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_dissector.h> 38#include <linux/splice.h> 39#include <linux/in6.h> 40#include <net/flow.h> 41 42/* A. Checksumming of received packets by device. 43 * 44 * CHECKSUM_NONE: 45 * 46 * Device failed to checksum this packet e.g. due to lack of capabilities. 47 * The packet contains full (though not verified) checksum in packet but 48 * not in skb->csum. Thus, skb->csum is undefined in this case. 49 * 50 * CHECKSUM_UNNECESSARY: 51 * 52 * The hardware you're dealing with doesn't calculate the full checksum 53 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 54 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 55 * if their checksums are okay. skb->csum is still undefined in this case 56 * though. It is a bad option, but, unfortunately, nowadays most vendors do 57 * this. Apparently with the secret goal to sell you new devices, when you 58 * will add new protocol to your host, f.e. IPv6 8) 59 * 60 * CHECKSUM_UNNECESSARY is applicable to following protocols: 61 * TCP: IPv6 and IPv4. 62 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 63 * zero UDP checksum for either IPv4 or IPv6, the networking stack 64 * may perform further validation in this case. 65 * GRE: only if the checksum is present in the header. 66 * SCTP: indicates the CRC in SCTP header has been validated. 67 * 68 * skb->csum_level indicates the number of consecutive checksums found in 69 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 70 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 71 * and a device is able to verify the checksums for UDP (possibly zero), 72 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 73 * two. If the device were only able to verify the UDP checksum and not 74 * GRE, either because it doesn't support GRE checksum of because GRE 75 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 76 * not considered in this case). 77 * 78 * CHECKSUM_COMPLETE: 79 * 80 * This is the most generic way. The device supplied checksum of the _whole_ 81 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 82 * hardware doesn't need to parse L3/L4 headers to implement this. 83 * 84 * Note: Even if device supports only some protocols, but is able to produce 85 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 86 * 87 * CHECKSUM_PARTIAL: 88 * 89 * A checksum is set up to be offloaded to a device as described in the 90 * output description for CHECKSUM_PARTIAL. This may occur on a packet 91 * received directly from another Linux OS, e.g., a virtualized Linux kernel 92 * on the same host, or it may be set in the input path in GRO or remote 93 * checksum offload. For the purposes of checksum verification, the checksum 94 * referred to by skb->csum_start + skb->csum_offset and any preceding 95 * checksums in the packet are considered verified. Any checksums in the 96 * packet that are after the checksum being offloaded are not considered to 97 * be verified. 98 * 99 * B. Checksumming on output. 100 * 101 * CHECKSUM_NONE: 102 * 103 * The skb was already checksummed by the protocol, or a checksum is not 104 * required. 105 * 106 * CHECKSUM_PARTIAL: 107 * 108 * The device is required to checksum the packet as seen by hard_start_xmit() 109 * from skb->csum_start up to the end, and to record/write the checksum at 110 * offset skb->csum_start + skb->csum_offset. 111 * 112 * The device must show its capabilities in dev->features, set up at device 113 * setup time, e.g. netdev_features.h: 114 * 115 * NETIF_F_HW_CSUM - It's a clever device, it's able to checksum everything. 116 * NETIF_F_IP_CSUM - Device is dumb, it's able to checksum only TCP/UDP over 117 * IPv4. Sigh. Vendors like this way for an unknown reason. 118 * Though, see comment above about CHECKSUM_UNNECESSARY. 8) 119 * NETIF_F_IPV6_CSUM - About as dumb as the last one but does IPv6 instead. 120 * NETIF_F_... - Well, you get the picture. 121 * 122 * CHECKSUM_UNNECESSARY: 123 * 124 * Normally, the device will do per protocol specific checksumming. Protocol 125 * implementations that do not want the NIC to perform the checksum 126 * calculation should use this flag in their outgoing skbs. 127 * 128 * NETIF_F_FCOE_CRC - This indicates that the device can do FCoE FC CRC 129 * offload. Correspondingly, the FCoE protocol driver 130 * stack should use CHECKSUM_UNNECESSARY. 131 * 132 * Any questions? No questions, good. --ANK 133 */ 134 135/* Don't change this without changing skb_csum_unnecessary! */ 136#define CHECKSUM_NONE 0 137#define CHECKSUM_UNNECESSARY 1 138#define CHECKSUM_COMPLETE 2 139#define CHECKSUM_PARTIAL 3 140 141/* Maximum value in skb->csum_level */ 142#define SKB_MAX_CSUM_LEVEL 3 143 144#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 145#define SKB_WITH_OVERHEAD(X) \ 146 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 147#define SKB_MAX_ORDER(X, ORDER) \ 148 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 149#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 150#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 151 152/* return minimum truesize of one skb containing X bytes of data */ 153#define SKB_TRUESIZE(X) ((X) + \ 154 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 155 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 156 157struct net_device; 158struct scatterlist; 159struct pipe_inode_info; 160struct iov_iter; 161struct napi_struct; 162 163#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 164struct nf_conntrack { 165 atomic_t use; 166}; 167#endif 168 169#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 170struct nf_bridge_info { 171 atomic_t use; 172 enum { 173 BRNF_PROTO_UNCHANGED, 174 BRNF_PROTO_8021Q, 175 BRNF_PROTO_PPPOE 176 } orig_proto:8; 177 u8 pkt_otherhost:1; 178 u8 in_prerouting:1; 179 u8 bridged_dnat:1; 180 __u16 frag_max_size; 181 struct net_device *physindev; 182 183 /* always valid & non-NULL from FORWARD on, for physdev match */ 184 struct net_device *physoutdev; 185 union { 186 /* prerouting: detect dnat in orig/reply direction */ 187 __be32 ipv4_daddr; 188 struct in6_addr ipv6_daddr; 189 190 /* after prerouting + nat detected: store original source 191 * mac since neigh resolution overwrites it, only used while 192 * skb is out in neigh layer. 193 */ 194 char neigh_header[8]; 195 }; 196}; 197#endif 198 199struct sk_buff_head { 200 /* These two members must be first. */ 201 struct sk_buff *next; 202 struct sk_buff *prev; 203 204 __u32 qlen; 205 spinlock_t lock; 206}; 207 208struct sk_buff; 209 210/* To allow 64K frame to be packed as single skb without frag_list we 211 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 212 * buffers which do not start on a page boundary. 213 * 214 * Since GRO uses frags we allocate at least 16 regardless of page 215 * size. 216 */ 217#if (65536/PAGE_SIZE + 1) < 16 218#define MAX_SKB_FRAGS 16UL 219#else 220#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 221#endif 222extern int sysctl_max_skb_frags; 223 224typedef struct skb_frag_struct skb_frag_t; 225 226struct skb_frag_struct { 227 struct { 228 struct page *p; 229 } page; 230#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 231 __u32 page_offset; 232 __u32 size; 233#else 234 __u16 page_offset; 235 __u16 size; 236#endif 237}; 238 239static inline unsigned int skb_frag_size(const skb_frag_t *frag) 240{ 241 return frag->size; 242} 243 244static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 245{ 246 frag->size = size; 247} 248 249static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 250{ 251 frag->size += delta; 252} 253 254static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 255{ 256 frag->size -= delta; 257} 258 259#define HAVE_HW_TIME_STAMP 260 261/** 262 * struct skb_shared_hwtstamps - hardware time stamps 263 * @hwtstamp: hardware time stamp transformed into duration 264 * since arbitrary point in time 265 * 266 * Software time stamps generated by ktime_get_real() are stored in 267 * skb->tstamp. 268 * 269 * hwtstamps can only be compared against other hwtstamps from 270 * the same device. 271 * 272 * This structure is attached to packets as part of the 273 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 274 */ 275struct skb_shared_hwtstamps { 276 ktime_t hwtstamp; 277}; 278 279/* Definitions for tx_flags in struct skb_shared_info */ 280enum { 281 /* generate hardware time stamp */ 282 SKBTX_HW_TSTAMP = 1 << 0, 283 284 /* generate software time stamp when queueing packet to NIC */ 285 SKBTX_SW_TSTAMP = 1 << 1, 286 287 /* device driver is going to provide hardware time stamp */ 288 SKBTX_IN_PROGRESS = 1 << 2, 289 290 /* device driver supports TX zero-copy buffers */ 291 SKBTX_DEV_ZEROCOPY = 1 << 3, 292 293 /* generate wifi status information (where possible) */ 294 SKBTX_WIFI_STATUS = 1 << 4, 295 296 /* This indicates at least one fragment might be overwritten 297 * (as in vmsplice(), sendfile() ...) 298 * If we need to compute a TX checksum, we'll need to copy 299 * all frags to avoid possible bad checksum 300 */ 301 SKBTX_SHARED_FRAG = 1 << 5, 302 303 /* generate software time stamp when entering packet scheduling */ 304 SKBTX_SCHED_TSTAMP = 1 << 6, 305 306 /* generate software timestamp on peer data acknowledgment */ 307 SKBTX_ACK_TSTAMP = 1 << 7, 308}; 309 310#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 311 SKBTX_SCHED_TSTAMP | \ 312 SKBTX_ACK_TSTAMP) 313#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 314 315/* 316 * The callback notifies userspace to release buffers when skb DMA is done in 317 * lower device, the skb last reference should be 0 when calling this. 318 * The zerocopy_success argument is true if zero copy transmit occurred, 319 * false on data copy or out of memory error caused by data copy attempt. 320 * The ctx field is used to track device context. 321 * The desc field is used to track userspace buffer index. 322 */ 323struct ubuf_info { 324 void (*callback)(struct ubuf_info *, bool zerocopy_success); 325 void *ctx; 326 unsigned long desc; 327}; 328 329/* This data is invariant across clones and lives at 330 * the end of the header data, ie. at skb->end. 331 */ 332struct skb_shared_info { 333 unsigned char nr_frags; 334 __u8 tx_flags; 335 unsigned short gso_size; 336 /* Warning: this field is not always filled in (UFO)! */ 337 unsigned short gso_segs; 338 unsigned short gso_type; 339 struct sk_buff *frag_list; 340 struct skb_shared_hwtstamps hwtstamps; 341 u32 tskey; 342 __be32 ip6_frag_id; 343 344 /* 345 * Warning : all fields before dataref are cleared in __alloc_skb() 346 */ 347 atomic_t dataref; 348 349 /* Intermediate layers must ensure that destructor_arg 350 * remains valid until skb destructor */ 351 void * destructor_arg; 352 353 /* must be last field, see pskb_expand_head() */ 354 skb_frag_t frags[MAX_SKB_FRAGS]; 355}; 356 357/* We divide dataref into two halves. The higher 16 bits hold references 358 * to the payload part of skb->data. The lower 16 bits hold references to 359 * the entire skb->data. A clone of a headerless skb holds the length of 360 * the header in skb->hdr_len. 361 * 362 * All users must obey the rule that the skb->data reference count must be 363 * greater than or equal to the payload reference count. 364 * 365 * Holding a reference to the payload part means that the user does not 366 * care about modifications to the header part of skb->data. 367 */ 368#define SKB_DATAREF_SHIFT 16 369#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 370 371 372enum { 373 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 374 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 375 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 376}; 377 378enum { 379 SKB_GSO_TCPV4 = 1 << 0, 380 SKB_GSO_UDP = 1 << 1, 381 382 /* This indicates the skb is from an untrusted source. */ 383 SKB_GSO_DODGY = 1 << 2, 384 385 /* This indicates the tcp segment has CWR set. */ 386 SKB_GSO_TCP_ECN = 1 << 3, 387 388 SKB_GSO_TCPV6 = 1 << 4, 389 390 SKB_GSO_FCOE = 1 << 5, 391 392 SKB_GSO_GRE = 1 << 6, 393 394 SKB_GSO_GRE_CSUM = 1 << 7, 395 396 SKB_GSO_IPIP = 1 << 8, 397 398 SKB_GSO_SIT = 1 << 9, 399 400 SKB_GSO_UDP_TUNNEL = 1 << 10, 401 402 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 403 404 SKB_GSO_TUNNEL_REMCSUM = 1 << 12, 405}; 406 407#if BITS_PER_LONG > 32 408#define NET_SKBUFF_DATA_USES_OFFSET 1 409#endif 410 411#ifdef NET_SKBUFF_DATA_USES_OFFSET 412typedef unsigned int sk_buff_data_t; 413#else 414typedef unsigned char *sk_buff_data_t; 415#endif 416 417/** 418 * struct skb_mstamp - multi resolution time stamps 419 * @stamp_us: timestamp in us resolution 420 * @stamp_jiffies: timestamp in jiffies 421 */ 422struct skb_mstamp { 423 union { 424 u64 v64; 425 struct { 426 u32 stamp_us; 427 u32 stamp_jiffies; 428 }; 429 }; 430}; 431 432/** 433 * skb_mstamp_get - get current timestamp 434 * @cl: place to store timestamps 435 */ 436static inline void skb_mstamp_get(struct skb_mstamp *cl) 437{ 438 u64 val = local_clock(); 439 440 do_div(val, NSEC_PER_USEC); 441 cl->stamp_us = (u32)val; 442 cl->stamp_jiffies = (u32)jiffies; 443} 444 445/** 446 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp 447 * @t1: pointer to newest sample 448 * @t0: pointer to oldest sample 449 */ 450static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1, 451 const struct skb_mstamp *t0) 452{ 453 s32 delta_us = t1->stamp_us - t0->stamp_us; 454 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies; 455 456 /* If delta_us is negative, this might be because interval is too big, 457 * or local_clock() drift is too big : fallback using jiffies. 458 */ 459 if (delta_us <= 0 || 460 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ))) 461 462 delta_us = jiffies_to_usecs(delta_jiffies); 463 464 return delta_us; 465} 466 467static inline bool skb_mstamp_after(const struct skb_mstamp *t1, 468 const struct skb_mstamp *t0) 469{ 470 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies; 471 472 if (!diff) 473 diff = t1->stamp_us - t0->stamp_us; 474 return diff > 0; 475} 476 477/** 478 * struct sk_buff - socket buffer 479 * @next: Next buffer in list 480 * @prev: Previous buffer in list 481 * @tstamp: Time we arrived/left 482 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 483 * @sk: Socket we are owned by 484 * @dev: Device we arrived on/are leaving by 485 * @cb: Control buffer. Free for use by every layer. Put private vars here 486 * @_skb_refdst: destination entry (with norefcount bit) 487 * @sp: the security path, used for xfrm 488 * @len: Length of actual data 489 * @data_len: Data length 490 * @mac_len: Length of link layer header 491 * @hdr_len: writable header length of cloned skb 492 * @csum: Checksum (must include start/offset pair) 493 * @csum_start: Offset from skb->head where checksumming should start 494 * @csum_offset: Offset from csum_start where checksum should be stored 495 * @priority: Packet queueing priority 496 * @ignore_df: allow local fragmentation 497 * @cloned: Head may be cloned (check refcnt to be sure) 498 * @ip_summed: Driver fed us an IP checksum 499 * @nohdr: Payload reference only, must not modify header 500 * @nfctinfo: Relationship of this skb to the connection 501 * @pkt_type: Packet class 502 * @fclone: skbuff clone status 503 * @ipvs_property: skbuff is owned by ipvs 504 * @peeked: this packet has been seen already, so stats have been 505 * done for it, don't do them again 506 * @nf_trace: netfilter packet trace flag 507 * @protocol: Packet protocol from driver 508 * @destructor: Destruct function 509 * @nfct: Associated connection, if any 510 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 511 * @skb_iif: ifindex of device we arrived on 512 * @tc_index: Traffic control index 513 * @tc_verd: traffic control verdict 514 * @hash: the packet hash 515 * @queue_mapping: Queue mapping for multiqueue devices 516 * @xmit_more: More SKBs are pending for this queue 517 * @ndisc_nodetype: router type (from link layer) 518 * @ooo_okay: allow the mapping of a socket to a queue to be changed 519 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 520 * ports. 521 * @sw_hash: indicates hash was computed in software stack 522 * @wifi_acked_valid: wifi_acked was set 523 * @wifi_acked: whether frame was acked on wifi or not 524 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 525 * @napi_id: id of the NAPI struct this skb came from 526 * @secmark: security marking 527 * @offload_fwd_mark: fwding offload mark 528 * @mark: Generic packet mark 529 * @vlan_proto: vlan encapsulation protocol 530 * @vlan_tci: vlan tag control information 531 * @inner_protocol: Protocol (encapsulation) 532 * @inner_transport_header: Inner transport layer header (encapsulation) 533 * @inner_network_header: Network layer header (encapsulation) 534 * @inner_mac_header: Link layer header (encapsulation) 535 * @transport_header: Transport layer header 536 * @network_header: Network layer header 537 * @mac_header: Link layer header 538 * @tail: Tail pointer 539 * @end: End pointer 540 * @head: Head of buffer 541 * @data: Data head pointer 542 * @truesize: Buffer size 543 * @users: User count - see {datagram,tcp}.c 544 */ 545 546struct sk_buff { 547 union { 548 struct { 549 /* These two members must be first. */ 550 struct sk_buff *next; 551 struct sk_buff *prev; 552 553 union { 554 ktime_t tstamp; 555 struct skb_mstamp skb_mstamp; 556 }; 557 }; 558 struct rb_node rbnode; /* used in netem & tcp stack */ 559 }; 560 struct sock *sk; 561 struct net_device *dev; 562 563 /* 564 * This is the control buffer. It is free to use for every 565 * layer. Please put your private variables there. If you 566 * want to keep them across layers you have to do a skb_clone() 567 * first. This is owned by whoever has the skb queued ATM. 568 */ 569 char cb[48] __aligned(8); 570 571 unsigned long _skb_refdst; 572 void (*destructor)(struct sk_buff *skb); 573#ifdef CONFIG_XFRM 574 struct sec_path *sp; 575#endif 576#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 577 struct nf_conntrack *nfct; 578#endif 579#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 580 struct nf_bridge_info *nf_bridge; 581#endif 582 unsigned int len, 583 data_len; 584 __u16 mac_len, 585 hdr_len; 586 587 /* Following fields are _not_ copied in __copy_skb_header() 588 * Note that queue_mapping is here mostly to fill a hole. 589 */ 590 kmemcheck_bitfield_begin(flags1); 591 __u16 queue_mapping; 592 __u8 cloned:1, 593 nohdr:1, 594 fclone:2, 595 peeked:1, 596 head_frag:1, 597 xmit_more:1; 598 /* one bit hole */ 599 kmemcheck_bitfield_end(flags1); 600 601 /* fields enclosed in headers_start/headers_end are copied 602 * using a single memcpy() in __copy_skb_header() 603 */ 604 /* private: */ 605 __u32 headers_start[0]; 606 /* public: */ 607 608/* if you move pkt_type around you also must adapt those constants */ 609#ifdef __BIG_ENDIAN_BITFIELD 610#define PKT_TYPE_MAX (7 << 5) 611#else 612#define PKT_TYPE_MAX 7 613#endif 614#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 615 616 __u8 __pkt_type_offset[0]; 617 __u8 pkt_type:3; 618 __u8 pfmemalloc:1; 619 __u8 ignore_df:1; 620 __u8 nfctinfo:3; 621 622 __u8 nf_trace:1; 623 __u8 ip_summed:2; 624 __u8 ooo_okay:1; 625 __u8 l4_hash:1; 626 __u8 sw_hash:1; 627 __u8 wifi_acked_valid:1; 628 __u8 wifi_acked:1; 629 630 __u8 no_fcs:1; 631 /* Indicates the inner headers are valid in the skbuff. */ 632 __u8 encapsulation:1; 633 __u8 encap_hdr_csum:1; 634 __u8 csum_valid:1; 635 __u8 csum_complete_sw:1; 636 __u8 csum_level:2; 637 __u8 csum_bad:1; 638 639#ifdef CONFIG_IPV6_NDISC_NODETYPE 640 __u8 ndisc_nodetype:2; 641#endif 642 __u8 ipvs_property:1; 643 __u8 inner_protocol_type:1; 644 __u8 remcsum_offload:1; 645 /* 3 or 5 bit hole */ 646 647#ifdef CONFIG_NET_SCHED 648 __u16 tc_index; /* traffic control index */ 649#ifdef CONFIG_NET_CLS_ACT 650 __u16 tc_verd; /* traffic control verdict */ 651#endif 652#endif 653 654 union { 655 __wsum csum; 656 struct { 657 __u16 csum_start; 658 __u16 csum_offset; 659 }; 660 }; 661 __u32 priority; 662 int skb_iif; 663 __u32 hash; 664 __be16 vlan_proto; 665 __u16 vlan_tci; 666#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 667 union { 668 unsigned int napi_id; 669 unsigned int sender_cpu; 670 }; 671#endif 672 union { 673#ifdef CONFIG_NETWORK_SECMARK 674 __u32 secmark; 675#endif 676#ifdef CONFIG_NET_SWITCHDEV 677 __u32 offload_fwd_mark; 678#endif 679 }; 680 681 union { 682 __u32 mark; 683 __u32 reserved_tailroom; 684 }; 685 686 union { 687 __be16 inner_protocol; 688 __u8 inner_ipproto; 689 }; 690 691 __u16 inner_transport_header; 692 __u16 inner_network_header; 693 __u16 inner_mac_header; 694 695 __be16 protocol; 696 __u16 transport_header; 697 __u16 network_header; 698 __u16 mac_header; 699 700 /* private: */ 701 __u32 headers_end[0]; 702 /* public: */ 703 704 /* These elements must be at the end, see alloc_skb() for details. */ 705 sk_buff_data_t tail; 706 sk_buff_data_t end; 707 unsigned char *head, 708 *data; 709 unsigned int truesize; 710 atomic_t users; 711}; 712 713#ifdef __KERNEL__ 714/* 715 * Handling routines are only of interest to the kernel 716 */ 717#include <linux/slab.h> 718 719 720#define SKB_ALLOC_FCLONE 0x01 721#define SKB_ALLOC_RX 0x02 722#define SKB_ALLOC_NAPI 0x04 723 724/* Returns true if the skb was allocated from PFMEMALLOC reserves */ 725static inline bool skb_pfmemalloc(const struct sk_buff *skb) 726{ 727 return unlikely(skb->pfmemalloc); 728} 729 730/* 731 * skb might have a dst pointer attached, refcounted or not. 732 * _skb_refdst low order bit is set if refcount was _not_ taken 733 */ 734#define SKB_DST_NOREF 1UL 735#define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 736 737/** 738 * skb_dst - returns skb dst_entry 739 * @skb: buffer 740 * 741 * Returns skb dst_entry, regardless of reference taken or not. 742 */ 743static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 744{ 745 /* If refdst was not refcounted, check we still are in a 746 * rcu_read_lock section 747 */ 748 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 749 !rcu_read_lock_held() && 750 !rcu_read_lock_bh_held()); 751 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 752} 753 754/** 755 * skb_dst_set - sets skb dst 756 * @skb: buffer 757 * @dst: dst entry 758 * 759 * Sets skb dst, assuming a reference was taken on dst and should 760 * be released by skb_dst_drop() 761 */ 762static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 763{ 764 skb->_skb_refdst = (unsigned long)dst; 765} 766 767/** 768 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 769 * @skb: buffer 770 * @dst: dst entry 771 * 772 * Sets skb dst, assuming a reference was not taken on dst. 773 * If dst entry is cached, we do not take reference and dst_release 774 * will be avoided by refdst_drop. If dst entry is not cached, we take 775 * reference, so that last dst_release can destroy the dst immediately. 776 */ 777static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 778{ 779 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 780 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 781} 782 783/** 784 * skb_dst_is_noref - Test if skb dst isn't refcounted 785 * @skb: buffer 786 */ 787static inline bool skb_dst_is_noref(const struct sk_buff *skb) 788{ 789 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 790} 791 792static inline struct rtable *skb_rtable(const struct sk_buff *skb) 793{ 794 return (struct rtable *)skb_dst(skb); 795} 796 797void kfree_skb(struct sk_buff *skb); 798void kfree_skb_list(struct sk_buff *segs); 799void skb_tx_error(struct sk_buff *skb); 800void consume_skb(struct sk_buff *skb); 801void __kfree_skb(struct sk_buff *skb); 802extern struct kmem_cache *skbuff_head_cache; 803 804void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 805bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 806 bool *fragstolen, int *delta_truesize); 807 808struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 809 int node); 810struct sk_buff *__build_skb(void *data, unsigned int frag_size); 811struct sk_buff *build_skb(void *data, unsigned int frag_size); 812static inline struct sk_buff *alloc_skb(unsigned int size, 813 gfp_t priority) 814{ 815 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 816} 817 818struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 819 unsigned long data_len, 820 int max_page_order, 821 int *errcode, 822 gfp_t gfp_mask); 823 824/* Layout of fast clones : [skb1][skb2][fclone_ref] */ 825struct sk_buff_fclones { 826 struct sk_buff skb1; 827 828 struct sk_buff skb2; 829 830 atomic_t fclone_ref; 831}; 832 833/** 834 * skb_fclone_busy - check if fclone is busy 835 * @skb: buffer 836 * 837 * Returns true is skb is a fast clone, and its clone is not freed. 838 * Some drivers call skb_orphan() in their ndo_start_xmit(), 839 * so we also check that this didnt happen. 840 */ 841static inline bool skb_fclone_busy(const struct sock *sk, 842 const struct sk_buff *skb) 843{ 844 const struct sk_buff_fclones *fclones; 845 846 fclones = container_of(skb, struct sk_buff_fclones, skb1); 847 848 return skb->fclone == SKB_FCLONE_ORIG && 849 atomic_read(&fclones->fclone_ref) > 1 && 850 fclones->skb2.sk == sk; 851} 852 853static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 854 gfp_t priority) 855{ 856 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 857} 858 859struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 860static inline struct sk_buff *alloc_skb_head(gfp_t priority) 861{ 862 return __alloc_skb_head(priority, -1); 863} 864 865struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 866int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 867struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 868struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 869struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 870 gfp_t gfp_mask, bool fclone); 871static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 872 gfp_t gfp_mask) 873{ 874 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 875} 876 877int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 878struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 879 unsigned int headroom); 880struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 881 int newtailroom, gfp_t priority); 882int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 883 int offset, int len); 884int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, 885 int len); 886int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 887int skb_pad(struct sk_buff *skb, int pad); 888#define dev_kfree_skb(a) consume_skb(a) 889 890int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 891 int getfrag(void *from, char *to, int offset, 892 int len, int odd, struct sk_buff *skb), 893 void *from, int length); 894 895int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 896 int offset, size_t size); 897 898struct skb_seq_state { 899 __u32 lower_offset; 900 __u32 upper_offset; 901 __u32 frag_idx; 902 __u32 stepped_offset; 903 struct sk_buff *root_skb; 904 struct sk_buff *cur_skb; 905 __u8 *frag_data; 906}; 907 908void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 909 unsigned int to, struct skb_seq_state *st); 910unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 911 struct skb_seq_state *st); 912void skb_abort_seq_read(struct skb_seq_state *st); 913 914unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 915 unsigned int to, struct ts_config *config); 916 917/* 918 * Packet hash types specify the type of hash in skb_set_hash. 919 * 920 * Hash types refer to the protocol layer addresses which are used to 921 * construct a packet's hash. The hashes are used to differentiate or identify 922 * flows of the protocol layer for the hash type. Hash types are either 923 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 924 * 925 * Properties of hashes: 926 * 927 * 1) Two packets in different flows have different hash values 928 * 2) Two packets in the same flow should have the same hash value 929 * 930 * A hash at a higher layer is considered to be more specific. A driver should 931 * set the most specific hash possible. 932 * 933 * A driver cannot indicate a more specific hash than the layer at which a hash 934 * was computed. For instance an L3 hash cannot be set as an L4 hash. 935 * 936 * A driver may indicate a hash level which is less specific than the 937 * actual layer the hash was computed on. For instance, a hash computed 938 * at L4 may be considered an L3 hash. This should only be done if the 939 * driver can't unambiguously determine that the HW computed the hash at 940 * the higher layer. Note that the "should" in the second property above 941 * permits this. 942 */ 943enum pkt_hash_types { 944 PKT_HASH_TYPE_NONE, /* Undefined type */ 945 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 946 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 947 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 948}; 949 950static inline void skb_clear_hash(struct sk_buff *skb) 951{ 952 skb->hash = 0; 953 skb->sw_hash = 0; 954 skb->l4_hash = 0; 955} 956 957static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 958{ 959 if (!skb->l4_hash) 960 skb_clear_hash(skb); 961} 962 963static inline void 964__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 965{ 966 skb->l4_hash = is_l4; 967 skb->sw_hash = is_sw; 968 skb->hash = hash; 969} 970 971static inline void 972skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 973{ 974 /* Used by drivers to set hash from HW */ 975 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 976} 977 978static inline void 979__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 980{ 981 __skb_set_hash(skb, hash, true, is_l4); 982} 983 984void __skb_get_hash(struct sk_buff *skb); 985u32 skb_get_poff(const struct sk_buff *skb); 986u32 __skb_get_poff(const struct sk_buff *skb, void *data, 987 const struct flow_keys *keys, int hlen); 988__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 989 void *data, int hlen_proto); 990 991static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 992 int thoff, u8 ip_proto) 993{ 994 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 995} 996 997void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 998 const struct flow_dissector_key *key, 999 unsigned int key_count); 1000 1001bool __skb_flow_dissect(const struct sk_buff *skb, 1002 struct flow_dissector *flow_dissector, 1003 void *target_container, 1004 void *data, __be16 proto, int nhoff, int hlen, 1005 unsigned int flags); 1006 1007static inline bool skb_flow_dissect(const struct sk_buff *skb, 1008 struct flow_dissector *flow_dissector, 1009 void *target_container, unsigned int flags) 1010{ 1011 return __skb_flow_dissect(skb, flow_dissector, target_container, 1012 NULL, 0, 0, 0, flags); 1013} 1014 1015static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1016 struct flow_keys *flow, 1017 unsigned int flags) 1018{ 1019 memset(flow, 0, sizeof(*flow)); 1020 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1021 NULL, 0, 0, 0, flags); 1022} 1023 1024static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow, 1025 void *data, __be16 proto, 1026 int nhoff, int hlen, 1027 unsigned int flags) 1028{ 1029 memset(flow, 0, sizeof(*flow)); 1030 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow, 1031 data, proto, nhoff, hlen, flags); 1032} 1033 1034static inline __u32 skb_get_hash(struct sk_buff *skb) 1035{ 1036 if (!skb->l4_hash && !skb->sw_hash) 1037 __skb_get_hash(skb); 1038 1039 return skb->hash; 1040} 1041 1042__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6); 1043 1044static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1045{ 1046 if (!skb->l4_hash && !skb->sw_hash) { 1047 struct flow_keys keys; 1048 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1049 1050 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1051 } 1052 1053 return skb->hash; 1054} 1055 1056__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl); 1057 1058static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4) 1059{ 1060 if (!skb->l4_hash && !skb->sw_hash) { 1061 struct flow_keys keys; 1062 __u32 hash = __get_hash_from_flowi4(fl4, &keys); 1063 1064 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1065 } 1066 1067 return skb->hash; 1068} 1069 1070__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1071 1072static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1073{ 1074 return skb->hash; 1075} 1076 1077static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1078{ 1079 to->hash = from->hash; 1080 to->sw_hash = from->sw_hash; 1081 to->l4_hash = from->l4_hash; 1082}; 1083 1084static inline void skb_sender_cpu_clear(struct sk_buff *skb) 1085{ 1086#ifdef CONFIG_XPS 1087 skb->sender_cpu = 0; 1088#endif 1089} 1090 1091#ifdef NET_SKBUFF_DATA_USES_OFFSET 1092static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1093{ 1094 return skb->head + skb->end; 1095} 1096 1097static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1098{ 1099 return skb->end; 1100} 1101#else 1102static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1103{ 1104 return skb->end; 1105} 1106 1107static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1108{ 1109 return skb->end - skb->head; 1110} 1111#endif 1112 1113/* Internal */ 1114#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1115 1116static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1117{ 1118 return &skb_shinfo(skb)->hwtstamps; 1119} 1120 1121/** 1122 * skb_queue_empty - check if a queue is empty 1123 * @list: queue head 1124 * 1125 * Returns true if the queue is empty, false otherwise. 1126 */ 1127static inline int skb_queue_empty(const struct sk_buff_head *list) 1128{ 1129 return list->next == (const struct sk_buff *) list; 1130} 1131 1132/** 1133 * skb_queue_is_last - check if skb is the last entry in the queue 1134 * @list: queue head 1135 * @skb: buffer 1136 * 1137 * Returns true if @skb is the last buffer on the list. 1138 */ 1139static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1140 const struct sk_buff *skb) 1141{ 1142 return skb->next == (const struct sk_buff *) list; 1143} 1144 1145/** 1146 * skb_queue_is_first - check if skb is the first entry in the queue 1147 * @list: queue head 1148 * @skb: buffer 1149 * 1150 * Returns true if @skb is the first buffer on the list. 1151 */ 1152static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1153 const struct sk_buff *skb) 1154{ 1155 return skb->prev == (const struct sk_buff *) list; 1156} 1157 1158/** 1159 * skb_queue_next - return the next packet in the queue 1160 * @list: queue head 1161 * @skb: current buffer 1162 * 1163 * Return the next packet in @list after @skb. It is only valid to 1164 * call this if skb_queue_is_last() evaluates to false. 1165 */ 1166static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1167 const struct sk_buff *skb) 1168{ 1169 /* This BUG_ON may seem severe, but if we just return then we 1170 * are going to dereference garbage. 1171 */ 1172 BUG_ON(skb_queue_is_last(list, skb)); 1173 return skb->next; 1174} 1175 1176/** 1177 * skb_queue_prev - return the prev packet in the queue 1178 * @list: queue head 1179 * @skb: current buffer 1180 * 1181 * Return the prev packet in @list before @skb. It is only valid to 1182 * call this if skb_queue_is_first() evaluates to false. 1183 */ 1184static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1185 const struct sk_buff *skb) 1186{ 1187 /* This BUG_ON may seem severe, but if we just return then we 1188 * are going to dereference garbage. 1189 */ 1190 BUG_ON(skb_queue_is_first(list, skb)); 1191 return skb->prev; 1192} 1193 1194/** 1195 * skb_get - reference buffer 1196 * @skb: buffer to reference 1197 * 1198 * Makes another reference to a socket buffer and returns a pointer 1199 * to the buffer. 1200 */ 1201static inline struct sk_buff *skb_get(struct sk_buff *skb) 1202{ 1203 atomic_inc(&skb->users); 1204 return skb; 1205} 1206 1207/* 1208 * If users == 1, we are the only owner and are can avoid redundant 1209 * atomic change. 1210 */ 1211 1212/** 1213 * skb_cloned - is the buffer a clone 1214 * @skb: buffer to check 1215 * 1216 * Returns true if the buffer was generated with skb_clone() and is 1217 * one of multiple shared copies of the buffer. Cloned buffers are 1218 * shared data so must not be written to under normal circumstances. 1219 */ 1220static inline int skb_cloned(const struct sk_buff *skb) 1221{ 1222 return skb->cloned && 1223 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1224} 1225 1226static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1227{ 1228 might_sleep_if(gfpflags_allow_blocking(pri)); 1229 1230 if (skb_cloned(skb)) 1231 return pskb_expand_head(skb, 0, 0, pri); 1232 1233 return 0; 1234} 1235 1236/** 1237 * skb_header_cloned - is the header a clone 1238 * @skb: buffer to check 1239 * 1240 * Returns true if modifying the header part of the buffer requires 1241 * the data to be copied. 1242 */ 1243static inline int skb_header_cloned(const struct sk_buff *skb) 1244{ 1245 int dataref; 1246 1247 if (!skb->cloned) 1248 return 0; 1249 1250 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1251 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1252 return dataref != 1; 1253} 1254 1255/** 1256 * skb_header_release - release reference to header 1257 * @skb: buffer to operate on 1258 * 1259 * Drop a reference to the header part of the buffer. This is done 1260 * by acquiring a payload reference. You must not read from the header 1261 * part of skb->data after this. 1262 * Note : Check if you can use __skb_header_release() instead. 1263 */ 1264static inline void skb_header_release(struct sk_buff *skb) 1265{ 1266 BUG_ON(skb->nohdr); 1267 skb->nohdr = 1; 1268 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1269} 1270 1271/** 1272 * __skb_header_release - release reference to header 1273 * @skb: buffer to operate on 1274 * 1275 * Variant of skb_header_release() assuming skb is private to caller. 1276 * We can avoid one atomic operation. 1277 */ 1278static inline void __skb_header_release(struct sk_buff *skb) 1279{ 1280 skb->nohdr = 1; 1281 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1282} 1283 1284 1285/** 1286 * skb_shared - is the buffer shared 1287 * @skb: buffer to check 1288 * 1289 * Returns true if more than one person has a reference to this 1290 * buffer. 1291 */ 1292static inline int skb_shared(const struct sk_buff *skb) 1293{ 1294 return atomic_read(&skb->users) != 1; 1295} 1296 1297/** 1298 * skb_share_check - check if buffer is shared and if so clone it 1299 * @skb: buffer to check 1300 * @pri: priority for memory allocation 1301 * 1302 * If the buffer is shared the buffer is cloned and the old copy 1303 * drops a reference. A new clone with a single reference is returned. 1304 * If the buffer is not shared the original buffer is returned. When 1305 * being called from interrupt status or with spinlocks held pri must 1306 * be GFP_ATOMIC. 1307 * 1308 * NULL is returned on a memory allocation failure. 1309 */ 1310static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1311{ 1312 might_sleep_if(gfpflags_allow_blocking(pri)); 1313 if (skb_shared(skb)) { 1314 struct sk_buff *nskb = skb_clone(skb, pri); 1315 1316 if (likely(nskb)) 1317 consume_skb(skb); 1318 else 1319 kfree_skb(skb); 1320 skb = nskb; 1321 } 1322 return skb; 1323} 1324 1325/* 1326 * Copy shared buffers into a new sk_buff. We effectively do COW on 1327 * packets to handle cases where we have a local reader and forward 1328 * and a couple of other messy ones. The normal one is tcpdumping 1329 * a packet thats being forwarded. 1330 */ 1331 1332/** 1333 * skb_unshare - make a copy of a shared buffer 1334 * @skb: buffer to check 1335 * @pri: priority for memory allocation 1336 * 1337 * If the socket buffer is a clone then this function creates a new 1338 * copy of the data, drops a reference count on the old copy and returns 1339 * the new copy with the reference count at 1. If the buffer is not a clone 1340 * the original buffer is returned. When called with a spinlock held or 1341 * from interrupt state @pri must be %GFP_ATOMIC 1342 * 1343 * %NULL is returned on a memory allocation failure. 1344 */ 1345static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1346 gfp_t pri) 1347{ 1348 might_sleep_if(gfpflags_allow_blocking(pri)); 1349 if (skb_cloned(skb)) { 1350 struct sk_buff *nskb = skb_copy(skb, pri); 1351 1352 /* Free our shared copy */ 1353 if (likely(nskb)) 1354 consume_skb(skb); 1355 else 1356 kfree_skb(skb); 1357 skb = nskb; 1358 } 1359 return skb; 1360} 1361 1362/** 1363 * skb_peek - peek at the head of an &sk_buff_head 1364 * @list_: list to peek at 1365 * 1366 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1367 * be careful with this one. A peek leaves the buffer on the 1368 * list and someone else may run off with it. You must hold 1369 * the appropriate locks or have a private queue to do this. 1370 * 1371 * Returns %NULL for an empty list or a pointer to the head element. 1372 * The reference count is not incremented and the reference is therefore 1373 * volatile. Use with caution. 1374 */ 1375static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1376{ 1377 struct sk_buff *skb = list_->next; 1378 1379 if (skb == (struct sk_buff *)list_) 1380 skb = NULL; 1381 return skb; 1382} 1383 1384/** 1385 * skb_peek_next - peek skb following the given one from a queue 1386 * @skb: skb to start from 1387 * @list_: list to peek at 1388 * 1389 * Returns %NULL when the end of the list is met or a pointer to the 1390 * next element. The reference count is not incremented and the 1391 * reference is therefore volatile. Use with caution. 1392 */ 1393static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1394 const struct sk_buff_head *list_) 1395{ 1396 struct sk_buff *next = skb->next; 1397 1398 if (next == (struct sk_buff *)list_) 1399 next = NULL; 1400 return next; 1401} 1402 1403/** 1404 * skb_peek_tail - peek at the tail of an &sk_buff_head 1405 * @list_: list to peek at 1406 * 1407 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1408 * be careful with this one. A peek leaves the buffer on the 1409 * list and someone else may run off with it. You must hold 1410 * the appropriate locks or have a private queue to do this. 1411 * 1412 * Returns %NULL for an empty list or a pointer to the tail element. 1413 * The reference count is not incremented and the reference is therefore 1414 * volatile. Use with caution. 1415 */ 1416static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1417{ 1418 struct sk_buff *skb = list_->prev; 1419 1420 if (skb == (struct sk_buff *)list_) 1421 skb = NULL; 1422 return skb; 1423 1424} 1425 1426/** 1427 * skb_queue_len - get queue length 1428 * @list_: list to measure 1429 * 1430 * Return the length of an &sk_buff queue. 1431 */ 1432static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1433{ 1434 return list_->qlen; 1435} 1436 1437/** 1438 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1439 * @list: queue to initialize 1440 * 1441 * This initializes only the list and queue length aspects of 1442 * an sk_buff_head object. This allows to initialize the list 1443 * aspects of an sk_buff_head without reinitializing things like 1444 * the spinlock. It can also be used for on-stack sk_buff_head 1445 * objects where the spinlock is known to not be used. 1446 */ 1447static inline void __skb_queue_head_init(struct sk_buff_head *list) 1448{ 1449 list->prev = list->next = (struct sk_buff *)list; 1450 list->qlen = 0; 1451} 1452 1453/* 1454 * This function creates a split out lock class for each invocation; 1455 * this is needed for now since a whole lot of users of the skb-queue 1456 * infrastructure in drivers have different locking usage (in hardirq) 1457 * than the networking core (in softirq only). In the long run either the 1458 * network layer or drivers should need annotation to consolidate the 1459 * main types of usage into 3 classes. 1460 */ 1461static inline void skb_queue_head_init(struct sk_buff_head *list) 1462{ 1463 spin_lock_init(&list->lock); 1464 __skb_queue_head_init(list); 1465} 1466 1467static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1468 struct lock_class_key *class) 1469{ 1470 skb_queue_head_init(list); 1471 lockdep_set_class(&list->lock, class); 1472} 1473 1474/* 1475 * Insert an sk_buff on a list. 1476 * 1477 * The "__skb_xxxx()" functions are the non-atomic ones that 1478 * can only be called with interrupts disabled. 1479 */ 1480void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1481 struct sk_buff_head *list); 1482static inline void __skb_insert(struct sk_buff *newsk, 1483 struct sk_buff *prev, struct sk_buff *next, 1484 struct sk_buff_head *list) 1485{ 1486 newsk->next = next; 1487 newsk->prev = prev; 1488 next->prev = prev->next = newsk; 1489 list->qlen++; 1490} 1491 1492static inline void __skb_queue_splice(const struct sk_buff_head *list, 1493 struct sk_buff *prev, 1494 struct sk_buff *next) 1495{ 1496 struct sk_buff *first = list->next; 1497 struct sk_buff *last = list->prev; 1498 1499 first->prev = prev; 1500 prev->next = first; 1501 1502 last->next = next; 1503 next->prev = last; 1504} 1505 1506/** 1507 * skb_queue_splice - join two skb lists, this is designed for stacks 1508 * @list: the new list to add 1509 * @head: the place to add it in the first list 1510 */ 1511static inline void skb_queue_splice(const struct sk_buff_head *list, 1512 struct sk_buff_head *head) 1513{ 1514 if (!skb_queue_empty(list)) { 1515 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1516 head->qlen += list->qlen; 1517 } 1518} 1519 1520/** 1521 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1522 * @list: the new list to add 1523 * @head: the place to add it in the first list 1524 * 1525 * The list at @list is reinitialised 1526 */ 1527static inline void skb_queue_splice_init(struct sk_buff_head *list, 1528 struct sk_buff_head *head) 1529{ 1530 if (!skb_queue_empty(list)) { 1531 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1532 head->qlen += list->qlen; 1533 __skb_queue_head_init(list); 1534 } 1535} 1536 1537/** 1538 * skb_queue_splice_tail - join two skb lists, each list being a queue 1539 * @list: the new list to add 1540 * @head: the place to add it in the first list 1541 */ 1542static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1543 struct sk_buff_head *head) 1544{ 1545 if (!skb_queue_empty(list)) { 1546 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1547 head->qlen += list->qlen; 1548 } 1549} 1550 1551/** 1552 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1553 * @list: the new list to add 1554 * @head: the place to add it in the first list 1555 * 1556 * Each of the lists is a queue. 1557 * The list at @list is reinitialised 1558 */ 1559static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1560 struct sk_buff_head *head) 1561{ 1562 if (!skb_queue_empty(list)) { 1563 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1564 head->qlen += list->qlen; 1565 __skb_queue_head_init(list); 1566 } 1567} 1568 1569/** 1570 * __skb_queue_after - queue a buffer at the list head 1571 * @list: list to use 1572 * @prev: place after this buffer 1573 * @newsk: buffer to queue 1574 * 1575 * Queue a buffer int the middle of a list. This function takes no locks 1576 * and you must therefore hold required locks before calling it. 1577 * 1578 * A buffer cannot be placed on two lists at the same time. 1579 */ 1580static inline void __skb_queue_after(struct sk_buff_head *list, 1581 struct sk_buff *prev, 1582 struct sk_buff *newsk) 1583{ 1584 __skb_insert(newsk, prev, prev->next, list); 1585} 1586 1587void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1588 struct sk_buff_head *list); 1589 1590static inline void __skb_queue_before(struct sk_buff_head *list, 1591 struct sk_buff *next, 1592 struct sk_buff *newsk) 1593{ 1594 __skb_insert(newsk, next->prev, next, list); 1595} 1596 1597/** 1598 * __skb_queue_head - queue a buffer at the list head 1599 * @list: list to use 1600 * @newsk: buffer to queue 1601 * 1602 * Queue a buffer at the start of a list. This function takes no locks 1603 * and you must therefore hold required locks before calling it. 1604 * 1605 * A buffer cannot be placed on two lists at the same time. 1606 */ 1607void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1608static inline void __skb_queue_head(struct sk_buff_head *list, 1609 struct sk_buff *newsk) 1610{ 1611 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1612} 1613 1614/** 1615 * __skb_queue_tail - queue a buffer at the list tail 1616 * @list: list to use 1617 * @newsk: buffer to queue 1618 * 1619 * Queue a buffer at the end of a list. This function takes no locks 1620 * and you must therefore hold required locks before calling it. 1621 * 1622 * A buffer cannot be placed on two lists at the same time. 1623 */ 1624void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1625static inline void __skb_queue_tail(struct sk_buff_head *list, 1626 struct sk_buff *newsk) 1627{ 1628 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1629} 1630 1631/* 1632 * remove sk_buff from list. _Must_ be called atomically, and with 1633 * the list known.. 1634 */ 1635void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1636static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1637{ 1638 struct sk_buff *next, *prev; 1639 1640 list->qlen--; 1641 next = skb->next; 1642 prev = skb->prev; 1643 skb->next = skb->prev = NULL; 1644 next->prev = prev; 1645 prev->next = next; 1646} 1647 1648/** 1649 * __skb_dequeue - remove from the head of the queue 1650 * @list: list to dequeue from 1651 * 1652 * Remove the head of the list. This function does not take any locks 1653 * so must be used with appropriate locks held only. The head item is 1654 * returned or %NULL if the list is empty. 1655 */ 1656struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1657static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1658{ 1659 struct sk_buff *skb = skb_peek(list); 1660 if (skb) 1661 __skb_unlink(skb, list); 1662 return skb; 1663} 1664 1665/** 1666 * __skb_dequeue_tail - remove from the tail of the queue 1667 * @list: list to dequeue from 1668 * 1669 * Remove the tail of the list. This function does not take any locks 1670 * so must be used with appropriate locks held only. The tail item is 1671 * returned or %NULL if the list is empty. 1672 */ 1673struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1674static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1675{ 1676 struct sk_buff *skb = skb_peek_tail(list); 1677 if (skb) 1678 __skb_unlink(skb, list); 1679 return skb; 1680} 1681 1682 1683static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1684{ 1685 return skb->data_len; 1686} 1687 1688static inline unsigned int skb_headlen(const struct sk_buff *skb) 1689{ 1690 return skb->len - skb->data_len; 1691} 1692 1693static inline int skb_pagelen(const struct sk_buff *skb) 1694{ 1695 int i, len = 0; 1696 1697 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1698 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1699 return len + skb_headlen(skb); 1700} 1701 1702/** 1703 * __skb_fill_page_desc - initialise a paged fragment in an skb 1704 * @skb: buffer containing fragment to be initialised 1705 * @i: paged fragment index to initialise 1706 * @page: the page to use for this fragment 1707 * @off: the offset to the data with @page 1708 * @size: the length of the data 1709 * 1710 * Initialises the @i'th fragment of @skb to point to &size bytes at 1711 * offset @off within @page. 1712 * 1713 * Does not take any additional reference on the fragment. 1714 */ 1715static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1716 struct page *page, int off, int size) 1717{ 1718 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1719 1720 /* 1721 * Propagate page pfmemalloc to the skb if we can. The problem is 1722 * that not all callers have unique ownership of the page but rely 1723 * on page_is_pfmemalloc doing the right thing(tm). 1724 */ 1725 frag->page.p = page; 1726 frag->page_offset = off; 1727 skb_frag_size_set(frag, size); 1728 1729 page = compound_head(page); 1730 if (page_is_pfmemalloc(page)) 1731 skb->pfmemalloc = true; 1732} 1733 1734/** 1735 * skb_fill_page_desc - initialise a paged fragment in an skb 1736 * @skb: buffer containing fragment to be initialised 1737 * @i: paged fragment index to initialise 1738 * @page: the page to use for this fragment 1739 * @off: the offset to the data with @page 1740 * @size: the length of the data 1741 * 1742 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1743 * @skb to point to @size bytes at offset @off within @page. In 1744 * addition updates @skb such that @i is the last fragment. 1745 * 1746 * Does not take any additional reference on the fragment. 1747 */ 1748static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1749 struct page *page, int off, int size) 1750{ 1751 __skb_fill_page_desc(skb, i, page, off, size); 1752 skb_shinfo(skb)->nr_frags = i + 1; 1753} 1754 1755void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1756 int size, unsigned int truesize); 1757 1758void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1759 unsigned int truesize); 1760 1761#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1762#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1763#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1764 1765#ifdef NET_SKBUFF_DATA_USES_OFFSET 1766static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1767{ 1768 return skb->head + skb->tail; 1769} 1770 1771static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1772{ 1773 skb->tail = skb->data - skb->head; 1774} 1775 1776static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1777{ 1778 skb_reset_tail_pointer(skb); 1779 skb->tail += offset; 1780} 1781 1782#else /* NET_SKBUFF_DATA_USES_OFFSET */ 1783static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1784{ 1785 return skb->tail; 1786} 1787 1788static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1789{ 1790 skb->tail = skb->data; 1791} 1792 1793static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1794{ 1795 skb->tail = skb->data + offset; 1796} 1797 1798#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1799 1800/* 1801 * Add data to an sk_buff 1802 */ 1803unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1804unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1805static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1806{ 1807 unsigned char *tmp = skb_tail_pointer(skb); 1808 SKB_LINEAR_ASSERT(skb); 1809 skb->tail += len; 1810 skb->len += len; 1811 return tmp; 1812} 1813 1814unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1815static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1816{ 1817 skb->data -= len; 1818 skb->len += len; 1819 return skb->data; 1820} 1821 1822unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1823static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1824{ 1825 skb->len -= len; 1826 BUG_ON(skb->len < skb->data_len); 1827 return skb->data += len; 1828} 1829 1830static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1831{ 1832 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1833} 1834 1835unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1836 1837static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1838{ 1839 if (len > skb_headlen(skb) && 1840 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1841 return NULL; 1842 skb->len -= len; 1843 return skb->data += len; 1844} 1845 1846static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1847{ 1848 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1849} 1850 1851static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1852{ 1853 if (likely(len <= skb_headlen(skb))) 1854 return 1; 1855 if (unlikely(len > skb->len)) 1856 return 0; 1857 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1858} 1859 1860/** 1861 * skb_headroom - bytes at buffer head 1862 * @skb: buffer to check 1863 * 1864 * Return the number of bytes of free space at the head of an &sk_buff. 1865 */ 1866static inline unsigned int skb_headroom(const struct sk_buff *skb) 1867{ 1868 return skb->data - skb->head; 1869} 1870 1871/** 1872 * skb_tailroom - bytes at buffer end 1873 * @skb: buffer to check 1874 * 1875 * Return the number of bytes of free space at the tail of an sk_buff 1876 */ 1877static inline int skb_tailroom(const struct sk_buff *skb) 1878{ 1879 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1880} 1881 1882/** 1883 * skb_availroom - bytes at buffer end 1884 * @skb: buffer to check 1885 * 1886 * Return the number of bytes of free space at the tail of an sk_buff 1887 * allocated by sk_stream_alloc() 1888 */ 1889static inline int skb_availroom(const struct sk_buff *skb) 1890{ 1891 if (skb_is_nonlinear(skb)) 1892 return 0; 1893 1894 return skb->end - skb->tail - skb->reserved_tailroom; 1895} 1896 1897/** 1898 * skb_reserve - adjust headroom 1899 * @skb: buffer to alter 1900 * @len: bytes to move 1901 * 1902 * Increase the headroom of an empty &sk_buff by reducing the tail 1903 * room. This is only allowed for an empty buffer. 1904 */ 1905static inline void skb_reserve(struct sk_buff *skb, int len) 1906{ 1907 skb->data += len; 1908 skb->tail += len; 1909} 1910 1911/** 1912 * skb_tailroom_reserve - adjust reserved_tailroom 1913 * @skb: buffer to alter 1914 * @mtu: maximum amount of headlen permitted 1915 * @needed_tailroom: minimum amount of reserved_tailroom 1916 * 1917 * Set reserved_tailroom so that headlen can be as large as possible but 1918 * not larger than mtu and tailroom cannot be smaller than 1919 * needed_tailroom. 1920 * The required headroom should already have been reserved before using 1921 * this function. 1922 */ 1923static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 1924 unsigned int needed_tailroom) 1925{ 1926 SKB_LINEAR_ASSERT(skb); 1927 if (mtu < skb_tailroom(skb) - needed_tailroom) 1928 /* use at most mtu */ 1929 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 1930 else 1931 /* use up to all available space */ 1932 skb->reserved_tailroom = needed_tailroom; 1933} 1934 1935#define ENCAP_TYPE_ETHER 0 1936#define ENCAP_TYPE_IPPROTO 1 1937 1938static inline void skb_set_inner_protocol(struct sk_buff *skb, 1939 __be16 protocol) 1940{ 1941 skb->inner_protocol = protocol; 1942 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 1943} 1944 1945static inline void skb_set_inner_ipproto(struct sk_buff *skb, 1946 __u8 ipproto) 1947{ 1948 skb->inner_ipproto = ipproto; 1949 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 1950} 1951 1952static inline void skb_reset_inner_headers(struct sk_buff *skb) 1953{ 1954 skb->inner_mac_header = skb->mac_header; 1955 skb->inner_network_header = skb->network_header; 1956 skb->inner_transport_header = skb->transport_header; 1957} 1958 1959static inline void skb_reset_mac_len(struct sk_buff *skb) 1960{ 1961 skb->mac_len = skb->network_header - skb->mac_header; 1962} 1963 1964static inline unsigned char *skb_inner_transport_header(const struct sk_buff 1965 *skb) 1966{ 1967 return skb->head + skb->inner_transport_header; 1968} 1969 1970static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 1971{ 1972 skb->inner_transport_header = skb->data - skb->head; 1973} 1974 1975static inline void skb_set_inner_transport_header(struct sk_buff *skb, 1976 const int offset) 1977{ 1978 skb_reset_inner_transport_header(skb); 1979 skb->inner_transport_header += offset; 1980} 1981 1982static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 1983{ 1984 return skb->head + skb->inner_network_header; 1985} 1986 1987static inline void skb_reset_inner_network_header(struct sk_buff *skb) 1988{ 1989 skb->inner_network_header = skb->data - skb->head; 1990} 1991 1992static inline void skb_set_inner_network_header(struct sk_buff *skb, 1993 const int offset) 1994{ 1995 skb_reset_inner_network_header(skb); 1996 skb->inner_network_header += offset; 1997} 1998 1999static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2000{ 2001 return skb->head + skb->inner_mac_header; 2002} 2003 2004static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2005{ 2006 skb->inner_mac_header = skb->data - skb->head; 2007} 2008 2009static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2010 const int offset) 2011{ 2012 skb_reset_inner_mac_header(skb); 2013 skb->inner_mac_header += offset; 2014} 2015static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2016{ 2017 return skb->transport_header != (typeof(skb->transport_header))~0U; 2018} 2019 2020static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2021{ 2022 return skb->head + skb->transport_header; 2023} 2024 2025static inline void skb_reset_transport_header(struct sk_buff *skb) 2026{ 2027 skb->transport_header = skb->data - skb->head; 2028} 2029 2030static inline void skb_set_transport_header(struct sk_buff *skb, 2031 const int offset) 2032{ 2033 skb_reset_transport_header(skb); 2034 skb->transport_header += offset; 2035} 2036 2037static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2038{ 2039 return skb->head + skb->network_header; 2040} 2041 2042static inline void skb_reset_network_header(struct sk_buff *skb) 2043{ 2044 skb->network_header = skb->data - skb->head; 2045} 2046 2047static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2048{ 2049 skb_reset_network_header(skb); 2050 skb->network_header += offset; 2051} 2052 2053static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2054{ 2055 return skb->head + skb->mac_header; 2056} 2057 2058static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2059{ 2060 return skb->mac_header != (typeof(skb->mac_header))~0U; 2061} 2062 2063static inline void skb_reset_mac_header(struct sk_buff *skb) 2064{ 2065 skb->mac_header = skb->data - skb->head; 2066} 2067 2068static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2069{ 2070 skb_reset_mac_header(skb); 2071 skb->mac_header += offset; 2072} 2073 2074static inline void skb_pop_mac_header(struct sk_buff *skb) 2075{ 2076 skb->mac_header = skb->network_header; 2077} 2078 2079static inline void skb_probe_transport_header(struct sk_buff *skb, 2080 const int offset_hint) 2081{ 2082 struct flow_keys keys; 2083 2084 if (skb_transport_header_was_set(skb)) 2085 return; 2086 else if (skb_flow_dissect_flow_keys(skb, &keys, 0)) 2087 skb_set_transport_header(skb, keys.control.thoff); 2088 else 2089 skb_set_transport_header(skb, offset_hint); 2090} 2091 2092static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2093{ 2094 if (skb_mac_header_was_set(skb)) { 2095 const unsigned char *old_mac = skb_mac_header(skb); 2096 2097 skb_set_mac_header(skb, -skb->mac_len); 2098 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2099 } 2100} 2101 2102static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2103{ 2104 return skb->csum_start - skb_headroom(skb); 2105} 2106 2107static inline int skb_transport_offset(const struct sk_buff *skb) 2108{ 2109 return skb_transport_header(skb) - skb->data; 2110} 2111 2112static inline u32 skb_network_header_len(const struct sk_buff *skb) 2113{ 2114 return skb->transport_header - skb->network_header; 2115} 2116 2117static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2118{ 2119 return skb->inner_transport_header - skb->inner_network_header; 2120} 2121 2122static inline int skb_network_offset(const struct sk_buff *skb) 2123{ 2124 return skb_network_header(skb) - skb->data; 2125} 2126 2127static inline int skb_inner_network_offset(const struct sk_buff *skb) 2128{ 2129 return skb_inner_network_header(skb) - skb->data; 2130} 2131 2132static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2133{ 2134 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2135} 2136 2137/* 2138 * CPUs often take a performance hit when accessing unaligned memory 2139 * locations. The actual performance hit varies, it can be small if the 2140 * hardware handles it or large if we have to take an exception and fix it 2141 * in software. 2142 * 2143 * Since an ethernet header is 14 bytes network drivers often end up with 2144 * the IP header at an unaligned offset. The IP header can be aligned by 2145 * shifting the start of the packet by 2 bytes. Drivers should do this 2146 * with: 2147 * 2148 * skb_reserve(skb, NET_IP_ALIGN); 2149 * 2150 * The downside to this alignment of the IP header is that the DMA is now 2151 * unaligned. On some architectures the cost of an unaligned DMA is high 2152 * and this cost outweighs the gains made by aligning the IP header. 2153 * 2154 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2155 * to be overridden. 2156 */ 2157#ifndef NET_IP_ALIGN 2158#define NET_IP_ALIGN 2 2159#endif 2160 2161/* 2162 * The networking layer reserves some headroom in skb data (via 2163 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2164 * the header has to grow. In the default case, if the header has to grow 2165 * 32 bytes or less we avoid the reallocation. 2166 * 2167 * Unfortunately this headroom changes the DMA alignment of the resulting 2168 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2169 * on some architectures. An architecture can override this value, 2170 * perhaps setting it to a cacheline in size (since that will maintain 2171 * cacheline alignment of the DMA). It must be a power of 2. 2172 * 2173 * Various parts of the networking layer expect at least 32 bytes of 2174 * headroom, you should not reduce this. 2175 * 2176 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2177 * to reduce average number of cache lines per packet. 2178 * get_rps_cpus() for example only access one 64 bytes aligned block : 2179 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2180 */ 2181#ifndef NET_SKB_PAD 2182#define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2183#endif 2184 2185int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2186 2187static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2188{ 2189 if (unlikely(skb_is_nonlinear(skb))) { 2190 WARN_ON(1); 2191 return; 2192 } 2193 skb->len = len; 2194 skb_set_tail_pointer(skb, len); 2195} 2196 2197void skb_trim(struct sk_buff *skb, unsigned int len); 2198 2199static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2200{ 2201 if (skb->data_len) 2202 return ___pskb_trim(skb, len); 2203 __skb_trim(skb, len); 2204 return 0; 2205} 2206 2207static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2208{ 2209 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2210} 2211 2212/** 2213 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2214 * @skb: buffer to alter 2215 * @len: new length 2216 * 2217 * This is identical to pskb_trim except that the caller knows that 2218 * the skb is not cloned so we should never get an error due to out- 2219 * of-memory. 2220 */ 2221static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2222{ 2223 int err = pskb_trim(skb, len); 2224 BUG_ON(err); 2225} 2226 2227/** 2228 * skb_orphan - orphan a buffer 2229 * @skb: buffer to orphan 2230 * 2231 * If a buffer currently has an owner then we call the owner's 2232 * destructor function and make the @skb unowned. The buffer continues 2233 * to exist but is no longer charged to its former owner. 2234 */ 2235static inline void skb_orphan(struct sk_buff *skb) 2236{ 2237 if (skb->destructor) { 2238 skb->destructor(skb); 2239 skb->destructor = NULL; 2240 skb->sk = NULL; 2241 } else { 2242 BUG_ON(skb->sk); 2243 } 2244} 2245 2246/** 2247 * skb_orphan_frags - orphan the frags contained in a buffer 2248 * @skb: buffer to orphan frags from 2249 * @gfp_mask: allocation mask for replacement pages 2250 * 2251 * For each frag in the SKB which needs a destructor (i.e. has an 2252 * owner) create a copy of that frag and release the original 2253 * page by calling the destructor. 2254 */ 2255static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2256{ 2257 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2258 return 0; 2259 return skb_copy_ubufs(skb, gfp_mask); 2260} 2261 2262/** 2263 * __skb_queue_purge - empty a list 2264 * @list: list to empty 2265 * 2266 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2267 * the list and one reference dropped. This function does not take the 2268 * list lock and the caller must hold the relevant locks to use it. 2269 */ 2270void skb_queue_purge(struct sk_buff_head *list); 2271static inline void __skb_queue_purge(struct sk_buff_head *list) 2272{ 2273 struct sk_buff *skb; 2274 while ((skb = __skb_dequeue(list)) != NULL) 2275 kfree_skb(skb); 2276} 2277 2278void *netdev_alloc_frag(unsigned int fragsz); 2279 2280struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2281 gfp_t gfp_mask); 2282 2283/** 2284 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2285 * @dev: network device to receive on 2286 * @length: length to allocate 2287 * 2288 * Allocate a new &sk_buff and assign it a usage count of one. The 2289 * buffer has unspecified headroom built in. Users should allocate 2290 * the headroom they think they need without accounting for the 2291 * built in space. The built in space is used for optimisations. 2292 * 2293 * %NULL is returned if there is no free memory. Although this function 2294 * allocates memory it can be called from an interrupt. 2295 */ 2296static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2297 unsigned int length) 2298{ 2299 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2300} 2301 2302/* legacy helper around __netdev_alloc_skb() */ 2303static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2304 gfp_t gfp_mask) 2305{ 2306 return __netdev_alloc_skb(NULL, length, gfp_mask); 2307} 2308 2309/* legacy helper around netdev_alloc_skb() */ 2310static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2311{ 2312 return netdev_alloc_skb(NULL, length); 2313} 2314 2315 2316static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2317 unsigned int length, gfp_t gfp) 2318{ 2319 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2320 2321 if (NET_IP_ALIGN && skb) 2322 skb_reserve(skb, NET_IP_ALIGN); 2323 return skb; 2324} 2325 2326static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2327 unsigned int length) 2328{ 2329 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2330} 2331 2332static inline void skb_free_frag(void *addr) 2333{ 2334 __free_page_frag(addr); 2335} 2336 2337void *napi_alloc_frag(unsigned int fragsz); 2338struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2339 unsigned int length, gfp_t gfp_mask); 2340static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2341 unsigned int length) 2342{ 2343 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2344} 2345 2346/** 2347 * __dev_alloc_pages - allocate page for network Rx 2348 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2349 * @order: size of the allocation 2350 * 2351 * Allocate a new page. 2352 * 2353 * %NULL is returned if there is no free memory. 2354*/ 2355static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2356 unsigned int order) 2357{ 2358 /* This piece of code contains several assumptions. 2359 * 1. This is for device Rx, therefor a cold page is preferred. 2360 * 2. The expectation is the user wants a compound page. 2361 * 3. If requesting a order 0 page it will not be compound 2362 * due to the check to see if order has a value in prep_new_page 2363 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2364 * code in gfp_to_alloc_flags that should be enforcing this. 2365 */ 2366 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2367 2368 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2369} 2370 2371static inline struct page *dev_alloc_pages(unsigned int order) 2372{ 2373 return __dev_alloc_pages(GFP_ATOMIC, order); 2374} 2375 2376/** 2377 * __dev_alloc_page - allocate a page for network Rx 2378 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2379 * 2380 * Allocate a new page. 2381 * 2382 * %NULL is returned if there is no free memory. 2383 */ 2384static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2385{ 2386 return __dev_alloc_pages(gfp_mask, 0); 2387} 2388 2389static inline struct page *dev_alloc_page(void) 2390{ 2391 return __dev_alloc_page(GFP_ATOMIC); 2392} 2393 2394/** 2395 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2396 * @page: The page that was allocated from skb_alloc_page 2397 * @skb: The skb that may need pfmemalloc set 2398 */ 2399static inline void skb_propagate_pfmemalloc(struct page *page, 2400 struct sk_buff *skb) 2401{ 2402 if (page_is_pfmemalloc(page)) 2403 skb->pfmemalloc = true; 2404} 2405 2406/** 2407 * skb_frag_page - retrieve the page referred to by a paged fragment 2408 * @frag: the paged fragment 2409 * 2410 * Returns the &struct page associated with @frag. 2411 */ 2412static inline struct page *skb_frag_page(const skb_frag_t *frag) 2413{ 2414 return frag->page.p; 2415} 2416 2417/** 2418 * __skb_frag_ref - take an addition reference on a paged fragment. 2419 * @frag: the paged fragment 2420 * 2421 * Takes an additional reference on the paged fragment @frag. 2422 */ 2423static inline void __skb_frag_ref(skb_frag_t *frag) 2424{ 2425 get_page(skb_frag_page(frag)); 2426} 2427 2428/** 2429 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2430 * @skb: the buffer 2431 * @f: the fragment offset. 2432 * 2433 * Takes an additional reference on the @f'th paged fragment of @skb. 2434 */ 2435static inline void skb_frag_ref(struct sk_buff *skb, int f) 2436{ 2437 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2438} 2439 2440/** 2441 * __skb_frag_unref - release a reference on a paged fragment. 2442 * @frag: the paged fragment 2443 * 2444 * Releases a reference on the paged fragment @frag. 2445 */ 2446static inline void __skb_frag_unref(skb_frag_t *frag) 2447{ 2448 put_page(skb_frag_page(frag)); 2449} 2450 2451/** 2452 * skb_frag_unref - release a reference on a paged fragment of an skb. 2453 * @skb: the buffer 2454 * @f: the fragment offset 2455 * 2456 * Releases a reference on the @f'th paged fragment of @skb. 2457 */ 2458static inline void skb_frag_unref(struct sk_buff *skb, int f) 2459{ 2460 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2461} 2462 2463/** 2464 * skb_frag_address - gets the address of the data contained in a paged fragment 2465 * @frag: the paged fragment buffer 2466 * 2467 * Returns the address of the data within @frag. The page must already 2468 * be mapped. 2469 */ 2470static inline void *skb_frag_address(const skb_frag_t *frag) 2471{ 2472 return page_address(skb_frag_page(frag)) + frag->page_offset; 2473} 2474 2475/** 2476 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2477 * @frag: the paged fragment buffer 2478 * 2479 * Returns the address of the data within @frag. Checks that the page 2480 * is mapped and returns %NULL otherwise. 2481 */ 2482static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2483{ 2484 void *ptr = page_address(skb_frag_page(frag)); 2485 if (unlikely(!ptr)) 2486 return NULL; 2487 2488 return ptr + frag->page_offset; 2489} 2490 2491/** 2492 * __skb_frag_set_page - sets the page contained in a paged fragment 2493 * @frag: the paged fragment 2494 * @page: the page to set 2495 * 2496 * Sets the fragment @frag to contain @page. 2497 */ 2498static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2499{ 2500 frag->page.p = page; 2501} 2502 2503/** 2504 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2505 * @skb: the buffer 2506 * @f: the fragment offset 2507 * @page: the page to set 2508 * 2509 * Sets the @f'th fragment of @skb to contain @page. 2510 */ 2511static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2512 struct page *page) 2513{ 2514 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2515} 2516 2517bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2518 2519/** 2520 * skb_frag_dma_map - maps a paged fragment via the DMA API 2521 * @dev: the device to map the fragment to 2522 * @frag: the paged fragment to map 2523 * @offset: the offset within the fragment (starting at the 2524 * fragment's own offset) 2525 * @size: the number of bytes to map 2526 * @dir: the direction of the mapping (%PCI_DMA_*) 2527 * 2528 * Maps the page associated with @frag to @device. 2529 */ 2530static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2531 const skb_frag_t *frag, 2532 size_t offset, size_t size, 2533 enum dma_data_direction dir) 2534{ 2535 return dma_map_page(dev, skb_frag_page(frag), 2536 frag->page_offset + offset, size, dir); 2537} 2538 2539static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2540 gfp_t gfp_mask) 2541{ 2542 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2543} 2544 2545 2546static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2547 gfp_t gfp_mask) 2548{ 2549 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2550} 2551 2552 2553/** 2554 * skb_clone_writable - is the header of a clone writable 2555 * @skb: buffer to check 2556 * @len: length up to which to write 2557 * 2558 * Returns true if modifying the header part of the cloned buffer 2559 * does not requires the data to be copied. 2560 */ 2561static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2562{ 2563 return !skb_header_cloned(skb) && 2564 skb_headroom(skb) + len <= skb->hdr_len; 2565} 2566 2567static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2568 int cloned) 2569{ 2570 int delta = 0; 2571 2572 if (headroom > skb_headroom(skb)) 2573 delta = headroom - skb_headroom(skb); 2574 2575 if (delta || cloned) 2576 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2577 GFP_ATOMIC); 2578 return 0; 2579} 2580 2581/** 2582 * skb_cow - copy header of skb when it is required 2583 * @skb: buffer to cow 2584 * @headroom: needed headroom 2585 * 2586 * If the skb passed lacks sufficient headroom or its data part 2587 * is shared, data is reallocated. If reallocation fails, an error 2588 * is returned and original skb is not changed. 2589 * 2590 * The result is skb with writable area skb->head...skb->tail 2591 * and at least @headroom of space at head. 2592 */ 2593static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2594{ 2595 return __skb_cow(skb, headroom, skb_cloned(skb)); 2596} 2597 2598/** 2599 * skb_cow_head - skb_cow but only making the head writable 2600 * @skb: buffer to cow 2601 * @headroom: needed headroom 2602 * 2603 * This function is identical to skb_cow except that we replace the 2604 * skb_cloned check by skb_header_cloned. It should be used when 2605 * you only need to push on some header and do not need to modify 2606 * the data. 2607 */ 2608static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2609{ 2610 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2611} 2612 2613/** 2614 * skb_padto - pad an skbuff up to a minimal size 2615 * @skb: buffer to pad 2616 * @len: minimal length 2617 * 2618 * Pads up a buffer to ensure the trailing bytes exist and are 2619 * blanked. If the buffer already contains sufficient data it 2620 * is untouched. Otherwise it is extended. Returns zero on 2621 * success. The skb is freed on error. 2622 */ 2623static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2624{ 2625 unsigned int size = skb->len; 2626 if (likely(size >= len)) 2627 return 0; 2628 return skb_pad(skb, len - size); 2629} 2630 2631/** 2632 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2633 * @skb: buffer to pad 2634 * @len: minimal length 2635 * 2636 * Pads up a buffer to ensure the trailing bytes exist and are 2637 * blanked. If the buffer already contains sufficient data it 2638 * is untouched. Otherwise it is extended. Returns zero on 2639 * success. The skb is freed on error. 2640 */ 2641static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2642{ 2643 unsigned int size = skb->len; 2644 2645 if (unlikely(size < len)) { 2646 len -= size; 2647 if (skb_pad(skb, len)) 2648 return -ENOMEM; 2649 __skb_put(skb, len); 2650 } 2651 return 0; 2652} 2653 2654static inline int skb_add_data(struct sk_buff *skb, 2655 struct iov_iter *from, int copy) 2656{ 2657 const int off = skb->len; 2658 2659 if (skb->ip_summed == CHECKSUM_NONE) { 2660 __wsum csum = 0; 2661 if (csum_and_copy_from_iter(skb_put(skb, copy), copy, 2662 &csum, from) == copy) { 2663 skb->csum = csum_block_add(skb->csum, csum, off); 2664 return 0; 2665 } 2666 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy) 2667 return 0; 2668 2669 __skb_trim(skb, off); 2670 return -EFAULT; 2671} 2672 2673static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2674 const struct page *page, int off) 2675{ 2676 if (i) { 2677 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2678 2679 return page == skb_frag_page(frag) && 2680 off == frag->page_offset + skb_frag_size(frag); 2681 } 2682 return false; 2683} 2684 2685static inline int __skb_linearize(struct sk_buff *skb) 2686{ 2687 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2688} 2689 2690/** 2691 * skb_linearize - convert paged skb to linear one 2692 * @skb: buffer to linarize 2693 * 2694 * If there is no free memory -ENOMEM is returned, otherwise zero 2695 * is returned and the old skb data released. 2696 */ 2697static inline int skb_linearize(struct sk_buff *skb) 2698{ 2699 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2700} 2701 2702/** 2703 * skb_has_shared_frag - can any frag be overwritten 2704 * @skb: buffer to test 2705 * 2706 * Return true if the skb has at least one frag that might be modified 2707 * by an external entity (as in vmsplice()/sendfile()) 2708 */ 2709static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2710{ 2711 return skb_is_nonlinear(skb) && 2712 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2713} 2714 2715/** 2716 * skb_linearize_cow - make sure skb is linear and writable 2717 * @skb: buffer to process 2718 * 2719 * If there is no free memory -ENOMEM is returned, otherwise zero 2720 * is returned and the old skb data released. 2721 */ 2722static inline int skb_linearize_cow(struct sk_buff *skb) 2723{ 2724 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2725 __skb_linearize(skb) : 0; 2726} 2727 2728/** 2729 * skb_postpull_rcsum - update checksum for received skb after pull 2730 * @skb: buffer to update 2731 * @start: start of data before pull 2732 * @len: length of data pulled 2733 * 2734 * After doing a pull on a received packet, you need to call this to 2735 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2736 * CHECKSUM_NONE so that it can be recomputed from scratch. 2737 */ 2738 2739static inline void skb_postpull_rcsum(struct sk_buff *skb, 2740 const void *start, unsigned int len) 2741{ 2742 if (skb->ip_summed == CHECKSUM_COMPLETE) 2743 skb->csum = csum_sub(skb->csum, csum_partial(start, len, 0)); 2744 else if (skb->ip_summed == CHECKSUM_PARTIAL && 2745 skb_checksum_start_offset(skb) < 0) 2746 skb->ip_summed = CHECKSUM_NONE; 2747} 2748 2749unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2750 2751static inline void skb_postpush_rcsum(struct sk_buff *skb, 2752 const void *start, unsigned int len) 2753{ 2754 /* For performing the reverse operation to skb_postpull_rcsum(), 2755 * we can instead of ... 2756 * 2757 * skb->csum = csum_add(skb->csum, csum_partial(start, len, 0)); 2758 * 2759 * ... just use this equivalent version here to save a few 2760 * instructions. Feeding csum of 0 in csum_partial() and later 2761 * on adding skb->csum is equivalent to feed skb->csum in the 2762 * first place. 2763 */ 2764 if (skb->ip_summed == CHECKSUM_COMPLETE) 2765 skb->csum = csum_partial(start, len, skb->csum); 2766} 2767 2768/** 2769 * pskb_trim_rcsum - trim received skb and update checksum 2770 * @skb: buffer to trim 2771 * @len: new length 2772 * 2773 * This is exactly the same as pskb_trim except that it ensures the 2774 * checksum of received packets are still valid after the operation. 2775 */ 2776 2777static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2778{ 2779 if (likely(len >= skb->len)) 2780 return 0; 2781 if (skb->ip_summed == CHECKSUM_COMPLETE) 2782 skb->ip_summed = CHECKSUM_NONE; 2783 return __pskb_trim(skb, len); 2784} 2785 2786#define skb_queue_walk(queue, skb) \ 2787 for (skb = (queue)->next; \ 2788 skb != (struct sk_buff *)(queue); \ 2789 skb = skb->next) 2790 2791#define skb_queue_walk_safe(queue, skb, tmp) \ 2792 for (skb = (queue)->next, tmp = skb->next; \ 2793 skb != (struct sk_buff *)(queue); \ 2794 skb = tmp, tmp = skb->next) 2795 2796#define skb_queue_walk_from(queue, skb) \ 2797 for (; skb != (struct sk_buff *)(queue); \ 2798 skb = skb->next) 2799 2800#define skb_queue_walk_from_safe(queue, skb, tmp) \ 2801 for (tmp = skb->next; \ 2802 skb != (struct sk_buff *)(queue); \ 2803 skb = tmp, tmp = skb->next) 2804 2805#define skb_queue_reverse_walk(queue, skb) \ 2806 for (skb = (queue)->prev; \ 2807 skb != (struct sk_buff *)(queue); \ 2808 skb = skb->prev) 2809 2810#define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 2811 for (skb = (queue)->prev, tmp = skb->prev; \ 2812 skb != (struct sk_buff *)(queue); \ 2813 skb = tmp, tmp = skb->prev) 2814 2815#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 2816 for (tmp = skb->prev; \ 2817 skb != (struct sk_buff *)(queue); \ 2818 skb = tmp, tmp = skb->prev) 2819 2820static inline bool skb_has_frag_list(const struct sk_buff *skb) 2821{ 2822 return skb_shinfo(skb)->frag_list != NULL; 2823} 2824 2825static inline void skb_frag_list_init(struct sk_buff *skb) 2826{ 2827 skb_shinfo(skb)->frag_list = NULL; 2828} 2829 2830#define skb_walk_frags(skb, iter) \ 2831 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 2832 2833struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 2834 int *peeked, int *off, int *err); 2835struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 2836 int *err); 2837unsigned int datagram_poll(struct file *file, struct socket *sock, 2838 struct poll_table_struct *wait); 2839int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 2840 struct iov_iter *to, int size); 2841static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 2842 struct msghdr *msg, int size) 2843{ 2844 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 2845} 2846int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 2847 struct msghdr *msg); 2848int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 2849 struct iov_iter *from, int len); 2850int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 2851void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 2852void skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb); 2853int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 2854int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 2855int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 2856__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 2857 int len, __wsum csum); 2858ssize_t skb_socket_splice(struct sock *sk, 2859 struct pipe_inode_info *pipe, 2860 struct splice_pipe_desc *spd); 2861int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 2862 struct pipe_inode_info *pipe, unsigned int len, 2863 unsigned int flags, 2864 ssize_t (*splice_cb)(struct sock *, 2865 struct pipe_inode_info *, 2866 struct splice_pipe_desc *)); 2867void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 2868unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 2869int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 2870 int len, int hlen); 2871void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 2872int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 2873void skb_scrub_packet(struct sk_buff *skb, bool xnet); 2874unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 2875struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 2876struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 2877int skb_ensure_writable(struct sk_buff *skb, int write_len); 2878int skb_vlan_pop(struct sk_buff *skb); 2879int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 2880 2881static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 2882{ 2883 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2884} 2885 2886static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 2887{ 2888 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 2889} 2890 2891struct skb_checksum_ops { 2892 __wsum (*update)(const void *mem, int len, __wsum wsum); 2893 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 2894}; 2895 2896__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 2897 __wsum csum, const struct skb_checksum_ops *ops); 2898__wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 2899 __wsum csum); 2900 2901static inline void * __must_check 2902__skb_header_pointer(const struct sk_buff *skb, int offset, 2903 int len, void *data, int hlen, void *buffer) 2904{ 2905 if (hlen - offset >= len) 2906 return data + offset; 2907 2908 if (!skb || 2909 skb_copy_bits(skb, offset, buffer, len) < 0) 2910 return NULL; 2911 2912 return buffer; 2913} 2914 2915static inline void * __must_check 2916skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 2917{ 2918 return __skb_header_pointer(skb, offset, len, skb->data, 2919 skb_headlen(skb), buffer); 2920} 2921 2922/** 2923 * skb_needs_linearize - check if we need to linearize a given skb 2924 * depending on the given device features. 2925 * @skb: socket buffer to check 2926 * @features: net device features 2927 * 2928 * Returns true if either: 2929 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 2930 * 2. skb is fragmented and the device does not support SG. 2931 */ 2932static inline bool skb_needs_linearize(struct sk_buff *skb, 2933 netdev_features_t features) 2934{ 2935 return skb_is_nonlinear(skb) && 2936 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 2937 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 2938} 2939 2940static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 2941 void *to, 2942 const unsigned int len) 2943{ 2944 memcpy(to, skb->data, len); 2945} 2946 2947static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 2948 const int offset, void *to, 2949 const unsigned int len) 2950{ 2951 memcpy(to, skb->data + offset, len); 2952} 2953 2954static inline void skb_copy_to_linear_data(struct sk_buff *skb, 2955 const void *from, 2956 const unsigned int len) 2957{ 2958 memcpy(skb->data, from, len); 2959} 2960 2961static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 2962 const int offset, 2963 const void *from, 2964 const unsigned int len) 2965{ 2966 memcpy(skb->data + offset, from, len); 2967} 2968 2969void skb_init(void); 2970 2971static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 2972{ 2973 return skb->tstamp; 2974} 2975 2976/** 2977 * skb_get_timestamp - get timestamp from a skb 2978 * @skb: skb to get stamp from 2979 * @stamp: pointer to struct timeval to store stamp in 2980 * 2981 * Timestamps are stored in the skb as offsets to a base timestamp. 2982 * This function converts the offset back to a struct timeval and stores 2983 * it in stamp. 2984 */ 2985static inline void skb_get_timestamp(const struct sk_buff *skb, 2986 struct timeval *stamp) 2987{ 2988 *stamp = ktime_to_timeval(skb->tstamp); 2989} 2990 2991static inline void skb_get_timestampns(const struct sk_buff *skb, 2992 struct timespec *stamp) 2993{ 2994 *stamp = ktime_to_timespec(skb->tstamp); 2995} 2996 2997static inline void __net_timestamp(struct sk_buff *skb) 2998{ 2999 skb->tstamp = ktime_get_real(); 3000} 3001 3002static inline ktime_t net_timedelta(ktime_t t) 3003{ 3004 return ktime_sub(ktime_get_real(), t); 3005} 3006 3007static inline ktime_t net_invalid_timestamp(void) 3008{ 3009 return ktime_set(0, 0); 3010} 3011 3012struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3013 3014#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3015 3016void skb_clone_tx_timestamp(struct sk_buff *skb); 3017bool skb_defer_rx_timestamp(struct sk_buff *skb); 3018 3019#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3020 3021static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3022{ 3023} 3024 3025static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3026{ 3027 return false; 3028} 3029 3030#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3031 3032/** 3033 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3034 * 3035 * PHY drivers may accept clones of transmitted packets for 3036 * timestamping via their phy_driver.txtstamp method. These drivers 3037 * must call this function to return the skb back to the stack with a 3038 * timestamp. 3039 * 3040 * @skb: clone of the the original outgoing packet 3041 * @hwtstamps: hardware time stamps 3042 * 3043 */ 3044void skb_complete_tx_timestamp(struct sk_buff *skb, 3045 struct skb_shared_hwtstamps *hwtstamps); 3046 3047void __skb_tstamp_tx(struct sk_buff *orig_skb, 3048 struct skb_shared_hwtstamps *hwtstamps, 3049 struct sock *sk, int tstype); 3050 3051/** 3052 * skb_tstamp_tx - queue clone of skb with send time stamps 3053 * @orig_skb: the original outgoing packet 3054 * @hwtstamps: hardware time stamps, may be NULL if not available 3055 * 3056 * If the skb has a socket associated, then this function clones the 3057 * skb (thus sharing the actual data and optional structures), stores 3058 * the optional hardware time stamping information (if non NULL) or 3059 * generates a software time stamp (otherwise), then queues the clone 3060 * to the error queue of the socket. Errors are silently ignored. 3061 */ 3062void skb_tstamp_tx(struct sk_buff *orig_skb, 3063 struct skb_shared_hwtstamps *hwtstamps); 3064 3065static inline void sw_tx_timestamp(struct sk_buff *skb) 3066{ 3067 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 3068 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 3069 skb_tstamp_tx(skb, NULL); 3070} 3071 3072/** 3073 * skb_tx_timestamp() - Driver hook for transmit timestamping 3074 * 3075 * Ethernet MAC Drivers should call this function in their hard_xmit() 3076 * function immediately before giving the sk_buff to the MAC hardware. 3077 * 3078 * Specifically, one should make absolutely sure that this function is 3079 * called before TX completion of this packet can trigger. Otherwise 3080 * the packet could potentially already be freed. 3081 * 3082 * @skb: A socket buffer. 3083 */ 3084static inline void skb_tx_timestamp(struct sk_buff *skb) 3085{ 3086 skb_clone_tx_timestamp(skb); 3087 sw_tx_timestamp(skb); 3088} 3089 3090/** 3091 * skb_complete_wifi_ack - deliver skb with wifi status 3092 * 3093 * @skb: the original outgoing packet 3094 * @acked: ack status 3095 * 3096 */ 3097void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3098 3099__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3100__sum16 __skb_checksum_complete(struct sk_buff *skb); 3101 3102static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3103{ 3104 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3105 skb->csum_valid || 3106 (skb->ip_summed == CHECKSUM_PARTIAL && 3107 skb_checksum_start_offset(skb) >= 0)); 3108} 3109 3110/** 3111 * skb_checksum_complete - Calculate checksum of an entire packet 3112 * @skb: packet to process 3113 * 3114 * This function calculates the checksum over the entire packet plus 3115 * the value of skb->csum. The latter can be used to supply the 3116 * checksum of a pseudo header as used by TCP/UDP. It returns the 3117 * checksum. 3118 * 3119 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3120 * this function can be used to verify that checksum on received 3121 * packets. In that case the function should return zero if the 3122 * checksum is correct. In particular, this function will return zero 3123 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3124 * hardware has already verified the correctness of the checksum. 3125 */ 3126static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3127{ 3128 return skb_csum_unnecessary(skb) ? 3129 0 : __skb_checksum_complete(skb); 3130} 3131 3132static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3133{ 3134 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3135 if (skb->csum_level == 0) 3136 skb->ip_summed = CHECKSUM_NONE; 3137 else 3138 skb->csum_level--; 3139 } 3140} 3141 3142static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3143{ 3144 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3145 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3146 skb->csum_level++; 3147 } else if (skb->ip_summed == CHECKSUM_NONE) { 3148 skb->ip_summed = CHECKSUM_UNNECESSARY; 3149 skb->csum_level = 0; 3150 } 3151} 3152 3153static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 3154{ 3155 /* Mark current checksum as bad (typically called from GRO 3156 * path). In the case that ip_summed is CHECKSUM_NONE 3157 * this must be the first checksum encountered in the packet. 3158 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 3159 * checksum after the last one validated. For UDP, a zero 3160 * checksum can not be marked as bad. 3161 */ 3162 3163 if (skb->ip_summed == CHECKSUM_NONE || 3164 skb->ip_summed == CHECKSUM_UNNECESSARY) 3165 skb->csum_bad = 1; 3166} 3167 3168/* Check if we need to perform checksum complete validation. 3169 * 3170 * Returns true if checksum complete is needed, false otherwise 3171 * (either checksum is unnecessary or zero checksum is allowed). 3172 */ 3173static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3174 bool zero_okay, 3175 __sum16 check) 3176{ 3177 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3178 skb->csum_valid = 1; 3179 __skb_decr_checksum_unnecessary(skb); 3180 return false; 3181 } 3182 3183 return true; 3184} 3185 3186/* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3187 * in checksum_init. 3188 */ 3189#define CHECKSUM_BREAK 76 3190 3191/* Unset checksum-complete 3192 * 3193 * Unset checksum complete can be done when packet is being modified 3194 * (uncompressed for instance) and checksum-complete value is 3195 * invalidated. 3196 */ 3197static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3198{ 3199 if (skb->ip_summed == CHECKSUM_COMPLETE) 3200 skb->ip_summed = CHECKSUM_NONE; 3201} 3202 3203/* Validate (init) checksum based on checksum complete. 3204 * 3205 * Return values: 3206 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3207 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3208 * checksum is stored in skb->csum for use in __skb_checksum_complete 3209 * non-zero: value of invalid checksum 3210 * 3211 */ 3212static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3213 bool complete, 3214 __wsum psum) 3215{ 3216 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3217 if (!csum_fold(csum_add(psum, skb->csum))) { 3218 skb->csum_valid = 1; 3219 return 0; 3220 } 3221 } else if (skb->csum_bad) { 3222 /* ip_summed == CHECKSUM_NONE in this case */ 3223 return (__force __sum16)1; 3224 } 3225 3226 skb->csum = psum; 3227 3228 if (complete || skb->len <= CHECKSUM_BREAK) { 3229 __sum16 csum; 3230 3231 csum = __skb_checksum_complete(skb); 3232 skb->csum_valid = !csum; 3233 return csum; 3234 } 3235 3236 return 0; 3237} 3238 3239static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3240{ 3241 return 0; 3242} 3243 3244/* Perform checksum validate (init). Note that this is a macro since we only 3245 * want to calculate the pseudo header which is an input function if necessary. 3246 * First we try to validate without any computation (checksum unnecessary) and 3247 * then calculate based on checksum complete calling the function to compute 3248 * pseudo header. 3249 * 3250 * Return values: 3251 * 0: checksum is validated or try to in skb_checksum_complete 3252 * non-zero: value of invalid checksum 3253 */ 3254#define __skb_checksum_validate(skb, proto, complete, \ 3255 zero_okay, check, compute_pseudo) \ 3256({ \ 3257 __sum16 __ret = 0; \ 3258 skb->csum_valid = 0; \ 3259 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3260 __ret = __skb_checksum_validate_complete(skb, \ 3261 complete, compute_pseudo(skb, proto)); \ 3262 __ret; \ 3263}) 3264 3265#define skb_checksum_init(skb, proto, compute_pseudo) \ 3266 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3267 3268#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3269 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3270 3271#define skb_checksum_validate(skb, proto, compute_pseudo) \ 3272 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3273 3274#define skb_checksum_validate_zero_check(skb, proto, check, \ 3275 compute_pseudo) \ 3276 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3277 3278#define skb_checksum_simple_validate(skb) \ 3279 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3280 3281static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3282{ 3283 return (skb->ip_summed == CHECKSUM_NONE && 3284 skb->csum_valid && !skb->csum_bad); 3285} 3286 3287static inline void __skb_checksum_convert(struct sk_buff *skb, 3288 __sum16 check, __wsum pseudo) 3289{ 3290 skb->csum = ~pseudo; 3291 skb->ip_summed = CHECKSUM_COMPLETE; 3292} 3293 3294#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3295do { \ 3296 if (__skb_checksum_convert_check(skb)) \ 3297 __skb_checksum_convert(skb, check, \ 3298 compute_pseudo(skb, proto)); \ 3299} while (0) 3300 3301static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3302 u16 start, u16 offset) 3303{ 3304 skb->ip_summed = CHECKSUM_PARTIAL; 3305 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3306 skb->csum_offset = offset - start; 3307} 3308 3309/* Update skbuf and packet to reflect the remote checksum offload operation. 3310 * When called, ptr indicates the starting point for skb->csum when 3311 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3312 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3313 */ 3314static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3315 int start, int offset, bool nopartial) 3316{ 3317 __wsum delta; 3318 3319 if (!nopartial) { 3320 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3321 return; 3322 } 3323 3324 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3325 __skb_checksum_complete(skb); 3326 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3327 } 3328 3329 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3330 3331 /* Adjust skb->csum since we changed the packet */ 3332 skb->csum = csum_add(skb->csum, delta); 3333} 3334 3335#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3336void nf_conntrack_destroy(struct nf_conntrack *nfct); 3337static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3338{ 3339 if (nfct && atomic_dec_and_test(&nfct->use)) 3340 nf_conntrack_destroy(nfct); 3341} 3342static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3343{ 3344 if (nfct) 3345 atomic_inc(&nfct->use); 3346} 3347#endif 3348#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3349static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3350{ 3351 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3352 kfree(nf_bridge); 3353} 3354static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3355{ 3356 if (nf_bridge) 3357 atomic_inc(&nf_bridge->use); 3358} 3359#endif /* CONFIG_BRIDGE_NETFILTER */ 3360static inline void nf_reset(struct sk_buff *skb) 3361{ 3362#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3363 nf_conntrack_put(skb->nfct); 3364 skb->nfct = NULL; 3365#endif 3366#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3367 nf_bridge_put(skb->nf_bridge); 3368 skb->nf_bridge = NULL; 3369#endif 3370} 3371 3372static inline void nf_reset_trace(struct sk_buff *skb) 3373{ 3374#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3375 skb->nf_trace = 0; 3376#endif 3377} 3378 3379/* Note: This doesn't put any conntrack and bridge info in dst. */ 3380static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3381 bool copy) 3382{ 3383#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3384 dst->nfct = src->nfct; 3385 nf_conntrack_get(src->nfct); 3386 if (copy) 3387 dst->nfctinfo = src->nfctinfo; 3388#endif 3389#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3390 dst->nf_bridge = src->nf_bridge; 3391 nf_bridge_get(src->nf_bridge); 3392#endif 3393#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3394 if (copy) 3395 dst->nf_trace = src->nf_trace; 3396#endif 3397} 3398 3399static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3400{ 3401#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3402 nf_conntrack_put(dst->nfct); 3403#endif 3404#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3405 nf_bridge_put(dst->nf_bridge); 3406#endif 3407 __nf_copy(dst, src, true); 3408} 3409 3410#ifdef CONFIG_NETWORK_SECMARK 3411static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3412{ 3413 to->secmark = from->secmark; 3414} 3415 3416static inline void skb_init_secmark(struct sk_buff *skb) 3417{ 3418 skb->secmark = 0; 3419} 3420#else 3421static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3422{ } 3423 3424static inline void skb_init_secmark(struct sk_buff *skb) 3425{ } 3426#endif 3427 3428static inline bool skb_irq_freeable(const struct sk_buff *skb) 3429{ 3430 return !skb->destructor && 3431#if IS_ENABLED(CONFIG_XFRM) 3432 !skb->sp && 3433#endif 3434#if IS_ENABLED(CONFIG_NF_CONNTRACK) 3435 !skb->nfct && 3436#endif 3437 !skb->_skb_refdst && 3438 !skb_has_frag_list(skb); 3439} 3440 3441static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3442{ 3443 skb->queue_mapping = queue_mapping; 3444} 3445 3446static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3447{ 3448 return skb->queue_mapping; 3449} 3450 3451static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3452{ 3453 to->queue_mapping = from->queue_mapping; 3454} 3455 3456static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3457{ 3458 skb->queue_mapping = rx_queue + 1; 3459} 3460 3461static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3462{ 3463 return skb->queue_mapping - 1; 3464} 3465 3466static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3467{ 3468 return skb->queue_mapping != 0; 3469} 3470 3471static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3472{ 3473#ifdef CONFIG_XFRM 3474 return skb->sp; 3475#else 3476 return NULL; 3477#endif 3478} 3479 3480/* Keeps track of mac header offset relative to skb->head. 3481 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3482 * For non-tunnel skb it points to skb_mac_header() and for 3483 * tunnel skb it points to outer mac header. 3484 * Keeps track of level of encapsulation of network headers. 3485 */ 3486struct skb_gso_cb { 3487 int mac_offset; 3488 int encap_level; 3489 __u16 csum_start; 3490}; 3491#define SKB_SGO_CB_OFFSET 32 3492#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3493 3494static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3495{ 3496 return (skb_mac_header(inner_skb) - inner_skb->head) - 3497 SKB_GSO_CB(inner_skb)->mac_offset; 3498} 3499 3500static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3501{ 3502 int new_headroom, headroom; 3503 int ret; 3504 3505 headroom = skb_headroom(skb); 3506 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3507 if (ret) 3508 return ret; 3509 3510 new_headroom = skb_headroom(skb); 3511 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3512 return 0; 3513} 3514 3515/* Compute the checksum for a gso segment. First compute the checksum value 3516 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3517 * then add in skb->csum (checksum from csum_start to end of packet). 3518 * skb->csum and csum_start are then updated to reflect the checksum of the 3519 * resultant packet starting from the transport header-- the resultant checksum 3520 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3521 * header. 3522 */ 3523static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3524{ 3525 int plen = SKB_GSO_CB(skb)->csum_start - skb_headroom(skb) - 3526 skb_transport_offset(skb); 3527 __wsum partial; 3528 3529 partial = csum_partial(skb_transport_header(skb), plen, skb->csum); 3530 skb->csum = res; 3531 SKB_GSO_CB(skb)->csum_start -= plen; 3532 3533 return csum_fold(partial); 3534} 3535 3536static inline bool skb_is_gso(const struct sk_buff *skb) 3537{ 3538 return skb_shinfo(skb)->gso_size; 3539} 3540 3541/* Note: Should be called only if skb_is_gso(skb) is true */ 3542static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3543{ 3544 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3545} 3546 3547void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3548 3549static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3550{ 3551 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3552 * wanted then gso_type will be set. */ 3553 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3554 3555 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3556 unlikely(shinfo->gso_type == 0)) { 3557 __skb_warn_lro_forwarding(skb); 3558 return true; 3559 } 3560 return false; 3561} 3562 3563static inline void skb_forward_csum(struct sk_buff *skb) 3564{ 3565 /* Unfortunately we don't support this one. Any brave souls? */ 3566 if (skb->ip_summed == CHECKSUM_COMPLETE) 3567 skb->ip_summed = CHECKSUM_NONE; 3568} 3569 3570/** 3571 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3572 * @skb: skb to check 3573 * 3574 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3575 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3576 * use this helper, to document places where we make this assertion. 3577 */ 3578static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3579{ 3580#ifdef DEBUG 3581 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3582#endif 3583} 3584 3585bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3586 3587int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3588struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 3589 unsigned int transport_len, 3590 __sum16(*skb_chkf)(struct sk_buff *skb)); 3591 3592/** 3593 * skb_head_is_locked - Determine if the skb->head is locked down 3594 * @skb: skb to check 3595 * 3596 * The head on skbs build around a head frag can be removed if they are 3597 * not cloned. This function returns true if the skb head is locked down 3598 * due to either being allocated via kmalloc, or by being a clone with 3599 * multiple references to the head. 3600 */ 3601static inline bool skb_head_is_locked(const struct sk_buff *skb) 3602{ 3603 return !skb->head_frag || skb_cloned(skb); 3604} 3605 3606/** 3607 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3608 * 3609 * @skb: GSO skb 3610 * 3611 * skb_gso_network_seglen is used to determine the real size of the 3612 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3613 * 3614 * The MAC/L2 header is not accounted for. 3615 */ 3616static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3617{ 3618 unsigned int hdr_len = skb_transport_header(skb) - 3619 skb_network_header(skb); 3620 return hdr_len + skb_gso_transport_seglen(skb); 3621} 3622 3623#endif /* __KERNEL__ */ 3624#endif /* _LINUX_SKBUFF_H */ 3625