at v4.0 99 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#include <linux/rbtree.h> 24#include <linux/socket.h> 25 26#include <linux/atomic.h> 27#include <asm/types.h> 28#include <linux/spinlock.h> 29#include <linux/net.h> 30#include <linux/textsearch.h> 31#include <net/checksum.h> 32#include <linux/rcupdate.h> 33#include <linux/hrtimer.h> 34#include <linux/dma-mapping.h> 35#include <linux/netdev_features.h> 36#include <linux/sched.h> 37#include <net/flow_keys.h> 38 39/* A. Checksumming of received packets by device. 40 * 41 * CHECKSUM_NONE: 42 * 43 * Device failed to checksum this packet e.g. due to lack of capabilities. 44 * The packet contains full (though not verified) checksum in packet but 45 * not in skb->csum. Thus, skb->csum is undefined in this case. 46 * 47 * CHECKSUM_UNNECESSARY: 48 * 49 * The hardware you're dealing with doesn't calculate the full checksum 50 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 51 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 52 * if their checksums are okay. skb->csum is still undefined in this case 53 * though. It is a bad option, but, unfortunately, nowadays most vendors do 54 * this. Apparently with the secret goal to sell you new devices, when you 55 * will add new protocol to your host, f.e. IPv6 8) 56 * 57 * CHECKSUM_UNNECESSARY is applicable to following protocols: 58 * TCP: IPv6 and IPv4. 59 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 60 * zero UDP checksum for either IPv4 or IPv6, the networking stack 61 * may perform further validation in this case. 62 * GRE: only if the checksum is present in the header. 63 * SCTP: indicates the CRC in SCTP header has been validated. 64 * 65 * skb->csum_level indicates the number of consecutive checksums found in 66 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 67 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 68 * and a device is able to verify the checksums for UDP (possibly zero), 69 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 70 * two. If the device were only able to verify the UDP checksum and not 71 * GRE, either because it doesn't support GRE checksum of because GRE 72 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 73 * not considered in this case). 74 * 75 * CHECKSUM_COMPLETE: 76 * 77 * This is the most generic way. The device supplied checksum of the _whole_ 78 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 79 * hardware doesn't need to parse L3/L4 headers to implement this. 80 * 81 * Note: Even if device supports only some protocols, but is able to produce 82 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 83 * 84 * CHECKSUM_PARTIAL: 85 * 86 * A checksum is set up to be offloaded to a device as described in the 87 * output description for CHECKSUM_PARTIAL. This may occur on a packet 88 * received directly from another Linux OS, e.g., a virtualized Linux kernel 89 * on the same host, or it may be set in the input path in GRO or remote 90 * checksum offload. For the purposes of checksum verification, the checksum 91 * referred to by skb->csum_start + skb->csum_offset and any preceding 92 * checksums in the packet are considered verified. Any checksums in the 93 * packet that are after the checksum being offloaded are not considered to 94 * be verified. 95 * 96 * B. Checksumming on output. 97 * 98 * CHECKSUM_NONE: 99 * 100 * The skb was already checksummed by the protocol, or a checksum is not 101 * required. 102 * 103 * CHECKSUM_PARTIAL: 104 * 105 * The device is required to checksum the packet as seen by hard_start_xmit() 106 * from skb->csum_start up to the end, and to record/write the checksum at 107 * offset skb->csum_start + skb->csum_offset. 108 * 109 * The device must show its capabilities in dev->features, set up at device 110 * setup time, e.g. netdev_features.h: 111 * 112 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything. 113 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over 114 * IPv4. Sigh. Vendors like this way for an unknown reason. 115 * Though, see comment above about CHECKSUM_UNNECESSARY. 8) 116 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead. 117 * NETIF_F_... - Well, you get the picture. 118 * 119 * CHECKSUM_UNNECESSARY: 120 * 121 * Normally, the device will do per protocol specific checksumming. Protocol 122 * implementations that do not want the NIC to perform the checksum 123 * calculation should use this flag in their outgoing skbs. 124 * 125 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC 126 * offload. Correspondingly, the FCoE protocol driver 127 * stack should use CHECKSUM_UNNECESSARY. 128 * 129 * Any questions? No questions, good. --ANK 130 */ 131 132/* Don't change this without changing skb_csum_unnecessary! */ 133#define CHECKSUM_NONE 0 134#define CHECKSUM_UNNECESSARY 1 135#define CHECKSUM_COMPLETE 2 136#define CHECKSUM_PARTIAL 3 137 138/* Maximum value in skb->csum_level */ 139#define SKB_MAX_CSUM_LEVEL 3 140 141#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 142#define SKB_WITH_OVERHEAD(X) \ 143 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 144#define SKB_MAX_ORDER(X, ORDER) \ 145 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 146#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 147#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 148 149/* return minimum truesize of one skb containing X bytes of data */ 150#define SKB_TRUESIZE(X) ((X) + \ 151 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 152 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 153 154struct net_device; 155struct scatterlist; 156struct pipe_inode_info; 157struct iov_iter; 158struct napi_struct; 159 160#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 161struct nf_conntrack { 162 atomic_t use; 163}; 164#endif 165 166#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 167struct nf_bridge_info { 168 atomic_t use; 169 unsigned int mask; 170 struct net_device *physindev; 171 struct net_device *physoutdev; 172 unsigned long data[32 / sizeof(unsigned long)]; 173}; 174#endif 175 176struct sk_buff_head { 177 /* These two members must be first. */ 178 struct sk_buff *next; 179 struct sk_buff *prev; 180 181 __u32 qlen; 182 spinlock_t lock; 183}; 184 185struct sk_buff; 186 187/* To allow 64K frame to be packed as single skb without frag_list we 188 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 189 * buffers which do not start on a page boundary. 190 * 191 * Since GRO uses frags we allocate at least 16 regardless of page 192 * size. 193 */ 194#if (65536/PAGE_SIZE + 1) < 16 195#define MAX_SKB_FRAGS 16UL 196#else 197#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 198#endif 199 200typedef struct skb_frag_struct skb_frag_t; 201 202struct skb_frag_struct { 203 struct { 204 struct page *p; 205 } page; 206#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 207 __u32 page_offset; 208 __u32 size; 209#else 210 __u16 page_offset; 211 __u16 size; 212#endif 213}; 214 215static inline unsigned int skb_frag_size(const skb_frag_t *frag) 216{ 217 return frag->size; 218} 219 220static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 221{ 222 frag->size = size; 223} 224 225static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 226{ 227 frag->size += delta; 228} 229 230static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 231{ 232 frag->size -= delta; 233} 234 235#define HAVE_HW_TIME_STAMP 236 237/** 238 * struct skb_shared_hwtstamps - hardware time stamps 239 * @hwtstamp: hardware time stamp transformed into duration 240 * since arbitrary point in time 241 * 242 * Software time stamps generated by ktime_get_real() are stored in 243 * skb->tstamp. 244 * 245 * hwtstamps can only be compared against other hwtstamps from 246 * the same device. 247 * 248 * This structure is attached to packets as part of the 249 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 250 */ 251struct skb_shared_hwtstamps { 252 ktime_t hwtstamp; 253}; 254 255/* Definitions for tx_flags in struct skb_shared_info */ 256enum { 257 /* generate hardware time stamp */ 258 SKBTX_HW_TSTAMP = 1 << 0, 259 260 /* generate software time stamp when queueing packet to NIC */ 261 SKBTX_SW_TSTAMP = 1 << 1, 262 263 /* device driver is going to provide hardware time stamp */ 264 SKBTX_IN_PROGRESS = 1 << 2, 265 266 /* device driver supports TX zero-copy buffers */ 267 SKBTX_DEV_ZEROCOPY = 1 << 3, 268 269 /* generate wifi status information (where possible) */ 270 SKBTX_WIFI_STATUS = 1 << 4, 271 272 /* This indicates at least one fragment might be overwritten 273 * (as in vmsplice(), sendfile() ...) 274 * If we need to compute a TX checksum, we'll need to copy 275 * all frags to avoid possible bad checksum 276 */ 277 SKBTX_SHARED_FRAG = 1 << 5, 278 279 /* generate software time stamp when entering packet scheduling */ 280 SKBTX_SCHED_TSTAMP = 1 << 6, 281 282 /* generate software timestamp on peer data acknowledgment */ 283 SKBTX_ACK_TSTAMP = 1 << 7, 284}; 285 286#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 287 SKBTX_SCHED_TSTAMP | \ 288 SKBTX_ACK_TSTAMP) 289#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 290 291/* 292 * The callback notifies userspace to release buffers when skb DMA is done in 293 * lower device, the skb last reference should be 0 when calling this. 294 * The zerocopy_success argument is true if zero copy transmit occurred, 295 * false on data copy or out of memory error caused by data copy attempt. 296 * The ctx field is used to track device context. 297 * The desc field is used to track userspace buffer index. 298 */ 299struct ubuf_info { 300 void (*callback)(struct ubuf_info *, bool zerocopy_success); 301 void *ctx; 302 unsigned long desc; 303}; 304 305/* This data is invariant across clones and lives at 306 * the end of the header data, ie. at skb->end. 307 */ 308struct skb_shared_info { 309 unsigned char nr_frags; 310 __u8 tx_flags; 311 unsigned short gso_size; 312 /* Warning: this field is not always filled in (UFO)! */ 313 unsigned short gso_segs; 314 unsigned short gso_type; 315 struct sk_buff *frag_list; 316 struct skb_shared_hwtstamps hwtstamps; 317 u32 tskey; 318 __be32 ip6_frag_id; 319 320 /* 321 * Warning : all fields before dataref are cleared in __alloc_skb() 322 */ 323 atomic_t dataref; 324 325 /* Intermediate layers must ensure that destructor_arg 326 * remains valid until skb destructor */ 327 void * destructor_arg; 328 329 /* must be last field, see pskb_expand_head() */ 330 skb_frag_t frags[MAX_SKB_FRAGS]; 331}; 332 333/* We divide dataref into two halves. The higher 16 bits hold references 334 * to the payload part of skb->data. The lower 16 bits hold references to 335 * the entire skb->data. A clone of a headerless skb holds the length of 336 * the header in skb->hdr_len. 337 * 338 * All users must obey the rule that the skb->data reference count must be 339 * greater than or equal to the payload reference count. 340 * 341 * Holding a reference to the payload part means that the user does not 342 * care about modifications to the header part of skb->data. 343 */ 344#define SKB_DATAREF_SHIFT 16 345#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 346 347 348enum { 349 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 350 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 351 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 352}; 353 354enum { 355 SKB_GSO_TCPV4 = 1 << 0, 356 SKB_GSO_UDP = 1 << 1, 357 358 /* This indicates the skb is from an untrusted source. */ 359 SKB_GSO_DODGY = 1 << 2, 360 361 /* This indicates the tcp segment has CWR set. */ 362 SKB_GSO_TCP_ECN = 1 << 3, 363 364 SKB_GSO_TCPV6 = 1 << 4, 365 366 SKB_GSO_FCOE = 1 << 5, 367 368 SKB_GSO_GRE = 1 << 6, 369 370 SKB_GSO_GRE_CSUM = 1 << 7, 371 372 SKB_GSO_IPIP = 1 << 8, 373 374 SKB_GSO_SIT = 1 << 9, 375 376 SKB_GSO_UDP_TUNNEL = 1 << 10, 377 378 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 379 380 SKB_GSO_TUNNEL_REMCSUM = 1 << 12, 381}; 382 383#if BITS_PER_LONG > 32 384#define NET_SKBUFF_DATA_USES_OFFSET 1 385#endif 386 387#ifdef NET_SKBUFF_DATA_USES_OFFSET 388typedef unsigned int sk_buff_data_t; 389#else 390typedef unsigned char *sk_buff_data_t; 391#endif 392 393/** 394 * struct skb_mstamp - multi resolution time stamps 395 * @stamp_us: timestamp in us resolution 396 * @stamp_jiffies: timestamp in jiffies 397 */ 398struct skb_mstamp { 399 union { 400 u64 v64; 401 struct { 402 u32 stamp_us; 403 u32 stamp_jiffies; 404 }; 405 }; 406}; 407 408/** 409 * skb_mstamp_get - get current timestamp 410 * @cl: place to store timestamps 411 */ 412static inline void skb_mstamp_get(struct skb_mstamp *cl) 413{ 414 u64 val = local_clock(); 415 416 do_div(val, NSEC_PER_USEC); 417 cl->stamp_us = (u32)val; 418 cl->stamp_jiffies = (u32)jiffies; 419} 420 421/** 422 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp 423 * @t1: pointer to newest sample 424 * @t0: pointer to oldest sample 425 */ 426static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1, 427 const struct skb_mstamp *t0) 428{ 429 s32 delta_us = t1->stamp_us - t0->stamp_us; 430 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies; 431 432 /* If delta_us is negative, this might be because interval is too big, 433 * or local_clock() drift is too big : fallback using jiffies. 434 */ 435 if (delta_us <= 0 || 436 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ))) 437 438 delta_us = jiffies_to_usecs(delta_jiffies); 439 440 return delta_us; 441} 442 443 444/** 445 * struct sk_buff - socket buffer 446 * @next: Next buffer in list 447 * @prev: Previous buffer in list 448 * @tstamp: Time we arrived/left 449 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 450 * @sk: Socket we are owned by 451 * @dev: Device we arrived on/are leaving by 452 * @cb: Control buffer. Free for use by every layer. Put private vars here 453 * @_skb_refdst: destination entry (with norefcount bit) 454 * @sp: the security path, used for xfrm 455 * @len: Length of actual data 456 * @data_len: Data length 457 * @mac_len: Length of link layer header 458 * @hdr_len: writable header length of cloned skb 459 * @csum: Checksum (must include start/offset pair) 460 * @csum_start: Offset from skb->head where checksumming should start 461 * @csum_offset: Offset from csum_start where checksum should be stored 462 * @priority: Packet queueing priority 463 * @ignore_df: allow local fragmentation 464 * @cloned: Head may be cloned (check refcnt to be sure) 465 * @ip_summed: Driver fed us an IP checksum 466 * @nohdr: Payload reference only, must not modify header 467 * @nfctinfo: Relationship of this skb to the connection 468 * @pkt_type: Packet class 469 * @fclone: skbuff clone status 470 * @ipvs_property: skbuff is owned by ipvs 471 * @peeked: this packet has been seen already, so stats have been 472 * done for it, don't do them again 473 * @nf_trace: netfilter packet trace flag 474 * @protocol: Packet protocol from driver 475 * @destructor: Destruct function 476 * @nfct: Associated connection, if any 477 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 478 * @skb_iif: ifindex of device we arrived on 479 * @tc_index: Traffic control index 480 * @tc_verd: traffic control verdict 481 * @hash: the packet hash 482 * @queue_mapping: Queue mapping for multiqueue devices 483 * @xmit_more: More SKBs are pending for this queue 484 * @ndisc_nodetype: router type (from link layer) 485 * @ooo_okay: allow the mapping of a socket to a queue to be changed 486 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 487 * ports. 488 * @sw_hash: indicates hash was computed in software stack 489 * @wifi_acked_valid: wifi_acked was set 490 * @wifi_acked: whether frame was acked on wifi or not 491 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 492 * @napi_id: id of the NAPI struct this skb came from 493 * @secmark: security marking 494 * @mark: Generic packet mark 495 * @dropcount: total number of sk_receive_queue overflows 496 * @vlan_proto: vlan encapsulation protocol 497 * @vlan_tci: vlan tag control information 498 * @inner_protocol: Protocol (encapsulation) 499 * @inner_transport_header: Inner transport layer header (encapsulation) 500 * @inner_network_header: Network layer header (encapsulation) 501 * @inner_mac_header: Link layer header (encapsulation) 502 * @transport_header: Transport layer header 503 * @network_header: Network layer header 504 * @mac_header: Link layer header 505 * @tail: Tail pointer 506 * @end: End pointer 507 * @head: Head of buffer 508 * @data: Data head pointer 509 * @truesize: Buffer size 510 * @users: User count - see {datagram,tcp}.c 511 */ 512 513struct sk_buff { 514 union { 515 struct { 516 /* These two members must be first. */ 517 struct sk_buff *next; 518 struct sk_buff *prev; 519 520 union { 521 ktime_t tstamp; 522 struct skb_mstamp skb_mstamp; 523 }; 524 }; 525 struct rb_node rbnode; /* used in netem & tcp stack */ 526 }; 527 struct sock *sk; 528 struct net_device *dev; 529 530 /* 531 * This is the control buffer. It is free to use for every 532 * layer. Please put your private variables there. If you 533 * want to keep them across layers you have to do a skb_clone() 534 * first. This is owned by whoever has the skb queued ATM. 535 */ 536 char cb[48] __aligned(8); 537 538 unsigned long _skb_refdst; 539 void (*destructor)(struct sk_buff *skb); 540#ifdef CONFIG_XFRM 541 struct sec_path *sp; 542#endif 543#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 544 struct nf_conntrack *nfct; 545#endif 546#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 547 struct nf_bridge_info *nf_bridge; 548#endif 549 unsigned int len, 550 data_len; 551 __u16 mac_len, 552 hdr_len; 553 554 /* Following fields are _not_ copied in __copy_skb_header() 555 * Note that queue_mapping is here mostly to fill a hole. 556 */ 557 kmemcheck_bitfield_begin(flags1); 558 __u16 queue_mapping; 559 __u8 cloned:1, 560 nohdr:1, 561 fclone:2, 562 peeked:1, 563 head_frag:1, 564 xmit_more:1; 565 /* one bit hole */ 566 kmemcheck_bitfield_end(flags1); 567 568 /* fields enclosed in headers_start/headers_end are copied 569 * using a single memcpy() in __copy_skb_header() 570 */ 571 /* private: */ 572 __u32 headers_start[0]; 573 /* public: */ 574 575/* if you move pkt_type around you also must adapt those constants */ 576#ifdef __BIG_ENDIAN_BITFIELD 577#define PKT_TYPE_MAX (7 << 5) 578#else 579#define PKT_TYPE_MAX 7 580#endif 581#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 582 583 __u8 __pkt_type_offset[0]; 584 __u8 pkt_type:3; 585 __u8 pfmemalloc:1; 586 __u8 ignore_df:1; 587 __u8 nfctinfo:3; 588 589 __u8 nf_trace:1; 590 __u8 ip_summed:2; 591 __u8 ooo_okay:1; 592 __u8 l4_hash:1; 593 __u8 sw_hash:1; 594 __u8 wifi_acked_valid:1; 595 __u8 wifi_acked:1; 596 597 __u8 no_fcs:1; 598 /* Indicates the inner headers are valid in the skbuff. */ 599 __u8 encapsulation:1; 600 __u8 encap_hdr_csum:1; 601 __u8 csum_valid:1; 602 __u8 csum_complete_sw:1; 603 __u8 csum_level:2; 604 __u8 csum_bad:1; 605 606#ifdef CONFIG_IPV6_NDISC_NODETYPE 607 __u8 ndisc_nodetype:2; 608#endif 609 __u8 ipvs_property:1; 610 __u8 inner_protocol_type:1; 611 __u8 remcsum_offload:1; 612 /* 3 or 5 bit hole */ 613 614#ifdef CONFIG_NET_SCHED 615 __u16 tc_index; /* traffic control index */ 616#ifdef CONFIG_NET_CLS_ACT 617 __u16 tc_verd; /* traffic control verdict */ 618#endif 619#endif 620 621 union { 622 __wsum csum; 623 struct { 624 __u16 csum_start; 625 __u16 csum_offset; 626 }; 627 }; 628 __u32 priority; 629 int skb_iif; 630 __u32 hash; 631 __be16 vlan_proto; 632 __u16 vlan_tci; 633#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 634 union { 635 unsigned int napi_id; 636 unsigned int sender_cpu; 637 }; 638#endif 639#ifdef CONFIG_NETWORK_SECMARK 640 __u32 secmark; 641#endif 642 union { 643 __u32 mark; 644 __u32 dropcount; 645 __u32 reserved_tailroom; 646 }; 647 648 union { 649 __be16 inner_protocol; 650 __u8 inner_ipproto; 651 }; 652 653 __u16 inner_transport_header; 654 __u16 inner_network_header; 655 __u16 inner_mac_header; 656 657 __be16 protocol; 658 __u16 transport_header; 659 __u16 network_header; 660 __u16 mac_header; 661 662 /* private: */ 663 __u32 headers_end[0]; 664 /* public: */ 665 666 /* These elements must be at the end, see alloc_skb() for details. */ 667 sk_buff_data_t tail; 668 sk_buff_data_t end; 669 unsigned char *head, 670 *data; 671 unsigned int truesize; 672 atomic_t users; 673}; 674 675#ifdef __KERNEL__ 676/* 677 * Handling routines are only of interest to the kernel 678 */ 679#include <linux/slab.h> 680 681 682#define SKB_ALLOC_FCLONE 0x01 683#define SKB_ALLOC_RX 0x02 684#define SKB_ALLOC_NAPI 0x04 685 686/* Returns true if the skb was allocated from PFMEMALLOC reserves */ 687static inline bool skb_pfmemalloc(const struct sk_buff *skb) 688{ 689 return unlikely(skb->pfmemalloc); 690} 691 692/* 693 * skb might have a dst pointer attached, refcounted or not. 694 * _skb_refdst low order bit is set if refcount was _not_ taken 695 */ 696#define SKB_DST_NOREF 1UL 697#define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 698 699/** 700 * skb_dst - returns skb dst_entry 701 * @skb: buffer 702 * 703 * Returns skb dst_entry, regardless of reference taken or not. 704 */ 705static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 706{ 707 /* If refdst was not refcounted, check we still are in a 708 * rcu_read_lock section 709 */ 710 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 711 !rcu_read_lock_held() && 712 !rcu_read_lock_bh_held()); 713 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 714} 715 716/** 717 * skb_dst_set - sets skb dst 718 * @skb: buffer 719 * @dst: dst entry 720 * 721 * Sets skb dst, assuming a reference was taken on dst and should 722 * be released by skb_dst_drop() 723 */ 724static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 725{ 726 skb->_skb_refdst = (unsigned long)dst; 727} 728 729/** 730 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 731 * @skb: buffer 732 * @dst: dst entry 733 * 734 * Sets skb dst, assuming a reference was not taken on dst. 735 * If dst entry is cached, we do not take reference and dst_release 736 * will be avoided by refdst_drop. If dst entry is not cached, we take 737 * reference, so that last dst_release can destroy the dst immediately. 738 */ 739static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 740{ 741 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 742 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 743} 744 745/** 746 * skb_dst_is_noref - Test if skb dst isn't refcounted 747 * @skb: buffer 748 */ 749static inline bool skb_dst_is_noref(const struct sk_buff *skb) 750{ 751 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 752} 753 754static inline struct rtable *skb_rtable(const struct sk_buff *skb) 755{ 756 return (struct rtable *)skb_dst(skb); 757} 758 759void kfree_skb(struct sk_buff *skb); 760void kfree_skb_list(struct sk_buff *segs); 761void skb_tx_error(struct sk_buff *skb); 762void consume_skb(struct sk_buff *skb); 763void __kfree_skb(struct sk_buff *skb); 764extern struct kmem_cache *skbuff_head_cache; 765 766void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 767bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 768 bool *fragstolen, int *delta_truesize); 769 770struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 771 int node); 772struct sk_buff *build_skb(void *data, unsigned int frag_size); 773static inline struct sk_buff *alloc_skb(unsigned int size, 774 gfp_t priority) 775{ 776 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 777} 778 779struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 780 unsigned long data_len, 781 int max_page_order, 782 int *errcode, 783 gfp_t gfp_mask); 784 785/* Layout of fast clones : [skb1][skb2][fclone_ref] */ 786struct sk_buff_fclones { 787 struct sk_buff skb1; 788 789 struct sk_buff skb2; 790 791 atomic_t fclone_ref; 792}; 793 794/** 795 * skb_fclone_busy - check if fclone is busy 796 * @skb: buffer 797 * 798 * Returns true is skb is a fast clone, and its clone is not freed. 799 * Some drivers call skb_orphan() in their ndo_start_xmit(), 800 * so we also check that this didnt happen. 801 */ 802static inline bool skb_fclone_busy(const struct sock *sk, 803 const struct sk_buff *skb) 804{ 805 const struct sk_buff_fclones *fclones; 806 807 fclones = container_of(skb, struct sk_buff_fclones, skb1); 808 809 return skb->fclone == SKB_FCLONE_ORIG && 810 atomic_read(&fclones->fclone_ref) > 1 && 811 fclones->skb2.sk == sk; 812} 813 814static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 815 gfp_t priority) 816{ 817 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 818} 819 820struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 821static inline struct sk_buff *alloc_skb_head(gfp_t priority) 822{ 823 return __alloc_skb_head(priority, -1); 824} 825 826struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 827int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 828struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 829struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 830struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 831 gfp_t gfp_mask, bool fclone); 832static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 833 gfp_t gfp_mask) 834{ 835 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 836} 837 838int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 839struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 840 unsigned int headroom); 841struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 842 int newtailroom, gfp_t priority); 843int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 844 int offset, int len); 845int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, 846 int len); 847int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 848int skb_pad(struct sk_buff *skb, int pad); 849#define dev_kfree_skb(a) consume_skb(a) 850 851int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 852 int getfrag(void *from, char *to, int offset, 853 int len, int odd, struct sk_buff *skb), 854 void *from, int length); 855 856struct skb_seq_state { 857 __u32 lower_offset; 858 __u32 upper_offset; 859 __u32 frag_idx; 860 __u32 stepped_offset; 861 struct sk_buff *root_skb; 862 struct sk_buff *cur_skb; 863 __u8 *frag_data; 864}; 865 866void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 867 unsigned int to, struct skb_seq_state *st); 868unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 869 struct skb_seq_state *st); 870void skb_abort_seq_read(struct skb_seq_state *st); 871 872unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 873 unsigned int to, struct ts_config *config, 874 struct ts_state *state); 875 876/* 877 * Packet hash types specify the type of hash in skb_set_hash. 878 * 879 * Hash types refer to the protocol layer addresses which are used to 880 * construct a packet's hash. The hashes are used to differentiate or identify 881 * flows of the protocol layer for the hash type. Hash types are either 882 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 883 * 884 * Properties of hashes: 885 * 886 * 1) Two packets in different flows have different hash values 887 * 2) Two packets in the same flow should have the same hash value 888 * 889 * A hash at a higher layer is considered to be more specific. A driver should 890 * set the most specific hash possible. 891 * 892 * A driver cannot indicate a more specific hash than the layer at which a hash 893 * was computed. For instance an L3 hash cannot be set as an L4 hash. 894 * 895 * A driver may indicate a hash level which is less specific than the 896 * actual layer the hash was computed on. For instance, a hash computed 897 * at L4 may be considered an L3 hash. This should only be done if the 898 * driver can't unambiguously determine that the HW computed the hash at 899 * the higher layer. Note that the "should" in the second property above 900 * permits this. 901 */ 902enum pkt_hash_types { 903 PKT_HASH_TYPE_NONE, /* Undefined type */ 904 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 905 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 906 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 907}; 908 909static inline void 910skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 911{ 912 skb->l4_hash = (type == PKT_HASH_TYPE_L4); 913 skb->sw_hash = 0; 914 skb->hash = hash; 915} 916 917void __skb_get_hash(struct sk_buff *skb); 918static inline __u32 skb_get_hash(struct sk_buff *skb) 919{ 920 if (!skb->l4_hash && !skb->sw_hash) 921 __skb_get_hash(skb); 922 923 return skb->hash; 924} 925 926static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 927{ 928 return skb->hash; 929} 930 931static inline void skb_clear_hash(struct sk_buff *skb) 932{ 933 skb->hash = 0; 934 skb->sw_hash = 0; 935 skb->l4_hash = 0; 936} 937 938static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 939{ 940 if (!skb->l4_hash) 941 skb_clear_hash(skb); 942} 943 944static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 945{ 946 to->hash = from->hash; 947 to->sw_hash = from->sw_hash; 948 to->l4_hash = from->l4_hash; 949}; 950 951static inline void skb_sender_cpu_clear(struct sk_buff *skb) 952{ 953#ifdef CONFIG_XPS 954 skb->sender_cpu = 0; 955#endif 956} 957 958#ifdef NET_SKBUFF_DATA_USES_OFFSET 959static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 960{ 961 return skb->head + skb->end; 962} 963 964static inline unsigned int skb_end_offset(const struct sk_buff *skb) 965{ 966 return skb->end; 967} 968#else 969static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 970{ 971 return skb->end; 972} 973 974static inline unsigned int skb_end_offset(const struct sk_buff *skb) 975{ 976 return skb->end - skb->head; 977} 978#endif 979 980/* Internal */ 981#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 982 983static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 984{ 985 return &skb_shinfo(skb)->hwtstamps; 986} 987 988/** 989 * skb_queue_empty - check if a queue is empty 990 * @list: queue head 991 * 992 * Returns true if the queue is empty, false otherwise. 993 */ 994static inline int skb_queue_empty(const struct sk_buff_head *list) 995{ 996 return list->next == (const struct sk_buff *) list; 997} 998 999/** 1000 * skb_queue_is_last - check if skb is the last entry in the queue 1001 * @list: queue head 1002 * @skb: buffer 1003 * 1004 * Returns true if @skb is the last buffer on the list. 1005 */ 1006static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1007 const struct sk_buff *skb) 1008{ 1009 return skb->next == (const struct sk_buff *) list; 1010} 1011 1012/** 1013 * skb_queue_is_first - check if skb is the first entry in the queue 1014 * @list: queue head 1015 * @skb: buffer 1016 * 1017 * Returns true if @skb is the first buffer on the list. 1018 */ 1019static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1020 const struct sk_buff *skb) 1021{ 1022 return skb->prev == (const struct sk_buff *) list; 1023} 1024 1025/** 1026 * skb_queue_next - return the next packet in the queue 1027 * @list: queue head 1028 * @skb: current buffer 1029 * 1030 * Return the next packet in @list after @skb. It is only valid to 1031 * call this if skb_queue_is_last() evaluates to false. 1032 */ 1033static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1034 const struct sk_buff *skb) 1035{ 1036 /* This BUG_ON may seem severe, but if we just return then we 1037 * are going to dereference garbage. 1038 */ 1039 BUG_ON(skb_queue_is_last(list, skb)); 1040 return skb->next; 1041} 1042 1043/** 1044 * skb_queue_prev - return the prev packet in the queue 1045 * @list: queue head 1046 * @skb: current buffer 1047 * 1048 * Return the prev packet in @list before @skb. It is only valid to 1049 * call this if skb_queue_is_first() evaluates to false. 1050 */ 1051static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1052 const struct sk_buff *skb) 1053{ 1054 /* This BUG_ON may seem severe, but if we just return then we 1055 * are going to dereference garbage. 1056 */ 1057 BUG_ON(skb_queue_is_first(list, skb)); 1058 return skb->prev; 1059} 1060 1061/** 1062 * skb_get - reference buffer 1063 * @skb: buffer to reference 1064 * 1065 * Makes another reference to a socket buffer and returns a pointer 1066 * to the buffer. 1067 */ 1068static inline struct sk_buff *skb_get(struct sk_buff *skb) 1069{ 1070 atomic_inc(&skb->users); 1071 return skb; 1072} 1073 1074/* 1075 * If users == 1, we are the only owner and are can avoid redundant 1076 * atomic change. 1077 */ 1078 1079/** 1080 * skb_cloned - is the buffer a clone 1081 * @skb: buffer to check 1082 * 1083 * Returns true if the buffer was generated with skb_clone() and is 1084 * one of multiple shared copies of the buffer. Cloned buffers are 1085 * shared data so must not be written to under normal circumstances. 1086 */ 1087static inline int skb_cloned(const struct sk_buff *skb) 1088{ 1089 return skb->cloned && 1090 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1091} 1092 1093static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1094{ 1095 might_sleep_if(pri & __GFP_WAIT); 1096 1097 if (skb_cloned(skb)) 1098 return pskb_expand_head(skb, 0, 0, pri); 1099 1100 return 0; 1101} 1102 1103/** 1104 * skb_header_cloned - is the header a clone 1105 * @skb: buffer to check 1106 * 1107 * Returns true if modifying the header part of the buffer requires 1108 * the data to be copied. 1109 */ 1110static inline int skb_header_cloned(const struct sk_buff *skb) 1111{ 1112 int dataref; 1113 1114 if (!skb->cloned) 1115 return 0; 1116 1117 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1118 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1119 return dataref != 1; 1120} 1121 1122/** 1123 * skb_header_release - release reference to header 1124 * @skb: buffer to operate on 1125 * 1126 * Drop a reference to the header part of the buffer. This is done 1127 * by acquiring a payload reference. You must not read from the header 1128 * part of skb->data after this. 1129 * Note : Check if you can use __skb_header_release() instead. 1130 */ 1131static inline void skb_header_release(struct sk_buff *skb) 1132{ 1133 BUG_ON(skb->nohdr); 1134 skb->nohdr = 1; 1135 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1136} 1137 1138/** 1139 * __skb_header_release - release reference to header 1140 * @skb: buffer to operate on 1141 * 1142 * Variant of skb_header_release() assuming skb is private to caller. 1143 * We can avoid one atomic operation. 1144 */ 1145static inline void __skb_header_release(struct sk_buff *skb) 1146{ 1147 skb->nohdr = 1; 1148 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1149} 1150 1151 1152/** 1153 * skb_shared - is the buffer shared 1154 * @skb: buffer to check 1155 * 1156 * Returns true if more than one person has a reference to this 1157 * buffer. 1158 */ 1159static inline int skb_shared(const struct sk_buff *skb) 1160{ 1161 return atomic_read(&skb->users) != 1; 1162} 1163 1164/** 1165 * skb_share_check - check if buffer is shared and if so clone it 1166 * @skb: buffer to check 1167 * @pri: priority for memory allocation 1168 * 1169 * If the buffer is shared the buffer is cloned and the old copy 1170 * drops a reference. A new clone with a single reference is returned. 1171 * If the buffer is not shared the original buffer is returned. When 1172 * being called from interrupt status or with spinlocks held pri must 1173 * be GFP_ATOMIC. 1174 * 1175 * NULL is returned on a memory allocation failure. 1176 */ 1177static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1178{ 1179 might_sleep_if(pri & __GFP_WAIT); 1180 if (skb_shared(skb)) { 1181 struct sk_buff *nskb = skb_clone(skb, pri); 1182 1183 if (likely(nskb)) 1184 consume_skb(skb); 1185 else 1186 kfree_skb(skb); 1187 skb = nskb; 1188 } 1189 return skb; 1190} 1191 1192/* 1193 * Copy shared buffers into a new sk_buff. We effectively do COW on 1194 * packets to handle cases where we have a local reader and forward 1195 * and a couple of other messy ones. The normal one is tcpdumping 1196 * a packet thats being forwarded. 1197 */ 1198 1199/** 1200 * skb_unshare - make a copy of a shared buffer 1201 * @skb: buffer to check 1202 * @pri: priority for memory allocation 1203 * 1204 * If the socket buffer is a clone then this function creates a new 1205 * copy of the data, drops a reference count on the old copy and returns 1206 * the new copy with the reference count at 1. If the buffer is not a clone 1207 * the original buffer is returned. When called with a spinlock held or 1208 * from interrupt state @pri must be %GFP_ATOMIC 1209 * 1210 * %NULL is returned on a memory allocation failure. 1211 */ 1212static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1213 gfp_t pri) 1214{ 1215 might_sleep_if(pri & __GFP_WAIT); 1216 if (skb_cloned(skb)) { 1217 struct sk_buff *nskb = skb_copy(skb, pri); 1218 1219 /* Free our shared copy */ 1220 if (likely(nskb)) 1221 consume_skb(skb); 1222 else 1223 kfree_skb(skb); 1224 skb = nskb; 1225 } 1226 return skb; 1227} 1228 1229/** 1230 * skb_peek - peek at the head of an &sk_buff_head 1231 * @list_: list to peek at 1232 * 1233 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1234 * be careful with this one. A peek leaves the buffer on the 1235 * list and someone else may run off with it. You must hold 1236 * the appropriate locks or have a private queue to do this. 1237 * 1238 * Returns %NULL for an empty list or a pointer to the head element. 1239 * The reference count is not incremented and the reference is therefore 1240 * volatile. Use with caution. 1241 */ 1242static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1243{ 1244 struct sk_buff *skb = list_->next; 1245 1246 if (skb == (struct sk_buff *)list_) 1247 skb = NULL; 1248 return skb; 1249} 1250 1251/** 1252 * skb_peek_next - peek skb following the given one from a queue 1253 * @skb: skb to start from 1254 * @list_: list to peek at 1255 * 1256 * Returns %NULL when the end of the list is met or a pointer to the 1257 * next element. The reference count is not incremented and the 1258 * reference is therefore volatile. Use with caution. 1259 */ 1260static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1261 const struct sk_buff_head *list_) 1262{ 1263 struct sk_buff *next = skb->next; 1264 1265 if (next == (struct sk_buff *)list_) 1266 next = NULL; 1267 return next; 1268} 1269 1270/** 1271 * skb_peek_tail - peek at the tail of an &sk_buff_head 1272 * @list_: list to peek at 1273 * 1274 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1275 * be careful with this one. A peek leaves the buffer on the 1276 * list and someone else may run off with it. You must hold 1277 * the appropriate locks or have a private queue to do this. 1278 * 1279 * Returns %NULL for an empty list or a pointer to the tail element. 1280 * The reference count is not incremented and the reference is therefore 1281 * volatile. Use with caution. 1282 */ 1283static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1284{ 1285 struct sk_buff *skb = list_->prev; 1286 1287 if (skb == (struct sk_buff *)list_) 1288 skb = NULL; 1289 return skb; 1290 1291} 1292 1293/** 1294 * skb_queue_len - get queue length 1295 * @list_: list to measure 1296 * 1297 * Return the length of an &sk_buff queue. 1298 */ 1299static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1300{ 1301 return list_->qlen; 1302} 1303 1304/** 1305 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1306 * @list: queue to initialize 1307 * 1308 * This initializes only the list and queue length aspects of 1309 * an sk_buff_head object. This allows to initialize the list 1310 * aspects of an sk_buff_head without reinitializing things like 1311 * the spinlock. It can also be used for on-stack sk_buff_head 1312 * objects where the spinlock is known to not be used. 1313 */ 1314static inline void __skb_queue_head_init(struct sk_buff_head *list) 1315{ 1316 list->prev = list->next = (struct sk_buff *)list; 1317 list->qlen = 0; 1318} 1319 1320/* 1321 * This function creates a split out lock class for each invocation; 1322 * this is needed for now since a whole lot of users of the skb-queue 1323 * infrastructure in drivers have different locking usage (in hardirq) 1324 * than the networking core (in softirq only). In the long run either the 1325 * network layer or drivers should need annotation to consolidate the 1326 * main types of usage into 3 classes. 1327 */ 1328static inline void skb_queue_head_init(struct sk_buff_head *list) 1329{ 1330 spin_lock_init(&list->lock); 1331 __skb_queue_head_init(list); 1332} 1333 1334static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1335 struct lock_class_key *class) 1336{ 1337 skb_queue_head_init(list); 1338 lockdep_set_class(&list->lock, class); 1339} 1340 1341/* 1342 * Insert an sk_buff on a list. 1343 * 1344 * The "__skb_xxxx()" functions are the non-atomic ones that 1345 * can only be called with interrupts disabled. 1346 */ 1347void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1348 struct sk_buff_head *list); 1349static inline void __skb_insert(struct sk_buff *newsk, 1350 struct sk_buff *prev, struct sk_buff *next, 1351 struct sk_buff_head *list) 1352{ 1353 newsk->next = next; 1354 newsk->prev = prev; 1355 next->prev = prev->next = newsk; 1356 list->qlen++; 1357} 1358 1359static inline void __skb_queue_splice(const struct sk_buff_head *list, 1360 struct sk_buff *prev, 1361 struct sk_buff *next) 1362{ 1363 struct sk_buff *first = list->next; 1364 struct sk_buff *last = list->prev; 1365 1366 first->prev = prev; 1367 prev->next = first; 1368 1369 last->next = next; 1370 next->prev = last; 1371} 1372 1373/** 1374 * skb_queue_splice - join two skb lists, this is designed for stacks 1375 * @list: the new list to add 1376 * @head: the place to add it in the first list 1377 */ 1378static inline void skb_queue_splice(const struct sk_buff_head *list, 1379 struct sk_buff_head *head) 1380{ 1381 if (!skb_queue_empty(list)) { 1382 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1383 head->qlen += list->qlen; 1384 } 1385} 1386 1387/** 1388 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1389 * @list: the new list to add 1390 * @head: the place to add it in the first list 1391 * 1392 * The list at @list is reinitialised 1393 */ 1394static inline void skb_queue_splice_init(struct sk_buff_head *list, 1395 struct sk_buff_head *head) 1396{ 1397 if (!skb_queue_empty(list)) { 1398 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1399 head->qlen += list->qlen; 1400 __skb_queue_head_init(list); 1401 } 1402} 1403 1404/** 1405 * skb_queue_splice_tail - join two skb lists, each list being a queue 1406 * @list: the new list to add 1407 * @head: the place to add it in the first list 1408 */ 1409static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1410 struct sk_buff_head *head) 1411{ 1412 if (!skb_queue_empty(list)) { 1413 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1414 head->qlen += list->qlen; 1415 } 1416} 1417 1418/** 1419 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1420 * @list: the new list to add 1421 * @head: the place to add it in the first list 1422 * 1423 * Each of the lists is a queue. 1424 * The list at @list is reinitialised 1425 */ 1426static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1427 struct sk_buff_head *head) 1428{ 1429 if (!skb_queue_empty(list)) { 1430 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1431 head->qlen += list->qlen; 1432 __skb_queue_head_init(list); 1433 } 1434} 1435 1436/** 1437 * __skb_queue_after - queue a buffer at the list head 1438 * @list: list to use 1439 * @prev: place after this buffer 1440 * @newsk: buffer to queue 1441 * 1442 * Queue a buffer int the middle of a list. This function takes no locks 1443 * and you must therefore hold required locks before calling it. 1444 * 1445 * A buffer cannot be placed on two lists at the same time. 1446 */ 1447static inline void __skb_queue_after(struct sk_buff_head *list, 1448 struct sk_buff *prev, 1449 struct sk_buff *newsk) 1450{ 1451 __skb_insert(newsk, prev, prev->next, list); 1452} 1453 1454void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1455 struct sk_buff_head *list); 1456 1457static inline void __skb_queue_before(struct sk_buff_head *list, 1458 struct sk_buff *next, 1459 struct sk_buff *newsk) 1460{ 1461 __skb_insert(newsk, next->prev, next, list); 1462} 1463 1464/** 1465 * __skb_queue_head - queue a buffer at the list head 1466 * @list: list to use 1467 * @newsk: buffer to queue 1468 * 1469 * Queue a buffer at the start of a list. This function takes no locks 1470 * and you must therefore hold required locks before calling it. 1471 * 1472 * A buffer cannot be placed on two lists at the same time. 1473 */ 1474void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1475static inline void __skb_queue_head(struct sk_buff_head *list, 1476 struct sk_buff *newsk) 1477{ 1478 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1479} 1480 1481/** 1482 * __skb_queue_tail - queue a buffer at the list tail 1483 * @list: list to use 1484 * @newsk: buffer to queue 1485 * 1486 * Queue a buffer at the end of a list. This function takes no locks 1487 * and you must therefore hold required locks before calling it. 1488 * 1489 * A buffer cannot be placed on two lists at the same time. 1490 */ 1491void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1492static inline void __skb_queue_tail(struct sk_buff_head *list, 1493 struct sk_buff *newsk) 1494{ 1495 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1496} 1497 1498/* 1499 * remove sk_buff from list. _Must_ be called atomically, and with 1500 * the list known.. 1501 */ 1502void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1503static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1504{ 1505 struct sk_buff *next, *prev; 1506 1507 list->qlen--; 1508 next = skb->next; 1509 prev = skb->prev; 1510 skb->next = skb->prev = NULL; 1511 next->prev = prev; 1512 prev->next = next; 1513} 1514 1515/** 1516 * __skb_dequeue - remove from the head of the queue 1517 * @list: list to dequeue from 1518 * 1519 * Remove the head of the list. This function does not take any locks 1520 * so must be used with appropriate locks held only. The head item is 1521 * returned or %NULL if the list is empty. 1522 */ 1523struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1524static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1525{ 1526 struct sk_buff *skb = skb_peek(list); 1527 if (skb) 1528 __skb_unlink(skb, list); 1529 return skb; 1530} 1531 1532/** 1533 * __skb_dequeue_tail - remove from the tail of the queue 1534 * @list: list to dequeue from 1535 * 1536 * Remove the tail of the list. This function does not take any locks 1537 * so must be used with appropriate locks held only. The tail item is 1538 * returned or %NULL if the list is empty. 1539 */ 1540struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1541static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1542{ 1543 struct sk_buff *skb = skb_peek_tail(list); 1544 if (skb) 1545 __skb_unlink(skb, list); 1546 return skb; 1547} 1548 1549 1550static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1551{ 1552 return skb->data_len; 1553} 1554 1555static inline unsigned int skb_headlen(const struct sk_buff *skb) 1556{ 1557 return skb->len - skb->data_len; 1558} 1559 1560static inline int skb_pagelen(const struct sk_buff *skb) 1561{ 1562 int i, len = 0; 1563 1564 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1565 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1566 return len + skb_headlen(skb); 1567} 1568 1569/** 1570 * __skb_fill_page_desc - initialise a paged fragment in an skb 1571 * @skb: buffer containing fragment to be initialised 1572 * @i: paged fragment index to initialise 1573 * @page: the page to use for this fragment 1574 * @off: the offset to the data with @page 1575 * @size: the length of the data 1576 * 1577 * Initialises the @i'th fragment of @skb to point to &size bytes at 1578 * offset @off within @page. 1579 * 1580 * Does not take any additional reference on the fragment. 1581 */ 1582static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1583 struct page *page, int off, int size) 1584{ 1585 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1586 1587 /* 1588 * Propagate page->pfmemalloc to the skb if we can. The problem is 1589 * that not all callers have unique ownership of the page. If 1590 * pfmemalloc is set, we check the mapping as a mapping implies 1591 * page->index is set (index and pfmemalloc share space). 1592 * If it's a valid mapping, we cannot use page->pfmemalloc but we 1593 * do not lose pfmemalloc information as the pages would not be 1594 * allocated using __GFP_MEMALLOC. 1595 */ 1596 frag->page.p = page; 1597 frag->page_offset = off; 1598 skb_frag_size_set(frag, size); 1599 1600 page = compound_head(page); 1601 if (page->pfmemalloc && !page->mapping) 1602 skb->pfmemalloc = true; 1603} 1604 1605/** 1606 * skb_fill_page_desc - initialise a paged fragment in an skb 1607 * @skb: buffer containing fragment to be initialised 1608 * @i: paged fragment index to initialise 1609 * @page: the page to use for this fragment 1610 * @off: the offset to the data with @page 1611 * @size: the length of the data 1612 * 1613 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1614 * @skb to point to @size bytes at offset @off within @page. In 1615 * addition updates @skb such that @i is the last fragment. 1616 * 1617 * Does not take any additional reference on the fragment. 1618 */ 1619static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1620 struct page *page, int off, int size) 1621{ 1622 __skb_fill_page_desc(skb, i, page, off, size); 1623 skb_shinfo(skb)->nr_frags = i + 1; 1624} 1625 1626void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1627 int size, unsigned int truesize); 1628 1629void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1630 unsigned int truesize); 1631 1632#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1633#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1634#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1635 1636#ifdef NET_SKBUFF_DATA_USES_OFFSET 1637static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1638{ 1639 return skb->head + skb->tail; 1640} 1641 1642static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1643{ 1644 skb->tail = skb->data - skb->head; 1645} 1646 1647static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1648{ 1649 skb_reset_tail_pointer(skb); 1650 skb->tail += offset; 1651} 1652 1653#else /* NET_SKBUFF_DATA_USES_OFFSET */ 1654static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1655{ 1656 return skb->tail; 1657} 1658 1659static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1660{ 1661 skb->tail = skb->data; 1662} 1663 1664static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1665{ 1666 skb->tail = skb->data + offset; 1667} 1668 1669#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1670 1671/* 1672 * Add data to an sk_buff 1673 */ 1674unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1675unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1676static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1677{ 1678 unsigned char *tmp = skb_tail_pointer(skb); 1679 SKB_LINEAR_ASSERT(skb); 1680 skb->tail += len; 1681 skb->len += len; 1682 return tmp; 1683} 1684 1685unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1686static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1687{ 1688 skb->data -= len; 1689 skb->len += len; 1690 return skb->data; 1691} 1692 1693unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1694static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1695{ 1696 skb->len -= len; 1697 BUG_ON(skb->len < skb->data_len); 1698 return skb->data += len; 1699} 1700 1701static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1702{ 1703 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1704} 1705 1706unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1707 1708static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1709{ 1710 if (len > skb_headlen(skb) && 1711 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1712 return NULL; 1713 skb->len -= len; 1714 return skb->data += len; 1715} 1716 1717static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1718{ 1719 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1720} 1721 1722static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1723{ 1724 if (likely(len <= skb_headlen(skb))) 1725 return 1; 1726 if (unlikely(len > skb->len)) 1727 return 0; 1728 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1729} 1730 1731/** 1732 * skb_headroom - bytes at buffer head 1733 * @skb: buffer to check 1734 * 1735 * Return the number of bytes of free space at the head of an &sk_buff. 1736 */ 1737static inline unsigned int skb_headroom(const struct sk_buff *skb) 1738{ 1739 return skb->data - skb->head; 1740} 1741 1742/** 1743 * skb_tailroom - bytes at buffer end 1744 * @skb: buffer to check 1745 * 1746 * Return the number of bytes of free space at the tail of an sk_buff 1747 */ 1748static inline int skb_tailroom(const struct sk_buff *skb) 1749{ 1750 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1751} 1752 1753/** 1754 * skb_availroom - bytes at buffer end 1755 * @skb: buffer to check 1756 * 1757 * Return the number of bytes of free space at the tail of an sk_buff 1758 * allocated by sk_stream_alloc() 1759 */ 1760static inline int skb_availroom(const struct sk_buff *skb) 1761{ 1762 if (skb_is_nonlinear(skb)) 1763 return 0; 1764 1765 return skb->end - skb->tail - skb->reserved_tailroom; 1766} 1767 1768/** 1769 * skb_reserve - adjust headroom 1770 * @skb: buffer to alter 1771 * @len: bytes to move 1772 * 1773 * Increase the headroom of an empty &sk_buff by reducing the tail 1774 * room. This is only allowed for an empty buffer. 1775 */ 1776static inline void skb_reserve(struct sk_buff *skb, int len) 1777{ 1778 skb->data += len; 1779 skb->tail += len; 1780} 1781 1782#define ENCAP_TYPE_ETHER 0 1783#define ENCAP_TYPE_IPPROTO 1 1784 1785static inline void skb_set_inner_protocol(struct sk_buff *skb, 1786 __be16 protocol) 1787{ 1788 skb->inner_protocol = protocol; 1789 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 1790} 1791 1792static inline void skb_set_inner_ipproto(struct sk_buff *skb, 1793 __u8 ipproto) 1794{ 1795 skb->inner_ipproto = ipproto; 1796 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 1797} 1798 1799static inline void skb_reset_inner_headers(struct sk_buff *skb) 1800{ 1801 skb->inner_mac_header = skb->mac_header; 1802 skb->inner_network_header = skb->network_header; 1803 skb->inner_transport_header = skb->transport_header; 1804} 1805 1806static inline void skb_reset_mac_len(struct sk_buff *skb) 1807{ 1808 skb->mac_len = skb->network_header - skb->mac_header; 1809} 1810 1811static inline unsigned char *skb_inner_transport_header(const struct sk_buff 1812 *skb) 1813{ 1814 return skb->head + skb->inner_transport_header; 1815} 1816 1817static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 1818{ 1819 skb->inner_transport_header = skb->data - skb->head; 1820} 1821 1822static inline void skb_set_inner_transport_header(struct sk_buff *skb, 1823 const int offset) 1824{ 1825 skb_reset_inner_transport_header(skb); 1826 skb->inner_transport_header += offset; 1827} 1828 1829static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 1830{ 1831 return skb->head + skb->inner_network_header; 1832} 1833 1834static inline void skb_reset_inner_network_header(struct sk_buff *skb) 1835{ 1836 skb->inner_network_header = skb->data - skb->head; 1837} 1838 1839static inline void skb_set_inner_network_header(struct sk_buff *skb, 1840 const int offset) 1841{ 1842 skb_reset_inner_network_header(skb); 1843 skb->inner_network_header += offset; 1844} 1845 1846static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 1847{ 1848 return skb->head + skb->inner_mac_header; 1849} 1850 1851static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 1852{ 1853 skb->inner_mac_header = skb->data - skb->head; 1854} 1855 1856static inline void skb_set_inner_mac_header(struct sk_buff *skb, 1857 const int offset) 1858{ 1859 skb_reset_inner_mac_header(skb); 1860 skb->inner_mac_header += offset; 1861} 1862static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 1863{ 1864 return skb->transport_header != (typeof(skb->transport_header))~0U; 1865} 1866 1867static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 1868{ 1869 return skb->head + skb->transport_header; 1870} 1871 1872static inline void skb_reset_transport_header(struct sk_buff *skb) 1873{ 1874 skb->transport_header = skb->data - skb->head; 1875} 1876 1877static inline void skb_set_transport_header(struct sk_buff *skb, 1878 const int offset) 1879{ 1880 skb_reset_transport_header(skb); 1881 skb->transport_header += offset; 1882} 1883 1884static inline unsigned char *skb_network_header(const struct sk_buff *skb) 1885{ 1886 return skb->head + skb->network_header; 1887} 1888 1889static inline void skb_reset_network_header(struct sk_buff *skb) 1890{ 1891 skb->network_header = skb->data - skb->head; 1892} 1893 1894static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 1895{ 1896 skb_reset_network_header(skb); 1897 skb->network_header += offset; 1898} 1899 1900static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 1901{ 1902 return skb->head + skb->mac_header; 1903} 1904 1905static inline int skb_mac_header_was_set(const struct sk_buff *skb) 1906{ 1907 return skb->mac_header != (typeof(skb->mac_header))~0U; 1908} 1909 1910static inline void skb_reset_mac_header(struct sk_buff *skb) 1911{ 1912 skb->mac_header = skb->data - skb->head; 1913} 1914 1915static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 1916{ 1917 skb_reset_mac_header(skb); 1918 skb->mac_header += offset; 1919} 1920 1921static inline void skb_pop_mac_header(struct sk_buff *skb) 1922{ 1923 skb->mac_header = skb->network_header; 1924} 1925 1926static inline void skb_probe_transport_header(struct sk_buff *skb, 1927 const int offset_hint) 1928{ 1929 struct flow_keys keys; 1930 1931 if (skb_transport_header_was_set(skb)) 1932 return; 1933 else if (skb_flow_dissect(skb, &keys)) 1934 skb_set_transport_header(skb, keys.thoff); 1935 else 1936 skb_set_transport_header(skb, offset_hint); 1937} 1938 1939static inline void skb_mac_header_rebuild(struct sk_buff *skb) 1940{ 1941 if (skb_mac_header_was_set(skb)) { 1942 const unsigned char *old_mac = skb_mac_header(skb); 1943 1944 skb_set_mac_header(skb, -skb->mac_len); 1945 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 1946 } 1947} 1948 1949static inline int skb_checksum_start_offset(const struct sk_buff *skb) 1950{ 1951 return skb->csum_start - skb_headroom(skb); 1952} 1953 1954static inline int skb_transport_offset(const struct sk_buff *skb) 1955{ 1956 return skb_transport_header(skb) - skb->data; 1957} 1958 1959static inline u32 skb_network_header_len(const struct sk_buff *skb) 1960{ 1961 return skb->transport_header - skb->network_header; 1962} 1963 1964static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 1965{ 1966 return skb->inner_transport_header - skb->inner_network_header; 1967} 1968 1969static inline int skb_network_offset(const struct sk_buff *skb) 1970{ 1971 return skb_network_header(skb) - skb->data; 1972} 1973 1974static inline int skb_inner_network_offset(const struct sk_buff *skb) 1975{ 1976 return skb_inner_network_header(skb) - skb->data; 1977} 1978 1979static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 1980{ 1981 return pskb_may_pull(skb, skb_network_offset(skb) + len); 1982} 1983 1984/* 1985 * CPUs often take a performance hit when accessing unaligned memory 1986 * locations. The actual performance hit varies, it can be small if the 1987 * hardware handles it or large if we have to take an exception and fix it 1988 * in software. 1989 * 1990 * Since an ethernet header is 14 bytes network drivers often end up with 1991 * the IP header at an unaligned offset. The IP header can be aligned by 1992 * shifting the start of the packet by 2 bytes. Drivers should do this 1993 * with: 1994 * 1995 * skb_reserve(skb, NET_IP_ALIGN); 1996 * 1997 * The downside to this alignment of the IP header is that the DMA is now 1998 * unaligned. On some architectures the cost of an unaligned DMA is high 1999 * and this cost outweighs the gains made by aligning the IP header. 2000 * 2001 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2002 * to be overridden. 2003 */ 2004#ifndef NET_IP_ALIGN 2005#define NET_IP_ALIGN 2 2006#endif 2007 2008/* 2009 * The networking layer reserves some headroom in skb data (via 2010 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2011 * the header has to grow. In the default case, if the header has to grow 2012 * 32 bytes or less we avoid the reallocation. 2013 * 2014 * Unfortunately this headroom changes the DMA alignment of the resulting 2015 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2016 * on some architectures. An architecture can override this value, 2017 * perhaps setting it to a cacheline in size (since that will maintain 2018 * cacheline alignment of the DMA). It must be a power of 2. 2019 * 2020 * Various parts of the networking layer expect at least 32 bytes of 2021 * headroom, you should not reduce this. 2022 * 2023 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2024 * to reduce average number of cache lines per packet. 2025 * get_rps_cpus() for example only access one 64 bytes aligned block : 2026 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2027 */ 2028#ifndef NET_SKB_PAD 2029#define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2030#endif 2031 2032int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2033 2034static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2035{ 2036 if (unlikely(skb_is_nonlinear(skb))) { 2037 WARN_ON(1); 2038 return; 2039 } 2040 skb->len = len; 2041 skb_set_tail_pointer(skb, len); 2042} 2043 2044void skb_trim(struct sk_buff *skb, unsigned int len); 2045 2046static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2047{ 2048 if (skb->data_len) 2049 return ___pskb_trim(skb, len); 2050 __skb_trim(skb, len); 2051 return 0; 2052} 2053 2054static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2055{ 2056 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2057} 2058 2059/** 2060 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2061 * @skb: buffer to alter 2062 * @len: new length 2063 * 2064 * This is identical to pskb_trim except that the caller knows that 2065 * the skb is not cloned so we should never get an error due to out- 2066 * of-memory. 2067 */ 2068static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2069{ 2070 int err = pskb_trim(skb, len); 2071 BUG_ON(err); 2072} 2073 2074/** 2075 * skb_orphan - orphan a buffer 2076 * @skb: buffer to orphan 2077 * 2078 * If a buffer currently has an owner then we call the owner's 2079 * destructor function and make the @skb unowned. The buffer continues 2080 * to exist but is no longer charged to its former owner. 2081 */ 2082static inline void skb_orphan(struct sk_buff *skb) 2083{ 2084 if (skb->destructor) { 2085 skb->destructor(skb); 2086 skb->destructor = NULL; 2087 skb->sk = NULL; 2088 } else { 2089 BUG_ON(skb->sk); 2090 } 2091} 2092 2093/** 2094 * skb_orphan_frags - orphan the frags contained in a buffer 2095 * @skb: buffer to orphan frags from 2096 * @gfp_mask: allocation mask for replacement pages 2097 * 2098 * For each frag in the SKB which needs a destructor (i.e. has an 2099 * owner) create a copy of that frag and release the original 2100 * page by calling the destructor. 2101 */ 2102static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2103{ 2104 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2105 return 0; 2106 return skb_copy_ubufs(skb, gfp_mask); 2107} 2108 2109/** 2110 * __skb_queue_purge - empty a list 2111 * @list: list to empty 2112 * 2113 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2114 * the list and one reference dropped. This function does not take the 2115 * list lock and the caller must hold the relevant locks to use it. 2116 */ 2117void skb_queue_purge(struct sk_buff_head *list); 2118static inline void __skb_queue_purge(struct sk_buff_head *list) 2119{ 2120 struct sk_buff *skb; 2121 while ((skb = __skb_dequeue(list)) != NULL) 2122 kfree_skb(skb); 2123} 2124 2125#define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768) 2126#define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER) 2127#define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE 2128 2129void *netdev_alloc_frag(unsigned int fragsz); 2130 2131struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2132 gfp_t gfp_mask); 2133 2134/** 2135 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2136 * @dev: network device to receive on 2137 * @length: length to allocate 2138 * 2139 * Allocate a new &sk_buff and assign it a usage count of one. The 2140 * buffer has unspecified headroom built in. Users should allocate 2141 * the headroom they think they need without accounting for the 2142 * built in space. The built in space is used for optimisations. 2143 * 2144 * %NULL is returned if there is no free memory. Although this function 2145 * allocates memory it can be called from an interrupt. 2146 */ 2147static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2148 unsigned int length) 2149{ 2150 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2151} 2152 2153/* legacy helper around __netdev_alloc_skb() */ 2154static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2155 gfp_t gfp_mask) 2156{ 2157 return __netdev_alloc_skb(NULL, length, gfp_mask); 2158} 2159 2160/* legacy helper around netdev_alloc_skb() */ 2161static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2162{ 2163 return netdev_alloc_skb(NULL, length); 2164} 2165 2166 2167static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2168 unsigned int length, gfp_t gfp) 2169{ 2170 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2171 2172 if (NET_IP_ALIGN && skb) 2173 skb_reserve(skb, NET_IP_ALIGN); 2174 return skb; 2175} 2176 2177static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2178 unsigned int length) 2179{ 2180 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2181} 2182 2183void *napi_alloc_frag(unsigned int fragsz); 2184struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2185 unsigned int length, gfp_t gfp_mask); 2186static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2187 unsigned int length) 2188{ 2189 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2190} 2191 2192/** 2193 * __dev_alloc_pages - allocate page for network Rx 2194 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2195 * @order: size of the allocation 2196 * 2197 * Allocate a new page. 2198 * 2199 * %NULL is returned if there is no free memory. 2200*/ 2201static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2202 unsigned int order) 2203{ 2204 /* This piece of code contains several assumptions. 2205 * 1. This is for device Rx, therefor a cold page is preferred. 2206 * 2. The expectation is the user wants a compound page. 2207 * 3. If requesting a order 0 page it will not be compound 2208 * due to the check to see if order has a value in prep_new_page 2209 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2210 * code in gfp_to_alloc_flags that should be enforcing this. 2211 */ 2212 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2213 2214 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2215} 2216 2217static inline struct page *dev_alloc_pages(unsigned int order) 2218{ 2219 return __dev_alloc_pages(GFP_ATOMIC, order); 2220} 2221 2222/** 2223 * __dev_alloc_page - allocate a page for network Rx 2224 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2225 * 2226 * Allocate a new page. 2227 * 2228 * %NULL is returned if there is no free memory. 2229 */ 2230static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2231{ 2232 return __dev_alloc_pages(gfp_mask, 0); 2233} 2234 2235static inline struct page *dev_alloc_page(void) 2236{ 2237 return __dev_alloc_page(GFP_ATOMIC); 2238} 2239 2240/** 2241 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2242 * @page: The page that was allocated from skb_alloc_page 2243 * @skb: The skb that may need pfmemalloc set 2244 */ 2245static inline void skb_propagate_pfmemalloc(struct page *page, 2246 struct sk_buff *skb) 2247{ 2248 if (page && page->pfmemalloc) 2249 skb->pfmemalloc = true; 2250} 2251 2252/** 2253 * skb_frag_page - retrieve the page referred to by a paged fragment 2254 * @frag: the paged fragment 2255 * 2256 * Returns the &struct page associated with @frag. 2257 */ 2258static inline struct page *skb_frag_page(const skb_frag_t *frag) 2259{ 2260 return frag->page.p; 2261} 2262 2263/** 2264 * __skb_frag_ref - take an addition reference on a paged fragment. 2265 * @frag: the paged fragment 2266 * 2267 * Takes an additional reference on the paged fragment @frag. 2268 */ 2269static inline void __skb_frag_ref(skb_frag_t *frag) 2270{ 2271 get_page(skb_frag_page(frag)); 2272} 2273 2274/** 2275 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2276 * @skb: the buffer 2277 * @f: the fragment offset. 2278 * 2279 * Takes an additional reference on the @f'th paged fragment of @skb. 2280 */ 2281static inline void skb_frag_ref(struct sk_buff *skb, int f) 2282{ 2283 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2284} 2285 2286/** 2287 * __skb_frag_unref - release a reference on a paged fragment. 2288 * @frag: the paged fragment 2289 * 2290 * Releases a reference on the paged fragment @frag. 2291 */ 2292static inline void __skb_frag_unref(skb_frag_t *frag) 2293{ 2294 put_page(skb_frag_page(frag)); 2295} 2296 2297/** 2298 * skb_frag_unref - release a reference on a paged fragment of an skb. 2299 * @skb: the buffer 2300 * @f: the fragment offset 2301 * 2302 * Releases a reference on the @f'th paged fragment of @skb. 2303 */ 2304static inline void skb_frag_unref(struct sk_buff *skb, int f) 2305{ 2306 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2307} 2308 2309/** 2310 * skb_frag_address - gets the address of the data contained in a paged fragment 2311 * @frag: the paged fragment buffer 2312 * 2313 * Returns the address of the data within @frag. The page must already 2314 * be mapped. 2315 */ 2316static inline void *skb_frag_address(const skb_frag_t *frag) 2317{ 2318 return page_address(skb_frag_page(frag)) + frag->page_offset; 2319} 2320 2321/** 2322 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2323 * @frag: the paged fragment buffer 2324 * 2325 * Returns the address of the data within @frag. Checks that the page 2326 * is mapped and returns %NULL otherwise. 2327 */ 2328static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2329{ 2330 void *ptr = page_address(skb_frag_page(frag)); 2331 if (unlikely(!ptr)) 2332 return NULL; 2333 2334 return ptr + frag->page_offset; 2335} 2336 2337/** 2338 * __skb_frag_set_page - sets the page contained in a paged fragment 2339 * @frag: the paged fragment 2340 * @page: the page to set 2341 * 2342 * Sets the fragment @frag to contain @page. 2343 */ 2344static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2345{ 2346 frag->page.p = page; 2347} 2348 2349/** 2350 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2351 * @skb: the buffer 2352 * @f: the fragment offset 2353 * @page: the page to set 2354 * 2355 * Sets the @f'th fragment of @skb to contain @page. 2356 */ 2357static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2358 struct page *page) 2359{ 2360 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2361} 2362 2363bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2364 2365/** 2366 * skb_frag_dma_map - maps a paged fragment via the DMA API 2367 * @dev: the device to map the fragment to 2368 * @frag: the paged fragment to map 2369 * @offset: the offset within the fragment (starting at the 2370 * fragment's own offset) 2371 * @size: the number of bytes to map 2372 * @dir: the direction of the mapping (%PCI_DMA_*) 2373 * 2374 * Maps the page associated with @frag to @device. 2375 */ 2376static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2377 const skb_frag_t *frag, 2378 size_t offset, size_t size, 2379 enum dma_data_direction dir) 2380{ 2381 return dma_map_page(dev, skb_frag_page(frag), 2382 frag->page_offset + offset, size, dir); 2383} 2384 2385static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2386 gfp_t gfp_mask) 2387{ 2388 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2389} 2390 2391 2392static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2393 gfp_t gfp_mask) 2394{ 2395 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2396} 2397 2398 2399/** 2400 * skb_clone_writable - is the header of a clone writable 2401 * @skb: buffer to check 2402 * @len: length up to which to write 2403 * 2404 * Returns true if modifying the header part of the cloned buffer 2405 * does not requires the data to be copied. 2406 */ 2407static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2408{ 2409 return !skb_header_cloned(skb) && 2410 skb_headroom(skb) + len <= skb->hdr_len; 2411} 2412 2413static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2414 int cloned) 2415{ 2416 int delta = 0; 2417 2418 if (headroom > skb_headroom(skb)) 2419 delta = headroom - skb_headroom(skb); 2420 2421 if (delta || cloned) 2422 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2423 GFP_ATOMIC); 2424 return 0; 2425} 2426 2427/** 2428 * skb_cow - copy header of skb when it is required 2429 * @skb: buffer to cow 2430 * @headroom: needed headroom 2431 * 2432 * If the skb passed lacks sufficient headroom or its data part 2433 * is shared, data is reallocated. If reallocation fails, an error 2434 * is returned and original skb is not changed. 2435 * 2436 * The result is skb with writable area skb->head...skb->tail 2437 * and at least @headroom of space at head. 2438 */ 2439static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2440{ 2441 return __skb_cow(skb, headroom, skb_cloned(skb)); 2442} 2443 2444/** 2445 * skb_cow_head - skb_cow but only making the head writable 2446 * @skb: buffer to cow 2447 * @headroom: needed headroom 2448 * 2449 * This function is identical to skb_cow except that we replace the 2450 * skb_cloned check by skb_header_cloned. It should be used when 2451 * you only need to push on some header and do not need to modify 2452 * the data. 2453 */ 2454static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2455{ 2456 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2457} 2458 2459/** 2460 * skb_padto - pad an skbuff up to a minimal size 2461 * @skb: buffer to pad 2462 * @len: minimal length 2463 * 2464 * Pads up a buffer to ensure the trailing bytes exist and are 2465 * blanked. If the buffer already contains sufficient data it 2466 * is untouched. Otherwise it is extended. Returns zero on 2467 * success. The skb is freed on error. 2468 */ 2469static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2470{ 2471 unsigned int size = skb->len; 2472 if (likely(size >= len)) 2473 return 0; 2474 return skb_pad(skb, len - size); 2475} 2476 2477/** 2478 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2479 * @skb: buffer to pad 2480 * @len: minimal length 2481 * 2482 * Pads up a buffer to ensure the trailing bytes exist and are 2483 * blanked. If the buffer already contains sufficient data it 2484 * is untouched. Otherwise it is extended. Returns zero on 2485 * success. The skb is freed on error. 2486 */ 2487static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2488{ 2489 unsigned int size = skb->len; 2490 2491 if (unlikely(size < len)) { 2492 len -= size; 2493 if (skb_pad(skb, len)) 2494 return -ENOMEM; 2495 __skb_put(skb, len); 2496 } 2497 return 0; 2498} 2499 2500static inline int skb_add_data(struct sk_buff *skb, 2501 struct iov_iter *from, int copy) 2502{ 2503 const int off = skb->len; 2504 2505 if (skb->ip_summed == CHECKSUM_NONE) { 2506 __wsum csum = 0; 2507 if (csum_and_copy_from_iter(skb_put(skb, copy), copy, 2508 &csum, from) == copy) { 2509 skb->csum = csum_block_add(skb->csum, csum, off); 2510 return 0; 2511 } 2512 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy) 2513 return 0; 2514 2515 __skb_trim(skb, off); 2516 return -EFAULT; 2517} 2518 2519static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2520 const struct page *page, int off) 2521{ 2522 if (i) { 2523 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2524 2525 return page == skb_frag_page(frag) && 2526 off == frag->page_offset + skb_frag_size(frag); 2527 } 2528 return false; 2529} 2530 2531static inline int __skb_linearize(struct sk_buff *skb) 2532{ 2533 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2534} 2535 2536/** 2537 * skb_linearize - convert paged skb to linear one 2538 * @skb: buffer to linarize 2539 * 2540 * If there is no free memory -ENOMEM is returned, otherwise zero 2541 * is returned and the old skb data released. 2542 */ 2543static inline int skb_linearize(struct sk_buff *skb) 2544{ 2545 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2546} 2547 2548/** 2549 * skb_has_shared_frag - can any frag be overwritten 2550 * @skb: buffer to test 2551 * 2552 * Return true if the skb has at least one frag that might be modified 2553 * by an external entity (as in vmsplice()/sendfile()) 2554 */ 2555static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2556{ 2557 return skb_is_nonlinear(skb) && 2558 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2559} 2560 2561/** 2562 * skb_linearize_cow - make sure skb is linear and writable 2563 * @skb: buffer to process 2564 * 2565 * If there is no free memory -ENOMEM is returned, otherwise zero 2566 * is returned and the old skb data released. 2567 */ 2568static inline int skb_linearize_cow(struct sk_buff *skb) 2569{ 2570 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2571 __skb_linearize(skb) : 0; 2572} 2573 2574/** 2575 * skb_postpull_rcsum - update checksum for received skb after pull 2576 * @skb: buffer to update 2577 * @start: start of data before pull 2578 * @len: length of data pulled 2579 * 2580 * After doing a pull on a received packet, you need to call this to 2581 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2582 * CHECKSUM_NONE so that it can be recomputed from scratch. 2583 */ 2584 2585static inline void skb_postpull_rcsum(struct sk_buff *skb, 2586 const void *start, unsigned int len) 2587{ 2588 if (skb->ip_summed == CHECKSUM_COMPLETE) 2589 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0)); 2590} 2591 2592unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2593 2594/** 2595 * pskb_trim_rcsum - trim received skb and update checksum 2596 * @skb: buffer to trim 2597 * @len: new length 2598 * 2599 * This is exactly the same as pskb_trim except that it ensures the 2600 * checksum of received packets are still valid after the operation. 2601 */ 2602 2603static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2604{ 2605 if (likely(len >= skb->len)) 2606 return 0; 2607 if (skb->ip_summed == CHECKSUM_COMPLETE) 2608 skb->ip_summed = CHECKSUM_NONE; 2609 return __pskb_trim(skb, len); 2610} 2611 2612#define skb_queue_walk(queue, skb) \ 2613 for (skb = (queue)->next; \ 2614 skb != (struct sk_buff *)(queue); \ 2615 skb = skb->next) 2616 2617#define skb_queue_walk_safe(queue, skb, tmp) \ 2618 for (skb = (queue)->next, tmp = skb->next; \ 2619 skb != (struct sk_buff *)(queue); \ 2620 skb = tmp, tmp = skb->next) 2621 2622#define skb_queue_walk_from(queue, skb) \ 2623 for (; skb != (struct sk_buff *)(queue); \ 2624 skb = skb->next) 2625 2626#define skb_queue_walk_from_safe(queue, skb, tmp) \ 2627 for (tmp = skb->next; \ 2628 skb != (struct sk_buff *)(queue); \ 2629 skb = tmp, tmp = skb->next) 2630 2631#define skb_queue_reverse_walk(queue, skb) \ 2632 for (skb = (queue)->prev; \ 2633 skb != (struct sk_buff *)(queue); \ 2634 skb = skb->prev) 2635 2636#define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 2637 for (skb = (queue)->prev, tmp = skb->prev; \ 2638 skb != (struct sk_buff *)(queue); \ 2639 skb = tmp, tmp = skb->prev) 2640 2641#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 2642 for (tmp = skb->prev; \ 2643 skb != (struct sk_buff *)(queue); \ 2644 skb = tmp, tmp = skb->prev) 2645 2646static inline bool skb_has_frag_list(const struct sk_buff *skb) 2647{ 2648 return skb_shinfo(skb)->frag_list != NULL; 2649} 2650 2651static inline void skb_frag_list_init(struct sk_buff *skb) 2652{ 2653 skb_shinfo(skb)->frag_list = NULL; 2654} 2655 2656static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag) 2657{ 2658 frag->next = skb_shinfo(skb)->frag_list; 2659 skb_shinfo(skb)->frag_list = frag; 2660} 2661 2662#define skb_walk_frags(skb, iter) \ 2663 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 2664 2665struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 2666 int *peeked, int *off, int *err); 2667struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 2668 int *err); 2669unsigned int datagram_poll(struct file *file, struct socket *sock, 2670 struct poll_table_struct *wait); 2671int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 2672 struct iov_iter *to, int size); 2673static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 2674 struct msghdr *msg, int size) 2675{ 2676 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 2677} 2678int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 2679 struct msghdr *msg); 2680int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 2681 struct iov_iter *from, int len); 2682int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 2683void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 2684void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb); 2685int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 2686int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 2687int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 2688__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 2689 int len, __wsum csum); 2690int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 2691 struct pipe_inode_info *pipe, unsigned int len, 2692 unsigned int flags); 2693void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 2694unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 2695int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 2696 int len, int hlen); 2697void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 2698int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 2699void skb_scrub_packet(struct sk_buff *skb, bool xnet); 2700unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 2701struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 2702struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 2703int skb_ensure_writable(struct sk_buff *skb, int write_len); 2704int skb_vlan_pop(struct sk_buff *skb); 2705int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 2706 2707static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 2708{ 2709 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2710} 2711 2712static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 2713{ 2714 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2715} 2716 2717struct skb_checksum_ops { 2718 __wsum (*update)(const void *mem, int len, __wsum wsum); 2719 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 2720}; 2721 2722__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 2723 __wsum csum, const struct skb_checksum_ops *ops); 2724__wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 2725 __wsum csum); 2726 2727static inline void *__skb_header_pointer(const struct sk_buff *skb, int offset, 2728 int len, void *data, int hlen, void *buffer) 2729{ 2730 if (hlen - offset >= len) 2731 return data + offset; 2732 2733 if (!skb || 2734 skb_copy_bits(skb, offset, buffer, len) < 0) 2735 return NULL; 2736 2737 return buffer; 2738} 2739 2740static inline void *skb_header_pointer(const struct sk_buff *skb, int offset, 2741 int len, void *buffer) 2742{ 2743 return __skb_header_pointer(skb, offset, len, skb->data, 2744 skb_headlen(skb), buffer); 2745} 2746 2747/** 2748 * skb_needs_linearize - check if we need to linearize a given skb 2749 * depending on the given device features. 2750 * @skb: socket buffer to check 2751 * @features: net device features 2752 * 2753 * Returns true if either: 2754 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 2755 * 2. skb is fragmented and the device does not support SG. 2756 */ 2757static inline bool skb_needs_linearize(struct sk_buff *skb, 2758 netdev_features_t features) 2759{ 2760 return skb_is_nonlinear(skb) && 2761 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 2762 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 2763} 2764 2765static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 2766 void *to, 2767 const unsigned int len) 2768{ 2769 memcpy(to, skb->data, len); 2770} 2771 2772static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 2773 const int offset, void *to, 2774 const unsigned int len) 2775{ 2776 memcpy(to, skb->data + offset, len); 2777} 2778 2779static inline void skb_copy_to_linear_data(struct sk_buff *skb, 2780 const void *from, 2781 const unsigned int len) 2782{ 2783 memcpy(skb->data, from, len); 2784} 2785 2786static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 2787 const int offset, 2788 const void *from, 2789 const unsigned int len) 2790{ 2791 memcpy(skb->data + offset, from, len); 2792} 2793 2794void skb_init(void); 2795 2796static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 2797{ 2798 return skb->tstamp; 2799} 2800 2801/** 2802 * skb_get_timestamp - get timestamp from a skb 2803 * @skb: skb to get stamp from 2804 * @stamp: pointer to struct timeval to store stamp in 2805 * 2806 * Timestamps are stored in the skb as offsets to a base timestamp. 2807 * This function converts the offset back to a struct timeval and stores 2808 * it in stamp. 2809 */ 2810static inline void skb_get_timestamp(const struct sk_buff *skb, 2811 struct timeval *stamp) 2812{ 2813 *stamp = ktime_to_timeval(skb->tstamp); 2814} 2815 2816static inline void skb_get_timestampns(const struct sk_buff *skb, 2817 struct timespec *stamp) 2818{ 2819 *stamp = ktime_to_timespec(skb->tstamp); 2820} 2821 2822static inline void __net_timestamp(struct sk_buff *skb) 2823{ 2824 skb->tstamp = ktime_get_real(); 2825} 2826 2827static inline ktime_t net_timedelta(ktime_t t) 2828{ 2829 return ktime_sub(ktime_get_real(), t); 2830} 2831 2832static inline ktime_t net_invalid_timestamp(void) 2833{ 2834 return ktime_set(0, 0); 2835} 2836 2837struct sk_buff *skb_clone_sk(struct sk_buff *skb); 2838 2839#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 2840 2841void skb_clone_tx_timestamp(struct sk_buff *skb); 2842bool skb_defer_rx_timestamp(struct sk_buff *skb); 2843 2844#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 2845 2846static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 2847{ 2848} 2849 2850static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 2851{ 2852 return false; 2853} 2854 2855#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 2856 2857/** 2858 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 2859 * 2860 * PHY drivers may accept clones of transmitted packets for 2861 * timestamping via their phy_driver.txtstamp method. These drivers 2862 * must call this function to return the skb back to the stack, with 2863 * or without a timestamp. 2864 * 2865 * @skb: clone of the the original outgoing packet 2866 * @hwtstamps: hardware time stamps, may be NULL if not available 2867 * 2868 */ 2869void skb_complete_tx_timestamp(struct sk_buff *skb, 2870 struct skb_shared_hwtstamps *hwtstamps); 2871 2872void __skb_tstamp_tx(struct sk_buff *orig_skb, 2873 struct skb_shared_hwtstamps *hwtstamps, 2874 struct sock *sk, int tstype); 2875 2876/** 2877 * skb_tstamp_tx - queue clone of skb with send time stamps 2878 * @orig_skb: the original outgoing packet 2879 * @hwtstamps: hardware time stamps, may be NULL if not available 2880 * 2881 * If the skb has a socket associated, then this function clones the 2882 * skb (thus sharing the actual data and optional structures), stores 2883 * the optional hardware time stamping information (if non NULL) or 2884 * generates a software time stamp (otherwise), then queues the clone 2885 * to the error queue of the socket. Errors are silently ignored. 2886 */ 2887void skb_tstamp_tx(struct sk_buff *orig_skb, 2888 struct skb_shared_hwtstamps *hwtstamps); 2889 2890static inline void sw_tx_timestamp(struct sk_buff *skb) 2891{ 2892 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 2893 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 2894 skb_tstamp_tx(skb, NULL); 2895} 2896 2897/** 2898 * skb_tx_timestamp() - Driver hook for transmit timestamping 2899 * 2900 * Ethernet MAC Drivers should call this function in their hard_xmit() 2901 * function immediately before giving the sk_buff to the MAC hardware. 2902 * 2903 * Specifically, one should make absolutely sure that this function is 2904 * called before TX completion of this packet can trigger. Otherwise 2905 * the packet could potentially already be freed. 2906 * 2907 * @skb: A socket buffer. 2908 */ 2909static inline void skb_tx_timestamp(struct sk_buff *skb) 2910{ 2911 skb_clone_tx_timestamp(skb); 2912 sw_tx_timestamp(skb); 2913} 2914 2915/** 2916 * skb_complete_wifi_ack - deliver skb with wifi status 2917 * 2918 * @skb: the original outgoing packet 2919 * @acked: ack status 2920 * 2921 */ 2922void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 2923 2924__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 2925__sum16 __skb_checksum_complete(struct sk_buff *skb); 2926 2927static inline int skb_csum_unnecessary(const struct sk_buff *skb) 2928{ 2929 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 2930 skb->csum_valid || 2931 (skb->ip_summed == CHECKSUM_PARTIAL && 2932 skb_checksum_start_offset(skb) >= 0)); 2933} 2934 2935/** 2936 * skb_checksum_complete - Calculate checksum of an entire packet 2937 * @skb: packet to process 2938 * 2939 * This function calculates the checksum over the entire packet plus 2940 * the value of skb->csum. The latter can be used to supply the 2941 * checksum of a pseudo header as used by TCP/UDP. It returns the 2942 * checksum. 2943 * 2944 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 2945 * this function can be used to verify that checksum on received 2946 * packets. In that case the function should return zero if the 2947 * checksum is correct. In particular, this function will return zero 2948 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 2949 * hardware has already verified the correctness of the checksum. 2950 */ 2951static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 2952{ 2953 return skb_csum_unnecessary(skb) ? 2954 0 : __skb_checksum_complete(skb); 2955} 2956 2957static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 2958{ 2959 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2960 if (skb->csum_level == 0) 2961 skb->ip_summed = CHECKSUM_NONE; 2962 else 2963 skb->csum_level--; 2964 } 2965} 2966 2967static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 2968{ 2969 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2970 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 2971 skb->csum_level++; 2972 } else if (skb->ip_summed == CHECKSUM_NONE) { 2973 skb->ip_summed = CHECKSUM_UNNECESSARY; 2974 skb->csum_level = 0; 2975 } 2976} 2977 2978static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 2979{ 2980 /* Mark current checksum as bad (typically called from GRO 2981 * path). In the case that ip_summed is CHECKSUM_NONE 2982 * this must be the first checksum encountered in the packet. 2983 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 2984 * checksum after the last one validated. For UDP, a zero 2985 * checksum can not be marked as bad. 2986 */ 2987 2988 if (skb->ip_summed == CHECKSUM_NONE || 2989 skb->ip_summed == CHECKSUM_UNNECESSARY) 2990 skb->csum_bad = 1; 2991} 2992 2993/* Check if we need to perform checksum complete validation. 2994 * 2995 * Returns true if checksum complete is needed, false otherwise 2996 * (either checksum is unnecessary or zero checksum is allowed). 2997 */ 2998static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 2999 bool zero_okay, 3000 __sum16 check) 3001{ 3002 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3003 skb->csum_valid = 1; 3004 __skb_decr_checksum_unnecessary(skb); 3005 return false; 3006 } 3007 3008 return true; 3009} 3010 3011/* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3012 * in checksum_init. 3013 */ 3014#define CHECKSUM_BREAK 76 3015 3016/* Validate (init) checksum based on checksum complete. 3017 * 3018 * Return values: 3019 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3020 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3021 * checksum is stored in skb->csum for use in __skb_checksum_complete 3022 * non-zero: value of invalid checksum 3023 * 3024 */ 3025static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3026 bool complete, 3027 __wsum psum) 3028{ 3029 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3030 if (!csum_fold(csum_add(psum, skb->csum))) { 3031 skb->csum_valid = 1; 3032 return 0; 3033 } 3034 } else if (skb->csum_bad) { 3035 /* ip_summed == CHECKSUM_NONE in this case */ 3036 return 1; 3037 } 3038 3039 skb->csum = psum; 3040 3041 if (complete || skb->len <= CHECKSUM_BREAK) { 3042 __sum16 csum; 3043 3044 csum = __skb_checksum_complete(skb); 3045 skb->csum_valid = !csum; 3046 return csum; 3047 } 3048 3049 return 0; 3050} 3051 3052static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3053{ 3054 return 0; 3055} 3056 3057/* Perform checksum validate (init). Note that this is a macro since we only 3058 * want to calculate the pseudo header which is an input function if necessary. 3059 * First we try to validate without any computation (checksum unnecessary) and 3060 * then calculate based on checksum complete calling the function to compute 3061 * pseudo header. 3062 * 3063 * Return values: 3064 * 0: checksum is validated or try to in skb_checksum_complete 3065 * non-zero: value of invalid checksum 3066 */ 3067#define __skb_checksum_validate(skb, proto, complete, \ 3068 zero_okay, check, compute_pseudo) \ 3069({ \ 3070 __sum16 __ret = 0; \ 3071 skb->csum_valid = 0; \ 3072 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3073 __ret = __skb_checksum_validate_complete(skb, \ 3074 complete, compute_pseudo(skb, proto)); \ 3075 __ret; \ 3076}) 3077 3078#define skb_checksum_init(skb, proto, compute_pseudo) \ 3079 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3080 3081#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3082 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3083 3084#define skb_checksum_validate(skb, proto, compute_pseudo) \ 3085 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3086 3087#define skb_checksum_validate_zero_check(skb, proto, check, \ 3088 compute_pseudo) \ 3089 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3090 3091#define skb_checksum_simple_validate(skb) \ 3092 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3093 3094static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3095{ 3096 return (skb->ip_summed == CHECKSUM_NONE && 3097 skb->csum_valid && !skb->csum_bad); 3098} 3099 3100static inline void __skb_checksum_convert(struct sk_buff *skb, 3101 __sum16 check, __wsum pseudo) 3102{ 3103 skb->csum = ~pseudo; 3104 skb->ip_summed = CHECKSUM_COMPLETE; 3105} 3106 3107#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3108do { \ 3109 if (__skb_checksum_convert_check(skb)) \ 3110 __skb_checksum_convert(skb, check, \ 3111 compute_pseudo(skb, proto)); \ 3112} while (0) 3113 3114static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3115 u16 start, u16 offset) 3116{ 3117 skb->ip_summed = CHECKSUM_PARTIAL; 3118 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3119 skb->csum_offset = offset - start; 3120} 3121 3122/* Update skbuf and packet to reflect the remote checksum offload operation. 3123 * When called, ptr indicates the starting point for skb->csum when 3124 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3125 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3126 */ 3127static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3128 int start, int offset, bool nopartial) 3129{ 3130 __wsum delta; 3131 3132 if (!nopartial) { 3133 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3134 return; 3135 } 3136 3137 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3138 __skb_checksum_complete(skb); 3139 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3140 } 3141 3142 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3143 3144 /* Adjust skb->csum since we changed the packet */ 3145 skb->csum = csum_add(skb->csum, delta); 3146} 3147 3148#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3149void nf_conntrack_destroy(struct nf_conntrack *nfct); 3150static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3151{ 3152 if (nfct && atomic_dec_and_test(&nfct->use)) 3153 nf_conntrack_destroy(nfct); 3154} 3155static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3156{ 3157 if (nfct) 3158 atomic_inc(&nfct->use); 3159} 3160#endif 3161#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3162static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3163{ 3164 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3165 kfree(nf_bridge); 3166} 3167static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3168{ 3169 if (nf_bridge) 3170 atomic_inc(&nf_bridge->use); 3171} 3172#endif /* CONFIG_BRIDGE_NETFILTER */ 3173static inline void nf_reset(struct sk_buff *skb) 3174{ 3175#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3176 nf_conntrack_put(skb->nfct); 3177 skb->nfct = NULL; 3178#endif 3179#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3180 nf_bridge_put(skb->nf_bridge); 3181 skb->nf_bridge = NULL; 3182#endif 3183} 3184 3185static inline void nf_reset_trace(struct sk_buff *skb) 3186{ 3187#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3188 skb->nf_trace = 0; 3189#endif 3190} 3191 3192/* Note: This doesn't put any conntrack and bridge info in dst. */ 3193static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3194 bool copy) 3195{ 3196#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3197 dst->nfct = src->nfct; 3198 nf_conntrack_get(src->nfct); 3199 if (copy) 3200 dst->nfctinfo = src->nfctinfo; 3201#endif 3202#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3203 dst->nf_bridge = src->nf_bridge; 3204 nf_bridge_get(src->nf_bridge); 3205#endif 3206#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3207 if (copy) 3208 dst->nf_trace = src->nf_trace; 3209#endif 3210} 3211 3212static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3213{ 3214#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3215 nf_conntrack_put(dst->nfct); 3216#endif 3217#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3218 nf_bridge_put(dst->nf_bridge); 3219#endif 3220 __nf_copy(dst, src, true); 3221} 3222 3223#ifdef CONFIG_NETWORK_SECMARK 3224static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3225{ 3226 to->secmark = from->secmark; 3227} 3228 3229static inline void skb_init_secmark(struct sk_buff *skb) 3230{ 3231 skb->secmark = 0; 3232} 3233#else 3234static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3235{ } 3236 3237static inline void skb_init_secmark(struct sk_buff *skb) 3238{ } 3239#endif 3240 3241static inline bool skb_irq_freeable(const struct sk_buff *skb) 3242{ 3243 return !skb->destructor && 3244#if IS_ENABLED(CONFIG_XFRM) 3245 !skb->sp && 3246#endif 3247#if IS_ENABLED(CONFIG_NF_CONNTRACK) 3248 !skb->nfct && 3249#endif 3250 !skb->_skb_refdst && 3251 !skb_has_frag_list(skb); 3252} 3253 3254static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3255{ 3256 skb->queue_mapping = queue_mapping; 3257} 3258 3259static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3260{ 3261 return skb->queue_mapping; 3262} 3263 3264static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3265{ 3266 to->queue_mapping = from->queue_mapping; 3267} 3268 3269static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3270{ 3271 skb->queue_mapping = rx_queue + 1; 3272} 3273 3274static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3275{ 3276 return skb->queue_mapping - 1; 3277} 3278 3279static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3280{ 3281 return skb->queue_mapping != 0; 3282} 3283 3284u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, 3285 unsigned int num_tx_queues); 3286 3287static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3288{ 3289#ifdef CONFIG_XFRM 3290 return skb->sp; 3291#else 3292 return NULL; 3293#endif 3294} 3295 3296/* Keeps track of mac header offset relative to skb->head. 3297 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3298 * For non-tunnel skb it points to skb_mac_header() and for 3299 * tunnel skb it points to outer mac header. 3300 * Keeps track of level of encapsulation of network headers. 3301 */ 3302struct skb_gso_cb { 3303 int mac_offset; 3304 int encap_level; 3305 __u16 csum_start; 3306}; 3307#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb) 3308 3309static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3310{ 3311 return (skb_mac_header(inner_skb) - inner_skb->head) - 3312 SKB_GSO_CB(inner_skb)->mac_offset; 3313} 3314 3315static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3316{ 3317 int new_headroom, headroom; 3318 int ret; 3319 3320 headroom = skb_headroom(skb); 3321 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3322 if (ret) 3323 return ret; 3324 3325 new_headroom = skb_headroom(skb); 3326 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3327 return 0; 3328} 3329 3330/* Compute the checksum for a gso segment. First compute the checksum value 3331 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3332 * then add in skb->csum (checksum from csum_start to end of packet). 3333 * skb->csum and csum_start are then updated to reflect the checksum of the 3334 * resultant packet starting from the transport header-- the resultant checksum 3335 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3336 * header. 3337 */ 3338static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3339{ 3340 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) - 3341 skb_transport_offset(skb); 3342 __u16 csum; 3343 3344 csum = csum_fold(csum_partial(skb_transport_header(skb), 3345 plen, skb->csum)); 3346 skb->csum = res; 3347 SKB_GSO_CB(skb)->csum_start -= plen; 3348 3349 return csum; 3350} 3351 3352static inline bool skb_is_gso(const struct sk_buff *skb) 3353{ 3354 return skb_shinfo(skb)->gso_size; 3355} 3356 3357/* Note: Should be called only if skb_is_gso(skb) is true */ 3358static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3359{ 3360 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3361} 3362 3363void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3364 3365static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3366{ 3367 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3368 * wanted then gso_type will be set. */ 3369 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3370 3371 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3372 unlikely(shinfo->gso_type == 0)) { 3373 __skb_warn_lro_forwarding(skb); 3374 return true; 3375 } 3376 return false; 3377} 3378 3379static inline void skb_forward_csum(struct sk_buff *skb) 3380{ 3381 /* Unfortunately we don't support this one. Any brave souls? */ 3382 if (skb->ip_summed == CHECKSUM_COMPLETE) 3383 skb->ip_summed = CHECKSUM_NONE; 3384} 3385 3386/** 3387 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3388 * @skb: skb to check 3389 * 3390 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3391 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3392 * use this helper, to document places where we make this assertion. 3393 */ 3394static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3395{ 3396#ifdef DEBUG 3397 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3398#endif 3399} 3400 3401bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3402 3403int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3404 3405u32 skb_get_poff(const struct sk_buff *skb); 3406u32 __skb_get_poff(const struct sk_buff *skb, void *data, 3407 const struct flow_keys *keys, int hlen); 3408 3409/** 3410 * skb_head_is_locked - Determine if the skb->head is locked down 3411 * @skb: skb to check 3412 * 3413 * The head on skbs build around a head frag can be removed if they are 3414 * not cloned. This function returns true if the skb head is locked down 3415 * due to either being allocated via kmalloc, or by being a clone with 3416 * multiple references to the head. 3417 */ 3418static inline bool skb_head_is_locked(const struct sk_buff *skb) 3419{ 3420 return !skb->head_frag || skb_cloned(skb); 3421} 3422 3423/** 3424 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3425 * 3426 * @skb: GSO skb 3427 * 3428 * skb_gso_network_seglen is used to determine the real size of the 3429 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3430 * 3431 * The MAC/L2 header is not accounted for. 3432 */ 3433static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3434{ 3435 unsigned int hdr_len = skb_transport_header(skb) - 3436 skb_network_header(skb); 3437 return hdr_len + skb_gso_transport_seglen(skb); 3438} 3439#endif /* __KERNEL__ */ 3440#endif /* _LINUX_SKBUFF_H */