is a bad option, but, unfortunately, 53 * nowadays most vendors do this. Apparently with the secret goal to sell 54 * you new devices, when you will add new protocol to your host, f.e. IPv6 8) 55 * 56 * CHECKSUM_COMPLETE: 57 * 58 * This is the most generic way. The device supplied checksum of the _whole_ 59 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 60 * hardware doesn't need to parse L3/L4 headers to implement this. 61 * 62 * Note: Even if device supports only some protocols, but is able to produce 63 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 64 * 65 * CHECKSUM_PARTIAL: 66 * 67 * This is identical to the case for output below. This may occur on a packet 68 * received directly from another Linux OS, e.g., a virtualized Linux kernel 69 * on the same host. The packet can be treated in the same way as 70 * CHECKSUM_UNNECESSARY, except that on output (i.e., forwarding) the 71 * checksum must be filled in by the OS or the hardware. 72 * 73 * B. Checksumming on output. 74 * 75 * CHECKSUM_NONE: 76 * 77 * The skb was already checksummed by the protocol, or a checksum is not 78 * required. 79 * 80 * CHECKSUM_PARTIAL: 81 * 82 * The device is required to checksum the packet as seen by hard_start_xmit() 83 * from skb->csum_start up to the end, and to record/write the checksum at 84 * offset skb->csum_start + skb->csum_offset. 85 * 86 * The device must show its capabilities in dev->features, set up at device 87 * setup time, e.g. netdev_features.h: 88 * 89 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything. 90 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over 91 * IPv4. Sigh. Vendors like this way for an unknown reason. 92 * Though, see comment above about CHECKSUM_UNNECESSARY. 8) 93 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead. 94 * NETIF_F_... - Well, you get the picture. 95 * 96 * CHECKSUM_UNNECESSARY: 97 * 98 * Normally, the device will do per protocol specific checksumming. Protocol 99 * implementations that do not want the NIC to perform the checksum 100 * calculation should use this flag in their outgoing skbs. 101 * 102 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC 103 * offload. Correspondingly, the FCoE protocol driver 104 * stack should use CHECKSUM_UNNECESSARY. 105 * 106 * Any questions? No questions, good. --ANK 107 */ 108 109/* Don't change this without changing skb_csum_unnecessary! */ 110#define CHECKSUM_NONE 0 111#define CHECKSUM_UNNECESSARY 1 112#define CHECKSUM_COMPLETE 2 113#define CHECKSUM_PARTIAL 3 114 115#define SKB_DATA_ALIGN(X) (((X) + (SMP_CACHE_BYTES - 1)) & \ 116 ~(SMP_CACHE_BYTES - 1)) 117#define SKB_WITH_OVERHEAD(X) \ 118 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 119#define SKB_MAX_ORDER(X, ORDER) \ 120 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 121#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 122#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 123 124/* return minimum truesize of one skb containing X bytes of data */ 125#define SKB_TRUESIZE(X) ((X) + \ 126 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 127 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 128 129struct net_device; 130struct scatterlist; 131struct pipe_inode_info; 132 133#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 134struct nf_conntrack { 135 atomic_t use; 136}; 137#endif 138 139#ifdef CONFIG_BRIDGE_NETFILTER 140struct nf_bridge_info { 141 atomic_t use; 142 unsigned int mask; 143 struct net_device *physindev; 144 struct net_device *physoutdev; 145 unsigned long data[32 / sizeof(unsigned long)]; 146}; 147#endif 148 149struct sk_buff_head { 150 /* These two members must be first. */ 151 struct sk_buff *next; 152 struct sk_buff *prev; 153 154 __u32 qlen; 155 spinlock_t lock; 156}; 157 158struct sk_buff; 159 160/* To allow 64K frame to be packed as single skb without frag_list we 161 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 162 * buffers which do not start on a page boundary. 163 * 164 * Since GRO uses frags we allocate at least 16 regardless of page 165 * size. 166 */ 167#if (65536/PAGE_SIZE + 1) < 16 168#define MAX_SKB_FRAGS 16UL 169#else 170#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 171#endif 172 173typedef struct skb_frag_struct skb_frag_t; 174 175struct skb_frag_struct { 176 struct { 177 struct page *p; 178 } page; 179#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 180 __u32 page_offset; 181 __u32 size; 182#else 183 __u16 page_offset; 184 __u16 size; 185#endif 186}; 187 188static inline unsigned int skb_frag_size(const skb_frag_t *frag) 189{ 190 return frag->size; 191} 192 193static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 194{ 195 frag->size = size; 196} 197 198static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 199{ 200 frag->size += delta; 201} 202 203static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 204{ 205 frag->size -= delta; 206} 207 208#define HAVE_HW_TIME_STAMP 209 210/** 211 * struct skb_shared_hwtstamps - hardware time stamps 212 * @hwtstamp: hardware time stamp transformed into duration 213 * since arbitrary point in time 214 * @syststamp: hwtstamp transformed to system time base 215 * 216 * Software time stamps generated by ktime_get_real() are stored in 217 * skb->tstamp. The relation between the different kinds of time 218 * stamps is as follows: 219 * 220 * syststamp and tstamp can be compared against each other in 221 * arbitrary combinations. The accuracy of a 222 * syststamp/tstamp/"syststamp from other device" comparison is 223 * limited by the accuracy of the transformation into system time 224 * base. This depends on the device driver and its underlying 225 * hardware. 226 * 227 * hwtstamps can only be compared against other hwtstamps from 228 * the same device. 229 * 230 * This structure is attached to packets as part of the 231 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 232 */ 233struct skb_shared_hwtstamps { 234 ktime_t hwtstamp; 235 ktime_t syststamp; 236}; 237 238/* Definitions for tx_flags in struct skb_shared_info */ 239enum { 240 /* generate hardware time stamp */ 241 SKBTX_HW_TSTAMP = 1 << 0, 242 243 /* generate software time stamp */ 244 SKBTX_SW_TSTAMP = 1 << 1, 245 246 /* device driver is going to provide hardware time stamp */ 247 SKBTX_IN_PROGRESS = 1 << 2, 248 249 /* device driver supports TX zero-copy buffers */ 250 SKBTX_DEV_ZEROCOPY = 1 << 3, 251 252 /* generate wifi status information (where possible) */ 253 SKBTX_WIFI_STATUS = 1 << 4, 254 255 /* This indicates at least one fragment might be overwritten 256 * (as in vmsplice(), sendfile() ...) 257 * If we need to compute a TX checksum, we'll need to copy 258 * all frags to avoid possible bad checksum 259 */ 260 SKBTX_SHARED_FRAG = 1 << 5, 261}; 262 263/* 264 * The callback notifies userspace to release buffers when skb DMA is done in 265 * lower device, the skb last reference should be 0 when calling this. 266 * The zerocopy_success argument is true if zero copy transmit occurred, 267 * false on data copy or out of memory error caused by data copy attempt. 268 * The ctx field is used to track device context. 269 * The desc field is used to track userspace buffer index. 270 */ 271struct ubuf_info { 272 void (*callback)(struct ubuf_info *, bool zerocopy_success); 273 void *ctx; 274 unsigned long desc; 275}; 276 277/* This data is invariant across clones and lives at 278 * the end of the header data, ie. at skb->end. 279 */ 280struct skb_shared_info { 281 unsigned char nr_frags; 282 __u8 tx_flags; 283 unsigned short gso_size; 284 /* Warning: this field is not always filled in (UFO)! */ 285 unsigned short gso_segs; 286 unsigned short gso_type; 287 struct sk_buff *frag_list; 288 struct skb_shared_hwtstamps hwtstamps; 289 __be32 ip6_frag_id; 290 291 /* 292 * Warning : all fields before dataref are cleared in __alloc_skb() 293 */ 294 atomic_t dataref; 295 296 /* Intermediate layers must ensure that destructor_arg 297 * remains valid until skb destructor */ 298 void * destructor_arg; 299 300 /* must be last field, see pskb_expand_head() */ 301 skb_frag_t frags[MAX_SKB_FRAGS]; 302}; 303 304/* We divide dataref into two halves. The higher 16 bits hold references 305 * to the payload part of skb->data. The lower 16 bits hold references to 306 * the entire skb->data. A clone of a headerless skb holds the length of 307 * the header in skb->hdr_len. 308 * 309 * All users must obey the rule that the skb->data reference count must be 310 * greater than or equal to the payload reference count. 311 * 312 * Holding a reference to the payload part means that the user does not 313 * care about modifications to the header part of skb->data. 314 */ 315#define SKB_DATAREF_SHIFT 16 316#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 317 318 319enum { 320 SKB_FCLONE_UNAVAILABLE, 321 SKB_FCLONE_ORIG, 322 SKB_FCLONE_CLONE, 323}; 324 325enum { 326 SKB_GSO_TCPV4 = 1 << 0, 327 SKB_GSO_UDP = 1 << 1, 328 329 /* This indicates the skb is from an untrusted source. */ 330 SKB_GSO_DODGY = 1 << 2, 331 332 /* This indicates the tcp segment has CWR set. */ 333 SKB_GSO_TCP_ECN = 1 << 3, 334 335 SKB_GSO_TCPV6 = 1 << 4, 336 337 SKB_GSO_FCOE = 1 << 5, 338 339 SKB_GSO_GRE = 1 << 6, 340 341 SKB_GSO_GRE_CSUM = 1 << 7, 342 343 SKB_GSO_IPIP = 1 << 8, 344 345 SKB_GSO_SIT = 1 << 9, 346 347 SKB_GSO_UDP_TUNNEL = 1 << 10, 348 349 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 350 351 SKB_GSO_MPLS = 1 << 12, 352 353}; 354 355#if BITS_PER_LONG > 32 356#define NET_SKBUFF_DATA_USES_OFFSET 1 357#endif 358 359#ifdef NET_SKBUFF_DATA_USES_OFFSET 360typedef unsigned int sk_buff_data_t; 361#else 362typedef unsigned char *sk_buff_data_t; 363#endif 364 365/** 366 * struct skb_mstamp - multi resolution time stamps 367 * @stamp_us: timestamp in us resolution 368 * @stamp_jiffies: timestamp in jiffies 369 */ 370struct skb_mstamp { 371 union { 372 u64 v64; 373 struct { 374 u32 stamp_us; 375 u32 stamp_jiffies; 376 }; 377 }; 378}; 379 380/** 381 * skb_mstamp_get - get current timestamp 382 * @cl: place to store timestamps 383 */ 384static inline void skb_mstamp_get(struct skb_mstamp *cl) 385{ 386 u64 val = local_clock(); 387 388 do_div(val, NSEC_PER_USEC); 389 cl->stamp_us = (u32)val; 390 cl->stamp_jiffies = (u32)jiffies; 391} 392 393/** 394 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp 395 * @t1: pointer to newest sample 396 * @t0: pointer to oldest sample 397 */ 398static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1, 399 const struct skb_mstamp *t0) 400{ 401 s32 delta_us = t1->stamp_us - t0->stamp_us; 402 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies; 403 404 /* If delta_us is negative, this might be because interval is too big, 405 * or local_clock() drift is too big : fallback using jiffies. 406 */ 407 if (delta_us <= 0 || 408 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ))) 409 410 delta_us = jiffies_to_usecs(delta_jiffies); 411 412 return delta_us; 413} 414 415 416/** 417 * struct sk_buff - socket buffer 418 * @next: Next buffer in list 419 * @prev: Previous buffer in list 420 * @tstamp: Time we arrived/left 421 * @sk: Socket we are owned by 422 * @dev: Device we arrived on/are leaving by 423 * @cb: Control buffer. Free for use by every layer. Put private vars here 424 * @_skb_refdst: destination entry (with norefcount bit) 425 * @sp: the security path, used for xfrm 426 * @len: Length of actual data 427 * @data_len: Data length 428 * @mac_len: Length of link layer header 429 * @hdr_len: writable header length of cloned skb 430 * @csum: Checksum (must include start/offset pair) 431 * @csum_start: Offset from skb->head where checksumming should start 432 * @csum_offset: Offset from csum_start where checksum should be stored 433 * @priority: Packet queueing priority 434 * @ignore_df: allow local fragmentation 435 * @cloned: Head may be cloned (check refcnt to be sure) 436 * @ip_summed: Driver fed us an IP checksum 437 * @nohdr: Payload reference only, must not modify header 438 * @nfctinfo: Relationship of this skb to the connection 439 * @pkt_type: Packet class 440 * @fclone: skbuff clone status 441 * @ipvs_property: skbuff is owned by ipvs 442 * @peeked: this packet has been seen already, so stats have been 443 * done for it, don't do them again 444 * @nf_trace: netfilter packet trace flag 445 * @protocol: Packet protocol from driver 446 * @destructor: Destruct function 447 * @nfct: Associated connection, if any 448 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 449 * @skb_iif: ifindex of device we arrived on 450 * @tc_index: Traffic control index 451 * @tc_verd: traffic control verdict 452 * @hash: the packet hash 453 * @queue_mapping: Queue mapping for multiqueue devices 454 * @ndisc_nodetype: router type (from link layer) 455 * @ooo_okay: allow the mapping of a socket to a queue to be changed 456 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 457 * ports. 458 * @wifi_acked_valid: wifi_acked was set 459 * @wifi_acked: whether frame was acked on wifi or not 460 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 461 * @dma_cookie: a cookie to one of several possible DMA operations 462 * done by skb DMA functions 463 * @napi_id: id of the NAPI struct this skb came from 464 * @secmark: security marking 465 * @mark: Generic packet mark 466 * @dropcount: total number of sk_receive_queue overflows 467 * @vlan_proto: vlan encapsulation protocol 468 * @vlan_tci: vlan tag control information 469 * @inner_protocol: Protocol (encapsulation) 470 * @inner_transport_header: Inner transport layer header (encapsulation) 471 * @inner_network_header: Network layer header (encapsulation) 472 * @inner_mac_header: Link layer header (encapsulation) 473 * @transport_header: Transport layer header 474 * @network_header: Network layer header 475 * @mac_header: Link layer header 476 * @tail: Tail pointer 477 * @end: End pointer 478 * @head: Head of buffer 479 * @data: Data head pointer 480 * @truesize: Buffer size 481 * @users: User count - see {datagram,tcp}.c 482 */ 483 484struct sk_buff { 485 /* These two members must be first. */ 486 struct sk_buff *next; 487 struct sk_buff *prev; 488 489 union { 490 ktime_t tstamp; 491 struct skb_mstamp skb_mstamp; 492 }; 493 494 struct sock *sk; 495 struct net_device *dev; 496 497 /* 498 * This is the control buffer. It is free to use for every 499 * layer. Please put your private variables there. If you 500 * want to keep them across layers you have to do a skb_clone() 501 * first. This is owned by whoever has the skb queued ATM. 502 */ 503 char cb[48] __aligned(8); 504 505 unsigned long _skb_refdst; 506#ifdef CONFIG_XFRM 507 struct sec_path *sp; 508#endif 509 unsigned int len, 510 data_len; 511 __u16 mac_len, 512 hdr_len; 513 union { 514 __wsum csum; 515 struct { 516 __u16 csum_start; 517 __u16 csum_offset; 518 }; 519 }; 520 __u32 priority; 521 kmemcheck_bitfield_begin(flags1); 522 __u8 ignore_df:1, 523 cloned:1, 524 ip_summed:2, 525 nohdr:1, 526 nfctinfo:3; 527 __u8 pkt_type:3, 528 fclone:2, 529 ipvs_property:1, 530 peeked:1, 531 nf_trace:1; 532 kmemcheck_bitfield_end(flags1); 533 __be16 protocol; 534 535 void (*destructor)(struct sk_buff *skb); 536#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 537 struct nf_conntrack *nfct; 538#endif 539#ifdef CONFIG_BRIDGE_NETFILTER 540 struct nf_bridge_info *nf_bridge; 541#endif 542 543 int skb_iif; 544 545 __u32 hash; 546 547 __be16 vlan_proto; 548 __u16 vlan_tci; 549 550#ifdef CONFIG_NET_SCHED 551 __u16 tc_index; /* traffic control index */ 552#ifdef CONFIG_NET_CLS_ACT 553 __u16 tc_verd; /* traffic control verdict */ 554#endif 555#endif 556 557 __u16 queue_mapping; 558 kmemcheck_bitfield_begin(flags2); 559#ifdef CONFIG_IPV6_NDISC_NODETYPE 560 __u8 ndisc_nodetype:2; 561#endif 562 __u8 pfmemalloc:1; 563 __u8 ooo_okay:1; 564 __u8 l4_hash:1; 565 __u8 wifi_acked_valid:1; 566 __u8 wifi_acked:1; 567 __u8 no_fcs:1; 568 __u8 head_frag:1; 569 /* Encapsulation protocol and NIC drivers should use 570 * this flag to indicate to each other if the skb contains 571 * encapsulated packet or not and maybe use the inner packet 572 * headers if needed 573 */ 574 __u8 encapsulation:1; 575 __u8 encap_hdr_csum:1; 576 __u8 csum_valid:1; 577 __u8 csum_complete_sw:1; 578 /* 3/5 bit hole (depending on ndisc_nodetype presence) */ 579 kmemcheck_bitfield_end(flags2); 580 581#if defined CONFIG_NET_DMA || defined CONFIG_NET_RX_BUSY_POLL 582 union { 583 unsigned int napi_id; 584 dma_cookie_t dma_cookie; 585 }; 586#endif 587#ifdef CONFIG_NETWORK_SECMARK 588 __u32 secmark; 589#endif 590 union { 591 __u32 mark; 592 __u32 dropcount; 593 __u32 reserved_tailroom; 594 }; 595 596 __be16 inner_protocol; 597 __u16 inner_transport_header; 598 __u16 inner_network_header; 599 __u16 inner_mac_header; 600 __u16 transport_header; 601 __u16 network_header; 602 __u16 mac_header; 603 /* These elements must be at the end, see alloc_skb() for details. */ 604 sk_buff_data_t tail; 605 sk_buff_data_t end; 606 unsigned char *head, 607 *data; 608 unsigned int truesize; 609 atomic_t users; 610}; 611 612#ifdef __KERNEL__ 613/* 614 * Handling routines are only of interest to the kernel 615 */ 616#include <linux/slab.h> 617 618 619#define SKB_ALLOC_FCLONE 0x01 620#define SKB_ALLOC_RX 0x02 621 622/* Returns true if the skb was allocated from PFMEMALLOC reserves */ 623static inline bool skb_pfmemalloc(const struct sk_buff *skb) 624{ 625 return unlikely(skb->pfmemalloc); 626} 627 628/* 629 * skb might have a dst pointer attached, refcounted or not. 630 * _skb_refdst low order bit is set if refcount was _not_ taken 631 */ 632#define SKB_DST_NOREF 1UL 633#define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 634 635/** 636 * skb_dst - returns skb dst_entry 637 * @skb: buffer 638 * 639 * Returns skb dst_entry, regardless of reference taken or not. 640 */ 641static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 642{ 643 /* If refdst was not refcounted, check we still are in a 644 * rcu_read_lock section 645 */ 646 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 647 !rcu_read_lock_held() && 648 !rcu_read_lock_bh_held()); 649 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 650} 651 652/** 653 * skb_dst_set - sets skb dst 654 * @skb: buffer 655 * @dst: dst entry 656 * 657 * Sets skb dst, assuming a reference was taken on dst and should 658 * be released by skb_dst_drop() 659 */ 660static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 661{ 662 skb->_skb_refdst = (unsigned long)dst; 663} 664 665void __skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst, 666 bool force); 667 668/** 669 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 670 * @skb: buffer 671 * @dst: dst entry 672 * 673 * Sets skb dst, assuming a reference was not taken on dst. 674 * If dst entry is cached, we do not take reference and dst_release 675 * will be avoided by refdst_drop. If dst entry is not cached, we take 676 * reference, so that last dst_release can destroy the dst immediately. 677 */ 678static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 679{ 680 __skb_dst_set_noref(skb, dst, false); 681} 682 683/** 684 * skb_dst_set_noref_force - sets skb dst, without taking reference 685 * @skb: buffer 686 * @dst: dst entry 687 * 688 * Sets skb dst, assuming a reference was not taken on dst. 689 * No reference is taken and no dst_release will be called. While for 690 * cached dsts deferred reclaim is a basic feature, for entries that are 691 * not cached it is caller's job to guarantee that last dst_release for 692 * provided dst happens when nobody uses it, eg. after a RCU grace period. 693 */ 694static inline void skb_dst_set_noref_force(struct sk_buff *skb, 695 struct dst_entry *dst) 696{ 697 __skb_dst_set_noref(skb, dst, true); 698} 699 700/** 701 * skb_dst_is_noref - Test if skb dst isn't refcounted 702 * @skb: buffer 703 */ 704static inline bool skb_dst_is_noref(const struct sk_buff *skb) 705{ 706 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 707} 708 709static inline struct rtable *skb_rtable(const struct sk_buff *skb) 710{ 711 return (struct rtable *)skb_dst(skb); 712} 713 714void kfree_skb(struct sk_buff *skb); 715void kfree_skb_list(struct sk_buff *segs); 716void skb_tx_error(struct sk_buff *skb); 717void consume_skb(struct sk_buff *skb); 718void __kfree_skb(struct sk_buff *skb); 719extern struct kmem_cache *skbuff_head_cache; 720 721void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 722bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 723 bool *fragstolen, int *delta_truesize); 724 725struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 726 int node); 727struct sk_buff *build_skb(void *data, unsigned int frag_size); 728static inline struct sk_buff *alloc_skb(unsigned int size, 729 gfp_t priority) 730{ 731 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 732} 733 734static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 735 gfp_t priority) 736{ 737 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 738} 739 740struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 741static inline struct sk_buff *alloc_skb_head(gfp_t priority) 742{ 743 return __alloc_skb_head(priority, -1); 744} 745 746struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 747int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 748struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 749struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 750struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 751 gfp_t gfp_mask, bool fclone); 752static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 753 gfp_t gfp_mask) 754{ 755 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 756} 757 758int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 759struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 760 unsigned int headroom); 761struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 762 int newtailroom, gfp_t priority); 763int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 764 int offset, int len); 765int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, 766 int len); 767int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 768int skb_pad(struct sk_buff *skb, int pad); 769#define dev_kfree_skb(a) consume_skb(a) 770 771int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 772 int getfrag(void *from, char *to, int offset, 773 int len, int odd, struct sk_buff *skb), 774 void *from, int length); 775 776struct skb_seq_state { 777 __u32 lower_offset; 778 __u32 upper_offset; 779 __u32 frag_idx; 780 __u32 stepped_offset; 781 struct sk_buff *root_skb; 782 struct sk_buff *cur_skb; 783 __u8 *frag_data; 784}; 785 786void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 787 unsigned int to, struct skb_seq_state *st); 788unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 789 struct skb_seq_state *st); 790void skb_abort_seq_read(struct skb_seq_state *st); 791 792unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 793 unsigned int to, struct ts_config *config, 794 struct ts_state *state); 795 796/* 797 * Packet hash types specify the type of hash in skb_set_hash. 798 * 799 * Hash types refer to the protocol layer addresses which are used to 800 * construct a packet's hash. The hashes are used to differentiate or identify 801 * flows of the protocol layer for the hash type. Hash types are either 802 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 803 * 804 * Properties of hashes: 805 * 806 * 1) Two packets in different flows have different hash values 807 * 2) Two packets in the same flow should have the same hash value 808 * 809 * A hash at a higher layer is considered to be more specific. A driver should 810 * set the most specific hash possible. 811 * 812 * A driver cannot indicate a more specific hash than the layer at which a hash 813 * was computed. For instance an L3 hash cannot be set as an L4 hash. 814 * 815 * A driver may indicate a hash level which is less specific than the 816 * actual layer the hash was computed on. For instance, a hash computed 817 * at L4 may be considered an L3 hash. This should only be done if the 818 * driver can't unambiguously determine that the HW computed the hash at 819 * the higher layer. Note that the "should" in the second property above 820 * permits this. 821 */ 822enum pkt_hash_types { 823 PKT_HASH_TYPE_NONE, /* Undefined type */ 824 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 825 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 826 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 827}; 828 829static inline void 830skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 831{ 832 skb->l4_hash = (type == PKT_HASH_TYPE_L4); 833 skb->hash = hash; 834} 835 836void __skb_get_hash(struct sk_buff *skb); 837static inline __u32 skb_get_hash(struct sk_buff *skb) 838{ 839 if (!skb->l4_hash) 840 __skb_get_hash(skb); 841 842 return skb->hash; 843} 844 845static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 846{ 847 return skb->hash; 848} 849 850static inline void skb_clear_hash(struct sk_buff *skb) 851{ 852 skb->hash = 0; 853 skb->l4_hash = 0; 854} 855 856static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 857{ 858 if (!skb->l4_hash) 859 skb_clear_hash(skb); 860} 861 862static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 863{ 864 to->hash = from->hash; 865 to->l4_hash = from->l4_hash; 866}; 867 868#ifdef NET_SKBUFF_DATA_USES_OFFSET 869static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 870{ 871 return skb->head + skb->end; 872} 873 874static inline unsigned int skb_end_offset(const struct sk_buff *skb) 875{ 876 return skb->end; 877} 878#else 879static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 880{ 881 return skb->end; 882} 883 884static inline unsigned int skb_end_offset(const struct sk_buff *skb) 885{ 886 return skb->end - skb->head; 887} 888#endif 889 890/* Internal */ 891#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 892 893static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 894{ 895 return &skb_shinfo(skb)->hwtstamps; 896} 897 898/** 899 * skb_queue_empty - check if a queue is empty 900 * @list: queue head 901 * 902 * Returns true if the queue is empty, false otherwise. 903 */ 904static inline int skb_queue_empty(const struct sk_buff_head *list) 905{ 906 return list->next == (const struct sk_buff *) list; 907} 908 909/** 910 * skb_queue_is_last - check if skb is the last entry in the queue 911 * @list: queue head 912 * @skb: buffer 913 * 914 * Returns true if @skb is the last buffer on the list. 915 */ 916static inline bool skb_queue_is_last(const struct sk_buff_head *list, 917 const struct sk_buff *skb) 918{ 919 return skb->next == (const struct sk_buff *) list; 920} 921 922/** 923 * skb_queue_is_first - check if skb is the first entry in the queue 924 * @list: queue head 925 * @skb: buffer 926 * 927 * Returns true if @skb is the first buffer on the list. 928 */ 929static inline bool skb_queue_is_first(const struct sk_buff_head *list, 930 const struct sk_buff *skb) 931{ 932 return skb->prev == (const struct sk_buff *) list; 933} 934 935/** 936 * skb_queue_next - return the next packet in the queue 937 * @list: queue head 938 * @skb: current buffer 939 * 940 * Return the next packet in @list after @skb. It is only valid to 941 * call this if skb_queue_is_last() evaluates to false. 942 */ 943static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 944 const struct sk_buff *skb) 945{ 946 /* This BUG_ON may seem severe, but if we just return then we 947 * are going to dereference garbage. 948 */ 949 BUG_ON(skb_queue_is_last(list, skb)); 950 return skb->next; 951} 952 953/** 954 * skb_queue_prev - return the prev packet in the queue 955 * @list: queue head 956 * @skb: current buffer 957 * 958 * Return the prev packet in @list before @skb. It is only valid to 959 * call this if skb_queue_is_first() evaluates to false. 960 */ 961static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 962 const struct sk_buff *skb) 963{ 964 /* This BUG_ON may seem severe, but if we just return then we 965 * are going to dereference garbage. 966 */ 967 BUG_ON(skb_queue_is_first(list, skb)); 968 return skb->prev; 969} 970 971/** 972 * skb_get - reference buffer 973 * @skb: buffer to reference 974 * 975 * Makes another reference to a socket buffer and returns a pointer 976 * to the buffer. 977 */ 978static inline struct sk_buff *skb_get(struct sk_buff *skb) 979{ 980 atomic_inc(&skb->users); 981 return skb; 982} 983 984/* 985 * If users == 1, we are the only owner and are can avoid redundant 986 * atomic change. 987 */ 988 989/** 990 * skb_cloned - is the buffer a clone 991 * @skb: buffer to check 992 * 993 * Returns true if the buffer was generated with skb_clone() and is 994 * one of multiple shared copies of the buffer. Cloned buffers are 995 * shared data so must not be written to under normal circumstances. 996 */ 997static inline int skb_cloned(const struct sk_buff *skb) 998{ 999 return skb->cloned && 1000 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1001} 1002 1003static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1004{ 1005 might_sleep_if(pri & __GFP_WAIT); 1006 1007 if (skb_cloned(skb)) 1008 return pskb_expand_head(skb, 0, 0, pri); 1009 1010 return 0; 1011} 1012 1013/** 1014 * skb_header_cloned - is the header a clone 1015 * @skb: buffer to check 1016 * 1017 * Returns true if modifying the header part of the buffer requires 1018 * the data to be copied. 1019 */ 1020static inline int skb_header_cloned(const struct sk_buff *skb) 1021{ 1022 int dataref; 1023 1024 if (!skb->cloned) 1025 return 0; 1026 1027 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1028 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1029 return dataref != 1; 1030} 1031 1032/** 1033 * skb_header_release - release reference to header 1034 * @skb: buffer to operate on 1035 * 1036 * Drop a reference to the header part of the buffer. This is done 1037 * by acquiring a payload reference. You must not read from the header 1038 * part of skb->data after this. 1039 */ 1040static inline void skb_header_release(struct sk_buff *skb) 1041{ 1042 BUG_ON(skb->nohdr); 1043 skb->nohdr = 1; 1044 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1045} 1046 1047/** 1048 * skb_shared - is the buffer shared 1049 * @skb: buffer to check 1050 * 1051 * Returns true if more than one person has a reference to this 1052 * buffer. 1053 */ 1054static inline int skb_shared(const struct sk_buff *skb) 1055{ 1056 return atomic_read(&skb->users) != 1; 1057} 1058 1059/** 1060 * skb_share_check - check if buffer is shared and if so clone it 1061 * @skb: buffer to check 1062 * @pri: priority for memory allocation 1063 * 1064 * If the buffer is shared the buffer is cloned and the old copy 1065 * drops a reference. A new clone with a single reference is returned. 1066 * If the buffer is not shared the original buffer is returned. When 1067 * being called from interrupt status or with spinlocks held pri must 1068 * be GFP_ATOMIC. 1069 * 1070 * NULL is returned on a memory allocation failure. 1071 */ 1072static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1073{ 1074 might_sleep_if(pri & __GFP_WAIT); 1075 if (skb_shared(skb)) { 1076 struct sk_buff *nskb = skb_clone(skb, pri); 1077 1078 if (likely(nskb)) 1079 consume_skb(skb); 1080 else 1081 kfree_skb(skb); 1082 skb = nskb; 1083 } 1084 return skb; 1085} 1086 1087/* 1088 * Copy shared buffers into a new sk_buff. We effectively do COW on 1089 * packets to handle cases where we have a local reader and forward 1090 * and a couple of other messy ones. The normal one is tcpdumping 1091 * a packet thats being forwarded. 1092 */ 1093 1094/** 1095 * skb_unshare - make a copy of a shared buffer 1096 * @skb: buffer to check 1097 * @pri: priority for memory allocation 1098 * 1099 * If the socket buffer is a clone then this function creates a new 1100 * copy of the data, drops a reference count on the old copy and returns 1101 * the new copy with the reference count at 1. If the buffer is not a clone 1102 * the original buffer is returned. When called with a spinlock held or 1103 * from interrupt state @pri must be %GFP_ATOMIC 1104 * 1105 * %NULL is returned on a memory allocation failure. 1106 */ 1107static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1108 gfp_t pri) 1109{ 1110 might_sleep_if(pri & __GFP_WAIT); 1111 if (skb_cloned(skb)) { 1112 struct sk_buff *nskb = skb_copy(skb, pri); 1113 kfree_skb(skb); /* Free our shared copy */ 1114 skb = nskb; 1115 } 1116 return skb; 1117} 1118 1119/** 1120 * skb_peek - peek at the head of an &sk_buff_head 1121 * @list_: list to peek at 1122 * 1123 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1124 * be careful with this one. A peek leaves the buffer on the 1125 * list and someone else may run off with it. You must hold 1126 * the appropriate locks or have a private queue to do this. 1127 * 1128 * Returns %NULL for an empty list or a pointer to the head element. 1129 * The reference count is not incremented and the reference is therefore 1130 * volatile. Use with caution. 1131 */ 1132static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1133{ 1134 struct sk_buff *skb = list_->next; 1135 1136 if (skb == (struct sk_buff *)list_) 1137 skb = NULL; 1138 return skb; 1139} 1140 1141/** 1142 * skb_peek_next - peek skb following the given one from a queue 1143 * @skb: skb to start from 1144 * @list_: list to peek at 1145 * 1146 * Returns %NULL when the end of the list is met or a pointer to the 1147 * next element. The reference count is not incremented and the 1148 * reference is therefore volatile. Use with caution. 1149 */ 1150static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1151 const struct sk_buff_head *list_) 1152{ 1153 struct sk_buff *next = skb->next; 1154 1155 if (next == (struct sk_buff *)list_) 1156 next = NULL; 1157 return next; 1158} 1159 1160/** 1161 * skb_peek_tail - peek at the tail of an &sk_buff_head 1162 * @list_: list to peek at 1163 * 1164 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1165 * be careful with this one. A peek leaves the buffer on the 1166 * list and someone else may run off with it. You must hold 1167 * the appropriate locks or have a private queue to do this. 1168 * 1169 * Returns %NULL for an empty list or a pointer to the tail element. 1170 * The reference count is not incremented and the reference is therefore 1171 * volatile. Use with caution. 1172 */ 1173static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1174{ 1175 struct sk_buff *skb = list_->prev; 1176 1177 if (skb == (struct sk_buff *)list_) 1178 skb = NULL; 1179 return skb; 1180 1181} 1182 1183/** 1184 * skb_queue_len - get queue length 1185 * @list_: list to measure 1186 * 1187 * Return the length of an &sk_buff queue. 1188 */ 1189static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1190{ 1191 return list_->qlen; 1192} 1193 1194/** 1195 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1196 * @list: queue to initialize 1197 * 1198 * This initializes only the list and queue length aspects of 1199 * an sk_buff_head object. This allows to initialize the list 1200 * aspects of an sk_buff_head without reinitializing things like 1201 * the spinlock. It can also be used for on-stack sk_buff_head 1202 * objects where the spinlock is known to not be used. 1203 */ 1204static inline void __skb_queue_head_init(struct sk_buff_head *list) 1205{ 1206 list->prev = list->next = (struct sk_buff *)list; 1207 list->qlen = 0; 1208} 1209 1210/* 1211 * This function creates a split out lock class for each invocation; 1212 * this is needed for now since a whole lot of users of the skb-queue 1213 * infrastructure in drivers have different locking usage (in hardirq) 1214 * than the networking core (in softirq only). In the long run either the 1215 * network layer or drivers should need annotation to consolidate the 1216 * main types of usage into 3 classes. 1217 */ 1218static inline void skb_queue_head_init(struct sk_buff_head *list) 1219{ 1220 spin_lock_init(&list->lock); 1221 __skb_queue_head_init(list); 1222} 1223 1224static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1225 struct lock_class_key *class) 1226{ 1227 skb_queue_head_init(list); 1228 lockdep_set_class(&list->lock, class); 1229} 1230 1231/* 1232 * Insert an sk_buff on a list. 1233 * 1234 * The "__skb_xxxx()" functions are the non-atomic ones that 1235 * can only be called with interrupts disabled. 1236 */ 1237void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1238 struct sk_buff_head *list); 1239static inline void __skb_insert(struct sk_buff *newsk, 1240 struct sk_buff *prev, struct sk_buff *next, 1241 struct sk_buff_head *list) 1242{ 1243 newsk->next = next; 1244 newsk->prev = prev; 1245 next->prev = prev->next = newsk; 1246 list->qlen++; 1247} 1248 1249static inline void __skb_queue_splice(const struct sk_buff_head *list, 1250 struct sk_buff *prev, 1251 struct sk_buff *next) 1252{ 1253 struct sk_buff *first = list->next; 1254 struct sk_buff *last = list->prev; 1255 1256 first->prev = prev; 1257 prev->next = first; 1258 1259 last->next = next; 1260 next->prev = last; 1261} 1262 1263/** 1264 * skb_queue_splice - join two skb lists, this is designed for stacks 1265 * @list: the new list to add 1266 * @head: the place to add it in the first list 1267 */ 1268static inline void skb_queue_splice(const struct sk_buff_head *list, 1269 struct sk_buff_head *head) 1270{ 1271 if (!skb_queue_empty(list)) { 1272 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1273 head->qlen += list->qlen; 1274 } 1275} 1276 1277/** 1278 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1279 * @list: the new list to add 1280 * @head: the place to add it in the first list 1281 * 1282 * The list at @list is reinitialised 1283 */ 1284static inline void skb_queue_splice_init(struct sk_buff_head *list, 1285 struct sk_buff_head *head) 1286{ 1287 if (!skb_queue_empty(list)) { 1288 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1289 head->qlen += list->qlen; 1290 __skb_queue_head_init(list); 1291 } 1292} 1293 1294/** 1295 * skb_queue_splice_tail - join two skb lists, each list being a queue 1296 * @list: the new list to add 1297 * @head: the place to add it in the first list 1298 */ 1299static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1300 struct sk_buff_head *head) 1301{ 1302 if (!skb_queue_empty(list)) { 1303 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1304 head->qlen += list->qlen; 1305 } 1306} 1307 1308/** 1309 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1310 * @list: the new list to add 1311 * @head: the place to add it in the first list 1312 * 1313 * Each of the lists is a queue. 1314 * The list at @list is reinitialised 1315 */ 1316static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1317 struct sk_buff_head *head) 1318{ 1319 if (!skb_queue_empty(list)) { 1320 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1321 head->qlen += list->qlen; 1322 __skb_queue_head_init(list); 1323 } 1324} 1325 1326/** 1327 * __skb_queue_after - queue a buffer at the list head 1328 * @list: list to use 1329 * @prev: place after this buffer 1330 * @newsk: buffer to queue 1331 * 1332 * Queue a buffer int the middle of a list. This function takes no locks 1333 * and you must therefore hold required locks before calling it. 1334 * 1335 * A buffer cannot be placed on two lists at the same time. 1336 */ 1337static inline void __skb_queue_after(struct sk_buff_head *list, 1338 struct sk_buff *prev, 1339 struct sk_buff *newsk) 1340{ 1341 __skb_insert(newsk, prev, prev->next, list); 1342} 1343 1344void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1345 struct sk_buff_head *list); 1346 1347static inline void __skb_queue_before(struct sk_buff_head *list, 1348 struct sk_buff *next, 1349 struct sk_buff *newsk) 1350{ 1351 __skb_insert(newsk, next->prev, next, list); 1352} 1353 1354/** 1355 * __skb_queue_head - queue a buffer at the list head 1356 * @list: list to use 1357 * @newsk: buffer to queue 1358 * 1359 * Queue a buffer at the start of a list. This function takes no locks 1360 * and you must therefore hold required locks before calling it. 1361 * 1362 * A buffer cannot be placed on two lists at the same time. 1363 */ 1364void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1365static inline void __skb_queue_head(struct sk_buff_head *list, 1366 struct sk_buff *newsk) 1367{ 1368 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1369} 1370 1371/** 1372 * __skb_queue_tail - queue a buffer at the list tail 1373 * @list: list to use 1374 * @newsk: buffer to queue 1375 * 1376 * Queue a buffer at the end of a list. This function takes no locks 1377 * and you must therefore hold required locks before calling it. 1378 * 1379 * A buffer cannot be placed on two lists at the same time. 1380 */ 1381void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1382static inline void __skb_queue_tail(struct sk_buff_head *list, 1383 struct sk_buff *newsk) 1384{ 1385 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1386} 1387 1388/* 1389 * remove sk_buff from list. _Must_ be called atomically, and with 1390 * the list known.. 1391 */ 1392void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1393static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1394{ 1395 struct sk_buff *next, *prev; 1396 1397 list->qlen--; 1398 next = skb->next; 1399 prev = skb->prev; 1400 skb->next = skb->prev = NULL; 1401 next->prev = prev; 1402 prev->next = next; 1403} 1404 1405/** 1406 * __skb_dequeue - remove from the head of the queue 1407 * @list: list to dequeue from 1408 * 1409 * Remove the head of the list. This function does not take any locks 1410 * so must be used with appropriate locks held only. The head item is 1411 * returned or %NULL if the list is empty. 1412 */ 1413struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1414static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1415{ 1416 struct sk_buff *skb = skb_peek(list); 1417 if (skb) 1418 __skb_unlink(skb, list); 1419 return skb; 1420} 1421 1422/** 1423 * __skb_dequeue_tail - remove from the tail of the queue 1424 * @list: list to dequeue from 1425 * 1426 * Remove the tail of the list. This function does not take any locks 1427 * so must be used with appropriate locks held only. The tail item is 1428 * returned or %NULL if the list is empty. 1429 */ 1430struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1431static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1432{ 1433 struct sk_buff *skb = skb_peek_tail(list); 1434 if (skb) 1435 __skb_unlink(skb, list); 1436 return skb; 1437} 1438 1439 1440static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1441{ 1442 return skb->data_len; 1443} 1444 1445static inline unsigned int skb_headlen(const struct sk_buff *skb) 1446{ 1447 return skb->len - skb->data_len; 1448} 1449 1450static inline int skb_pagelen(const struct sk_buff *skb) 1451{ 1452 int i, len = 0; 1453 1454 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1455 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1456 return len + skb_headlen(skb); 1457} 1458 1459/** 1460 * __skb_fill_page_desc - initialise a paged fragment in an skb 1461 * @skb: buffer containing fragment to be initialised 1462 * @i: paged fragment index to initialise 1463 * @page: the page to use for this fragment 1464 * @off: the offset to the data with @page 1465 * @size: the length of the data 1466 * 1467 * Initialises the @i'th fragment of @skb to point to &size bytes at 1468 * offset @off within @page. 1469 * 1470 * Does not take any additional reference on the fragment. 1471 */ 1472static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1473 struct page *page, int off, int size) 1474{ 1475 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1476 1477 /* 1478 * Propagate page->pfmemalloc to the skb if we can. The problem is 1479 * that not all callers have unique ownership of the page. If 1480 * pfmemalloc is set, we check the mapping as a mapping implies 1481 * page->index is set (index and pfmemalloc share space). 1482 * If it's a valid mapping, we cannot use page->pfmemalloc but we 1483 * do not lose pfmemalloc information as the pages would not be 1484 * allocated using __GFP_MEMALLOC. 1485 */ 1486 frag->page.p = page; 1487 frag->page_offset = off; 1488 skb_frag_size_set(frag, size); 1489 1490 page = compound_head(page); 1491 if (page->pfmemalloc && !page->mapping) 1492 skb->pfmemalloc = true; 1493} 1494 1495/** 1496 * skb_fill_page_desc - initialise a paged fragment in an skb 1497 * @skb: buffer containing fragment to be initialised 1498 * @i: paged fragment index to initialise 1499 * @page: the page to use for this fragment 1500 * @off: the offset to the data with @page 1501 * @size: the length of the data 1502 * 1503 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1504 * @skb to point to @size bytes at offset @off within @page. In 1505 * addition updates @skb such that @i is the last fragment. 1506 * 1507 * Does not take any additional reference on the fragment. 1508 */ 1509static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1510 struct page *page, int off, int size) 1511{ 1512 __skb_fill_page_desc(skb, i, page, off, size); 1513 skb_shinfo(skb)->nr_frags = i + 1; 1514} 1515 1516void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1517 int size, unsigned int truesize); 1518 1519void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1520 unsigned int truesize); 1521 1522#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1523#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1524#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1525 1526#ifdef NET_SKBUFF_DATA_USES_OFFSET 1527static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1528{ 1529 return skb->head + skb->tail; 1530} 1531 1532static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1533{ 1534 skb->tail = skb->data - skb->head; 1535} 1536 1537static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1538{ 1539 skb_reset_tail_pointer(skb); 1540 skb->tail += offset; 1541} 1542 1543#else /* NET_SKBUFF_DATA_USES_OFFSET */ 1544static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1545{ 1546 return skb->tail; 1547} 1548 1549static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1550{ 1551 skb->tail = skb->data; 1552} 1553 1554static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1555{ 1556 skb->tail = skb->data + offset; 1557} 1558 1559#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1560 1561/* 1562 * Add data to an sk_buff 1563 */ 1564unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1565unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1566static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1567{ 1568 unsigned char *tmp = skb_tail_pointer(skb); 1569 SKB_LINEAR_ASSERT(skb); 1570 skb->tail += len; 1571 skb->len += len; 1572 return tmp; 1573} 1574 1575unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1576static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1577{ 1578 skb->data -= len; 1579 skb->len += len; 1580 return skb->data; 1581} 1582 1583unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1584static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1585{ 1586 skb->len -= len; 1587 BUG_ON(skb->len < skb->data_len); 1588 return skb->data += len; 1589} 1590 1591static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1592{ 1593 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1594} 1595 1596unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1597 1598static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1599{ 1600 if (len > skb_headlen(skb) && 1601 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1602 return NULL; 1603 skb->len -= len; 1604 return skb->data += len; 1605} 1606 1607static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1608{ 1609 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1610} 1611 1612static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1613{ 1614 if (likely(len <= skb_headlen(skb))) 1615 return 1; 1616 if (unlikely(len > skb->len)) 1617 return 0; 1618 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1619} 1620 1621/** 1622 * skb_headroom - bytes at buffer head 1623 * @skb: buffer to check 1624 * 1625 * Return the number of bytes of free space at the head of an &sk_buff. 1626 */ 1627static inline unsigned int skb_headroom(const struct sk_buff *skb) 1628{ 1629 return skb->data - skb->head; 1630} 1631 1632/** 1633 * skb_tailroom - bytes at buffer end 1634 * @skb: buffer to check 1635 * 1636 * Return the number of bytes of free space at the tail of an sk_buff 1637 */ 1638static inline int skb_tailroom(const struct sk_buff *skb) 1639{ 1640 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1641} 1642 1643/** 1644 * skb_availroom - bytes at buffer end 1645 * @skb: buffer to check 1646 * 1647 * Return the number of bytes of free space at the tail of an sk_buff 1648 * allocated by sk_stream_alloc() 1649 */ 1650static inline int skb_availroom(const struct sk_buff *skb) 1651{ 1652 if (skb_is_nonlinear(skb)) 1653 return 0; 1654 1655 return skb->end - skb->tail - skb->reserved_tailroom; 1656} 1657 1658/** 1659 * skb_reserve - adjust headroom 1660 * @skb: buffer to alter 1661 * @len: bytes to move 1662 * 1663 * Increase the headroom of an empty &sk_buff by reducing the tail 1664 * room. This is only allowed for an empty buffer. 1665 */ 1666static inline void skb_reserve(struct sk_buff *skb, int len) 1667{ 1668 skb->data += len; 1669 skb->tail += len; 1670} 1671 1672static inline void skb_reset_inner_headers(struct sk_buff *skb) 1673{ 1674 skb->inner_mac_header = skb->mac_header; 1675 skb->inner_network_header = skb->network_header; 1676 skb->inner_transport_header = skb->transport_header; 1677} 1678 1679static inline void skb_reset_mac_len(struct sk_buff *skb) 1680{ 1681 skb->mac_len = skb->network_header - skb->mac_header; 1682} 1683 1684static inline unsigned char *skb_inner_transport_header(const struct sk_buff 1685 *skb) 1686{ 1687 return skb->head + skb->inner_transport_header; 1688} 1689 1690static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 1691{ 1692 skb->inner_transport_header = skb->data - skb->head; 1693} 1694 1695static inline void skb_set_inner_transport_header(struct sk_buff *skb, 1696 const int offset) 1697{ 1698 skb_reset_inner_transport_header(skb); 1699 skb->inner_transport_header += offset; 1700} 1701 1702static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 1703{ 1704 return skb->head + skb->inner_network_header; 1705} 1706 1707static inline void skb_reset_inner_network_header(struct sk_buff *skb) 1708{ 1709 skb->inner_network_header = skb->data - skb->head; 1710} 1711 1712static inline void skb_set_inner_network_header(struct sk_buff *skb, 1713 const int offset) 1714{ 1715 skb_reset_inner_network_header(skb); 1716 skb->inner_network_header += offset; 1717} 1718 1719static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 1720{ 1721 return skb->head + skb->inner_mac_header; 1722} 1723 1724static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 1725{ 1726 skb->inner_mac_header = skb->data - skb->head; 1727} 1728 1729static inline void skb_set_inner_mac_header(struct sk_buff *skb, 1730 const int offset) 1731{ 1732 skb_reset_inner_mac_header(skb); 1733 skb->inner_mac_header += offset; 1734} 1735static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 1736{ 1737 return skb->transport_header != (typeof(skb->transport_header))~0U; 1738} 1739 1740static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 1741{ 1742 return skb->head + skb->transport_header; 1743} 1744 1745static inline void skb_reset_transport_header(struct sk_buff *skb) 1746{ 1747 skb->transport_header = skb->data - skb->head; 1748} 1749 1750static inline void skb_set_transport_header(struct sk_buff *skb, 1751 const int offset) 1752{ 1753 skb_reset_transport_header(skb); 1754 skb->transport_header += offset; 1755} 1756 1757static inline unsigned char *skb_network_header(const struct sk_buff *skb) 1758{ 1759 return skb->head + skb->network_header; 1760} 1761 1762static inline void skb_reset_network_header(struct sk_buff *skb) 1763{ 1764 skb->network_header = skb->data - skb->head; 1765} 1766 1767static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 1768{ 1769 skb_reset_network_header(skb); 1770 skb->network_header += offset; 1771} 1772 1773static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 1774{ 1775 return skb->head + skb->mac_header; 1776} 1777 1778static inline int skb_mac_header_was_set(const struct sk_buff *skb) 1779{ 1780 return skb->mac_header != (typeof(skb->mac_header))~0U; 1781} 1782 1783static inline void skb_reset_mac_header(struct sk_buff *skb) 1784{ 1785 skb->mac_header = skb->data - skb->head; 1786} 1787 1788static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 1789{ 1790 skb_reset_mac_header(skb); 1791 skb->mac_header += offset; 1792} 1793 1794static inline void skb_pop_mac_header(struct sk_buff *skb) 1795{ 1796 skb->mac_header = skb->network_header; 1797} 1798 1799static inline void skb_probe_transport_header(struct sk_buff *skb, 1800 const int offset_hint) 1801{ 1802 struct flow_keys keys; 1803 1804 if (skb_transport_header_was_set(skb)) 1805 return; 1806 else if (skb_flow_dissect(skb, &keys)) 1807 skb_set_transport_header(skb, keys.thoff); 1808 else 1809 skb_set_transport_header(skb, offset_hint); 1810} 1811 1812static inline void skb_mac_header_rebuild(struct sk_buff *skb) 1813{ 1814 if (skb_mac_header_was_set(skb)) { 1815 const unsigned char *old_mac = skb_mac_header(skb); 1816 1817 skb_set_mac_header(skb, -skb->mac_len); 1818 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 1819 } 1820} 1821 1822static inline int skb_checksum_start_offset(const struct sk_buff *skb) 1823{ 1824 return skb->csum_start - skb_headroom(skb); 1825} 1826 1827static inline int skb_transport_offset(const struct sk_buff *skb) 1828{ 1829 return skb_transport_header(skb) - skb->data; 1830} 1831 1832static inline u32 skb_network_header_len(const struct sk_buff *skb) 1833{ 1834 return skb->transport_header - skb->network_header; 1835} 1836 1837static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 1838{ 1839 return skb->inner_transport_header - skb->inner_network_header; 1840} 1841 1842static inline int skb_network_offset(const struct sk_buff *skb) 1843{ 1844 return skb_network_header(skb) - skb->data; 1845} 1846 1847static inline int skb_inner_network_offset(const struct sk_buff *skb) 1848{ 1849 return skb_inner_network_header(skb) - skb->data; 1850} 1851 1852static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 1853{ 1854 return pskb_may_pull(skb, skb_network_offset(skb) + len); 1855} 1856 1857static inline void skb_pop_rcv_encapsulation(struct sk_buff *skb) 1858{ 1859 /* Only continue with checksum unnecessary if device indicated 1860 * it is valid across encapsulation (skb->encapsulation was set). 1861 */ 1862 if (skb->ip_summed == CHECKSUM_UNNECESSARY && !skb->encapsulation) 1863 skb->ip_summed = CHECKSUM_NONE; 1864 1865 skb->encapsulation = 0; 1866 skb->csum_valid = 0; 1867} 1868 1869/* 1870 * CPUs often take a performance hit when accessing unaligned memory 1871 * locations. The actual performance hit varies, it can be small if the 1872 * hardware handles it or large if we have to take an exception and fix it 1873 * in software. 1874 * 1875 * Since an ethernet header is 14 bytes network drivers often end up with 1876 * the IP header at an unaligned offset. The IP header can be aligned by 1877 * shifting the start of the packet by 2 bytes. Drivers should do this 1878 * with: 1879 * 1880 * skb_reserve(skb, NET_IP_ALIGN); 1881 * 1882 * The downside to this alignment of the IP header is that the DMA is now 1883 * unaligned. On some architectures the cost of an unaligned DMA is high 1884 * and this cost outweighs the gains made by aligning the IP header. 1885 * 1886 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 1887 * to be overridden. 1888 */ 1889#ifndef NET_IP_ALIGN 1890#define NET_IP_ALIGN 2 1891#endif 1892 1893/* 1894 * The networking layer reserves some headroom in skb data (via 1895 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 1896 * the header has to grow. In the default case, if the header has to grow 1897 * 32 bytes or less we avoid the reallocation. 1898 * 1899 * Unfortunately this headroom changes the DMA alignment of the resulting 1900 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 1901 * on some architectures. An architecture can override this value, 1902 * perhaps setting it to a cacheline in size (since that will maintain 1903 * cacheline alignment of the DMA). It must be a power of 2. 1904 * 1905 * Various parts of the networking layer expect at least 32 bytes of 1906 * headroom, you should not reduce this. 1907 * 1908 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 1909 * to reduce average number of cache lines per packet. 1910 * get_rps_cpus() for example only access one 64 bytes aligned block : 1911 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 1912 */ 1913#ifndef NET_SKB_PAD 1914#define NET_SKB_PAD max(32, L1_CACHE_BYTES) 1915#endif 1916 1917int ___pskb_trim(struct sk_buff *skb, unsigned int len); 1918 1919static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 1920{ 1921 if (unlikely(skb_is_nonlinear(skb))) { 1922 WARN_ON(1); 1923 return; 1924 } 1925 skb->len = len; 1926 skb_set_tail_pointer(skb, len); 1927} 1928 1929void skb_trim(struct sk_buff *skb, unsigned int len); 1930 1931static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 1932{ 1933 if (skb->data_len) 1934 return ___pskb_trim(skb, len); 1935 __skb_trim(skb, len); 1936 return 0; 1937} 1938 1939static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 1940{ 1941 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 1942} 1943 1944/** 1945 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 1946 * @skb: buffer to alter 1947 * @len: new length 1948 * 1949 * This is identical to pskb_trim except that the caller knows that 1950 * the skb is not cloned so we should never get an error due to out- 1951 * of-memory. 1952 */ 1953static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 1954{ 1955 int err = pskb_trim(skb, len); 1956 BUG_ON(err); 1957} 1958 1959/** 1960 * skb_orphan - orphan a buffer 1961 * @skb: buffer to orphan 1962 * 1963 * If a buffer currently has an owner then we call the owner's 1964 * destructor function and make the @skb unowned. The buffer continues 1965 * to exist but is no longer charged to its former owner. 1966 */ 1967static inline void skb_orphan(struct sk_buff *skb) 1968{ 1969 if (skb->destructor) { 1970 skb->destructor(skb); 1971 skb->destructor = NULL; 1972 skb->sk = NULL; 1973 } else { 1974 BUG_ON(skb->sk); 1975 } 1976} 1977 1978/** 1979 * skb_orphan_frags - orphan the frags contained in a buffer 1980 * @skb: buffer to orphan frags from 1981 * @gfp_mask: allocation mask for replacement pages 1982 * 1983 * For each frag in the SKB which needs a destructor (i.e. has an 1984 * owner) create a copy of that frag and release the original 1985 * page by calling the destructor. 1986 */ 1987static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 1988{ 1989 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 1990 return 0; 1991 return skb_copy_ubufs(skb, gfp_mask); 1992} 1993 1994/** 1995 * __skb_queue_purge - empty a list 1996 * @list: list to empty 1997 * 1998 * Delete all buffers on an &sk_buff list. Each buffer is removed from 1999 * the list and one reference dropped. This function does not take the 2000 * list lock and the caller must hold the relevant locks to use it. 2001 */ 2002void skb_queue_purge(struct sk_buff_head *list); 2003static inline void __skb_queue_purge(struct sk_buff_head *list) 2004{ 2005 struct sk_buff *skb; 2006 while ((skb = __skb_dequeue(list)) != NULL) 2007 kfree_skb(skb); 2008} 2009 2010#define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768) 2011#define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER) 2012#define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE 2013 2014void *netdev_alloc_frag(unsigned int fragsz); 2015 2016struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2017 gfp_t gfp_mask); 2018 2019/** 2020 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2021 * @dev: network device to receive on 2022 * @length: length to allocate 2023 * 2024 * Allocate a new &sk_buff and assign it a usage count of one. The 2025 * buffer has unspecified headroom built in. Users should allocate 2026 * the headroom they think they need without accounting for the 2027 * built in space. The built in space is used for optimisations. 2028 * 2029 * %NULL is returned if there is no free memory. Although this function 2030 * allocates memory it can be called from an interrupt. 2031 */ 2032static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2033 unsigned int length) 2034{ 2035 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2036} 2037 2038/* legacy helper around __netdev_alloc_skb() */ 2039static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2040 gfp_t gfp_mask) 2041{ 2042 return __netdev_alloc_skb(NULL, length, gfp_mask); 2043} 2044 2045/* legacy helper around netdev_alloc_skb() */ 2046static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2047{ 2048 return netdev_alloc_skb(NULL, length); 2049} 2050 2051 2052static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2053 unsigned int length, gfp_t gfp) 2054{ 2055 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2056 2057 if (NET_IP_ALIGN && skb) 2058 skb_reserve(skb, NET_IP_ALIGN); 2059 return skb; 2060} 2061 2062static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2063 unsigned int length) 2064{ 2065 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2066} 2067 2068/** 2069 * __skb_alloc_pages - allocate pages for ps-rx on a skb and preserve pfmemalloc data 2070 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX 2071 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used 2072 * @order: size of the allocation 2073 * 2074 * Allocate a new page. 2075 * 2076 * %NULL is returned if there is no free memory. 2077*/ 2078static inline struct page *__skb_alloc_pages(gfp_t gfp_mask, 2079 struct sk_buff *skb, 2080 unsigned int order) 2081{ 2082 struct page *page; 2083 2084 gfp_mask |= __GFP_COLD; 2085 2086 if (!(gfp_mask & __GFP_NOMEMALLOC)) 2087 gfp_mask |= __GFP_MEMALLOC; 2088 2089 page = alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2090 if (skb && page && page->pfmemalloc) 2091 skb->pfmemalloc = true; 2092 2093 return page; 2094} 2095 2096/** 2097 * __skb_alloc_page - allocate a page for ps-rx for a given skb and preserve pfmemalloc data 2098 * @gfp_mask: alloc_pages_node mask. Set __GFP_NOMEMALLOC if not for network packet RX 2099 * @skb: skb to set pfmemalloc on if __GFP_MEMALLOC is used 2100 * 2101 * Allocate a new page. 2102 * 2103 * %NULL is returned if there is no free memory. 2104 */ 2105static inline struct page *__skb_alloc_page(gfp_t gfp_mask, 2106 struct sk_buff *skb) 2107{ 2108 return __skb_alloc_pages(gfp_mask, skb, 0); 2109} 2110 2111/** 2112 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2113 * @page: The page that was allocated from skb_alloc_page 2114 * @skb: The skb that may need pfmemalloc set 2115 */ 2116static inline void skb_propagate_pfmemalloc(struct page *page, 2117 struct sk_buff *skb) 2118{ 2119 if (page && page->pfmemalloc) 2120 skb->pfmemalloc = true; 2121} 2122 2123/** 2124 * skb_frag_page - retrieve the page referred to by a paged fragment 2125 * @frag: the paged fragment 2126 * 2127 * Returns the &struct page associated with @frag. 2128 */ 2129static inline struct page *skb_frag_page(const skb_frag_t *frag) 2130{ 2131 return frag->page.p; 2132} 2133 2134/** 2135 * __skb_frag_ref - take an addition reference on a paged fragment. 2136 * @frag: the paged fragment 2137 * 2138 * Takes an additional reference on the paged fragment @frag. 2139 */ 2140static inline void __skb_frag_ref(skb_frag_t *frag) 2141{ 2142 get_page(skb_frag_page(frag)); 2143} 2144 2145/** 2146 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2147 * @skb: the buffer 2148 * @f: the fragment offset. 2149 * 2150 * Takes an additional reference on the @f'th paged fragment of @skb. 2151 */ 2152static inline void skb_frag_ref(struct sk_buff *skb, int f) 2153{ 2154 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2155} 2156 2157/** 2158 * __skb_frag_unref - release a reference on a paged fragment. 2159 * @frag: the paged fragment 2160 * 2161 * Releases a reference on the paged fragment @frag. 2162 */ 2163static inline void __skb_frag_unref(skb_frag_t *frag) 2164{ 2165 put_page(skb_frag_page(frag)); 2166} 2167 2168/** 2169 * skb_frag_unref - release a reference on a paged fragment of an skb. 2170 * @skb: the buffer 2171 * @f: the fragment offset 2172 * 2173 * Releases a reference on the @f'th paged fragment of @skb. 2174 */ 2175static inline void skb_frag_unref(struct sk_buff *skb, int f) 2176{ 2177 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2178} 2179 2180/** 2181 * skb_frag_address - gets the address of the data contained in a paged fragment 2182 * @frag: the paged fragment buffer 2183 * 2184 * Returns the address of the data within @frag. The page must already 2185 * be mapped. 2186 */ 2187static inline void *skb_frag_address(const skb_frag_t *frag) 2188{ 2189 return page_address(skb_frag_page(frag)) + frag->page_offset; 2190} 2191 2192/** 2193 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2194 * @frag: the paged fragment buffer 2195 * 2196 * Returns the address of the data within @frag. Checks that the page 2197 * is mapped and returns %NULL otherwise. 2198 */ 2199static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2200{ 2201 void *ptr = page_address(skb_frag_page(frag)); 2202 if (unlikely(!ptr)) 2203 return NULL; 2204 2205 return ptr + frag->page_offset; 2206} 2207 2208/** 2209 * __skb_frag_set_page - sets the page contained in a paged fragment 2210 * @frag: the paged fragment 2211 * @page: the page to set 2212 * 2213 * Sets the fragment @frag to contain @page. 2214 */ 2215static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2216{ 2217 frag->page.p = page; 2218} 2219 2220/** 2221 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2222 * @skb: the buffer 2223 * @f: the fragment offset 2224 * @page: the page to set 2225 * 2226 * Sets the @f'th fragment of @skb to contain @page. 2227 */ 2228static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2229 struct page *page) 2230{ 2231 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2232} 2233 2234bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2235 2236/** 2237 * skb_frag_dma_map - maps a paged fragment via the DMA API 2238 * @dev: the device to map the fragment to 2239 * @frag: the paged fragment to map 2240 * @offset: the offset within the fragment (starting at the 2241 * fragment's own offset) 2242 * @size: the number of bytes to map 2243 * @dir: the direction of the mapping (%PCI_DMA_*) 2244 * 2245 * Maps the page associated with @frag to @device. 2246 */ 2247static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2248 const skb_frag_t *frag, 2249 size_t offset, size_t size, 2250 enum dma_data_direction dir) 2251{ 2252 return dma_map_page(dev, skb_frag_page(frag), 2253 frag->page_offset + offset, size, dir); 2254} 2255 2256static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2257 gfp_t gfp_mask) 2258{ 2259 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2260} 2261 2262 2263static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2264 gfp_t gfp_mask) 2265{ 2266 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2267} 2268 2269 2270/** 2271 * skb_clone_writable - is the header of a clone writable 2272 * @skb: buffer to check 2273 * @len: length up to which to write 2274 * 2275 * Returns true if modifying the header part of the cloned buffer 2276 * does not requires the data to be copied. 2277 */ 2278static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2279{ 2280 return !skb_header_cloned(skb) && 2281 skb_headroom(skb) + len <= skb->hdr_len; 2282} 2283 2284static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2285 int cloned) 2286{ 2287 int delta = 0; 2288 2289 if (headroom > skb_headroom(skb)) 2290 delta = headroom - skb_headroom(skb); 2291 2292 if (delta || cloned) 2293 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2294 GFP_ATOMIC); 2295 return 0; 2296} 2297 2298/** 2299 * skb_cow - copy header of skb when it is required 2300 * @skb: buffer to cow 2301 * @headroom: needed headroom 2302 * 2303 * If the skb passed lacks sufficient headroom or its data part 2304 * is shared, data is reallocated. If reallocation fails, an error 2305 * is returned and original skb is not changed. 2306 * 2307 * The result is skb with writable area skb->head...skb->tail 2308 * and at least @headroom of space at head. 2309 */ 2310static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2311{ 2312 return __skb_cow(skb, headroom, skb_cloned(skb)); 2313} 2314 2315/** 2316 * skb_cow_head - skb_cow but only making the head writable 2317 * @skb: buffer to cow 2318 * @headroom: needed headroom 2319 * 2320 * This function is identical to skb_cow except that we replace the 2321 * skb_cloned check by skb_header_cloned. It should be used when 2322 * you only need to push on some header and do not need to modify 2323 * the data. 2324 */ 2325static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2326{ 2327 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2328} 2329 2330/** 2331 * skb_padto - pad an skbuff up to a minimal size 2332 * @skb: buffer to pad 2333 * @len: minimal length 2334 * 2335 * Pads up a buffer to ensure the trailing bytes exist and are 2336 * blanked. If the buffer already contains sufficient data it 2337 * is untouched. Otherwise it is extended. Returns zero on 2338 * success. The skb is freed on error. 2339 */ 2340 2341static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2342{ 2343 unsigned int size = skb->len; 2344 if (likely(size >= len)) 2345 return 0; 2346 return skb_pad(skb, len - size); 2347} 2348 2349static inline int skb_add_data(struct sk_buff *skb, 2350 char __user *from, int copy) 2351{ 2352 const int off = skb->len; 2353 2354 if (skb->ip_summed == CHECKSUM_NONE) { 2355 int err = 0; 2356 __wsum csum = csum_and_copy_from_user(from, skb_put(skb, copy), 2357 copy, 0, &err); 2358 if (!err) { 2359 skb->csum = csum_block_add(skb->csum, csum, off); 2360 return 0; 2361 } 2362 } else if (!copy_from_user(skb_put(skb, copy), from, copy)) 2363 return 0; 2364 2365 __skb_trim(skb, off); 2366 return -EFAULT; 2367} 2368 2369static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2370 const struct page *page, int off) 2371{ 2372 if (i) { 2373 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2374 2375 return page == skb_frag_page(frag) && 2376 off == frag->page_offset + skb_frag_size(frag); 2377 } 2378 return false; 2379} 2380 2381static inline int __skb_linearize(struct sk_buff *skb) 2382{ 2383 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2384} 2385 2386/** 2387 * skb_linearize - convert paged skb to linear one 2388 * @skb: buffer to linarize 2389 * 2390 * If there is no free memory -ENOMEM is returned, otherwise zero 2391 * is returned and the old skb data released. 2392 */ 2393static inline int skb_linearize(struct sk_buff *skb) 2394{ 2395 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2396} 2397 2398/** 2399 * skb_has_shared_frag - can any frag be overwritten 2400 * @skb: buffer to test 2401 * 2402 * Return true if the skb has at least one frag that might be modified 2403 * by an external entity (as in vmsplice()/sendfile()) 2404 */ 2405static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2406{ 2407 return skb_is_nonlinear(skb) && 2408 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2409} 2410 2411/** 2412 * skb_linearize_cow - make sure skb is linear and writable 2413 * @skb: buffer to process 2414 * 2415 * If there is no free memory -ENOMEM is returned, otherwise zero 2416 * is returned and the old skb data released. 2417 */ 2418static inline int skb_linearize_cow(struct sk_buff *skb) 2419{ 2420 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2421 __skb_linearize(skb) : 0; 2422} 2423 2424/** 2425 * skb_postpull_rcsum - update checksum for received skb after pull 2426 * @skb: buffer to update 2427 * @start: start of data before pull 2428 * @len: length of data pulled 2429 * 2430 * After doing a pull on a received packet, you need to call this to 2431 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2432 * CHECKSUM_NONE so that it can be recomputed from scratch. 2433 */ 2434 2435static inline void skb_postpull_rcsum(struct sk_buff *skb, 2436 const void *start, unsigned int len) 2437{ 2438 if (skb->ip_summed == CHECKSUM_COMPLETE) 2439 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0)); 2440} 2441 2442unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2443 2444/** 2445 * pskb_trim_rcsum - trim received skb and update checksum 2446 * @skb: buffer to trim 2447 * @len: new length 2448 * 2449 * This is exactly the same as pskb_trim except that it ensures the 2450 * checksum of received packets are still valid after the operation. 2451 */ 2452 2453static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2454{ 2455 if (likely(len >= skb->len)) 2456 return 0; 2457 if (skb->ip_summed == CHECKSUM_COMPLETE) 2458 skb->ip_summed = CHECKSUM_NONE; 2459 return __pskb_trim(skb, len); 2460} 2461 2462#define skb_queue_walk(queue, skb) \ 2463 for (skb = (queue)->next; \ 2464 skb != (struct sk_buff *)(queue); \ 2465 skb = skb->next) 2466 2467#define skb_queue_walk_safe(queue, skb, tmp) \ 2468 for (skb = (queue)->next, tmp = skb->next; \ 2469 skb != (struct sk_buff *)(queue); \ 2470 skb = tmp, tmp = skb->next) 2471 2472#define skb_queue_walk_from(queue, skb) \ 2473 for (; skb != (struct sk_buff *)(queue); \ 2474 skb = skb->next) 2475 2476#define skb_queue_walk_from_safe(queue, skb, tmp) \ 2477 for (tmp = skb->next; \ 2478 skb != (struct sk_buff *)(queue); \ 2479 skb = tmp, tmp = skb->next) 2480 2481#define skb_queue_reverse_walk(queue, skb) \ 2482 for (skb = (queue)->prev; \ 2483 skb != (struct sk_buff *)(queue); \ 2484 skb = skb->prev) 2485 2486#define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 2487 for (skb = (queue)->prev, tmp = skb->prev; \ 2488 skb != (struct sk_buff *)(queue); \ 2489 skb = tmp, tmp = skb->prev) 2490 2491#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 2492 for (tmp = skb->prev; \ 2493 skb != (struct sk_buff *)(queue); \ 2494 skb = tmp, tmp = skb->prev) 2495 2496static inline bool skb_has_frag_list(const struct sk_buff *skb) 2497{ 2498 return skb_shinfo(skb)->frag_list != NULL; 2499} 2500 2501static inline void skb_frag_list_init(struct sk_buff *skb) 2502{ 2503 skb_shinfo(skb)->frag_list = NULL; 2504} 2505 2506static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag) 2507{ 2508 frag->next = skb_shinfo(skb)->frag_list; 2509 skb_shinfo(skb)->frag_list = frag; 2510} 2511 2512#define skb_walk_frags(skb, iter) \ 2513 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 2514 2515struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 2516 int *peeked, int *off, int *err); 2517struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 2518 int *err); 2519unsigned int datagram_poll(struct file *file, struct socket *sock, 2520 struct poll_table_struct *wait); 2521int skb_copy_datagram_iovec(const struct sk_buff *from, int offset, 2522 struct iovec *to, int size); 2523int skb_copy_and_csum_datagram_iovec(struct sk_buff *skb, int hlen, 2524 struct iovec *iov); 2525int skb_copy_datagram_from_iovec(struct sk_buff *skb, int offset, 2526 const struct iovec *from, int from_offset, 2527 int len); 2528int zerocopy_sg_from_iovec(struct sk_buff *skb, const struct iovec *frm, 2529 int offset, size_t count); 2530int skb_copy_datagram_const_iovec(const struct sk_buff *from, int offset, 2531 const struct iovec *to, int to_offset, 2532 int size); 2533void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 2534void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb); 2535int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 2536int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 2537int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 2538__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 2539 int len, __wsum csum); 2540int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 2541 struct pipe_inode_info *pipe, unsigned int len, 2542 unsigned int flags); 2543void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 2544unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 2545int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 2546 int len, int hlen); 2547void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 2548int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 2549void skb_scrub_packet(struct sk_buff *skb, bool xnet); 2550unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 2551struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 2552 2553struct skb_checksum_ops { 2554 __wsum (*update)(const void *mem, int len, __wsum wsum); 2555 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 2556}; 2557 2558__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 2559 __wsum csum, const struct skb_checksum_ops *ops); 2560__wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 2561 __wsum csum); 2562 2563static inline void *skb_header_pointer(const struct sk_buff *skb, int offset, 2564 int len, void *buffer) 2565{ 2566 int hlen = skb_headlen(skb); 2567 2568 if (hlen - offset >= len) 2569 return skb->data + offset; 2570 2571 if (skb_copy_bits(skb, offset, buffer, len) < 0) 2572 return NULL; 2573 2574 return buffer; 2575} 2576 2577/** 2578 * skb_needs_linearize - check if we need to linearize a given skb 2579 * depending on the given device features. 2580 * @skb: socket buffer to check 2581 * @features: net device features 2582 * 2583 * Returns true if either: 2584 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 2585 * 2. skb is fragmented and the device does not support SG. 2586 */ 2587static inline bool skb_needs_linearize(struct sk_buff *skb, 2588 netdev_features_t features) 2589{ 2590 return skb_is_nonlinear(skb) && 2591 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 2592 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 2593} 2594 2595static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 2596 void *to, 2597 const unsigned int len) 2598{ 2599 memcpy(to, skb->data, len); 2600} 2601 2602static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 2603 const int offset, void *to, 2604 const unsigned int len) 2605{ 2606 memcpy(to, skb->data + offset, len); 2607} 2608 2609static inline void skb_copy_to_linear_data(struct sk_buff *skb, 2610 const void *from, 2611 const unsigned int len) 2612{ 2613 memcpy(skb->data, from, len); 2614} 2615 2616static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 2617 const int offset, 2618 const void *from, 2619 const unsigned int len) 2620{ 2621 memcpy(skb->data + offset, from, len); 2622} 2623 2624void skb_init(void); 2625 2626static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 2627{ 2628 return skb->tstamp; 2629} 2630 2631/** 2632 * skb_get_timestamp - get timestamp from a skb 2633 * @skb: skb to get stamp from 2634 * @stamp: pointer to struct timeval to store stamp in 2635 * 2636 * Timestamps are stored in the skb as offsets to a base timestamp. 2637 * This function converts the offset back to a struct timeval and stores 2638 * it in stamp. 2639 */ 2640static inline void skb_get_timestamp(const struct sk_buff *skb, 2641 struct timeval *stamp) 2642{ 2643 *stamp = ktime_to_timeval(skb->tstamp); 2644} 2645 2646static inline void skb_get_timestampns(const struct sk_buff *skb, 2647 struct timespec *stamp) 2648{ 2649 *stamp = ktime_to_timespec(skb->tstamp); 2650} 2651 2652static inline void __net_timestamp(struct sk_buff *skb) 2653{ 2654 skb->tstamp = ktime_get_real(); 2655} 2656 2657static inline ktime_t net_timedelta(ktime_t t) 2658{ 2659 return ktime_sub(ktime_get_real(), t); 2660} 2661 2662static inline ktime_t net_invalid_timestamp(void) 2663{ 2664 return ktime_set(0, 0); 2665} 2666 2667#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 2668 2669void skb_clone_tx_timestamp(struct sk_buff *skb); 2670bool skb_defer_rx_timestamp(struct sk_buff *skb); 2671 2672#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 2673 2674static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 2675{ 2676} 2677 2678static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 2679{ 2680 return false; 2681} 2682 2683#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 2684 2685/** 2686 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 2687 * 2688 * PHY drivers may accept clones of transmitted packets for 2689 * timestamping via their phy_driver.txtstamp method. These drivers 2690 * must call this function to return the skb back to the stack, with 2691 * or without a timestamp. 2692 * 2693 * @skb: clone of the the original outgoing packet 2694 * @hwtstamps: hardware time stamps, may be NULL if not available 2695 * 2696 */ 2697void skb_complete_tx_timestamp(struct sk_buff *skb, 2698 struct skb_shared_hwtstamps *hwtstamps); 2699 2700/** 2701 * skb_tstamp_tx - queue clone of skb with send time stamps 2702 * @orig_skb: the original outgoing packet 2703 * @hwtstamps: hardware time stamps, may be NULL if not available 2704 * 2705 * If the skb has a socket associated, then this function clones the 2706 * skb (thus sharing the actual data and optional structures), stores 2707 * the optional hardware time stamping information (if non NULL) or 2708 * generates a software time stamp (otherwise), then queues the clone 2709 * to the error queue of the socket. Errors are silently ignored. 2710 */ 2711void skb_tstamp_tx(struct sk_buff *orig_skb, 2712 struct skb_shared_hwtstamps *hwtstamps); 2713 2714static inline void sw_tx_timestamp(struct sk_buff *skb) 2715{ 2716 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 2717 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 2718 skb_tstamp_tx(skb, NULL); 2719} 2720 2721/** 2722 * skb_tx_timestamp() - Driver hook for transmit timestamping 2723 * 2724 * Ethernet MAC Drivers should call this function in their hard_xmit() 2725 * function immediately before giving the sk_buff to the MAC hardware. 2726 * 2727 * Specifically, one should make absolutely sure that this function is 2728 * called before TX completion of this packet can trigger. Otherwise 2729 * the packet could potentially already be freed. 2730 * 2731 * @skb: A socket buffer. 2732 */ 2733static inline void skb_tx_timestamp(struct sk_buff *skb) 2734{ 2735 skb_clone_tx_timestamp(skb); 2736 sw_tx_timestamp(skb); 2737} 2738 2739/** 2740 * skb_complete_wifi_ack - deliver skb with wifi status 2741 * 2742 * @skb: the original outgoing packet 2743 * @acked: ack status 2744 * 2745 */ 2746void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 2747 2748__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 2749__sum16 __skb_checksum_complete(struct sk_buff *skb); 2750 2751static inline int skb_csum_unnecessary(const struct sk_buff *skb) 2752{ 2753 return ((skb->ip_summed & CHECKSUM_UNNECESSARY) || skb->csum_valid); 2754} 2755 2756/** 2757 * skb_checksum_complete - Calculate checksum of an entire packet 2758 * @skb: packet to process 2759 * 2760 * This function calculates the checksum over the entire packet plus 2761 * the value of skb->csum. The latter can be used to supply the 2762 * checksum of a pseudo header as used by TCP/UDP. It returns the 2763 * checksum. 2764 * 2765 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 2766 * this function can be used to verify that checksum on received 2767 * packets. In that case the function should return zero if the 2768 * checksum is correct. In particular, this function will return zero 2769 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 2770 * hardware has already verified the correctness of the checksum. 2771 */ 2772static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 2773{ 2774 return skb_csum_unnecessary(skb) ? 2775 0 : __skb_checksum_complete(skb); 2776} 2777 2778/* Check if we need to perform checksum complete validation. 2779 * 2780 * Returns true if checksum complete is needed, false otherwise 2781 * (either checksum is unnecessary or zero checksum is allowed). 2782 */ 2783static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 2784 bool zero_okay, 2785 __sum16 check) 2786{ 2787 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 2788 skb->csum_valid = 1; 2789 return false; 2790 } 2791 2792 return true; 2793} 2794 2795/* For small packets <= CHECKSUM_BREAK peform checksum complete directly 2796 * in checksum_init. 2797 */ 2798#define CHECKSUM_BREAK 76 2799 2800/* Validate (init) checksum based on checksum complete. 2801 * 2802 * Return values: 2803 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 2804 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 2805 * checksum is stored in skb->csum for use in __skb_checksum_complete 2806 * non-zero: value of invalid checksum 2807 * 2808 */ 2809static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 2810 bool complete, 2811 __wsum psum) 2812{ 2813 if (skb->ip_summed == CHECKSUM_COMPLETE) { 2814 if (!csum_fold(csum_add(psum, skb->csum))) { 2815 skb->csum_valid = 1; 2816 return 0; 2817 } 2818 } 2819 2820 skb->csum = psum; 2821 2822 if (complete || skb->len <= CHECKSUM_BREAK) { 2823 __sum16 csum; 2824 2825 csum = __skb_checksum_complete(skb); 2826 skb->csum_valid = !csum; 2827 return csum; 2828 } 2829 2830 return 0; 2831} 2832 2833static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 2834{ 2835 return 0; 2836} 2837 2838/* Perform checksum validate (init). Note that this is a macro since we only 2839 * want to calculate the pseudo header which is an input function if necessary. 2840 * First we try to validate without any computation (checksum unnecessary) and 2841 * then calculate based on checksum complete calling the function to compute 2842 * pseudo header. 2843 * 2844 * Return values: 2845 * 0: checksum is validated or try to in skb_checksum_complete 2846 * non-zero: value of invalid checksum 2847 */ 2848#define __skb_checksum_validate(skb, proto, complete, \ 2849 zero_okay, check, compute_pseudo) \ 2850({ \ 2851 __sum16 __ret = 0; \ 2852 skb->csum_valid = 0; \ 2853 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 2854 __ret = __skb_checksum_validate_complete(skb, \ 2855 complete, compute_pseudo(skb, proto)); \ 2856 __ret; \ 2857}) 2858 2859#define skb_checksum_init(skb, proto, compute_pseudo) \ 2860 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 2861 2862#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 2863 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 2864 2865#define skb_checksum_validate(skb, proto, compute_pseudo) \ 2866 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 2867 2868#define skb_checksum_validate_zero_check(skb, proto, check, \ 2869 compute_pseudo) \ 2870 __skb_checksum_validate_(skb, proto, true, true, check, compute_pseudo) 2871 2872#define skb_checksum_simple_validate(skb) \ 2873 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 2874 2875#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 2876void nf_conntrack_destroy(struct nf_conntrack *nfct); 2877static inline void nf_conntrack_put(struct nf_conntrack *nfct) 2878{ 2879 if (nfct && atomic_dec_and_test(&nfct->use)) 2880 nf_conntrack_destroy(nfct); 2881} 2882static inline void nf_conntrack_get(struct nf_conntrack *nfct) 2883{ 2884 if (nfct) 2885 atomic_inc(&nfct->use); 2886} 2887#endif 2888#ifdef CONFIG_BRIDGE_NETFILTER 2889static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 2890{ 2891 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 2892 kfree(nf_bridge); 2893} 2894static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 2895{ 2896 if (nf_bridge) 2897 atomic_inc(&nf_bridge->use); 2898} 2899#endif /* CONFIG_BRIDGE_NETFILTER */ 2900static inline void nf_reset(struct sk_buff *skb) 2901{ 2902#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 2903 nf_conntrack_put(skb->nfct); 2904 skb->nfct = NULL; 2905#endif 2906#ifdef CONFIG_BRIDGE_NETFILTER 2907 nf_bridge_put(skb->nf_bridge); 2908 skb->nf_bridge = NULL; 2909#endif 2910} 2911 2912static inline void nf_reset_trace(struct sk_buff *skb) 2913{ 2914#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 2915 skb->nf_trace = 0; 2916#endif 2917} 2918 2919/* Note: This doesn't put any conntrack and bridge info in dst. */ 2920static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src) 2921{ 2922#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 2923 dst->nfct = src->nfct; 2924 nf_conntrack_get(src->nfct); 2925 dst->nfctinfo = src->nfctinfo; 2926#endif 2927#ifdef CONFIG_BRIDGE_NETFILTER 2928 dst->nf_bridge = src->nf_bridge; 2929 nf_bridge_get(src->nf_bridge); 2930#endif 2931#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 2932 dst->nf_trace = src->nf_trace; 2933#endif 2934} 2935 2936static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 2937{ 2938#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 2939 nf_conntrack_put(dst->nfct); 2940#endif 2941#ifdef CONFIG_BRIDGE_NETFILTER 2942 nf_bridge_put(dst->nf_bridge); 2943#endif 2944 __nf_copy(dst, src); 2945} 2946 2947#ifdef CONFIG_NETWORK_SECMARK 2948static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 2949{ 2950 to->secmark = from->secmark; 2951} 2952 2953static inline void skb_init_secmark(struct sk_buff *skb) 2954{ 2955 skb->secmark = 0; 2956} 2957#else 2958static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 2959{ } 2960 2961static inline void skb_init_secmark(struct sk_buff *skb) 2962{ } 2963#endif 2964 2965static inline bool skb_irq_freeable(const struct sk_buff *skb) 2966{ 2967 return !skb->destructor && 2968#if IS_ENABLED(CONFIG_XFRM) 2969 !skb->sp && 2970#endif 2971#if IS_ENABLED(CONFIG_NF_CONNTRACK) 2972 !skb->nfct && 2973#endif 2974 !skb->_skb_refdst && 2975 !skb_has_frag_list(skb); 2976} 2977 2978static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 2979{ 2980 skb->queue_mapping = queue_mapping; 2981} 2982 2983static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 2984{ 2985 return skb->queue_mapping; 2986} 2987 2988static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 2989{ 2990 to->queue_mapping = from->queue_mapping; 2991} 2992 2993static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 2994{ 2995 skb->queue_mapping = rx_queue + 1; 2996} 2997 2998static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 2999{ 3000 return skb->queue_mapping - 1; 3001} 3002 3003static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3004{ 3005 return skb->queue_mapping != 0; 3006} 3007 3008u16 __skb_tx_hash(const struct net_device *dev, const struct sk_buff *skb, 3009 unsigned int num_tx_queues); 3010 3011static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3012{ 3013#ifdef CONFIG_XFRM 3014 return skb->sp; 3015#else 3016 return NULL; 3017#endif 3018} 3019 3020/* Keeps track of mac header offset relative to skb->head. 3021 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3022 * For non-tunnel skb it points to skb_mac_header() and for 3023 * tunnel skb it points to outer mac header. 3024 * Keeps track of level of encapsulation of network headers. 3025 */ 3026struct skb_gso_cb { 3027 int mac_offset; 3028 int encap_level; 3029 __u16 csum_start; 3030}; 3031#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb) 3032 3033static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3034{ 3035 return (skb_mac_header(inner_skb) - inner_skb->head) - 3036 SKB_GSO_CB(inner_skb)->mac_offset; 3037} 3038 3039static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3040{ 3041 int new_headroom, headroom; 3042 int ret; 3043 3044 headroom = skb_headroom(skb); 3045 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3046 if (ret) 3047 return ret; 3048 3049 new_headroom = skb_headroom(skb); 3050 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3051 return 0; 3052} 3053 3054/* Compute the checksum for a gso segment. First compute the checksum value 3055 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3056 * then add in skb->csum (checksum from csum_start to end of packet). 3057 * skb->csum and csum_start are then updated to reflect the checksum of the 3058 * resultant packet starting from the transport header-- the resultant checksum 3059 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3060 * header. 3061 */ 3062static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3063{ 3064 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) - 3065 skb_transport_offset(skb); 3066 __u16 csum; 3067 3068 csum = csum_fold(csum_partial(skb_transport_header(skb), 3069 plen, skb->csum)); 3070 skb->csum = res; 3071 SKB_GSO_CB(skb)->csum_start -= plen; 3072 3073 return csum; 3074} 3075 3076static inline bool skb_is_gso(const struct sk_buff *skb) 3077{ 3078 return skb_shinfo(skb)->gso_size; 3079} 3080 3081/* Note: Should be called only if skb_is_gso(skb) is true */ 3082static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3083{ 3084 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3085} 3086 3087void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3088 3089static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3090{ 3091 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3092 * wanted then gso_type will be set. */ 3093 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3094 3095 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3096 unlikely(shinfo->gso_type == 0)) { 3097 __skb_warn_lro_forwarding(skb); 3098 return true; 3099 } 3100 return false; 3101} 3102 3103static inline void skb_forward_csum(struct sk_buff *skb) 3104{ 3105 /* Unfortunately we don't support this one. Any brave souls? */ 3106 if (skb->ip_summed == CHECKSUM_COMPLETE) 3107 skb->ip_summed = CHECKSUM_NONE; 3108} 3109 3110/** 3111 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3112 * @skb: skb to check 3113 * 3114 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3115 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3116 * use this helper, to document places where we make this assertion. 3117 */ 3118static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3119{ 3120#ifdef DEBUG 3121 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3122#endif 3123} 3124 3125bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3126 3127int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3128 3129u32 __skb_get_poff(const struct sk_buff *skb); 3130 3131/** 3132 * skb_head_is_locked - Determine if the skb->head is locked down 3133 * @skb: skb to check 3134 * 3135 * The head on skbs build around a head frag can be removed if they are 3136 * not cloned. This function returns true if the skb head is locked down 3137 * due to either being allocated via kmalloc, or by being a clone with 3138 * multiple references to the head. 3139 */ 3140static inline bool skb_head_is_locked(const struct sk_buff *skb) 3141{ 3142 return !skb->head_frag || skb_cloned(skb); 3143} 3144 3145/** 3146 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3147 * 3148 * @skb: GSO skb 3149 * 3150 * skb_gso_network_seglen is used to determine the real size of the 3151 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3152 * 3153 * The MAC/L2 header is not accounted for. 3154 */ 3155static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3156{ 3157 unsigned int hdr_len = skb_transport_header(skb) - 3158 skb_network_header(skb); 3159 return hdr_len + skb_gso_transport_seglen(skb); 3160} 3161#endif /* __KERNEL__ */ 3162#endif /* _LINUX_SKBUFF_H */