at v4.14 119 kB view raw
1/* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <gw4pts@gw4pts.ampr.org> 6 * Florian La Roche, <rzsfl@rz.uni-sb.de> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14#ifndef _LINUX_SKBUFF_H 15#define _LINUX_SKBUFF_H 16 17#include <linux/kernel.h> 18#include <linux/kmemcheck.h> 19#include <linux/compiler.h> 20#include <linux/time.h> 21#include <linux/bug.h> 22#include <linux/cache.h> 23#include <linux/rbtree.h> 24#include <linux/socket.h> 25#include <linux/refcount.h> 26 27#include <linux/atomic.h> 28#include <asm/types.h> 29#include <linux/spinlock.h> 30#include <linux/net.h> 31#include <linux/textsearch.h> 32#include <net/checksum.h> 33#include <linux/rcupdate.h> 34#include <linux/hrtimer.h> 35#include <linux/dma-mapping.h> 36#include <linux/netdev_features.h> 37#include <linux/sched.h> 38#include <linux/sched/clock.h> 39#include <net/flow_dissector.h> 40#include <linux/splice.h> 41#include <linux/in6.h> 42#include <linux/if_packet.h> 43#include <net/flow.h> 44 45/* The interface for checksum offload between the stack and networking drivers 46 * is as follows... 47 * 48 * A. IP checksum related features 49 * 50 * Drivers advertise checksum offload capabilities in the features of a device. 51 * From the stack's point of view these are capabilities offered by the driver, 52 * a driver typically only advertises features that it is capable of offloading 53 * to its device. 54 * 55 * The checksum related features are: 56 * 57 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one 58 * IP (one's complement) checksum for any combination 59 * of protocols or protocol layering. The checksum is 60 * computed and set in a packet per the CHECKSUM_PARTIAL 61 * interface (see below). 62 * 63 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain 64 * TCP or UDP packets over IPv4. These are specifically 65 * unencapsulated packets of the form IPv4|TCP or 66 * IPv4|UDP where the Protocol field in the IPv4 header 67 * is TCP or UDP. The IPv4 header may contain IP options 68 * This feature cannot be set in features for a device 69 * with NETIF_F_HW_CSUM also set. This feature is being 70 * DEPRECATED (see below). 71 * 72 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain 73 * TCP or UDP packets over IPv6. These are specifically 74 * unencapsulated packets of the form IPv6|TCP or 75 * IPv4|UDP where the Next Header field in the IPv6 76 * header is either TCP or UDP. IPv6 extension headers 77 * are not supported with this feature. This feature 78 * cannot be set in features for a device with 79 * NETIF_F_HW_CSUM also set. This feature is being 80 * DEPRECATED (see below). 81 * 82 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload. 83 * This flag is used only used to disable the RX checksum 84 * feature for a device. The stack will accept receive 85 * checksum indication in packets received on a device 86 * regardless of whether NETIF_F_RXCSUM is set. 87 * 88 * B. Checksumming of received packets by device. Indication of checksum 89 * verification is in set skb->ip_summed. Possible values are: 90 * 91 * CHECKSUM_NONE: 92 * 93 * Device did not checksum this packet e.g. due to lack of capabilities. 94 * The packet contains full (though not verified) checksum in packet but 95 * not in skb->csum. Thus, skb->csum is undefined in this case. 96 * 97 * CHECKSUM_UNNECESSARY: 98 * 99 * The hardware you're dealing with doesn't calculate the full checksum 100 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 101 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 102 * if their checksums are okay. skb->csum is still undefined in this case 103 * though. A driver or device must never modify the checksum field in the 104 * packet even if checksum is verified. 105 * 106 * CHECKSUM_UNNECESSARY is applicable to following protocols: 107 * TCP: IPv6 and IPv4. 108 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 109 * zero UDP checksum for either IPv4 or IPv6, the networking stack 110 * may perform further validation in this case. 111 * GRE: only if the checksum is present in the header. 112 * SCTP: indicates the CRC in SCTP header has been validated. 113 * FCOE: indicates the CRC in FC frame has been validated. 114 * 115 * skb->csum_level indicates the number of consecutive checksums found in 116 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 117 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 118 * and a device is able to verify the checksums for UDP (possibly zero), 119 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 120 * two. If the device were only able to verify the UDP checksum and not 121 * GRE, either because it doesn't support GRE checksum of because GRE 122 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 123 * not considered in this case). 124 * 125 * CHECKSUM_COMPLETE: 126 * 127 * This is the most generic way. The device supplied checksum of the _whole_ 128 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 129 * hardware doesn't need to parse L3/L4 headers to implement this. 130 * 131 * Notes: 132 * - Even if device supports only some protocols, but is able to produce 133 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 134 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 135 * 136 * CHECKSUM_PARTIAL: 137 * 138 * A checksum is set up to be offloaded to a device as described in the 139 * output description for CHECKSUM_PARTIAL. This may occur on a packet 140 * received directly from another Linux OS, e.g., a virtualized Linux kernel 141 * on the same host, or it may be set in the input path in GRO or remote 142 * checksum offload. For the purposes of checksum verification, the checksum 143 * referred to by skb->csum_start + skb->csum_offset and any preceding 144 * checksums in the packet are considered verified. Any checksums in the 145 * packet that are after the checksum being offloaded are not considered to 146 * be verified. 147 * 148 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload 149 * in the skb->ip_summed for a packet. Values are: 150 * 151 * CHECKSUM_PARTIAL: 152 * 153 * The driver is required to checksum the packet as seen by hard_start_xmit() 154 * from skb->csum_start up to the end, and to record/write the checksum at 155 * offset skb->csum_start + skb->csum_offset. A driver may verify that the 156 * csum_start and csum_offset values are valid values given the length and 157 * offset of the packet, however they should not attempt to validate that the 158 * checksum refers to a legitimate transport layer checksum-- it is the 159 * purview of the stack to validate that csum_start and csum_offset are set 160 * correctly. 161 * 162 * When the stack requests checksum offload for a packet, the driver MUST 163 * ensure that the checksum is set correctly. A driver can either offload the 164 * checksum calculation to the device, or call skb_checksum_help (in the case 165 * that the device does not support offload for a particular checksum). 166 * 167 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of 168 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate 169 * checksum offload capability. 170 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based 171 * on network device checksumming capabilities: if a packet does not match 172 * them, skb_checksum_help or skb_crc32c_help (depending on the value of 173 * csum_not_inet, see item D.) is called to resolve the checksum. 174 * 175 * CHECKSUM_NONE: 176 * 177 * The skb was already checksummed by the protocol, or a checksum is not 178 * required. 179 * 180 * CHECKSUM_UNNECESSARY: 181 * 182 * This has the same meaning on as CHECKSUM_NONE for checksum offload on 183 * output. 184 * 185 * CHECKSUM_COMPLETE: 186 * Not used in checksum output. If a driver observes a packet with this value 187 * set in skbuff, if should treat as CHECKSUM_NONE being set. 188 * 189 * D. Non-IP checksum (CRC) offloads 190 * 191 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of 192 * offloading the SCTP CRC in a packet. To perform this offload the stack 193 * will set set csum_start and csum_offset accordingly, set ip_summed to 194 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in 195 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c. 196 * A driver that supports both IP checksum offload and SCTP CRC32c offload 197 * must verify which offload is configured for a packet by testing the 198 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve 199 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 200 * 201 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of 202 * offloading the FCOE CRC in a packet. To perform this offload the stack 203 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 204 * accordingly. Note the there is no indication in the skbuff that the 205 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports 206 * both IP checksum offload and FCOE CRC offload must verify which offload 207 * is configured for a packet presumably by inspecting packet headers. 208 * 209 * E. Checksumming on output with GSO. 210 * 211 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload 212 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 213 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as 214 * part of the GSO operation is implied. If a checksum is being offloaded 215 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset 216 * are set to refer to the outermost checksum being offload (two offloaded 217 * checksums are possible with UDP encapsulation). 218 */ 219 220/* Don't change this without changing skb_csum_unnecessary! */ 221#define CHECKSUM_NONE 0 222#define CHECKSUM_UNNECESSARY 1 223#define CHECKSUM_COMPLETE 2 224#define CHECKSUM_PARTIAL 3 225 226/* Maximum value in skb->csum_level */ 227#define SKB_MAX_CSUM_LEVEL 3 228 229#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 230#define SKB_WITH_OVERHEAD(X) \ 231 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 232#define SKB_MAX_ORDER(X, ORDER) \ 233 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 234#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 235#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 236 237/* return minimum truesize of one skb containing X bytes of data */ 238#define SKB_TRUESIZE(X) ((X) + \ 239 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 240 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 241 242struct net_device; 243struct scatterlist; 244struct pipe_inode_info; 245struct iov_iter; 246struct napi_struct; 247 248#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 249struct nf_conntrack { 250 atomic_t use; 251}; 252#endif 253 254#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 255struct nf_bridge_info { 256 refcount_t use; 257 enum { 258 BRNF_PROTO_UNCHANGED, 259 BRNF_PROTO_8021Q, 260 BRNF_PROTO_PPPOE 261 } orig_proto:8; 262 u8 pkt_otherhost:1; 263 u8 in_prerouting:1; 264 u8 bridged_dnat:1; 265 __u16 frag_max_size; 266 struct net_device *physindev; 267 268 /* always valid & non-NULL from FORWARD on, for physdev match */ 269 struct net_device *physoutdev; 270 union { 271 /* prerouting: detect dnat in orig/reply direction */ 272 __be32 ipv4_daddr; 273 struct in6_addr ipv6_daddr; 274 275 /* after prerouting + nat detected: store original source 276 * mac since neigh resolution overwrites it, only used while 277 * skb is out in neigh layer. 278 */ 279 char neigh_header[8]; 280 }; 281}; 282#endif 283 284struct sk_buff_head { 285 /* These two members must be first. */ 286 struct sk_buff *next; 287 struct sk_buff *prev; 288 289 __u32 qlen; 290 spinlock_t lock; 291}; 292 293struct sk_buff; 294 295/* To allow 64K frame to be packed as single skb without frag_list we 296 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 297 * buffers which do not start on a page boundary. 298 * 299 * Since GRO uses frags we allocate at least 16 regardless of page 300 * size. 301 */ 302#if (65536/PAGE_SIZE + 1) < 16 303#define MAX_SKB_FRAGS 16UL 304#else 305#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 306#endif 307extern int sysctl_max_skb_frags; 308 309/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 310 * segment using its current segmentation instead. 311 */ 312#define GSO_BY_FRAGS 0xFFFF 313 314typedef struct skb_frag_struct skb_frag_t; 315 316struct skb_frag_struct { 317 struct { 318 struct page *p; 319 } page; 320#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 321 __u32 page_offset; 322 __u32 size; 323#else 324 __u16 page_offset; 325 __u16 size; 326#endif 327}; 328 329static inline unsigned int skb_frag_size(const skb_frag_t *frag) 330{ 331 return frag->size; 332} 333 334static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 335{ 336 frag->size = size; 337} 338 339static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 340{ 341 frag->size += delta; 342} 343 344static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 345{ 346 frag->size -= delta; 347} 348 349static inline bool skb_frag_must_loop(struct page *p) 350{ 351#if defined(CONFIG_HIGHMEM) 352 if (PageHighMem(p)) 353 return true; 354#endif 355 return false; 356} 357 358/** 359 * skb_frag_foreach_page - loop over pages in a fragment 360 * 361 * @f: skb frag to operate on 362 * @f_off: offset from start of f->page.p 363 * @f_len: length from f_off to loop over 364 * @p: (temp var) current page 365 * @p_off: (temp var) offset from start of current page, 366 * non-zero only on first page. 367 * @p_len: (temp var) length in current page, 368 * < PAGE_SIZE only on first and last page. 369 * @copied: (temp var) length so far, excluding current p_len. 370 * 371 * A fragment can hold a compound page, in which case per-page 372 * operations, notably kmap_atomic, must be called for each 373 * regular page. 374 */ 375#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 376 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 377 p_off = (f_off) & (PAGE_SIZE - 1), \ 378 p_len = skb_frag_must_loop(p) ? \ 379 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 380 copied = 0; \ 381 copied < f_len; \ 382 copied += p_len, p++, p_off = 0, \ 383 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 384 385#define HAVE_HW_TIME_STAMP 386 387/** 388 * struct skb_shared_hwtstamps - hardware time stamps 389 * @hwtstamp: hardware time stamp transformed into duration 390 * since arbitrary point in time 391 * 392 * Software time stamps generated by ktime_get_real() are stored in 393 * skb->tstamp. 394 * 395 * hwtstamps can only be compared against other hwtstamps from 396 * the same device. 397 * 398 * This structure is attached to packets as part of the 399 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 400 */ 401struct skb_shared_hwtstamps { 402 ktime_t hwtstamp; 403}; 404 405/* Definitions for tx_flags in struct skb_shared_info */ 406enum { 407 /* generate hardware time stamp */ 408 SKBTX_HW_TSTAMP = 1 << 0, 409 410 /* generate software time stamp when queueing packet to NIC */ 411 SKBTX_SW_TSTAMP = 1 << 1, 412 413 /* device driver is going to provide hardware time stamp */ 414 SKBTX_IN_PROGRESS = 1 << 2, 415 416 /* device driver supports TX zero-copy buffers */ 417 SKBTX_DEV_ZEROCOPY = 1 << 3, 418 419 /* generate wifi status information (where possible) */ 420 SKBTX_WIFI_STATUS = 1 << 4, 421 422 /* This indicates at least one fragment might be overwritten 423 * (as in vmsplice(), sendfile() ...) 424 * If we need to compute a TX checksum, we'll need to copy 425 * all frags to avoid possible bad checksum 426 */ 427 SKBTX_SHARED_FRAG = 1 << 5, 428 429 /* generate software time stamp when entering packet scheduling */ 430 SKBTX_SCHED_TSTAMP = 1 << 6, 431}; 432 433#define SKBTX_ZEROCOPY_FRAG (SKBTX_DEV_ZEROCOPY | SKBTX_SHARED_FRAG) 434#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 435 SKBTX_SCHED_TSTAMP) 436#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 437 438/* 439 * The callback notifies userspace to release buffers when skb DMA is done in 440 * lower device, the skb last reference should be 0 when calling this. 441 * The zerocopy_success argument is true if zero copy transmit occurred, 442 * false on data copy or out of memory error caused by data copy attempt. 443 * The ctx field is used to track device context. 444 * The desc field is used to track userspace buffer index. 445 */ 446struct ubuf_info { 447 void (*callback)(struct ubuf_info *, bool zerocopy_success); 448 union { 449 struct { 450 unsigned long desc; 451 void *ctx; 452 }; 453 struct { 454 u32 id; 455 u16 len; 456 u16 zerocopy:1; 457 u32 bytelen; 458 }; 459 }; 460 refcount_t refcnt; 461 462 struct mmpin { 463 struct user_struct *user; 464 unsigned int num_pg; 465 } mmp; 466}; 467 468#define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 469 470struct ubuf_info *sock_zerocopy_alloc(struct sock *sk, size_t size); 471struct ubuf_info *sock_zerocopy_realloc(struct sock *sk, size_t size, 472 struct ubuf_info *uarg); 473 474static inline void sock_zerocopy_get(struct ubuf_info *uarg) 475{ 476 refcount_inc(&uarg->refcnt); 477} 478 479void sock_zerocopy_put(struct ubuf_info *uarg); 480void sock_zerocopy_put_abort(struct ubuf_info *uarg); 481 482void sock_zerocopy_callback(struct ubuf_info *uarg, bool success); 483 484int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 485 struct msghdr *msg, int len, 486 struct ubuf_info *uarg); 487 488/* This data is invariant across clones and lives at 489 * the end of the header data, ie. at skb->end. 490 */ 491struct skb_shared_info { 492 unsigned short _unused; 493 unsigned char nr_frags; 494 __u8 tx_flags; 495 unsigned short gso_size; 496 /* Warning: this field is not always filled in (UFO)! */ 497 unsigned short gso_segs; 498 struct sk_buff *frag_list; 499 struct skb_shared_hwtstamps hwtstamps; 500 unsigned int gso_type; 501 u32 tskey; 502 __be32 ip6_frag_id; 503 504 /* 505 * Warning : all fields before dataref are cleared in __alloc_skb() 506 */ 507 atomic_t dataref; 508 509 /* Intermediate layers must ensure that destructor_arg 510 * remains valid until skb destructor */ 511 void * destructor_arg; 512 513 /* must be last field, see pskb_expand_head() */ 514 skb_frag_t frags[MAX_SKB_FRAGS]; 515}; 516 517/* We divide dataref into two halves. The higher 16 bits hold references 518 * to the payload part of skb->data. The lower 16 bits hold references to 519 * the entire skb->data. A clone of a headerless skb holds the length of 520 * the header in skb->hdr_len. 521 * 522 * All users must obey the rule that the skb->data reference count must be 523 * greater than or equal to the payload reference count. 524 * 525 * Holding a reference to the payload part means that the user does not 526 * care about modifications to the header part of skb->data. 527 */ 528#define SKB_DATAREF_SHIFT 16 529#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 530 531 532enum { 533 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 534 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 535 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 536}; 537 538enum { 539 SKB_GSO_TCPV4 = 1 << 0, 540 541 /* This indicates the skb is from an untrusted source. */ 542 SKB_GSO_DODGY = 1 << 1, 543 544 /* This indicates the tcp segment has CWR set. */ 545 SKB_GSO_TCP_ECN = 1 << 2, 546 547 SKB_GSO_TCP_FIXEDID = 1 << 3, 548 549 SKB_GSO_TCPV6 = 1 << 4, 550 551 SKB_GSO_FCOE = 1 << 5, 552 553 SKB_GSO_GRE = 1 << 6, 554 555 SKB_GSO_GRE_CSUM = 1 << 7, 556 557 SKB_GSO_IPXIP4 = 1 << 8, 558 559 SKB_GSO_IPXIP6 = 1 << 9, 560 561 SKB_GSO_UDP_TUNNEL = 1 << 10, 562 563 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 564 565 SKB_GSO_PARTIAL = 1 << 12, 566 567 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 568 569 SKB_GSO_SCTP = 1 << 14, 570 571 SKB_GSO_ESP = 1 << 15, 572}; 573 574#if BITS_PER_LONG > 32 575#define NET_SKBUFF_DATA_USES_OFFSET 1 576#endif 577 578#ifdef NET_SKBUFF_DATA_USES_OFFSET 579typedef unsigned int sk_buff_data_t; 580#else 581typedef unsigned char *sk_buff_data_t; 582#endif 583 584/** 585 * struct sk_buff - socket buffer 586 * @next: Next buffer in list 587 * @prev: Previous buffer in list 588 * @tstamp: Time we arrived/left 589 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 590 * @sk: Socket we are owned by 591 * @dev: Device we arrived on/are leaving by 592 * @cb: Control buffer. Free for use by every layer. Put private vars here 593 * @_skb_refdst: destination entry (with norefcount bit) 594 * @sp: the security path, used for xfrm 595 * @len: Length of actual data 596 * @data_len: Data length 597 * @mac_len: Length of link layer header 598 * @hdr_len: writable header length of cloned skb 599 * @csum: Checksum (must include start/offset pair) 600 * @csum_start: Offset from skb->head where checksumming should start 601 * @csum_offset: Offset from csum_start where checksum should be stored 602 * @priority: Packet queueing priority 603 * @ignore_df: allow local fragmentation 604 * @cloned: Head may be cloned (check refcnt to be sure) 605 * @ip_summed: Driver fed us an IP checksum 606 * @nohdr: Payload reference only, must not modify header 607 * @pkt_type: Packet class 608 * @fclone: skbuff clone status 609 * @ipvs_property: skbuff is owned by ipvs 610 * @tc_skip_classify: do not classify packet. set by IFB device 611 * @tc_at_ingress: used within tc_classify to distinguish in/egress 612 * @tc_redirected: packet was redirected by a tc action 613 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect 614 * @peeked: this packet has been seen already, so stats have been 615 * done for it, don't do them again 616 * @nf_trace: netfilter packet trace flag 617 * @protocol: Packet protocol from driver 618 * @destructor: Destruct function 619 * @_nfct: Associated connection, if any (with nfctinfo bits) 620 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 621 * @skb_iif: ifindex of device we arrived on 622 * @tc_index: Traffic control index 623 * @hash: the packet hash 624 * @queue_mapping: Queue mapping for multiqueue devices 625 * @xmit_more: More SKBs are pending for this queue 626 * @ndisc_nodetype: router type (from link layer) 627 * @ooo_okay: allow the mapping of a socket to a queue to be changed 628 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 629 * ports. 630 * @sw_hash: indicates hash was computed in software stack 631 * @wifi_acked_valid: wifi_acked was set 632 * @wifi_acked: whether frame was acked on wifi or not 633 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 634 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 635 * @dst_pending_confirm: need to confirm neighbour 636 * @napi_id: id of the NAPI struct this skb came from 637 * @secmark: security marking 638 * @mark: Generic packet mark 639 * @vlan_proto: vlan encapsulation protocol 640 * @vlan_tci: vlan tag control information 641 * @inner_protocol: Protocol (encapsulation) 642 * @inner_transport_header: Inner transport layer header (encapsulation) 643 * @inner_network_header: Network layer header (encapsulation) 644 * @inner_mac_header: Link layer header (encapsulation) 645 * @transport_header: Transport layer header 646 * @network_header: Network layer header 647 * @mac_header: Link layer header 648 * @tail: Tail pointer 649 * @end: End pointer 650 * @head: Head of buffer 651 * @data: Data head pointer 652 * @truesize: Buffer size 653 * @users: User count - see {datagram,tcp}.c 654 */ 655 656struct sk_buff { 657 union { 658 struct { 659 /* These two members must be first. */ 660 struct sk_buff *next; 661 struct sk_buff *prev; 662 663 union { 664 ktime_t tstamp; 665 u64 skb_mstamp; 666 }; 667 }; 668 struct rb_node rbnode; /* used in netem & tcp stack */ 669 }; 670 struct sock *sk; 671 672 union { 673 struct net_device *dev; 674 /* Some protocols might use this space to store information, 675 * while device pointer would be NULL. 676 * UDP receive path is one user. 677 */ 678 unsigned long dev_scratch; 679 }; 680 /* 681 * This is the control buffer. It is free to use for every 682 * layer. Please put your private variables there. If you 683 * want to keep them across layers you have to do a skb_clone() 684 * first. This is owned by whoever has the skb queued ATM. 685 */ 686 char cb[48] __aligned(8); 687 688 unsigned long _skb_refdst; 689 void (*destructor)(struct sk_buff *skb); 690#ifdef CONFIG_XFRM 691 struct sec_path *sp; 692#endif 693#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 694 unsigned long _nfct; 695#endif 696#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 697 struct nf_bridge_info *nf_bridge; 698#endif 699 unsigned int len, 700 data_len; 701 __u16 mac_len, 702 hdr_len; 703 704 /* Following fields are _not_ copied in __copy_skb_header() 705 * Note that queue_mapping is here mostly to fill a hole. 706 */ 707 kmemcheck_bitfield_begin(flags1); 708 __u16 queue_mapping; 709 710/* if you move cloned around you also must adapt those constants */ 711#ifdef __BIG_ENDIAN_BITFIELD 712#define CLONED_MASK (1 << 7) 713#else 714#define CLONED_MASK 1 715#endif 716#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset) 717 718 __u8 __cloned_offset[0]; 719 __u8 cloned:1, 720 nohdr:1, 721 fclone:2, 722 peeked:1, 723 head_frag:1, 724 xmit_more:1, 725 __unused:1; /* one bit hole */ 726 kmemcheck_bitfield_end(flags1); 727 728 /* fields enclosed in headers_start/headers_end are copied 729 * using a single memcpy() in __copy_skb_header() 730 */ 731 /* private: */ 732 __u32 headers_start[0]; 733 /* public: */ 734 735/* if you move pkt_type around you also must adapt those constants */ 736#ifdef __BIG_ENDIAN_BITFIELD 737#define PKT_TYPE_MAX (7 << 5) 738#else 739#define PKT_TYPE_MAX 7 740#endif 741#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 742 743 __u8 __pkt_type_offset[0]; 744 __u8 pkt_type:3; 745 __u8 pfmemalloc:1; 746 __u8 ignore_df:1; 747 748 __u8 nf_trace:1; 749 __u8 ip_summed:2; 750 __u8 ooo_okay:1; 751 __u8 l4_hash:1; 752 __u8 sw_hash:1; 753 __u8 wifi_acked_valid:1; 754 __u8 wifi_acked:1; 755 756 __u8 no_fcs:1; 757 /* Indicates the inner headers are valid in the skbuff. */ 758 __u8 encapsulation:1; 759 __u8 encap_hdr_csum:1; 760 __u8 csum_valid:1; 761 __u8 csum_complete_sw:1; 762 __u8 csum_level:2; 763 __u8 csum_not_inet:1; 764 765 __u8 dst_pending_confirm:1; 766#ifdef CONFIG_IPV6_NDISC_NODETYPE 767 __u8 ndisc_nodetype:2; 768#endif 769 __u8 ipvs_property:1; 770 __u8 inner_protocol_type:1; 771 __u8 remcsum_offload:1; 772#ifdef CONFIG_NET_SWITCHDEV 773 __u8 offload_fwd_mark:1; 774#endif 775#ifdef CONFIG_NET_CLS_ACT 776 __u8 tc_skip_classify:1; 777 __u8 tc_at_ingress:1; 778 __u8 tc_redirected:1; 779 __u8 tc_from_ingress:1; 780#endif 781 782#ifdef CONFIG_NET_SCHED 783 __u16 tc_index; /* traffic control index */ 784#endif 785 786 union { 787 __wsum csum; 788 struct { 789 __u16 csum_start; 790 __u16 csum_offset; 791 }; 792 }; 793 __u32 priority; 794 int skb_iif; 795 __u32 hash; 796 __be16 vlan_proto; 797 __u16 vlan_tci; 798#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 799 union { 800 unsigned int napi_id; 801 unsigned int sender_cpu; 802 }; 803#endif 804#ifdef CONFIG_NETWORK_SECMARK 805 __u32 secmark; 806#endif 807 808 union { 809 __u32 mark; 810 __u32 reserved_tailroom; 811 }; 812 813 union { 814 __be16 inner_protocol; 815 __u8 inner_ipproto; 816 }; 817 818 __u16 inner_transport_header; 819 __u16 inner_network_header; 820 __u16 inner_mac_header; 821 822 __be16 protocol; 823 __u16 transport_header; 824 __u16 network_header; 825 __u16 mac_header; 826 827 /* private: */ 828 __u32 headers_end[0]; 829 /* public: */ 830 831 /* These elements must be at the end, see alloc_skb() for details. */ 832 sk_buff_data_t tail; 833 sk_buff_data_t end; 834 unsigned char *head, 835 *data; 836 unsigned int truesize; 837 refcount_t users; 838}; 839 840#ifdef __KERNEL__ 841/* 842 * Handling routines are only of interest to the kernel 843 */ 844#include <linux/slab.h> 845 846 847#define SKB_ALLOC_FCLONE 0x01 848#define SKB_ALLOC_RX 0x02 849#define SKB_ALLOC_NAPI 0x04 850 851/* Returns true if the skb was allocated from PFMEMALLOC reserves */ 852static inline bool skb_pfmemalloc(const struct sk_buff *skb) 853{ 854 return unlikely(skb->pfmemalloc); 855} 856 857/* 858 * skb might have a dst pointer attached, refcounted or not. 859 * _skb_refdst low order bit is set if refcount was _not_ taken 860 */ 861#define SKB_DST_NOREF 1UL 862#define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 863 864#define SKB_NFCT_PTRMASK ~(7UL) 865/** 866 * skb_dst - returns skb dst_entry 867 * @skb: buffer 868 * 869 * Returns skb dst_entry, regardless of reference taken or not. 870 */ 871static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 872{ 873 /* If refdst was not refcounted, check we still are in a 874 * rcu_read_lock section 875 */ 876 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 877 !rcu_read_lock_held() && 878 !rcu_read_lock_bh_held()); 879 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 880} 881 882/** 883 * skb_dst_set - sets skb dst 884 * @skb: buffer 885 * @dst: dst entry 886 * 887 * Sets skb dst, assuming a reference was taken on dst and should 888 * be released by skb_dst_drop() 889 */ 890static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 891{ 892 skb->_skb_refdst = (unsigned long)dst; 893} 894 895/** 896 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 897 * @skb: buffer 898 * @dst: dst entry 899 * 900 * Sets skb dst, assuming a reference was not taken on dst. 901 * If dst entry is cached, we do not take reference and dst_release 902 * will be avoided by refdst_drop. If dst entry is not cached, we take 903 * reference, so that last dst_release can destroy the dst immediately. 904 */ 905static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 906{ 907 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 908 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 909} 910 911/** 912 * skb_dst_is_noref - Test if skb dst isn't refcounted 913 * @skb: buffer 914 */ 915static inline bool skb_dst_is_noref(const struct sk_buff *skb) 916{ 917 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 918} 919 920static inline struct rtable *skb_rtable(const struct sk_buff *skb) 921{ 922 return (struct rtable *)skb_dst(skb); 923} 924 925/* For mangling skb->pkt_type from user space side from applications 926 * such as nft, tc, etc, we only allow a conservative subset of 927 * possible pkt_types to be set. 928*/ 929static inline bool skb_pkt_type_ok(u32 ptype) 930{ 931 return ptype <= PACKET_OTHERHOST; 932} 933 934static inline unsigned int skb_napi_id(const struct sk_buff *skb) 935{ 936#ifdef CONFIG_NET_RX_BUSY_POLL 937 return skb->napi_id; 938#else 939 return 0; 940#endif 941} 942 943/* decrement the reference count and return true if we can free the skb */ 944static inline bool skb_unref(struct sk_buff *skb) 945{ 946 if (unlikely(!skb)) 947 return false; 948 if (likely(refcount_read(&skb->users) == 1)) 949 smp_rmb(); 950 else if (likely(!refcount_dec_and_test(&skb->users))) 951 return false; 952 953 return true; 954} 955 956void skb_release_head_state(struct sk_buff *skb); 957void kfree_skb(struct sk_buff *skb); 958void kfree_skb_list(struct sk_buff *segs); 959void skb_tx_error(struct sk_buff *skb); 960void consume_skb(struct sk_buff *skb); 961void __consume_stateless_skb(struct sk_buff *skb); 962void __kfree_skb(struct sk_buff *skb); 963extern struct kmem_cache *skbuff_head_cache; 964 965void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 966bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 967 bool *fragstolen, int *delta_truesize); 968 969struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 970 int node); 971struct sk_buff *__build_skb(void *data, unsigned int frag_size); 972struct sk_buff *build_skb(void *data, unsigned int frag_size); 973static inline struct sk_buff *alloc_skb(unsigned int size, 974 gfp_t priority) 975{ 976 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 977} 978 979struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 980 unsigned long data_len, 981 int max_page_order, 982 int *errcode, 983 gfp_t gfp_mask); 984 985/* Layout of fast clones : [skb1][skb2][fclone_ref] */ 986struct sk_buff_fclones { 987 struct sk_buff skb1; 988 989 struct sk_buff skb2; 990 991 refcount_t fclone_ref; 992}; 993 994/** 995 * skb_fclone_busy - check if fclone is busy 996 * @sk: socket 997 * @skb: buffer 998 * 999 * Returns true if skb is a fast clone, and its clone is not freed. 1000 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1001 * so we also check that this didnt happen. 1002 */ 1003static inline bool skb_fclone_busy(const struct sock *sk, 1004 const struct sk_buff *skb) 1005{ 1006 const struct sk_buff_fclones *fclones; 1007 1008 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1009 1010 return skb->fclone == SKB_FCLONE_ORIG && 1011 refcount_read(&fclones->fclone_ref) > 1 && 1012 fclones->skb2.sk == sk; 1013} 1014 1015static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1016 gfp_t priority) 1017{ 1018 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1019} 1020 1021struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1022int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1023struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1024struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1025struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1026 gfp_t gfp_mask, bool fclone); 1027static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1028 gfp_t gfp_mask) 1029{ 1030 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1031} 1032 1033int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1034struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1035 unsigned int headroom); 1036struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1037 int newtailroom, gfp_t priority); 1038int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1039 int offset, int len); 1040int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1041 int offset, int len); 1042int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1043int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1044 1045/** 1046 * skb_pad - zero pad the tail of an skb 1047 * @skb: buffer to pad 1048 * @pad: space to pad 1049 * 1050 * Ensure that a buffer is followed by a padding area that is zero 1051 * filled. Used by network drivers which may DMA or transfer data 1052 * beyond the buffer end onto the wire. 1053 * 1054 * May return error in out of memory cases. The skb is freed on error. 1055 */ 1056static inline int skb_pad(struct sk_buff *skb, int pad) 1057{ 1058 return __skb_pad(skb, pad, true); 1059} 1060#define dev_kfree_skb(a) consume_skb(a) 1061 1062int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 1063 int getfrag(void *from, char *to, int offset, 1064 int len, int odd, struct sk_buff *skb), 1065 void *from, int length); 1066 1067int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1068 int offset, size_t size); 1069 1070struct skb_seq_state { 1071 __u32 lower_offset; 1072 __u32 upper_offset; 1073 __u32 frag_idx; 1074 __u32 stepped_offset; 1075 struct sk_buff *root_skb; 1076 struct sk_buff *cur_skb; 1077 __u8 *frag_data; 1078}; 1079 1080void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1081 unsigned int to, struct skb_seq_state *st); 1082unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1083 struct skb_seq_state *st); 1084void skb_abort_seq_read(struct skb_seq_state *st); 1085 1086unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1087 unsigned int to, struct ts_config *config); 1088 1089/* 1090 * Packet hash types specify the type of hash in skb_set_hash. 1091 * 1092 * Hash types refer to the protocol layer addresses which are used to 1093 * construct a packet's hash. The hashes are used to differentiate or identify 1094 * flows of the protocol layer for the hash type. Hash types are either 1095 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1096 * 1097 * Properties of hashes: 1098 * 1099 * 1) Two packets in different flows have different hash values 1100 * 2) Two packets in the same flow should have the same hash value 1101 * 1102 * A hash at a higher layer is considered to be more specific. A driver should 1103 * set the most specific hash possible. 1104 * 1105 * A driver cannot indicate a more specific hash than the layer at which a hash 1106 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1107 * 1108 * A driver may indicate a hash level which is less specific than the 1109 * actual layer the hash was computed on. For instance, a hash computed 1110 * at L4 may be considered an L3 hash. This should only be done if the 1111 * driver can't unambiguously determine that the HW computed the hash at 1112 * the higher layer. Note that the "should" in the second property above 1113 * permits this. 1114 */ 1115enum pkt_hash_types { 1116 PKT_HASH_TYPE_NONE, /* Undefined type */ 1117 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1118 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1119 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1120}; 1121 1122static inline void skb_clear_hash(struct sk_buff *skb) 1123{ 1124 skb->hash = 0; 1125 skb->sw_hash = 0; 1126 skb->l4_hash = 0; 1127} 1128 1129static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1130{ 1131 if (!skb->l4_hash) 1132 skb_clear_hash(skb); 1133} 1134 1135static inline void 1136__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1137{ 1138 skb->l4_hash = is_l4; 1139 skb->sw_hash = is_sw; 1140 skb->hash = hash; 1141} 1142 1143static inline void 1144skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1145{ 1146 /* Used by drivers to set hash from HW */ 1147 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1148} 1149 1150static inline void 1151__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1152{ 1153 __skb_set_hash(skb, hash, true, is_l4); 1154} 1155 1156void __skb_get_hash(struct sk_buff *skb); 1157u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1158u32 skb_get_poff(const struct sk_buff *skb); 1159u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1160 const struct flow_keys *keys, int hlen); 1161__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1162 void *data, int hlen_proto); 1163 1164static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1165 int thoff, u8 ip_proto) 1166{ 1167 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1168} 1169 1170void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1171 const struct flow_dissector_key *key, 1172 unsigned int key_count); 1173 1174bool __skb_flow_dissect(const struct sk_buff *skb, 1175 struct flow_dissector *flow_dissector, 1176 void *target_container, 1177 void *data, __be16 proto, int nhoff, int hlen, 1178 unsigned int flags); 1179 1180static inline bool skb_flow_dissect(const struct sk_buff *skb, 1181 struct flow_dissector *flow_dissector, 1182 void *target_container, unsigned int flags) 1183{ 1184 return __skb_flow_dissect(skb, flow_dissector, target_container, 1185 NULL, 0, 0, 0, flags); 1186} 1187 1188static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1189 struct flow_keys *flow, 1190 unsigned int flags) 1191{ 1192 memset(flow, 0, sizeof(*flow)); 1193 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1194 NULL, 0, 0, 0, flags); 1195} 1196 1197static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow, 1198 void *data, __be16 proto, 1199 int nhoff, int hlen, 1200 unsigned int flags) 1201{ 1202 memset(flow, 0, sizeof(*flow)); 1203 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow, 1204 data, proto, nhoff, hlen, flags); 1205} 1206 1207static inline __u32 skb_get_hash(struct sk_buff *skb) 1208{ 1209 if (!skb->l4_hash && !skb->sw_hash) 1210 __skb_get_hash(skb); 1211 1212 return skb->hash; 1213} 1214 1215static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1216{ 1217 if (!skb->l4_hash && !skb->sw_hash) { 1218 struct flow_keys keys; 1219 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1220 1221 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1222 } 1223 1224 return skb->hash; 1225} 1226 1227__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1228 1229static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1230{ 1231 return skb->hash; 1232} 1233 1234static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1235{ 1236 to->hash = from->hash; 1237 to->sw_hash = from->sw_hash; 1238 to->l4_hash = from->l4_hash; 1239}; 1240 1241#ifdef NET_SKBUFF_DATA_USES_OFFSET 1242static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1243{ 1244 return skb->head + skb->end; 1245} 1246 1247static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1248{ 1249 return skb->end; 1250} 1251#else 1252static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1253{ 1254 return skb->end; 1255} 1256 1257static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1258{ 1259 return skb->end - skb->head; 1260} 1261#endif 1262 1263/* Internal */ 1264#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1265 1266static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1267{ 1268 return &skb_shinfo(skb)->hwtstamps; 1269} 1270 1271static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1272{ 1273 bool is_zcopy = skb && skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY; 1274 1275 return is_zcopy ? skb_uarg(skb) : NULL; 1276} 1277 1278static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg) 1279{ 1280 if (skb && uarg && !skb_zcopy(skb)) { 1281 sock_zerocopy_get(uarg); 1282 skb_shinfo(skb)->destructor_arg = uarg; 1283 skb_shinfo(skb)->tx_flags |= SKBTX_ZEROCOPY_FRAG; 1284 } 1285} 1286 1287/* Release a reference on a zerocopy structure */ 1288static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy) 1289{ 1290 struct ubuf_info *uarg = skb_zcopy(skb); 1291 1292 if (uarg) { 1293 if (uarg->callback == sock_zerocopy_callback) { 1294 uarg->zerocopy = uarg->zerocopy && zerocopy; 1295 sock_zerocopy_put(uarg); 1296 } else { 1297 uarg->callback(uarg, zerocopy); 1298 } 1299 1300 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1301 } 1302} 1303 1304/* Abort a zerocopy operation and revert zckey on error in send syscall */ 1305static inline void skb_zcopy_abort(struct sk_buff *skb) 1306{ 1307 struct ubuf_info *uarg = skb_zcopy(skb); 1308 1309 if (uarg) { 1310 sock_zerocopy_put_abort(uarg); 1311 skb_shinfo(skb)->tx_flags &= ~SKBTX_ZEROCOPY_FRAG; 1312 } 1313} 1314 1315/** 1316 * skb_queue_empty - check if a queue is empty 1317 * @list: queue head 1318 * 1319 * Returns true if the queue is empty, false otherwise. 1320 */ 1321static inline int skb_queue_empty(const struct sk_buff_head *list) 1322{ 1323 return list->next == (const struct sk_buff *) list; 1324} 1325 1326/** 1327 * skb_queue_is_last - check if skb is the last entry in the queue 1328 * @list: queue head 1329 * @skb: buffer 1330 * 1331 * Returns true if @skb is the last buffer on the list. 1332 */ 1333static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1334 const struct sk_buff *skb) 1335{ 1336 return skb->next == (const struct sk_buff *) list; 1337} 1338 1339/** 1340 * skb_queue_is_first - check if skb is the first entry in the queue 1341 * @list: queue head 1342 * @skb: buffer 1343 * 1344 * Returns true if @skb is the first buffer on the list. 1345 */ 1346static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1347 const struct sk_buff *skb) 1348{ 1349 return skb->prev == (const struct sk_buff *) list; 1350} 1351 1352/** 1353 * skb_queue_next - return the next packet in the queue 1354 * @list: queue head 1355 * @skb: current buffer 1356 * 1357 * Return the next packet in @list after @skb. It is only valid to 1358 * call this if skb_queue_is_last() evaluates to false. 1359 */ 1360static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1361 const struct sk_buff *skb) 1362{ 1363 /* This BUG_ON may seem severe, but if we just return then we 1364 * are going to dereference garbage. 1365 */ 1366 BUG_ON(skb_queue_is_last(list, skb)); 1367 return skb->next; 1368} 1369 1370/** 1371 * skb_queue_prev - return the prev packet in the queue 1372 * @list: queue head 1373 * @skb: current buffer 1374 * 1375 * Return the prev packet in @list before @skb. It is only valid to 1376 * call this if skb_queue_is_first() evaluates to false. 1377 */ 1378static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1379 const struct sk_buff *skb) 1380{ 1381 /* This BUG_ON may seem severe, but if we just return then we 1382 * are going to dereference garbage. 1383 */ 1384 BUG_ON(skb_queue_is_first(list, skb)); 1385 return skb->prev; 1386} 1387 1388/** 1389 * skb_get - reference buffer 1390 * @skb: buffer to reference 1391 * 1392 * Makes another reference to a socket buffer and returns a pointer 1393 * to the buffer. 1394 */ 1395static inline struct sk_buff *skb_get(struct sk_buff *skb) 1396{ 1397 refcount_inc(&skb->users); 1398 return skb; 1399} 1400 1401/* 1402 * If users == 1, we are the only owner and are can avoid redundant 1403 * atomic change. 1404 */ 1405 1406/** 1407 * skb_cloned - is the buffer a clone 1408 * @skb: buffer to check 1409 * 1410 * Returns true if the buffer was generated with skb_clone() and is 1411 * one of multiple shared copies of the buffer. Cloned buffers are 1412 * shared data so must not be written to under normal circumstances. 1413 */ 1414static inline int skb_cloned(const struct sk_buff *skb) 1415{ 1416 return skb->cloned && 1417 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1418} 1419 1420static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1421{ 1422 might_sleep_if(gfpflags_allow_blocking(pri)); 1423 1424 if (skb_cloned(skb)) 1425 return pskb_expand_head(skb, 0, 0, pri); 1426 1427 return 0; 1428} 1429 1430/** 1431 * skb_header_cloned - is the header a clone 1432 * @skb: buffer to check 1433 * 1434 * Returns true if modifying the header part of the buffer requires 1435 * the data to be copied. 1436 */ 1437static inline int skb_header_cloned(const struct sk_buff *skb) 1438{ 1439 int dataref; 1440 1441 if (!skb->cloned) 1442 return 0; 1443 1444 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1445 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1446 return dataref != 1; 1447} 1448 1449static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1450{ 1451 might_sleep_if(gfpflags_allow_blocking(pri)); 1452 1453 if (skb_header_cloned(skb)) 1454 return pskb_expand_head(skb, 0, 0, pri); 1455 1456 return 0; 1457} 1458 1459/** 1460 * skb_header_release - release reference to header 1461 * @skb: buffer to operate on 1462 * 1463 * Drop a reference to the header part of the buffer. This is done 1464 * by acquiring a payload reference. You must not read from the header 1465 * part of skb->data after this. 1466 * Note : Check if you can use __skb_header_release() instead. 1467 */ 1468static inline void skb_header_release(struct sk_buff *skb) 1469{ 1470 BUG_ON(skb->nohdr); 1471 skb->nohdr = 1; 1472 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1473} 1474 1475/** 1476 * __skb_header_release - release reference to header 1477 * @skb: buffer to operate on 1478 * 1479 * Variant of skb_header_release() assuming skb is private to caller. 1480 * We can avoid one atomic operation. 1481 */ 1482static inline void __skb_header_release(struct sk_buff *skb) 1483{ 1484 skb->nohdr = 1; 1485 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1486} 1487 1488 1489/** 1490 * skb_shared - is the buffer shared 1491 * @skb: buffer to check 1492 * 1493 * Returns true if more than one person has a reference to this 1494 * buffer. 1495 */ 1496static inline int skb_shared(const struct sk_buff *skb) 1497{ 1498 return refcount_read(&skb->users) != 1; 1499} 1500 1501/** 1502 * skb_share_check - check if buffer is shared and if so clone it 1503 * @skb: buffer to check 1504 * @pri: priority for memory allocation 1505 * 1506 * If the buffer is shared the buffer is cloned and the old copy 1507 * drops a reference. A new clone with a single reference is returned. 1508 * If the buffer is not shared the original buffer is returned. When 1509 * being called from interrupt status or with spinlocks held pri must 1510 * be GFP_ATOMIC. 1511 * 1512 * NULL is returned on a memory allocation failure. 1513 */ 1514static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1515{ 1516 might_sleep_if(gfpflags_allow_blocking(pri)); 1517 if (skb_shared(skb)) { 1518 struct sk_buff *nskb = skb_clone(skb, pri); 1519 1520 if (likely(nskb)) 1521 consume_skb(skb); 1522 else 1523 kfree_skb(skb); 1524 skb = nskb; 1525 } 1526 return skb; 1527} 1528 1529/* 1530 * Copy shared buffers into a new sk_buff. We effectively do COW on 1531 * packets to handle cases where we have a local reader and forward 1532 * and a couple of other messy ones. The normal one is tcpdumping 1533 * a packet thats being forwarded. 1534 */ 1535 1536/** 1537 * skb_unshare - make a copy of a shared buffer 1538 * @skb: buffer to check 1539 * @pri: priority for memory allocation 1540 * 1541 * If the socket buffer is a clone then this function creates a new 1542 * copy of the data, drops a reference count on the old copy and returns 1543 * the new copy with the reference count at 1. If the buffer is not a clone 1544 * the original buffer is returned. When called with a spinlock held or 1545 * from interrupt state @pri must be %GFP_ATOMIC 1546 * 1547 * %NULL is returned on a memory allocation failure. 1548 */ 1549static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1550 gfp_t pri) 1551{ 1552 might_sleep_if(gfpflags_allow_blocking(pri)); 1553 if (skb_cloned(skb)) { 1554 struct sk_buff *nskb = skb_copy(skb, pri); 1555 1556 /* Free our shared copy */ 1557 if (likely(nskb)) 1558 consume_skb(skb); 1559 else 1560 kfree_skb(skb); 1561 skb = nskb; 1562 } 1563 return skb; 1564} 1565 1566/** 1567 * skb_peek - peek at the head of an &sk_buff_head 1568 * @list_: list to peek at 1569 * 1570 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1571 * be careful with this one. A peek leaves the buffer on the 1572 * list and someone else may run off with it. You must hold 1573 * the appropriate locks or have a private queue to do this. 1574 * 1575 * Returns %NULL for an empty list or a pointer to the head element. 1576 * The reference count is not incremented and the reference is therefore 1577 * volatile. Use with caution. 1578 */ 1579static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1580{ 1581 struct sk_buff *skb = list_->next; 1582 1583 if (skb == (struct sk_buff *)list_) 1584 skb = NULL; 1585 return skb; 1586} 1587 1588/** 1589 * skb_peek_next - peek skb following the given one from a queue 1590 * @skb: skb to start from 1591 * @list_: list to peek at 1592 * 1593 * Returns %NULL when the end of the list is met or a pointer to the 1594 * next element. The reference count is not incremented and the 1595 * reference is therefore volatile. Use with caution. 1596 */ 1597static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1598 const struct sk_buff_head *list_) 1599{ 1600 struct sk_buff *next = skb->next; 1601 1602 if (next == (struct sk_buff *)list_) 1603 next = NULL; 1604 return next; 1605} 1606 1607/** 1608 * skb_peek_tail - peek at the tail of an &sk_buff_head 1609 * @list_: list to peek at 1610 * 1611 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1612 * be careful with this one. A peek leaves the buffer on the 1613 * list and someone else may run off with it. You must hold 1614 * the appropriate locks or have a private queue to do this. 1615 * 1616 * Returns %NULL for an empty list or a pointer to the tail element. 1617 * The reference count is not incremented and the reference is therefore 1618 * volatile. Use with caution. 1619 */ 1620static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1621{ 1622 struct sk_buff *skb = list_->prev; 1623 1624 if (skb == (struct sk_buff *)list_) 1625 skb = NULL; 1626 return skb; 1627 1628} 1629 1630/** 1631 * skb_queue_len - get queue length 1632 * @list_: list to measure 1633 * 1634 * Return the length of an &sk_buff queue. 1635 */ 1636static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1637{ 1638 return list_->qlen; 1639} 1640 1641/** 1642 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1643 * @list: queue to initialize 1644 * 1645 * This initializes only the list and queue length aspects of 1646 * an sk_buff_head object. This allows to initialize the list 1647 * aspects of an sk_buff_head without reinitializing things like 1648 * the spinlock. It can also be used for on-stack sk_buff_head 1649 * objects where the spinlock is known to not be used. 1650 */ 1651static inline void __skb_queue_head_init(struct sk_buff_head *list) 1652{ 1653 list->prev = list->next = (struct sk_buff *)list; 1654 list->qlen = 0; 1655} 1656 1657/* 1658 * This function creates a split out lock class for each invocation; 1659 * this is needed for now since a whole lot of users of the skb-queue 1660 * infrastructure in drivers have different locking usage (in hardirq) 1661 * than the networking core (in softirq only). In the long run either the 1662 * network layer or drivers should need annotation to consolidate the 1663 * main types of usage into 3 classes. 1664 */ 1665static inline void skb_queue_head_init(struct sk_buff_head *list) 1666{ 1667 spin_lock_init(&list->lock); 1668 __skb_queue_head_init(list); 1669} 1670 1671static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1672 struct lock_class_key *class) 1673{ 1674 skb_queue_head_init(list); 1675 lockdep_set_class(&list->lock, class); 1676} 1677 1678/* 1679 * Insert an sk_buff on a list. 1680 * 1681 * The "__skb_xxxx()" functions are the non-atomic ones that 1682 * can only be called with interrupts disabled. 1683 */ 1684void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1685 struct sk_buff_head *list); 1686static inline void __skb_insert(struct sk_buff *newsk, 1687 struct sk_buff *prev, struct sk_buff *next, 1688 struct sk_buff_head *list) 1689{ 1690 newsk->next = next; 1691 newsk->prev = prev; 1692 next->prev = prev->next = newsk; 1693 list->qlen++; 1694} 1695 1696static inline void __skb_queue_splice(const struct sk_buff_head *list, 1697 struct sk_buff *prev, 1698 struct sk_buff *next) 1699{ 1700 struct sk_buff *first = list->next; 1701 struct sk_buff *last = list->prev; 1702 1703 first->prev = prev; 1704 prev->next = first; 1705 1706 last->next = next; 1707 next->prev = last; 1708} 1709 1710/** 1711 * skb_queue_splice - join two skb lists, this is designed for stacks 1712 * @list: the new list to add 1713 * @head: the place to add it in the first list 1714 */ 1715static inline void skb_queue_splice(const struct sk_buff_head *list, 1716 struct sk_buff_head *head) 1717{ 1718 if (!skb_queue_empty(list)) { 1719 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1720 head->qlen += list->qlen; 1721 } 1722} 1723 1724/** 1725 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1726 * @list: the new list to add 1727 * @head: the place to add it in the first list 1728 * 1729 * The list at @list is reinitialised 1730 */ 1731static inline void skb_queue_splice_init(struct sk_buff_head *list, 1732 struct sk_buff_head *head) 1733{ 1734 if (!skb_queue_empty(list)) { 1735 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1736 head->qlen += list->qlen; 1737 __skb_queue_head_init(list); 1738 } 1739} 1740 1741/** 1742 * skb_queue_splice_tail - join two skb lists, each list being a queue 1743 * @list: the new list to add 1744 * @head: the place to add it in the first list 1745 */ 1746static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1747 struct sk_buff_head *head) 1748{ 1749 if (!skb_queue_empty(list)) { 1750 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1751 head->qlen += list->qlen; 1752 } 1753} 1754 1755/** 1756 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1757 * @list: the new list to add 1758 * @head: the place to add it in the first list 1759 * 1760 * Each of the lists is a queue. 1761 * The list at @list is reinitialised 1762 */ 1763static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1764 struct sk_buff_head *head) 1765{ 1766 if (!skb_queue_empty(list)) { 1767 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1768 head->qlen += list->qlen; 1769 __skb_queue_head_init(list); 1770 } 1771} 1772 1773/** 1774 * __skb_queue_after - queue a buffer at the list head 1775 * @list: list to use 1776 * @prev: place after this buffer 1777 * @newsk: buffer to queue 1778 * 1779 * Queue a buffer int the middle of a list. This function takes no locks 1780 * and you must therefore hold required locks before calling it. 1781 * 1782 * A buffer cannot be placed on two lists at the same time. 1783 */ 1784static inline void __skb_queue_after(struct sk_buff_head *list, 1785 struct sk_buff *prev, 1786 struct sk_buff *newsk) 1787{ 1788 __skb_insert(newsk, prev, prev->next, list); 1789} 1790 1791void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1792 struct sk_buff_head *list); 1793 1794static inline void __skb_queue_before(struct sk_buff_head *list, 1795 struct sk_buff *next, 1796 struct sk_buff *newsk) 1797{ 1798 __skb_insert(newsk, next->prev, next, list); 1799} 1800 1801/** 1802 * __skb_queue_head - queue a buffer at the list head 1803 * @list: list to use 1804 * @newsk: buffer to queue 1805 * 1806 * Queue a buffer at the start of a list. This function takes no locks 1807 * and you must therefore hold required locks before calling it. 1808 * 1809 * A buffer cannot be placed on two lists at the same time. 1810 */ 1811void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1812static inline void __skb_queue_head(struct sk_buff_head *list, 1813 struct sk_buff *newsk) 1814{ 1815 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1816} 1817 1818/** 1819 * __skb_queue_tail - queue a buffer at the list tail 1820 * @list: list to use 1821 * @newsk: buffer to queue 1822 * 1823 * Queue a buffer at the end of a list. This function takes no locks 1824 * and you must therefore hold required locks before calling it. 1825 * 1826 * A buffer cannot be placed on two lists at the same time. 1827 */ 1828void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1829static inline void __skb_queue_tail(struct sk_buff_head *list, 1830 struct sk_buff *newsk) 1831{ 1832 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1833} 1834 1835/* 1836 * remove sk_buff from list. _Must_ be called atomically, and with 1837 * the list known.. 1838 */ 1839void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1840static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1841{ 1842 struct sk_buff *next, *prev; 1843 1844 list->qlen--; 1845 next = skb->next; 1846 prev = skb->prev; 1847 skb->next = skb->prev = NULL; 1848 next->prev = prev; 1849 prev->next = next; 1850} 1851 1852/** 1853 * __skb_dequeue - remove from the head of the queue 1854 * @list: list to dequeue from 1855 * 1856 * Remove the head of the list. This function does not take any locks 1857 * so must be used with appropriate locks held only. The head item is 1858 * returned or %NULL if the list is empty. 1859 */ 1860struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1861static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1862{ 1863 struct sk_buff *skb = skb_peek(list); 1864 if (skb) 1865 __skb_unlink(skb, list); 1866 return skb; 1867} 1868 1869/** 1870 * __skb_dequeue_tail - remove from the tail of the queue 1871 * @list: list to dequeue from 1872 * 1873 * Remove the tail of the list. This function does not take any locks 1874 * so must be used with appropriate locks held only. The tail item is 1875 * returned or %NULL if the list is empty. 1876 */ 1877struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1878static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1879{ 1880 struct sk_buff *skb = skb_peek_tail(list); 1881 if (skb) 1882 __skb_unlink(skb, list); 1883 return skb; 1884} 1885 1886 1887static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1888{ 1889 return skb->data_len; 1890} 1891 1892static inline unsigned int skb_headlen(const struct sk_buff *skb) 1893{ 1894 return skb->len - skb->data_len; 1895} 1896 1897static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 1898{ 1899 unsigned int i, len = 0; 1900 1901 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 1902 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1903 return len; 1904} 1905 1906static inline unsigned int skb_pagelen(const struct sk_buff *skb) 1907{ 1908 return skb_headlen(skb) + __skb_pagelen(skb); 1909} 1910 1911/** 1912 * __skb_fill_page_desc - initialise a paged fragment in an skb 1913 * @skb: buffer containing fragment to be initialised 1914 * @i: paged fragment index to initialise 1915 * @page: the page to use for this fragment 1916 * @off: the offset to the data with @page 1917 * @size: the length of the data 1918 * 1919 * Initialises the @i'th fragment of @skb to point to &size bytes at 1920 * offset @off within @page. 1921 * 1922 * Does not take any additional reference on the fragment. 1923 */ 1924static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1925 struct page *page, int off, int size) 1926{ 1927 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1928 1929 /* 1930 * Propagate page pfmemalloc to the skb if we can. The problem is 1931 * that not all callers have unique ownership of the page but rely 1932 * on page_is_pfmemalloc doing the right thing(tm). 1933 */ 1934 frag->page.p = page; 1935 frag->page_offset = off; 1936 skb_frag_size_set(frag, size); 1937 1938 page = compound_head(page); 1939 if (page_is_pfmemalloc(page)) 1940 skb->pfmemalloc = true; 1941} 1942 1943/** 1944 * skb_fill_page_desc - initialise a paged fragment in an skb 1945 * @skb: buffer containing fragment to be initialised 1946 * @i: paged fragment index to initialise 1947 * @page: the page to use for this fragment 1948 * @off: the offset to the data with @page 1949 * @size: the length of the data 1950 * 1951 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1952 * @skb to point to @size bytes at offset @off within @page. In 1953 * addition updates @skb such that @i is the last fragment. 1954 * 1955 * Does not take any additional reference on the fragment. 1956 */ 1957static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1958 struct page *page, int off, int size) 1959{ 1960 __skb_fill_page_desc(skb, i, page, off, size); 1961 skb_shinfo(skb)->nr_frags = i + 1; 1962} 1963 1964void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1965 int size, unsigned int truesize); 1966 1967void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1968 unsigned int truesize); 1969 1970#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1971#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1972#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1973 1974#ifdef NET_SKBUFF_DATA_USES_OFFSET 1975static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1976{ 1977 return skb->head + skb->tail; 1978} 1979 1980static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1981{ 1982 skb->tail = skb->data - skb->head; 1983} 1984 1985static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1986{ 1987 skb_reset_tail_pointer(skb); 1988 skb->tail += offset; 1989} 1990 1991#else /* NET_SKBUFF_DATA_USES_OFFSET */ 1992static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1993{ 1994 return skb->tail; 1995} 1996 1997static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1998{ 1999 skb->tail = skb->data; 2000} 2001 2002static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2003{ 2004 skb->tail = skb->data + offset; 2005} 2006 2007#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2008 2009/* 2010 * Add data to an sk_buff 2011 */ 2012void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2013void *skb_put(struct sk_buff *skb, unsigned int len); 2014static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2015{ 2016 void *tmp = skb_tail_pointer(skb); 2017 SKB_LINEAR_ASSERT(skb); 2018 skb->tail += len; 2019 skb->len += len; 2020 return tmp; 2021} 2022 2023static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2024{ 2025 void *tmp = __skb_put(skb, len); 2026 2027 memset(tmp, 0, len); 2028 return tmp; 2029} 2030 2031static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2032 unsigned int len) 2033{ 2034 void *tmp = __skb_put(skb, len); 2035 2036 memcpy(tmp, data, len); 2037 return tmp; 2038} 2039 2040static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2041{ 2042 *(u8 *)__skb_put(skb, 1) = val; 2043} 2044 2045static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2046{ 2047 void *tmp = skb_put(skb, len); 2048 2049 memset(tmp, 0, len); 2050 2051 return tmp; 2052} 2053 2054static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2055 unsigned int len) 2056{ 2057 void *tmp = skb_put(skb, len); 2058 2059 memcpy(tmp, data, len); 2060 2061 return tmp; 2062} 2063 2064static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2065{ 2066 *(u8 *)skb_put(skb, 1) = val; 2067} 2068 2069void *skb_push(struct sk_buff *skb, unsigned int len); 2070static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2071{ 2072 skb->data -= len; 2073 skb->len += len; 2074 return skb->data; 2075} 2076 2077void *skb_pull(struct sk_buff *skb, unsigned int len); 2078static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2079{ 2080 skb->len -= len; 2081 BUG_ON(skb->len < skb->data_len); 2082 return skb->data += len; 2083} 2084 2085static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2086{ 2087 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2088} 2089 2090void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2091 2092static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len) 2093{ 2094 if (len > skb_headlen(skb) && 2095 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 2096 return NULL; 2097 skb->len -= len; 2098 return skb->data += len; 2099} 2100 2101static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2102{ 2103 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 2104} 2105 2106static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 2107{ 2108 if (likely(len <= skb_headlen(skb))) 2109 return 1; 2110 if (unlikely(len > skb->len)) 2111 return 0; 2112 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2113} 2114 2115void skb_condense(struct sk_buff *skb); 2116 2117/** 2118 * skb_headroom - bytes at buffer head 2119 * @skb: buffer to check 2120 * 2121 * Return the number of bytes of free space at the head of an &sk_buff. 2122 */ 2123static inline unsigned int skb_headroom(const struct sk_buff *skb) 2124{ 2125 return skb->data - skb->head; 2126} 2127 2128/** 2129 * skb_tailroom - bytes at buffer end 2130 * @skb: buffer to check 2131 * 2132 * Return the number of bytes of free space at the tail of an sk_buff 2133 */ 2134static inline int skb_tailroom(const struct sk_buff *skb) 2135{ 2136 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2137} 2138 2139/** 2140 * skb_availroom - bytes at buffer end 2141 * @skb: buffer to check 2142 * 2143 * Return the number of bytes of free space at the tail of an sk_buff 2144 * allocated by sk_stream_alloc() 2145 */ 2146static inline int skb_availroom(const struct sk_buff *skb) 2147{ 2148 if (skb_is_nonlinear(skb)) 2149 return 0; 2150 2151 return skb->end - skb->tail - skb->reserved_tailroom; 2152} 2153 2154/** 2155 * skb_reserve - adjust headroom 2156 * @skb: buffer to alter 2157 * @len: bytes to move 2158 * 2159 * Increase the headroom of an empty &sk_buff by reducing the tail 2160 * room. This is only allowed for an empty buffer. 2161 */ 2162static inline void skb_reserve(struct sk_buff *skb, int len) 2163{ 2164 skb->data += len; 2165 skb->tail += len; 2166} 2167 2168/** 2169 * skb_tailroom_reserve - adjust reserved_tailroom 2170 * @skb: buffer to alter 2171 * @mtu: maximum amount of headlen permitted 2172 * @needed_tailroom: minimum amount of reserved_tailroom 2173 * 2174 * Set reserved_tailroom so that headlen can be as large as possible but 2175 * not larger than mtu and tailroom cannot be smaller than 2176 * needed_tailroom. 2177 * The required headroom should already have been reserved before using 2178 * this function. 2179 */ 2180static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2181 unsigned int needed_tailroom) 2182{ 2183 SKB_LINEAR_ASSERT(skb); 2184 if (mtu < skb_tailroom(skb) - needed_tailroom) 2185 /* use at most mtu */ 2186 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2187 else 2188 /* use up to all available space */ 2189 skb->reserved_tailroom = needed_tailroom; 2190} 2191 2192#define ENCAP_TYPE_ETHER 0 2193#define ENCAP_TYPE_IPPROTO 1 2194 2195static inline void skb_set_inner_protocol(struct sk_buff *skb, 2196 __be16 protocol) 2197{ 2198 skb->inner_protocol = protocol; 2199 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2200} 2201 2202static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2203 __u8 ipproto) 2204{ 2205 skb->inner_ipproto = ipproto; 2206 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2207} 2208 2209static inline void skb_reset_inner_headers(struct sk_buff *skb) 2210{ 2211 skb->inner_mac_header = skb->mac_header; 2212 skb->inner_network_header = skb->network_header; 2213 skb->inner_transport_header = skb->transport_header; 2214} 2215 2216static inline void skb_reset_mac_len(struct sk_buff *skb) 2217{ 2218 skb->mac_len = skb->network_header - skb->mac_header; 2219} 2220 2221static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2222 *skb) 2223{ 2224 return skb->head + skb->inner_transport_header; 2225} 2226 2227static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2228{ 2229 return skb_inner_transport_header(skb) - skb->data; 2230} 2231 2232static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2233{ 2234 skb->inner_transport_header = skb->data - skb->head; 2235} 2236 2237static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2238 const int offset) 2239{ 2240 skb_reset_inner_transport_header(skb); 2241 skb->inner_transport_header += offset; 2242} 2243 2244static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2245{ 2246 return skb->head + skb->inner_network_header; 2247} 2248 2249static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2250{ 2251 skb->inner_network_header = skb->data - skb->head; 2252} 2253 2254static inline void skb_set_inner_network_header(struct sk_buff *skb, 2255 const int offset) 2256{ 2257 skb_reset_inner_network_header(skb); 2258 skb->inner_network_header += offset; 2259} 2260 2261static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2262{ 2263 return skb->head + skb->inner_mac_header; 2264} 2265 2266static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2267{ 2268 skb->inner_mac_header = skb->data - skb->head; 2269} 2270 2271static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2272 const int offset) 2273{ 2274 skb_reset_inner_mac_header(skb); 2275 skb->inner_mac_header += offset; 2276} 2277static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2278{ 2279 return skb->transport_header != (typeof(skb->transport_header))~0U; 2280} 2281 2282static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2283{ 2284 return skb->head + skb->transport_header; 2285} 2286 2287static inline void skb_reset_transport_header(struct sk_buff *skb) 2288{ 2289 skb->transport_header = skb->data - skb->head; 2290} 2291 2292static inline void skb_set_transport_header(struct sk_buff *skb, 2293 const int offset) 2294{ 2295 skb_reset_transport_header(skb); 2296 skb->transport_header += offset; 2297} 2298 2299static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2300{ 2301 return skb->head + skb->network_header; 2302} 2303 2304static inline void skb_reset_network_header(struct sk_buff *skb) 2305{ 2306 skb->network_header = skb->data - skb->head; 2307} 2308 2309static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2310{ 2311 skb_reset_network_header(skb); 2312 skb->network_header += offset; 2313} 2314 2315static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2316{ 2317 return skb->head + skb->mac_header; 2318} 2319 2320static inline int skb_mac_offset(const struct sk_buff *skb) 2321{ 2322 return skb_mac_header(skb) - skb->data; 2323} 2324 2325static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2326{ 2327 return skb->network_header - skb->mac_header; 2328} 2329 2330static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2331{ 2332 return skb->mac_header != (typeof(skb->mac_header))~0U; 2333} 2334 2335static inline void skb_reset_mac_header(struct sk_buff *skb) 2336{ 2337 skb->mac_header = skb->data - skb->head; 2338} 2339 2340static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2341{ 2342 skb_reset_mac_header(skb); 2343 skb->mac_header += offset; 2344} 2345 2346static inline void skb_pop_mac_header(struct sk_buff *skb) 2347{ 2348 skb->mac_header = skb->network_header; 2349} 2350 2351static inline void skb_probe_transport_header(struct sk_buff *skb, 2352 const int offset_hint) 2353{ 2354 struct flow_keys keys; 2355 2356 if (skb_transport_header_was_set(skb)) 2357 return; 2358 else if (skb_flow_dissect_flow_keys(skb, &keys, 0)) 2359 skb_set_transport_header(skb, keys.control.thoff); 2360 else 2361 skb_set_transport_header(skb, offset_hint); 2362} 2363 2364static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2365{ 2366 if (skb_mac_header_was_set(skb)) { 2367 const unsigned char *old_mac = skb_mac_header(skb); 2368 2369 skb_set_mac_header(skb, -skb->mac_len); 2370 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2371 } 2372} 2373 2374static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2375{ 2376 return skb->csum_start - skb_headroom(skb); 2377} 2378 2379static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2380{ 2381 return skb->head + skb->csum_start; 2382} 2383 2384static inline int skb_transport_offset(const struct sk_buff *skb) 2385{ 2386 return skb_transport_header(skb) - skb->data; 2387} 2388 2389static inline u32 skb_network_header_len(const struct sk_buff *skb) 2390{ 2391 return skb->transport_header - skb->network_header; 2392} 2393 2394static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2395{ 2396 return skb->inner_transport_header - skb->inner_network_header; 2397} 2398 2399static inline int skb_network_offset(const struct sk_buff *skb) 2400{ 2401 return skb_network_header(skb) - skb->data; 2402} 2403 2404static inline int skb_inner_network_offset(const struct sk_buff *skb) 2405{ 2406 return skb_inner_network_header(skb) - skb->data; 2407} 2408 2409static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2410{ 2411 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2412} 2413 2414/* 2415 * CPUs often take a performance hit when accessing unaligned memory 2416 * locations. The actual performance hit varies, it can be small if the 2417 * hardware handles it or large if we have to take an exception and fix it 2418 * in software. 2419 * 2420 * Since an ethernet header is 14 bytes network drivers often end up with 2421 * the IP header at an unaligned offset. The IP header can be aligned by 2422 * shifting the start of the packet by 2 bytes. Drivers should do this 2423 * with: 2424 * 2425 * skb_reserve(skb, NET_IP_ALIGN); 2426 * 2427 * The downside to this alignment of the IP header is that the DMA is now 2428 * unaligned. On some architectures the cost of an unaligned DMA is high 2429 * and this cost outweighs the gains made by aligning the IP header. 2430 * 2431 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2432 * to be overridden. 2433 */ 2434#ifndef NET_IP_ALIGN 2435#define NET_IP_ALIGN 2 2436#endif 2437 2438/* 2439 * The networking layer reserves some headroom in skb data (via 2440 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2441 * the header has to grow. In the default case, if the header has to grow 2442 * 32 bytes or less we avoid the reallocation. 2443 * 2444 * Unfortunately this headroom changes the DMA alignment of the resulting 2445 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2446 * on some architectures. An architecture can override this value, 2447 * perhaps setting it to a cacheline in size (since that will maintain 2448 * cacheline alignment of the DMA). It must be a power of 2. 2449 * 2450 * Various parts of the networking layer expect at least 32 bytes of 2451 * headroom, you should not reduce this. 2452 * 2453 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2454 * to reduce average number of cache lines per packet. 2455 * get_rps_cpus() for example only access one 64 bytes aligned block : 2456 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2457 */ 2458#ifndef NET_SKB_PAD 2459#define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2460#endif 2461 2462int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2463 2464static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2465{ 2466 if (unlikely(skb_is_nonlinear(skb))) { 2467 WARN_ON(1); 2468 return; 2469 } 2470 skb->len = len; 2471 skb_set_tail_pointer(skb, len); 2472} 2473 2474static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2475{ 2476 __skb_set_length(skb, len); 2477} 2478 2479void skb_trim(struct sk_buff *skb, unsigned int len); 2480 2481static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2482{ 2483 if (skb->data_len) 2484 return ___pskb_trim(skb, len); 2485 __skb_trim(skb, len); 2486 return 0; 2487} 2488 2489static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2490{ 2491 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2492} 2493 2494/** 2495 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2496 * @skb: buffer to alter 2497 * @len: new length 2498 * 2499 * This is identical to pskb_trim except that the caller knows that 2500 * the skb is not cloned so we should never get an error due to out- 2501 * of-memory. 2502 */ 2503static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2504{ 2505 int err = pskb_trim(skb, len); 2506 BUG_ON(err); 2507} 2508 2509static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2510{ 2511 unsigned int diff = len - skb->len; 2512 2513 if (skb_tailroom(skb) < diff) { 2514 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2515 GFP_ATOMIC); 2516 if (ret) 2517 return ret; 2518 } 2519 __skb_set_length(skb, len); 2520 return 0; 2521} 2522 2523/** 2524 * skb_orphan - orphan a buffer 2525 * @skb: buffer to orphan 2526 * 2527 * If a buffer currently has an owner then we call the owner's 2528 * destructor function and make the @skb unowned. The buffer continues 2529 * to exist but is no longer charged to its former owner. 2530 */ 2531static inline void skb_orphan(struct sk_buff *skb) 2532{ 2533 if (skb->destructor) { 2534 skb->destructor(skb); 2535 skb->destructor = NULL; 2536 skb->sk = NULL; 2537 } else { 2538 BUG_ON(skb->sk); 2539 } 2540} 2541 2542/** 2543 * skb_orphan_frags - orphan the frags contained in a buffer 2544 * @skb: buffer to orphan frags from 2545 * @gfp_mask: allocation mask for replacement pages 2546 * 2547 * For each frag in the SKB which needs a destructor (i.e. has an 2548 * owner) create a copy of that frag and release the original 2549 * page by calling the destructor. 2550 */ 2551static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2552{ 2553 if (likely(!skb_zcopy(skb))) 2554 return 0; 2555 if (skb_uarg(skb)->callback == sock_zerocopy_callback) 2556 return 0; 2557 return skb_copy_ubufs(skb, gfp_mask); 2558} 2559 2560/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 2561static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 2562{ 2563 if (likely(!skb_zcopy(skb))) 2564 return 0; 2565 return skb_copy_ubufs(skb, gfp_mask); 2566} 2567 2568/** 2569 * __skb_queue_purge - empty a list 2570 * @list: list to empty 2571 * 2572 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2573 * the list and one reference dropped. This function does not take the 2574 * list lock and the caller must hold the relevant locks to use it. 2575 */ 2576void skb_queue_purge(struct sk_buff_head *list); 2577static inline void __skb_queue_purge(struct sk_buff_head *list) 2578{ 2579 struct sk_buff *skb; 2580 while ((skb = __skb_dequeue(list)) != NULL) 2581 kfree_skb(skb); 2582} 2583 2584void skb_rbtree_purge(struct rb_root *root); 2585 2586void *netdev_alloc_frag(unsigned int fragsz); 2587 2588struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2589 gfp_t gfp_mask); 2590 2591/** 2592 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2593 * @dev: network device to receive on 2594 * @length: length to allocate 2595 * 2596 * Allocate a new &sk_buff and assign it a usage count of one. The 2597 * buffer has unspecified headroom built in. Users should allocate 2598 * the headroom they think they need without accounting for the 2599 * built in space. The built in space is used for optimisations. 2600 * 2601 * %NULL is returned if there is no free memory. Although this function 2602 * allocates memory it can be called from an interrupt. 2603 */ 2604static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2605 unsigned int length) 2606{ 2607 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2608} 2609 2610/* legacy helper around __netdev_alloc_skb() */ 2611static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2612 gfp_t gfp_mask) 2613{ 2614 return __netdev_alloc_skb(NULL, length, gfp_mask); 2615} 2616 2617/* legacy helper around netdev_alloc_skb() */ 2618static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2619{ 2620 return netdev_alloc_skb(NULL, length); 2621} 2622 2623 2624static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2625 unsigned int length, gfp_t gfp) 2626{ 2627 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2628 2629 if (NET_IP_ALIGN && skb) 2630 skb_reserve(skb, NET_IP_ALIGN); 2631 return skb; 2632} 2633 2634static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2635 unsigned int length) 2636{ 2637 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2638} 2639 2640static inline void skb_free_frag(void *addr) 2641{ 2642 page_frag_free(addr); 2643} 2644 2645void *napi_alloc_frag(unsigned int fragsz); 2646struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2647 unsigned int length, gfp_t gfp_mask); 2648static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2649 unsigned int length) 2650{ 2651 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2652} 2653void napi_consume_skb(struct sk_buff *skb, int budget); 2654 2655void __kfree_skb_flush(void); 2656void __kfree_skb_defer(struct sk_buff *skb); 2657 2658/** 2659 * __dev_alloc_pages - allocate page for network Rx 2660 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2661 * @order: size of the allocation 2662 * 2663 * Allocate a new page. 2664 * 2665 * %NULL is returned if there is no free memory. 2666*/ 2667static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2668 unsigned int order) 2669{ 2670 /* This piece of code contains several assumptions. 2671 * 1. This is for device Rx, therefor a cold page is preferred. 2672 * 2. The expectation is the user wants a compound page. 2673 * 3. If requesting a order 0 page it will not be compound 2674 * due to the check to see if order has a value in prep_new_page 2675 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2676 * code in gfp_to_alloc_flags that should be enforcing this. 2677 */ 2678 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2679 2680 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2681} 2682 2683static inline struct page *dev_alloc_pages(unsigned int order) 2684{ 2685 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 2686} 2687 2688/** 2689 * __dev_alloc_page - allocate a page for network Rx 2690 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2691 * 2692 * Allocate a new page. 2693 * 2694 * %NULL is returned if there is no free memory. 2695 */ 2696static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2697{ 2698 return __dev_alloc_pages(gfp_mask, 0); 2699} 2700 2701static inline struct page *dev_alloc_page(void) 2702{ 2703 return dev_alloc_pages(0); 2704} 2705 2706/** 2707 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2708 * @page: The page that was allocated from skb_alloc_page 2709 * @skb: The skb that may need pfmemalloc set 2710 */ 2711static inline void skb_propagate_pfmemalloc(struct page *page, 2712 struct sk_buff *skb) 2713{ 2714 if (page_is_pfmemalloc(page)) 2715 skb->pfmemalloc = true; 2716} 2717 2718/** 2719 * skb_frag_page - retrieve the page referred to by a paged fragment 2720 * @frag: the paged fragment 2721 * 2722 * Returns the &struct page associated with @frag. 2723 */ 2724static inline struct page *skb_frag_page(const skb_frag_t *frag) 2725{ 2726 return frag->page.p; 2727} 2728 2729/** 2730 * __skb_frag_ref - take an addition reference on a paged fragment. 2731 * @frag: the paged fragment 2732 * 2733 * Takes an additional reference on the paged fragment @frag. 2734 */ 2735static inline void __skb_frag_ref(skb_frag_t *frag) 2736{ 2737 get_page(skb_frag_page(frag)); 2738} 2739 2740/** 2741 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2742 * @skb: the buffer 2743 * @f: the fragment offset. 2744 * 2745 * Takes an additional reference on the @f'th paged fragment of @skb. 2746 */ 2747static inline void skb_frag_ref(struct sk_buff *skb, int f) 2748{ 2749 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2750} 2751 2752/** 2753 * __skb_frag_unref - release a reference on a paged fragment. 2754 * @frag: the paged fragment 2755 * 2756 * Releases a reference on the paged fragment @frag. 2757 */ 2758static inline void __skb_frag_unref(skb_frag_t *frag) 2759{ 2760 put_page(skb_frag_page(frag)); 2761} 2762 2763/** 2764 * skb_frag_unref - release a reference on a paged fragment of an skb. 2765 * @skb: the buffer 2766 * @f: the fragment offset 2767 * 2768 * Releases a reference on the @f'th paged fragment of @skb. 2769 */ 2770static inline void skb_frag_unref(struct sk_buff *skb, int f) 2771{ 2772 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2773} 2774 2775/** 2776 * skb_frag_address - gets the address of the data contained in a paged fragment 2777 * @frag: the paged fragment buffer 2778 * 2779 * Returns the address of the data within @frag. The page must already 2780 * be mapped. 2781 */ 2782static inline void *skb_frag_address(const skb_frag_t *frag) 2783{ 2784 return page_address(skb_frag_page(frag)) + frag->page_offset; 2785} 2786 2787/** 2788 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2789 * @frag: the paged fragment buffer 2790 * 2791 * Returns the address of the data within @frag. Checks that the page 2792 * is mapped and returns %NULL otherwise. 2793 */ 2794static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2795{ 2796 void *ptr = page_address(skb_frag_page(frag)); 2797 if (unlikely(!ptr)) 2798 return NULL; 2799 2800 return ptr + frag->page_offset; 2801} 2802 2803/** 2804 * __skb_frag_set_page - sets the page contained in a paged fragment 2805 * @frag: the paged fragment 2806 * @page: the page to set 2807 * 2808 * Sets the fragment @frag to contain @page. 2809 */ 2810static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2811{ 2812 frag->page.p = page; 2813} 2814 2815/** 2816 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2817 * @skb: the buffer 2818 * @f: the fragment offset 2819 * @page: the page to set 2820 * 2821 * Sets the @f'th fragment of @skb to contain @page. 2822 */ 2823static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2824 struct page *page) 2825{ 2826 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2827} 2828 2829bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2830 2831/** 2832 * skb_frag_dma_map - maps a paged fragment via the DMA API 2833 * @dev: the device to map the fragment to 2834 * @frag: the paged fragment to map 2835 * @offset: the offset within the fragment (starting at the 2836 * fragment's own offset) 2837 * @size: the number of bytes to map 2838 * @dir: the direction of the mapping (``PCI_DMA_*``) 2839 * 2840 * Maps the page associated with @frag to @device. 2841 */ 2842static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2843 const skb_frag_t *frag, 2844 size_t offset, size_t size, 2845 enum dma_data_direction dir) 2846{ 2847 return dma_map_page(dev, skb_frag_page(frag), 2848 frag->page_offset + offset, size, dir); 2849} 2850 2851static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2852 gfp_t gfp_mask) 2853{ 2854 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2855} 2856 2857 2858static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2859 gfp_t gfp_mask) 2860{ 2861 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2862} 2863 2864 2865/** 2866 * skb_clone_writable - is the header of a clone writable 2867 * @skb: buffer to check 2868 * @len: length up to which to write 2869 * 2870 * Returns true if modifying the header part of the cloned buffer 2871 * does not requires the data to be copied. 2872 */ 2873static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2874{ 2875 return !skb_header_cloned(skb) && 2876 skb_headroom(skb) + len <= skb->hdr_len; 2877} 2878 2879static inline int skb_try_make_writable(struct sk_buff *skb, 2880 unsigned int write_len) 2881{ 2882 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 2883 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2884} 2885 2886static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2887 int cloned) 2888{ 2889 int delta = 0; 2890 2891 if (headroom > skb_headroom(skb)) 2892 delta = headroom - skb_headroom(skb); 2893 2894 if (delta || cloned) 2895 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2896 GFP_ATOMIC); 2897 return 0; 2898} 2899 2900/** 2901 * skb_cow - copy header of skb when it is required 2902 * @skb: buffer to cow 2903 * @headroom: needed headroom 2904 * 2905 * If the skb passed lacks sufficient headroom or its data part 2906 * is shared, data is reallocated. If reallocation fails, an error 2907 * is returned and original skb is not changed. 2908 * 2909 * The result is skb with writable area skb->head...skb->tail 2910 * and at least @headroom of space at head. 2911 */ 2912static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2913{ 2914 return __skb_cow(skb, headroom, skb_cloned(skb)); 2915} 2916 2917/** 2918 * skb_cow_head - skb_cow but only making the head writable 2919 * @skb: buffer to cow 2920 * @headroom: needed headroom 2921 * 2922 * This function is identical to skb_cow except that we replace the 2923 * skb_cloned check by skb_header_cloned. It should be used when 2924 * you only need to push on some header and do not need to modify 2925 * the data. 2926 */ 2927static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2928{ 2929 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2930} 2931 2932/** 2933 * skb_padto - pad an skbuff up to a minimal size 2934 * @skb: buffer to pad 2935 * @len: minimal length 2936 * 2937 * Pads up a buffer to ensure the trailing bytes exist and are 2938 * blanked. If the buffer already contains sufficient data it 2939 * is untouched. Otherwise it is extended. Returns zero on 2940 * success. The skb is freed on error. 2941 */ 2942static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2943{ 2944 unsigned int size = skb->len; 2945 if (likely(size >= len)) 2946 return 0; 2947 return skb_pad(skb, len - size); 2948} 2949 2950/** 2951 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2952 * @skb: buffer to pad 2953 * @len: minimal length 2954 * @free_on_error: free buffer on error 2955 * 2956 * Pads up a buffer to ensure the trailing bytes exist and are 2957 * blanked. If the buffer already contains sufficient data it 2958 * is untouched. Otherwise it is extended. Returns zero on 2959 * success. The skb is freed on error if @free_on_error is true. 2960 */ 2961static inline int __skb_put_padto(struct sk_buff *skb, unsigned int len, 2962 bool free_on_error) 2963{ 2964 unsigned int size = skb->len; 2965 2966 if (unlikely(size < len)) { 2967 len -= size; 2968 if (__skb_pad(skb, len, free_on_error)) 2969 return -ENOMEM; 2970 __skb_put(skb, len); 2971 } 2972 return 0; 2973} 2974 2975/** 2976 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2977 * @skb: buffer to pad 2978 * @len: minimal length 2979 * 2980 * Pads up a buffer to ensure the trailing bytes exist and are 2981 * blanked. If the buffer already contains sufficient data it 2982 * is untouched. Otherwise it is extended. Returns zero on 2983 * success. The skb is freed on error. 2984 */ 2985static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2986{ 2987 return __skb_put_padto(skb, len, true); 2988} 2989 2990static inline int skb_add_data(struct sk_buff *skb, 2991 struct iov_iter *from, int copy) 2992{ 2993 const int off = skb->len; 2994 2995 if (skb->ip_summed == CHECKSUM_NONE) { 2996 __wsum csum = 0; 2997 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 2998 &csum, from)) { 2999 skb->csum = csum_block_add(skb->csum, csum, off); 3000 return 0; 3001 } 3002 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3003 return 0; 3004 3005 __skb_trim(skb, off); 3006 return -EFAULT; 3007} 3008 3009static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3010 const struct page *page, int off) 3011{ 3012 if (skb_zcopy(skb)) 3013 return false; 3014 if (i) { 3015 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 3016 3017 return page == skb_frag_page(frag) && 3018 off == frag->page_offset + skb_frag_size(frag); 3019 } 3020 return false; 3021} 3022 3023static inline int __skb_linearize(struct sk_buff *skb) 3024{ 3025 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3026} 3027 3028/** 3029 * skb_linearize - convert paged skb to linear one 3030 * @skb: buffer to linarize 3031 * 3032 * If there is no free memory -ENOMEM is returned, otherwise zero 3033 * is returned and the old skb data released. 3034 */ 3035static inline int skb_linearize(struct sk_buff *skb) 3036{ 3037 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3038} 3039 3040/** 3041 * skb_has_shared_frag - can any frag be overwritten 3042 * @skb: buffer to test 3043 * 3044 * Return true if the skb has at least one frag that might be modified 3045 * by an external entity (as in vmsplice()/sendfile()) 3046 */ 3047static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3048{ 3049 return skb_is_nonlinear(skb) && 3050 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 3051} 3052 3053/** 3054 * skb_linearize_cow - make sure skb is linear and writable 3055 * @skb: buffer to process 3056 * 3057 * If there is no free memory -ENOMEM is returned, otherwise zero 3058 * is returned and the old skb data released. 3059 */ 3060static inline int skb_linearize_cow(struct sk_buff *skb) 3061{ 3062 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3063 __skb_linearize(skb) : 0; 3064} 3065 3066static __always_inline void 3067__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3068 unsigned int off) 3069{ 3070 if (skb->ip_summed == CHECKSUM_COMPLETE) 3071 skb->csum = csum_block_sub(skb->csum, 3072 csum_partial(start, len, 0), off); 3073 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3074 skb_checksum_start_offset(skb) < 0) 3075 skb->ip_summed = CHECKSUM_NONE; 3076} 3077 3078/** 3079 * skb_postpull_rcsum - update checksum for received skb after pull 3080 * @skb: buffer to update 3081 * @start: start of data before pull 3082 * @len: length of data pulled 3083 * 3084 * After doing a pull on a received packet, you need to call this to 3085 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3086 * CHECKSUM_NONE so that it can be recomputed from scratch. 3087 */ 3088static inline void skb_postpull_rcsum(struct sk_buff *skb, 3089 const void *start, unsigned int len) 3090{ 3091 __skb_postpull_rcsum(skb, start, len, 0); 3092} 3093 3094static __always_inline void 3095__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3096 unsigned int off) 3097{ 3098 if (skb->ip_summed == CHECKSUM_COMPLETE) 3099 skb->csum = csum_block_add(skb->csum, 3100 csum_partial(start, len, 0), off); 3101} 3102 3103/** 3104 * skb_postpush_rcsum - update checksum for received skb after push 3105 * @skb: buffer to update 3106 * @start: start of data after push 3107 * @len: length of data pushed 3108 * 3109 * After doing a push on a received packet, you need to call this to 3110 * update the CHECKSUM_COMPLETE checksum. 3111 */ 3112static inline void skb_postpush_rcsum(struct sk_buff *skb, 3113 const void *start, unsigned int len) 3114{ 3115 __skb_postpush_rcsum(skb, start, len, 0); 3116} 3117 3118void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3119 3120/** 3121 * skb_push_rcsum - push skb and update receive checksum 3122 * @skb: buffer to update 3123 * @len: length of data pulled 3124 * 3125 * This function performs an skb_push on the packet and updates 3126 * the CHECKSUM_COMPLETE checksum. It should be used on 3127 * receive path processing instead of skb_push unless you know 3128 * that the checksum difference is zero (e.g., a valid IP header) 3129 * or you are setting ip_summed to CHECKSUM_NONE. 3130 */ 3131static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3132{ 3133 skb_push(skb, len); 3134 skb_postpush_rcsum(skb, skb->data, len); 3135 return skb->data; 3136} 3137 3138/** 3139 * pskb_trim_rcsum - trim received skb and update checksum 3140 * @skb: buffer to trim 3141 * @len: new length 3142 * 3143 * This is exactly the same as pskb_trim except that it ensures the 3144 * checksum of received packets are still valid after the operation. 3145 */ 3146 3147static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3148{ 3149 if (likely(len >= skb->len)) 3150 return 0; 3151 if (skb->ip_summed == CHECKSUM_COMPLETE) 3152 skb->ip_summed = CHECKSUM_NONE; 3153 return __pskb_trim(skb, len); 3154} 3155 3156static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3157{ 3158 if (skb->ip_summed == CHECKSUM_COMPLETE) 3159 skb->ip_summed = CHECKSUM_NONE; 3160 __skb_trim(skb, len); 3161 return 0; 3162} 3163 3164static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3165{ 3166 if (skb->ip_summed == CHECKSUM_COMPLETE) 3167 skb->ip_summed = CHECKSUM_NONE; 3168 return __skb_grow(skb, len); 3169} 3170 3171#define skb_queue_walk(queue, skb) \ 3172 for (skb = (queue)->next; \ 3173 skb != (struct sk_buff *)(queue); \ 3174 skb = skb->next) 3175 3176#define skb_queue_walk_safe(queue, skb, tmp) \ 3177 for (skb = (queue)->next, tmp = skb->next; \ 3178 skb != (struct sk_buff *)(queue); \ 3179 skb = tmp, tmp = skb->next) 3180 3181#define skb_queue_walk_from(queue, skb) \ 3182 for (; skb != (struct sk_buff *)(queue); \ 3183 skb = skb->next) 3184 3185#define skb_queue_walk_from_safe(queue, skb, tmp) \ 3186 for (tmp = skb->next; \ 3187 skb != (struct sk_buff *)(queue); \ 3188 skb = tmp, tmp = skb->next) 3189 3190#define skb_queue_reverse_walk(queue, skb) \ 3191 for (skb = (queue)->prev; \ 3192 skb != (struct sk_buff *)(queue); \ 3193 skb = skb->prev) 3194 3195#define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3196 for (skb = (queue)->prev, tmp = skb->prev; \ 3197 skb != (struct sk_buff *)(queue); \ 3198 skb = tmp, tmp = skb->prev) 3199 3200#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3201 for (tmp = skb->prev; \ 3202 skb != (struct sk_buff *)(queue); \ 3203 skb = tmp, tmp = skb->prev) 3204 3205static inline bool skb_has_frag_list(const struct sk_buff *skb) 3206{ 3207 return skb_shinfo(skb)->frag_list != NULL; 3208} 3209 3210static inline void skb_frag_list_init(struct sk_buff *skb) 3211{ 3212 skb_shinfo(skb)->frag_list = NULL; 3213} 3214 3215#define skb_walk_frags(skb, iter) \ 3216 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3217 3218 3219int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 3220 const struct sk_buff *skb); 3221struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3222 struct sk_buff_head *queue, 3223 unsigned int flags, 3224 void (*destructor)(struct sock *sk, 3225 struct sk_buff *skb), 3226 int *peeked, int *off, int *err, 3227 struct sk_buff **last); 3228struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 3229 void (*destructor)(struct sock *sk, 3230 struct sk_buff *skb), 3231 int *peeked, int *off, int *err, 3232 struct sk_buff **last); 3233struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 3234 void (*destructor)(struct sock *sk, 3235 struct sk_buff *skb), 3236 int *peeked, int *off, int *err); 3237struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 3238 int *err); 3239unsigned int datagram_poll(struct file *file, struct socket *sock, 3240 struct poll_table_struct *wait); 3241int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3242 struct iov_iter *to, int size); 3243static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3244 struct msghdr *msg, int size) 3245{ 3246 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3247} 3248int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3249 struct msghdr *msg); 3250int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3251 struct iov_iter *from, int len); 3252int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3253void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3254void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3255static inline void skb_free_datagram_locked(struct sock *sk, 3256 struct sk_buff *skb) 3257{ 3258 __skb_free_datagram_locked(sk, skb, 0); 3259} 3260int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3261int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3262int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3263__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3264 int len, __wsum csum); 3265int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3266 struct pipe_inode_info *pipe, unsigned int len, 3267 unsigned int flags); 3268int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3269 int len); 3270int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 3271void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3272unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3273int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3274 int len, int hlen); 3275void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3276int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3277void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3278unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 3279bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu); 3280struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3281struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3282int skb_ensure_writable(struct sk_buff *skb, int write_len); 3283int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3284int skb_vlan_pop(struct sk_buff *skb); 3285int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3286struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3287 gfp_t gfp); 3288 3289static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3290{ 3291 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 3292} 3293 3294static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3295{ 3296 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3297} 3298 3299struct skb_checksum_ops { 3300 __wsum (*update)(const void *mem, int len, __wsum wsum); 3301 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3302}; 3303 3304extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 3305 3306__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3307 __wsum csum, const struct skb_checksum_ops *ops); 3308__wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3309 __wsum csum); 3310 3311static inline void * __must_check 3312__skb_header_pointer(const struct sk_buff *skb, int offset, 3313 int len, void *data, int hlen, void *buffer) 3314{ 3315 if (hlen - offset >= len) 3316 return data + offset; 3317 3318 if (!skb || 3319 skb_copy_bits(skb, offset, buffer, len) < 0) 3320 return NULL; 3321 3322 return buffer; 3323} 3324 3325static inline void * __must_check 3326skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3327{ 3328 return __skb_header_pointer(skb, offset, len, skb->data, 3329 skb_headlen(skb), buffer); 3330} 3331 3332/** 3333 * skb_needs_linearize - check if we need to linearize a given skb 3334 * depending on the given device features. 3335 * @skb: socket buffer to check 3336 * @features: net device features 3337 * 3338 * Returns true if either: 3339 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3340 * 2. skb is fragmented and the device does not support SG. 3341 */ 3342static inline bool skb_needs_linearize(struct sk_buff *skb, 3343 netdev_features_t features) 3344{ 3345 return skb_is_nonlinear(skb) && 3346 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3347 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3348} 3349 3350static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3351 void *to, 3352 const unsigned int len) 3353{ 3354 memcpy(to, skb->data, len); 3355} 3356 3357static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3358 const int offset, void *to, 3359 const unsigned int len) 3360{ 3361 memcpy(to, skb->data + offset, len); 3362} 3363 3364static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3365 const void *from, 3366 const unsigned int len) 3367{ 3368 memcpy(skb->data, from, len); 3369} 3370 3371static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3372 const int offset, 3373 const void *from, 3374 const unsigned int len) 3375{ 3376 memcpy(skb->data + offset, from, len); 3377} 3378 3379void skb_init(void); 3380 3381static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3382{ 3383 return skb->tstamp; 3384} 3385 3386/** 3387 * skb_get_timestamp - get timestamp from a skb 3388 * @skb: skb to get stamp from 3389 * @stamp: pointer to struct timeval to store stamp in 3390 * 3391 * Timestamps are stored in the skb as offsets to a base timestamp. 3392 * This function converts the offset back to a struct timeval and stores 3393 * it in stamp. 3394 */ 3395static inline void skb_get_timestamp(const struct sk_buff *skb, 3396 struct timeval *stamp) 3397{ 3398 *stamp = ktime_to_timeval(skb->tstamp); 3399} 3400 3401static inline void skb_get_timestampns(const struct sk_buff *skb, 3402 struct timespec *stamp) 3403{ 3404 *stamp = ktime_to_timespec(skb->tstamp); 3405} 3406 3407static inline void __net_timestamp(struct sk_buff *skb) 3408{ 3409 skb->tstamp = ktime_get_real(); 3410} 3411 3412static inline ktime_t net_timedelta(ktime_t t) 3413{ 3414 return ktime_sub(ktime_get_real(), t); 3415} 3416 3417static inline ktime_t net_invalid_timestamp(void) 3418{ 3419 return 0; 3420} 3421 3422struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3423 3424#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3425 3426void skb_clone_tx_timestamp(struct sk_buff *skb); 3427bool skb_defer_rx_timestamp(struct sk_buff *skb); 3428 3429#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3430 3431static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3432{ 3433} 3434 3435static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3436{ 3437 return false; 3438} 3439 3440#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3441 3442/** 3443 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3444 * 3445 * PHY drivers may accept clones of transmitted packets for 3446 * timestamping via their phy_driver.txtstamp method. These drivers 3447 * must call this function to return the skb back to the stack with a 3448 * timestamp. 3449 * 3450 * @skb: clone of the the original outgoing packet 3451 * @hwtstamps: hardware time stamps 3452 * 3453 */ 3454void skb_complete_tx_timestamp(struct sk_buff *skb, 3455 struct skb_shared_hwtstamps *hwtstamps); 3456 3457void __skb_tstamp_tx(struct sk_buff *orig_skb, 3458 struct skb_shared_hwtstamps *hwtstamps, 3459 struct sock *sk, int tstype); 3460 3461/** 3462 * skb_tstamp_tx - queue clone of skb with send time stamps 3463 * @orig_skb: the original outgoing packet 3464 * @hwtstamps: hardware time stamps, may be NULL if not available 3465 * 3466 * If the skb has a socket associated, then this function clones the 3467 * skb (thus sharing the actual data and optional structures), stores 3468 * the optional hardware time stamping information (if non NULL) or 3469 * generates a software time stamp (otherwise), then queues the clone 3470 * to the error queue of the socket. Errors are silently ignored. 3471 */ 3472void skb_tstamp_tx(struct sk_buff *orig_skb, 3473 struct skb_shared_hwtstamps *hwtstamps); 3474 3475/** 3476 * skb_tx_timestamp() - Driver hook for transmit timestamping 3477 * 3478 * Ethernet MAC Drivers should call this function in their hard_xmit() 3479 * function immediately before giving the sk_buff to the MAC hardware. 3480 * 3481 * Specifically, one should make absolutely sure that this function is 3482 * called before TX completion of this packet can trigger. Otherwise 3483 * the packet could potentially already be freed. 3484 * 3485 * @skb: A socket buffer. 3486 */ 3487static inline void skb_tx_timestamp(struct sk_buff *skb) 3488{ 3489 skb_clone_tx_timestamp(skb); 3490 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 3491 skb_tstamp_tx(skb, NULL); 3492} 3493 3494/** 3495 * skb_complete_wifi_ack - deliver skb with wifi status 3496 * 3497 * @skb: the original outgoing packet 3498 * @acked: ack status 3499 * 3500 */ 3501void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3502 3503__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3504__sum16 __skb_checksum_complete(struct sk_buff *skb); 3505 3506static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3507{ 3508 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3509 skb->csum_valid || 3510 (skb->ip_summed == CHECKSUM_PARTIAL && 3511 skb_checksum_start_offset(skb) >= 0)); 3512} 3513 3514/** 3515 * skb_checksum_complete - Calculate checksum of an entire packet 3516 * @skb: packet to process 3517 * 3518 * This function calculates the checksum over the entire packet plus 3519 * the value of skb->csum. The latter can be used to supply the 3520 * checksum of a pseudo header as used by TCP/UDP. It returns the 3521 * checksum. 3522 * 3523 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3524 * this function can be used to verify that checksum on received 3525 * packets. In that case the function should return zero if the 3526 * checksum is correct. In particular, this function will return zero 3527 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3528 * hardware has already verified the correctness of the checksum. 3529 */ 3530static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3531{ 3532 return skb_csum_unnecessary(skb) ? 3533 0 : __skb_checksum_complete(skb); 3534} 3535 3536static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3537{ 3538 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3539 if (skb->csum_level == 0) 3540 skb->ip_summed = CHECKSUM_NONE; 3541 else 3542 skb->csum_level--; 3543 } 3544} 3545 3546static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3547{ 3548 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3549 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3550 skb->csum_level++; 3551 } else if (skb->ip_summed == CHECKSUM_NONE) { 3552 skb->ip_summed = CHECKSUM_UNNECESSARY; 3553 skb->csum_level = 0; 3554 } 3555} 3556 3557/* Check if we need to perform checksum complete validation. 3558 * 3559 * Returns true if checksum complete is needed, false otherwise 3560 * (either checksum is unnecessary or zero checksum is allowed). 3561 */ 3562static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3563 bool zero_okay, 3564 __sum16 check) 3565{ 3566 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3567 skb->csum_valid = 1; 3568 __skb_decr_checksum_unnecessary(skb); 3569 return false; 3570 } 3571 3572 return true; 3573} 3574 3575/* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3576 * in checksum_init. 3577 */ 3578#define CHECKSUM_BREAK 76 3579 3580/* Unset checksum-complete 3581 * 3582 * Unset checksum complete can be done when packet is being modified 3583 * (uncompressed for instance) and checksum-complete value is 3584 * invalidated. 3585 */ 3586static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3587{ 3588 if (skb->ip_summed == CHECKSUM_COMPLETE) 3589 skb->ip_summed = CHECKSUM_NONE; 3590} 3591 3592/* Validate (init) checksum based on checksum complete. 3593 * 3594 * Return values: 3595 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3596 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3597 * checksum is stored in skb->csum for use in __skb_checksum_complete 3598 * non-zero: value of invalid checksum 3599 * 3600 */ 3601static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3602 bool complete, 3603 __wsum psum) 3604{ 3605 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3606 if (!csum_fold(csum_add(psum, skb->csum))) { 3607 skb->csum_valid = 1; 3608 return 0; 3609 } 3610 } 3611 3612 skb->csum = psum; 3613 3614 if (complete || skb->len <= CHECKSUM_BREAK) { 3615 __sum16 csum; 3616 3617 csum = __skb_checksum_complete(skb); 3618 skb->csum_valid = !csum; 3619 return csum; 3620 } 3621 3622 return 0; 3623} 3624 3625static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3626{ 3627 return 0; 3628} 3629 3630/* Perform checksum validate (init). Note that this is a macro since we only 3631 * want to calculate the pseudo header which is an input function if necessary. 3632 * First we try to validate without any computation (checksum unnecessary) and 3633 * then calculate based on checksum complete calling the function to compute 3634 * pseudo header. 3635 * 3636 * Return values: 3637 * 0: checksum is validated or try to in skb_checksum_complete 3638 * non-zero: value of invalid checksum 3639 */ 3640#define __skb_checksum_validate(skb, proto, complete, \ 3641 zero_okay, check, compute_pseudo) \ 3642({ \ 3643 __sum16 __ret = 0; \ 3644 skb->csum_valid = 0; \ 3645 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3646 __ret = __skb_checksum_validate_complete(skb, \ 3647 complete, compute_pseudo(skb, proto)); \ 3648 __ret; \ 3649}) 3650 3651#define skb_checksum_init(skb, proto, compute_pseudo) \ 3652 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3653 3654#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3655 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3656 3657#define skb_checksum_validate(skb, proto, compute_pseudo) \ 3658 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3659 3660#define skb_checksum_validate_zero_check(skb, proto, check, \ 3661 compute_pseudo) \ 3662 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3663 3664#define skb_checksum_simple_validate(skb) \ 3665 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3666 3667static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3668{ 3669 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 3670} 3671 3672static inline void __skb_checksum_convert(struct sk_buff *skb, 3673 __sum16 check, __wsum pseudo) 3674{ 3675 skb->csum = ~pseudo; 3676 skb->ip_summed = CHECKSUM_COMPLETE; 3677} 3678 3679#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3680do { \ 3681 if (__skb_checksum_convert_check(skb)) \ 3682 __skb_checksum_convert(skb, check, \ 3683 compute_pseudo(skb, proto)); \ 3684} while (0) 3685 3686static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3687 u16 start, u16 offset) 3688{ 3689 skb->ip_summed = CHECKSUM_PARTIAL; 3690 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3691 skb->csum_offset = offset - start; 3692} 3693 3694/* Update skbuf and packet to reflect the remote checksum offload operation. 3695 * When called, ptr indicates the starting point for skb->csum when 3696 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3697 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3698 */ 3699static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3700 int start, int offset, bool nopartial) 3701{ 3702 __wsum delta; 3703 3704 if (!nopartial) { 3705 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3706 return; 3707 } 3708 3709 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3710 __skb_checksum_complete(skb); 3711 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3712 } 3713 3714 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3715 3716 /* Adjust skb->csum since we changed the packet */ 3717 skb->csum = csum_add(skb->csum, delta); 3718} 3719 3720static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 3721{ 3722#if IS_ENABLED(CONFIG_NF_CONNTRACK) 3723 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK); 3724#else 3725 return NULL; 3726#endif 3727} 3728 3729#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3730void nf_conntrack_destroy(struct nf_conntrack *nfct); 3731static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3732{ 3733 if (nfct && atomic_dec_and_test(&nfct->use)) 3734 nf_conntrack_destroy(nfct); 3735} 3736static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3737{ 3738 if (nfct) 3739 atomic_inc(&nfct->use); 3740} 3741#endif 3742#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3743static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3744{ 3745 if (nf_bridge && refcount_dec_and_test(&nf_bridge->use)) 3746 kfree(nf_bridge); 3747} 3748static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3749{ 3750 if (nf_bridge) 3751 refcount_inc(&nf_bridge->use); 3752} 3753#endif /* CONFIG_BRIDGE_NETFILTER */ 3754static inline void nf_reset(struct sk_buff *skb) 3755{ 3756#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3757 nf_conntrack_put(skb_nfct(skb)); 3758 skb->_nfct = 0; 3759#endif 3760#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3761 nf_bridge_put(skb->nf_bridge); 3762 skb->nf_bridge = NULL; 3763#endif 3764} 3765 3766static inline void nf_reset_trace(struct sk_buff *skb) 3767{ 3768#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3769 skb->nf_trace = 0; 3770#endif 3771} 3772 3773static inline void ipvs_reset(struct sk_buff *skb) 3774{ 3775#if IS_ENABLED(CONFIG_IP_VS) 3776 skb->ipvs_property = 0; 3777#endif 3778} 3779 3780/* Note: This doesn't put any conntrack and bridge info in dst. */ 3781static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3782 bool copy) 3783{ 3784#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3785 dst->_nfct = src->_nfct; 3786 nf_conntrack_get(skb_nfct(src)); 3787#endif 3788#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3789 dst->nf_bridge = src->nf_bridge; 3790 nf_bridge_get(src->nf_bridge); 3791#endif 3792#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3793 if (copy) 3794 dst->nf_trace = src->nf_trace; 3795#endif 3796} 3797 3798static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3799{ 3800#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3801 nf_conntrack_put(skb_nfct(dst)); 3802#endif 3803#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3804 nf_bridge_put(dst->nf_bridge); 3805#endif 3806 __nf_copy(dst, src, true); 3807} 3808 3809#ifdef CONFIG_NETWORK_SECMARK 3810static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3811{ 3812 to->secmark = from->secmark; 3813} 3814 3815static inline void skb_init_secmark(struct sk_buff *skb) 3816{ 3817 skb->secmark = 0; 3818} 3819#else 3820static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3821{ } 3822 3823static inline void skb_init_secmark(struct sk_buff *skb) 3824{ } 3825#endif 3826 3827static inline bool skb_irq_freeable(const struct sk_buff *skb) 3828{ 3829 return !skb->destructor && 3830#if IS_ENABLED(CONFIG_XFRM) 3831 !skb->sp && 3832#endif 3833 !skb_nfct(skb) && 3834 !skb->_skb_refdst && 3835 !skb_has_frag_list(skb); 3836} 3837 3838static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3839{ 3840 skb->queue_mapping = queue_mapping; 3841} 3842 3843static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3844{ 3845 return skb->queue_mapping; 3846} 3847 3848static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3849{ 3850 to->queue_mapping = from->queue_mapping; 3851} 3852 3853static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3854{ 3855 skb->queue_mapping = rx_queue + 1; 3856} 3857 3858static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3859{ 3860 return skb->queue_mapping - 1; 3861} 3862 3863static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3864{ 3865 return skb->queue_mapping != 0; 3866} 3867 3868static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 3869{ 3870 skb->dst_pending_confirm = val; 3871} 3872 3873static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 3874{ 3875 return skb->dst_pending_confirm != 0; 3876} 3877 3878static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3879{ 3880#ifdef CONFIG_XFRM 3881 return skb->sp; 3882#else 3883 return NULL; 3884#endif 3885} 3886 3887/* Keeps track of mac header offset relative to skb->head. 3888 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3889 * For non-tunnel skb it points to skb_mac_header() and for 3890 * tunnel skb it points to outer mac header. 3891 * Keeps track of level of encapsulation of network headers. 3892 */ 3893struct skb_gso_cb { 3894 union { 3895 int mac_offset; 3896 int data_offset; 3897 }; 3898 int encap_level; 3899 __wsum csum; 3900 __u16 csum_start; 3901}; 3902#define SKB_SGO_CB_OFFSET 32 3903#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3904 3905static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3906{ 3907 return (skb_mac_header(inner_skb) - inner_skb->head) - 3908 SKB_GSO_CB(inner_skb)->mac_offset; 3909} 3910 3911static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3912{ 3913 int new_headroom, headroom; 3914 int ret; 3915 3916 headroom = skb_headroom(skb); 3917 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3918 if (ret) 3919 return ret; 3920 3921 new_headroom = skb_headroom(skb); 3922 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3923 return 0; 3924} 3925 3926static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 3927{ 3928 /* Do not update partial checksums if remote checksum is enabled. */ 3929 if (skb->remcsum_offload) 3930 return; 3931 3932 SKB_GSO_CB(skb)->csum = res; 3933 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 3934} 3935 3936/* Compute the checksum for a gso segment. First compute the checksum value 3937 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3938 * then add in skb->csum (checksum from csum_start to end of packet). 3939 * skb->csum and csum_start are then updated to reflect the checksum of the 3940 * resultant packet starting from the transport header-- the resultant checksum 3941 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3942 * header. 3943 */ 3944static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3945{ 3946 unsigned char *csum_start = skb_transport_header(skb); 3947 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 3948 __wsum partial = SKB_GSO_CB(skb)->csum; 3949 3950 SKB_GSO_CB(skb)->csum = res; 3951 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 3952 3953 return csum_fold(csum_partial(csum_start, plen, partial)); 3954} 3955 3956static inline bool skb_is_gso(const struct sk_buff *skb) 3957{ 3958 return skb_shinfo(skb)->gso_size; 3959} 3960 3961/* Note: Should be called only if skb_is_gso(skb) is true */ 3962static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3963{ 3964 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3965} 3966 3967static inline void skb_gso_reset(struct sk_buff *skb) 3968{ 3969 skb_shinfo(skb)->gso_size = 0; 3970 skb_shinfo(skb)->gso_segs = 0; 3971 skb_shinfo(skb)->gso_type = 0; 3972} 3973 3974void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3975 3976static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3977{ 3978 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3979 * wanted then gso_type will be set. */ 3980 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3981 3982 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3983 unlikely(shinfo->gso_type == 0)) { 3984 __skb_warn_lro_forwarding(skb); 3985 return true; 3986 } 3987 return false; 3988} 3989 3990static inline void skb_forward_csum(struct sk_buff *skb) 3991{ 3992 /* Unfortunately we don't support this one. Any brave souls? */ 3993 if (skb->ip_summed == CHECKSUM_COMPLETE) 3994 skb->ip_summed = CHECKSUM_NONE; 3995} 3996 3997/** 3998 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3999 * @skb: skb to check 4000 * 4001 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4002 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4003 * use this helper, to document places where we make this assertion. 4004 */ 4005static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4006{ 4007#ifdef DEBUG 4008 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 4009#endif 4010} 4011 4012bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4013 4014int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4015struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4016 unsigned int transport_len, 4017 __sum16(*skb_chkf)(struct sk_buff *skb)); 4018 4019/** 4020 * skb_head_is_locked - Determine if the skb->head is locked down 4021 * @skb: skb to check 4022 * 4023 * The head on skbs build around a head frag can be removed if they are 4024 * not cloned. This function returns true if the skb head is locked down 4025 * due to either being allocated via kmalloc, or by being a clone with 4026 * multiple references to the head. 4027 */ 4028static inline bool skb_head_is_locked(const struct sk_buff *skb) 4029{ 4030 return !skb->head_frag || skb_cloned(skb); 4031} 4032 4033/** 4034 * skb_gso_network_seglen - Return length of individual segments of a gso packet 4035 * 4036 * @skb: GSO skb 4037 * 4038 * skb_gso_network_seglen is used to determine the real size of the 4039 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 4040 * 4041 * The MAC/L2 header is not accounted for. 4042 */ 4043static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 4044{ 4045 unsigned int hdr_len = skb_transport_header(skb) - 4046 skb_network_header(skb); 4047 return hdr_len + skb_gso_transport_seglen(skb); 4048} 4049 4050/* Local Checksum Offload. 4051 * Compute outer checksum based on the assumption that the 4052 * inner checksum will be offloaded later. 4053 * See Documentation/networking/checksum-offloads.txt for 4054 * explanation of how this works. 4055 * Fill in outer checksum adjustment (e.g. with sum of outer 4056 * pseudo-header) before calling. 4057 * Also ensure that inner checksum is in linear data area. 4058 */ 4059static inline __wsum lco_csum(struct sk_buff *skb) 4060{ 4061 unsigned char *csum_start = skb_checksum_start(skb); 4062 unsigned char *l4_hdr = skb_transport_header(skb); 4063 __wsum partial; 4064 4065 /* Start with complement of inner checksum adjustment */ 4066 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 4067 skb->csum_offset)); 4068 4069 /* Add in checksum of our headers (incl. outer checksum 4070 * adjustment filled in by caller) and return result. 4071 */ 4072 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 4073} 4074 4075#endif /* __KERNEL__ */ 4076#endif /* _LINUX_SKBUFF_H */