at v5.6-rc2 4553 lines 132 kB view raw
1/* SPDX-License-Identifier: GPL-2.0-or-later */ 2/* 3 * Definitions for the 'struct sk_buff' memory handlers. 4 * 5 * Authors: 6 * Alan Cox, <gw4pts@gw4pts.ampr.org> 7 * Florian La Roche, <rzsfl@rz.uni-sb.de> 8 */ 9 10#ifndef _LINUX_SKBUFF_H 11#define _LINUX_SKBUFF_H 12 13#include <linux/kernel.h> 14#include <linux/compiler.h> 15#include <linux/time.h> 16#include <linux/bug.h> 17#include <linux/bvec.h> 18#include <linux/cache.h> 19#include <linux/rbtree.h> 20#include <linux/socket.h> 21#include <linux/refcount.h> 22 23#include <linux/atomic.h> 24#include <asm/types.h> 25#include <linux/spinlock.h> 26#include <linux/net.h> 27#include <linux/textsearch.h> 28#include <net/checksum.h> 29#include <linux/rcupdate.h> 30#include <linux/hrtimer.h> 31#include <linux/dma-mapping.h> 32#include <linux/netdev_features.h> 33#include <linux/sched.h> 34#include <linux/sched/clock.h> 35#include <net/flow_dissector.h> 36#include <linux/splice.h> 37#include <linux/in6.h> 38#include <linux/if_packet.h> 39#include <net/flow.h> 40#if IS_ENABLED(CONFIG_NF_CONNTRACK) 41#include <linux/netfilter/nf_conntrack_common.h> 42#endif 43 44/* The interface for checksum offload between the stack and networking drivers 45 * is as follows... 46 * 47 * A. IP checksum related features 48 * 49 * Drivers advertise checksum offload capabilities in the features of a device. 50 * From the stack's point of view these are capabilities offered by the driver, 51 * a driver typically only advertises features that it is capable of offloading 52 * to its device. 53 * 54 * The checksum related features are: 55 * 56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one 57 * IP (one's complement) checksum for any combination 58 * of protocols or protocol layering. The checksum is 59 * computed and set in a packet per the CHECKSUM_PARTIAL 60 * interface (see below). 61 * 62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain 63 * TCP or UDP packets over IPv4. These are specifically 64 * unencapsulated packets of the form IPv4|TCP or 65 * IPv4|UDP where the Protocol field in the IPv4 header 66 * is TCP or UDP. The IPv4 header may contain IP options 67 * This feature cannot be set in features for a device 68 * with NETIF_F_HW_CSUM also set. This feature is being 69 * DEPRECATED (see below). 70 * 71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain 72 * TCP or UDP packets over IPv6. These are specifically 73 * unencapsulated packets of the form IPv6|TCP or 74 * IPv4|UDP where the Next Header field in the IPv6 75 * header is either TCP or UDP. IPv6 extension headers 76 * are not supported with this feature. This feature 77 * cannot be set in features for a device with 78 * NETIF_F_HW_CSUM also set. This feature is being 79 * DEPRECATED (see below). 80 * 81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload. 82 * This flag is used only used to disable the RX checksum 83 * feature for a device. The stack will accept receive 84 * checksum indication in packets received on a device 85 * regardless of whether NETIF_F_RXCSUM is set. 86 * 87 * B. Checksumming of received packets by device. Indication of checksum 88 * verification is in set skb->ip_summed. Possible values are: 89 * 90 * CHECKSUM_NONE: 91 * 92 * Device did not checksum this packet e.g. due to lack of capabilities. 93 * The packet contains full (though not verified) checksum in packet but 94 * not in skb->csum. Thus, skb->csum is undefined in this case. 95 * 96 * CHECKSUM_UNNECESSARY: 97 * 98 * The hardware you're dealing with doesn't calculate the full checksum 99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 101 * if their checksums are okay. skb->csum is still undefined in this case 102 * though. A driver or device must never modify the checksum field in the 103 * packet even if checksum is verified. 104 * 105 * CHECKSUM_UNNECESSARY is applicable to following protocols: 106 * TCP: IPv6 and IPv4. 107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 108 * zero UDP checksum for either IPv4 or IPv6, the networking stack 109 * may perform further validation in this case. 110 * GRE: only if the checksum is present in the header. 111 * SCTP: indicates the CRC in SCTP header has been validated. 112 * FCOE: indicates the CRC in FC frame has been validated. 113 * 114 * skb->csum_level indicates the number of consecutive checksums found in 115 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 116 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 117 * and a device is able to verify the checksums for UDP (possibly zero), 118 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 119 * two. If the device were only able to verify the UDP checksum and not 120 * GRE, either because it doesn't support GRE checksum of because GRE 121 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 122 * not considered in this case). 123 * 124 * CHECKSUM_COMPLETE: 125 * 126 * This is the most generic way. The device supplied checksum of the _whole_ 127 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 128 * hardware doesn't need to parse L3/L4 headers to implement this. 129 * 130 * Notes: 131 * - Even if device supports only some protocols, but is able to produce 132 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 133 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 134 * 135 * CHECKSUM_PARTIAL: 136 * 137 * A checksum is set up to be offloaded to a device as described in the 138 * output description for CHECKSUM_PARTIAL. This may occur on a packet 139 * received directly from another Linux OS, e.g., a virtualized Linux kernel 140 * on the same host, or it may be set in the input path in GRO or remote 141 * checksum offload. For the purposes of checksum verification, the checksum 142 * referred to by skb->csum_start + skb->csum_offset and any preceding 143 * checksums in the packet are considered verified. Any checksums in the 144 * packet that are after the checksum being offloaded are not considered to 145 * be verified. 146 * 147 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload 148 * in the skb->ip_summed for a packet. Values are: 149 * 150 * CHECKSUM_PARTIAL: 151 * 152 * The driver is required to checksum the packet as seen by hard_start_xmit() 153 * from skb->csum_start up to the end, and to record/write the checksum at 154 * offset skb->csum_start + skb->csum_offset. A driver may verify that the 155 * csum_start and csum_offset values are valid values given the length and 156 * offset of the packet, however they should not attempt to validate that the 157 * checksum refers to a legitimate transport layer checksum-- it is the 158 * purview of the stack to validate that csum_start and csum_offset are set 159 * correctly. 160 * 161 * When the stack requests checksum offload for a packet, the driver MUST 162 * ensure that the checksum is set correctly. A driver can either offload the 163 * checksum calculation to the device, or call skb_checksum_help (in the case 164 * that the device does not support offload for a particular checksum). 165 * 166 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of 167 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate 168 * checksum offload capability. 169 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based 170 * on network device checksumming capabilities: if a packet does not match 171 * them, skb_checksum_help or skb_crc32c_help (depending on the value of 172 * csum_not_inet, see item D.) is called to resolve the checksum. 173 * 174 * CHECKSUM_NONE: 175 * 176 * The skb was already checksummed by the protocol, or a checksum is not 177 * required. 178 * 179 * CHECKSUM_UNNECESSARY: 180 * 181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on 182 * output. 183 * 184 * CHECKSUM_COMPLETE: 185 * Not used in checksum output. If a driver observes a packet with this value 186 * set in skbuff, if should treat as CHECKSUM_NONE being set. 187 * 188 * D. Non-IP checksum (CRC) offloads 189 * 190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of 191 * offloading the SCTP CRC in a packet. To perform this offload the stack 192 * will set set csum_start and csum_offset accordingly, set ip_summed to 193 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in 194 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c. 195 * A driver that supports both IP checksum offload and SCTP CRC32c offload 196 * must verify which offload is configured for a packet by testing the 197 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve 198 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 199 * 200 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of 201 * offloading the FCOE CRC in a packet. To perform this offload the stack 202 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 203 * accordingly. Note the there is no indication in the skbuff that the 204 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports 205 * both IP checksum offload and FCOE CRC offload must verify which offload 206 * is configured for a packet presumably by inspecting packet headers. 207 * 208 * E. Checksumming on output with GSO. 209 * 210 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload 211 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 212 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as 213 * part of the GSO operation is implied. If a checksum is being offloaded 214 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset 215 * are set to refer to the outermost checksum being offload (two offloaded 216 * checksums are possible with UDP encapsulation). 217 */ 218 219/* Don't change this without changing skb_csum_unnecessary! */ 220#define CHECKSUM_NONE 0 221#define CHECKSUM_UNNECESSARY 1 222#define CHECKSUM_COMPLETE 2 223#define CHECKSUM_PARTIAL 3 224 225/* Maximum value in skb->csum_level */ 226#define SKB_MAX_CSUM_LEVEL 3 227 228#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 229#define SKB_WITH_OVERHEAD(X) \ 230 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 231#define SKB_MAX_ORDER(X, ORDER) \ 232 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 233#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 234#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 235 236/* return minimum truesize of one skb containing X bytes of data */ 237#define SKB_TRUESIZE(X) ((X) + \ 238 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 239 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 240 241struct net_device; 242struct scatterlist; 243struct pipe_inode_info; 244struct iov_iter; 245struct napi_struct; 246struct bpf_prog; 247union bpf_attr; 248struct skb_ext; 249 250#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 251struct nf_bridge_info { 252 enum { 253 BRNF_PROTO_UNCHANGED, 254 BRNF_PROTO_8021Q, 255 BRNF_PROTO_PPPOE 256 } orig_proto:8; 257 u8 pkt_otherhost:1; 258 u8 in_prerouting:1; 259 u8 bridged_dnat:1; 260 __u16 frag_max_size; 261 struct net_device *physindev; 262 263 /* always valid & non-NULL from FORWARD on, for physdev match */ 264 struct net_device *physoutdev; 265 union { 266 /* prerouting: detect dnat in orig/reply direction */ 267 __be32 ipv4_daddr; 268 struct in6_addr ipv6_daddr; 269 270 /* after prerouting + nat detected: store original source 271 * mac since neigh resolution overwrites it, only used while 272 * skb is out in neigh layer. 273 */ 274 char neigh_header[8]; 275 }; 276}; 277#endif 278 279#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 280/* Chain in tc_skb_ext will be used to share the tc chain with 281 * ovs recirc_id. It will be set to the current chain by tc 282 * and read by ovs to recirc_id. 283 */ 284struct tc_skb_ext { 285 __u32 chain; 286}; 287#endif 288 289struct sk_buff_head { 290 /* These two members must be first. */ 291 struct sk_buff *next; 292 struct sk_buff *prev; 293 294 __u32 qlen; 295 spinlock_t lock; 296}; 297 298struct sk_buff; 299 300/* To allow 64K frame to be packed as single skb without frag_list we 301 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 302 * buffers which do not start on a page boundary. 303 * 304 * Since GRO uses frags we allocate at least 16 regardless of page 305 * size. 306 */ 307#if (65536/PAGE_SIZE + 1) < 16 308#define MAX_SKB_FRAGS 16UL 309#else 310#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 311#endif 312extern int sysctl_max_skb_frags; 313 314/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 315 * segment using its current segmentation instead. 316 */ 317#define GSO_BY_FRAGS 0xFFFF 318 319typedef struct bio_vec skb_frag_t; 320 321/** 322 * skb_frag_size() - Returns the size of a skb fragment 323 * @frag: skb fragment 324 */ 325static inline unsigned int skb_frag_size(const skb_frag_t *frag) 326{ 327 return frag->bv_len; 328} 329 330/** 331 * skb_frag_size_set() - Sets the size of a skb fragment 332 * @frag: skb fragment 333 * @size: size of fragment 334 */ 335static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 336{ 337 frag->bv_len = size; 338} 339 340/** 341 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 342 * @frag: skb fragment 343 * @delta: value to add 344 */ 345static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 346{ 347 frag->bv_len += delta; 348} 349 350/** 351 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 352 * @frag: skb fragment 353 * @delta: value to subtract 354 */ 355static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 356{ 357 frag->bv_len -= delta; 358} 359 360/** 361 * skb_frag_must_loop - Test if %p is a high memory page 362 * @p: fragment's page 363 */ 364static inline bool skb_frag_must_loop(struct page *p) 365{ 366#if defined(CONFIG_HIGHMEM) 367 if (PageHighMem(p)) 368 return true; 369#endif 370 return false; 371} 372 373/** 374 * skb_frag_foreach_page - loop over pages in a fragment 375 * 376 * @f: skb frag to operate on 377 * @f_off: offset from start of f->bv_page 378 * @f_len: length from f_off to loop over 379 * @p: (temp var) current page 380 * @p_off: (temp var) offset from start of current page, 381 * non-zero only on first page. 382 * @p_len: (temp var) length in current page, 383 * < PAGE_SIZE only on first and last page. 384 * @copied: (temp var) length so far, excluding current p_len. 385 * 386 * A fragment can hold a compound page, in which case per-page 387 * operations, notably kmap_atomic, must be called for each 388 * regular page. 389 */ 390#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 391 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 392 p_off = (f_off) & (PAGE_SIZE - 1), \ 393 p_len = skb_frag_must_loop(p) ? \ 394 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 395 copied = 0; \ 396 copied < f_len; \ 397 copied += p_len, p++, p_off = 0, \ 398 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 399 400#define HAVE_HW_TIME_STAMP 401 402/** 403 * struct skb_shared_hwtstamps - hardware time stamps 404 * @hwtstamp: hardware time stamp transformed into duration 405 * since arbitrary point in time 406 * 407 * Software time stamps generated by ktime_get_real() are stored in 408 * skb->tstamp. 409 * 410 * hwtstamps can only be compared against other hwtstamps from 411 * the same device. 412 * 413 * This structure is attached to packets as part of the 414 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 415 */ 416struct skb_shared_hwtstamps { 417 ktime_t hwtstamp; 418}; 419 420/* Definitions for tx_flags in struct skb_shared_info */ 421enum { 422 /* generate hardware time stamp */ 423 SKBTX_HW_TSTAMP = 1 << 0, 424 425 /* generate software time stamp when queueing packet to NIC */ 426 SKBTX_SW_TSTAMP = 1 << 1, 427 428 /* device driver is going to provide hardware time stamp */ 429 SKBTX_IN_PROGRESS = 1 << 2, 430 431 /* device driver supports TX zero-copy buffers */ 432 SKBTX_DEV_ZEROCOPY = 1 << 3, 433 434 /* generate wifi status information (where possible) */ 435 SKBTX_WIFI_STATUS = 1 << 4, 436 437 /* This indicates at least one fragment might be overwritten 438 * (as in vmsplice(), sendfile() ...) 439 * If we need to compute a TX checksum, we'll need to copy 440 * all frags to avoid possible bad checksum 441 */ 442 SKBTX_SHARED_FRAG = 1 << 5, 443 444 /* generate software time stamp when entering packet scheduling */ 445 SKBTX_SCHED_TSTAMP = 1 << 6, 446}; 447 448#define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG) 449#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 450 SKBTX_SCHED_TSTAMP) 451#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 452 453/* 454 * The callback notifies userspace to release buffers when skb DMA is done in 455 * lower device, the skb last reference should be 0 when calling this. 456 * The zerocopy_success argument is true if zero copy transmit occurred, 457 * false on data copy or out of memory error caused by data copy attempt. 458 * The ctx field is used to track device context. 459 * The desc field is used to track userspace buffer index. 460 */ 461struct ubuf_info { 462 void (*callback)(struct ubuf_info *, bool zerocopy_success); 463 union { 464 struct { 465 unsigned long desc; 466 void *ctx; 467 }; 468 struct { 469 u32 id; 470 u16 len; 471 u16 zerocopy:1; 472 u32 bytelen; 473 }; 474 }; 475 refcount_t refcnt; 476 477 struct mmpin { 478 struct user_struct *user; 479 unsigned int num_pg; 480 } mmp; 481}; 482 483#define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 484 485int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 486void mm_unaccount_pinned_pages(struct mmpin *mmp); 487 488struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size); 489struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, 490 struct ubuf_info *uarg); 491 492static inline void sock_zerocopy_get(struct ubuf_info *uarg) 493{ 494 refcount_inc(&uarg->refcnt); 495} 496 497void sock_zerocopy_put(struct ubuf_info *uarg); 498void sock_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 499 500void sock_zerocopy_callback(struct ubuf_info *uarg, bool success); 501 502int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len); 503int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 504 struct msghdr *msg, int len, 505 struct ubuf_info *uarg); 506 507/* This data is invariant across clones and lives at 508 * the end of the header data, ie. at skb->end. 509 */ 510struct skb_shared_info { 511 __u8 __unused; 512 __u8 meta_len; 513 __u8 nr_frags; 514 __u8 tx_flags; 515 unsigned short gso_size; 516 /* Warning: this field is not always filled in (UFO)! */ 517 unsigned short gso_segs; 518 struct sk_buff *frag_list; 519 struct skb_shared_hwtstamps hwtstamps; 520 unsigned int gso_type; 521 u32 tskey; 522 523 /* 524 * Warning : all fields before dataref are cleared in __alloc_skb() 525 */ 526 atomic_t dataref; 527 528 /* Intermediate layers must ensure that destructor_arg 529 * remains valid until skb destructor */ 530 void * destructor_arg; 531 532 /* must be last field, see pskb_expand_head() */ 533 skb_frag_t frags[MAX_SKB_FRAGS]; 534}; 535 536/* We divide dataref into two halves. The higher 16 bits hold references 537 * to the payload part of skb->data. The lower 16 bits hold references to 538 * the entire skb->data. A clone of a headerless skb holds the length of 539 * the header in skb->hdr_len. 540 * 541 * All users must obey the rule that the skb->data reference count must be 542 * greater than or equal to the payload reference count. 543 * 544 * Holding a reference to the payload part means that the user does not 545 * care about modifications to the header part of skb->data. 546 */ 547#define SKB_DATAREF_SHIFT 16 548#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 549 550 551enum { 552 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 553 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 554 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 555}; 556 557enum { 558 SKB_GSO_TCPV4 = 1 << 0, 559 560 /* This indicates the skb is from an untrusted source. */ 561 SKB_GSO_DODGY = 1 << 1, 562 563 /* This indicates the tcp segment has CWR set. */ 564 SKB_GSO_TCP_ECN = 1 << 2, 565 566 SKB_GSO_TCP_FIXEDID = 1 << 3, 567 568 SKB_GSO_TCPV6 = 1 << 4, 569 570 SKB_GSO_FCOE = 1 << 5, 571 572 SKB_GSO_GRE = 1 << 6, 573 574 SKB_GSO_GRE_CSUM = 1 << 7, 575 576 SKB_GSO_IPXIP4 = 1 << 8, 577 578 SKB_GSO_IPXIP6 = 1 << 9, 579 580 SKB_GSO_UDP_TUNNEL = 1 << 10, 581 582 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 583 584 SKB_GSO_PARTIAL = 1 << 12, 585 586 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 587 588 SKB_GSO_SCTP = 1 << 14, 589 590 SKB_GSO_ESP = 1 << 15, 591 592 SKB_GSO_UDP = 1 << 16, 593 594 SKB_GSO_UDP_L4 = 1 << 17, 595 596 SKB_GSO_FRAGLIST = 1 << 18, 597}; 598 599#if BITS_PER_LONG > 32 600#define NET_SKBUFF_DATA_USES_OFFSET 1 601#endif 602 603#ifdef NET_SKBUFF_DATA_USES_OFFSET 604typedef unsigned int sk_buff_data_t; 605#else 606typedef unsigned char *sk_buff_data_t; 607#endif 608 609/** 610 * struct sk_buff - socket buffer 611 * @next: Next buffer in list 612 * @prev: Previous buffer in list 613 * @tstamp: Time we arrived/left 614 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 615 * @sk: Socket we are owned by 616 * @dev: Device we arrived on/are leaving by 617 * @cb: Control buffer. Free for use by every layer. Put private vars here 618 * @_skb_refdst: destination entry (with norefcount bit) 619 * @sp: the security path, used for xfrm 620 * @len: Length of actual data 621 * @data_len: Data length 622 * @mac_len: Length of link layer header 623 * @hdr_len: writable header length of cloned skb 624 * @csum: Checksum (must include start/offset pair) 625 * @csum_start: Offset from skb->head where checksumming should start 626 * @csum_offset: Offset from csum_start where checksum should be stored 627 * @priority: Packet queueing priority 628 * @ignore_df: allow local fragmentation 629 * @cloned: Head may be cloned (check refcnt to be sure) 630 * @ip_summed: Driver fed us an IP checksum 631 * @nohdr: Payload reference only, must not modify header 632 * @pkt_type: Packet class 633 * @fclone: skbuff clone status 634 * @ipvs_property: skbuff is owned by ipvs 635 * @offload_fwd_mark: Packet was L2-forwarded in hardware 636 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 637 * @tc_skip_classify: do not classify packet. set by IFB device 638 * @tc_at_ingress: used within tc_classify to distinguish in/egress 639 * @tc_redirected: packet was redirected by a tc action 640 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect 641 * @peeked: this packet has been seen already, so stats have been 642 * done for it, don't do them again 643 * @nf_trace: netfilter packet trace flag 644 * @protocol: Packet protocol from driver 645 * @destructor: Destruct function 646 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 647 * @_nfct: Associated connection, if any (with nfctinfo bits) 648 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 649 * @skb_iif: ifindex of device we arrived on 650 * @tc_index: Traffic control index 651 * @hash: the packet hash 652 * @queue_mapping: Queue mapping for multiqueue devices 653 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 654 * @active_extensions: active extensions (skb_ext_id types) 655 * @ndisc_nodetype: router type (from link layer) 656 * @ooo_okay: allow the mapping of a socket to a queue to be changed 657 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 658 * ports. 659 * @sw_hash: indicates hash was computed in software stack 660 * @wifi_acked_valid: wifi_acked was set 661 * @wifi_acked: whether frame was acked on wifi or not 662 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 663 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 664 * @dst_pending_confirm: need to confirm neighbour 665 * @decrypted: Decrypted SKB 666 * @napi_id: id of the NAPI struct this skb came from 667 * @secmark: security marking 668 * @mark: Generic packet mark 669 * @vlan_proto: vlan encapsulation protocol 670 * @vlan_tci: vlan tag control information 671 * @inner_protocol: Protocol (encapsulation) 672 * @inner_transport_header: Inner transport layer header (encapsulation) 673 * @inner_network_header: Network layer header (encapsulation) 674 * @inner_mac_header: Link layer header (encapsulation) 675 * @transport_header: Transport layer header 676 * @network_header: Network layer header 677 * @mac_header: Link layer header 678 * @tail: Tail pointer 679 * @end: End pointer 680 * @head: Head of buffer 681 * @data: Data head pointer 682 * @truesize: Buffer size 683 * @users: User count - see {datagram,tcp}.c 684 * @extensions: allocated extensions, valid if active_extensions is nonzero 685 */ 686 687struct sk_buff { 688 union { 689 struct { 690 /* These two members must be first. */ 691 struct sk_buff *next; 692 struct sk_buff *prev; 693 694 union { 695 struct net_device *dev; 696 /* Some protocols might use this space to store information, 697 * while device pointer would be NULL. 698 * UDP receive path is one user. 699 */ 700 unsigned long dev_scratch; 701 }; 702 }; 703 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 704 struct list_head list; 705 }; 706 707 union { 708 struct sock *sk; 709 int ip_defrag_offset; 710 }; 711 712 union { 713 ktime_t tstamp; 714 u64 skb_mstamp_ns; /* earliest departure time */ 715 }; 716 /* 717 * This is the control buffer. It is free to use for every 718 * layer. Please put your private variables there. If you 719 * want to keep them across layers you have to do a skb_clone() 720 * first. This is owned by whoever has the skb queued ATM. 721 */ 722 char cb[48] __aligned(8); 723 724 union { 725 struct { 726 unsigned long _skb_refdst; 727 void (*destructor)(struct sk_buff *skb); 728 }; 729 struct list_head tcp_tsorted_anchor; 730 }; 731 732#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 733 unsigned long _nfct; 734#endif 735 unsigned int len, 736 data_len; 737 __u16 mac_len, 738 hdr_len; 739 740 /* Following fields are _not_ copied in __copy_skb_header() 741 * Note that queue_mapping is here mostly to fill a hole. 742 */ 743 __u16 queue_mapping; 744 745/* if you move cloned around you also must adapt those constants */ 746#ifdef __BIG_ENDIAN_BITFIELD 747#define CLONED_MASK (1 << 7) 748#else 749#define CLONED_MASK 1 750#endif 751#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset) 752 753 __u8 __cloned_offset[0]; 754 __u8 cloned:1, 755 nohdr:1, 756 fclone:2, 757 peeked:1, 758 head_frag:1, 759 pfmemalloc:1; 760#ifdef CONFIG_SKB_EXTENSIONS 761 __u8 active_extensions; 762#endif 763 /* fields enclosed in headers_start/headers_end are copied 764 * using a single memcpy() in __copy_skb_header() 765 */ 766 /* private: */ 767 __u32 headers_start[0]; 768 /* public: */ 769 770/* if you move pkt_type around you also must adapt those constants */ 771#ifdef __BIG_ENDIAN_BITFIELD 772#define PKT_TYPE_MAX (7 << 5) 773#else 774#define PKT_TYPE_MAX 7 775#endif 776#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 777 778 __u8 __pkt_type_offset[0]; 779 __u8 pkt_type:3; 780 __u8 ignore_df:1; 781 __u8 nf_trace:1; 782 __u8 ip_summed:2; 783 __u8 ooo_okay:1; 784 785 __u8 l4_hash:1; 786 __u8 sw_hash:1; 787 __u8 wifi_acked_valid:1; 788 __u8 wifi_acked:1; 789 __u8 no_fcs:1; 790 /* Indicates the inner headers are valid in the skbuff. */ 791 __u8 encapsulation:1; 792 __u8 encap_hdr_csum:1; 793 __u8 csum_valid:1; 794 795#ifdef __BIG_ENDIAN_BITFIELD 796#define PKT_VLAN_PRESENT_BIT 7 797#else 798#define PKT_VLAN_PRESENT_BIT 0 799#endif 800#define PKT_VLAN_PRESENT_OFFSET() offsetof(struct sk_buff, __pkt_vlan_present_offset) 801 __u8 __pkt_vlan_present_offset[0]; 802 __u8 vlan_present:1; 803 __u8 csum_complete_sw:1; 804 __u8 csum_level:2; 805 __u8 csum_not_inet:1; 806 __u8 dst_pending_confirm:1; 807#ifdef CONFIG_IPV6_NDISC_NODETYPE 808 __u8 ndisc_nodetype:2; 809#endif 810 811 __u8 ipvs_property:1; 812 __u8 inner_protocol_type:1; 813 __u8 remcsum_offload:1; 814#ifdef CONFIG_NET_SWITCHDEV 815 __u8 offload_fwd_mark:1; 816 __u8 offload_l3_fwd_mark:1; 817#endif 818#ifdef CONFIG_NET_CLS_ACT 819 __u8 tc_skip_classify:1; 820 __u8 tc_at_ingress:1; 821 __u8 tc_redirected:1; 822 __u8 tc_from_ingress:1; 823#endif 824#ifdef CONFIG_TLS_DEVICE 825 __u8 decrypted:1; 826#endif 827 828#ifdef CONFIG_NET_SCHED 829 __u16 tc_index; /* traffic control index */ 830#endif 831 832 union { 833 __wsum csum; 834 struct { 835 __u16 csum_start; 836 __u16 csum_offset; 837 }; 838 }; 839 __u32 priority; 840 int skb_iif; 841 __u32 hash; 842 __be16 vlan_proto; 843 __u16 vlan_tci; 844#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 845 union { 846 unsigned int napi_id; 847 unsigned int sender_cpu; 848 }; 849#endif 850#ifdef CONFIG_NETWORK_SECMARK 851 __u32 secmark; 852#endif 853 854 union { 855 __u32 mark; 856 __u32 reserved_tailroom; 857 }; 858 859 union { 860 __be16 inner_protocol; 861 __u8 inner_ipproto; 862 }; 863 864 __u16 inner_transport_header; 865 __u16 inner_network_header; 866 __u16 inner_mac_header; 867 868 __be16 protocol; 869 __u16 transport_header; 870 __u16 network_header; 871 __u16 mac_header; 872 873 /* private: */ 874 __u32 headers_end[0]; 875 /* public: */ 876 877 /* These elements must be at the end, see alloc_skb() for details. */ 878 sk_buff_data_t tail; 879 sk_buff_data_t end; 880 unsigned char *head, 881 *data; 882 unsigned int truesize; 883 refcount_t users; 884 885#ifdef CONFIG_SKB_EXTENSIONS 886 /* only useable after checking ->active_extensions != 0 */ 887 struct skb_ext *extensions; 888#endif 889}; 890 891#ifdef __KERNEL__ 892/* 893 * Handling routines are only of interest to the kernel 894 */ 895 896#define SKB_ALLOC_FCLONE 0x01 897#define SKB_ALLOC_RX 0x02 898#define SKB_ALLOC_NAPI 0x04 899 900/** 901 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 902 * @skb: buffer 903 */ 904static inline bool skb_pfmemalloc(const struct sk_buff *skb) 905{ 906 return unlikely(skb->pfmemalloc); 907} 908 909/* 910 * skb might have a dst pointer attached, refcounted or not. 911 * _skb_refdst low order bit is set if refcount was _not_ taken 912 */ 913#define SKB_DST_NOREF 1UL 914#define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 915 916/** 917 * skb_dst - returns skb dst_entry 918 * @skb: buffer 919 * 920 * Returns skb dst_entry, regardless of reference taken or not. 921 */ 922static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 923{ 924 /* If refdst was not refcounted, check we still are in a 925 * rcu_read_lock section 926 */ 927 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 928 !rcu_read_lock_held() && 929 !rcu_read_lock_bh_held()); 930 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 931} 932 933/** 934 * skb_dst_set - sets skb dst 935 * @skb: buffer 936 * @dst: dst entry 937 * 938 * Sets skb dst, assuming a reference was taken on dst and should 939 * be released by skb_dst_drop() 940 */ 941static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 942{ 943 skb->_skb_refdst = (unsigned long)dst; 944} 945 946/** 947 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 948 * @skb: buffer 949 * @dst: dst entry 950 * 951 * Sets skb dst, assuming a reference was not taken on dst. 952 * If dst entry is cached, we do not take reference and dst_release 953 * will be avoided by refdst_drop. If dst entry is not cached, we take 954 * reference, so that last dst_release can destroy the dst immediately. 955 */ 956static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 957{ 958 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 959 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 960} 961 962/** 963 * skb_dst_is_noref - Test if skb dst isn't refcounted 964 * @skb: buffer 965 */ 966static inline bool skb_dst_is_noref(const struct sk_buff *skb) 967{ 968 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 969} 970 971/** 972 * skb_rtable - Returns the skb &rtable 973 * @skb: buffer 974 */ 975static inline struct rtable *skb_rtable(const struct sk_buff *skb) 976{ 977 return (struct rtable *)skb_dst(skb); 978} 979 980/* For mangling skb->pkt_type from user space side from applications 981 * such as nft, tc, etc, we only allow a conservative subset of 982 * possible pkt_types to be set. 983*/ 984static inline bool skb_pkt_type_ok(u32 ptype) 985{ 986 return ptype <= PACKET_OTHERHOST; 987} 988 989/** 990 * skb_napi_id - Returns the skb's NAPI id 991 * @skb: buffer 992 */ 993static inline unsigned int skb_napi_id(const struct sk_buff *skb) 994{ 995#ifdef CONFIG_NET_RX_BUSY_POLL 996 return skb->napi_id; 997#else 998 return 0; 999#endif 1000} 1001 1002/** 1003 * skb_unref - decrement the skb's reference count 1004 * @skb: buffer 1005 * 1006 * Returns true if we can free the skb. 1007 */ 1008static inline bool skb_unref(struct sk_buff *skb) 1009{ 1010 if (unlikely(!skb)) 1011 return false; 1012 if (likely(refcount_read(&skb->users) == 1)) 1013 smp_rmb(); 1014 else if (likely(!refcount_dec_and_test(&skb->users))) 1015 return false; 1016 1017 return true; 1018} 1019 1020void skb_release_head_state(struct sk_buff *skb); 1021void kfree_skb(struct sk_buff *skb); 1022void kfree_skb_list(struct sk_buff *segs); 1023void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1024void skb_tx_error(struct sk_buff *skb); 1025void consume_skb(struct sk_buff *skb); 1026void __consume_stateless_skb(struct sk_buff *skb); 1027void __kfree_skb(struct sk_buff *skb); 1028extern struct kmem_cache *skbuff_head_cache; 1029 1030void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1031bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1032 bool *fragstolen, int *delta_truesize); 1033 1034struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1035 int node); 1036struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1037struct sk_buff *build_skb(void *data, unsigned int frag_size); 1038struct sk_buff *build_skb_around(struct sk_buff *skb, 1039 void *data, unsigned int frag_size); 1040 1041/** 1042 * alloc_skb - allocate a network buffer 1043 * @size: size to allocate 1044 * @priority: allocation mask 1045 * 1046 * This function is a convenient wrapper around __alloc_skb(). 1047 */ 1048static inline struct sk_buff *alloc_skb(unsigned int size, 1049 gfp_t priority) 1050{ 1051 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1052} 1053 1054struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1055 unsigned long data_len, 1056 int max_page_order, 1057 int *errcode, 1058 gfp_t gfp_mask); 1059struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1060 1061/* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1062struct sk_buff_fclones { 1063 struct sk_buff skb1; 1064 1065 struct sk_buff skb2; 1066 1067 refcount_t fclone_ref; 1068}; 1069 1070/** 1071 * skb_fclone_busy - check if fclone is busy 1072 * @sk: socket 1073 * @skb: buffer 1074 * 1075 * Returns true if skb is a fast clone, and its clone is not freed. 1076 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1077 * so we also check that this didnt happen. 1078 */ 1079static inline bool skb_fclone_busy(const struct sock *sk, 1080 const struct sk_buff *skb) 1081{ 1082 const struct sk_buff_fclones *fclones; 1083 1084 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1085 1086 return skb->fclone == SKB_FCLONE_ORIG && 1087 refcount_read(&fclones->fclone_ref) > 1 && 1088 fclones->skb2.sk == sk; 1089} 1090 1091/** 1092 * alloc_skb_fclone - allocate a network buffer from fclone cache 1093 * @size: size to allocate 1094 * @priority: allocation mask 1095 * 1096 * This function is a convenient wrapper around __alloc_skb(). 1097 */ 1098static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1099 gfp_t priority) 1100{ 1101 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1102} 1103 1104struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1105void skb_headers_offset_update(struct sk_buff *skb, int off); 1106int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1107struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1108void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1109struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1110struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1111 gfp_t gfp_mask, bool fclone); 1112static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1113 gfp_t gfp_mask) 1114{ 1115 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1116} 1117 1118int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1119struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1120 unsigned int headroom); 1121struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1122 int newtailroom, gfp_t priority); 1123int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1124 int offset, int len); 1125int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1126 int offset, int len); 1127int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1128int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1129 1130/** 1131 * skb_pad - zero pad the tail of an skb 1132 * @skb: buffer to pad 1133 * @pad: space to pad 1134 * 1135 * Ensure that a buffer is followed by a padding area that is zero 1136 * filled. Used by network drivers which may DMA or transfer data 1137 * beyond the buffer end onto the wire. 1138 * 1139 * May return error in out of memory cases. The skb is freed on error. 1140 */ 1141static inline int skb_pad(struct sk_buff *skb, int pad) 1142{ 1143 return __skb_pad(skb, pad, true); 1144} 1145#define dev_kfree_skb(a) consume_skb(a) 1146 1147int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1148 int offset, size_t size); 1149 1150struct skb_seq_state { 1151 __u32 lower_offset; 1152 __u32 upper_offset; 1153 __u32 frag_idx; 1154 __u32 stepped_offset; 1155 struct sk_buff *root_skb; 1156 struct sk_buff *cur_skb; 1157 __u8 *frag_data; 1158}; 1159 1160void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1161 unsigned int to, struct skb_seq_state *st); 1162unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1163 struct skb_seq_state *st); 1164void skb_abort_seq_read(struct skb_seq_state *st); 1165 1166unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1167 unsigned int to, struct ts_config *config); 1168 1169/* 1170 * Packet hash types specify the type of hash in skb_set_hash. 1171 * 1172 * Hash types refer to the protocol layer addresses which are used to 1173 * construct a packet's hash. The hashes are used to differentiate or identify 1174 * flows of the protocol layer for the hash type. Hash types are either 1175 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1176 * 1177 * Properties of hashes: 1178 * 1179 * 1) Two packets in different flows have different hash values 1180 * 2) Two packets in the same flow should have the same hash value 1181 * 1182 * A hash at a higher layer is considered to be more specific. A driver should 1183 * set the most specific hash possible. 1184 * 1185 * A driver cannot indicate a more specific hash than the layer at which a hash 1186 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1187 * 1188 * A driver may indicate a hash level which is less specific than the 1189 * actual layer the hash was computed on. For instance, a hash computed 1190 * at L4 may be considered an L3 hash. This should only be done if the 1191 * driver can't unambiguously determine that the HW computed the hash at 1192 * the higher layer. Note that the "should" in the second property above 1193 * permits this. 1194 */ 1195enum pkt_hash_types { 1196 PKT_HASH_TYPE_NONE, /* Undefined type */ 1197 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1198 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1199 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1200}; 1201 1202static inline void skb_clear_hash(struct sk_buff *skb) 1203{ 1204 skb->hash = 0; 1205 skb->sw_hash = 0; 1206 skb->l4_hash = 0; 1207} 1208 1209static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1210{ 1211 if (!skb->l4_hash) 1212 skb_clear_hash(skb); 1213} 1214 1215static inline void 1216__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1217{ 1218 skb->l4_hash = is_l4; 1219 skb->sw_hash = is_sw; 1220 skb->hash = hash; 1221} 1222 1223static inline void 1224skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1225{ 1226 /* Used by drivers to set hash from HW */ 1227 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1228} 1229 1230static inline void 1231__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1232{ 1233 __skb_set_hash(skb, hash, true, is_l4); 1234} 1235 1236void __skb_get_hash(struct sk_buff *skb); 1237u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1238u32 skb_get_poff(const struct sk_buff *skb); 1239u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1240 const struct flow_keys_basic *keys, int hlen); 1241__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1242 void *data, int hlen_proto); 1243 1244static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1245 int thoff, u8 ip_proto) 1246{ 1247 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1248} 1249 1250void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1251 const struct flow_dissector_key *key, 1252 unsigned int key_count); 1253 1254#ifdef CONFIG_NET 1255int skb_flow_dissector_prog_query(const union bpf_attr *attr, 1256 union bpf_attr __user *uattr); 1257int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr, 1258 struct bpf_prog *prog); 1259 1260int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr); 1261#else 1262static inline int skb_flow_dissector_prog_query(const union bpf_attr *attr, 1263 union bpf_attr __user *uattr) 1264{ 1265 return -EOPNOTSUPP; 1266} 1267 1268static inline int skb_flow_dissector_bpf_prog_attach(const union bpf_attr *attr, 1269 struct bpf_prog *prog) 1270{ 1271 return -EOPNOTSUPP; 1272} 1273 1274static inline int skb_flow_dissector_bpf_prog_detach(const union bpf_attr *attr) 1275{ 1276 return -EOPNOTSUPP; 1277} 1278#endif 1279 1280struct bpf_flow_dissector; 1281bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1282 __be16 proto, int nhoff, int hlen, unsigned int flags); 1283 1284bool __skb_flow_dissect(const struct net *net, 1285 const struct sk_buff *skb, 1286 struct flow_dissector *flow_dissector, 1287 void *target_container, 1288 void *data, __be16 proto, int nhoff, int hlen, 1289 unsigned int flags); 1290 1291static inline bool skb_flow_dissect(const struct sk_buff *skb, 1292 struct flow_dissector *flow_dissector, 1293 void *target_container, unsigned int flags) 1294{ 1295 return __skb_flow_dissect(NULL, skb, flow_dissector, 1296 target_container, NULL, 0, 0, 0, flags); 1297} 1298 1299static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1300 struct flow_keys *flow, 1301 unsigned int flags) 1302{ 1303 memset(flow, 0, sizeof(*flow)); 1304 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1305 flow, NULL, 0, 0, 0, flags); 1306} 1307 1308static inline bool 1309skb_flow_dissect_flow_keys_basic(const struct net *net, 1310 const struct sk_buff *skb, 1311 struct flow_keys_basic *flow, void *data, 1312 __be16 proto, int nhoff, int hlen, 1313 unsigned int flags) 1314{ 1315 memset(flow, 0, sizeof(*flow)); 1316 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1317 data, proto, nhoff, hlen, flags); 1318} 1319 1320void skb_flow_dissect_meta(const struct sk_buff *skb, 1321 struct flow_dissector *flow_dissector, 1322 void *target_container); 1323 1324/* Gets a skb connection tracking info, ctinfo map should be a 1325 * a map of mapsize to translate enum ip_conntrack_info states 1326 * to user states. 1327 */ 1328void 1329skb_flow_dissect_ct(const struct sk_buff *skb, 1330 struct flow_dissector *flow_dissector, 1331 void *target_container, 1332 u16 *ctinfo_map, 1333 size_t mapsize); 1334void 1335skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1336 struct flow_dissector *flow_dissector, 1337 void *target_container); 1338 1339static inline __u32 skb_get_hash(struct sk_buff *skb) 1340{ 1341 if (!skb->l4_hash && !skb->sw_hash) 1342 __skb_get_hash(skb); 1343 1344 return skb->hash; 1345} 1346 1347static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1348{ 1349 if (!skb->l4_hash && !skb->sw_hash) { 1350 struct flow_keys keys; 1351 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1352 1353 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1354 } 1355 1356 return skb->hash; 1357} 1358 1359__u32 skb_get_hash_perturb(const struct sk_buff *skb, 1360 const siphash_key_t *perturb); 1361 1362static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1363{ 1364 return skb->hash; 1365} 1366 1367static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1368{ 1369 to->hash = from->hash; 1370 to->sw_hash = from->sw_hash; 1371 to->l4_hash = from->l4_hash; 1372}; 1373 1374static inline void skb_copy_decrypted(struct sk_buff *to, 1375 const struct sk_buff *from) 1376{ 1377#ifdef CONFIG_TLS_DEVICE 1378 to->decrypted = from->decrypted; 1379#endif 1380} 1381 1382#ifdef NET_SKBUFF_DATA_USES_OFFSET 1383static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1384{ 1385 return skb->head + skb->end; 1386} 1387 1388static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1389{ 1390 return skb->end; 1391} 1392#else 1393static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1394{ 1395 return skb->end; 1396} 1397 1398static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1399{ 1400 return skb->end - skb->head; 1401} 1402#endif 1403 1404/* Internal */ 1405#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1406 1407static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1408{ 1409 return &skb_shinfo(skb)->hwtstamps; 1410} 1411 1412static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1413{ 1414 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY; 1415 1416 return is_zcopy ? skb_uarg(skb) : NULL; 1417} 1418 1419static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1420 bool *have_ref) 1421{ 1422 if (skb && uarg && !skb_zcopy(skb)) { 1423 if (unlikely(have_ref && *have_ref)) 1424 *have_ref = false; 1425 else 1426 sock_zerocopy_get(uarg); 1427 skb_shinfo(skb)->destructor_arg = uarg; 1428 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; 1429 } 1430} 1431 1432static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1433{ 1434 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1435 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; 1436} 1437 1438static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1439{ 1440 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1441} 1442 1443static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1444{ 1445 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1446} 1447 1448/* Release a reference on a zerocopy structure */ 1449static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy) 1450{ 1451 struct ubuf_info *uarg = skb_zcopy(skb); 1452 1453 if (uarg) { 1454 if (skb_zcopy_is_nouarg(skb)) { 1455 /* no notification callback */ 1456 } else if (uarg->callback == sock_zerocopy_callback) { 1457 uarg->zerocopy = uarg->zerocopy && zerocopy; 1458 sock_zerocopy_put(uarg); 1459 } else { 1460 uarg->callback(uarg, zerocopy); 1461 } 1462 1463 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1464 } 1465} 1466 1467/* Abort a zerocopy operation and revert zckey on error in send syscall */ 1468static inline void skb_zcopy_abort(struct sk_buff *skb) 1469{ 1470 struct ubuf_info *uarg = skb_zcopy(skb); 1471 1472 if (uarg) { 1473 sock_zerocopy_put_abort(uarg, false); 1474 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1475 } 1476} 1477 1478static inline void skb_mark_not_on_list(struct sk_buff *skb) 1479{ 1480 skb->next = NULL; 1481} 1482 1483/* Iterate through singly-linked GSO fragments of an skb. */ 1484#define skb_list_walk_safe(first, skb, next_skb) \ 1485 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1486 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1487 1488static inline void skb_list_del_init(struct sk_buff *skb) 1489{ 1490 __list_del_entry(&skb->list); 1491 skb_mark_not_on_list(skb); 1492} 1493 1494/** 1495 * skb_queue_empty - check if a queue is empty 1496 * @list: queue head 1497 * 1498 * Returns true if the queue is empty, false otherwise. 1499 */ 1500static inline int skb_queue_empty(const struct sk_buff_head *list) 1501{ 1502 return list->next == (const struct sk_buff *) list; 1503} 1504 1505/** 1506 * skb_queue_empty_lockless - check if a queue is empty 1507 * @list: queue head 1508 * 1509 * Returns true if the queue is empty, false otherwise. 1510 * This variant can be used in lockless contexts. 1511 */ 1512static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1513{ 1514 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1515} 1516 1517 1518/** 1519 * skb_queue_is_last - check if skb is the last entry in the queue 1520 * @list: queue head 1521 * @skb: buffer 1522 * 1523 * Returns true if @skb is the last buffer on the list. 1524 */ 1525static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1526 const struct sk_buff *skb) 1527{ 1528 return skb->next == (const struct sk_buff *) list; 1529} 1530 1531/** 1532 * skb_queue_is_first - check if skb is the first entry in the queue 1533 * @list: queue head 1534 * @skb: buffer 1535 * 1536 * Returns true if @skb is the first buffer on the list. 1537 */ 1538static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1539 const struct sk_buff *skb) 1540{ 1541 return skb->prev == (const struct sk_buff *) list; 1542} 1543 1544/** 1545 * skb_queue_next - return the next packet in the queue 1546 * @list: queue head 1547 * @skb: current buffer 1548 * 1549 * Return the next packet in @list after @skb. It is only valid to 1550 * call this if skb_queue_is_last() evaluates to false. 1551 */ 1552static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1553 const struct sk_buff *skb) 1554{ 1555 /* This BUG_ON may seem severe, but if we just return then we 1556 * are going to dereference garbage. 1557 */ 1558 BUG_ON(skb_queue_is_last(list, skb)); 1559 return skb->next; 1560} 1561 1562/** 1563 * skb_queue_prev - return the prev packet in the queue 1564 * @list: queue head 1565 * @skb: current buffer 1566 * 1567 * Return the prev packet in @list before @skb. It is only valid to 1568 * call this if skb_queue_is_first() evaluates to false. 1569 */ 1570static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1571 const struct sk_buff *skb) 1572{ 1573 /* This BUG_ON may seem severe, but if we just return then we 1574 * are going to dereference garbage. 1575 */ 1576 BUG_ON(skb_queue_is_first(list, skb)); 1577 return skb->prev; 1578} 1579 1580/** 1581 * skb_get - reference buffer 1582 * @skb: buffer to reference 1583 * 1584 * Makes another reference to a socket buffer and returns a pointer 1585 * to the buffer. 1586 */ 1587static inline struct sk_buff *skb_get(struct sk_buff *skb) 1588{ 1589 refcount_inc(&skb->users); 1590 return skb; 1591} 1592 1593/* 1594 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1595 */ 1596 1597/** 1598 * skb_cloned - is the buffer a clone 1599 * @skb: buffer to check 1600 * 1601 * Returns true if the buffer was generated with skb_clone() and is 1602 * one of multiple shared copies of the buffer. Cloned buffers are 1603 * shared data so must not be written to under normal circumstances. 1604 */ 1605static inline int skb_cloned(const struct sk_buff *skb) 1606{ 1607 return skb->cloned && 1608 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1609} 1610 1611static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1612{ 1613 might_sleep_if(gfpflags_allow_blocking(pri)); 1614 1615 if (skb_cloned(skb)) 1616 return pskb_expand_head(skb, 0, 0, pri); 1617 1618 return 0; 1619} 1620 1621/** 1622 * skb_header_cloned - is the header a clone 1623 * @skb: buffer to check 1624 * 1625 * Returns true if modifying the header part of the buffer requires 1626 * the data to be copied. 1627 */ 1628static inline int skb_header_cloned(const struct sk_buff *skb) 1629{ 1630 int dataref; 1631 1632 if (!skb->cloned) 1633 return 0; 1634 1635 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1636 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1637 return dataref != 1; 1638} 1639 1640static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1641{ 1642 might_sleep_if(gfpflags_allow_blocking(pri)); 1643 1644 if (skb_header_cloned(skb)) 1645 return pskb_expand_head(skb, 0, 0, pri); 1646 1647 return 0; 1648} 1649 1650/** 1651 * __skb_header_release - release reference to header 1652 * @skb: buffer to operate on 1653 */ 1654static inline void __skb_header_release(struct sk_buff *skb) 1655{ 1656 skb->nohdr = 1; 1657 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1658} 1659 1660 1661/** 1662 * skb_shared - is the buffer shared 1663 * @skb: buffer to check 1664 * 1665 * Returns true if more than one person has a reference to this 1666 * buffer. 1667 */ 1668static inline int skb_shared(const struct sk_buff *skb) 1669{ 1670 return refcount_read(&skb->users) != 1; 1671} 1672 1673/** 1674 * skb_share_check - check if buffer is shared and if so clone it 1675 * @skb: buffer to check 1676 * @pri: priority for memory allocation 1677 * 1678 * If the buffer is shared the buffer is cloned and the old copy 1679 * drops a reference. A new clone with a single reference is returned. 1680 * If the buffer is not shared the original buffer is returned. When 1681 * being called from interrupt status or with spinlocks held pri must 1682 * be GFP_ATOMIC. 1683 * 1684 * NULL is returned on a memory allocation failure. 1685 */ 1686static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1687{ 1688 might_sleep_if(gfpflags_allow_blocking(pri)); 1689 if (skb_shared(skb)) { 1690 struct sk_buff *nskb = skb_clone(skb, pri); 1691 1692 if (likely(nskb)) 1693 consume_skb(skb); 1694 else 1695 kfree_skb(skb); 1696 skb = nskb; 1697 } 1698 return skb; 1699} 1700 1701/* 1702 * Copy shared buffers into a new sk_buff. We effectively do COW on 1703 * packets to handle cases where we have a local reader and forward 1704 * and a couple of other messy ones. The normal one is tcpdumping 1705 * a packet thats being forwarded. 1706 */ 1707 1708/** 1709 * skb_unshare - make a copy of a shared buffer 1710 * @skb: buffer to check 1711 * @pri: priority for memory allocation 1712 * 1713 * If the socket buffer is a clone then this function creates a new 1714 * copy of the data, drops a reference count on the old copy and returns 1715 * the new copy with the reference count at 1. If the buffer is not a clone 1716 * the original buffer is returned. When called with a spinlock held or 1717 * from interrupt state @pri must be %GFP_ATOMIC 1718 * 1719 * %NULL is returned on a memory allocation failure. 1720 */ 1721static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1722 gfp_t pri) 1723{ 1724 might_sleep_if(gfpflags_allow_blocking(pri)); 1725 if (skb_cloned(skb)) { 1726 struct sk_buff *nskb = skb_copy(skb, pri); 1727 1728 /* Free our shared copy */ 1729 if (likely(nskb)) 1730 consume_skb(skb); 1731 else 1732 kfree_skb(skb); 1733 skb = nskb; 1734 } 1735 return skb; 1736} 1737 1738/** 1739 * skb_peek - peek at the head of an &sk_buff_head 1740 * @list_: list to peek at 1741 * 1742 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1743 * be careful with this one. A peek leaves the buffer on the 1744 * list and someone else may run off with it. You must hold 1745 * the appropriate locks or have a private queue to do this. 1746 * 1747 * Returns %NULL for an empty list or a pointer to the head element. 1748 * The reference count is not incremented and the reference is therefore 1749 * volatile. Use with caution. 1750 */ 1751static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1752{ 1753 struct sk_buff *skb = list_->next; 1754 1755 if (skb == (struct sk_buff *)list_) 1756 skb = NULL; 1757 return skb; 1758} 1759 1760/** 1761 * __skb_peek - peek at the head of a non-empty &sk_buff_head 1762 * @list_: list to peek at 1763 * 1764 * Like skb_peek(), but the caller knows that the list is not empty. 1765 */ 1766static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 1767{ 1768 return list_->next; 1769} 1770 1771/** 1772 * skb_peek_next - peek skb following the given one from a queue 1773 * @skb: skb to start from 1774 * @list_: list to peek at 1775 * 1776 * Returns %NULL when the end of the list is met or a pointer to the 1777 * next element. The reference count is not incremented and the 1778 * reference is therefore volatile. Use with caution. 1779 */ 1780static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1781 const struct sk_buff_head *list_) 1782{ 1783 struct sk_buff *next = skb->next; 1784 1785 if (next == (struct sk_buff *)list_) 1786 next = NULL; 1787 return next; 1788} 1789 1790/** 1791 * skb_peek_tail - peek at the tail of an &sk_buff_head 1792 * @list_: list to peek at 1793 * 1794 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1795 * be careful with this one. A peek leaves the buffer on the 1796 * list and someone else may run off with it. You must hold 1797 * the appropriate locks or have a private queue to do this. 1798 * 1799 * Returns %NULL for an empty list or a pointer to the tail element. 1800 * The reference count is not incremented and the reference is therefore 1801 * volatile. Use with caution. 1802 */ 1803static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1804{ 1805 struct sk_buff *skb = READ_ONCE(list_->prev); 1806 1807 if (skb == (struct sk_buff *)list_) 1808 skb = NULL; 1809 return skb; 1810 1811} 1812 1813/** 1814 * skb_queue_len - get queue length 1815 * @list_: list to measure 1816 * 1817 * Return the length of an &sk_buff queue. 1818 */ 1819static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1820{ 1821 return list_->qlen; 1822} 1823 1824/** 1825 * skb_queue_len_lockless - get queue length 1826 * @list_: list to measure 1827 * 1828 * Return the length of an &sk_buff queue. 1829 * This variant can be used in lockless contexts. 1830 */ 1831static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 1832{ 1833 return READ_ONCE(list_->qlen); 1834} 1835 1836/** 1837 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1838 * @list: queue to initialize 1839 * 1840 * This initializes only the list and queue length aspects of 1841 * an sk_buff_head object. This allows to initialize the list 1842 * aspects of an sk_buff_head without reinitializing things like 1843 * the spinlock. It can also be used for on-stack sk_buff_head 1844 * objects where the spinlock is known to not be used. 1845 */ 1846static inline void __skb_queue_head_init(struct sk_buff_head *list) 1847{ 1848 list->prev = list->next = (struct sk_buff *)list; 1849 list->qlen = 0; 1850} 1851 1852/* 1853 * This function creates a split out lock class for each invocation; 1854 * this is needed for now since a whole lot of users of the skb-queue 1855 * infrastructure in drivers have different locking usage (in hardirq) 1856 * than the networking core (in softirq only). In the long run either the 1857 * network layer or drivers should need annotation to consolidate the 1858 * main types of usage into 3 classes. 1859 */ 1860static inline void skb_queue_head_init(struct sk_buff_head *list) 1861{ 1862 spin_lock_init(&list->lock); 1863 __skb_queue_head_init(list); 1864} 1865 1866static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1867 struct lock_class_key *class) 1868{ 1869 skb_queue_head_init(list); 1870 lockdep_set_class(&list->lock, class); 1871} 1872 1873/* 1874 * Insert an sk_buff on a list. 1875 * 1876 * The "__skb_xxxx()" functions are the non-atomic ones that 1877 * can only be called with interrupts disabled. 1878 */ 1879static inline void __skb_insert(struct sk_buff *newsk, 1880 struct sk_buff *prev, struct sk_buff *next, 1881 struct sk_buff_head *list) 1882{ 1883 /* See skb_queue_empty_lockless() and skb_peek_tail() 1884 * for the opposite READ_ONCE() 1885 */ 1886 WRITE_ONCE(newsk->next, next); 1887 WRITE_ONCE(newsk->prev, prev); 1888 WRITE_ONCE(next->prev, newsk); 1889 WRITE_ONCE(prev->next, newsk); 1890 list->qlen++; 1891} 1892 1893static inline void __skb_queue_splice(const struct sk_buff_head *list, 1894 struct sk_buff *prev, 1895 struct sk_buff *next) 1896{ 1897 struct sk_buff *first = list->next; 1898 struct sk_buff *last = list->prev; 1899 1900 WRITE_ONCE(first->prev, prev); 1901 WRITE_ONCE(prev->next, first); 1902 1903 WRITE_ONCE(last->next, next); 1904 WRITE_ONCE(next->prev, last); 1905} 1906 1907/** 1908 * skb_queue_splice - join two skb lists, this is designed for stacks 1909 * @list: the new list to add 1910 * @head: the place to add it in the first list 1911 */ 1912static inline void skb_queue_splice(const struct sk_buff_head *list, 1913 struct sk_buff_head *head) 1914{ 1915 if (!skb_queue_empty(list)) { 1916 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1917 head->qlen += list->qlen; 1918 } 1919} 1920 1921/** 1922 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1923 * @list: the new list to add 1924 * @head: the place to add it in the first list 1925 * 1926 * The list at @list is reinitialised 1927 */ 1928static inline void skb_queue_splice_init(struct sk_buff_head *list, 1929 struct sk_buff_head *head) 1930{ 1931 if (!skb_queue_empty(list)) { 1932 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1933 head->qlen += list->qlen; 1934 __skb_queue_head_init(list); 1935 } 1936} 1937 1938/** 1939 * skb_queue_splice_tail - join two skb lists, each list being a queue 1940 * @list: the new list to add 1941 * @head: the place to add it in the first list 1942 */ 1943static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1944 struct sk_buff_head *head) 1945{ 1946 if (!skb_queue_empty(list)) { 1947 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1948 head->qlen += list->qlen; 1949 } 1950} 1951 1952/** 1953 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1954 * @list: the new list to add 1955 * @head: the place to add it in the first list 1956 * 1957 * Each of the lists is a queue. 1958 * The list at @list is reinitialised 1959 */ 1960static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1961 struct sk_buff_head *head) 1962{ 1963 if (!skb_queue_empty(list)) { 1964 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1965 head->qlen += list->qlen; 1966 __skb_queue_head_init(list); 1967 } 1968} 1969 1970/** 1971 * __skb_queue_after - queue a buffer at the list head 1972 * @list: list to use 1973 * @prev: place after this buffer 1974 * @newsk: buffer to queue 1975 * 1976 * Queue a buffer int the middle of a list. This function takes no locks 1977 * and you must therefore hold required locks before calling it. 1978 * 1979 * A buffer cannot be placed on two lists at the same time. 1980 */ 1981static inline void __skb_queue_after(struct sk_buff_head *list, 1982 struct sk_buff *prev, 1983 struct sk_buff *newsk) 1984{ 1985 __skb_insert(newsk, prev, prev->next, list); 1986} 1987 1988void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1989 struct sk_buff_head *list); 1990 1991static inline void __skb_queue_before(struct sk_buff_head *list, 1992 struct sk_buff *next, 1993 struct sk_buff *newsk) 1994{ 1995 __skb_insert(newsk, next->prev, next, list); 1996} 1997 1998/** 1999 * __skb_queue_head - queue a buffer at the list head 2000 * @list: list to use 2001 * @newsk: buffer to queue 2002 * 2003 * Queue a buffer at the start of a list. This function takes no locks 2004 * and you must therefore hold required locks before calling it. 2005 * 2006 * A buffer cannot be placed on two lists at the same time. 2007 */ 2008static inline void __skb_queue_head(struct sk_buff_head *list, 2009 struct sk_buff *newsk) 2010{ 2011 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2012} 2013void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2014 2015/** 2016 * __skb_queue_tail - queue a buffer at the list tail 2017 * @list: list to use 2018 * @newsk: buffer to queue 2019 * 2020 * Queue a buffer at the end of a list. This function takes no locks 2021 * and you must therefore hold required locks before calling it. 2022 * 2023 * A buffer cannot be placed on two lists at the same time. 2024 */ 2025static inline void __skb_queue_tail(struct sk_buff_head *list, 2026 struct sk_buff *newsk) 2027{ 2028 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2029} 2030void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2031 2032/* 2033 * remove sk_buff from list. _Must_ be called atomically, and with 2034 * the list known.. 2035 */ 2036void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2037static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2038{ 2039 struct sk_buff *next, *prev; 2040 2041 WRITE_ONCE(list->qlen, list->qlen - 1); 2042 next = skb->next; 2043 prev = skb->prev; 2044 skb->next = skb->prev = NULL; 2045 WRITE_ONCE(next->prev, prev); 2046 WRITE_ONCE(prev->next, next); 2047} 2048 2049/** 2050 * __skb_dequeue - remove from the head of the queue 2051 * @list: list to dequeue from 2052 * 2053 * Remove the head of the list. This function does not take any locks 2054 * so must be used with appropriate locks held only. The head item is 2055 * returned or %NULL if the list is empty. 2056 */ 2057static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2058{ 2059 struct sk_buff *skb = skb_peek(list); 2060 if (skb) 2061 __skb_unlink(skb, list); 2062 return skb; 2063} 2064struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2065 2066/** 2067 * __skb_dequeue_tail - remove from the tail of the queue 2068 * @list: list to dequeue from 2069 * 2070 * Remove the tail of the list. This function does not take any locks 2071 * so must be used with appropriate locks held only. The tail item is 2072 * returned or %NULL if the list is empty. 2073 */ 2074static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2075{ 2076 struct sk_buff *skb = skb_peek_tail(list); 2077 if (skb) 2078 __skb_unlink(skb, list); 2079 return skb; 2080} 2081struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2082 2083 2084static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2085{ 2086 return skb->data_len; 2087} 2088 2089static inline unsigned int skb_headlen(const struct sk_buff *skb) 2090{ 2091 return skb->len - skb->data_len; 2092} 2093 2094static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2095{ 2096 unsigned int i, len = 0; 2097 2098 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2099 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2100 return len; 2101} 2102 2103static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2104{ 2105 return skb_headlen(skb) + __skb_pagelen(skb); 2106} 2107 2108/** 2109 * __skb_fill_page_desc - initialise a paged fragment in an skb 2110 * @skb: buffer containing fragment to be initialised 2111 * @i: paged fragment index to initialise 2112 * @page: the page to use for this fragment 2113 * @off: the offset to the data with @page 2114 * @size: the length of the data 2115 * 2116 * Initialises the @i'th fragment of @skb to point to &size bytes at 2117 * offset @off within @page. 2118 * 2119 * Does not take any additional reference on the fragment. 2120 */ 2121static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2122 struct page *page, int off, int size) 2123{ 2124 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2125 2126 /* 2127 * Propagate page pfmemalloc to the skb if we can. The problem is 2128 * that not all callers have unique ownership of the page but rely 2129 * on page_is_pfmemalloc doing the right thing(tm). 2130 */ 2131 frag->bv_page = page; 2132 frag->bv_offset = off; 2133 skb_frag_size_set(frag, size); 2134 2135 page = compound_head(page); 2136 if (page_is_pfmemalloc(page)) 2137 skb->pfmemalloc = true; 2138} 2139 2140/** 2141 * skb_fill_page_desc - initialise a paged fragment in an skb 2142 * @skb: buffer containing fragment to be initialised 2143 * @i: paged fragment index to initialise 2144 * @page: the page to use for this fragment 2145 * @off: the offset to the data with @page 2146 * @size: the length of the data 2147 * 2148 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2149 * @skb to point to @size bytes at offset @off within @page. In 2150 * addition updates @skb such that @i is the last fragment. 2151 * 2152 * Does not take any additional reference on the fragment. 2153 */ 2154static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2155 struct page *page, int off, int size) 2156{ 2157 __skb_fill_page_desc(skb, i, page, off, size); 2158 skb_shinfo(skb)->nr_frags = i + 1; 2159} 2160 2161void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 2162 int size, unsigned int truesize); 2163 2164void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2165 unsigned int truesize); 2166 2167#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2168 2169#ifdef NET_SKBUFF_DATA_USES_OFFSET 2170static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2171{ 2172 return skb->head + skb->tail; 2173} 2174 2175static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2176{ 2177 skb->tail = skb->data - skb->head; 2178} 2179 2180static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2181{ 2182 skb_reset_tail_pointer(skb); 2183 skb->tail += offset; 2184} 2185 2186#else /* NET_SKBUFF_DATA_USES_OFFSET */ 2187static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2188{ 2189 return skb->tail; 2190} 2191 2192static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2193{ 2194 skb->tail = skb->data; 2195} 2196 2197static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2198{ 2199 skb->tail = skb->data + offset; 2200} 2201 2202#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2203 2204/* 2205 * Add data to an sk_buff 2206 */ 2207void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2208void *skb_put(struct sk_buff *skb, unsigned int len); 2209static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2210{ 2211 void *tmp = skb_tail_pointer(skb); 2212 SKB_LINEAR_ASSERT(skb); 2213 skb->tail += len; 2214 skb->len += len; 2215 return tmp; 2216} 2217 2218static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2219{ 2220 void *tmp = __skb_put(skb, len); 2221 2222 memset(tmp, 0, len); 2223 return tmp; 2224} 2225 2226static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2227 unsigned int len) 2228{ 2229 void *tmp = __skb_put(skb, len); 2230 2231 memcpy(tmp, data, len); 2232 return tmp; 2233} 2234 2235static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2236{ 2237 *(u8 *)__skb_put(skb, 1) = val; 2238} 2239 2240static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2241{ 2242 void *tmp = skb_put(skb, len); 2243 2244 memset(tmp, 0, len); 2245 2246 return tmp; 2247} 2248 2249static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2250 unsigned int len) 2251{ 2252 void *tmp = skb_put(skb, len); 2253 2254 memcpy(tmp, data, len); 2255 2256 return tmp; 2257} 2258 2259static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2260{ 2261 *(u8 *)skb_put(skb, 1) = val; 2262} 2263 2264void *skb_push(struct sk_buff *skb, unsigned int len); 2265static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2266{ 2267 skb->data -= len; 2268 skb->len += len; 2269 return skb->data; 2270} 2271 2272void *skb_pull(struct sk_buff *skb, unsigned int len); 2273static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2274{ 2275 skb->len -= len; 2276 BUG_ON(skb->len < skb->data_len); 2277 return skb->data += len; 2278} 2279 2280static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2281{ 2282 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2283} 2284 2285void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2286 2287static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len) 2288{ 2289 if (len > skb_headlen(skb) && 2290 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 2291 return NULL; 2292 skb->len -= len; 2293 return skb->data += len; 2294} 2295 2296static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2297{ 2298 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 2299} 2300 2301static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2302{ 2303 if (likely(len <= skb_headlen(skb))) 2304 return true; 2305 if (unlikely(len > skb->len)) 2306 return false; 2307 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2308} 2309 2310void skb_condense(struct sk_buff *skb); 2311 2312/** 2313 * skb_headroom - bytes at buffer head 2314 * @skb: buffer to check 2315 * 2316 * Return the number of bytes of free space at the head of an &sk_buff. 2317 */ 2318static inline unsigned int skb_headroom(const struct sk_buff *skb) 2319{ 2320 return skb->data - skb->head; 2321} 2322 2323/** 2324 * skb_tailroom - bytes at buffer end 2325 * @skb: buffer to check 2326 * 2327 * Return the number of bytes of free space at the tail of an sk_buff 2328 */ 2329static inline int skb_tailroom(const struct sk_buff *skb) 2330{ 2331 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2332} 2333 2334/** 2335 * skb_availroom - bytes at buffer end 2336 * @skb: buffer to check 2337 * 2338 * Return the number of bytes of free space at the tail of an sk_buff 2339 * allocated by sk_stream_alloc() 2340 */ 2341static inline int skb_availroom(const struct sk_buff *skb) 2342{ 2343 if (skb_is_nonlinear(skb)) 2344 return 0; 2345 2346 return skb->end - skb->tail - skb->reserved_tailroom; 2347} 2348 2349/** 2350 * skb_reserve - adjust headroom 2351 * @skb: buffer to alter 2352 * @len: bytes to move 2353 * 2354 * Increase the headroom of an empty &sk_buff by reducing the tail 2355 * room. This is only allowed for an empty buffer. 2356 */ 2357static inline void skb_reserve(struct sk_buff *skb, int len) 2358{ 2359 skb->data += len; 2360 skb->tail += len; 2361} 2362 2363/** 2364 * skb_tailroom_reserve - adjust reserved_tailroom 2365 * @skb: buffer to alter 2366 * @mtu: maximum amount of headlen permitted 2367 * @needed_tailroom: minimum amount of reserved_tailroom 2368 * 2369 * Set reserved_tailroom so that headlen can be as large as possible but 2370 * not larger than mtu and tailroom cannot be smaller than 2371 * needed_tailroom. 2372 * The required headroom should already have been reserved before using 2373 * this function. 2374 */ 2375static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2376 unsigned int needed_tailroom) 2377{ 2378 SKB_LINEAR_ASSERT(skb); 2379 if (mtu < skb_tailroom(skb) - needed_tailroom) 2380 /* use at most mtu */ 2381 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2382 else 2383 /* use up to all available space */ 2384 skb->reserved_tailroom = needed_tailroom; 2385} 2386 2387#define ENCAP_TYPE_ETHER 0 2388#define ENCAP_TYPE_IPPROTO 1 2389 2390static inline void skb_set_inner_protocol(struct sk_buff *skb, 2391 __be16 protocol) 2392{ 2393 skb->inner_protocol = protocol; 2394 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2395} 2396 2397static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2398 __u8 ipproto) 2399{ 2400 skb->inner_ipproto = ipproto; 2401 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2402} 2403 2404static inline void skb_reset_inner_headers(struct sk_buff *skb) 2405{ 2406 skb->inner_mac_header = skb->mac_header; 2407 skb->inner_network_header = skb->network_header; 2408 skb->inner_transport_header = skb->transport_header; 2409} 2410 2411static inline void skb_reset_mac_len(struct sk_buff *skb) 2412{ 2413 skb->mac_len = skb->network_header - skb->mac_header; 2414} 2415 2416static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2417 *skb) 2418{ 2419 return skb->head + skb->inner_transport_header; 2420} 2421 2422static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2423{ 2424 return skb_inner_transport_header(skb) - skb->data; 2425} 2426 2427static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2428{ 2429 skb->inner_transport_header = skb->data - skb->head; 2430} 2431 2432static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2433 const int offset) 2434{ 2435 skb_reset_inner_transport_header(skb); 2436 skb->inner_transport_header += offset; 2437} 2438 2439static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2440{ 2441 return skb->head + skb->inner_network_header; 2442} 2443 2444static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2445{ 2446 skb->inner_network_header = skb->data - skb->head; 2447} 2448 2449static inline void skb_set_inner_network_header(struct sk_buff *skb, 2450 const int offset) 2451{ 2452 skb_reset_inner_network_header(skb); 2453 skb->inner_network_header += offset; 2454} 2455 2456static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2457{ 2458 return skb->head + skb->inner_mac_header; 2459} 2460 2461static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2462{ 2463 skb->inner_mac_header = skb->data - skb->head; 2464} 2465 2466static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2467 const int offset) 2468{ 2469 skb_reset_inner_mac_header(skb); 2470 skb->inner_mac_header += offset; 2471} 2472static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2473{ 2474 return skb->transport_header != (typeof(skb->transport_header))~0U; 2475} 2476 2477static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2478{ 2479 return skb->head + skb->transport_header; 2480} 2481 2482static inline void skb_reset_transport_header(struct sk_buff *skb) 2483{ 2484 skb->transport_header = skb->data - skb->head; 2485} 2486 2487static inline void skb_set_transport_header(struct sk_buff *skb, 2488 const int offset) 2489{ 2490 skb_reset_transport_header(skb); 2491 skb->transport_header += offset; 2492} 2493 2494static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2495{ 2496 return skb->head + skb->network_header; 2497} 2498 2499static inline void skb_reset_network_header(struct sk_buff *skb) 2500{ 2501 skb->network_header = skb->data - skb->head; 2502} 2503 2504static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2505{ 2506 skb_reset_network_header(skb); 2507 skb->network_header += offset; 2508} 2509 2510static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2511{ 2512 return skb->head + skb->mac_header; 2513} 2514 2515static inline int skb_mac_offset(const struct sk_buff *skb) 2516{ 2517 return skb_mac_header(skb) - skb->data; 2518} 2519 2520static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2521{ 2522 return skb->network_header - skb->mac_header; 2523} 2524 2525static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2526{ 2527 return skb->mac_header != (typeof(skb->mac_header))~0U; 2528} 2529 2530static inline void skb_reset_mac_header(struct sk_buff *skb) 2531{ 2532 skb->mac_header = skb->data - skb->head; 2533} 2534 2535static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2536{ 2537 skb_reset_mac_header(skb); 2538 skb->mac_header += offset; 2539} 2540 2541static inline void skb_pop_mac_header(struct sk_buff *skb) 2542{ 2543 skb->mac_header = skb->network_header; 2544} 2545 2546static inline void skb_probe_transport_header(struct sk_buff *skb) 2547{ 2548 struct flow_keys_basic keys; 2549 2550 if (skb_transport_header_was_set(skb)) 2551 return; 2552 2553 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 2554 NULL, 0, 0, 0, 0)) 2555 skb_set_transport_header(skb, keys.control.thoff); 2556} 2557 2558static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2559{ 2560 if (skb_mac_header_was_set(skb)) { 2561 const unsigned char *old_mac = skb_mac_header(skb); 2562 2563 skb_set_mac_header(skb, -skb->mac_len); 2564 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2565 } 2566} 2567 2568static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2569{ 2570 return skb->csum_start - skb_headroom(skb); 2571} 2572 2573static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2574{ 2575 return skb->head + skb->csum_start; 2576} 2577 2578static inline int skb_transport_offset(const struct sk_buff *skb) 2579{ 2580 return skb_transport_header(skb) - skb->data; 2581} 2582 2583static inline u32 skb_network_header_len(const struct sk_buff *skb) 2584{ 2585 return skb->transport_header - skb->network_header; 2586} 2587 2588static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2589{ 2590 return skb->inner_transport_header - skb->inner_network_header; 2591} 2592 2593static inline int skb_network_offset(const struct sk_buff *skb) 2594{ 2595 return skb_network_header(skb) - skb->data; 2596} 2597 2598static inline int skb_inner_network_offset(const struct sk_buff *skb) 2599{ 2600 return skb_inner_network_header(skb) - skb->data; 2601} 2602 2603static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2604{ 2605 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2606} 2607 2608/* 2609 * CPUs often take a performance hit when accessing unaligned memory 2610 * locations. The actual performance hit varies, it can be small if the 2611 * hardware handles it or large if we have to take an exception and fix it 2612 * in software. 2613 * 2614 * Since an ethernet header is 14 bytes network drivers often end up with 2615 * the IP header at an unaligned offset. The IP header can be aligned by 2616 * shifting the start of the packet by 2 bytes. Drivers should do this 2617 * with: 2618 * 2619 * skb_reserve(skb, NET_IP_ALIGN); 2620 * 2621 * The downside to this alignment of the IP header is that the DMA is now 2622 * unaligned. On some architectures the cost of an unaligned DMA is high 2623 * and this cost outweighs the gains made by aligning the IP header. 2624 * 2625 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2626 * to be overridden. 2627 */ 2628#ifndef NET_IP_ALIGN 2629#define NET_IP_ALIGN 2 2630#endif 2631 2632/* 2633 * The networking layer reserves some headroom in skb data (via 2634 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2635 * the header has to grow. In the default case, if the header has to grow 2636 * 32 bytes or less we avoid the reallocation. 2637 * 2638 * Unfortunately this headroom changes the DMA alignment of the resulting 2639 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2640 * on some architectures. An architecture can override this value, 2641 * perhaps setting it to a cacheline in size (since that will maintain 2642 * cacheline alignment of the DMA). It must be a power of 2. 2643 * 2644 * Various parts of the networking layer expect at least 32 bytes of 2645 * headroom, you should not reduce this. 2646 * 2647 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2648 * to reduce average number of cache lines per packet. 2649 * get_rps_cpus() for example only access one 64 bytes aligned block : 2650 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2651 */ 2652#ifndef NET_SKB_PAD 2653#define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2654#endif 2655 2656int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2657 2658static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2659{ 2660 if (WARN_ON(skb_is_nonlinear(skb))) 2661 return; 2662 skb->len = len; 2663 skb_set_tail_pointer(skb, len); 2664} 2665 2666static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2667{ 2668 __skb_set_length(skb, len); 2669} 2670 2671void skb_trim(struct sk_buff *skb, unsigned int len); 2672 2673static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2674{ 2675 if (skb->data_len) 2676 return ___pskb_trim(skb, len); 2677 __skb_trim(skb, len); 2678 return 0; 2679} 2680 2681static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2682{ 2683 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2684} 2685 2686/** 2687 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2688 * @skb: buffer to alter 2689 * @len: new length 2690 * 2691 * This is identical to pskb_trim except that the caller knows that 2692 * the skb is not cloned so we should never get an error due to out- 2693 * of-memory. 2694 */ 2695static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2696{ 2697 int err = pskb_trim(skb, len); 2698 BUG_ON(err); 2699} 2700 2701static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2702{ 2703 unsigned int diff = len - skb->len; 2704 2705 if (skb_tailroom(skb) < diff) { 2706 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2707 GFP_ATOMIC); 2708 if (ret) 2709 return ret; 2710 } 2711 __skb_set_length(skb, len); 2712 return 0; 2713} 2714 2715/** 2716 * skb_orphan - orphan a buffer 2717 * @skb: buffer to orphan 2718 * 2719 * If a buffer currently has an owner then we call the owner's 2720 * destructor function and make the @skb unowned. The buffer continues 2721 * to exist but is no longer charged to its former owner. 2722 */ 2723static inline void skb_orphan(struct sk_buff *skb) 2724{ 2725 if (skb->destructor) { 2726 skb->destructor(skb); 2727 skb->destructor = NULL; 2728 skb->sk = NULL; 2729 } else { 2730 BUG_ON(skb->sk); 2731 } 2732} 2733 2734/** 2735 * skb_orphan_frags - orphan the frags contained in a buffer 2736 * @skb: buffer to orphan frags from 2737 * @gfp_mask: allocation mask for replacement pages 2738 * 2739 * For each frag in the SKB which needs a destructor (i.e. has an 2740 * owner) create a copy of that frag and release the original 2741 * page by calling the destructor. 2742 */ 2743static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2744{ 2745 if (likely(!skb_zcopy(skb))) 2746 return 0; 2747 if (!skb_zcopy_is_nouarg(skb) && 2748 skb_uarg(skb)->callback == sock_zerocopy_callback) 2749 return 0; 2750 return skb_copy_ubufs(skb, gfp_mask); 2751} 2752 2753/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 2754static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 2755{ 2756 if (likely(!skb_zcopy(skb))) 2757 return 0; 2758 return skb_copy_ubufs(skb, gfp_mask); 2759} 2760 2761/** 2762 * __skb_queue_purge - empty a list 2763 * @list: list to empty 2764 * 2765 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2766 * the list and one reference dropped. This function does not take the 2767 * list lock and the caller must hold the relevant locks to use it. 2768 */ 2769static inline void __skb_queue_purge(struct sk_buff_head *list) 2770{ 2771 struct sk_buff *skb; 2772 while ((skb = __skb_dequeue(list)) != NULL) 2773 kfree_skb(skb); 2774} 2775void skb_queue_purge(struct sk_buff_head *list); 2776 2777unsigned int skb_rbtree_purge(struct rb_root *root); 2778 2779void *netdev_alloc_frag(unsigned int fragsz); 2780 2781struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2782 gfp_t gfp_mask); 2783 2784/** 2785 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2786 * @dev: network device to receive on 2787 * @length: length to allocate 2788 * 2789 * Allocate a new &sk_buff and assign it a usage count of one. The 2790 * buffer has unspecified headroom built in. Users should allocate 2791 * the headroom they think they need without accounting for the 2792 * built in space. The built in space is used for optimisations. 2793 * 2794 * %NULL is returned if there is no free memory. Although this function 2795 * allocates memory it can be called from an interrupt. 2796 */ 2797static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2798 unsigned int length) 2799{ 2800 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2801} 2802 2803/* legacy helper around __netdev_alloc_skb() */ 2804static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2805 gfp_t gfp_mask) 2806{ 2807 return __netdev_alloc_skb(NULL, length, gfp_mask); 2808} 2809 2810/* legacy helper around netdev_alloc_skb() */ 2811static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2812{ 2813 return netdev_alloc_skb(NULL, length); 2814} 2815 2816 2817static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2818 unsigned int length, gfp_t gfp) 2819{ 2820 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2821 2822 if (NET_IP_ALIGN && skb) 2823 skb_reserve(skb, NET_IP_ALIGN); 2824 return skb; 2825} 2826 2827static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2828 unsigned int length) 2829{ 2830 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2831} 2832 2833static inline void skb_free_frag(void *addr) 2834{ 2835 page_frag_free(addr); 2836} 2837 2838void *napi_alloc_frag(unsigned int fragsz); 2839struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2840 unsigned int length, gfp_t gfp_mask); 2841static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2842 unsigned int length) 2843{ 2844 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2845} 2846void napi_consume_skb(struct sk_buff *skb, int budget); 2847 2848void __kfree_skb_flush(void); 2849void __kfree_skb_defer(struct sk_buff *skb); 2850 2851/** 2852 * __dev_alloc_pages - allocate page for network Rx 2853 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2854 * @order: size of the allocation 2855 * 2856 * Allocate a new page. 2857 * 2858 * %NULL is returned if there is no free memory. 2859*/ 2860static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2861 unsigned int order) 2862{ 2863 /* This piece of code contains several assumptions. 2864 * 1. This is for device Rx, therefor a cold page is preferred. 2865 * 2. The expectation is the user wants a compound page. 2866 * 3. If requesting a order 0 page it will not be compound 2867 * due to the check to see if order has a value in prep_new_page 2868 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2869 * code in gfp_to_alloc_flags that should be enforcing this. 2870 */ 2871 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 2872 2873 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2874} 2875 2876static inline struct page *dev_alloc_pages(unsigned int order) 2877{ 2878 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 2879} 2880 2881/** 2882 * __dev_alloc_page - allocate a page for network Rx 2883 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2884 * 2885 * Allocate a new page. 2886 * 2887 * %NULL is returned if there is no free memory. 2888 */ 2889static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2890{ 2891 return __dev_alloc_pages(gfp_mask, 0); 2892} 2893 2894static inline struct page *dev_alloc_page(void) 2895{ 2896 return dev_alloc_pages(0); 2897} 2898 2899/** 2900 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2901 * @page: The page that was allocated from skb_alloc_page 2902 * @skb: The skb that may need pfmemalloc set 2903 */ 2904static inline void skb_propagate_pfmemalloc(struct page *page, 2905 struct sk_buff *skb) 2906{ 2907 if (page_is_pfmemalloc(page)) 2908 skb->pfmemalloc = true; 2909} 2910 2911/** 2912 * skb_frag_off() - Returns the offset of a skb fragment 2913 * @frag: the paged fragment 2914 */ 2915static inline unsigned int skb_frag_off(const skb_frag_t *frag) 2916{ 2917 return frag->bv_offset; 2918} 2919 2920/** 2921 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 2922 * @frag: skb fragment 2923 * @delta: value to add 2924 */ 2925static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 2926{ 2927 frag->bv_offset += delta; 2928} 2929 2930/** 2931 * skb_frag_off_set() - Sets the offset of a skb fragment 2932 * @frag: skb fragment 2933 * @offset: offset of fragment 2934 */ 2935static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 2936{ 2937 frag->bv_offset = offset; 2938} 2939 2940/** 2941 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 2942 * @fragto: skb fragment where offset is set 2943 * @fragfrom: skb fragment offset is copied from 2944 */ 2945static inline void skb_frag_off_copy(skb_frag_t *fragto, 2946 const skb_frag_t *fragfrom) 2947{ 2948 fragto->bv_offset = fragfrom->bv_offset; 2949} 2950 2951/** 2952 * skb_frag_page - retrieve the page referred to by a paged fragment 2953 * @frag: the paged fragment 2954 * 2955 * Returns the &struct page associated with @frag. 2956 */ 2957static inline struct page *skb_frag_page(const skb_frag_t *frag) 2958{ 2959 return frag->bv_page; 2960} 2961 2962/** 2963 * __skb_frag_ref - take an addition reference on a paged fragment. 2964 * @frag: the paged fragment 2965 * 2966 * Takes an additional reference on the paged fragment @frag. 2967 */ 2968static inline void __skb_frag_ref(skb_frag_t *frag) 2969{ 2970 get_page(skb_frag_page(frag)); 2971} 2972 2973/** 2974 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2975 * @skb: the buffer 2976 * @f: the fragment offset. 2977 * 2978 * Takes an additional reference on the @f'th paged fragment of @skb. 2979 */ 2980static inline void skb_frag_ref(struct sk_buff *skb, int f) 2981{ 2982 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2983} 2984 2985/** 2986 * __skb_frag_unref - release a reference on a paged fragment. 2987 * @frag: the paged fragment 2988 * 2989 * Releases a reference on the paged fragment @frag. 2990 */ 2991static inline void __skb_frag_unref(skb_frag_t *frag) 2992{ 2993 put_page(skb_frag_page(frag)); 2994} 2995 2996/** 2997 * skb_frag_unref - release a reference on a paged fragment of an skb. 2998 * @skb: the buffer 2999 * @f: the fragment offset 3000 * 3001 * Releases a reference on the @f'th paged fragment of @skb. 3002 */ 3003static inline void skb_frag_unref(struct sk_buff *skb, int f) 3004{ 3005 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 3006} 3007 3008/** 3009 * skb_frag_address - gets the address of the data contained in a paged fragment 3010 * @frag: the paged fragment buffer 3011 * 3012 * Returns the address of the data within @frag. The page must already 3013 * be mapped. 3014 */ 3015static inline void *skb_frag_address(const skb_frag_t *frag) 3016{ 3017 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3018} 3019 3020/** 3021 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3022 * @frag: the paged fragment buffer 3023 * 3024 * Returns the address of the data within @frag. Checks that the page 3025 * is mapped and returns %NULL otherwise. 3026 */ 3027static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3028{ 3029 void *ptr = page_address(skb_frag_page(frag)); 3030 if (unlikely(!ptr)) 3031 return NULL; 3032 3033 return ptr + skb_frag_off(frag); 3034} 3035 3036/** 3037 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3038 * @fragto: skb fragment where page is set 3039 * @fragfrom: skb fragment page is copied from 3040 */ 3041static inline void skb_frag_page_copy(skb_frag_t *fragto, 3042 const skb_frag_t *fragfrom) 3043{ 3044 fragto->bv_page = fragfrom->bv_page; 3045} 3046 3047/** 3048 * __skb_frag_set_page - sets the page contained in a paged fragment 3049 * @frag: the paged fragment 3050 * @page: the page to set 3051 * 3052 * Sets the fragment @frag to contain @page. 3053 */ 3054static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 3055{ 3056 frag->bv_page = page; 3057} 3058 3059/** 3060 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 3061 * @skb: the buffer 3062 * @f: the fragment offset 3063 * @page: the page to set 3064 * 3065 * Sets the @f'th fragment of @skb to contain @page. 3066 */ 3067static inline void skb_frag_set_page(struct sk_buff *skb, int f, 3068 struct page *page) 3069{ 3070 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 3071} 3072 3073bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3074 3075/** 3076 * skb_frag_dma_map - maps a paged fragment via the DMA API 3077 * @dev: the device to map the fragment to 3078 * @frag: the paged fragment to map 3079 * @offset: the offset within the fragment (starting at the 3080 * fragment's own offset) 3081 * @size: the number of bytes to map 3082 * @dir: the direction of the mapping (``PCI_DMA_*``) 3083 * 3084 * Maps the page associated with @frag to @device. 3085 */ 3086static inline dma_addr_t skb_frag_dma_map(struct device *dev, 3087 const skb_frag_t *frag, 3088 size_t offset, size_t size, 3089 enum dma_data_direction dir) 3090{ 3091 return dma_map_page(dev, skb_frag_page(frag), 3092 skb_frag_off(frag) + offset, size, dir); 3093} 3094 3095static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3096 gfp_t gfp_mask) 3097{ 3098 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3099} 3100 3101 3102static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3103 gfp_t gfp_mask) 3104{ 3105 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3106} 3107 3108 3109/** 3110 * skb_clone_writable - is the header of a clone writable 3111 * @skb: buffer to check 3112 * @len: length up to which to write 3113 * 3114 * Returns true if modifying the header part of the cloned buffer 3115 * does not requires the data to be copied. 3116 */ 3117static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3118{ 3119 return !skb_header_cloned(skb) && 3120 skb_headroom(skb) + len <= skb->hdr_len; 3121} 3122 3123static inline int skb_try_make_writable(struct sk_buff *skb, 3124 unsigned int write_len) 3125{ 3126 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3127 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3128} 3129 3130static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3131 int cloned) 3132{ 3133 int delta = 0; 3134 3135 if (headroom > skb_headroom(skb)) 3136 delta = headroom - skb_headroom(skb); 3137 3138 if (delta || cloned) 3139 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3140 GFP_ATOMIC); 3141 return 0; 3142} 3143 3144/** 3145 * skb_cow - copy header of skb when it is required 3146 * @skb: buffer to cow 3147 * @headroom: needed headroom 3148 * 3149 * If the skb passed lacks sufficient headroom or its data part 3150 * is shared, data is reallocated. If reallocation fails, an error 3151 * is returned and original skb is not changed. 3152 * 3153 * The result is skb with writable area skb->head...skb->tail 3154 * and at least @headroom of space at head. 3155 */ 3156static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3157{ 3158 return __skb_cow(skb, headroom, skb_cloned(skb)); 3159} 3160 3161/** 3162 * skb_cow_head - skb_cow but only making the head writable 3163 * @skb: buffer to cow 3164 * @headroom: needed headroom 3165 * 3166 * This function is identical to skb_cow except that we replace the 3167 * skb_cloned check by skb_header_cloned. It should be used when 3168 * you only need to push on some header and do not need to modify 3169 * the data. 3170 */ 3171static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3172{ 3173 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3174} 3175 3176/** 3177 * skb_padto - pad an skbuff up to a minimal size 3178 * @skb: buffer to pad 3179 * @len: minimal length 3180 * 3181 * Pads up a buffer to ensure the trailing bytes exist and are 3182 * blanked. If the buffer already contains sufficient data it 3183 * is untouched. Otherwise it is extended. Returns zero on 3184 * success. The skb is freed on error. 3185 */ 3186static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3187{ 3188 unsigned int size = skb->len; 3189 if (likely(size >= len)) 3190 return 0; 3191 return skb_pad(skb, len - size); 3192} 3193 3194/** 3195 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3196 * @skb: buffer to pad 3197 * @len: minimal length 3198 * @free_on_error: free buffer on error 3199 * 3200 * Pads up a buffer to ensure the trailing bytes exist and are 3201 * blanked. If the buffer already contains sufficient data it 3202 * is untouched. Otherwise it is extended. Returns zero on 3203 * success. The skb is freed on error if @free_on_error is true. 3204 */ 3205static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len, 3206 bool free_on_error) 3207{ 3208 unsigned int size = skb->len; 3209 3210 if (unlikely(size < len)) { 3211 len -= size; 3212 if (__skb_pad(skb, len, free_on_error)) 3213 return -ENOMEM; 3214 __skb_put(skb, len); 3215 } 3216 return 0; 3217} 3218 3219/** 3220 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3221 * @skb: buffer to pad 3222 * @len: minimal length 3223 * 3224 * Pads up a buffer to ensure the trailing bytes exist and are 3225 * blanked. If the buffer already contains sufficient data it 3226 * is untouched. Otherwise it is extended. Returns zero on 3227 * success. The skb is freed on error. 3228 */ 3229static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 3230{ 3231 return __skb_put_padto(skb, len, true); 3232} 3233 3234static inline int skb_add_data(struct sk_buff *skb, 3235 struct iov_iter *from, int copy) 3236{ 3237 const int off = skb->len; 3238 3239 if (skb->ip_summed == CHECKSUM_NONE) { 3240 __wsum csum = 0; 3241 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3242 &csum, from)) { 3243 skb->csum = csum_block_add(skb->csum, csum, off); 3244 return 0; 3245 } 3246 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3247 return 0; 3248 3249 __skb_trim(skb, off); 3250 return -EFAULT; 3251} 3252 3253static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3254 const struct page *page, int off) 3255{ 3256 if (skb_zcopy(skb)) 3257 return false; 3258 if (i) { 3259 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3260 3261 return page == skb_frag_page(frag) && 3262 off == skb_frag_off(frag) + skb_frag_size(frag); 3263 } 3264 return false; 3265} 3266 3267static inline int __skb_linearize(struct sk_buff *skb) 3268{ 3269 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3270} 3271 3272/** 3273 * skb_linearize - convert paged skb to linear one 3274 * @skb: buffer to linarize 3275 * 3276 * If there is no free memory -ENOMEM is returned, otherwise zero 3277 * is returned and the old skb data released. 3278 */ 3279static inline int skb_linearize(struct sk_buff *skb) 3280{ 3281 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3282} 3283 3284/** 3285 * skb_has_shared_frag - can any frag be overwritten 3286 * @skb: buffer to test 3287 * 3288 * Return true if the skb has at least one frag that might be modified 3289 * by an external entity (as in vmsplice()/sendfile()) 3290 */ 3291static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3292{ 3293 return skb_is_nonlinear(skb) && 3294 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 3295} 3296 3297/** 3298 * skb_linearize_cow - make sure skb is linear and writable 3299 * @skb: buffer to process 3300 * 3301 * If there is no free memory -ENOMEM is returned, otherwise zero 3302 * is returned and the old skb data released. 3303 */ 3304static inline int skb_linearize_cow(struct sk_buff *skb) 3305{ 3306 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3307 __skb_linearize(skb) : 0; 3308} 3309 3310static __always_inline void 3311__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3312 unsigned int off) 3313{ 3314 if (skb->ip_summed == CHECKSUM_COMPLETE) 3315 skb->csum = csum_block_sub(skb->csum, 3316 csum_partial(start, len, 0), off); 3317 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3318 skb_checksum_start_offset(skb) < 0) 3319 skb->ip_summed = CHECKSUM_NONE; 3320} 3321 3322/** 3323 * skb_postpull_rcsum - update checksum for received skb after pull 3324 * @skb: buffer to update 3325 * @start: start of data before pull 3326 * @len: length of data pulled 3327 * 3328 * After doing a pull on a received packet, you need to call this to 3329 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3330 * CHECKSUM_NONE so that it can be recomputed from scratch. 3331 */ 3332static inline void skb_postpull_rcsum(struct sk_buff *skb, 3333 const void *start, unsigned int len) 3334{ 3335 __skb_postpull_rcsum(skb, start, len, 0); 3336} 3337 3338static __always_inline void 3339__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3340 unsigned int off) 3341{ 3342 if (skb->ip_summed == CHECKSUM_COMPLETE) 3343 skb->csum = csum_block_add(skb->csum, 3344 csum_partial(start, len, 0), off); 3345} 3346 3347/** 3348 * skb_postpush_rcsum - update checksum for received skb after push 3349 * @skb: buffer to update 3350 * @start: start of data after push 3351 * @len: length of data pushed 3352 * 3353 * After doing a push on a received packet, you need to call this to 3354 * update the CHECKSUM_COMPLETE checksum. 3355 */ 3356static inline void skb_postpush_rcsum(struct sk_buff *skb, 3357 const void *start, unsigned int len) 3358{ 3359 __skb_postpush_rcsum(skb, start, len, 0); 3360} 3361 3362void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3363 3364/** 3365 * skb_push_rcsum - push skb and update receive checksum 3366 * @skb: buffer to update 3367 * @len: length of data pulled 3368 * 3369 * This function performs an skb_push on the packet and updates 3370 * the CHECKSUM_COMPLETE checksum. It should be used on 3371 * receive path processing instead of skb_push unless you know 3372 * that the checksum difference is zero (e.g., a valid IP header) 3373 * or you are setting ip_summed to CHECKSUM_NONE. 3374 */ 3375static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3376{ 3377 skb_push(skb, len); 3378 skb_postpush_rcsum(skb, skb->data, len); 3379 return skb->data; 3380} 3381 3382int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3383/** 3384 * pskb_trim_rcsum - trim received skb and update checksum 3385 * @skb: buffer to trim 3386 * @len: new length 3387 * 3388 * This is exactly the same as pskb_trim except that it ensures the 3389 * checksum of received packets are still valid after the operation. 3390 * It can change skb pointers. 3391 */ 3392 3393static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3394{ 3395 if (likely(len >= skb->len)) 3396 return 0; 3397 return pskb_trim_rcsum_slow(skb, len); 3398} 3399 3400static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3401{ 3402 if (skb->ip_summed == CHECKSUM_COMPLETE) 3403 skb->ip_summed = CHECKSUM_NONE; 3404 __skb_trim(skb, len); 3405 return 0; 3406} 3407 3408static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3409{ 3410 if (skb->ip_summed == CHECKSUM_COMPLETE) 3411 skb->ip_summed = CHECKSUM_NONE; 3412 return __skb_grow(skb, len); 3413} 3414 3415#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3416#define skb_rb_first(root) rb_to_skb(rb_first(root)) 3417#define skb_rb_last(root) rb_to_skb(rb_last(root)) 3418#define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3419#define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3420 3421#define skb_queue_walk(queue, skb) \ 3422 for (skb = (queue)->next; \ 3423 skb != (struct sk_buff *)(queue); \ 3424 skb = skb->next) 3425 3426#define skb_queue_walk_safe(queue, skb, tmp) \ 3427 for (skb = (queue)->next, tmp = skb->next; \ 3428 skb != (struct sk_buff *)(queue); \ 3429 skb = tmp, tmp = skb->next) 3430 3431#define skb_queue_walk_from(queue, skb) \ 3432 for (; skb != (struct sk_buff *)(queue); \ 3433 skb = skb->next) 3434 3435#define skb_rbtree_walk(skb, root) \ 3436 for (skb = skb_rb_first(root); skb != NULL; \ 3437 skb = skb_rb_next(skb)) 3438 3439#define skb_rbtree_walk_from(skb) \ 3440 for (; skb != NULL; \ 3441 skb = skb_rb_next(skb)) 3442 3443#define skb_rbtree_walk_from_safe(skb, tmp) \ 3444 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3445 skb = tmp) 3446 3447#define skb_queue_walk_from_safe(queue, skb, tmp) \ 3448 for (tmp = skb->next; \ 3449 skb != (struct sk_buff *)(queue); \ 3450 skb = tmp, tmp = skb->next) 3451 3452#define skb_queue_reverse_walk(queue, skb) \ 3453 for (skb = (queue)->prev; \ 3454 skb != (struct sk_buff *)(queue); \ 3455 skb = skb->prev) 3456 3457#define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3458 for (skb = (queue)->prev, tmp = skb->prev; \ 3459 skb != (struct sk_buff *)(queue); \ 3460 skb = tmp, tmp = skb->prev) 3461 3462#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3463 for (tmp = skb->prev; \ 3464 skb != (struct sk_buff *)(queue); \ 3465 skb = tmp, tmp = skb->prev) 3466 3467static inline bool skb_has_frag_list(const struct sk_buff *skb) 3468{ 3469 return skb_shinfo(skb)->frag_list != NULL; 3470} 3471 3472static inline void skb_frag_list_init(struct sk_buff *skb) 3473{ 3474 skb_shinfo(skb)->frag_list = NULL; 3475} 3476 3477#define skb_walk_frags(skb, iter) \ 3478 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3479 3480 3481int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 3482 int *err, long *timeo_p, 3483 const struct sk_buff *skb); 3484struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3485 struct sk_buff_head *queue, 3486 unsigned int flags, 3487 void (*destructor)(struct sock *sk, 3488 struct sk_buff *skb), 3489 int *off, int *err, 3490 struct sk_buff **last); 3491struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 3492 struct sk_buff_head *queue, 3493 unsigned int flags, 3494 void (*destructor)(struct sock *sk, 3495 struct sk_buff *skb), 3496 int *off, int *err, 3497 struct sk_buff **last); 3498struct sk_buff *__skb_recv_datagram(struct sock *sk, 3499 struct sk_buff_head *sk_queue, 3500 unsigned int flags, 3501 void (*destructor)(struct sock *sk, 3502 struct sk_buff *skb), 3503 int *off, int *err); 3504struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 3505 int *err); 3506__poll_t datagram_poll(struct file *file, struct socket *sock, 3507 struct poll_table_struct *wait); 3508int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3509 struct iov_iter *to, int size); 3510static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3511 struct msghdr *msg, int size) 3512{ 3513 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3514} 3515int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3516 struct msghdr *msg); 3517int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 3518 struct iov_iter *to, int len, 3519 struct ahash_request *hash); 3520int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3521 struct iov_iter *from, int len); 3522int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3523void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3524void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3525static inline void skb_free_datagram_locked(struct sock *sk, 3526 struct sk_buff *skb) 3527{ 3528 __skb_free_datagram_locked(sk, skb, 0); 3529} 3530int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3531int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3532int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3533__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3534 int len, __wsum csum); 3535int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3536 struct pipe_inode_info *pipe, unsigned int len, 3537 unsigned int flags); 3538int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3539 int len); 3540void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3541unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3542int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3543 int len, int hlen); 3544void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3545int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3546void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3547bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu); 3548bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len); 3549struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3550struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 3551 unsigned int offset); 3552struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3553int skb_ensure_writable(struct sk_buff *skb, int write_len); 3554int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3555int skb_vlan_pop(struct sk_buff *skb); 3556int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3557int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 3558 int mac_len, bool ethernet); 3559int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 3560 bool ethernet); 3561int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 3562int skb_mpls_dec_ttl(struct sk_buff *skb); 3563struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3564 gfp_t gfp); 3565 3566static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3567{ 3568 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 3569} 3570 3571static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3572{ 3573 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3574} 3575 3576struct skb_checksum_ops { 3577 __wsum (*update)(const void *mem, int len, __wsum wsum); 3578 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3579}; 3580 3581extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 3582 3583__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3584 __wsum csum, const struct skb_checksum_ops *ops); 3585__wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3586 __wsum csum); 3587 3588static inline void * __must_check 3589__skb_header_pointer(const struct sk_buff *skb, int offset, 3590 int len, void *data, int hlen, void *buffer) 3591{ 3592 if (hlen - offset >= len) 3593 return data + offset; 3594 3595 if (!skb || 3596 skb_copy_bits(skb, offset, buffer, len) < 0) 3597 return NULL; 3598 3599 return buffer; 3600} 3601 3602static inline void * __must_check 3603skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3604{ 3605 return __skb_header_pointer(skb, offset, len, skb->data, 3606 skb_headlen(skb), buffer); 3607} 3608 3609/** 3610 * skb_needs_linearize - check if we need to linearize a given skb 3611 * depending on the given device features. 3612 * @skb: socket buffer to check 3613 * @features: net device features 3614 * 3615 * Returns true if either: 3616 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3617 * 2. skb is fragmented and the device does not support SG. 3618 */ 3619static inline bool skb_needs_linearize(struct sk_buff *skb, 3620 netdev_features_t features) 3621{ 3622 return skb_is_nonlinear(skb) && 3623 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3624 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3625} 3626 3627static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3628 void *to, 3629 const unsigned int len) 3630{ 3631 memcpy(to, skb->data, len); 3632} 3633 3634static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3635 const int offset, void *to, 3636 const unsigned int len) 3637{ 3638 memcpy(to, skb->data + offset, len); 3639} 3640 3641static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3642 const void *from, 3643 const unsigned int len) 3644{ 3645 memcpy(skb->data, from, len); 3646} 3647 3648static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3649 const int offset, 3650 const void *from, 3651 const unsigned int len) 3652{ 3653 memcpy(skb->data + offset, from, len); 3654} 3655 3656void skb_init(void); 3657 3658static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3659{ 3660 return skb->tstamp; 3661} 3662 3663/** 3664 * skb_get_timestamp - get timestamp from a skb 3665 * @skb: skb to get stamp from 3666 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 3667 * 3668 * Timestamps are stored in the skb as offsets to a base timestamp. 3669 * This function converts the offset back to a struct timeval and stores 3670 * it in stamp. 3671 */ 3672static inline void skb_get_timestamp(const struct sk_buff *skb, 3673 struct __kernel_old_timeval *stamp) 3674{ 3675 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 3676} 3677 3678static inline void skb_get_new_timestamp(const struct sk_buff *skb, 3679 struct __kernel_sock_timeval *stamp) 3680{ 3681 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 3682 3683 stamp->tv_sec = ts.tv_sec; 3684 stamp->tv_usec = ts.tv_nsec / 1000; 3685} 3686 3687static inline void skb_get_timestampns(const struct sk_buff *skb, 3688 struct __kernel_old_timespec *stamp) 3689{ 3690 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 3691 3692 stamp->tv_sec = ts.tv_sec; 3693 stamp->tv_nsec = ts.tv_nsec; 3694} 3695 3696static inline void skb_get_new_timestampns(const struct sk_buff *skb, 3697 struct __kernel_timespec *stamp) 3698{ 3699 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 3700 3701 stamp->tv_sec = ts.tv_sec; 3702 stamp->tv_nsec = ts.tv_nsec; 3703} 3704 3705static inline void __net_timestamp(struct sk_buff *skb) 3706{ 3707 skb->tstamp = ktime_get_real(); 3708} 3709 3710static inline ktime_t net_timedelta(ktime_t t) 3711{ 3712 return ktime_sub(ktime_get_real(), t); 3713} 3714 3715static inline ktime_t net_invalid_timestamp(void) 3716{ 3717 return 0; 3718} 3719 3720static inline u8 skb_metadata_len(const struct sk_buff *skb) 3721{ 3722 return skb_shinfo(skb)->meta_len; 3723} 3724 3725static inline void *skb_metadata_end(const struct sk_buff *skb) 3726{ 3727 return skb_mac_header(skb); 3728} 3729 3730static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 3731 const struct sk_buff *skb_b, 3732 u8 meta_len) 3733{ 3734 const void *a = skb_metadata_end(skb_a); 3735 const void *b = skb_metadata_end(skb_b); 3736 /* Using more efficient varaiant than plain call to memcmp(). */ 3737#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64 3738 u64 diffs = 0; 3739 3740 switch (meta_len) { 3741#define __it(x, op) (x -= sizeof(u##op)) 3742#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 3743 case 32: diffs |= __it_diff(a, b, 64); 3744 /* fall through */ 3745 case 24: diffs |= __it_diff(a, b, 64); 3746 /* fall through */ 3747 case 16: diffs |= __it_diff(a, b, 64); 3748 /* fall through */ 3749 case 8: diffs |= __it_diff(a, b, 64); 3750 break; 3751 case 28: diffs |= __it_diff(a, b, 64); 3752 /* fall through */ 3753 case 20: diffs |= __it_diff(a, b, 64); 3754 /* fall through */ 3755 case 12: diffs |= __it_diff(a, b, 64); 3756 /* fall through */ 3757 case 4: diffs |= __it_diff(a, b, 32); 3758 break; 3759 } 3760 return diffs; 3761#else 3762 return memcmp(a - meta_len, b - meta_len, meta_len); 3763#endif 3764} 3765 3766static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 3767 const struct sk_buff *skb_b) 3768{ 3769 u8 len_a = skb_metadata_len(skb_a); 3770 u8 len_b = skb_metadata_len(skb_b); 3771 3772 if (!(len_a | len_b)) 3773 return false; 3774 3775 return len_a != len_b ? 3776 true : __skb_metadata_differs(skb_a, skb_b, len_a); 3777} 3778 3779static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 3780{ 3781 skb_shinfo(skb)->meta_len = meta_len; 3782} 3783 3784static inline void skb_metadata_clear(struct sk_buff *skb) 3785{ 3786 skb_metadata_set(skb, 0); 3787} 3788 3789struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3790 3791#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3792 3793void skb_clone_tx_timestamp(struct sk_buff *skb); 3794bool skb_defer_rx_timestamp(struct sk_buff *skb); 3795 3796#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3797 3798static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3799{ 3800} 3801 3802static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3803{ 3804 return false; 3805} 3806 3807#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3808 3809/** 3810 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3811 * 3812 * PHY drivers may accept clones of transmitted packets for 3813 * timestamping via their phy_driver.txtstamp method. These drivers 3814 * must call this function to return the skb back to the stack with a 3815 * timestamp. 3816 * 3817 * @skb: clone of the the original outgoing packet 3818 * @hwtstamps: hardware time stamps 3819 * 3820 */ 3821void skb_complete_tx_timestamp(struct sk_buff *skb, 3822 struct skb_shared_hwtstamps *hwtstamps); 3823 3824void __skb_tstamp_tx(struct sk_buff *orig_skb, 3825 struct skb_shared_hwtstamps *hwtstamps, 3826 struct sock *sk, int tstype); 3827 3828/** 3829 * skb_tstamp_tx - queue clone of skb with send time stamps 3830 * @orig_skb: the original outgoing packet 3831 * @hwtstamps: hardware time stamps, may be NULL if not available 3832 * 3833 * If the skb has a socket associated, then this function clones the 3834 * skb (thus sharing the actual data and optional structures), stores 3835 * the optional hardware time stamping information (if non NULL) or 3836 * generates a software time stamp (otherwise), then queues the clone 3837 * to the error queue of the socket. Errors are silently ignored. 3838 */ 3839void skb_tstamp_tx(struct sk_buff *orig_skb, 3840 struct skb_shared_hwtstamps *hwtstamps); 3841 3842/** 3843 * skb_tx_timestamp() - Driver hook for transmit timestamping 3844 * 3845 * Ethernet MAC Drivers should call this function in their hard_xmit() 3846 * function immediately before giving the sk_buff to the MAC hardware. 3847 * 3848 * Specifically, one should make absolutely sure that this function is 3849 * called before TX completion of this packet can trigger. Otherwise 3850 * the packet could potentially already be freed. 3851 * 3852 * @skb: A socket buffer. 3853 */ 3854static inline void skb_tx_timestamp(struct sk_buff *skb) 3855{ 3856 skb_clone_tx_timestamp(skb); 3857 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 3858 skb_tstamp_tx(skb, NULL); 3859} 3860 3861/** 3862 * skb_complete_wifi_ack - deliver skb with wifi status 3863 * 3864 * @skb: the original outgoing packet 3865 * @acked: ack status 3866 * 3867 */ 3868void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3869 3870__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3871__sum16 __skb_checksum_complete(struct sk_buff *skb); 3872 3873static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3874{ 3875 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3876 skb->csum_valid || 3877 (skb->ip_summed == CHECKSUM_PARTIAL && 3878 skb_checksum_start_offset(skb) >= 0)); 3879} 3880 3881/** 3882 * skb_checksum_complete - Calculate checksum of an entire packet 3883 * @skb: packet to process 3884 * 3885 * This function calculates the checksum over the entire packet plus 3886 * the value of skb->csum. The latter can be used to supply the 3887 * checksum of a pseudo header as used by TCP/UDP. It returns the 3888 * checksum. 3889 * 3890 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3891 * this function can be used to verify that checksum on received 3892 * packets. In that case the function should return zero if the 3893 * checksum is correct. In particular, this function will return zero 3894 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3895 * hardware has already verified the correctness of the checksum. 3896 */ 3897static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3898{ 3899 return skb_csum_unnecessary(skb) ? 3900 0 : __skb_checksum_complete(skb); 3901} 3902 3903static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3904{ 3905 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3906 if (skb->csum_level == 0) 3907 skb->ip_summed = CHECKSUM_NONE; 3908 else 3909 skb->csum_level--; 3910 } 3911} 3912 3913static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3914{ 3915 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3916 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3917 skb->csum_level++; 3918 } else if (skb->ip_summed == CHECKSUM_NONE) { 3919 skb->ip_summed = CHECKSUM_UNNECESSARY; 3920 skb->csum_level = 0; 3921 } 3922} 3923 3924/* Check if we need to perform checksum complete validation. 3925 * 3926 * Returns true if checksum complete is needed, false otherwise 3927 * (either checksum is unnecessary or zero checksum is allowed). 3928 */ 3929static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3930 bool zero_okay, 3931 __sum16 check) 3932{ 3933 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3934 skb->csum_valid = 1; 3935 __skb_decr_checksum_unnecessary(skb); 3936 return false; 3937 } 3938 3939 return true; 3940} 3941 3942/* For small packets <= CHECKSUM_BREAK perform checksum complete directly 3943 * in checksum_init. 3944 */ 3945#define CHECKSUM_BREAK 76 3946 3947/* Unset checksum-complete 3948 * 3949 * Unset checksum complete can be done when packet is being modified 3950 * (uncompressed for instance) and checksum-complete value is 3951 * invalidated. 3952 */ 3953static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3954{ 3955 if (skb->ip_summed == CHECKSUM_COMPLETE) 3956 skb->ip_summed = CHECKSUM_NONE; 3957} 3958 3959/* Validate (init) checksum based on checksum complete. 3960 * 3961 * Return values: 3962 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3963 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3964 * checksum is stored in skb->csum for use in __skb_checksum_complete 3965 * non-zero: value of invalid checksum 3966 * 3967 */ 3968static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3969 bool complete, 3970 __wsum psum) 3971{ 3972 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3973 if (!csum_fold(csum_add(psum, skb->csum))) { 3974 skb->csum_valid = 1; 3975 return 0; 3976 } 3977 } 3978 3979 skb->csum = psum; 3980 3981 if (complete || skb->len <= CHECKSUM_BREAK) { 3982 __sum16 csum; 3983 3984 csum = __skb_checksum_complete(skb); 3985 skb->csum_valid = !csum; 3986 return csum; 3987 } 3988 3989 return 0; 3990} 3991 3992static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3993{ 3994 return 0; 3995} 3996 3997/* Perform checksum validate (init). Note that this is a macro since we only 3998 * want to calculate the pseudo header which is an input function if necessary. 3999 * First we try to validate without any computation (checksum unnecessary) and 4000 * then calculate based on checksum complete calling the function to compute 4001 * pseudo header. 4002 * 4003 * Return values: 4004 * 0: checksum is validated or try to in skb_checksum_complete 4005 * non-zero: value of invalid checksum 4006 */ 4007#define __skb_checksum_validate(skb, proto, complete, \ 4008 zero_okay, check, compute_pseudo) \ 4009({ \ 4010 __sum16 __ret = 0; \ 4011 skb->csum_valid = 0; \ 4012 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4013 __ret = __skb_checksum_validate_complete(skb, \ 4014 complete, compute_pseudo(skb, proto)); \ 4015 __ret; \ 4016}) 4017 4018#define skb_checksum_init(skb, proto, compute_pseudo) \ 4019 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4020 4021#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4022 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4023 4024#define skb_checksum_validate(skb, proto, compute_pseudo) \ 4025 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4026 4027#define skb_checksum_validate_zero_check(skb, proto, check, \ 4028 compute_pseudo) \ 4029 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4030 4031#define skb_checksum_simple_validate(skb) \ 4032 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4033 4034static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4035{ 4036 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4037} 4038 4039static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4040{ 4041 skb->csum = ~pseudo; 4042 skb->ip_summed = CHECKSUM_COMPLETE; 4043} 4044 4045#define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4046do { \ 4047 if (__skb_checksum_convert_check(skb)) \ 4048 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4049} while (0) 4050 4051static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4052 u16 start, u16 offset) 4053{ 4054 skb->ip_summed = CHECKSUM_PARTIAL; 4055 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4056 skb->csum_offset = offset - start; 4057} 4058 4059/* Update skbuf and packet to reflect the remote checksum offload operation. 4060 * When called, ptr indicates the starting point for skb->csum when 4061 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4062 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4063 */ 4064static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4065 int start, int offset, bool nopartial) 4066{ 4067 __wsum delta; 4068 4069 if (!nopartial) { 4070 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4071 return; 4072 } 4073 4074 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4075 __skb_checksum_complete(skb); 4076 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4077 } 4078 4079 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4080 4081 /* Adjust skb->csum since we changed the packet */ 4082 skb->csum = csum_add(skb->csum, delta); 4083} 4084 4085static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4086{ 4087#if IS_ENABLED(CONFIG_NF_CONNTRACK) 4088 return (void *)(skb->_nfct & NFCT_PTRMASK); 4089#else 4090 return NULL; 4091#endif 4092} 4093 4094static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4095{ 4096#if IS_ENABLED(CONFIG_NF_CONNTRACK) 4097 return skb->_nfct; 4098#else 4099 return 0UL; 4100#endif 4101} 4102 4103static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4104{ 4105#if IS_ENABLED(CONFIG_NF_CONNTRACK) 4106 skb->_nfct = nfct; 4107#endif 4108} 4109 4110#ifdef CONFIG_SKB_EXTENSIONS 4111enum skb_ext_id { 4112#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4113 SKB_EXT_BRIDGE_NF, 4114#endif 4115#ifdef CONFIG_XFRM 4116 SKB_EXT_SEC_PATH, 4117#endif 4118#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4119 TC_SKB_EXT, 4120#endif 4121#if IS_ENABLED(CONFIG_MPTCP) 4122 SKB_EXT_MPTCP, 4123#endif 4124 SKB_EXT_NUM, /* must be last */ 4125}; 4126 4127/** 4128 * struct skb_ext - sk_buff extensions 4129 * @refcnt: 1 on allocation, deallocated on 0 4130 * @offset: offset to add to @data to obtain extension address 4131 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4132 * @data: start of extension data, variable sized 4133 * 4134 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4135 * to use 'u8' types while allowing up to 2kb worth of extension data. 4136 */ 4137struct skb_ext { 4138 refcount_t refcnt; 4139 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4140 u8 chunks; /* same */ 4141 char data[0] __aligned(8); 4142}; 4143 4144struct skb_ext *__skb_ext_alloc(void); 4145void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4146 struct skb_ext *ext); 4147void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4148void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4149void __skb_ext_put(struct skb_ext *ext); 4150 4151static inline void skb_ext_put(struct sk_buff *skb) 4152{ 4153 if (skb->active_extensions) 4154 __skb_ext_put(skb->extensions); 4155} 4156 4157static inline void __skb_ext_copy(struct sk_buff *dst, 4158 const struct sk_buff *src) 4159{ 4160 dst->active_extensions = src->active_extensions; 4161 4162 if (src->active_extensions) { 4163 struct skb_ext *ext = src->extensions; 4164 4165 refcount_inc(&ext->refcnt); 4166 dst->extensions = ext; 4167 } 4168} 4169 4170static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4171{ 4172 skb_ext_put(dst); 4173 __skb_ext_copy(dst, src); 4174} 4175 4176static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4177{ 4178 return !!ext->offset[i]; 4179} 4180 4181static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4182{ 4183 return skb->active_extensions & (1 << id); 4184} 4185 4186static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4187{ 4188 if (skb_ext_exist(skb, id)) 4189 __skb_ext_del(skb, id); 4190} 4191 4192static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4193{ 4194 if (skb_ext_exist(skb, id)) { 4195 struct skb_ext *ext = skb->extensions; 4196 4197 return (void *)ext + (ext->offset[id] << 3); 4198 } 4199 4200 return NULL; 4201} 4202 4203static inline void skb_ext_reset(struct sk_buff *skb) 4204{ 4205 if (unlikely(skb->active_extensions)) { 4206 __skb_ext_put(skb->extensions); 4207 skb->active_extensions = 0; 4208 } 4209} 4210 4211static inline bool skb_has_extensions(struct sk_buff *skb) 4212{ 4213 return unlikely(skb->active_extensions); 4214} 4215#else 4216static inline void skb_ext_put(struct sk_buff *skb) {} 4217static inline void skb_ext_reset(struct sk_buff *skb) {} 4218static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4219static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4220static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4221static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4222#endif /* CONFIG_SKB_EXTENSIONS */ 4223 4224static inline void nf_reset_ct(struct sk_buff *skb) 4225{ 4226#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4227 nf_conntrack_put(skb_nfct(skb)); 4228 skb->_nfct = 0; 4229#endif 4230} 4231 4232static inline void nf_reset_trace(struct sk_buff *skb) 4233{ 4234#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4235 skb->nf_trace = 0; 4236#endif 4237} 4238 4239static inline void ipvs_reset(struct sk_buff *skb) 4240{ 4241#if IS_ENABLED(CONFIG_IP_VS) 4242 skb->ipvs_property = 0; 4243#endif 4244} 4245 4246/* Note: This doesn't put any conntrack info in dst. */ 4247static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4248 bool copy) 4249{ 4250#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4251 dst->_nfct = src->_nfct; 4252 nf_conntrack_get(skb_nfct(src)); 4253#endif 4254#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4255 if (copy) 4256 dst->nf_trace = src->nf_trace; 4257#endif 4258} 4259 4260static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4261{ 4262#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4263 nf_conntrack_put(skb_nfct(dst)); 4264#endif 4265 __nf_copy(dst, src, true); 4266} 4267 4268#ifdef CONFIG_NETWORK_SECMARK 4269static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4270{ 4271 to->secmark = from->secmark; 4272} 4273 4274static inline void skb_init_secmark(struct sk_buff *skb) 4275{ 4276 skb->secmark = 0; 4277} 4278#else 4279static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4280{ } 4281 4282static inline void skb_init_secmark(struct sk_buff *skb) 4283{ } 4284#endif 4285 4286static inline int secpath_exists(const struct sk_buff *skb) 4287{ 4288#ifdef CONFIG_XFRM 4289 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4290#else 4291 return 0; 4292#endif 4293} 4294 4295static inline bool skb_irq_freeable(const struct sk_buff *skb) 4296{ 4297 return !skb->destructor && 4298 !secpath_exists(skb) && 4299 !skb_nfct(skb) && 4300 !skb->_skb_refdst && 4301 !skb_has_frag_list(skb); 4302} 4303 4304static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 4305{ 4306 skb->queue_mapping = queue_mapping; 4307} 4308 4309static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 4310{ 4311 return skb->queue_mapping; 4312} 4313 4314static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 4315{ 4316 to->queue_mapping = from->queue_mapping; 4317} 4318 4319static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 4320{ 4321 skb->queue_mapping = rx_queue + 1; 4322} 4323 4324static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 4325{ 4326 return skb->queue_mapping - 1; 4327} 4328 4329static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 4330{ 4331 return skb->queue_mapping != 0; 4332} 4333 4334static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 4335{ 4336 skb->dst_pending_confirm = val; 4337} 4338 4339static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 4340{ 4341 return skb->dst_pending_confirm != 0; 4342} 4343 4344static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 4345{ 4346#ifdef CONFIG_XFRM 4347 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 4348#else 4349 return NULL; 4350#endif 4351} 4352 4353/* Keeps track of mac header offset relative to skb->head. 4354 * It is useful for TSO of Tunneling protocol. e.g. GRE. 4355 * For non-tunnel skb it points to skb_mac_header() and for 4356 * tunnel skb it points to outer mac header. 4357 * Keeps track of level of encapsulation of network headers. 4358 */ 4359struct skb_gso_cb { 4360 union { 4361 int mac_offset; 4362 int data_offset; 4363 }; 4364 int encap_level; 4365 __wsum csum; 4366 __u16 csum_start; 4367}; 4368#define SKB_SGO_CB_OFFSET 32 4369#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 4370 4371static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 4372{ 4373 return (skb_mac_header(inner_skb) - inner_skb->head) - 4374 SKB_GSO_CB(inner_skb)->mac_offset; 4375} 4376 4377static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 4378{ 4379 int new_headroom, headroom; 4380 int ret; 4381 4382 headroom = skb_headroom(skb); 4383 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 4384 if (ret) 4385 return ret; 4386 4387 new_headroom = skb_headroom(skb); 4388 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 4389 return 0; 4390} 4391 4392static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 4393{ 4394 /* Do not update partial checksums if remote checksum is enabled. */ 4395 if (skb->remcsum_offload) 4396 return; 4397 4398 SKB_GSO_CB(skb)->csum = res; 4399 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 4400} 4401 4402/* Compute the checksum for a gso segment. First compute the checksum value 4403 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 4404 * then add in skb->csum (checksum from csum_start to end of packet). 4405 * skb->csum and csum_start are then updated to reflect the checksum of the 4406 * resultant packet starting from the transport header-- the resultant checksum 4407 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 4408 * header. 4409 */ 4410static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 4411{ 4412 unsigned char *csum_start = skb_transport_header(skb); 4413 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 4414 __wsum partial = SKB_GSO_CB(skb)->csum; 4415 4416 SKB_GSO_CB(skb)->csum = res; 4417 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 4418 4419 return csum_fold(csum_partial(csum_start, plen, partial)); 4420} 4421 4422static inline bool skb_is_gso(const struct sk_buff *skb) 4423{ 4424 return skb_shinfo(skb)->gso_size; 4425} 4426 4427/* Note: Should be called only if skb_is_gso(skb) is true */ 4428static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4429{ 4430 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4431} 4432 4433/* Note: Should be called only if skb_is_gso(skb) is true */ 4434static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4435{ 4436 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4437} 4438 4439/* Note: Should be called only if skb_is_gso(skb) is true */ 4440static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 4441{ 4442 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 4443} 4444 4445static inline void skb_gso_reset(struct sk_buff *skb) 4446{ 4447 skb_shinfo(skb)->gso_size = 0; 4448 skb_shinfo(skb)->gso_segs = 0; 4449 skb_shinfo(skb)->gso_type = 0; 4450} 4451 4452static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4453 u16 increment) 4454{ 4455 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4456 return; 4457 shinfo->gso_size += increment; 4458} 4459 4460static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4461 u16 decrement) 4462{ 4463 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4464 return; 4465 shinfo->gso_size -= decrement; 4466} 4467 4468void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4469 4470static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4471{ 4472 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4473 * wanted then gso_type will be set. */ 4474 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4475 4476 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4477 unlikely(shinfo->gso_type == 0)) { 4478 __skb_warn_lro_forwarding(skb); 4479 return true; 4480 } 4481 return false; 4482} 4483 4484static inline void skb_forward_csum(struct sk_buff *skb) 4485{ 4486 /* Unfortunately we don't support this one. Any brave souls? */ 4487 if (skb->ip_summed == CHECKSUM_COMPLETE) 4488 skb->ip_summed = CHECKSUM_NONE; 4489} 4490 4491/** 4492 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4493 * @skb: skb to check 4494 * 4495 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4496 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4497 * use this helper, to document places where we make this assertion. 4498 */ 4499static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4500{ 4501#ifdef DEBUG 4502 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 4503#endif 4504} 4505 4506bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4507 4508int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4509struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4510 unsigned int transport_len, 4511 __sum16(*skb_chkf)(struct sk_buff *skb)); 4512 4513/** 4514 * skb_head_is_locked - Determine if the skb->head is locked down 4515 * @skb: skb to check 4516 * 4517 * The head on skbs build around a head frag can be removed if they are 4518 * not cloned. This function returns true if the skb head is locked down 4519 * due to either being allocated via kmalloc, or by being a clone with 4520 * multiple references to the head. 4521 */ 4522static inline bool skb_head_is_locked(const struct sk_buff *skb) 4523{ 4524 return !skb->head_frag || skb_cloned(skb); 4525} 4526 4527/* Local Checksum Offload. 4528 * Compute outer checksum based on the assumption that the 4529 * inner checksum will be offloaded later. 4530 * See Documentation/networking/checksum-offloads.rst for 4531 * explanation of how this works. 4532 * Fill in outer checksum adjustment (e.g. with sum of outer 4533 * pseudo-header) before calling. 4534 * Also ensure that inner checksum is in linear data area. 4535 */ 4536static inline __wsum lco_csum(struct sk_buff *skb) 4537{ 4538 unsigned char *csum_start = skb_checksum_start(skb); 4539 unsigned char *l4_hdr = skb_transport_header(skb); 4540 __wsum partial; 4541 4542 /* Start with complement of inner checksum adjustment */ 4543 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 4544 skb->csum_offset)); 4545 4546 /* Add in checksum of our headers (incl. outer checksum 4547 * adjustment filled in by caller) and return result. 4548 */ 4549 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 4550} 4551 4552#endif /* __KERNEL__ */ 4553#endif /* _LINUX_SKBUFF_H */