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