at v4.1 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 enum { 170 BRNF_PROTO_UNCHANGED, 171 BRNF_PROTO_8021Q, 172 BRNF_PROTO_PPPOE 173 } orig_proto; 174 bool pkt_otherhost; 175 unsigned int mask; 176 struct net_device *physindev; 177 struct net_device *physoutdev; 178 char neigh_header[8]; 179 __be32 ipv4_daddr; 180}; 181#endif 182 183struct sk_buff_head { 184 /* These two members must be first. */ 185 struct sk_buff *next; 186 struct sk_buff *prev; 187 188 __u32 qlen; 189 spinlock_t lock; 190}; 191 192struct sk_buff; 193 194/* To allow 64K frame to be packed as single skb without frag_list we 195 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 196 * buffers which do not start on a page boundary. 197 * 198 * Since GRO uses frags we allocate at least 16 regardless of page 199 * size. 200 */ 201#if (65536/PAGE_SIZE + 1) < 16 202#define MAX_SKB_FRAGS 16UL 203#else 204#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 205#endif 206 207typedef struct skb_frag_struct skb_frag_t; 208 209struct skb_frag_struct { 210 struct { 211 struct page *p; 212 } page; 213#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 214 __u32 page_offset; 215 __u32 size; 216#else 217 __u16 page_offset; 218 __u16 size; 219#endif 220}; 221 222static inline unsigned int skb_frag_size(const skb_frag_t *frag) 223{ 224 return frag->size; 225} 226 227static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 228{ 229 frag->size = size; 230} 231 232static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 233{ 234 frag->size += delta; 235} 236 237static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 238{ 239 frag->size -= delta; 240} 241 242#define HAVE_HW_TIME_STAMP 243 244/** 245 * struct skb_shared_hwtstamps - hardware time stamps 246 * @hwtstamp: hardware time stamp transformed into duration 247 * since arbitrary point in time 248 * 249 * Software time stamps generated by ktime_get_real() are stored in 250 * skb->tstamp. 251 * 252 * hwtstamps can only be compared against other hwtstamps from 253 * the same device. 254 * 255 * This structure is attached to packets as part of the 256 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 257 */ 258struct skb_shared_hwtstamps { 259 ktime_t hwtstamp; 260}; 261 262/* Definitions for tx_flags in struct skb_shared_info */ 263enum { 264 /* generate hardware time stamp */ 265 SKBTX_HW_TSTAMP = 1 << 0, 266 267 /* generate software time stamp when queueing packet to NIC */ 268 SKBTX_SW_TSTAMP = 1 << 1, 269 270 /* device driver is going to provide hardware time stamp */ 271 SKBTX_IN_PROGRESS = 1 << 2, 272 273 /* device driver supports TX zero-copy buffers */ 274 SKBTX_DEV_ZEROCOPY = 1 << 3, 275 276 /* generate wifi status information (where possible) */ 277 SKBTX_WIFI_STATUS = 1 << 4, 278 279 /* This indicates at least one fragment might be overwritten 280 * (as in vmsplice(), sendfile() ...) 281 * If we need to compute a TX checksum, we'll need to copy 282 * all frags to avoid possible bad checksum 283 */ 284 SKBTX_SHARED_FRAG = 1 << 5, 285 286 /* generate software time stamp when entering packet scheduling */ 287 SKBTX_SCHED_TSTAMP = 1 << 6, 288 289 /* generate software timestamp on peer data acknowledgment */ 290 SKBTX_ACK_TSTAMP = 1 << 7, 291}; 292 293#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 294 SKBTX_SCHED_TSTAMP | \ 295 SKBTX_ACK_TSTAMP) 296#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 297 298/* 299 * The callback notifies userspace to release buffers when skb DMA is done in 300 * lower device, the skb last reference should be 0 when calling this. 301 * The zerocopy_success argument is true if zero copy transmit occurred, 302 * false on data copy or out of memory error caused by data copy attempt. 303 * The ctx field is used to track device context. 304 * The desc field is used to track userspace buffer index. 305 */ 306struct ubuf_info { 307 void (*callback)(struct ubuf_info *, bool zerocopy_success); 308 void *ctx; 309 unsigned long desc; 310}; 311 312/* This data is invariant across clones and lives at 313 * the end of the header data, ie. at skb->end. 314 */ 315struct skb_shared_info { 316 unsigned char nr_frags; 317 __u8 tx_flags; 318 unsigned short gso_size; 319 /* Warning: this field is not always filled in (UFO)! */ 320 unsigned short gso_segs; 321 unsigned short gso_type; 322 struct sk_buff *frag_list; 323 struct skb_shared_hwtstamps hwtstamps; 324 u32 tskey; 325 __be32 ip6_frag_id; 326 327 /* 328 * Warning : all fields before dataref are cleared in __alloc_skb() 329 */ 330 atomic_t dataref; 331 332 /* Intermediate layers must ensure that destructor_arg 333 * remains valid until skb destructor */ 334 void * destructor_arg; 335 336 /* must be last field, see pskb_expand_head() */ 337 skb_frag_t frags[MAX_SKB_FRAGS]; 338}; 339 340/* We divide dataref into two halves. The higher 16 bits hold references 341 * to the payload part of skb->data. The lower 16 bits hold references to 342 * the entire skb->data. A clone of a headerless skb holds the length of 343 * the header in skb->hdr_len. 344 * 345 * All users must obey the rule that the skb->data reference count must be 346 * greater than or equal to the payload reference count. 347 * 348 * Holding a reference to the payload part means that the user does not 349 * care about modifications to the header part of skb->data. 350 */ 351#define SKB_DATAREF_SHIFT 16 352#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 353 354 355enum { 356 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 357 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 358 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 359}; 360 361enum { 362 SKB_GSO_TCPV4 = 1 << 0, 363 SKB_GSO_UDP = 1 << 1, 364 365 /* This indicates the skb is from an untrusted source. */ 366 SKB_GSO_DODGY = 1 << 2, 367 368 /* This indicates the tcp segment has CWR set. */ 369 SKB_GSO_TCP_ECN = 1 << 3, 370 371 SKB_GSO_TCPV6 = 1 << 4, 372 373 SKB_GSO_FCOE = 1 << 5, 374 375 SKB_GSO_GRE = 1 << 6, 376 377 SKB_GSO_GRE_CSUM = 1 << 7, 378 379 SKB_GSO_IPIP = 1 << 8, 380 381 SKB_GSO_SIT = 1 << 9, 382 383 SKB_GSO_UDP_TUNNEL = 1 << 10, 384 385 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 386 387 SKB_GSO_TUNNEL_REMCSUM = 1 << 12, 388}; 389 390#if BITS_PER_LONG > 32 391#define NET_SKBUFF_DATA_USES_OFFSET 1 392#endif 393 394#ifdef NET_SKBUFF_DATA_USES_OFFSET 395typedef unsigned int sk_buff_data_t; 396#else 397typedef unsigned char *sk_buff_data_t; 398#endif 399 400/** 401 * struct skb_mstamp - multi resolution time stamps 402 * @stamp_us: timestamp in us resolution 403 * @stamp_jiffies: timestamp in jiffies 404 */ 405struct skb_mstamp { 406 union { 407 u64 v64; 408 struct { 409 u32 stamp_us; 410 u32 stamp_jiffies; 411 }; 412 }; 413}; 414 415/** 416 * skb_mstamp_get - get current timestamp 417 * @cl: place to store timestamps 418 */ 419static inline void skb_mstamp_get(struct skb_mstamp *cl) 420{ 421 u64 val = local_clock(); 422 423 do_div(val, NSEC_PER_USEC); 424 cl->stamp_us = (u32)val; 425 cl->stamp_jiffies = (u32)jiffies; 426} 427 428/** 429 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp 430 * @t1: pointer to newest sample 431 * @t0: pointer to oldest sample 432 */ 433static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1, 434 const struct skb_mstamp *t0) 435{ 436 s32 delta_us = t1->stamp_us - t0->stamp_us; 437 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies; 438 439 /* If delta_us is negative, this might be because interval is too big, 440 * or local_clock() drift is too big : fallback using jiffies. 441 */ 442 if (delta_us <= 0 || 443 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ))) 444 445 delta_us = jiffies_to_usecs(delta_jiffies); 446 447 return delta_us; 448} 449 450 451/** 452 * struct sk_buff - socket buffer 453 * @next: Next buffer in list 454 * @prev: Previous buffer in list 455 * @tstamp: Time we arrived/left 456 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 457 * @sk: Socket we are owned by 458 * @dev: Device we arrived on/are leaving by 459 * @cb: Control buffer. Free for use by every layer. Put private vars here 460 * @_skb_refdst: destination entry (with norefcount bit) 461 * @sp: the security path, used for xfrm 462 * @len: Length of actual data 463 * @data_len: Data length 464 * @mac_len: Length of link layer header 465 * @hdr_len: writable header length of cloned skb 466 * @csum: Checksum (must include start/offset pair) 467 * @csum_start: Offset from skb->head where checksumming should start 468 * @csum_offset: Offset from csum_start where checksum should be stored 469 * @priority: Packet queueing priority 470 * @ignore_df: allow local fragmentation 471 * @cloned: Head may be cloned (check refcnt to be sure) 472 * @ip_summed: Driver fed us an IP checksum 473 * @nohdr: Payload reference only, must not modify header 474 * @nfctinfo: Relationship of this skb to the connection 475 * @pkt_type: Packet class 476 * @fclone: skbuff clone status 477 * @ipvs_property: skbuff is owned by ipvs 478 * @peeked: this packet has been seen already, so stats have been 479 * done for it, don't do them again 480 * @nf_trace: netfilter packet trace flag 481 * @protocol: Packet protocol from driver 482 * @destructor: Destruct function 483 * @nfct: Associated connection, if any 484 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 485 * @skb_iif: ifindex of device we arrived on 486 * @tc_index: Traffic control index 487 * @tc_verd: traffic control verdict 488 * @hash: the packet hash 489 * @queue_mapping: Queue mapping for multiqueue devices 490 * @xmit_more: More SKBs are pending for this queue 491 * @ndisc_nodetype: router type (from link layer) 492 * @ooo_okay: allow the mapping of a socket to a queue to be changed 493 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 494 * ports. 495 * @sw_hash: indicates hash was computed in software stack 496 * @wifi_acked_valid: wifi_acked was set 497 * @wifi_acked: whether frame was acked on wifi or not 498 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 499 * @napi_id: id of the NAPI struct this skb came from 500 * @secmark: security marking 501 * @mark: Generic packet mark 502 * @vlan_proto: vlan encapsulation protocol 503 * @vlan_tci: vlan tag control information 504 * @inner_protocol: Protocol (encapsulation) 505 * @inner_transport_header: Inner transport layer header (encapsulation) 506 * @inner_network_header: Network layer header (encapsulation) 507 * @inner_mac_header: Link layer header (encapsulation) 508 * @transport_header: Transport layer header 509 * @network_header: Network layer header 510 * @mac_header: Link layer header 511 * @tail: Tail pointer 512 * @end: End pointer 513 * @head: Head of buffer 514 * @data: Data head pointer 515 * @truesize: Buffer size 516 * @users: User count - see {datagram,tcp}.c 517 */ 518 519struct sk_buff { 520 union { 521 struct { 522 /* These two members must be first. */ 523 struct sk_buff *next; 524 struct sk_buff *prev; 525 526 union { 527 ktime_t tstamp; 528 struct skb_mstamp skb_mstamp; 529 }; 530 }; 531 struct rb_node rbnode; /* used in netem & tcp stack */ 532 }; 533 struct sock *sk; 534 struct net_device *dev; 535 536 /* 537 * This is the control buffer. It is free to use for every 538 * layer. Please put your private variables there. If you 539 * want to keep them across layers you have to do a skb_clone() 540 * first. This is owned by whoever has the skb queued ATM. 541 */ 542 char cb[48] __aligned(8); 543 544 unsigned long _skb_refdst; 545 void (*destructor)(struct sk_buff *skb); 546#ifdef CONFIG_XFRM 547 struct sec_path *sp; 548#endif 549#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 550 struct nf_conntrack *nfct; 551#endif 552#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 553 struct nf_bridge_info *nf_bridge; 554#endif 555 unsigned int len, 556 data_len; 557 __u16 mac_len, 558 hdr_len; 559 560 /* Following fields are _not_ copied in __copy_skb_header() 561 * Note that queue_mapping is here mostly to fill a hole. 562 */ 563 kmemcheck_bitfield_begin(flags1); 564 __u16 queue_mapping; 565 __u8 cloned:1, 566 nohdr:1, 567 fclone:2, 568 peeked:1, 569 head_frag:1, 570 xmit_more:1; 571 /* one bit hole */ 572 kmemcheck_bitfield_end(flags1); 573 574 /* fields enclosed in headers_start/headers_end are copied 575 * using a single memcpy() in __copy_skb_header() 576 */ 577 /* private: */ 578 __u32 headers_start[0]; 579 /* public: */ 580 581/* if you move pkt_type around you also must adapt those constants */ 582#ifdef __BIG_ENDIAN_BITFIELD 583#define PKT_TYPE_MAX (7 << 5) 584#else 585#define PKT_TYPE_MAX 7 586#endif 587#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 588 589 __u8 __pkt_type_offset[0]; 590 __u8 pkt_type:3; 591 __u8 pfmemalloc:1; 592 __u8 ignore_df:1; 593 __u8 nfctinfo:3; 594 595 __u8 nf_trace:1; 596 __u8 ip_summed:2; 597 __u8 ooo_okay:1; 598 __u8 l4_hash:1; 599 __u8 sw_hash:1; 600 __u8 wifi_acked_valid:1; 601 __u8 wifi_acked:1; 602 603 __u8 no_fcs:1; 604 /* Indicates the inner headers are valid in the skbuff. */ 605 __u8 encapsulation:1; 606 __u8 encap_hdr_csum:1; 607 __u8 csum_valid:1; 608 __u8 csum_complete_sw:1; 609 __u8 csum_level:2; 610 __u8 csum_bad:1; 611 612#ifdef CONFIG_IPV6_NDISC_NODETYPE 613 __u8 ndisc_nodetype:2; 614#endif 615 __u8 ipvs_property:1; 616 __u8 inner_protocol_type:1; 617 __u8 remcsum_offload:1; 618 /* 3 or 5 bit hole */ 619 620#ifdef CONFIG_NET_SCHED 621 __u16 tc_index; /* traffic control index */ 622#ifdef CONFIG_NET_CLS_ACT 623 __u16 tc_verd; /* traffic control verdict */ 624#endif 625#endif 626 627 union { 628 __wsum csum; 629 struct { 630 __u16 csum_start; 631 __u16 csum_offset; 632 }; 633 }; 634 __u32 priority; 635 int skb_iif; 636 __u32 hash; 637 __be16 vlan_proto; 638 __u16 vlan_tci; 639#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 640 union { 641 unsigned int napi_id; 642 unsigned int sender_cpu; 643 }; 644#endif 645#ifdef CONFIG_NETWORK_SECMARK 646 __u32 secmark; 647#endif 648 union { 649 __u32 mark; 650 __u32 reserved_tailroom; 651 }; 652 653 union { 654 __be16 inner_protocol; 655 __u8 inner_ipproto; 656 }; 657 658 __u16 inner_transport_header; 659 __u16 inner_network_header; 660 __u16 inner_mac_header; 661 662 __be16 protocol; 663 __u16 transport_header; 664 __u16 network_header; 665 __u16 mac_header; 666 667 /* private: */ 668 __u32 headers_end[0]; 669 /* public: */ 670 671 /* These elements must be at the end, see alloc_skb() for details. */ 672 sk_buff_data_t tail; 673 sk_buff_data_t end; 674 unsigned char *head, 675 *data; 676 unsigned int truesize; 677 atomic_t users; 678}; 679 680#ifdef __KERNEL__ 681/* 682 * Handling routines are only of interest to the kernel 683 */ 684#include <linux/slab.h> 685 686 687#define SKB_ALLOC_FCLONE 0x01 688#define SKB_ALLOC_RX 0x02 689#define SKB_ALLOC_NAPI 0x04 690 691/* Returns true if the skb was allocated from PFMEMALLOC reserves */ 692static inline bool skb_pfmemalloc(const struct sk_buff *skb) 693{ 694 return unlikely(skb->pfmemalloc); 695} 696 697/* 698 * skb might have a dst pointer attached, refcounted or not. 699 * _skb_refdst low order bit is set if refcount was _not_ taken 700 */ 701#define SKB_DST_NOREF 1UL 702#define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 703 704/** 705 * skb_dst - returns skb dst_entry 706 * @skb: buffer 707 * 708 * Returns skb dst_entry, regardless of reference taken or not. 709 */ 710static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 711{ 712 /* If refdst was not refcounted, check we still are in a 713 * rcu_read_lock section 714 */ 715 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 716 !rcu_read_lock_held() && 717 !rcu_read_lock_bh_held()); 718 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 719} 720 721/** 722 * skb_dst_set - sets skb dst 723 * @skb: buffer 724 * @dst: dst entry 725 * 726 * Sets skb dst, assuming a reference was taken on dst and should 727 * be released by skb_dst_drop() 728 */ 729static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 730{ 731 skb->_skb_refdst = (unsigned long)dst; 732} 733 734/** 735 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 736 * @skb: buffer 737 * @dst: dst entry 738 * 739 * Sets skb dst, assuming a reference was not taken on dst. 740 * If dst entry is cached, we do not take reference and dst_release 741 * will be avoided by refdst_drop. If dst entry is not cached, we take 742 * reference, so that last dst_release can destroy the dst immediately. 743 */ 744static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 745{ 746 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 747 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 748} 749 750/** 751 * skb_dst_is_noref - Test if skb dst isn't refcounted 752 * @skb: buffer 753 */ 754static inline bool skb_dst_is_noref(const struct sk_buff *skb) 755{ 756 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 757} 758 759static inline struct rtable *skb_rtable(const struct sk_buff *skb) 760{ 761 return (struct rtable *)skb_dst(skb); 762} 763 764void kfree_skb(struct sk_buff *skb); 765void kfree_skb_list(struct sk_buff *segs); 766void skb_tx_error(struct sk_buff *skb); 767void consume_skb(struct sk_buff *skb); 768void __kfree_skb(struct sk_buff *skb); 769extern struct kmem_cache *skbuff_head_cache; 770 771void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 772bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 773 bool *fragstolen, int *delta_truesize); 774 775struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 776 int node); 777struct sk_buff *__build_skb(void *data, unsigned int frag_size); 778struct sk_buff *build_skb(void *data, unsigned int frag_size); 779static inline struct sk_buff *alloc_skb(unsigned int size, 780 gfp_t priority) 781{ 782 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 783} 784 785struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 786 unsigned long data_len, 787 int max_page_order, 788 int *errcode, 789 gfp_t gfp_mask); 790 791/* Layout of fast clones : [skb1][skb2][fclone_ref] */ 792struct sk_buff_fclones { 793 struct sk_buff skb1; 794 795 struct sk_buff skb2; 796 797 atomic_t fclone_ref; 798}; 799 800/** 801 * skb_fclone_busy - check if fclone is busy 802 * @skb: buffer 803 * 804 * Returns true is skb is a fast clone, and its clone is not freed. 805 * Some drivers call skb_orphan() in their ndo_start_xmit(), 806 * so we also check that this didnt happen. 807 */ 808static inline bool skb_fclone_busy(const struct sock *sk, 809 const struct sk_buff *skb) 810{ 811 const struct sk_buff_fclones *fclones; 812 813 fclones = container_of(skb, struct sk_buff_fclones, skb1); 814 815 return skb->fclone == SKB_FCLONE_ORIG && 816 atomic_read(&fclones->fclone_ref) > 1 && 817 fclones->skb2.sk == sk; 818} 819 820static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 821 gfp_t priority) 822{ 823 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 824} 825 826struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 827static inline struct sk_buff *alloc_skb_head(gfp_t priority) 828{ 829 return __alloc_skb_head(priority, -1); 830} 831 832struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 833int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 834struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 835struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 836struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 837 gfp_t gfp_mask, bool fclone); 838static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 839 gfp_t gfp_mask) 840{ 841 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 842} 843 844int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 845struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 846 unsigned int headroom); 847struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 848 int newtailroom, gfp_t priority); 849int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 850 int offset, int len); 851int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, 852 int len); 853int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 854int skb_pad(struct sk_buff *skb, int pad); 855#define dev_kfree_skb(a) consume_skb(a) 856 857int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 858 int getfrag(void *from, char *to, int offset, 859 int len, int odd, struct sk_buff *skb), 860 void *from, int length); 861 862struct skb_seq_state { 863 __u32 lower_offset; 864 __u32 upper_offset; 865 __u32 frag_idx; 866 __u32 stepped_offset; 867 struct sk_buff *root_skb; 868 struct sk_buff *cur_skb; 869 __u8 *frag_data; 870}; 871 872void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 873 unsigned int to, struct skb_seq_state *st); 874unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 875 struct skb_seq_state *st); 876void skb_abort_seq_read(struct skb_seq_state *st); 877 878unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 879 unsigned int to, struct ts_config *config); 880 881/* 882 * Packet hash types specify the type of hash in skb_set_hash. 883 * 884 * Hash types refer to the protocol layer addresses which are used to 885 * construct a packet's hash. The hashes are used to differentiate or identify 886 * flows of the protocol layer for the hash type. Hash types are either 887 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 888 * 889 * Properties of hashes: 890 * 891 * 1) Two packets in different flows have different hash values 892 * 2) Two packets in the same flow should have the same hash value 893 * 894 * A hash at a higher layer is considered to be more specific. A driver should 895 * set the most specific hash possible. 896 * 897 * A driver cannot indicate a more specific hash than the layer at which a hash 898 * was computed. For instance an L3 hash cannot be set as an L4 hash. 899 * 900 * A driver may indicate a hash level which is less specific than the 901 * actual layer the hash was computed on. For instance, a hash computed 902 * at L4 may be considered an L3 hash. This should only be done if the 903 * driver can't unambiguously determine that the HW computed the hash at 904 * the higher layer. Note that the "should" in the second property above 905 * permits this. 906 */ 907enum pkt_hash_types { 908 PKT_HASH_TYPE_NONE, /* Undefined type */ 909 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 910 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 911 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 912}; 913 914static inline void 915skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 916{ 917 skb->l4_hash = (type == PKT_HASH_TYPE_L4); 918 skb->sw_hash = 0; 919 skb->hash = hash; 920} 921 922void __skb_get_hash(struct sk_buff *skb); 923static inline __u32 skb_get_hash(struct sk_buff *skb) 924{ 925 if (!skb->l4_hash && !skb->sw_hash) 926 __skb_get_hash(skb); 927 928 return skb->hash; 929} 930 931static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 932{ 933 return skb->hash; 934} 935 936static inline void skb_clear_hash(struct sk_buff *skb) 937{ 938 skb->hash = 0; 939 skb->sw_hash = 0; 940 skb->l4_hash = 0; 941} 942 943static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 944{ 945 if (!skb->l4_hash) 946 skb_clear_hash(skb); 947} 948 949static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 950{ 951 to->hash = from->hash; 952 to->sw_hash = from->sw_hash; 953 to->l4_hash = from->l4_hash; 954}; 955 956static inline void skb_sender_cpu_clear(struct sk_buff *skb) 957{ 958#ifdef CONFIG_XPS 959 skb->sender_cpu = 0; 960#endif 961} 962 963#ifdef NET_SKBUFF_DATA_USES_OFFSET 964static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 965{ 966 return skb->head + skb->end; 967} 968 969static inline unsigned int skb_end_offset(const struct sk_buff *skb) 970{ 971 return skb->end; 972} 973#else 974static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 975{ 976 return skb->end; 977} 978 979static inline unsigned int skb_end_offset(const struct sk_buff *skb) 980{ 981 return skb->end - skb->head; 982} 983#endif 984 985/* Internal */ 986#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 987 988static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 989{ 990 return &skb_shinfo(skb)->hwtstamps; 991} 992 993/** 994 * skb_queue_empty - check if a queue is empty 995 * @list: queue head 996 * 997 * Returns true if the queue is empty, false otherwise. 998 */ 999static inline int skb_queue_empty(const struct sk_buff_head *list) 1000{ 1001 return list->next == (const struct sk_buff *) list; 1002} 1003 1004/** 1005 * skb_queue_is_last - check if skb is the last entry in the queue 1006 * @list: queue head 1007 * @skb: buffer 1008 * 1009 * Returns true if @skb is the last buffer on the list. 1010 */ 1011static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1012 const struct sk_buff *skb) 1013{ 1014 return skb->next == (const struct sk_buff *) list; 1015} 1016 1017/** 1018 * skb_queue_is_first - check if skb is the first entry in the queue 1019 * @list: queue head 1020 * @skb: buffer 1021 * 1022 * Returns true if @skb is the first buffer on the list. 1023 */ 1024static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1025 const struct sk_buff *skb) 1026{ 1027 return skb->prev == (const struct sk_buff *) list; 1028} 1029 1030/** 1031 * skb_queue_next - return the next packet in the queue 1032 * @list: queue head 1033 * @skb: current buffer 1034 * 1035 * Return the next packet in @list after @skb. It is only valid to 1036 * call this if skb_queue_is_last() evaluates to false. 1037 */ 1038static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1039 const struct sk_buff *skb) 1040{ 1041 /* This BUG_ON may seem severe, but if we just return then we 1042 * are going to dereference garbage. 1043 */ 1044 BUG_ON(skb_queue_is_last(list, skb)); 1045 return skb->next; 1046} 1047 1048/** 1049 * skb_queue_prev - return the prev packet in the queue 1050 * @list: queue head 1051 * @skb: current buffer 1052 * 1053 * Return the prev packet in @list before @skb. It is only valid to 1054 * call this if skb_queue_is_first() evaluates to false. 1055 */ 1056static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1057 const struct sk_buff *skb) 1058{ 1059 /* This BUG_ON may seem severe, but if we just return then we 1060 * are going to dereference garbage. 1061 */ 1062 BUG_ON(skb_queue_is_first(list, skb)); 1063 return skb->prev; 1064} 1065 1066/** 1067 * skb_get - reference buffer 1068 * @skb: buffer to reference 1069 * 1070 * Makes another reference to a socket buffer and returns a pointer 1071 * to the buffer. 1072 */ 1073static inline struct sk_buff *skb_get(struct sk_buff *skb) 1074{ 1075 atomic_inc(&skb->users); 1076 return skb; 1077} 1078 1079/* 1080 * If users == 1, we are the only owner and are can avoid redundant 1081 * atomic change. 1082 */ 1083 1084/** 1085 * skb_cloned - is the buffer a clone 1086 * @skb: buffer to check 1087 * 1088 * Returns true if the buffer was generated with skb_clone() and is 1089 * one of multiple shared copies of the buffer. Cloned buffers are 1090 * shared data so must not be written to under normal circumstances. 1091 */ 1092static inline int skb_cloned(const struct sk_buff *skb) 1093{ 1094 return skb->cloned && 1095 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1096} 1097 1098static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1099{ 1100 might_sleep_if(pri & __GFP_WAIT); 1101 1102 if (skb_cloned(skb)) 1103 return pskb_expand_head(skb, 0, 0, pri); 1104 1105 return 0; 1106} 1107 1108/** 1109 * skb_header_cloned - is the header a clone 1110 * @skb: buffer to check 1111 * 1112 * Returns true if modifying the header part of the buffer requires 1113 * the data to be copied. 1114 */ 1115static inline int skb_header_cloned(const struct sk_buff *skb) 1116{ 1117 int dataref; 1118 1119 if (!skb->cloned) 1120 return 0; 1121 1122 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1123 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1124 return dataref != 1; 1125} 1126 1127/** 1128 * skb_header_release - release reference to header 1129 * @skb: buffer to operate on 1130 * 1131 * Drop a reference to the header part of the buffer. This is done 1132 * by acquiring a payload reference. You must not read from the header 1133 * part of skb->data after this. 1134 * Note : Check if you can use __skb_header_release() instead. 1135 */ 1136static inline void skb_header_release(struct sk_buff *skb) 1137{ 1138 BUG_ON(skb->nohdr); 1139 skb->nohdr = 1; 1140 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1141} 1142 1143/** 1144 * __skb_header_release - release reference to header 1145 * @skb: buffer to operate on 1146 * 1147 * Variant of skb_header_release() assuming skb is private to caller. 1148 * We can avoid one atomic operation. 1149 */ 1150static inline void __skb_header_release(struct sk_buff *skb) 1151{ 1152 skb->nohdr = 1; 1153 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1154} 1155 1156 1157/** 1158 * skb_shared - is the buffer shared 1159 * @skb: buffer to check 1160 * 1161 * Returns true if more than one person has a reference to this 1162 * buffer. 1163 */ 1164static inline int skb_shared(const struct sk_buff *skb) 1165{ 1166 return atomic_read(&skb->users) != 1; 1167} 1168 1169/** 1170 * skb_share_check - check if buffer is shared and if so clone it 1171 * @skb: buffer to check 1172 * @pri: priority for memory allocation 1173 * 1174 * If the buffer is shared the buffer is cloned and the old copy 1175 * drops a reference. A new clone with a single reference is returned. 1176 * If the buffer is not shared the original buffer is returned. When 1177 * being called from interrupt status or with spinlocks held pri must 1178 * be GFP_ATOMIC. 1179 * 1180 * NULL is returned on a memory allocation failure. 1181 */ 1182static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1183{ 1184 might_sleep_if(pri & __GFP_WAIT); 1185 if (skb_shared(skb)) { 1186 struct sk_buff *nskb = skb_clone(skb, pri); 1187 1188 if (likely(nskb)) 1189 consume_skb(skb); 1190 else 1191 kfree_skb(skb); 1192 skb = nskb; 1193 } 1194 return skb; 1195} 1196 1197/* 1198 * Copy shared buffers into a new sk_buff. We effectively do COW on 1199 * packets to handle cases where we have a local reader and forward 1200 * and a couple of other messy ones. The normal one is tcpdumping 1201 * a packet thats being forwarded. 1202 */ 1203 1204/** 1205 * skb_unshare - make a copy of a shared buffer 1206 * @skb: buffer to check 1207 * @pri: priority for memory allocation 1208 * 1209 * If the socket buffer is a clone then this function creates a new 1210 * copy of the data, drops a reference count on the old copy and returns 1211 * the new copy with the reference count at 1. If the buffer is not a clone 1212 * the original buffer is returned. When called with a spinlock held or 1213 * from interrupt state @pri must be %GFP_ATOMIC 1214 * 1215 * %NULL is returned on a memory allocation failure. 1216 */ 1217static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1218 gfp_t pri) 1219{ 1220 might_sleep_if(pri & __GFP_WAIT); 1221 if (skb_cloned(skb)) { 1222 struct sk_buff *nskb = skb_copy(skb, pri); 1223 1224 /* Free our shared copy */ 1225 if (likely(nskb)) 1226 consume_skb(skb); 1227 else 1228 kfree_skb(skb); 1229 skb = nskb; 1230 } 1231 return skb; 1232} 1233 1234/** 1235 * skb_peek - peek at the head of an &sk_buff_head 1236 * @list_: list to peek at 1237 * 1238 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1239 * be careful with this one. A peek leaves the buffer on the 1240 * list and someone else may run off with it. You must hold 1241 * the appropriate locks or have a private queue to do this. 1242 * 1243 * Returns %NULL for an empty list or a pointer to the head element. 1244 * The reference count is not incremented and the reference is therefore 1245 * volatile. Use with caution. 1246 */ 1247static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1248{ 1249 struct sk_buff *skb = list_->next; 1250 1251 if (skb == (struct sk_buff *)list_) 1252 skb = NULL; 1253 return skb; 1254} 1255 1256/** 1257 * skb_peek_next - peek skb following the given one from a queue 1258 * @skb: skb to start from 1259 * @list_: list to peek at 1260 * 1261 * Returns %NULL when the end of the list is met or a pointer to the 1262 * next element. The reference count is not incremented and the 1263 * reference is therefore volatile. Use with caution. 1264 */ 1265static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1266 const struct sk_buff_head *list_) 1267{ 1268 struct sk_buff *next = skb->next; 1269 1270 if (next == (struct sk_buff *)list_) 1271 next = NULL; 1272 return next; 1273} 1274 1275/** 1276 * skb_peek_tail - peek at the tail of an &sk_buff_head 1277 * @list_: list to peek at 1278 * 1279 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1280 * be careful with this one. A peek leaves the buffer on the 1281 * list and someone else may run off with it. You must hold 1282 * the appropriate locks or have a private queue to do this. 1283 * 1284 * Returns %NULL for an empty list or a pointer to the tail element. 1285 * The reference count is not incremented and the reference is therefore 1286 * volatile. Use with caution. 1287 */ 1288static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1289{ 1290 struct sk_buff *skb = list_->prev; 1291 1292 if (skb == (struct sk_buff *)list_) 1293 skb = NULL; 1294 return skb; 1295 1296} 1297 1298/** 1299 * skb_queue_len - get queue length 1300 * @list_: list to measure 1301 * 1302 * Return the length of an &sk_buff queue. 1303 */ 1304static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1305{ 1306 return list_->qlen; 1307} 1308 1309/** 1310 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1311 * @list: queue to initialize 1312 * 1313 * This initializes only the list and queue length aspects of 1314 * an sk_buff_head object. This allows to initialize the list 1315 * aspects of an sk_buff_head without reinitializing things like 1316 * the spinlock. It can also be used for on-stack sk_buff_head 1317 * objects where the spinlock is known to not be used. 1318 */ 1319static inline void __skb_queue_head_init(struct sk_buff_head *list) 1320{ 1321 list->prev = list->next = (struct sk_buff *)list; 1322 list->qlen = 0; 1323} 1324 1325/* 1326 * This function creates a split out lock class for each invocation; 1327 * this is needed for now since a whole lot of users of the skb-queue 1328 * infrastructure in drivers have different locking usage (in hardirq) 1329 * than the networking core (in softirq only). In the long run either the 1330 * network layer or drivers should need annotation to consolidate the 1331 * main types of usage into 3 classes. 1332 */ 1333static inline void skb_queue_head_init(struct sk_buff_head *list) 1334{ 1335 spin_lock_init(&list->lock); 1336 __skb_queue_head_init(list); 1337} 1338 1339static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1340 struct lock_class_key *class) 1341{ 1342 skb_queue_head_init(list); 1343 lockdep_set_class(&list->lock, class); 1344} 1345 1346/* 1347 * Insert an sk_buff on a list. 1348 * 1349 * The "__skb_xxxx()" functions are the non-atomic ones that 1350 * can only be called with interrupts disabled. 1351 */ 1352void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1353 struct sk_buff_head *list); 1354static inline void __skb_insert(struct sk_buff *newsk, 1355 struct sk_buff *prev, struct sk_buff *next, 1356 struct sk_buff_head *list) 1357{ 1358 newsk->next = next; 1359 newsk->prev = prev; 1360 next->prev = prev->next = newsk; 1361 list->qlen++; 1362} 1363 1364static inline void __skb_queue_splice(const struct sk_buff_head *list, 1365 struct sk_buff *prev, 1366 struct sk_buff *next) 1367{ 1368 struct sk_buff *first = list->next; 1369 struct sk_buff *last = list->prev; 1370 1371 first->prev = prev; 1372 prev->next = first; 1373 1374 last->next = next; 1375 next->prev = last; 1376} 1377 1378/** 1379 * skb_queue_splice - join two skb lists, this is designed for stacks 1380 * @list: the new list to add 1381 * @head: the place to add it in the first list 1382 */ 1383static inline void skb_queue_splice(const struct sk_buff_head *list, 1384 struct sk_buff_head *head) 1385{ 1386 if (!skb_queue_empty(list)) { 1387 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1388 head->qlen += list->qlen; 1389 } 1390} 1391 1392/** 1393 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1394 * @list: the new list to add 1395 * @head: the place to add it in the first list 1396 * 1397 * The list at @list is reinitialised 1398 */ 1399static inline void skb_queue_splice_init(struct sk_buff_head *list, 1400 struct sk_buff_head *head) 1401{ 1402 if (!skb_queue_empty(list)) { 1403 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1404 head->qlen += list->qlen; 1405 __skb_queue_head_init(list); 1406 } 1407} 1408 1409/** 1410 * skb_queue_splice_tail - join two skb lists, each list being a queue 1411 * @list: the new list to add 1412 * @head: the place to add it in the first list 1413 */ 1414static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1415 struct sk_buff_head *head) 1416{ 1417 if (!skb_queue_empty(list)) { 1418 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1419 head->qlen += list->qlen; 1420 } 1421} 1422 1423/** 1424 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1425 * @list: the new list to add 1426 * @head: the place to add it in the first list 1427 * 1428 * Each of the lists is a queue. 1429 * The list at @list is reinitialised 1430 */ 1431static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1432 struct sk_buff_head *head) 1433{ 1434 if (!skb_queue_empty(list)) { 1435 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1436 head->qlen += list->qlen; 1437 __skb_queue_head_init(list); 1438 } 1439} 1440 1441/** 1442 * __skb_queue_after - queue a buffer at the list head 1443 * @list: list to use 1444 * @prev: place after this buffer 1445 * @newsk: buffer to queue 1446 * 1447 * Queue a buffer int the middle of a list. This function takes no locks 1448 * and you must therefore hold required locks before calling it. 1449 * 1450 * A buffer cannot be placed on two lists at the same time. 1451 */ 1452static inline void __skb_queue_after(struct sk_buff_head *list, 1453 struct sk_buff *prev, 1454 struct sk_buff *newsk) 1455{ 1456 __skb_insert(newsk, prev, prev->next, list); 1457} 1458 1459void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1460 struct sk_buff_head *list); 1461 1462static inline void __skb_queue_before(struct sk_buff_head *list, 1463 struct sk_buff *next, 1464 struct sk_buff *newsk) 1465{ 1466 __skb_insert(newsk, next->prev, next, list); 1467} 1468 1469/** 1470 * __skb_queue_head - queue a buffer at the list head 1471 * @list: list to use 1472 * @newsk: buffer to queue 1473 * 1474 * Queue a buffer at the start of a list. This function takes no locks 1475 * and you must therefore hold required locks before calling it. 1476 * 1477 * A buffer cannot be placed on two lists at the same time. 1478 */ 1479void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1480static inline void __skb_queue_head(struct sk_buff_head *list, 1481 struct sk_buff *newsk) 1482{ 1483 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1484} 1485 1486/** 1487 * __skb_queue_tail - queue a buffer at the list tail 1488 * @list: list to use 1489 * @newsk: buffer to queue 1490 * 1491 * Queue a buffer at the end of a list. This function takes no locks 1492 * and you must therefore hold required locks before calling it. 1493 * 1494 * A buffer cannot be placed on two lists at the same time. 1495 */ 1496void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1497static inline void __skb_queue_tail(struct sk_buff_head *list, 1498 struct sk_buff *newsk) 1499{ 1500 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1501} 1502 1503/* 1504 * remove sk_buff from list. _Must_ be called atomically, and with 1505 * the list known.. 1506 */ 1507void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1508static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1509{ 1510 struct sk_buff *next, *prev; 1511 1512 list->qlen--; 1513 next = skb->next; 1514 prev = skb->prev; 1515 skb->next = skb->prev = NULL; 1516 next->prev = prev; 1517 prev->next = next; 1518} 1519 1520/** 1521 * __skb_dequeue - remove from the head of the queue 1522 * @list: list to dequeue from 1523 * 1524 * Remove the head of the list. This function does not take any locks 1525 * so must be used with appropriate locks held only. The head item is 1526 * returned or %NULL if the list is empty. 1527 */ 1528struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1529static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1530{ 1531 struct sk_buff *skb = skb_peek(list); 1532 if (skb) 1533 __skb_unlink(skb, list); 1534 return skb; 1535} 1536 1537/** 1538 * __skb_dequeue_tail - remove from the tail of the queue 1539 * @list: list to dequeue from 1540 * 1541 * Remove the tail of the list. This function does not take any locks 1542 * so must be used with appropriate locks held only. The tail item is 1543 * returned or %NULL if the list is empty. 1544 */ 1545struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1546static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1547{ 1548 struct sk_buff *skb = skb_peek_tail(list); 1549 if (skb) 1550 __skb_unlink(skb, list); 1551 return skb; 1552} 1553 1554 1555static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1556{ 1557 return skb->data_len; 1558} 1559 1560static inline unsigned int skb_headlen(const struct sk_buff *skb) 1561{ 1562 return skb->len - skb->data_len; 1563} 1564 1565static inline int skb_pagelen(const struct sk_buff *skb) 1566{ 1567 int i, len = 0; 1568 1569 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1570 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1571 return len + skb_headlen(skb); 1572} 1573 1574/** 1575 * __skb_fill_page_desc - initialise a paged fragment in an skb 1576 * @skb: buffer containing fragment to be initialised 1577 * @i: paged fragment index to initialise 1578 * @page: the page to use for this fragment 1579 * @off: the offset to the data with @page 1580 * @size: the length of the data 1581 * 1582 * Initialises the @i'th fragment of @skb to point to &size bytes at 1583 * offset @off within @page. 1584 * 1585 * Does not take any additional reference on the fragment. 1586 */ 1587static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1588 struct page *page, int off, int size) 1589{ 1590 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1591 1592 /* 1593 * Propagate page->pfmemalloc to the skb if we can. The problem is 1594 * that not all callers have unique ownership of the page. If 1595 * pfmemalloc is set, we check the mapping as a mapping implies 1596 * page->index is set (index and pfmemalloc share space). 1597 * If it's a valid mapping, we cannot use page->pfmemalloc but we 1598 * do not lose pfmemalloc information as the pages would not be 1599 * allocated using __GFP_MEMALLOC. 1600 */ 1601 frag->page.p = page; 1602 frag->page_offset = off; 1603 skb_frag_size_set(frag, size); 1604 1605 page = compound_head(page); 1606 if (page->pfmemalloc && !page->mapping) 1607 skb->pfmemalloc = true; 1608} 1609 1610/** 1611 * skb_fill_page_desc - initialise a paged fragment in an skb 1612 * @skb: buffer containing fragment to be initialised 1613 * @i: paged fragment index to initialise 1614 * @page: the page to use for this fragment 1615 * @off: the offset to the data with @page 1616 * @size: the length of the data 1617 * 1618 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1619 * @skb to point to @size bytes at offset @off within @page. In 1620 * addition updates @skb such that @i is the last fragment. 1621 * 1622 * Does not take any additional reference on the fragment. 1623 */ 1624static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1625 struct page *page, int off, int size) 1626{ 1627 __skb_fill_page_desc(skb, i, page, off, size); 1628 skb_shinfo(skb)->nr_frags = i + 1; 1629} 1630 1631void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1632 int size, unsigned int truesize); 1633 1634void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1635 unsigned int truesize); 1636 1637#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1638#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1639#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1640 1641#ifdef NET_SKBUFF_DATA_USES_OFFSET 1642static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1643{ 1644 return skb->head + skb->tail; 1645} 1646 1647static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1648{ 1649 skb->tail = skb->data - skb->head; 1650} 1651 1652static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1653{ 1654 skb_reset_tail_pointer(skb); 1655 skb->tail += offset; 1656} 1657 1658#else /* NET_SKBUFF_DATA_USES_OFFSET */ 1659static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1660{ 1661 return skb->tail; 1662} 1663 1664static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1665{ 1666 skb->tail = skb->data; 1667} 1668 1669static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1670{ 1671 skb->tail = skb->data + offset; 1672} 1673 1674#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1675 1676/* 1677 * Add data to an sk_buff 1678 */ 1679unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1680unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1681static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1682{ 1683 unsigned char *tmp = skb_tail_pointer(skb); 1684 SKB_LINEAR_ASSERT(skb); 1685 skb->tail += len; 1686 skb->len += len; 1687 return tmp; 1688} 1689 1690unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1691static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1692{ 1693 skb->data -= len; 1694 skb->len += len; 1695 return skb->data; 1696} 1697 1698unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1699static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1700{ 1701 skb->len -= len; 1702 BUG_ON(skb->len < skb->data_len); 1703 return skb->data += len; 1704} 1705 1706static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1707{ 1708 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1709} 1710 1711unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1712 1713static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1714{ 1715 if (len > skb_headlen(skb) && 1716 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1717 return NULL; 1718 skb->len -= len; 1719 return skb->data += len; 1720} 1721 1722static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1723{ 1724 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1725} 1726 1727static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1728{ 1729 if (likely(len <= skb_headlen(skb))) 1730 return 1; 1731 if (unlikely(len > skb->len)) 1732 return 0; 1733 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1734} 1735 1736/** 1737 * skb_headroom - bytes at buffer head 1738 * @skb: buffer to check 1739 * 1740 * Return the number of bytes of free space at the head of an &sk_buff. 1741 */ 1742static inline unsigned int skb_headroom(const struct sk_buff *skb) 1743{ 1744 return skb->data - skb->head; 1745} 1746 1747/** 1748 * skb_tailroom - bytes at buffer end 1749 * @skb: buffer to check 1750 * 1751 * Return the number of bytes of free space at the tail of an sk_buff 1752 */ 1753static inline int skb_tailroom(const struct sk_buff *skb) 1754{ 1755 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1756} 1757 1758/** 1759 * skb_availroom - bytes at buffer end 1760 * @skb: buffer to check 1761 * 1762 * Return the number of bytes of free space at the tail of an sk_buff 1763 * allocated by sk_stream_alloc() 1764 */ 1765static inline int skb_availroom(const struct sk_buff *skb) 1766{ 1767 if (skb_is_nonlinear(skb)) 1768 return 0; 1769 1770 return skb->end - skb->tail - skb->reserved_tailroom; 1771} 1772 1773/** 1774 * skb_reserve - adjust headroom 1775 * @skb: buffer to alter 1776 * @len: bytes to move 1777 * 1778 * Increase the headroom of an empty &sk_buff by reducing the tail 1779 * room. This is only allowed for an empty buffer. 1780 */ 1781static inline void skb_reserve(struct sk_buff *skb, int len) 1782{ 1783 skb->data += len; 1784 skb->tail += len; 1785} 1786 1787#define ENCAP_TYPE_ETHER 0 1788#define ENCAP_TYPE_IPPROTO 1 1789 1790static inline void skb_set_inner_protocol(struct sk_buff *skb, 1791 __be16 protocol) 1792{ 1793 skb->inner_protocol = protocol; 1794 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 1795} 1796 1797static inline void skb_set_inner_ipproto(struct sk_buff *skb, 1798 __u8 ipproto) 1799{ 1800 skb->inner_ipproto = ipproto; 1801 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 1802} 1803 1804static inline void skb_reset_inner_headers(struct sk_buff *skb) 1805{ 1806 skb->inner_mac_header = skb->mac_header; 1807 skb->inner_network_header = skb->network_header; 1808 skb->inner_transport_header = skb->transport_header; 1809} 1810 1811static inline void skb_reset_mac_len(struct sk_buff *skb) 1812{ 1813 skb->mac_len = skb->network_header - skb->mac_header; 1814} 1815 1816static inline unsigned char *skb_inner_transport_header(const struct sk_buff 1817 *skb) 1818{ 1819 return skb->head + skb->inner_transport_header; 1820} 1821 1822static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 1823{ 1824 skb->inner_transport_header = skb->data - skb->head; 1825} 1826 1827static inline void skb_set_inner_transport_header(struct sk_buff *skb, 1828 const int offset) 1829{ 1830 skb_reset_inner_transport_header(skb); 1831 skb->inner_transport_header += offset; 1832} 1833 1834static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 1835{ 1836 return skb->head + skb->inner_network_header; 1837} 1838 1839static inline void skb_reset_inner_network_header(struct sk_buff *skb) 1840{ 1841 skb->inner_network_header = skb->data - skb->head; 1842} 1843 1844static inline void skb_set_inner_network_header(struct sk_buff *skb, 1845 const int offset) 1846{ 1847 skb_reset_inner_network_header(skb); 1848 skb->inner_network_header += offset; 1849} 1850 1851static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 1852{ 1853 return skb->head + skb->inner_mac_header; 1854} 1855 1856static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 1857{ 1858 skb->inner_mac_header = skb->data - skb->head; 1859} 1860 1861static inline void skb_set_inner_mac_header(struct sk_buff *skb, 1862 const int offset) 1863{ 1864 skb_reset_inner_mac_header(skb); 1865 skb->inner_mac_header += offset; 1866} 1867static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 1868{ 1869 return skb->transport_header != (typeof(skb->transport_header))~0U; 1870} 1871 1872static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 1873{ 1874 return skb->head + skb->transport_header; 1875} 1876 1877static inline void skb_reset_transport_header(struct sk_buff *skb) 1878{ 1879 skb->transport_header = skb->data - skb->head; 1880} 1881 1882static inline void skb_set_transport_header(struct sk_buff *skb, 1883 const int offset) 1884{ 1885 skb_reset_transport_header(skb); 1886 skb->transport_header += offset; 1887} 1888 1889static inline unsigned char *skb_network_header(const struct sk_buff *skb) 1890{ 1891 return skb->head + skb->network_header; 1892} 1893 1894static inline void skb_reset_network_header(struct sk_buff *skb) 1895{ 1896 skb->network_header = skb->data - skb->head; 1897} 1898 1899static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 1900{ 1901 skb_reset_network_header(skb); 1902 skb->network_header += offset; 1903} 1904 1905static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 1906{ 1907 return skb->head + skb->mac_header; 1908} 1909 1910static inline int skb_mac_header_was_set(const struct sk_buff *skb) 1911{ 1912 return skb->mac_header != (typeof(skb->mac_header))~0U; 1913} 1914 1915static inline void skb_reset_mac_header(struct sk_buff *skb) 1916{ 1917 skb->mac_header = skb->data - skb->head; 1918} 1919 1920static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 1921{ 1922 skb_reset_mac_header(skb); 1923 skb->mac_header += offset; 1924} 1925 1926static inline void skb_pop_mac_header(struct sk_buff *skb) 1927{ 1928 skb->mac_header = skb->network_header; 1929} 1930 1931static inline void skb_probe_transport_header(struct sk_buff *skb, 1932 const int offset_hint) 1933{ 1934 struct flow_keys keys; 1935 1936 if (skb_transport_header_was_set(skb)) 1937 return; 1938 else if (skb_flow_dissect(skb, &keys)) 1939 skb_set_transport_header(skb, keys.thoff); 1940 else 1941 skb_set_transport_header(skb, offset_hint); 1942} 1943 1944static inline void skb_mac_header_rebuild(struct sk_buff *skb) 1945{ 1946 if (skb_mac_header_was_set(skb)) { 1947 const unsigned char *old_mac = skb_mac_header(skb); 1948 1949 skb_set_mac_header(skb, -skb->mac_len); 1950 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 1951 } 1952} 1953 1954static inline int skb_checksum_start_offset(const struct sk_buff *skb) 1955{ 1956 return skb->csum_start - skb_headroom(skb); 1957} 1958 1959static inline int skb_transport_offset(const struct sk_buff *skb) 1960{ 1961 return skb_transport_header(skb) - skb->data; 1962} 1963 1964static inline u32 skb_network_header_len(const struct sk_buff *skb) 1965{ 1966 return skb->transport_header - skb->network_header; 1967} 1968 1969static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 1970{ 1971 return skb->inner_transport_header - skb->inner_network_header; 1972} 1973 1974static inline int skb_network_offset(const struct sk_buff *skb) 1975{ 1976 return skb_network_header(skb) - skb->data; 1977} 1978 1979static inline int skb_inner_network_offset(const struct sk_buff *skb) 1980{ 1981 return skb_inner_network_header(skb) - skb->data; 1982} 1983 1984static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 1985{ 1986 return pskb_may_pull(skb, skb_network_offset(skb) + len); 1987} 1988 1989/* 1990 * CPUs often take a performance hit when accessing unaligned memory 1991 * locations. The actual performance hit varies, it can be small if the 1992 * hardware handles it or large if we have to take an exception and fix it 1993 * in software. 1994 * 1995 * Since an ethernet header is 14 bytes network drivers often end up with 1996 * the IP header at an unaligned offset. The IP header can be aligned by 1997 * shifting the start of the packet by 2 bytes. Drivers should do this 1998 * with: 1999 * 2000 * skb_reserve(skb, NET_IP_ALIGN); 2001 * 2002 * The downside to this alignment of the IP header is that the DMA is now 2003 * unaligned. On some architectures the cost of an unaligned DMA is high 2004 * and this cost outweighs the gains made by aligning the IP header. 2005 * 2006 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2007 * to be overridden. 2008 */ 2009#ifndef NET_IP_ALIGN 2010#define NET_IP_ALIGN 2 2011#endif 2012 2013/* 2014 * The networking layer reserves some headroom in skb data (via 2015 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2016 * the header has to grow. In the default case, if the header has to grow 2017 * 32 bytes or less we avoid the reallocation. 2018 * 2019 * Unfortunately this headroom changes the DMA alignment of the resulting 2020 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2021 * on some architectures. An architecture can override this value, 2022 * perhaps setting it to a cacheline in size (since that will maintain 2023 * cacheline alignment of the DMA). It must be a power of 2. 2024 * 2025 * Various parts of the networking layer expect at least 32 bytes of 2026 * headroom, you should not reduce this. 2027 * 2028 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2029 * to reduce average number of cache lines per packet. 2030 * get_rps_cpus() for example only access one 64 bytes aligned block : 2031 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2032 */ 2033#ifndef NET_SKB_PAD 2034#define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2035#endif 2036 2037int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2038 2039static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2040{ 2041 if (unlikely(skb_is_nonlinear(skb))) { 2042 WARN_ON(1); 2043 return; 2044 } 2045 skb->len = len; 2046 skb_set_tail_pointer(skb, len); 2047} 2048 2049void skb_trim(struct sk_buff *skb, unsigned int len); 2050 2051static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2052{ 2053 if (skb->data_len) 2054 return ___pskb_trim(skb, len); 2055 __skb_trim(skb, len); 2056 return 0; 2057} 2058 2059static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2060{ 2061 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2062} 2063 2064/** 2065 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2066 * @skb: buffer to alter 2067 * @len: new length 2068 * 2069 * This is identical to pskb_trim except that the caller knows that 2070 * the skb is not cloned so we should never get an error due to out- 2071 * of-memory. 2072 */ 2073static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2074{ 2075 int err = pskb_trim(skb, len); 2076 BUG_ON(err); 2077} 2078 2079/** 2080 * skb_orphan - orphan a buffer 2081 * @skb: buffer to orphan 2082 * 2083 * If a buffer currently has an owner then we call the owner's 2084 * destructor function and make the @skb unowned. The buffer continues 2085 * to exist but is no longer charged to its former owner. 2086 */ 2087static inline void skb_orphan(struct sk_buff *skb) 2088{ 2089 if (skb->destructor) { 2090 skb->destructor(skb); 2091 skb->destructor = NULL; 2092 skb->sk = NULL; 2093 } else { 2094 BUG_ON(skb->sk); 2095 } 2096} 2097 2098/** 2099 * skb_orphan_frags - orphan the frags contained in a buffer 2100 * @skb: buffer to orphan frags from 2101 * @gfp_mask: allocation mask for replacement pages 2102 * 2103 * For each frag in the SKB which needs a destructor (i.e. has an 2104 * owner) create a copy of that frag and release the original 2105 * page by calling the destructor. 2106 */ 2107static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2108{ 2109 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2110 return 0; 2111 return skb_copy_ubufs(skb, gfp_mask); 2112} 2113 2114/** 2115 * __skb_queue_purge - empty a list 2116 * @list: list to empty 2117 * 2118 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2119 * the list and one reference dropped. This function does not take the 2120 * list lock and the caller must hold the relevant locks to use it. 2121 */ 2122void skb_queue_purge(struct sk_buff_head *list); 2123static inline void __skb_queue_purge(struct sk_buff_head *list) 2124{ 2125 struct sk_buff *skb; 2126 while ((skb = __skb_dequeue(list)) != NULL) 2127 kfree_skb(skb); 2128} 2129 2130#define NETDEV_FRAG_PAGE_MAX_ORDER get_order(32768) 2131#define NETDEV_FRAG_PAGE_MAX_SIZE (PAGE_SIZE << NETDEV_FRAG_PAGE_MAX_ORDER) 2132#define NETDEV_PAGECNT_MAX_BIAS NETDEV_FRAG_PAGE_MAX_SIZE 2133 2134void *netdev_alloc_frag(unsigned int fragsz); 2135 2136struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2137 gfp_t gfp_mask); 2138 2139/** 2140 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2141 * @dev: network device to receive on 2142 * @length: length to allocate 2143 * 2144 * Allocate a new &sk_buff and assign it a usage count of one. The 2145 * buffer has unspecified headroom built in. Users should allocate 2146 * the headroom they think they need without accounting for the 2147 * built in space. The built in space is used for optimisations. 2148 * 2149 * %NULL is returned if there is no free memory. Although this function 2150 * allocates memory it can be called from an interrupt. 2151 */ 2152static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2153 unsigned int length) 2154{ 2155 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2156} 2157 2158/* legacy helper around __netdev_alloc_skb() */ 2159static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2160 gfp_t gfp_mask) 2161{ 2162 return __netdev_alloc_skb(NULL, length, gfp_mask); 2163} 2164 2165/* legacy helper around netdev_alloc_skb() */ 2166static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2167{ 2168 return netdev_alloc_skb(NULL, length); 2169} 2170 2171 2172static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2173 unsigned int length, gfp_t gfp) 2174{ 2175 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2176 2177 if (NET_IP_ALIGN && skb) 2178 skb_reserve(skb, NET_IP_ALIGN); 2179 return skb; 2180} 2181 2182static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2183 unsigned int length) 2184{ 2185 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2186} 2187 2188void *napi_alloc_frag(unsigned int fragsz); 2189struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2190 unsigned int length, gfp_t gfp_mask); 2191static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2192 unsigned int length) 2193{ 2194 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2195} 2196 2197/** 2198 * __dev_alloc_pages - allocate page for network Rx 2199 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2200 * @order: size of the allocation 2201 * 2202 * Allocate a new page. 2203 * 2204 * %NULL is returned if there is no free memory. 2205*/ 2206static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2207 unsigned int order) 2208{ 2209 /* This piece of code contains several assumptions. 2210 * 1. This is for device Rx, therefor a cold page is preferred. 2211 * 2. The expectation is the user wants a compound page. 2212 * 3. If requesting a order 0 page it will not be compound 2213 * due to the check to see if order has a value in prep_new_page 2214 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2215 * code in gfp_to_alloc_flags that should be enforcing this. 2216 */ 2217 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2218 2219 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2220} 2221 2222static inline struct page *dev_alloc_pages(unsigned int order) 2223{ 2224 return __dev_alloc_pages(GFP_ATOMIC, order); 2225} 2226 2227/** 2228 * __dev_alloc_page - allocate a page for network Rx 2229 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2230 * 2231 * Allocate a new page. 2232 * 2233 * %NULL is returned if there is no free memory. 2234 */ 2235static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2236{ 2237 return __dev_alloc_pages(gfp_mask, 0); 2238} 2239 2240static inline struct page *dev_alloc_page(void) 2241{ 2242 return __dev_alloc_page(GFP_ATOMIC); 2243} 2244 2245/** 2246 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2247 * @page: The page that was allocated from skb_alloc_page 2248 * @skb: The skb that may need pfmemalloc set 2249 */ 2250static inline void skb_propagate_pfmemalloc(struct page *page, 2251 struct sk_buff *skb) 2252{ 2253 if (page && page->pfmemalloc) 2254 skb->pfmemalloc = true; 2255} 2256 2257/** 2258 * skb_frag_page - retrieve the page referred to by a paged fragment 2259 * @frag: the paged fragment 2260 * 2261 * Returns the &struct page associated with @frag. 2262 */ 2263static inline struct page *skb_frag_page(const skb_frag_t *frag) 2264{ 2265 return frag->page.p; 2266} 2267 2268/** 2269 * __skb_frag_ref - take an addition reference on a paged fragment. 2270 * @frag: the paged fragment 2271 * 2272 * Takes an additional reference on the paged fragment @frag. 2273 */ 2274static inline void __skb_frag_ref(skb_frag_t *frag) 2275{ 2276 get_page(skb_frag_page(frag)); 2277} 2278 2279/** 2280 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2281 * @skb: the buffer 2282 * @f: the fragment offset. 2283 * 2284 * Takes an additional reference on the @f'th paged fragment of @skb. 2285 */ 2286static inline void skb_frag_ref(struct sk_buff *skb, int f) 2287{ 2288 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2289} 2290 2291/** 2292 * __skb_frag_unref - release a reference on a paged fragment. 2293 * @frag: the paged fragment 2294 * 2295 * Releases a reference on the paged fragment @frag. 2296 */ 2297static inline void __skb_frag_unref(skb_frag_t *frag) 2298{ 2299 put_page(skb_frag_page(frag)); 2300} 2301 2302/** 2303 * skb_frag_unref - release a reference on a paged fragment of an skb. 2304 * @skb: the buffer 2305 * @f: the fragment offset 2306 * 2307 * Releases a reference on the @f'th paged fragment of @skb. 2308 */ 2309static inline void skb_frag_unref(struct sk_buff *skb, int f) 2310{ 2311 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2312} 2313 2314/** 2315 * skb_frag_address - gets the address of the data contained in a paged fragment 2316 * @frag: the paged fragment buffer 2317 * 2318 * Returns the address of the data within @frag. The page must already 2319 * be mapped. 2320 */ 2321static inline void *skb_frag_address(const skb_frag_t *frag) 2322{ 2323 return page_address(skb_frag_page(frag)) + frag->page_offset; 2324} 2325 2326/** 2327 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2328 * @frag: the paged fragment buffer 2329 * 2330 * Returns the address of the data within @frag. Checks that the page 2331 * is mapped and returns %NULL otherwise. 2332 */ 2333static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2334{ 2335 void *ptr = page_address(skb_frag_page(frag)); 2336 if (unlikely(!ptr)) 2337 return NULL; 2338 2339 return ptr + frag->page_offset; 2340} 2341 2342/** 2343 * __skb_frag_set_page - sets the page contained in a paged fragment 2344 * @frag: the paged fragment 2345 * @page: the page to set 2346 * 2347 * Sets the fragment @frag to contain @page. 2348 */ 2349static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2350{ 2351 frag->page.p = page; 2352} 2353 2354/** 2355 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2356 * @skb: the buffer 2357 * @f: the fragment offset 2358 * @page: the page to set 2359 * 2360 * Sets the @f'th fragment of @skb to contain @page. 2361 */ 2362static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2363 struct page *page) 2364{ 2365 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2366} 2367 2368bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2369 2370/** 2371 * skb_frag_dma_map - maps a paged fragment via the DMA API 2372 * @dev: the device to map the fragment to 2373 * @frag: the paged fragment to map 2374 * @offset: the offset within the fragment (starting at the 2375 * fragment's own offset) 2376 * @size: the number of bytes to map 2377 * @dir: the direction of the mapping (%PCI_DMA_*) 2378 * 2379 * Maps the page associated with @frag to @device. 2380 */ 2381static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2382 const skb_frag_t *frag, 2383 size_t offset, size_t size, 2384 enum dma_data_direction dir) 2385{ 2386 return dma_map_page(dev, skb_frag_page(frag), 2387 frag->page_offset + offset, size, dir); 2388} 2389 2390static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2391 gfp_t gfp_mask) 2392{ 2393 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2394} 2395 2396 2397static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2398 gfp_t gfp_mask) 2399{ 2400 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2401} 2402 2403 2404/** 2405 * skb_clone_writable - is the header of a clone writable 2406 * @skb: buffer to check 2407 * @len: length up to which to write 2408 * 2409 * Returns true if modifying the header part of the cloned buffer 2410 * does not requires the data to be copied. 2411 */ 2412static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2413{ 2414 return !skb_header_cloned(skb) && 2415 skb_headroom(skb) + len <= skb->hdr_len; 2416} 2417 2418static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2419 int cloned) 2420{ 2421 int delta = 0; 2422 2423 if (headroom > skb_headroom(skb)) 2424 delta = headroom - skb_headroom(skb); 2425 2426 if (delta || cloned) 2427 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2428 GFP_ATOMIC); 2429 return 0; 2430} 2431 2432/** 2433 * skb_cow - copy header of skb when it is required 2434 * @skb: buffer to cow 2435 * @headroom: needed headroom 2436 * 2437 * If the skb passed lacks sufficient headroom or its data part 2438 * is shared, data is reallocated. If reallocation fails, an error 2439 * is returned and original skb is not changed. 2440 * 2441 * The result is skb with writable area skb->head...skb->tail 2442 * and at least @headroom of space at head. 2443 */ 2444static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2445{ 2446 return __skb_cow(skb, headroom, skb_cloned(skb)); 2447} 2448 2449/** 2450 * skb_cow_head - skb_cow but only making the head writable 2451 * @skb: buffer to cow 2452 * @headroom: needed headroom 2453 * 2454 * This function is identical to skb_cow except that we replace the 2455 * skb_cloned check by skb_header_cloned. It should be used when 2456 * you only need to push on some header and do not need to modify 2457 * the data. 2458 */ 2459static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2460{ 2461 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2462} 2463 2464/** 2465 * skb_padto - pad an skbuff up to a minimal size 2466 * @skb: buffer to pad 2467 * @len: minimal length 2468 * 2469 * Pads up a buffer to ensure the trailing bytes exist and are 2470 * blanked. If the buffer already contains sufficient data it 2471 * is untouched. Otherwise it is extended. Returns zero on 2472 * success. The skb is freed on error. 2473 */ 2474static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2475{ 2476 unsigned int size = skb->len; 2477 if (likely(size >= len)) 2478 return 0; 2479 return skb_pad(skb, len - size); 2480} 2481 2482/** 2483 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2484 * @skb: buffer to pad 2485 * @len: minimal length 2486 * 2487 * Pads up a buffer to ensure the trailing bytes exist and are 2488 * blanked. If the buffer already contains sufficient data it 2489 * is untouched. Otherwise it is extended. Returns zero on 2490 * success. The skb is freed on error. 2491 */ 2492static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2493{ 2494 unsigned int size = skb->len; 2495 2496 if (unlikely(size < len)) { 2497 len -= size; 2498 if (skb_pad(skb, len)) 2499 return -ENOMEM; 2500 __skb_put(skb, len); 2501 } 2502 return 0; 2503} 2504 2505static inline int skb_add_data(struct sk_buff *skb, 2506 struct iov_iter *from, int copy) 2507{ 2508 const int off = skb->len; 2509 2510 if (skb->ip_summed == CHECKSUM_NONE) { 2511 __wsum csum = 0; 2512 if (csum_and_copy_from_iter(skb_put(skb, copy), copy, 2513 &csum, from) == copy) { 2514 skb->csum = csum_block_add(skb->csum, csum, off); 2515 return 0; 2516 } 2517 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy) 2518 return 0; 2519 2520 __skb_trim(skb, off); 2521 return -EFAULT; 2522} 2523 2524static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2525 const struct page *page, int off) 2526{ 2527 if (i) { 2528 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2529 2530 return page == skb_frag_page(frag) && 2531 off == frag->page_offset + skb_frag_size(frag); 2532 } 2533 return false; 2534} 2535 2536static inline int __skb_linearize(struct sk_buff *skb) 2537{ 2538 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2539} 2540 2541/** 2542 * skb_linearize - convert paged skb to linear one 2543 * @skb: buffer to linarize 2544 * 2545 * If there is no free memory -ENOMEM is returned, otherwise zero 2546 * is returned and the old skb data released. 2547 */ 2548static inline int skb_linearize(struct sk_buff *skb) 2549{ 2550 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2551} 2552 2553/** 2554 * skb_has_shared_frag - can any frag be overwritten 2555 * @skb: buffer to test 2556 * 2557 * Return true if the skb has at least one frag that might be modified 2558 * by an external entity (as in vmsplice()/sendfile()) 2559 */ 2560static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2561{ 2562 return skb_is_nonlinear(skb) && 2563 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2564} 2565 2566/** 2567 * skb_linearize_cow - make sure skb is linear and writable 2568 * @skb: buffer to process 2569 * 2570 * If there is no free memory -ENOMEM is returned, otherwise zero 2571 * is returned and the old skb data released. 2572 */ 2573static inline int skb_linearize_cow(struct sk_buff *skb) 2574{ 2575 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2576 __skb_linearize(skb) : 0; 2577} 2578 2579/** 2580 * skb_postpull_rcsum - update checksum for received skb after pull 2581 * @skb: buffer to update 2582 * @start: start of data before pull 2583 * @len: length of data pulled 2584 * 2585 * After doing a pull on a received packet, you need to call this to 2586 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2587 * CHECKSUM_NONE so that it can be recomputed from scratch. 2588 */ 2589 2590static inline void skb_postpull_rcsum(struct sk_buff *skb, 2591 const void *start, unsigned int len) 2592{ 2593 if (skb->ip_summed == CHECKSUM_COMPLETE) 2594 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0)); 2595} 2596 2597unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2598 2599/** 2600 * pskb_trim_rcsum - trim received skb and update checksum 2601 * @skb: buffer to trim 2602 * @len: new length 2603 * 2604 * This is exactly the same as pskb_trim except that it ensures the 2605 * checksum of received packets are still valid after the operation. 2606 */ 2607 2608static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2609{ 2610 if (likely(len >= skb->len)) 2611 return 0; 2612 if (skb->ip_summed == CHECKSUM_COMPLETE) 2613 skb->ip_summed = CHECKSUM_NONE; 2614 return __pskb_trim(skb, len); 2615} 2616 2617#define skb_queue_walk(queue, skb) \ 2618 for (skb = (queue)->next; \ 2619 skb != (struct sk_buff *)(queue); \ 2620 skb = skb->next) 2621 2622#define skb_queue_walk_safe(queue, skb, tmp) \ 2623 for (skb = (queue)->next, tmp = skb->next; \ 2624 skb != (struct sk_buff *)(queue); \ 2625 skb = tmp, tmp = skb->next) 2626 2627#define skb_queue_walk_from(queue, skb) \ 2628 for (; skb != (struct sk_buff *)(queue); \ 2629 skb = skb->next) 2630 2631#define skb_queue_walk_from_safe(queue, skb, tmp) \ 2632 for (tmp = skb->next; \ 2633 skb != (struct sk_buff *)(queue); \ 2634 skb = tmp, tmp = skb->next) 2635 2636#define skb_queue_reverse_walk(queue, skb) \ 2637 for (skb = (queue)->prev; \ 2638 skb != (struct sk_buff *)(queue); \ 2639 skb = skb->prev) 2640 2641#define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 2642 for (skb = (queue)->prev, tmp = skb->prev; \ 2643 skb != (struct sk_buff *)(queue); \ 2644 skb = tmp, tmp = skb->prev) 2645 2646#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 2647 for (tmp = skb->prev; \ 2648 skb != (struct sk_buff *)(queue); \ 2649 skb = tmp, tmp = skb->prev) 2650 2651static inline bool skb_has_frag_list(const struct sk_buff *skb) 2652{ 2653 return skb_shinfo(skb)->frag_list != NULL; 2654} 2655 2656static inline void skb_frag_list_init(struct sk_buff *skb) 2657{ 2658 skb_shinfo(skb)->frag_list = NULL; 2659} 2660 2661static inline void skb_frag_add_head(struct sk_buff *skb, struct sk_buff *frag) 2662{ 2663 frag->next = skb_shinfo(skb)->frag_list; 2664 skb_shinfo(skb)->frag_list = frag; 2665} 2666 2667#define skb_walk_frags(skb, iter) \ 2668 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 2669 2670struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 2671 int *peeked, int *off, int *err); 2672struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 2673 int *err); 2674unsigned int datagram_poll(struct file *file, struct socket *sock, 2675 struct poll_table_struct *wait); 2676int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 2677 struct iov_iter *to, int size); 2678static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 2679 struct msghdr *msg, int size) 2680{ 2681 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 2682} 2683int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 2684 struct msghdr *msg); 2685int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 2686 struct iov_iter *from, int len); 2687int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 2688void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 2689void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb); 2690int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 2691int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 2692int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 2693__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 2694 int len, __wsum csum); 2695int skb_splice_bits(struct sk_buff *skb, unsigned int offset, 2696 struct pipe_inode_info *pipe, unsigned int len, 2697 unsigned int flags); 2698void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 2699unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 2700int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 2701 int len, int hlen); 2702void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 2703int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 2704void skb_scrub_packet(struct sk_buff *skb, bool xnet); 2705unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 2706struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 2707struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 2708int skb_ensure_writable(struct sk_buff *skb, int write_len); 2709int skb_vlan_pop(struct sk_buff *skb); 2710int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 2711 2712static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 2713{ 2714 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2715} 2716 2717static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 2718{ 2719 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2720} 2721 2722struct skb_checksum_ops { 2723 __wsum (*update)(const void *mem, int len, __wsum wsum); 2724 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 2725}; 2726 2727__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 2728 __wsum csum, const struct skb_checksum_ops *ops); 2729__wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 2730 __wsum csum); 2731 2732static inline void *__skb_header_pointer(const struct sk_buff *skb, int offset, 2733 int len, void *data, int hlen, void *buffer) 2734{ 2735 if (hlen - offset >= len) 2736 return data + offset; 2737 2738 if (!skb || 2739 skb_copy_bits(skb, offset, buffer, len) < 0) 2740 return NULL; 2741 2742 return buffer; 2743} 2744 2745static inline void *skb_header_pointer(const struct sk_buff *skb, int offset, 2746 int len, void *buffer) 2747{ 2748 return __skb_header_pointer(skb, offset, len, skb->data, 2749 skb_headlen(skb), buffer); 2750} 2751 2752/** 2753 * skb_needs_linearize - check if we need to linearize a given skb 2754 * depending on the given device features. 2755 * @skb: socket buffer to check 2756 * @features: net device features 2757 * 2758 * Returns true if either: 2759 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 2760 * 2. skb is fragmented and the device does not support SG. 2761 */ 2762static inline bool skb_needs_linearize(struct sk_buff *skb, 2763 netdev_features_t features) 2764{ 2765 return skb_is_nonlinear(skb) && 2766 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 2767 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 2768} 2769 2770static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 2771 void *to, 2772 const unsigned int len) 2773{ 2774 memcpy(to, skb->data, len); 2775} 2776 2777static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 2778 const int offset, void *to, 2779 const unsigned int len) 2780{ 2781 memcpy(to, skb->data + offset, len); 2782} 2783 2784static inline void skb_copy_to_linear_data(struct sk_buff *skb, 2785 const void *from, 2786 const unsigned int len) 2787{ 2788 memcpy(skb->data, from, len); 2789} 2790 2791static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 2792 const int offset, 2793 const void *from, 2794 const unsigned int len) 2795{ 2796 memcpy(skb->data + offset, from, len); 2797} 2798 2799void skb_init(void); 2800 2801static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 2802{ 2803 return skb->tstamp; 2804} 2805 2806/** 2807 * skb_get_timestamp - get timestamp from a skb 2808 * @skb: skb to get stamp from 2809 * @stamp: pointer to struct timeval to store stamp in 2810 * 2811 * Timestamps are stored in the skb as offsets to a base timestamp. 2812 * This function converts the offset back to a struct timeval and stores 2813 * it in stamp. 2814 */ 2815static inline void skb_get_timestamp(const struct sk_buff *skb, 2816 struct timeval *stamp) 2817{ 2818 *stamp = ktime_to_timeval(skb->tstamp); 2819} 2820 2821static inline void skb_get_timestampns(const struct sk_buff *skb, 2822 struct timespec *stamp) 2823{ 2824 *stamp = ktime_to_timespec(skb->tstamp); 2825} 2826 2827static inline void __net_timestamp(struct sk_buff *skb) 2828{ 2829 skb->tstamp = ktime_get_real(); 2830} 2831 2832static inline ktime_t net_timedelta(ktime_t t) 2833{ 2834 return ktime_sub(ktime_get_real(), t); 2835} 2836 2837static inline ktime_t net_invalid_timestamp(void) 2838{ 2839 return ktime_set(0, 0); 2840} 2841 2842struct sk_buff *skb_clone_sk(struct sk_buff *skb); 2843 2844#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 2845 2846void skb_clone_tx_timestamp(struct sk_buff *skb); 2847bool skb_defer_rx_timestamp(struct sk_buff *skb); 2848 2849#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 2850 2851static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 2852{ 2853} 2854 2855static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 2856{ 2857 return false; 2858} 2859 2860#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 2861 2862/** 2863 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 2864 * 2865 * PHY drivers may accept clones of transmitted packets for 2866 * timestamping via their phy_driver.txtstamp method. These drivers 2867 * must call this function to return the skb back to the stack, with 2868 * or without a timestamp. 2869 * 2870 * @skb: clone of the the original outgoing packet 2871 * @hwtstamps: hardware time stamps, may be NULL if not available 2872 * 2873 */ 2874void skb_complete_tx_timestamp(struct sk_buff *skb, 2875 struct skb_shared_hwtstamps *hwtstamps); 2876 2877void __skb_tstamp_tx(struct sk_buff *orig_skb, 2878 struct skb_shared_hwtstamps *hwtstamps, 2879 struct sock *sk, int tstype); 2880 2881/** 2882 * skb_tstamp_tx - queue clone of skb with send time stamps 2883 * @orig_skb: the original outgoing packet 2884 * @hwtstamps: hardware time stamps, may be NULL if not available 2885 * 2886 * If the skb has a socket associated, then this function clones the 2887 * skb (thus sharing the actual data and optional structures), stores 2888 * the optional hardware time stamping information (if non NULL) or 2889 * generates a software time stamp (otherwise), then queues the clone 2890 * to the error queue of the socket. Errors are silently ignored. 2891 */ 2892void skb_tstamp_tx(struct sk_buff *orig_skb, 2893 struct skb_shared_hwtstamps *hwtstamps); 2894 2895static inline void sw_tx_timestamp(struct sk_buff *skb) 2896{ 2897 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 2898 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 2899 skb_tstamp_tx(skb, NULL); 2900} 2901 2902/** 2903 * skb_tx_timestamp() - Driver hook for transmit timestamping 2904 * 2905 * Ethernet MAC Drivers should call this function in their hard_xmit() 2906 * function immediately before giving the sk_buff to the MAC hardware. 2907 * 2908 * Specifically, one should make absolutely sure that this function is 2909 * called before TX completion of this packet can trigger. Otherwise 2910 * the packet could potentially already be freed. 2911 * 2912 * @skb: A socket buffer. 2913 */ 2914static inline void skb_tx_timestamp(struct sk_buff *skb) 2915{ 2916 skb_clone_tx_timestamp(skb); 2917 sw_tx_timestamp(skb); 2918} 2919 2920/** 2921 * skb_complete_wifi_ack - deliver skb with wifi status 2922 * 2923 * @skb: the original outgoing packet 2924 * @acked: ack status 2925 * 2926 */ 2927void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 2928 2929__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 2930__sum16 __skb_checksum_complete(struct sk_buff *skb); 2931 2932static inline int skb_csum_unnecessary(const struct sk_buff *skb) 2933{ 2934 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 2935 skb->csum_valid || 2936 (skb->ip_summed == CHECKSUM_PARTIAL && 2937 skb_checksum_start_offset(skb) >= 0)); 2938} 2939 2940/** 2941 * skb_checksum_complete - Calculate checksum of an entire packet 2942 * @skb: packet to process 2943 * 2944 * This function calculates the checksum over the entire packet plus 2945 * the value of skb->csum. The latter can be used to supply the 2946 * checksum of a pseudo header as used by TCP/UDP. It returns the 2947 * checksum. 2948 * 2949 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 2950 * this function can be used to verify that checksum on received 2951 * packets. In that case the function should return zero if the 2952 * checksum is correct. In particular, this function will return zero 2953 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 2954 * hardware has already verified the correctness of the checksum. 2955 */ 2956static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 2957{ 2958 return skb_csum_unnecessary(skb) ? 2959 0 : __skb_checksum_complete(skb); 2960} 2961 2962static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 2963{ 2964 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2965 if (skb->csum_level == 0) 2966 skb->ip_summed = CHECKSUM_NONE; 2967 else 2968 skb->csum_level--; 2969 } 2970} 2971 2972static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 2973{ 2974 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 2975 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 2976 skb->csum_level++; 2977 } else if (skb->ip_summed == CHECKSUM_NONE) { 2978 skb->ip_summed = CHECKSUM_UNNECESSARY; 2979 skb->csum_level = 0; 2980 } 2981} 2982 2983static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 2984{ 2985 /* Mark current checksum as bad (typically called from GRO 2986 * path). In the case that ip_summed is CHECKSUM_NONE 2987 * this must be the first checksum encountered in the packet. 2988 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 2989 * checksum after the last one validated. For UDP, a zero 2990 * checksum can not be marked as bad. 2991 */ 2992 2993 if (skb->ip_summed == CHECKSUM_NONE || 2994 skb->ip_summed == CHECKSUM_UNNECESSARY) 2995 skb->csum_bad = 1; 2996} 2997 2998/* Check if we need to perform checksum complete validation. 2999 * 3000 * Returns true if checksum complete is needed, false otherwise 3001 * (either checksum is unnecessary or zero checksum is allowed). 3002 */ 3003static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3004 bool zero_okay, 3005 __sum16 check) 3006{ 3007 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3008 skb->csum_valid = 1; 3009 __skb_decr_checksum_unnecessary(skb); 3010 return false; 3011 } 3012 3013 return true; 3014} 3015 3016/* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3017 * in checksum_init. 3018 */ 3019#define CHECKSUM_BREAK 76 3020 3021/* Unset checksum-complete 3022 * 3023 * Unset checksum complete can be done when packet is being modified 3024 * (uncompressed for instance) and checksum-complete value is 3025 * invalidated. 3026 */ 3027static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3028{ 3029 if (skb->ip_summed == CHECKSUM_COMPLETE) 3030 skb->ip_summed = CHECKSUM_NONE; 3031} 3032 3033/* Validate (init) checksum based on checksum complete. 3034 * 3035 * Return values: 3036 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3037 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3038 * checksum is stored in skb->csum for use in __skb_checksum_complete 3039 * non-zero: value of invalid checksum 3040 * 3041 */ 3042static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3043 bool complete, 3044 __wsum psum) 3045{ 3046 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3047 if (!csum_fold(csum_add(psum, skb->csum))) { 3048 skb->csum_valid = 1; 3049 return 0; 3050 } 3051 } else if (skb->csum_bad) { 3052 /* ip_summed == CHECKSUM_NONE in this case */ 3053 return 1; 3054 } 3055 3056 skb->csum = psum; 3057 3058 if (complete || skb->len <= CHECKSUM_BREAK) { 3059 __sum16 csum; 3060 3061 csum = __skb_checksum_complete(skb); 3062 skb->csum_valid = !csum; 3063 return csum; 3064 } 3065 3066 return 0; 3067} 3068 3069static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3070{ 3071 return 0; 3072} 3073 3074/* Perform checksum validate (init). Note that this is a macro since we only 3075 * want to calculate the pseudo header which is an input function if necessary. 3076 * First we try to validate without any computation (checksum unnecessary) and 3077 * then calculate based on checksum complete calling the function to compute 3078 * pseudo header. 3079 * 3080 * Return values: 3081 * 0: checksum is validated or try to in skb_checksum_complete 3082 * non-zero: value of invalid checksum 3083 */ 3084#define __skb_checksum_validate(skb, proto, complete, \ 3085 zero_okay, check, compute_pseudo) \ 3086({ \ 3087 __sum16 __ret = 0; \ 3088 skb->csum_valid = 0; \ 3089 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3090 __ret = __skb_checksum_validate_complete(skb, \ 3091 complete, compute_pseudo(skb, proto)); \ 3092 __ret; \ 3093}) 3094 3095#define skb_checksum_init(skb, proto, compute_pseudo) \ 3096 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3097 3098#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3099 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3100 3101#define skb_checksum_validate(skb, proto, compute_pseudo) \ 3102 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3103 3104#define skb_checksum_validate_zero_check(skb, proto, check, \ 3105 compute_pseudo) \ 3106 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3107 3108#define skb_checksum_simple_validate(skb) \ 3109 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3110 3111static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3112{ 3113 return (skb->ip_summed == CHECKSUM_NONE && 3114 skb->csum_valid && !skb->csum_bad); 3115} 3116 3117static inline void __skb_checksum_convert(struct sk_buff *skb, 3118 __sum16 check, __wsum pseudo) 3119{ 3120 skb->csum = ~pseudo; 3121 skb->ip_summed = CHECKSUM_COMPLETE; 3122} 3123 3124#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3125do { \ 3126 if (__skb_checksum_convert_check(skb)) \ 3127 __skb_checksum_convert(skb, check, \ 3128 compute_pseudo(skb, proto)); \ 3129} while (0) 3130 3131static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3132 u16 start, u16 offset) 3133{ 3134 skb->ip_summed = CHECKSUM_PARTIAL; 3135 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3136 skb->csum_offset = offset - start; 3137} 3138 3139/* Update skbuf and packet to reflect the remote checksum offload operation. 3140 * When called, ptr indicates the starting point for skb->csum when 3141 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3142 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3143 */ 3144static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3145 int start, int offset, bool nopartial) 3146{ 3147 __wsum delta; 3148 3149 if (!nopartial) { 3150 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3151 return; 3152 } 3153 3154 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3155 __skb_checksum_complete(skb); 3156 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3157 } 3158 3159 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3160 3161 /* Adjust skb->csum since we changed the packet */ 3162 skb->csum = csum_add(skb->csum, delta); 3163} 3164 3165#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3166void nf_conntrack_destroy(struct nf_conntrack *nfct); 3167static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3168{ 3169 if (nfct && atomic_dec_and_test(&nfct->use)) 3170 nf_conntrack_destroy(nfct); 3171} 3172static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3173{ 3174 if (nfct) 3175 atomic_inc(&nfct->use); 3176} 3177#endif 3178#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3179static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3180{ 3181 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3182 kfree(nf_bridge); 3183} 3184static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3185{ 3186 if (nf_bridge) 3187 atomic_inc(&nf_bridge->use); 3188} 3189#endif /* CONFIG_BRIDGE_NETFILTER */ 3190static inline void nf_reset(struct sk_buff *skb) 3191{ 3192#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3193 nf_conntrack_put(skb->nfct); 3194 skb->nfct = NULL; 3195#endif 3196#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3197 nf_bridge_put(skb->nf_bridge); 3198 skb->nf_bridge = NULL; 3199#endif 3200} 3201 3202static inline void nf_reset_trace(struct sk_buff *skb) 3203{ 3204#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3205 skb->nf_trace = 0; 3206#endif 3207} 3208 3209/* Note: This doesn't put any conntrack and bridge info in dst. */ 3210static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3211 bool copy) 3212{ 3213#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3214 dst->nfct = src->nfct; 3215 nf_conntrack_get(src->nfct); 3216 if (copy) 3217 dst->nfctinfo = src->nfctinfo; 3218#endif 3219#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3220 dst->nf_bridge = src->nf_bridge; 3221 nf_bridge_get(src->nf_bridge); 3222#endif 3223#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3224 if (copy) 3225 dst->nf_trace = src->nf_trace; 3226#endif 3227} 3228 3229static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3230{ 3231#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3232 nf_conntrack_put(dst->nfct); 3233#endif 3234#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3235 nf_bridge_put(dst->nf_bridge); 3236#endif 3237 __nf_copy(dst, src, true); 3238} 3239 3240#ifdef CONFIG_NETWORK_SECMARK 3241static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3242{ 3243 to->secmark = from->secmark; 3244} 3245 3246static inline void skb_init_secmark(struct sk_buff *skb) 3247{ 3248 skb->secmark = 0; 3249} 3250#else 3251static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3252{ } 3253 3254static inline void skb_init_secmark(struct sk_buff *skb) 3255{ } 3256#endif 3257 3258static inline bool skb_irq_freeable(const struct sk_buff *skb) 3259{ 3260 return !skb->destructor && 3261#if IS_ENABLED(CONFIG_XFRM) 3262 !skb->sp && 3263#endif 3264#if IS_ENABLED(CONFIG_NF_CONNTRACK) 3265 !skb->nfct && 3266#endif 3267 !skb->_skb_refdst && 3268 !skb_has_frag_list(skb); 3269} 3270 3271static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3272{ 3273 skb->queue_mapping = queue_mapping; 3274} 3275 3276static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3277{ 3278 return skb->queue_mapping; 3279} 3280 3281static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3282{ 3283 to->queue_mapping = from->queue_mapping; 3284} 3285 3286static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3287{ 3288 skb->queue_mapping = rx_queue + 1; 3289} 3290 3291static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3292{ 3293 return skb->queue_mapping - 1; 3294} 3295 3296static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3297{ 3298 return skb->queue_mapping != 0; 3299} 3300 3301u16 __skb_tx_hash(const struct net_device *dev, struct sk_buff *skb, 3302 unsigned int num_tx_queues); 3303 3304static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3305{ 3306#ifdef CONFIG_XFRM 3307 return skb->sp; 3308#else 3309 return NULL; 3310#endif 3311} 3312 3313/* Keeps track of mac header offset relative to skb->head. 3314 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3315 * For non-tunnel skb it points to skb_mac_header() and for 3316 * tunnel skb it points to outer mac header. 3317 * Keeps track of level of encapsulation of network headers. 3318 */ 3319struct skb_gso_cb { 3320 int mac_offset; 3321 int encap_level; 3322 __u16 csum_start; 3323}; 3324#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)(skb)->cb) 3325 3326static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3327{ 3328 return (skb_mac_header(inner_skb) - inner_skb->head) - 3329 SKB_GSO_CB(inner_skb)->mac_offset; 3330} 3331 3332static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3333{ 3334 int new_headroom, headroom; 3335 int ret; 3336 3337 headroom = skb_headroom(skb); 3338 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3339 if (ret) 3340 return ret; 3341 3342 new_headroom = skb_headroom(skb); 3343 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3344 return 0; 3345} 3346 3347/* Compute the checksum for a gso segment. First compute the checksum value 3348 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3349 * then add in skb->csum (checksum from csum_start to end of packet). 3350 * skb->csum and csum_start are then updated to reflect the checksum of the 3351 * resultant packet starting from the transport header-- the resultant checksum 3352 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3353 * header. 3354 */ 3355static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3356{ 3357 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) - 3358 skb_transport_offset(skb); 3359 __u16 csum; 3360 3361 csum = csum_fold(csum_partial(skb_transport_header(skb), 3362 plen, skb->csum)); 3363 skb->csum = res; 3364 SKB_GSO_CB(skb)->csum_start -= plen; 3365 3366 return csum; 3367} 3368 3369static inline bool skb_is_gso(const struct sk_buff *skb) 3370{ 3371 return skb_shinfo(skb)->gso_size; 3372} 3373 3374/* Note: Should be called only if skb_is_gso(skb) is true */ 3375static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3376{ 3377 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3378} 3379 3380void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3381 3382static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3383{ 3384 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3385 * wanted then gso_type will be set. */ 3386 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3387 3388 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3389 unlikely(shinfo->gso_type == 0)) { 3390 __skb_warn_lro_forwarding(skb); 3391 return true; 3392 } 3393 return false; 3394} 3395 3396static inline void skb_forward_csum(struct sk_buff *skb) 3397{ 3398 /* Unfortunately we don't support this one. Any brave souls? */ 3399 if (skb->ip_summed == CHECKSUM_COMPLETE) 3400 skb->ip_summed = CHECKSUM_NONE; 3401} 3402 3403/** 3404 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3405 * @skb: skb to check 3406 * 3407 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3408 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3409 * use this helper, to document places where we make this assertion. 3410 */ 3411static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3412{ 3413#ifdef DEBUG 3414 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3415#endif 3416} 3417 3418bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3419 3420int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3421 3422u32 skb_get_poff(const struct sk_buff *skb); 3423u32 __skb_get_poff(const struct sk_buff *skb, void *data, 3424 const struct flow_keys *keys, int hlen); 3425 3426/** 3427 * skb_head_is_locked - Determine if the skb->head is locked down 3428 * @skb: skb to check 3429 * 3430 * The head on skbs build around a head frag can be removed if they are 3431 * not cloned. This function returns true if the skb head is locked down 3432 * due to either being allocated via kmalloc, or by being a clone with 3433 * multiple references to the head. 3434 */ 3435static inline bool skb_head_is_locked(const struct sk_buff *skb) 3436{ 3437 return !skb->head_frag || skb_cloned(skb); 3438} 3439 3440/** 3441 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3442 * 3443 * @skb: GSO skb 3444 * 3445 * skb_gso_network_seglen is used to determine the real size of the 3446 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3447 * 3448 * The MAC/L2 header is not accounted for. 3449 */ 3450static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3451{ 3452 unsigned int hdr_len = skb_transport_header(skb) - 3453 skb_network_header(skb); 3454 return hdr_len + skb_gso_transport_seglen(skb); 3455} 3456#endif /* __KERNEL__ */ 3457#endif /* _LINUX_SKBUFF_H */