at v4.9 114 kB view raw
1/* 2 * Definitions for the 'struct sk_buff' memory handlers. 3 * 4 * Authors: 5 * Alan Cox, <gw4pts@gw4pts.ampr.org> 6 * Florian La Roche, <rzsfl@rz.uni-sb.de> 7 * 8 * This program is free software; you can redistribute it and/or 9 * modify it under the terms of the GNU General Public License 10 * as published by the Free Software Foundation; either version 11 * 2 of the License, or (at your option) any later version. 12 */ 13 14#ifndef _LINUX_SKBUFF_H 15#define _LINUX_SKBUFF_H 16 17#include <linux/kernel.h> 18#include <linux/kmemcheck.h> 19#include <linux/compiler.h> 20#include <linux/time.h> 21#include <linux/bug.h> 22#include <linux/cache.h> 23#include <linux/rbtree.h> 24#include <linux/socket.h> 25 26#include <linux/atomic.h> 27#include <asm/types.h> 28#include <linux/spinlock.h> 29#include <linux/net.h> 30#include <linux/textsearch.h> 31#include <net/checksum.h> 32#include <linux/rcupdate.h> 33#include <linux/hrtimer.h> 34#include <linux/dma-mapping.h> 35#include <linux/netdev_features.h> 36#include <linux/sched.h> 37#include <net/flow_dissector.h> 38#include <linux/splice.h> 39#include <linux/in6.h> 40#include <linux/if_packet.h> 41#include <net/flow.h> 42 43/* The interface for checksum offload between the stack and networking drivers 44 * is as follows... 45 * 46 * A. IP checksum related features 47 * 48 * Drivers advertise checksum offload capabilities in the features of a device. 49 * From the stack's point of view these are capabilities offered by the driver, 50 * a driver typically only advertises features that it is capable of offloading 51 * to its device. 52 * 53 * The checksum related features are: 54 * 55 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one 56 * IP (one's complement) checksum for any combination 57 * of protocols or protocol layering. The checksum is 58 * computed and set in a packet per the CHECKSUM_PARTIAL 59 * interface (see below). 60 * 61 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain 62 * TCP or UDP packets over IPv4. These are specifically 63 * unencapsulated packets of the form IPv4|TCP or 64 * IPv4|UDP where the Protocol field in the IPv4 header 65 * is TCP or UDP. The IPv4 header may contain IP options 66 * This feature cannot be set in features for a device 67 * with NETIF_F_HW_CSUM also set. This feature is being 68 * DEPRECATED (see below). 69 * 70 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain 71 * TCP or UDP packets over IPv6. These are specifically 72 * unencapsulated packets of the form IPv6|TCP or 73 * IPv4|UDP where the Next Header field in the IPv6 74 * header is either TCP or UDP. IPv6 extension headers 75 * are not supported with this feature. This feature 76 * cannot be set in features for a device with 77 * NETIF_F_HW_CSUM also set. This feature is being 78 * DEPRECATED (see below). 79 * 80 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload. 81 * This flag is used only used to disable the RX checksum 82 * feature for a device. The stack will accept receive 83 * checksum indication in packets received on a device 84 * regardless of whether NETIF_F_RXCSUM is set. 85 * 86 * B. Checksumming of received packets by device. Indication of checksum 87 * verification is in set skb->ip_summed. Possible values are: 88 * 89 * CHECKSUM_NONE: 90 * 91 * Device did not checksum this packet e.g. due to lack of capabilities. 92 * The packet contains full (though not verified) checksum in packet but 93 * not in skb->csum. Thus, skb->csum is undefined in this case. 94 * 95 * CHECKSUM_UNNECESSARY: 96 * 97 * The hardware you're dealing with doesn't calculate the full checksum 98 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 99 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 100 * if their checksums are okay. skb->csum is still undefined in this case 101 * though. A driver or device must never modify the checksum field in the 102 * packet even if checksum is verified. 103 * 104 * CHECKSUM_UNNECESSARY is applicable to following protocols: 105 * TCP: IPv6 and IPv4. 106 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 107 * zero UDP checksum for either IPv4 or IPv6, the networking stack 108 * may perform further validation in this case. 109 * GRE: only if the checksum is present in the header. 110 * SCTP: indicates the CRC in SCTP header has been validated. 111 * 112 * skb->csum_level indicates the number of consecutive checksums found in 113 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 114 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 115 * and a device is able to verify the checksums for UDP (possibly zero), 116 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 117 * two. If the device were only able to verify the UDP checksum and not 118 * GRE, either because it doesn't support GRE checksum of because GRE 119 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 120 * not considered in this case). 121 * 122 * CHECKSUM_COMPLETE: 123 * 124 * This is the most generic way. The device supplied checksum of the _whole_ 125 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 126 * hardware doesn't need to parse L3/L4 headers to implement this. 127 * 128 * Note: Even if device supports only some protocols, but is able to produce 129 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 130 * 131 * CHECKSUM_PARTIAL: 132 * 133 * A checksum is set up to be offloaded to a device as described in the 134 * output description for CHECKSUM_PARTIAL. This may occur on a packet 135 * received directly from another Linux OS, e.g., a virtualized Linux kernel 136 * on the same host, or it may be set in the input path in GRO or remote 137 * checksum offload. For the purposes of checksum verification, the checksum 138 * referred to by skb->csum_start + skb->csum_offset and any preceding 139 * checksums in the packet are considered verified. Any checksums in the 140 * packet that are after the checksum being offloaded are not considered to 141 * be verified. 142 * 143 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload 144 * in the skb->ip_summed for a packet. Values are: 145 * 146 * CHECKSUM_PARTIAL: 147 * 148 * The driver is required to checksum the packet as seen by hard_start_xmit() 149 * from skb->csum_start up to the end, and to record/write the checksum at 150 * offset skb->csum_start + skb->csum_offset. A driver may verify that the 151 * csum_start and csum_offset values are valid values given the length and 152 * offset of the packet, however they should not attempt to validate that the 153 * checksum refers to a legitimate transport layer checksum-- it is the 154 * purview of the stack to validate that csum_start and csum_offset are set 155 * correctly. 156 * 157 * When the stack requests checksum offload for a packet, the driver MUST 158 * ensure that the checksum is set correctly. A driver can either offload the 159 * checksum calculation to the device, or call skb_checksum_help (in the case 160 * that the device does not support offload for a particular checksum). 161 * 162 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of 163 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate 164 * checksum offload capability. If a device has limited checksum capabilities 165 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as 166 * described above) a helper function can be called to resolve 167 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper 168 * function takes a spec argument that describes the protocol layer that is 169 * supported for checksum offload and can be called for each packet. If a 170 * packet does not match the specification for offload, skb_checksum_help 171 * is called to resolve the checksum. 172 * 173 * CHECKSUM_NONE: 174 * 175 * The skb was already checksummed by the protocol, or a checksum is not 176 * required. 177 * 178 * CHECKSUM_UNNECESSARY: 179 * 180 * This has the same meaning on as CHECKSUM_NONE for checksum offload on 181 * output. 182 * 183 * CHECKSUM_COMPLETE: 184 * Not used in checksum output. If a driver observes a packet with this value 185 * set in skbuff, if should treat as CHECKSUM_NONE being set. 186 * 187 * D. Non-IP checksum (CRC) offloads 188 * 189 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of 190 * offloading the SCTP CRC in a packet. To perform this offload the stack 191 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 192 * accordingly. Note the there is no indication in the skbuff that the 193 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports 194 * both IP checksum offload and SCTP CRC offload must verify which offload 195 * is configured for a packet presumably by inspecting packet headers. 196 * 197 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of 198 * offloading the FCOE CRC in a packet. To perform this offload the stack 199 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 200 * accordingly. Note the there is no indication in the skbuff that the 201 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports 202 * both IP checksum offload and FCOE CRC offload must verify which offload 203 * is configured for a packet presumably by inspecting packet headers. 204 * 205 * E. Checksumming on output with GSO. 206 * 207 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload 208 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 209 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as 210 * part of the GSO operation is implied. If a checksum is being offloaded 211 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset 212 * are set to refer to the outermost checksum being offload (two offloaded 213 * checksums are possible with UDP encapsulation). 214 */ 215 216/* Don't change this without changing skb_csum_unnecessary! */ 217#define CHECKSUM_NONE 0 218#define CHECKSUM_UNNECESSARY 1 219#define CHECKSUM_COMPLETE 2 220#define CHECKSUM_PARTIAL 3 221 222/* Maximum value in skb->csum_level */ 223#define SKB_MAX_CSUM_LEVEL 3 224 225#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 226#define SKB_WITH_OVERHEAD(X) \ 227 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 228#define SKB_MAX_ORDER(X, ORDER) \ 229 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 230#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 231#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 232 233/* return minimum truesize of one skb containing X bytes of data */ 234#define SKB_TRUESIZE(X) ((X) + \ 235 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 236 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 237 238struct net_device; 239struct scatterlist; 240struct pipe_inode_info; 241struct iov_iter; 242struct napi_struct; 243 244#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 245struct nf_conntrack { 246 atomic_t use; 247}; 248#endif 249 250#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 251struct nf_bridge_info { 252 atomic_t use; 253 enum { 254 BRNF_PROTO_UNCHANGED, 255 BRNF_PROTO_8021Q, 256 BRNF_PROTO_PPPOE 257 } orig_proto:8; 258 u8 pkt_otherhost:1; 259 u8 in_prerouting:1; 260 u8 bridged_dnat:1; 261 __u16 frag_max_size; 262 struct net_device *physindev; 263 264 /* always valid & non-NULL from FORWARD on, for physdev match */ 265 struct net_device *physoutdev; 266 union { 267 /* prerouting: detect dnat in orig/reply direction */ 268 __be32 ipv4_daddr; 269 struct in6_addr ipv6_daddr; 270 271 /* after prerouting + nat detected: store original source 272 * mac since neigh resolution overwrites it, only used while 273 * skb is out in neigh layer. 274 */ 275 char neigh_header[8]; 276 }; 277}; 278#endif 279 280struct sk_buff_head { 281 /* These two members must be first. */ 282 struct sk_buff *next; 283 struct sk_buff *prev; 284 285 __u32 qlen; 286 spinlock_t lock; 287}; 288 289struct sk_buff; 290 291/* To allow 64K frame to be packed as single skb without frag_list we 292 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 293 * buffers which do not start on a page boundary. 294 * 295 * Since GRO uses frags we allocate at least 16 regardless of page 296 * size. 297 */ 298#if (65536/PAGE_SIZE + 1) < 16 299#define MAX_SKB_FRAGS 16UL 300#else 301#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 302#endif 303extern int sysctl_max_skb_frags; 304 305/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 306 * segment using its current segmentation instead. 307 */ 308#define GSO_BY_FRAGS 0xFFFF 309 310typedef struct skb_frag_struct skb_frag_t; 311 312struct skb_frag_struct { 313 struct { 314 struct page *p; 315 } page; 316#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 317 __u32 page_offset; 318 __u32 size; 319#else 320 __u16 page_offset; 321 __u16 size; 322#endif 323}; 324 325static inline unsigned int skb_frag_size(const skb_frag_t *frag) 326{ 327 return frag->size; 328} 329 330static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 331{ 332 frag->size = size; 333} 334 335static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 336{ 337 frag->size += delta; 338} 339 340static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 341{ 342 frag->size -= delta; 343} 344 345#define HAVE_HW_TIME_STAMP 346 347/** 348 * struct skb_shared_hwtstamps - hardware time stamps 349 * @hwtstamp: hardware time stamp transformed into duration 350 * since arbitrary point in time 351 * 352 * Software time stamps generated by ktime_get_real() are stored in 353 * skb->tstamp. 354 * 355 * hwtstamps can only be compared against other hwtstamps from 356 * the same device. 357 * 358 * This structure is attached to packets as part of the 359 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 360 */ 361struct skb_shared_hwtstamps { 362 ktime_t hwtstamp; 363}; 364 365/* Definitions for tx_flags in struct skb_shared_info */ 366enum { 367 /* generate hardware time stamp */ 368 SKBTX_HW_TSTAMP = 1 << 0, 369 370 /* generate software time stamp when queueing packet to NIC */ 371 SKBTX_SW_TSTAMP = 1 << 1, 372 373 /* device driver is going to provide hardware time stamp */ 374 SKBTX_IN_PROGRESS = 1 << 2, 375 376 /* device driver supports TX zero-copy buffers */ 377 SKBTX_DEV_ZEROCOPY = 1 << 3, 378 379 /* generate wifi status information (where possible) */ 380 SKBTX_WIFI_STATUS = 1 << 4, 381 382 /* This indicates at least one fragment might be overwritten 383 * (as in vmsplice(), sendfile() ...) 384 * If we need to compute a TX checksum, we'll need to copy 385 * all frags to avoid possible bad checksum 386 */ 387 SKBTX_SHARED_FRAG = 1 << 5, 388 389 /* generate software time stamp when entering packet scheduling */ 390 SKBTX_SCHED_TSTAMP = 1 << 6, 391}; 392 393#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 394 SKBTX_SCHED_TSTAMP) 395#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 396 397/* 398 * The callback notifies userspace to release buffers when skb DMA is done in 399 * lower device, the skb last reference should be 0 when calling this. 400 * The zerocopy_success argument is true if zero copy transmit occurred, 401 * false on data copy or out of memory error caused by data copy attempt. 402 * The ctx field is used to track device context. 403 * The desc field is used to track userspace buffer index. 404 */ 405struct ubuf_info { 406 void (*callback)(struct ubuf_info *, bool zerocopy_success); 407 void *ctx; 408 unsigned long desc; 409}; 410 411/* This data is invariant across clones and lives at 412 * the end of the header data, ie. at skb->end. 413 */ 414struct skb_shared_info { 415 unsigned char nr_frags; 416 __u8 tx_flags; 417 unsigned short gso_size; 418 /* Warning: this field is not always filled in (UFO)! */ 419 unsigned short gso_segs; 420 unsigned short gso_type; 421 struct sk_buff *frag_list; 422 struct skb_shared_hwtstamps hwtstamps; 423 u32 tskey; 424 __be32 ip6_frag_id; 425 426 /* 427 * Warning : all fields before dataref are cleared in __alloc_skb() 428 */ 429 atomic_t dataref; 430 431 /* Intermediate layers must ensure that destructor_arg 432 * remains valid until skb destructor */ 433 void * destructor_arg; 434 435 /* must be last field, see pskb_expand_head() */ 436 skb_frag_t frags[MAX_SKB_FRAGS]; 437}; 438 439/* We divide dataref into two halves. The higher 16 bits hold references 440 * to the payload part of skb->data. The lower 16 bits hold references to 441 * the entire skb->data. A clone of a headerless skb holds the length of 442 * the header in skb->hdr_len. 443 * 444 * All users must obey the rule that the skb->data reference count must be 445 * greater than or equal to the payload reference count. 446 * 447 * Holding a reference to the payload part means that the user does not 448 * care about modifications to the header part of skb->data. 449 */ 450#define SKB_DATAREF_SHIFT 16 451#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 452 453 454enum { 455 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 456 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 457 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 458}; 459 460enum { 461 SKB_GSO_TCPV4 = 1 << 0, 462 SKB_GSO_UDP = 1 << 1, 463 464 /* This indicates the skb is from an untrusted source. */ 465 SKB_GSO_DODGY = 1 << 2, 466 467 /* This indicates the tcp segment has CWR set. */ 468 SKB_GSO_TCP_ECN = 1 << 3, 469 470 SKB_GSO_TCP_FIXEDID = 1 << 4, 471 472 SKB_GSO_TCPV6 = 1 << 5, 473 474 SKB_GSO_FCOE = 1 << 6, 475 476 SKB_GSO_GRE = 1 << 7, 477 478 SKB_GSO_GRE_CSUM = 1 << 8, 479 480 SKB_GSO_IPXIP4 = 1 << 9, 481 482 SKB_GSO_IPXIP6 = 1 << 10, 483 484 SKB_GSO_UDP_TUNNEL = 1 << 11, 485 486 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12, 487 488 SKB_GSO_PARTIAL = 1 << 13, 489 490 SKB_GSO_TUNNEL_REMCSUM = 1 << 14, 491 492 SKB_GSO_SCTP = 1 << 15, 493}; 494 495#if BITS_PER_LONG > 32 496#define NET_SKBUFF_DATA_USES_OFFSET 1 497#endif 498 499#ifdef NET_SKBUFF_DATA_USES_OFFSET 500typedef unsigned int sk_buff_data_t; 501#else 502typedef unsigned char *sk_buff_data_t; 503#endif 504 505/** 506 * struct skb_mstamp - multi resolution time stamps 507 * @stamp_us: timestamp in us resolution 508 * @stamp_jiffies: timestamp in jiffies 509 */ 510struct skb_mstamp { 511 union { 512 u64 v64; 513 struct { 514 u32 stamp_us; 515 u32 stamp_jiffies; 516 }; 517 }; 518}; 519 520/** 521 * skb_mstamp_get - get current timestamp 522 * @cl: place to store timestamps 523 */ 524static inline void skb_mstamp_get(struct skb_mstamp *cl) 525{ 526 u64 val = local_clock(); 527 528 do_div(val, NSEC_PER_USEC); 529 cl->stamp_us = (u32)val; 530 cl->stamp_jiffies = (u32)jiffies; 531} 532 533/** 534 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp 535 * @t1: pointer to newest sample 536 * @t0: pointer to oldest sample 537 */ 538static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1, 539 const struct skb_mstamp *t0) 540{ 541 s32 delta_us = t1->stamp_us - t0->stamp_us; 542 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies; 543 544 /* If delta_us is negative, this might be because interval is too big, 545 * or local_clock() drift is too big : fallback using jiffies. 546 */ 547 if (delta_us <= 0 || 548 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ))) 549 550 delta_us = jiffies_to_usecs(delta_jiffies); 551 552 return delta_us; 553} 554 555static inline bool skb_mstamp_after(const struct skb_mstamp *t1, 556 const struct skb_mstamp *t0) 557{ 558 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies; 559 560 if (!diff) 561 diff = t1->stamp_us - t0->stamp_us; 562 return diff > 0; 563} 564 565/** 566 * struct sk_buff - socket buffer 567 * @next: Next buffer in list 568 * @prev: Previous buffer in list 569 * @tstamp: Time we arrived/left 570 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 571 * @sk: Socket we are owned by 572 * @dev: Device we arrived on/are leaving by 573 * @cb: Control buffer. Free for use by every layer. Put private vars here 574 * @_skb_refdst: destination entry (with norefcount bit) 575 * @sp: the security path, used for xfrm 576 * @len: Length of actual data 577 * @data_len: Data length 578 * @mac_len: Length of link layer header 579 * @hdr_len: writable header length of cloned skb 580 * @csum: Checksum (must include start/offset pair) 581 * @csum_start: Offset from skb->head where checksumming should start 582 * @csum_offset: Offset from csum_start where checksum should be stored 583 * @priority: Packet queueing priority 584 * @ignore_df: allow local fragmentation 585 * @cloned: Head may be cloned (check refcnt to be sure) 586 * @ip_summed: Driver fed us an IP checksum 587 * @nohdr: Payload reference only, must not modify header 588 * @nfctinfo: Relationship of this skb to the connection 589 * @pkt_type: Packet class 590 * @fclone: skbuff clone status 591 * @ipvs_property: skbuff is owned by ipvs 592 * @peeked: this packet has been seen already, so stats have been 593 * done for it, don't do them again 594 * @nf_trace: netfilter packet trace flag 595 * @protocol: Packet protocol from driver 596 * @destructor: Destruct function 597 * @nfct: Associated connection, if any 598 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 599 * @skb_iif: ifindex of device we arrived on 600 * @tc_index: Traffic control index 601 * @tc_verd: traffic control verdict 602 * @hash: the packet hash 603 * @queue_mapping: Queue mapping for multiqueue devices 604 * @xmit_more: More SKBs are pending for this queue 605 * @ndisc_nodetype: router type (from link layer) 606 * @ooo_okay: allow the mapping of a socket to a queue to be changed 607 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 608 * ports. 609 * @sw_hash: indicates hash was computed in software stack 610 * @wifi_acked_valid: wifi_acked was set 611 * @wifi_acked: whether frame was acked on wifi or not 612 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 613 * @napi_id: id of the NAPI struct this skb came from 614 * @secmark: security marking 615 * @mark: Generic packet mark 616 * @vlan_proto: vlan encapsulation protocol 617 * @vlan_tci: vlan tag control information 618 * @inner_protocol: Protocol (encapsulation) 619 * @inner_transport_header: Inner transport layer header (encapsulation) 620 * @inner_network_header: Network layer header (encapsulation) 621 * @inner_mac_header: Link layer header (encapsulation) 622 * @transport_header: Transport layer header 623 * @network_header: Network layer header 624 * @mac_header: Link layer header 625 * @tail: Tail pointer 626 * @end: End pointer 627 * @head: Head of buffer 628 * @data: Data head pointer 629 * @truesize: Buffer size 630 * @users: User count - see {datagram,tcp}.c 631 */ 632 633struct sk_buff { 634 union { 635 struct { 636 /* These two members must be first. */ 637 struct sk_buff *next; 638 struct sk_buff *prev; 639 640 union { 641 ktime_t tstamp; 642 struct skb_mstamp skb_mstamp; 643 }; 644 }; 645 struct rb_node rbnode; /* used in netem & tcp stack */ 646 }; 647 struct sock *sk; 648 struct net_device *dev; 649 650 /* 651 * This is the control buffer. It is free to use for every 652 * layer. Please put your private variables there. If you 653 * want to keep them across layers you have to do a skb_clone() 654 * first. This is owned by whoever has the skb queued ATM. 655 */ 656 char cb[48] __aligned(8); 657 658 unsigned long _skb_refdst; 659 void (*destructor)(struct sk_buff *skb); 660#ifdef CONFIG_XFRM 661 struct sec_path *sp; 662#endif 663#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 664 struct nf_conntrack *nfct; 665#endif 666#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 667 struct nf_bridge_info *nf_bridge; 668#endif 669 unsigned int len, 670 data_len; 671 __u16 mac_len, 672 hdr_len; 673 674 /* Following fields are _not_ copied in __copy_skb_header() 675 * Note that queue_mapping is here mostly to fill a hole. 676 */ 677 kmemcheck_bitfield_begin(flags1); 678 __u16 queue_mapping; 679 680/* if you move cloned around you also must adapt those constants */ 681#ifdef __BIG_ENDIAN_BITFIELD 682#define CLONED_MASK (1 << 7) 683#else 684#define CLONED_MASK 1 685#endif 686#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset) 687 688 __u8 __cloned_offset[0]; 689 __u8 cloned:1, 690 nohdr:1, 691 fclone:2, 692 peeked:1, 693 head_frag:1, 694 xmit_more:1, 695 __unused:1; /* one bit hole */ 696 kmemcheck_bitfield_end(flags1); 697 698 /* fields enclosed in headers_start/headers_end are copied 699 * using a single memcpy() in __copy_skb_header() 700 */ 701 /* private: */ 702 __u32 headers_start[0]; 703 /* public: */ 704 705/* if you move pkt_type around you also must adapt those constants */ 706#ifdef __BIG_ENDIAN_BITFIELD 707#define PKT_TYPE_MAX (7 << 5) 708#else 709#define PKT_TYPE_MAX 7 710#endif 711#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 712 713 __u8 __pkt_type_offset[0]; 714 __u8 pkt_type:3; 715 __u8 pfmemalloc:1; 716 __u8 ignore_df:1; 717 __u8 nfctinfo:3; 718 719 __u8 nf_trace:1; 720 __u8 ip_summed:2; 721 __u8 ooo_okay:1; 722 __u8 l4_hash:1; 723 __u8 sw_hash:1; 724 __u8 wifi_acked_valid:1; 725 __u8 wifi_acked:1; 726 727 __u8 no_fcs:1; 728 /* Indicates the inner headers are valid in the skbuff. */ 729 __u8 encapsulation:1; 730 __u8 encap_hdr_csum:1; 731 __u8 csum_valid:1; 732 __u8 csum_complete_sw:1; 733 __u8 csum_level:2; 734 __u8 csum_bad:1; 735 736#ifdef CONFIG_IPV6_NDISC_NODETYPE 737 __u8 ndisc_nodetype:2; 738#endif 739 __u8 ipvs_property:1; 740 __u8 inner_protocol_type:1; 741 __u8 remcsum_offload:1; 742#ifdef CONFIG_NET_SWITCHDEV 743 __u8 offload_fwd_mark:1; 744#endif 745 /* 2, 4 or 5 bit hole */ 746 747#ifdef CONFIG_NET_SCHED 748 __u16 tc_index; /* traffic control index */ 749#ifdef CONFIG_NET_CLS_ACT 750 __u16 tc_verd; /* traffic control verdict */ 751#endif 752#endif 753 754 union { 755 __wsum csum; 756 struct { 757 __u16 csum_start; 758 __u16 csum_offset; 759 }; 760 }; 761 __u32 priority; 762 int skb_iif; 763 __u32 hash; 764 __be16 vlan_proto; 765 __u16 vlan_tci; 766#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 767 union { 768 unsigned int napi_id; 769 unsigned int sender_cpu; 770 }; 771#endif 772#ifdef CONFIG_NETWORK_SECMARK 773 __u32 secmark; 774#endif 775 776 union { 777 __u32 mark; 778 __u32 reserved_tailroom; 779 }; 780 781 union { 782 __be16 inner_protocol; 783 __u8 inner_ipproto; 784 }; 785 786 __u16 inner_transport_header; 787 __u16 inner_network_header; 788 __u16 inner_mac_header; 789 790 __be16 protocol; 791 __u16 transport_header; 792 __u16 network_header; 793 __u16 mac_header; 794 795 /* private: */ 796 __u32 headers_end[0]; 797 /* public: */ 798 799 /* These elements must be at the end, see alloc_skb() for details. */ 800 sk_buff_data_t tail; 801 sk_buff_data_t end; 802 unsigned char *head, 803 *data; 804 unsigned int truesize; 805 atomic_t users; 806}; 807 808#ifdef __KERNEL__ 809/* 810 * Handling routines are only of interest to the kernel 811 */ 812#include <linux/slab.h> 813 814 815#define SKB_ALLOC_FCLONE 0x01 816#define SKB_ALLOC_RX 0x02 817#define SKB_ALLOC_NAPI 0x04 818 819/* Returns true if the skb was allocated from PFMEMALLOC reserves */ 820static inline bool skb_pfmemalloc(const struct sk_buff *skb) 821{ 822 return unlikely(skb->pfmemalloc); 823} 824 825/* 826 * skb might have a dst pointer attached, refcounted or not. 827 * _skb_refdst low order bit is set if refcount was _not_ taken 828 */ 829#define SKB_DST_NOREF 1UL 830#define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 831 832/** 833 * skb_dst - returns skb dst_entry 834 * @skb: buffer 835 * 836 * Returns skb dst_entry, regardless of reference taken or not. 837 */ 838static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 839{ 840 /* If refdst was not refcounted, check we still are in a 841 * rcu_read_lock section 842 */ 843 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 844 !rcu_read_lock_held() && 845 !rcu_read_lock_bh_held()); 846 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 847} 848 849/** 850 * skb_dst_set - sets skb dst 851 * @skb: buffer 852 * @dst: dst entry 853 * 854 * Sets skb dst, assuming a reference was taken on dst and should 855 * be released by skb_dst_drop() 856 */ 857static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 858{ 859 skb->_skb_refdst = (unsigned long)dst; 860} 861 862/** 863 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 864 * @skb: buffer 865 * @dst: dst entry 866 * 867 * Sets skb dst, assuming a reference was not taken on dst. 868 * If dst entry is cached, we do not take reference and dst_release 869 * will be avoided by refdst_drop. If dst entry is not cached, we take 870 * reference, so that last dst_release can destroy the dst immediately. 871 */ 872static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 873{ 874 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 875 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 876} 877 878/** 879 * skb_dst_is_noref - Test if skb dst isn't refcounted 880 * @skb: buffer 881 */ 882static inline bool skb_dst_is_noref(const struct sk_buff *skb) 883{ 884 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 885} 886 887static inline struct rtable *skb_rtable(const struct sk_buff *skb) 888{ 889 return (struct rtable *)skb_dst(skb); 890} 891 892/* For mangling skb->pkt_type from user space side from applications 893 * such as nft, tc, etc, we only allow a conservative subset of 894 * possible pkt_types to be set. 895*/ 896static inline bool skb_pkt_type_ok(u32 ptype) 897{ 898 return ptype <= PACKET_OTHERHOST; 899} 900 901void kfree_skb(struct sk_buff *skb); 902void kfree_skb_list(struct sk_buff *segs); 903void skb_tx_error(struct sk_buff *skb); 904void consume_skb(struct sk_buff *skb); 905void __kfree_skb(struct sk_buff *skb); 906extern struct kmem_cache *skbuff_head_cache; 907 908void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 909bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 910 bool *fragstolen, int *delta_truesize); 911 912struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 913 int node); 914struct sk_buff *__build_skb(void *data, unsigned int frag_size); 915struct sk_buff *build_skb(void *data, unsigned int frag_size); 916static inline struct sk_buff *alloc_skb(unsigned int size, 917 gfp_t priority) 918{ 919 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 920} 921 922struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 923 unsigned long data_len, 924 int max_page_order, 925 int *errcode, 926 gfp_t gfp_mask); 927 928/* Layout of fast clones : [skb1][skb2][fclone_ref] */ 929struct sk_buff_fclones { 930 struct sk_buff skb1; 931 932 struct sk_buff skb2; 933 934 atomic_t fclone_ref; 935}; 936 937/** 938 * skb_fclone_busy - check if fclone is busy 939 * @sk: socket 940 * @skb: buffer 941 * 942 * Returns true if skb is a fast clone, and its clone is not freed. 943 * Some drivers call skb_orphan() in their ndo_start_xmit(), 944 * so we also check that this didnt happen. 945 */ 946static inline bool skb_fclone_busy(const struct sock *sk, 947 const struct sk_buff *skb) 948{ 949 const struct sk_buff_fclones *fclones; 950 951 fclones = container_of(skb, struct sk_buff_fclones, skb1); 952 953 return skb->fclone == SKB_FCLONE_ORIG && 954 atomic_read(&fclones->fclone_ref) > 1 && 955 fclones->skb2.sk == sk; 956} 957 958static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 959 gfp_t priority) 960{ 961 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 962} 963 964struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 965static inline struct sk_buff *alloc_skb_head(gfp_t priority) 966{ 967 return __alloc_skb_head(priority, -1); 968} 969 970struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 971int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 972struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 973struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 974struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 975 gfp_t gfp_mask, bool fclone); 976static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 977 gfp_t gfp_mask) 978{ 979 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 980} 981 982int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 983struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 984 unsigned int headroom); 985struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 986 int newtailroom, gfp_t priority); 987int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 988 int offset, int len); 989int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, 990 int len); 991int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 992int skb_pad(struct sk_buff *skb, int pad); 993#define dev_kfree_skb(a) consume_skb(a) 994 995int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 996 int getfrag(void *from, char *to, int offset, 997 int len, int odd, struct sk_buff *skb), 998 void *from, int length); 999 1000int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1001 int offset, size_t size); 1002 1003struct skb_seq_state { 1004 __u32 lower_offset; 1005 __u32 upper_offset; 1006 __u32 frag_idx; 1007 __u32 stepped_offset; 1008 struct sk_buff *root_skb; 1009 struct sk_buff *cur_skb; 1010 __u8 *frag_data; 1011}; 1012 1013void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1014 unsigned int to, struct skb_seq_state *st); 1015unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1016 struct skb_seq_state *st); 1017void skb_abort_seq_read(struct skb_seq_state *st); 1018 1019unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1020 unsigned int to, struct ts_config *config); 1021 1022/* 1023 * Packet hash types specify the type of hash in skb_set_hash. 1024 * 1025 * Hash types refer to the protocol layer addresses which are used to 1026 * construct a packet's hash. The hashes are used to differentiate or identify 1027 * flows of the protocol layer for the hash type. Hash types are either 1028 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1029 * 1030 * Properties of hashes: 1031 * 1032 * 1) Two packets in different flows have different hash values 1033 * 2) Two packets in the same flow should have the same hash value 1034 * 1035 * A hash at a higher layer is considered to be more specific. A driver should 1036 * set the most specific hash possible. 1037 * 1038 * A driver cannot indicate a more specific hash than the layer at which a hash 1039 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1040 * 1041 * A driver may indicate a hash level which is less specific than the 1042 * actual layer the hash was computed on. For instance, a hash computed 1043 * at L4 may be considered an L3 hash. This should only be done if the 1044 * driver can't unambiguously determine that the HW computed the hash at 1045 * the higher layer. Note that the "should" in the second property above 1046 * permits this. 1047 */ 1048enum pkt_hash_types { 1049 PKT_HASH_TYPE_NONE, /* Undefined type */ 1050 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1051 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1052 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1053}; 1054 1055static inline void skb_clear_hash(struct sk_buff *skb) 1056{ 1057 skb->hash = 0; 1058 skb->sw_hash = 0; 1059 skb->l4_hash = 0; 1060} 1061 1062static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1063{ 1064 if (!skb->l4_hash) 1065 skb_clear_hash(skb); 1066} 1067 1068static inline void 1069__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1070{ 1071 skb->l4_hash = is_l4; 1072 skb->sw_hash = is_sw; 1073 skb->hash = hash; 1074} 1075 1076static inline void 1077skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1078{ 1079 /* Used by drivers to set hash from HW */ 1080 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1081} 1082 1083static inline void 1084__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1085{ 1086 __skb_set_hash(skb, hash, true, is_l4); 1087} 1088 1089void __skb_get_hash(struct sk_buff *skb); 1090u32 __skb_get_hash_symmetric(struct sk_buff *skb); 1091u32 skb_get_poff(const struct sk_buff *skb); 1092u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1093 const struct flow_keys *keys, int hlen); 1094__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1095 void *data, int hlen_proto); 1096 1097static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1098 int thoff, u8 ip_proto) 1099{ 1100 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1101} 1102 1103void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1104 const struct flow_dissector_key *key, 1105 unsigned int key_count); 1106 1107bool __skb_flow_dissect(const struct sk_buff *skb, 1108 struct flow_dissector *flow_dissector, 1109 void *target_container, 1110 void *data, __be16 proto, int nhoff, int hlen, 1111 unsigned int flags); 1112 1113static inline bool skb_flow_dissect(const struct sk_buff *skb, 1114 struct flow_dissector *flow_dissector, 1115 void *target_container, unsigned int flags) 1116{ 1117 return __skb_flow_dissect(skb, flow_dissector, target_container, 1118 NULL, 0, 0, 0, flags); 1119} 1120 1121static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1122 struct flow_keys *flow, 1123 unsigned int flags) 1124{ 1125 memset(flow, 0, sizeof(*flow)); 1126 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1127 NULL, 0, 0, 0, flags); 1128} 1129 1130static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow, 1131 void *data, __be16 proto, 1132 int nhoff, int hlen, 1133 unsigned int flags) 1134{ 1135 memset(flow, 0, sizeof(*flow)); 1136 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow, 1137 data, proto, nhoff, hlen, flags); 1138} 1139 1140static inline __u32 skb_get_hash(struct sk_buff *skb) 1141{ 1142 if (!skb->l4_hash && !skb->sw_hash) 1143 __skb_get_hash(skb); 1144 1145 return skb->hash; 1146} 1147 1148__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6); 1149 1150static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1151{ 1152 if (!skb->l4_hash && !skb->sw_hash) { 1153 struct flow_keys keys; 1154 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1155 1156 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1157 } 1158 1159 return skb->hash; 1160} 1161 1162__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl); 1163 1164static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4) 1165{ 1166 if (!skb->l4_hash && !skb->sw_hash) { 1167 struct flow_keys keys; 1168 __u32 hash = __get_hash_from_flowi4(fl4, &keys); 1169 1170 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1171 } 1172 1173 return skb->hash; 1174} 1175 1176__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1177 1178static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1179{ 1180 return skb->hash; 1181} 1182 1183static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1184{ 1185 to->hash = from->hash; 1186 to->sw_hash = from->sw_hash; 1187 to->l4_hash = from->l4_hash; 1188}; 1189 1190#ifdef NET_SKBUFF_DATA_USES_OFFSET 1191static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1192{ 1193 return skb->head + skb->end; 1194} 1195 1196static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1197{ 1198 return skb->end; 1199} 1200#else 1201static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1202{ 1203 return skb->end; 1204} 1205 1206static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1207{ 1208 return skb->end - skb->head; 1209} 1210#endif 1211 1212/* Internal */ 1213#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1214 1215static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1216{ 1217 return &skb_shinfo(skb)->hwtstamps; 1218} 1219 1220/** 1221 * skb_queue_empty - check if a queue is empty 1222 * @list: queue head 1223 * 1224 * Returns true if the queue is empty, false otherwise. 1225 */ 1226static inline int skb_queue_empty(const struct sk_buff_head *list) 1227{ 1228 return list->next == (const struct sk_buff *) list; 1229} 1230 1231/** 1232 * skb_queue_is_last - check if skb is the last entry in the queue 1233 * @list: queue head 1234 * @skb: buffer 1235 * 1236 * Returns true if @skb is the last buffer on the list. 1237 */ 1238static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1239 const struct sk_buff *skb) 1240{ 1241 return skb->next == (const struct sk_buff *) list; 1242} 1243 1244/** 1245 * skb_queue_is_first - check if skb is the first entry in the queue 1246 * @list: queue head 1247 * @skb: buffer 1248 * 1249 * Returns true if @skb is the first buffer on the list. 1250 */ 1251static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1252 const struct sk_buff *skb) 1253{ 1254 return skb->prev == (const struct sk_buff *) list; 1255} 1256 1257/** 1258 * skb_queue_next - return the next packet in the queue 1259 * @list: queue head 1260 * @skb: current buffer 1261 * 1262 * Return the next packet in @list after @skb. It is only valid to 1263 * call this if skb_queue_is_last() evaluates to false. 1264 */ 1265static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1266 const struct sk_buff *skb) 1267{ 1268 /* This BUG_ON may seem severe, but if we just return then we 1269 * are going to dereference garbage. 1270 */ 1271 BUG_ON(skb_queue_is_last(list, skb)); 1272 return skb->next; 1273} 1274 1275/** 1276 * skb_queue_prev - return the prev packet in the queue 1277 * @list: queue head 1278 * @skb: current buffer 1279 * 1280 * Return the prev packet in @list before @skb. It is only valid to 1281 * call this if skb_queue_is_first() evaluates to false. 1282 */ 1283static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1284 const struct sk_buff *skb) 1285{ 1286 /* This BUG_ON may seem severe, but if we just return then we 1287 * are going to dereference garbage. 1288 */ 1289 BUG_ON(skb_queue_is_first(list, skb)); 1290 return skb->prev; 1291} 1292 1293/** 1294 * skb_get - reference buffer 1295 * @skb: buffer to reference 1296 * 1297 * Makes another reference to a socket buffer and returns a pointer 1298 * to the buffer. 1299 */ 1300static inline struct sk_buff *skb_get(struct sk_buff *skb) 1301{ 1302 atomic_inc(&skb->users); 1303 return skb; 1304} 1305 1306/* 1307 * If users == 1, we are the only owner and are can avoid redundant 1308 * atomic change. 1309 */ 1310 1311/** 1312 * skb_cloned - is the buffer a clone 1313 * @skb: buffer to check 1314 * 1315 * Returns true if the buffer was generated with skb_clone() and is 1316 * one of multiple shared copies of the buffer. Cloned buffers are 1317 * shared data so must not be written to under normal circumstances. 1318 */ 1319static inline int skb_cloned(const struct sk_buff *skb) 1320{ 1321 return skb->cloned && 1322 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1323} 1324 1325static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1326{ 1327 might_sleep_if(gfpflags_allow_blocking(pri)); 1328 1329 if (skb_cloned(skb)) 1330 return pskb_expand_head(skb, 0, 0, pri); 1331 1332 return 0; 1333} 1334 1335/** 1336 * skb_header_cloned - is the header a clone 1337 * @skb: buffer to check 1338 * 1339 * Returns true if modifying the header part of the buffer requires 1340 * the data to be copied. 1341 */ 1342static inline int skb_header_cloned(const struct sk_buff *skb) 1343{ 1344 int dataref; 1345 1346 if (!skb->cloned) 1347 return 0; 1348 1349 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1350 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1351 return dataref != 1; 1352} 1353 1354static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1355{ 1356 might_sleep_if(gfpflags_allow_blocking(pri)); 1357 1358 if (skb_header_cloned(skb)) 1359 return pskb_expand_head(skb, 0, 0, pri); 1360 1361 return 0; 1362} 1363 1364/** 1365 * skb_header_release - release reference to header 1366 * @skb: buffer to operate on 1367 * 1368 * Drop a reference to the header part of the buffer. This is done 1369 * by acquiring a payload reference. You must not read from the header 1370 * part of skb->data after this. 1371 * Note : Check if you can use __skb_header_release() instead. 1372 */ 1373static inline void skb_header_release(struct sk_buff *skb) 1374{ 1375 BUG_ON(skb->nohdr); 1376 skb->nohdr = 1; 1377 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1378} 1379 1380/** 1381 * __skb_header_release - release reference to header 1382 * @skb: buffer to operate on 1383 * 1384 * Variant of skb_header_release() assuming skb is private to caller. 1385 * We can avoid one atomic operation. 1386 */ 1387static inline void __skb_header_release(struct sk_buff *skb) 1388{ 1389 skb->nohdr = 1; 1390 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1391} 1392 1393 1394/** 1395 * skb_shared - is the buffer shared 1396 * @skb: buffer to check 1397 * 1398 * Returns true if more than one person has a reference to this 1399 * buffer. 1400 */ 1401static inline int skb_shared(const struct sk_buff *skb) 1402{ 1403 return atomic_read(&skb->users) != 1; 1404} 1405 1406/** 1407 * skb_share_check - check if buffer is shared and if so clone it 1408 * @skb: buffer to check 1409 * @pri: priority for memory allocation 1410 * 1411 * If the buffer is shared the buffer is cloned and the old copy 1412 * drops a reference. A new clone with a single reference is returned. 1413 * If the buffer is not shared the original buffer is returned. When 1414 * being called from interrupt status or with spinlocks held pri must 1415 * be GFP_ATOMIC. 1416 * 1417 * NULL is returned on a memory allocation failure. 1418 */ 1419static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1420{ 1421 might_sleep_if(gfpflags_allow_blocking(pri)); 1422 if (skb_shared(skb)) { 1423 struct sk_buff *nskb = skb_clone(skb, pri); 1424 1425 if (likely(nskb)) 1426 consume_skb(skb); 1427 else 1428 kfree_skb(skb); 1429 skb = nskb; 1430 } 1431 return skb; 1432} 1433 1434/* 1435 * Copy shared buffers into a new sk_buff. We effectively do COW on 1436 * packets to handle cases where we have a local reader and forward 1437 * and a couple of other messy ones. The normal one is tcpdumping 1438 * a packet thats being forwarded. 1439 */ 1440 1441/** 1442 * skb_unshare - make a copy of a shared buffer 1443 * @skb: buffer to check 1444 * @pri: priority for memory allocation 1445 * 1446 * If the socket buffer is a clone then this function creates a new 1447 * copy of the data, drops a reference count on the old copy and returns 1448 * the new copy with the reference count at 1. If the buffer is not a clone 1449 * the original buffer is returned. When called with a spinlock held or 1450 * from interrupt state @pri must be %GFP_ATOMIC 1451 * 1452 * %NULL is returned on a memory allocation failure. 1453 */ 1454static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1455 gfp_t pri) 1456{ 1457 might_sleep_if(gfpflags_allow_blocking(pri)); 1458 if (skb_cloned(skb)) { 1459 struct sk_buff *nskb = skb_copy(skb, pri); 1460 1461 /* Free our shared copy */ 1462 if (likely(nskb)) 1463 consume_skb(skb); 1464 else 1465 kfree_skb(skb); 1466 skb = nskb; 1467 } 1468 return skb; 1469} 1470 1471/** 1472 * skb_peek - peek at the head of an &sk_buff_head 1473 * @list_: list to peek at 1474 * 1475 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1476 * be careful with this one. A peek leaves the buffer on the 1477 * list and someone else may run off with it. You must hold 1478 * the appropriate locks or have a private queue to do this. 1479 * 1480 * Returns %NULL for an empty list or a pointer to the head element. 1481 * The reference count is not incremented and the reference is therefore 1482 * volatile. Use with caution. 1483 */ 1484static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1485{ 1486 struct sk_buff *skb = list_->next; 1487 1488 if (skb == (struct sk_buff *)list_) 1489 skb = NULL; 1490 return skb; 1491} 1492 1493/** 1494 * skb_peek_next - peek skb following the given one from a queue 1495 * @skb: skb to start from 1496 * @list_: list to peek at 1497 * 1498 * Returns %NULL when the end of the list is met or a pointer to the 1499 * next element. The reference count is not incremented and the 1500 * reference is therefore volatile. Use with caution. 1501 */ 1502static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1503 const struct sk_buff_head *list_) 1504{ 1505 struct sk_buff *next = skb->next; 1506 1507 if (next == (struct sk_buff *)list_) 1508 next = NULL; 1509 return next; 1510} 1511 1512/** 1513 * skb_peek_tail - peek at the tail of an &sk_buff_head 1514 * @list_: list to peek at 1515 * 1516 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1517 * be careful with this one. A peek leaves the buffer on the 1518 * list and someone else may run off with it. You must hold 1519 * the appropriate locks or have a private queue to do this. 1520 * 1521 * Returns %NULL for an empty list or a pointer to the tail element. 1522 * The reference count is not incremented and the reference is therefore 1523 * volatile. Use with caution. 1524 */ 1525static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1526{ 1527 struct sk_buff *skb = list_->prev; 1528 1529 if (skb == (struct sk_buff *)list_) 1530 skb = NULL; 1531 return skb; 1532 1533} 1534 1535/** 1536 * skb_queue_len - get queue length 1537 * @list_: list to measure 1538 * 1539 * Return the length of an &sk_buff queue. 1540 */ 1541static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1542{ 1543 return list_->qlen; 1544} 1545 1546/** 1547 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1548 * @list: queue to initialize 1549 * 1550 * This initializes only the list and queue length aspects of 1551 * an sk_buff_head object. This allows to initialize the list 1552 * aspects of an sk_buff_head without reinitializing things like 1553 * the spinlock. It can also be used for on-stack sk_buff_head 1554 * objects where the spinlock is known to not be used. 1555 */ 1556static inline void __skb_queue_head_init(struct sk_buff_head *list) 1557{ 1558 list->prev = list->next = (struct sk_buff *)list; 1559 list->qlen = 0; 1560} 1561 1562/* 1563 * This function creates a split out lock class for each invocation; 1564 * this is needed for now since a whole lot of users of the skb-queue 1565 * infrastructure in drivers have different locking usage (in hardirq) 1566 * than the networking core (in softirq only). In the long run either the 1567 * network layer or drivers should need annotation to consolidate the 1568 * main types of usage into 3 classes. 1569 */ 1570static inline void skb_queue_head_init(struct sk_buff_head *list) 1571{ 1572 spin_lock_init(&list->lock); 1573 __skb_queue_head_init(list); 1574} 1575 1576static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1577 struct lock_class_key *class) 1578{ 1579 skb_queue_head_init(list); 1580 lockdep_set_class(&list->lock, class); 1581} 1582 1583/* 1584 * Insert an sk_buff on a list. 1585 * 1586 * The "__skb_xxxx()" functions are the non-atomic ones that 1587 * can only be called with interrupts disabled. 1588 */ 1589void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1590 struct sk_buff_head *list); 1591static inline void __skb_insert(struct sk_buff *newsk, 1592 struct sk_buff *prev, struct sk_buff *next, 1593 struct sk_buff_head *list) 1594{ 1595 newsk->next = next; 1596 newsk->prev = prev; 1597 next->prev = prev->next = newsk; 1598 list->qlen++; 1599} 1600 1601static inline void __skb_queue_splice(const struct sk_buff_head *list, 1602 struct sk_buff *prev, 1603 struct sk_buff *next) 1604{ 1605 struct sk_buff *first = list->next; 1606 struct sk_buff *last = list->prev; 1607 1608 first->prev = prev; 1609 prev->next = first; 1610 1611 last->next = next; 1612 next->prev = last; 1613} 1614 1615/** 1616 * skb_queue_splice - join two skb lists, this is designed for stacks 1617 * @list: the new list to add 1618 * @head: the place to add it in the first list 1619 */ 1620static inline void skb_queue_splice(const struct sk_buff_head *list, 1621 struct sk_buff_head *head) 1622{ 1623 if (!skb_queue_empty(list)) { 1624 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1625 head->qlen += list->qlen; 1626 } 1627} 1628 1629/** 1630 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1631 * @list: the new list to add 1632 * @head: the place to add it in the first list 1633 * 1634 * The list at @list is reinitialised 1635 */ 1636static inline void skb_queue_splice_init(struct sk_buff_head *list, 1637 struct sk_buff_head *head) 1638{ 1639 if (!skb_queue_empty(list)) { 1640 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1641 head->qlen += list->qlen; 1642 __skb_queue_head_init(list); 1643 } 1644} 1645 1646/** 1647 * skb_queue_splice_tail - join two skb lists, each list being a queue 1648 * @list: the new list to add 1649 * @head: the place to add it in the first list 1650 */ 1651static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1652 struct sk_buff_head *head) 1653{ 1654 if (!skb_queue_empty(list)) { 1655 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1656 head->qlen += list->qlen; 1657 } 1658} 1659 1660/** 1661 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1662 * @list: the new list to add 1663 * @head: the place to add it in the first list 1664 * 1665 * Each of the lists is a queue. 1666 * The list at @list is reinitialised 1667 */ 1668static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1669 struct sk_buff_head *head) 1670{ 1671 if (!skb_queue_empty(list)) { 1672 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1673 head->qlen += list->qlen; 1674 __skb_queue_head_init(list); 1675 } 1676} 1677 1678/** 1679 * __skb_queue_after - queue a buffer at the list head 1680 * @list: list to use 1681 * @prev: place after this buffer 1682 * @newsk: buffer to queue 1683 * 1684 * Queue a buffer int the middle of a list. This function takes no locks 1685 * and you must therefore hold required locks before calling it. 1686 * 1687 * A buffer cannot be placed on two lists at the same time. 1688 */ 1689static inline void __skb_queue_after(struct sk_buff_head *list, 1690 struct sk_buff *prev, 1691 struct sk_buff *newsk) 1692{ 1693 __skb_insert(newsk, prev, prev->next, list); 1694} 1695 1696void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1697 struct sk_buff_head *list); 1698 1699static inline void __skb_queue_before(struct sk_buff_head *list, 1700 struct sk_buff *next, 1701 struct sk_buff *newsk) 1702{ 1703 __skb_insert(newsk, next->prev, next, list); 1704} 1705 1706/** 1707 * __skb_queue_head - queue a buffer at the list head 1708 * @list: list to use 1709 * @newsk: buffer to queue 1710 * 1711 * Queue a buffer at the start of a list. This function takes no locks 1712 * and you must therefore hold required locks before calling it. 1713 * 1714 * A buffer cannot be placed on two lists at the same time. 1715 */ 1716void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1717static inline void __skb_queue_head(struct sk_buff_head *list, 1718 struct sk_buff *newsk) 1719{ 1720 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1721} 1722 1723/** 1724 * __skb_queue_tail - queue a buffer at the list tail 1725 * @list: list to use 1726 * @newsk: buffer to queue 1727 * 1728 * Queue a buffer at the end of a list. This function takes no locks 1729 * and you must therefore hold required locks before calling it. 1730 * 1731 * A buffer cannot be placed on two lists at the same time. 1732 */ 1733void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1734static inline void __skb_queue_tail(struct sk_buff_head *list, 1735 struct sk_buff *newsk) 1736{ 1737 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1738} 1739 1740/* 1741 * remove sk_buff from list. _Must_ be called atomically, and with 1742 * the list known.. 1743 */ 1744void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1745static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1746{ 1747 struct sk_buff *next, *prev; 1748 1749 list->qlen--; 1750 next = skb->next; 1751 prev = skb->prev; 1752 skb->next = skb->prev = NULL; 1753 next->prev = prev; 1754 prev->next = next; 1755} 1756 1757/** 1758 * __skb_dequeue - remove from the head of the queue 1759 * @list: list to dequeue from 1760 * 1761 * Remove the head of the list. This function does not take any locks 1762 * so must be used with appropriate locks held only. The head item is 1763 * returned or %NULL if the list is empty. 1764 */ 1765struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1766static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1767{ 1768 struct sk_buff *skb = skb_peek(list); 1769 if (skb) 1770 __skb_unlink(skb, list); 1771 return skb; 1772} 1773 1774/** 1775 * __skb_dequeue_tail - remove from the tail of the queue 1776 * @list: list to dequeue from 1777 * 1778 * Remove the tail of the list. This function does not take any locks 1779 * so must be used with appropriate locks held only. The tail item is 1780 * returned or %NULL if the list is empty. 1781 */ 1782struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1783static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1784{ 1785 struct sk_buff *skb = skb_peek_tail(list); 1786 if (skb) 1787 __skb_unlink(skb, list); 1788 return skb; 1789} 1790 1791 1792static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1793{ 1794 return skb->data_len; 1795} 1796 1797static inline unsigned int skb_headlen(const struct sk_buff *skb) 1798{ 1799 return skb->len - skb->data_len; 1800} 1801 1802static inline int skb_pagelen(const struct sk_buff *skb) 1803{ 1804 int i, len = 0; 1805 1806 for (i = (int)skb_shinfo(skb)->nr_frags - 1; i >= 0; i--) 1807 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1808 return len + skb_headlen(skb); 1809} 1810 1811/** 1812 * __skb_fill_page_desc - initialise a paged fragment in an skb 1813 * @skb: buffer containing fragment to be initialised 1814 * @i: paged fragment index to initialise 1815 * @page: the page to use for this fragment 1816 * @off: the offset to the data with @page 1817 * @size: the length of the data 1818 * 1819 * Initialises the @i'th fragment of @skb to point to &size bytes at 1820 * offset @off within @page. 1821 * 1822 * Does not take any additional reference on the fragment. 1823 */ 1824static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1825 struct page *page, int off, int size) 1826{ 1827 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1828 1829 /* 1830 * Propagate page pfmemalloc to the skb if we can. The problem is 1831 * that not all callers have unique ownership of the page but rely 1832 * on page_is_pfmemalloc doing the right thing(tm). 1833 */ 1834 frag->page.p = page; 1835 frag->page_offset = off; 1836 skb_frag_size_set(frag, size); 1837 1838 page = compound_head(page); 1839 if (page_is_pfmemalloc(page)) 1840 skb->pfmemalloc = true; 1841} 1842 1843/** 1844 * skb_fill_page_desc - initialise a paged fragment in an skb 1845 * @skb: buffer containing fragment to be initialised 1846 * @i: paged fragment index to initialise 1847 * @page: the page to use for this fragment 1848 * @off: the offset to the data with @page 1849 * @size: the length of the data 1850 * 1851 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1852 * @skb to point to @size bytes at offset @off within @page. In 1853 * addition updates @skb such that @i is the last fragment. 1854 * 1855 * Does not take any additional reference on the fragment. 1856 */ 1857static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1858 struct page *page, int off, int size) 1859{ 1860 __skb_fill_page_desc(skb, i, page, off, size); 1861 skb_shinfo(skb)->nr_frags = i + 1; 1862} 1863 1864void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1865 int size, unsigned int truesize); 1866 1867void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1868 unsigned int truesize); 1869 1870#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1871#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1872#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1873 1874#ifdef NET_SKBUFF_DATA_USES_OFFSET 1875static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1876{ 1877 return skb->head + skb->tail; 1878} 1879 1880static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1881{ 1882 skb->tail = skb->data - skb->head; 1883} 1884 1885static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1886{ 1887 skb_reset_tail_pointer(skb); 1888 skb->tail += offset; 1889} 1890 1891#else /* NET_SKBUFF_DATA_USES_OFFSET */ 1892static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1893{ 1894 return skb->tail; 1895} 1896 1897static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1898{ 1899 skb->tail = skb->data; 1900} 1901 1902static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1903{ 1904 skb->tail = skb->data + offset; 1905} 1906 1907#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1908 1909/* 1910 * Add data to an sk_buff 1911 */ 1912unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1913unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1914static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1915{ 1916 unsigned char *tmp = skb_tail_pointer(skb); 1917 SKB_LINEAR_ASSERT(skb); 1918 skb->tail += len; 1919 skb->len += len; 1920 return tmp; 1921} 1922 1923unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1924static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1925{ 1926 skb->data -= len; 1927 skb->len += len; 1928 return skb->data; 1929} 1930 1931unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1932static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1933{ 1934 skb->len -= len; 1935 BUG_ON(skb->len < skb->data_len); 1936 return skb->data += len; 1937} 1938 1939static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1940{ 1941 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1942} 1943 1944unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1945 1946static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1947{ 1948 if (len > skb_headlen(skb) && 1949 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1950 return NULL; 1951 skb->len -= len; 1952 return skb->data += len; 1953} 1954 1955static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1956{ 1957 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1958} 1959 1960static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1961{ 1962 if (likely(len <= skb_headlen(skb))) 1963 return 1; 1964 if (unlikely(len > skb->len)) 1965 return 0; 1966 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1967} 1968 1969/** 1970 * skb_headroom - bytes at buffer head 1971 * @skb: buffer to check 1972 * 1973 * Return the number of bytes of free space at the head of an &sk_buff. 1974 */ 1975static inline unsigned int skb_headroom(const struct sk_buff *skb) 1976{ 1977 return skb->data - skb->head; 1978} 1979 1980/** 1981 * skb_tailroom - bytes at buffer end 1982 * @skb: buffer to check 1983 * 1984 * Return the number of bytes of free space at the tail of an sk_buff 1985 */ 1986static inline int skb_tailroom(const struct sk_buff *skb) 1987{ 1988 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 1989} 1990 1991/** 1992 * skb_availroom - bytes at buffer end 1993 * @skb: buffer to check 1994 * 1995 * Return the number of bytes of free space at the tail of an sk_buff 1996 * allocated by sk_stream_alloc() 1997 */ 1998static inline int skb_availroom(const struct sk_buff *skb) 1999{ 2000 if (skb_is_nonlinear(skb)) 2001 return 0; 2002 2003 return skb->end - skb->tail - skb->reserved_tailroom; 2004} 2005 2006/** 2007 * skb_reserve - adjust headroom 2008 * @skb: buffer to alter 2009 * @len: bytes to move 2010 * 2011 * Increase the headroom of an empty &sk_buff by reducing the tail 2012 * room. This is only allowed for an empty buffer. 2013 */ 2014static inline void skb_reserve(struct sk_buff *skb, int len) 2015{ 2016 skb->data += len; 2017 skb->tail += len; 2018} 2019 2020/** 2021 * skb_tailroom_reserve - adjust reserved_tailroom 2022 * @skb: buffer to alter 2023 * @mtu: maximum amount of headlen permitted 2024 * @needed_tailroom: minimum amount of reserved_tailroom 2025 * 2026 * Set reserved_tailroom so that headlen can be as large as possible but 2027 * not larger than mtu and tailroom cannot be smaller than 2028 * needed_tailroom. 2029 * The required headroom should already have been reserved before using 2030 * this function. 2031 */ 2032static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2033 unsigned int needed_tailroom) 2034{ 2035 SKB_LINEAR_ASSERT(skb); 2036 if (mtu < skb_tailroom(skb) - needed_tailroom) 2037 /* use at most mtu */ 2038 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2039 else 2040 /* use up to all available space */ 2041 skb->reserved_tailroom = needed_tailroom; 2042} 2043 2044#define ENCAP_TYPE_ETHER 0 2045#define ENCAP_TYPE_IPPROTO 1 2046 2047static inline void skb_set_inner_protocol(struct sk_buff *skb, 2048 __be16 protocol) 2049{ 2050 skb->inner_protocol = protocol; 2051 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2052} 2053 2054static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2055 __u8 ipproto) 2056{ 2057 skb->inner_ipproto = ipproto; 2058 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2059} 2060 2061static inline void skb_reset_inner_headers(struct sk_buff *skb) 2062{ 2063 skb->inner_mac_header = skb->mac_header; 2064 skb->inner_network_header = skb->network_header; 2065 skb->inner_transport_header = skb->transport_header; 2066} 2067 2068static inline void skb_reset_mac_len(struct sk_buff *skb) 2069{ 2070 skb->mac_len = skb->network_header - skb->mac_header; 2071} 2072 2073static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2074 *skb) 2075{ 2076 return skb->head + skb->inner_transport_header; 2077} 2078 2079static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2080{ 2081 return skb_inner_transport_header(skb) - skb->data; 2082} 2083 2084static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2085{ 2086 skb->inner_transport_header = skb->data - skb->head; 2087} 2088 2089static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2090 const int offset) 2091{ 2092 skb_reset_inner_transport_header(skb); 2093 skb->inner_transport_header += offset; 2094} 2095 2096static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2097{ 2098 return skb->head + skb->inner_network_header; 2099} 2100 2101static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2102{ 2103 skb->inner_network_header = skb->data - skb->head; 2104} 2105 2106static inline void skb_set_inner_network_header(struct sk_buff *skb, 2107 const int offset) 2108{ 2109 skb_reset_inner_network_header(skb); 2110 skb->inner_network_header += offset; 2111} 2112 2113static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2114{ 2115 return skb->head + skb->inner_mac_header; 2116} 2117 2118static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2119{ 2120 skb->inner_mac_header = skb->data - skb->head; 2121} 2122 2123static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2124 const int offset) 2125{ 2126 skb_reset_inner_mac_header(skb); 2127 skb->inner_mac_header += offset; 2128} 2129static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2130{ 2131 return skb->transport_header != (typeof(skb->transport_header))~0U; 2132} 2133 2134static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2135{ 2136 return skb->head + skb->transport_header; 2137} 2138 2139static inline void skb_reset_transport_header(struct sk_buff *skb) 2140{ 2141 skb->transport_header = skb->data - skb->head; 2142} 2143 2144static inline void skb_set_transport_header(struct sk_buff *skb, 2145 const int offset) 2146{ 2147 skb_reset_transport_header(skb); 2148 skb->transport_header += offset; 2149} 2150 2151static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2152{ 2153 return skb->head + skb->network_header; 2154} 2155 2156static inline void skb_reset_network_header(struct sk_buff *skb) 2157{ 2158 skb->network_header = skb->data - skb->head; 2159} 2160 2161static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2162{ 2163 skb_reset_network_header(skb); 2164 skb->network_header += offset; 2165} 2166 2167static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2168{ 2169 return skb->head + skb->mac_header; 2170} 2171 2172static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2173{ 2174 return skb->mac_header != (typeof(skb->mac_header))~0U; 2175} 2176 2177static inline void skb_reset_mac_header(struct sk_buff *skb) 2178{ 2179 skb->mac_header = skb->data - skb->head; 2180} 2181 2182static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2183{ 2184 skb_reset_mac_header(skb); 2185 skb->mac_header += offset; 2186} 2187 2188static inline void skb_pop_mac_header(struct sk_buff *skb) 2189{ 2190 skb->mac_header = skb->network_header; 2191} 2192 2193static inline void skb_probe_transport_header(struct sk_buff *skb, 2194 const int offset_hint) 2195{ 2196 struct flow_keys keys; 2197 2198 if (skb_transport_header_was_set(skb)) 2199 return; 2200 else if (skb_flow_dissect_flow_keys(skb, &keys, 0)) 2201 skb_set_transport_header(skb, keys.control.thoff); 2202 else 2203 skb_set_transport_header(skb, offset_hint); 2204} 2205 2206static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2207{ 2208 if (skb_mac_header_was_set(skb)) { 2209 const unsigned char *old_mac = skb_mac_header(skb); 2210 2211 skb_set_mac_header(skb, -skb->mac_len); 2212 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2213 } 2214} 2215 2216static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2217{ 2218 return skb->csum_start - skb_headroom(skb); 2219} 2220 2221static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2222{ 2223 return skb->head + skb->csum_start; 2224} 2225 2226static inline int skb_transport_offset(const struct sk_buff *skb) 2227{ 2228 return skb_transport_header(skb) - skb->data; 2229} 2230 2231static inline u32 skb_network_header_len(const struct sk_buff *skb) 2232{ 2233 return skb->transport_header - skb->network_header; 2234} 2235 2236static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2237{ 2238 return skb->inner_transport_header - skb->inner_network_header; 2239} 2240 2241static inline int skb_network_offset(const struct sk_buff *skb) 2242{ 2243 return skb_network_header(skb) - skb->data; 2244} 2245 2246static inline int skb_inner_network_offset(const struct sk_buff *skb) 2247{ 2248 return skb_inner_network_header(skb) - skb->data; 2249} 2250 2251static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2252{ 2253 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2254} 2255 2256/* 2257 * CPUs often take a performance hit when accessing unaligned memory 2258 * locations. The actual performance hit varies, it can be small if the 2259 * hardware handles it or large if we have to take an exception and fix it 2260 * in software. 2261 * 2262 * Since an ethernet header is 14 bytes network drivers often end up with 2263 * the IP header at an unaligned offset. The IP header can be aligned by 2264 * shifting the start of the packet by 2 bytes. Drivers should do this 2265 * with: 2266 * 2267 * skb_reserve(skb, NET_IP_ALIGN); 2268 * 2269 * The downside to this alignment of the IP header is that the DMA is now 2270 * unaligned. On some architectures the cost of an unaligned DMA is high 2271 * and this cost outweighs the gains made by aligning the IP header. 2272 * 2273 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2274 * to be overridden. 2275 */ 2276#ifndef NET_IP_ALIGN 2277#define NET_IP_ALIGN 2 2278#endif 2279 2280/* 2281 * The networking layer reserves some headroom in skb data (via 2282 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2283 * the header has to grow. In the default case, if the header has to grow 2284 * 32 bytes or less we avoid the reallocation. 2285 * 2286 * Unfortunately this headroom changes the DMA alignment of the resulting 2287 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2288 * on some architectures. An architecture can override this value, 2289 * perhaps setting it to a cacheline in size (since that will maintain 2290 * cacheline alignment of the DMA). It must be a power of 2. 2291 * 2292 * Various parts of the networking layer expect at least 32 bytes of 2293 * headroom, you should not reduce this. 2294 * 2295 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2296 * to reduce average number of cache lines per packet. 2297 * get_rps_cpus() for example only access one 64 bytes aligned block : 2298 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2299 */ 2300#ifndef NET_SKB_PAD 2301#define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2302#endif 2303 2304int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2305 2306static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2307{ 2308 if (unlikely(skb_is_nonlinear(skb))) { 2309 WARN_ON(1); 2310 return; 2311 } 2312 skb->len = len; 2313 skb_set_tail_pointer(skb, len); 2314} 2315 2316static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2317{ 2318 __skb_set_length(skb, len); 2319} 2320 2321void skb_trim(struct sk_buff *skb, unsigned int len); 2322 2323static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2324{ 2325 if (skb->data_len) 2326 return ___pskb_trim(skb, len); 2327 __skb_trim(skb, len); 2328 return 0; 2329} 2330 2331static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2332{ 2333 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2334} 2335 2336/** 2337 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2338 * @skb: buffer to alter 2339 * @len: new length 2340 * 2341 * This is identical to pskb_trim except that the caller knows that 2342 * the skb is not cloned so we should never get an error due to out- 2343 * of-memory. 2344 */ 2345static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2346{ 2347 int err = pskb_trim(skb, len); 2348 BUG_ON(err); 2349} 2350 2351static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2352{ 2353 unsigned int diff = len - skb->len; 2354 2355 if (skb_tailroom(skb) < diff) { 2356 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2357 GFP_ATOMIC); 2358 if (ret) 2359 return ret; 2360 } 2361 __skb_set_length(skb, len); 2362 return 0; 2363} 2364 2365/** 2366 * skb_orphan - orphan a buffer 2367 * @skb: buffer to orphan 2368 * 2369 * If a buffer currently has an owner then we call the owner's 2370 * destructor function and make the @skb unowned. The buffer continues 2371 * to exist but is no longer charged to its former owner. 2372 */ 2373static inline void skb_orphan(struct sk_buff *skb) 2374{ 2375 if (skb->destructor) { 2376 skb->destructor(skb); 2377 skb->destructor = NULL; 2378 skb->sk = NULL; 2379 } else { 2380 BUG_ON(skb->sk); 2381 } 2382} 2383 2384/** 2385 * skb_orphan_frags - orphan the frags contained in a buffer 2386 * @skb: buffer to orphan frags from 2387 * @gfp_mask: allocation mask for replacement pages 2388 * 2389 * For each frag in the SKB which needs a destructor (i.e. has an 2390 * owner) create a copy of that frag and release the original 2391 * page by calling the destructor. 2392 */ 2393static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2394{ 2395 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2396 return 0; 2397 return skb_copy_ubufs(skb, gfp_mask); 2398} 2399 2400/** 2401 * __skb_queue_purge - empty a list 2402 * @list: list to empty 2403 * 2404 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2405 * the list and one reference dropped. This function does not take the 2406 * list lock and the caller must hold the relevant locks to use it. 2407 */ 2408void skb_queue_purge(struct sk_buff_head *list); 2409static inline void __skb_queue_purge(struct sk_buff_head *list) 2410{ 2411 struct sk_buff *skb; 2412 while ((skb = __skb_dequeue(list)) != NULL) 2413 kfree_skb(skb); 2414} 2415 2416void skb_rbtree_purge(struct rb_root *root); 2417 2418void *netdev_alloc_frag(unsigned int fragsz); 2419 2420struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2421 gfp_t gfp_mask); 2422 2423/** 2424 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2425 * @dev: network device to receive on 2426 * @length: length to allocate 2427 * 2428 * Allocate a new &sk_buff and assign it a usage count of one. The 2429 * buffer has unspecified headroom built in. Users should allocate 2430 * the headroom they think they need without accounting for the 2431 * built in space. The built in space is used for optimisations. 2432 * 2433 * %NULL is returned if there is no free memory. Although this function 2434 * allocates memory it can be called from an interrupt. 2435 */ 2436static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2437 unsigned int length) 2438{ 2439 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2440} 2441 2442/* legacy helper around __netdev_alloc_skb() */ 2443static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2444 gfp_t gfp_mask) 2445{ 2446 return __netdev_alloc_skb(NULL, length, gfp_mask); 2447} 2448 2449/* legacy helper around netdev_alloc_skb() */ 2450static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2451{ 2452 return netdev_alloc_skb(NULL, length); 2453} 2454 2455 2456static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2457 unsigned int length, gfp_t gfp) 2458{ 2459 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2460 2461 if (NET_IP_ALIGN && skb) 2462 skb_reserve(skb, NET_IP_ALIGN); 2463 return skb; 2464} 2465 2466static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2467 unsigned int length) 2468{ 2469 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2470} 2471 2472static inline void skb_free_frag(void *addr) 2473{ 2474 __free_page_frag(addr); 2475} 2476 2477void *napi_alloc_frag(unsigned int fragsz); 2478struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2479 unsigned int length, gfp_t gfp_mask); 2480static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2481 unsigned int length) 2482{ 2483 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2484} 2485void napi_consume_skb(struct sk_buff *skb, int budget); 2486 2487void __kfree_skb_flush(void); 2488void __kfree_skb_defer(struct sk_buff *skb); 2489 2490/** 2491 * __dev_alloc_pages - allocate page for network Rx 2492 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2493 * @order: size of the allocation 2494 * 2495 * Allocate a new page. 2496 * 2497 * %NULL is returned if there is no free memory. 2498*/ 2499static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2500 unsigned int order) 2501{ 2502 /* This piece of code contains several assumptions. 2503 * 1. This is for device Rx, therefor a cold page is preferred. 2504 * 2. The expectation is the user wants a compound page. 2505 * 3. If requesting a order 0 page it will not be compound 2506 * due to the check to see if order has a value in prep_new_page 2507 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2508 * code in gfp_to_alloc_flags that should be enforcing this. 2509 */ 2510 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2511 2512 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2513} 2514 2515static inline struct page *dev_alloc_pages(unsigned int order) 2516{ 2517 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 2518} 2519 2520/** 2521 * __dev_alloc_page - allocate a page for network Rx 2522 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2523 * 2524 * Allocate a new page. 2525 * 2526 * %NULL is returned if there is no free memory. 2527 */ 2528static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2529{ 2530 return __dev_alloc_pages(gfp_mask, 0); 2531} 2532 2533static inline struct page *dev_alloc_page(void) 2534{ 2535 return dev_alloc_pages(0); 2536} 2537 2538/** 2539 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2540 * @page: The page that was allocated from skb_alloc_page 2541 * @skb: The skb that may need pfmemalloc set 2542 */ 2543static inline void skb_propagate_pfmemalloc(struct page *page, 2544 struct sk_buff *skb) 2545{ 2546 if (page_is_pfmemalloc(page)) 2547 skb->pfmemalloc = true; 2548} 2549 2550/** 2551 * skb_frag_page - retrieve the page referred to by a paged fragment 2552 * @frag: the paged fragment 2553 * 2554 * Returns the &struct page associated with @frag. 2555 */ 2556static inline struct page *skb_frag_page(const skb_frag_t *frag) 2557{ 2558 return frag->page.p; 2559} 2560 2561/** 2562 * __skb_frag_ref - take an addition reference on a paged fragment. 2563 * @frag: the paged fragment 2564 * 2565 * Takes an additional reference on the paged fragment @frag. 2566 */ 2567static inline void __skb_frag_ref(skb_frag_t *frag) 2568{ 2569 get_page(skb_frag_page(frag)); 2570} 2571 2572/** 2573 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2574 * @skb: the buffer 2575 * @f: the fragment offset. 2576 * 2577 * Takes an additional reference on the @f'th paged fragment of @skb. 2578 */ 2579static inline void skb_frag_ref(struct sk_buff *skb, int f) 2580{ 2581 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2582} 2583 2584/** 2585 * __skb_frag_unref - release a reference on a paged fragment. 2586 * @frag: the paged fragment 2587 * 2588 * Releases a reference on the paged fragment @frag. 2589 */ 2590static inline void __skb_frag_unref(skb_frag_t *frag) 2591{ 2592 put_page(skb_frag_page(frag)); 2593} 2594 2595/** 2596 * skb_frag_unref - release a reference on a paged fragment of an skb. 2597 * @skb: the buffer 2598 * @f: the fragment offset 2599 * 2600 * Releases a reference on the @f'th paged fragment of @skb. 2601 */ 2602static inline void skb_frag_unref(struct sk_buff *skb, int f) 2603{ 2604 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2605} 2606 2607/** 2608 * skb_frag_address - gets the address of the data contained in a paged fragment 2609 * @frag: the paged fragment buffer 2610 * 2611 * Returns the address of the data within @frag. The page must already 2612 * be mapped. 2613 */ 2614static inline void *skb_frag_address(const skb_frag_t *frag) 2615{ 2616 return page_address(skb_frag_page(frag)) + frag->page_offset; 2617} 2618 2619/** 2620 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2621 * @frag: the paged fragment buffer 2622 * 2623 * Returns the address of the data within @frag. Checks that the page 2624 * is mapped and returns %NULL otherwise. 2625 */ 2626static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2627{ 2628 void *ptr = page_address(skb_frag_page(frag)); 2629 if (unlikely(!ptr)) 2630 return NULL; 2631 2632 return ptr + frag->page_offset; 2633} 2634 2635/** 2636 * __skb_frag_set_page - sets the page contained in a paged fragment 2637 * @frag: the paged fragment 2638 * @page: the page to set 2639 * 2640 * Sets the fragment @frag to contain @page. 2641 */ 2642static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2643{ 2644 frag->page.p = page; 2645} 2646 2647/** 2648 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2649 * @skb: the buffer 2650 * @f: the fragment offset 2651 * @page: the page to set 2652 * 2653 * Sets the @f'th fragment of @skb to contain @page. 2654 */ 2655static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2656 struct page *page) 2657{ 2658 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2659} 2660 2661bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2662 2663/** 2664 * skb_frag_dma_map - maps a paged fragment via the DMA API 2665 * @dev: the device to map the fragment to 2666 * @frag: the paged fragment to map 2667 * @offset: the offset within the fragment (starting at the 2668 * fragment's own offset) 2669 * @size: the number of bytes to map 2670 * @dir: the direction of the mapping (%PCI_DMA_*) 2671 * 2672 * Maps the page associated with @frag to @device. 2673 */ 2674static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2675 const skb_frag_t *frag, 2676 size_t offset, size_t size, 2677 enum dma_data_direction dir) 2678{ 2679 return dma_map_page(dev, skb_frag_page(frag), 2680 frag->page_offset + offset, size, dir); 2681} 2682 2683static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2684 gfp_t gfp_mask) 2685{ 2686 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2687} 2688 2689 2690static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2691 gfp_t gfp_mask) 2692{ 2693 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2694} 2695 2696 2697/** 2698 * skb_clone_writable - is the header of a clone writable 2699 * @skb: buffer to check 2700 * @len: length up to which to write 2701 * 2702 * Returns true if modifying the header part of the cloned buffer 2703 * does not requires the data to be copied. 2704 */ 2705static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2706{ 2707 return !skb_header_cloned(skb) && 2708 skb_headroom(skb) + len <= skb->hdr_len; 2709} 2710 2711static inline int skb_try_make_writable(struct sk_buff *skb, 2712 unsigned int write_len) 2713{ 2714 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 2715 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2716} 2717 2718static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2719 int cloned) 2720{ 2721 int delta = 0; 2722 2723 if (headroom > skb_headroom(skb)) 2724 delta = headroom - skb_headroom(skb); 2725 2726 if (delta || cloned) 2727 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2728 GFP_ATOMIC); 2729 return 0; 2730} 2731 2732/** 2733 * skb_cow - copy header of skb when it is required 2734 * @skb: buffer to cow 2735 * @headroom: needed headroom 2736 * 2737 * If the skb passed lacks sufficient headroom or its data part 2738 * is shared, data is reallocated. If reallocation fails, an error 2739 * is returned and original skb is not changed. 2740 * 2741 * The result is skb with writable area skb->head...skb->tail 2742 * and at least @headroom of space at head. 2743 */ 2744static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2745{ 2746 return __skb_cow(skb, headroom, skb_cloned(skb)); 2747} 2748 2749/** 2750 * skb_cow_head - skb_cow but only making the head writable 2751 * @skb: buffer to cow 2752 * @headroom: needed headroom 2753 * 2754 * This function is identical to skb_cow except that we replace the 2755 * skb_cloned check by skb_header_cloned. It should be used when 2756 * you only need to push on some header and do not need to modify 2757 * the data. 2758 */ 2759static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2760{ 2761 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2762} 2763 2764/** 2765 * skb_padto - pad an skbuff up to a minimal size 2766 * @skb: buffer to pad 2767 * @len: minimal length 2768 * 2769 * Pads up a buffer to ensure the trailing bytes exist and are 2770 * blanked. If the buffer already contains sufficient data it 2771 * is untouched. Otherwise it is extended. Returns zero on 2772 * success. The skb is freed on error. 2773 */ 2774static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2775{ 2776 unsigned int size = skb->len; 2777 if (likely(size >= len)) 2778 return 0; 2779 return skb_pad(skb, len - size); 2780} 2781 2782/** 2783 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2784 * @skb: buffer to pad 2785 * @len: minimal length 2786 * 2787 * Pads up a buffer to ensure the trailing bytes exist and are 2788 * blanked. If the buffer already contains sufficient data it 2789 * is untouched. Otherwise it is extended. Returns zero on 2790 * success. The skb is freed on error. 2791 */ 2792static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2793{ 2794 unsigned int size = skb->len; 2795 2796 if (unlikely(size < len)) { 2797 len -= size; 2798 if (skb_pad(skb, len)) 2799 return -ENOMEM; 2800 __skb_put(skb, len); 2801 } 2802 return 0; 2803} 2804 2805static inline int skb_add_data(struct sk_buff *skb, 2806 struct iov_iter *from, int copy) 2807{ 2808 const int off = skb->len; 2809 2810 if (skb->ip_summed == CHECKSUM_NONE) { 2811 __wsum csum = 0; 2812 if (csum_and_copy_from_iter(skb_put(skb, copy), copy, 2813 &csum, from) == copy) { 2814 skb->csum = csum_block_add(skb->csum, csum, off); 2815 return 0; 2816 } 2817 } else if (copy_from_iter(skb_put(skb, copy), copy, from) == copy) 2818 return 0; 2819 2820 __skb_trim(skb, off); 2821 return -EFAULT; 2822} 2823 2824static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2825 const struct page *page, int off) 2826{ 2827 if (i) { 2828 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2829 2830 return page == skb_frag_page(frag) && 2831 off == frag->page_offset + skb_frag_size(frag); 2832 } 2833 return false; 2834} 2835 2836static inline int __skb_linearize(struct sk_buff *skb) 2837{ 2838 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2839} 2840 2841/** 2842 * skb_linearize - convert paged skb to linear one 2843 * @skb: buffer to linarize 2844 * 2845 * If there is no free memory -ENOMEM is returned, otherwise zero 2846 * is returned and the old skb data released. 2847 */ 2848static inline int skb_linearize(struct sk_buff *skb) 2849{ 2850 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2851} 2852 2853/** 2854 * skb_has_shared_frag - can any frag be overwritten 2855 * @skb: buffer to test 2856 * 2857 * Return true if the skb has at least one frag that might be modified 2858 * by an external entity (as in vmsplice()/sendfile()) 2859 */ 2860static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2861{ 2862 return skb_is_nonlinear(skb) && 2863 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2864} 2865 2866/** 2867 * skb_linearize_cow - make sure skb is linear and writable 2868 * @skb: buffer to process 2869 * 2870 * If there is no free memory -ENOMEM is returned, otherwise zero 2871 * is returned and the old skb data released. 2872 */ 2873static inline int skb_linearize_cow(struct sk_buff *skb) 2874{ 2875 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2876 __skb_linearize(skb) : 0; 2877} 2878 2879static __always_inline void 2880__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 2881 unsigned int off) 2882{ 2883 if (skb->ip_summed == CHECKSUM_COMPLETE) 2884 skb->csum = csum_block_sub(skb->csum, 2885 csum_partial(start, len, 0), off); 2886 else if (skb->ip_summed == CHECKSUM_PARTIAL && 2887 skb_checksum_start_offset(skb) < 0) 2888 skb->ip_summed = CHECKSUM_NONE; 2889} 2890 2891/** 2892 * skb_postpull_rcsum - update checksum for received skb after pull 2893 * @skb: buffer to update 2894 * @start: start of data before pull 2895 * @len: length of data pulled 2896 * 2897 * After doing a pull on a received packet, you need to call this to 2898 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2899 * CHECKSUM_NONE so that it can be recomputed from scratch. 2900 */ 2901static inline void skb_postpull_rcsum(struct sk_buff *skb, 2902 const void *start, unsigned int len) 2903{ 2904 __skb_postpull_rcsum(skb, start, len, 0); 2905} 2906 2907static __always_inline void 2908__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 2909 unsigned int off) 2910{ 2911 if (skb->ip_summed == CHECKSUM_COMPLETE) 2912 skb->csum = csum_block_add(skb->csum, 2913 csum_partial(start, len, 0), off); 2914} 2915 2916/** 2917 * skb_postpush_rcsum - update checksum for received skb after push 2918 * @skb: buffer to update 2919 * @start: start of data after push 2920 * @len: length of data pushed 2921 * 2922 * After doing a push on a received packet, you need to call this to 2923 * update the CHECKSUM_COMPLETE checksum. 2924 */ 2925static inline void skb_postpush_rcsum(struct sk_buff *skb, 2926 const void *start, unsigned int len) 2927{ 2928 __skb_postpush_rcsum(skb, start, len, 0); 2929} 2930 2931unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2932 2933/** 2934 * skb_push_rcsum - push skb and update receive checksum 2935 * @skb: buffer to update 2936 * @len: length of data pulled 2937 * 2938 * This function performs an skb_push on the packet and updates 2939 * the CHECKSUM_COMPLETE checksum. It should be used on 2940 * receive path processing instead of skb_push unless you know 2941 * that the checksum difference is zero (e.g., a valid IP header) 2942 * or you are setting ip_summed to CHECKSUM_NONE. 2943 */ 2944static inline unsigned char *skb_push_rcsum(struct sk_buff *skb, 2945 unsigned int len) 2946{ 2947 skb_push(skb, len); 2948 skb_postpush_rcsum(skb, skb->data, len); 2949 return skb->data; 2950} 2951 2952/** 2953 * pskb_trim_rcsum - trim received skb and update checksum 2954 * @skb: buffer to trim 2955 * @len: new length 2956 * 2957 * This is exactly the same as pskb_trim except that it ensures the 2958 * checksum of received packets are still valid after the operation. 2959 */ 2960 2961static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2962{ 2963 if (likely(len >= skb->len)) 2964 return 0; 2965 if (skb->ip_summed == CHECKSUM_COMPLETE) 2966 skb->ip_summed = CHECKSUM_NONE; 2967 return __pskb_trim(skb, len); 2968} 2969 2970static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2971{ 2972 if (skb->ip_summed == CHECKSUM_COMPLETE) 2973 skb->ip_summed = CHECKSUM_NONE; 2974 __skb_trim(skb, len); 2975 return 0; 2976} 2977 2978static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 2979{ 2980 if (skb->ip_summed == CHECKSUM_COMPLETE) 2981 skb->ip_summed = CHECKSUM_NONE; 2982 return __skb_grow(skb, len); 2983} 2984 2985#define skb_queue_walk(queue, skb) \ 2986 for (skb = (queue)->next; \ 2987 skb != (struct sk_buff *)(queue); \ 2988 skb = skb->next) 2989 2990#define skb_queue_walk_safe(queue, skb, tmp) \ 2991 for (skb = (queue)->next, tmp = skb->next; \ 2992 skb != (struct sk_buff *)(queue); \ 2993 skb = tmp, tmp = skb->next) 2994 2995#define skb_queue_walk_from(queue, skb) \ 2996 for (; skb != (struct sk_buff *)(queue); \ 2997 skb = skb->next) 2998 2999#define skb_queue_walk_from_safe(queue, skb, tmp) \ 3000 for (tmp = skb->next; \ 3001 skb != (struct sk_buff *)(queue); \ 3002 skb = tmp, tmp = skb->next) 3003 3004#define skb_queue_reverse_walk(queue, skb) \ 3005 for (skb = (queue)->prev; \ 3006 skb != (struct sk_buff *)(queue); \ 3007 skb = skb->prev) 3008 3009#define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3010 for (skb = (queue)->prev, tmp = skb->prev; \ 3011 skb != (struct sk_buff *)(queue); \ 3012 skb = tmp, tmp = skb->prev) 3013 3014#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3015 for (tmp = skb->prev; \ 3016 skb != (struct sk_buff *)(queue); \ 3017 skb = tmp, tmp = skb->prev) 3018 3019static inline bool skb_has_frag_list(const struct sk_buff *skb) 3020{ 3021 return skb_shinfo(skb)->frag_list != NULL; 3022} 3023 3024static inline void skb_frag_list_init(struct sk_buff *skb) 3025{ 3026 skb_shinfo(skb)->frag_list = NULL; 3027} 3028 3029#define skb_walk_frags(skb, iter) \ 3030 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3031 3032 3033int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 3034 const struct sk_buff *skb); 3035struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 3036 int *peeked, int *off, int *err, 3037 struct sk_buff **last); 3038struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 3039 int *peeked, int *off, int *err); 3040struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 3041 int *err); 3042unsigned int datagram_poll(struct file *file, struct socket *sock, 3043 struct poll_table_struct *wait); 3044int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3045 struct iov_iter *to, int size); 3046static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3047 struct msghdr *msg, int size) 3048{ 3049 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3050} 3051int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3052 struct msghdr *msg); 3053int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3054 struct iov_iter *from, int len); 3055int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3056void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3057void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3058static inline void skb_free_datagram_locked(struct sock *sk, 3059 struct sk_buff *skb) 3060{ 3061 __skb_free_datagram_locked(sk, skb, 0); 3062} 3063int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3064int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3065int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3066__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3067 int len, __wsum csum); 3068int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3069 struct pipe_inode_info *pipe, unsigned int len, 3070 unsigned int flags); 3071void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3072unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3073int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3074 int len, int hlen); 3075void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3076int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3077void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3078unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 3079bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu); 3080struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3081struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3082int skb_ensure_writable(struct sk_buff *skb, int write_len); 3083int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3084int skb_vlan_pop(struct sk_buff *skb); 3085int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3086struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3087 gfp_t gfp); 3088 3089static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3090{ 3091 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3092} 3093 3094static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3095{ 3096 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3097} 3098 3099struct skb_checksum_ops { 3100 __wsum (*update)(const void *mem, int len, __wsum wsum); 3101 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3102}; 3103 3104__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3105 __wsum csum, const struct skb_checksum_ops *ops); 3106__wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3107 __wsum csum); 3108 3109static inline void * __must_check 3110__skb_header_pointer(const struct sk_buff *skb, int offset, 3111 int len, void *data, int hlen, void *buffer) 3112{ 3113 if (hlen - offset >= len) 3114 return data + offset; 3115 3116 if (!skb || 3117 skb_copy_bits(skb, offset, buffer, len) < 0) 3118 return NULL; 3119 3120 return buffer; 3121} 3122 3123static inline void * __must_check 3124skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3125{ 3126 return __skb_header_pointer(skb, offset, len, skb->data, 3127 skb_headlen(skb), buffer); 3128} 3129 3130/** 3131 * skb_needs_linearize - check if we need to linearize a given skb 3132 * depending on the given device features. 3133 * @skb: socket buffer to check 3134 * @features: net device features 3135 * 3136 * Returns true if either: 3137 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3138 * 2. skb is fragmented and the device does not support SG. 3139 */ 3140static inline bool skb_needs_linearize(struct sk_buff *skb, 3141 netdev_features_t features) 3142{ 3143 return skb_is_nonlinear(skb) && 3144 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3145 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3146} 3147 3148static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3149 void *to, 3150 const unsigned int len) 3151{ 3152 memcpy(to, skb->data, len); 3153} 3154 3155static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3156 const int offset, void *to, 3157 const unsigned int len) 3158{ 3159 memcpy(to, skb->data + offset, len); 3160} 3161 3162static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3163 const void *from, 3164 const unsigned int len) 3165{ 3166 memcpy(skb->data, from, len); 3167} 3168 3169static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3170 const int offset, 3171 const void *from, 3172 const unsigned int len) 3173{ 3174 memcpy(skb->data + offset, from, len); 3175} 3176 3177void skb_init(void); 3178 3179static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3180{ 3181 return skb->tstamp; 3182} 3183 3184/** 3185 * skb_get_timestamp - get timestamp from a skb 3186 * @skb: skb to get stamp from 3187 * @stamp: pointer to struct timeval to store stamp in 3188 * 3189 * Timestamps are stored in the skb as offsets to a base timestamp. 3190 * This function converts the offset back to a struct timeval and stores 3191 * it in stamp. 3192 */ 3193static inline void skb_get_timestamp(const struct sk_buff *skb, 3194 struct timeval *stamp) 3195{ 3196 *stamp = ktime_to_timeval(skb->tstamp); 3197} 3198 3199static inline void skb_get_timestampns(const struct sk_buff *skb, 3200 struct timespec *stamp) 3201{ 3202 *stamp = ktime_to_timespec(skb->tstamp); 3203} 3204 3205static inline void __net_timestamp(struct sk_buff *skb) 3206{ 3207 skb->tstamp = ktime_get_real(); 3208} 3209 3210static inline ktime_t net_timedelta(ktime_t t) 3211{ 3212 return ktime_sub(ktime_get_real(), t); 3213} 3214 3215static inline ktime_t net_invalid_timestamp(void) 3216{ 3217 return ktime_set(0, 0); 3218} 3219 3220struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3221 3222#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3223 3224void skb_clone_tx_timestamp(struct sk_buff *skb); 3225bool skb_defer_rx_timestamp(struct sk_buff *skb); 3226 3227#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3228 3229static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3230{ 3231} 3232 3233static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3234{ 3235 return false; 3236} 3237 3238#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3239 3240/** 3241 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3242 * 3243 * PHY drivers may accept clones of transmitted packets for 3244 * timestamping via their phy_driver.txtstamp method. These drivers 3245 * must call this function to return the skb back to the stack with a 3246 * timestamp. 3247 * 3248 * @skb: clone of the the original outgoing packet 3249 * @hwtstamps: hardware time stamps 3250 * 3251 */ 3252void skb_complete_tx_timestamp(struct sk_buff *skb, 3253 struct skb_shared_hwtstamps *hwtstamps); 3254 3255void __skb_tstamp_tx(struct sk_buff *orig_skb, 3256 struct skb_shared_hwtstamps *hwtstamps, 3257 struct sock *sk, int tstype); 3258 3259/** 3260 * skb_tstamp_tx - queue clone of skb with send time stamps 3261 * @orig_skb: the original outgoing packet 3262 * @hwtstamps: hardware time stamps, may be NULL if not available 3263 * 3264 * If the skb has a socket associated, then this function clones the 3265 * skb (thus sharing the actual data and optional structures), stores 3266 * the optional hardware time stamping information (if non NULL) or 3267 * generates a software time stamp (otherwise), then queues the clone 3268 * to the error queue of the socket. Errors are silently ignored. 3269 */ 3270void skb_tstamp_tx(struct sk_buff *orig_skb, 3271 struct skb_shared_hwtstamps *hwtstamps); 3272 3273static inline void sw_tx_timestamp(struct sk_buff *skb) 3274{ 3275 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 3276 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 3277 skb_tstamp_tx(skb, NULL); 3278} 3279 3280/** 3281 * skb_tx_timestamp() - Driver hook for transmit timestamping 3282 * 3283 * Ethernet MAC Drivers should call this function in their hard_xmit() 3284 * function immediately before giving the sk_buff to the MAC hardware. 3285 * 3286 * Specifically, one should make absolutely sure that this function is 3287 * called before TX completion of this packet can trigger. Otherwise 3288 * the packet could potentially already be freed. 3289 * 3290 * @skb: A socket buffer. 3291 */ 3292static inline void skb_tx_timestamp(struct sk_buff *skb) 3293{ 3294 skb_clone_tx_timestamp(skb); 3295 sw_tx_timestamp(skb); 3296} 3297 3298/** 3299 * skb_complete_wifi_ack - deliver skb with wifi status 3300 * 3301 * @skb: the original outgoing packet 3302 * @acked: ack status 3303 * 3304 */ 3305void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3306 3307__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3308__sum16 __skb_checksum_complete(struct sk_buff *skb); 3309 3310static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3311{ 3312 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3313 skb->csum_valid || 3314 (skb->ip_summed == CHECKSUM_PARTIAL && 3315 skb_checksum_start_offset(skb) >= 0)); 3316} 3317 3318/** 3319 * skb_checksum_complete - Calculate checksum of an entire packet 3320 * @skb: packet to process 3321 * 3322 * This function calculates the checksum over the entire packet plus 3323 * the value of skb->csum. The latter can be used to supply the 3324 * checksum of a pseudo header as used by TCP/UDP. It returns the 3325 * checksum. 3326 * 3327 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3328 * this function can be used to verify that checksum on received 3329 * packets. In that case the function should return zero if the 3330 * checksum is correct. In particular, this function will return zero 3331 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3332 * hardware has already verified the correctness of the checksum. 3333 */ 3334static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3335{ 3336 return skb_csum_unnecessary(skb) ? 3337 0 : __skb_checksum_complete(skb); 3338} 3339 3340static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3341{ 3342 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3343 if (skb->csum_level == 0) 3344 skb->ip_summed = CHECKSUM_NONE; 3345 else 3346 skb->csum_level--; 3347 } 3348} 3349 3350static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3351{ 3352 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3353 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3354 skb->csum_level++; 3355 } else if (skb->ip_summed == CHECKSUM_NONE) { 3356 skb->ip_summed = CHECKSUM_UNNECESSARY; 3357 skb->csum_level = 0; 3358 } 3359} 3360 3361static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 3362{ 3363 /* Mark current checksum as bad (typically called from GRO 3364 * path). In the case that ip_summed is CHECKSUM_NONE 3365 * this must be the first checksum encountered in the packet. 3366 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 3367 * checksum after the last one validated. For UDP, a zero 3368 * checksum can not be marked as bad. 3369 */ 3370 3371 if (skb->ip_summed == CHECKSUM_NONE || 3372 skb->ip_summed == CHECKSUM_UNNECESSARY) 3373 skb->csum_bad = 1; 3374} 3375 3376/* Check if we need to perform checksum complete validation. 3377 * 3378 * Returns true if checksum complete is needed, false otherwise 3379 * (either checksum is unnecessary or zero checksum is allowed). 3380 */ 3381static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3382 bool zero_okay, 3383 __sum16 check) 3384{ 3385 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3386 skb->csum_valid = 1; 3387 __skb_decr_checksum_unnecessary(skb); 3388 return false; 3389 } 3390 3391 return true; 3392} 3393 3394/* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3395 * in checksum_init. 3396 */ 3397#define CHECKSUM_BREAK 76 3398 3399/* Unset checksum-complete 3400 * 3401 * Unset checksum complete can be done when packet is being modified 3402 * (uncompressed for instance) and checksum-complete value is 3403 * invalidated. 3404 */ 3405static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3406{ 3407 if (skb->ip_summed == CHECKSUM_COMPLETE) 3408 skb->ip_summed = CHECKSUM_NONE; 3409} 3410 3411/* Validate (init) checksum based on checksum complete. 3412 * 3413 * Return values: 3414 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3415 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3416 * checksum is stored in skb->csum for use in __skb_checksum_complete 3417 * non-zero: value of invalid checksum 3418 * 3419 */ 3420static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3421 bool complete, 3422 __wsum psum) 3423{ 3424 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3425 if (!csum_fold(csum_add(psum, skb->csum))) { 3426 skb->csum_valid = 1; 3427 return 0; 3428 } 3429 } else if (skb->csum_bad) { 3430 /* ip_summed == CHECKSUM_NONE in this case */ 3431 return (__force __sum16)1; 3432 } 3433 3434 skb->csum = psum; 3435 3436 if (complete || skb->len <= CHECKSUM_BREAK) { 3437 __sum16 csum; 3438 3439 csum = __skb_checksum_complete(skb); 3440 skb->csum_valid = !csum; 3441 return csum; 3442 } 3443 3444 return 0; 3445} 3446 3447static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3448{ 3449 return 0; 3450} 3451 3452/* Perform checksum validate (init). Note that this is a macro since we only 3453 * want to calculate the pseudo header which is an input function if necessary. 3454 * First we try to validate without any computation (checksum unnecessary) and 3455 * then calculate based on checksum complete calling the function to compute 3456 * pseudo header. 3457 * 3458 * Return values: 3459 * 0: checksum is validated or try to in skb_checksum_complete 3460 * non-zero: value of invalid checksum 3461 */ 3462#define __skb_checksum_validate(skb, proto, complete, \ 3463 zero_okay, check, compute_pseudo) \ 3464({ \ 3465 __sum16 __ret = 0; \ 3466 skb->csum_valid = 0; \ 3467 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3468 __ret = __skb_checksum_validate_complete(skb, \ 3469 complete, compute_pseudo(skb, proto)); \ 3470 __ret; \ 3471}) 3472 3473#define skb_checksum_init(skb, proto, compute_pseudo) \ 3474 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3475 3476#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3477 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3478 3479#define skb_checksum_validate(skb, proto, compute_pseudo) \ 3480 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3481 3482#define skb_checksum_validate_zero_check(skb, proto, check, \ 3483 compute_pseudo) \ 3484 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3485 3486#define skb_checksum_simple_validate(skb) \ 3487 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3488 3489static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3490{ 3491 return (skb->ip_summed == CHECKSUM_NONE && 3492 skb->csum_valid && !skb->csum_bad); 3493} 3494 3495static inline void __skb_checksum_convert(struct sk_buff *skb, 3496 __sum16 check, __wsum pseudo) 3497{ 3498 skb->csum = ~pseudo; 3499 skb->ip_summed = CHECKSUM_COMPLETE; 3500} 3501 3502#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3503do { \ 3504 if (__skb_checksum_convert_check(skb)) \ 3505 __skb_checksum_convert(skb, check, \ 3506 compute_pseudo(skb, proto)); \ 3507} while (0) 3508 3509static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3510 u16 start, u16 offset) 3511{ 3512 skb->ip_summed = CHECKSUM_PARTIAL; 3513 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3514 skb->csum_offset = offset - start; 3515} 3516 3517/* Update skbuf and packet to reflect the remote checksum offload operation. 3518 * When called, ptr indicates the starting point for skb->csum when 3519 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3520 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3521 */ 3522static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3523 int start, int offset, bool nopartial) 3524{ 3525 __wsum delta; 3526 3527 if (!nopartial) { 3528 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3529 return; 3530 } 3531 3532 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3533 __skb_checksum_complete(skb); 3534 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3535 } 3536 3537 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3538 3539 /* Adjust skb->csum since we changed the packet */ 3540 skb->csum = csum_add(skb->csum, delta); 3541} 3542 3543#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3544void nf_conntrack_destroy(struct nf_conntrack *nfct); 3545static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3546{ 3547 if (nfct && atomic_dec_and_test(&nfct->use)) 3548 nf_conntrack_destroy(nfct); 3549} 3550static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3551{ 3552 if (nfct) 3553 atomic_inc(&nfct->use); 3554} 3555#endif 3556#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3557static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3558{ 3559 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3560 kfree(nf_bridge); 3561} 3562static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3563{ 3564 if (nf_bridge) 3565 atomic_inc(&nf_bridge->use); 3566} 3567#endif /* CONFIG_BRIDGE_NETFILTER */ 3568static inline void nf_reset(struct sk_buff *skb) 3569{ 3570#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3571 nf_conntrack_put(skb->nfct); 3572 skb->nfct = NULL; 3573#endif 3574#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3575 nf_bridge_put(skb->nf_bridge); 3576 skb->nf_bridge = NULL; 3577#endif 3578} 3579 3580static inline void nf_reset_trace(struct sk_buff *skb) 3581{ 3582#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3583 skb->nf_trace = 0; 3584#endif 3585} 3586 3587/* Note: This doesn't put any conntrack and bridge info in dst. */ 3588static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3589 bool copy) 3590{ 3591#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3592 dst->nfct = src->nfct; 3593 nf_conntrack_get(src->nfct); 3594 if (copy) 3595 dst->nfctinfo = src->nfctinfo; 3596#endif 3597#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3598 dst->nf_bridge = src->nf_bridge; 3599 nf_bridge_get(src->nf_bridge); 3600#endif 3601#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3602 if (copy) 3603 dst->nf_trace = src->nf_trace; 3604#endif 3605} 3606 3607static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3608{ 3609#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3610 nf_conntrack_put(dst->nfct); 3611#endif 3612#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3613 nf_bridge_put(dst->nf_bridge); 3614#endif 3615 __nf_copy(dst, src, true); 3616} 3617 3618#ifdef CONFIG_NETWORK_SECMARK 3619static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3620{ 3621 to->secmark = from->secmark; 3622} 3623 3624static inline void skb_init_secmark(struct sk_buff *skb) 3625{ 3626 skb->secmark = 0; 3627} 3628#else 3629static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3630{ } 3631 3632static inline void skb_init_secmark(struct sk_buff *skb) 3633{ } 3634#endif 3635 3636static inline bool skb_irq_freeable(const struct sk_buff *skb) 3637{ 3638 return !skb->destructor && 3639#if IS_ENABLED(CONFIG_XFRM) 3640 !skb->sp && 3641#endif 3642#if IS_ENABLED(CONFIG_NF_CONNTRACK) 3643 !skb->nfct && 3644#endif 3645 !skb->_skb_refdst && 3646 !skb_has_frag_list(skb); 3647} 3648 3649static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3650{ 3651 skb->queue_mapping = queue_mapping; 3652} 3653 3654static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3655{ 3656 return skb->queue_mapping; 3657} 3658 3659static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3660{ 3661 to->queue_mapping = from->queue_mapping; 3662} 3663 3664static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3665{ 3666 skb->queue_mapping = rx_queue + 1; 3667} 3668 3669static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3670{ 3671 return skb->queue_mapping - 1; 3672} 3673 3674static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3675{ 3676 return skb->queue_mapping != 0; 3677} 3678 3679static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3680{ 3681#ifdef CONFIG_XFRM 3682 return skb->sp; 3683#else 3684 return NULL; 3685#endif 3686} 3687 3688/* Keeps track of mac header offset relative to skb->head. 3689 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3690 * For non-tunnel skb it points to skb_mac_header() and for 3691 * tunnel skb it points to outer mac header. 3692 * Keeps track of level of encapsulation of network headers. 3693 */ 3694struct skb_gso_cb { 3695 union { 3696 int mac_offset; 3697 int data_offset; 3698 }; 3699 int encap_level; 3700 __wsum csum; 3701 __u16 csum_start; 3702}; 3703#define SKB_SGO_CB_OFFSET 32 3704#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3705 3706static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3707{ 3708 return (skb_mac_header(inner_skb) - inner_skb->head) - 3709 SKB_GSO_CB(inner_skb)->mac_offset; 3710} 3711 3712static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3713{ 3714 int new_headroom, headroom; 3715 int ret; 3716 3717 headroom = skb_headroom(skb); 3718 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3719 if (ret) 3720 return ret; 3721 3722 new_headroom = skb_headroom(skb); 3723 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3724 return 0; 3725} 3726 3727static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 3728{ 3729 /* Do not update partial checksums if remote checksum is enabled. */ 3730 if (skb->remcsum_offload) 3731 return; 3732 3733 SKB_GSO_CB(skb)->csum = res; 3734 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 3735} 3736 3737/* Compute the checksum for a gso segment. First compute the checksum value 3738 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3739 * then add in skb->csum (checksum from csum_start to end of packet). 3740 * skb->csum and csum_start are then updated to reflect the checksum of the 3741 * resultant packet starting from the transport header-- the resultant checksum 3742 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3743 * header. 3744 */ 3745static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3746{ 3747 unsigned char *csum_start = skb_transport_header(skb); 3748 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 3749 __wsum partial = SKB_GSO_CB(skb)->csum; 3750 3751 SKB_GSO_CB(skb)->csum = res; 3752 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 3753 3754 return csum_fold(csum_partial(csum_start, plen, partial)); 3755} 3756 3757static inline bool skb_is_gso(const struct sk_buff *skb) 3758{ 3759 return skb_shinfo(skb)->gso_size; 3760} 3761 3762/* Note: Should be called only if skb_is_gso(skb) is true */ 3763static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3764{ 3765 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3766} 3767 3768static inline void skb_gso_reset(struct sk_buff *skb) 3769{ 3770 skb_shinfo(skb)->gso_size = 0; 3771 skb_shinfo(skb)->gso_segs = 0; 3772 skb_shinfo(skb)->gso_type = 0; 3773} 3774 3775void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3776 3777static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3778{ 3779 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3780 * wanted then gso_type will be set. */ 3781 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3782 3783 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3784 unlikely(shinfo->gso_type == 0)) { 3785 __skb_warn_lro_forwarding(skb); 3786 return true; 3787 } 3788 return false; 3789} 3790 3791static inline void skb_forward_csum(struct sk_buff *skb) 3792{ 3793 /* Unfortunately we don't support this one. Any brave souls? */ 3794 if (skb->ip_summed == CHECKSUM_COMPLETE) 3795 skb->ip_summed = CHECKSUM_NONE; 3796} 3797 3798/** 3799 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3800 * @skb: skb to check 3801 * 3802 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3803 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3804 * use this helper, to document places where we make this assertion. 3805 */ 3806static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3807{ 3808#ifdef DEBUG 3809 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3810#endif 3811} 3812 3813bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3814 3815int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3816struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 3817 unsigned int transport_len, 3818 __sum16(*skb_chkf)(struct sk_buff *skb)); 3819 3820/** 3821 * skb_head_is_locked - Determine if the skb->head is locked down 3822 * @skb: skb to check 3823 * 3824 * The head on skbs build around a head frag can be removed if they are 3825 * not cloned. This function returns true if the skb head is locked down 3826 * due to either being allocated via kmalloc, or by being a clone with 3827 * multiple references to the head. 3828 */ 3829static inline bool skb_head_is_locked(const struct sk_buff *skb) 3830{ 3831 return !skb->head_frag || skb_cloned(skb); 3832} 3833 3834/** 3835 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3836 * 3837 * @skb: GSO skb 3838 * 3839 * skb_gso_network_seglen is used to determine the real size of the 3840 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3841 * 3842 * The MAC/L2 header is not accounted for. 3843 */ 3844static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3845{ 3846 unsigned int hdr_len = skb_transport_header(skb) - 3847 skb_network_header(skb); 3848 return hdr_len + skb_gso_transport_seglen(skb); 3849} 3850 3851/* Local Checksum Offload. 3852 * Compute outer checksum based on the assumption that the 3853 * inner checksum will be offloaded later. 3854 * See Documentation/networking/checksum-offloads.txt for 3855 * explanation of how this works. 3856 * Fill in outer checksum adjustment (e.g. with sum of outer 3857 * pseudo-header) before calling. 3858 * Also ensure that inner checksum is in linear data area. 3859 */ 3860static inline __wsum lco_csum(struct sk_buff *skb) 3861{ 3862 unsigned char *csum_start = skb_checksum_start(skb); 3863 unsigned char *l4_hdr = skb_transport_header(skb); 3864 __wsum partial; 3865 3866 /* Start with complement of inner checksum adjustment */ 3867 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 3868 skb->csum_offset)); 3869 3870 /* Add in checksum of our headers (incl. outer checksum 3871 * adjustment filled in by caller) and return result. 3872 */ 3873 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 3874} 3875 3876#endif /* __KERNEL__ */ 3877#endif /* _LINUX_SKBUFF_H */