at v4.11 115 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 <linux/sched/clock.h> 38#include <net/flow_dissector.h> 39#include <linux/splice.h> 40#include <linux/in6.h> 41#include <linux/if_packet.h> 42#include <net/flow.h> 43 44/* The interface for checksum offload between the stack and networking drivers 45 * is as follows... 46 * 47 * A. IP checksum related features 48 * 49 * Drivers advertise checksum offload capabilities in the features of a device. 50 * From the stack's point of view these are capabilities offered by the driver, 51 * a driver typically only advertises features that it is capable of offloading 52 * to its device. 53 * 54 * The checksum related features are: 55 * 56 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one 57 * IP (one's complement) checksum for any combination 58 * of protocols or protocol layering. The checksum is 59 * computed and set in a packet per the CHECKSUM_PARTIAL 60 * interface (see below). 61 * 62 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain 63 * TCP or UDP packets over IPv4. These are specifically 64 * unencapsulated packets of the form IPv4|TCP or 65 * IPv4|UDP where the Protocol field in the IPv4 header 66 * is TCP or UDP. The IPv4 header may contain IP options 67 * This feature cannot be set in features for a device 68 * with NETIF_F_HW_CSUM also set. This feature is being 69 * DEPRECATED (see below). 70 * 71 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain 72 * TCP or UDP packets over IPv6. These are specifically 73 * unencapsulated packets of the form IPv6|TCP or 74 * IPv4|UDP where the Next Header field in the IPv6 75 * header is either TCP or UDP. IPv6 extension headers 76 * are not supported with this feature. This feature 77 * cannot be set in features for a device with 78 * NETIF_F_HW_CSUM also set. This feature is being 79 * DEPRECATED (see below). 80 * 81 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload. 82 * This flag is used only used to disable the RX checksum 83 * feature for a device. The stack will accept receive 84 * checksum indication in packets received on a device 85 * regardless of whether NETIF_F_RXCSUM is set. 86 * 87 * B. Checksumming of received packets by device. Indication of checksum 88 * verification is in set skb->ip_summed. Possible values are: 89 * 90 * CHECKSUM_NONE: 91 * 92 * Device did not checksum this packet e.g. due to lack of capabilities. 93 * The packet contains full (though not verified) checksum in packet but 94 * not in skb->csum. Thus, skb->csum is undefined in this case. 95 * 96 * CHECKSUM_UNNECESSARY: 97 * 98 * The hardware you're dealing with doesn't calculate the full checksum 99 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 100 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 101 * if their checksums are okay. skb->csum is still undefined in this case 102 * though. A driver or device must never modify the checksum field in the 103 * packet even if checksum is verified. 104 * 105 * CHECKSUM_UNNECESSARY is applicable to following protocols: 106 * TCP: IPv6 and IPv4. 107 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 108 * zero UDP checksum for either IPv4 or IPv6, the networking stack 109 * may perform further validation in this case. 110 * GRE: only if the checksum is present in the header. 111 * SCTP: indicates the CRC in SCTP header has been validated. 112 * 113 * skb->csum_level indicates the number of consecutive checksums found in 114 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 115 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 116 * and a device is able to verify the checksums for UDP (possibly zero), 117 * GRE (checksum flag is set), and TCP-- skb->csum_level would be set to 118 * two. If the device were only able to verify the UDP checksum and not 119 * GRE, either because it doesn't support GRE checksum of because GRE 120 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 121 * not considered in this case). 122 * 123 * CHECKSUM_COMPLETE: 124 * 125 * This is the most generic way. The device supplied checksum of the _whole_ 126 * packet as seen by netif_rx() and fills out in skb->csum. Meaning, the 127 * hardware doesn't need to parse L3/L4 headers to implement this. 128 * 129 * Note: Even if device supports only some protocols, but is able to produce 130 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 131 * 132 * CHECKSUM_PARTIAL: 133 * 134 * A checksum is set up to be offloaded to a device as described in the 135 * output description for CHECKSUM_PARTIAL. This may occur on a packet 136 * received directly from another Linux OS, e.g., a virtualized Linux kernel 137 * on the same host, or it may be set in the input path in GRO or remote 138 * checksum offload. For the purposes of checksum verification, the checksum 139 * referred to by skb->csum_start + skb->csum_offset and any preceding 140 * checksums in the packet are considered verified. Any checksums in the 141 * packet that are after the checksum being offloaded are not considered to 142 * be verified. 143 * 144 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload 145 * in the skb->ip_summed for a packet. Values are: 146 * 147 * CHECKSUM_PARTIAL: 148 * 149 * The driver is required to checksum the packet as seen by hard_start_xmit() 150 * from skb->csum_start up to the end, and to record/write the checksum at 151 * offset skb->csum_start + skb->csum_offset. A driver may verify that the 152 * csum_start and csum_offset values are valid values given the length and 153 * offset of the packet, however they should not attempt to validate that the 154 * checksum refers to a legitimate transport layer checksum-- it is the 155 * purview of the stack to validate that csum_start and csum_offset are set 156 * correctly. 157 * 158 * When the stack requests checksum offload for a packet, the driver MUST 159 * ensure that the checksum is set correctly. A driver can either offload the 160 * checksum calculation to the device, or call skb_checksum_help (in the case 161 * that the device does not support offload for a particular checksum). 162 * 163 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of 164 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate 165 * checksum offload capability. If a device has limited checksum capabilities 166 * (for instance can only perform NETIF_F_IP_CSUM or NETIF_F_IPV6_CSUM as 167 * described above) a helper function can be called to resolve 168 * CHECKSUM_PARTIAL. The helper functions are skb_csum_off_chk*. The helper 169 * function takes a spec argument that describes the protocol layer that is 170 * supported for checksum offload and can be called for each packet. If a 171 * packet does not match the specification for offload, skb_checksum_help 172 * is called to resolve the checksum. 173 * 174 * CHECKSUM_NONE: 175 * 176 * The skb was already checksummed by the protocol, or a checksum is not 177 * required. 178 * 179 * CHECKSUM_UNNECESSARY: 180 * 181 * This has the same meaning on as CHECKSUM_NONE for checksum offload on 182 * output. 183 * 184 * CHECKSUM_COMPLETE: 185 * Not used in checksum output. If a driver observes a packet with this value 186 * set in skbuff, if should treat as CHECKSUM_NONE being set. 187 * 188 * D. Non-IP checksum (CRC) offloads 189 * 190 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of 191 * offloading the SCTP CRC in a packet. To perform this offload the stack 192 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 193 * accordingly. Note the there is no indication in the skbuff that the 194 * CHECKSUM_PARTIAL refers to an SCTP checksum, a driver that supports 195 * both IP checksum offload and SCTP CRC offload must verify which offload 196 * is configured for a packet presumably by inspecting packet headers. 197 * 198 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of 199 * offloading the FCOE CRC in a packet. To perform this offload the stack 200 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 201 * accordingly. Note the there is no indication in the skbuff that the 202 * CHECKSUM_PARTIAL refers to an FCOE checksum, a driver that supports 203 * both IP checksum offload and FCOE CRC offload must verify which offload 204 * is configured for a packet presumably by inspecting packet headers. 205 * 206 * E. Checksumming on output with GSO. 207 * 208 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload 209 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 210 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as 211 * part of the GSO operation is implied. If a checksum is being offloaded 212 * with GSO then ip_summed is CHECKSUM_PARTIAL, csum_start and csum_offset 213 * are set to refer to the outermost checksum being offload (two offloaded 214 * checksums are possible with UDP encapsulation). 215 */ 216 217/* Don't change this without changing skb_csum_unnecessary! */ 218#define CHECKSUM_NONE 0 219#define CHECKSUM_UNNECESSARY 1 220#define CHECKSUM_COMPLETE 2 221#define CHECKSUM_PARTIAL 3 222 223/* Maximum value in skb->csum_level */ 224#define SKB_MAX_CSUM_LEVEL 3 225 226#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 227#define SKB_WITH_OVERHEAD(X) \ 228 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 229#define SKB_MAX_ORDER(X, ORDER) \ 230 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 231#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 232#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 233 234/* return minimum truesize of one skb containing X bytes of data */ 235#define SKB_TRUESIZE(X) ((X) + \ 236 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 237 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 238 239struct net_device; 240struct scatterlist; 241struct pipe_inode_info; 242struct iov_iter; 243struct napi_struct; 244 245#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 246struct nf_conntrack { 247 atomic_t use; 248}; 249#endif 250 251#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 252struct nf_bridge_info { 253 atomic_t use; 254 enum { 255 BRNF_PROTO_UNCHANGED, 256 BRNF_PROTO_8021Q, 257 BRNF_PROTO_PPPOE 258 } orig_proto:8; 259 u8 pkt_otherhost:1; 260 u8 in_prerouting:1; 261 u8 bridged_dnat:1; 262 __u16 frag_max_size; 263 struct net_device *physindev; 264 265 /* always valid & non-NULL from FORWARD on, for physdev match */ 266 struct net_device *physoutdev; 267 union { 268 /* prerouting: detect dnat in orig/reply direction */ 269 __be32 ipv4_daddr; 270 struct in6_addr ipv6_daddr; 271 272 /* after prerouting + nat detected: store original source 273 * mac since neigh resolution overwrites it, only used while 274 * skb is out in neigh layer. 275 */ 276 char neigh_header[8]; 277 }; 278}; 279#endif 280 281struct sk_buff_head { 282 /* These two members must be first. */ 283 struct sk_buff *next; 284 struct sk_buff *prev; 285 286 __u32 qlen; 287 spinlock_t lock; 288}; 289 290struct sk_buff; 291 292/* To allow 64K frame to be packed as single skb without frag_list we 293 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 294 * buffers which do not start on a page boundary. 295 * 296 * Since GRO uses frags we allocate at least 16 regardless of page 297 * size. 298 */ 299#if (65536/PAGE_SIZE + 1) < 16 300#define MAX_SKB_FRAGS 16UL 301#else 302#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 303#endif 304extern int sysctl_max_skb_frags; 305 306/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 307 * segment using its current segmentation instead. 308 */ 309#define GSO_BY_FRAGS 0xFFFF 310 311typedef struct skb_frag_struct skb_frag_t; 312 313struct skb_frag_struct { 314 struct { 315 struct page *p; 316 } page; 317#if (BITS_PER_LONG > 32) || (PAGE_SIZE >= 65536) 318 __u32 page_offset; 319 __u32 size; 320#else 321 __u16 page_offset; 322 __u16 size; 323#endif 324}; 325 326static inline unsigned int skb_frag_size(const skb_frag_t *frag) 327{ 328 return frag->size; 329} 330 331static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 332{ 333 frag->size = size; 334} 335 336static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 337{ 338 frag->size += delta; 339} 340 341static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 342{ 343 frag->size -= delta; 344} 345 346#define HAVE_HW_TIME_STAMP 347 348/** 349 * struct skb_shared_hwtstamps - hardware time stamps 350 * @hwtstamp: hardware time stamp transformed into duration 351 * since arbitrary point in time 352 * 353 * Software time stamps generated by ktime_get_real() are stored in 354 * skb->tstamp. 355 * 356 * hwtstamps can only be compared against other hwtstamps from 357 * the same device. 358 * 359 * This structure is attached to packets as part of the 360 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 361 */ 362struct skb_shared_hwtstamps { 363 ktime_t hwtstamp; 364}; 365 366/* Definitions for tx_flags in struct skb_shared_info */ 367enum { 368 /* generate hardware time stamp */ 369 SKBTX_HW_TSTAMP = 1 << 0, 370 371 /* generate software time stamp when queueing packet to NIC */ 372 SKBTX_SW_TSTAMP = 1 << 1, 373 374 /* device driver is going to provide hardware time stamp */ 375 SKBTX_IN_PROGRESS = 1 << 2, 376 377 /* device driver supports TX zero-copy buffers */ 378 SKBTX_DEV_ZEROCOPY = 1 << 3, 379 380 /* generate wifi status information (where possible) */ 381 SKBTX_WIFI_STATUS = 1 << 4, 382 383 /* This indicates at least one fragment might be overwritten 384 * (as in vmsplice(), sendfile() ...) 385 * If we need to compute a TX checksum, we'll need to copy 386 * all frags to avoid possible bad checksum 387 */ 388 SKBTX_SHARED_FRAG = 1 << 5, 389 390 /* generate software time stamp when entering packet scheduling */ 391 SKBTX_SCHED_TSTAMP = 1 << 6, 392}; 393 394#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 395 SKBTX_SCHED_TSTAMP) 396#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 397 398/* 399 * The callback notifies userspace to release buffers when skb DMA is done in 400 * lower device, the skb last reference should be 0 when calling this. 401 * The zerocopy_success argument is true if zero copy transmit occurred, 402 * false on data copy or out of memory error caused by data copy attempt. 403 * The ctx field is used to track device context. 404 * The desc field is used to track userspace buffer index. 405 */ 406struct ubuf_info { 407 void (*callback)(struct ubuf_info *, bool zerocopy_success); 408 void *ctx; 409 unsigned long desc; 410}; 411 412/* This data is invariant across clones and lives at 413 * the end of the header data, ie. at skb->end. 414 */ 415struct skb_shared_info { 416 unsigned char nr_frags; 417 __u8 tx_flags; 418 unsigned short gso_size; 419 /* Warning: this field is not always filled in (UFO)! */ 420 unsigned short gso_segs; 421 unsigned short gso_type; 422 struct sk_buff *frag_list; 423 struct skb_shared_hwtstamps hwtstamps; 424 u32 tskey; 425 __be32 ip6_frag_id; 426 427 /* 428 * Warning : all fields before dataref are cleared in __alloc_skb() 429 */ 430 atomic_t dataref; 431 432 /* Intermediate layers must ensure that destructor_arg 433 * remains valid until skb destructor */ 434 void * destructor_arg; 435 436 /* must be last field, see pskb_expand_head() */ 437 skb_frag_t frags[MAX_SKB_FRAGS]; 438}; 439 440/* We divide dataref into two halves. The higher 16 bits hold references 441 * to the payload part of skb->data. The lower 16 bits hold references to 442 * the entire skb->data. A clone of a headerless skb holds the length of 443 * the header in skb->hdr_len. 444 * 445 * All users must obey the rule that the skb->data reference count must be 446 * greater than or equal to the payload reference count. 447 * 448 * Holding a reference to the payload part means that the user does not 449 * care about modifications to the header part of skb->data. 450 */ 451#define SKB_DATAREF_SHIFT 16 452#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 453 454 455enum { 456 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 457 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 458 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 459}; 460 461enum { 462 SKB_GSO_TCPV4 = 1 << 0, 463 SKB_GSO_UDP = 1 << 1, 464 465 /* This indicates the skb is from an untrusted source. */ 466 SKB_GSO_DODGY = 1 << 2, 467 468 /* This indicates the tcp segment has CWR set. */ 469 SKB_GSO_TCP_ECN = 1 << 3, 470 471 SKB_GSO_TCP_FIXEDID = 1 << 4, 472 473 SKB_GSO_TCPV6 = 1 << 5, 474 475 SKB_GSO_FCOE = 1 << 6, 476 477 SKB_GSO_GRE = 1 << 7, 478 479 SKB_GSO_GRE_CSUM = 1 << 8, 480 481 SKB_GSO_IPXIP4 = 1 << 9, 482 483 SKB_GSO_IPXIP6 = 1 << 10, 484 485 SKB_GSO_UDP_TUNNEL = 1 << 11, 486 487 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 12, 488 489 SKB_GSO_PARTIAL = 1 << 13, 490 491 SKB_GSO_TUNNEL_REMCSUM = 1 << 14, 492 493 SKB_GSO_SCTP = 1 << 15, 494}; 495 496#if BITS_PER_LONG > 32 497#define NET_SKBUFF_DATA_USES_OFFSET 1 498#endif 499 500#ifdef NET_SKBUFF_DATA_USES_OFFSET 501typedef unsigned int sk_buff_data_t; 502#else 503typedef unsigned char *sk_buff_data_t; 504#endif 505 506/** 507 * struct skb_mstamp - multi resolution time stamps 508 * @stamp_us: timestamp in us resolution 509 * @stamp_jiffies: timestamp in jiffies 510 */ 511struct skb_mstamp { 512 union { 513 u64 v64; 514 struct { 515 u32 stamp_us; 516 u32 stamp_jiffies; 517 }; 518 }; 519}; 520 521/** 522 * skb_mstamp_get - get current timestamp 523 * @cl: place to store timestamps 524 */ 525static inline void skb_mstamp_get(struct skb_mstamp *cl) 526{ 527 u64 val = local_clock(); 528 529 do_div(val, NSEC_PER_USEC); 530 cl->stamp_us = (u32)val; 531 cl->stamp_jiffies = (u32)jiffies; 532} 533 534/** 535 * skb_mstamp_delta - compute the difference in usec between two skb_mstamp 536 * @t1: pointer to newest sample 537 * @t0: pointer to oldest sample 538 */ 539static inline u32 skb_mstamp_us_delta(const struct skb_mstamp *t1, 540 const struct skb_mstamp *t0) 541{ 542 s32 delta_us = t1->stamp_us - t0->stamp_us; 543 u32 delta_jiffies = t1->stamp_jiffies - t0->stamp_jiffies; 544 545 /* If delta_us is negative, this might be because interval is too big, 546 * or local_clock() drift is too big : fallback using jiffies. 547 */ 548 if (delta_us <= 0 || 549 delta_jiffies >= (INT_MAX / (USEC_PER_SEC / HZ))) 550 551 delta_us = jiffies_to_usecs(delta_jiffies); 552 553 return delta_us; 554} 555 556static inline bool skb_mstamp_after(const struct skb_mstamp *t1, 557 const struct skb_mstamp *t0) 558{ 559 s32 diff = t1->stamp_jiffies - t0->stamp_jiffies; 560 561 if (!diff) 562 diff = t1->stamp_us - t0->stamp_us; 563 return diff > 0; 564} 565 566/** 567 * struct sk_buff - socket buffer 568 * @next: Next buffer in list 569 * @prev: Previous buffer in list 570 * @tstamp: Time we arrived/left 571 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 572 * @sk: Socket we are owned by 573 * @dev: Device we arrived on/are leaving by 574 * @cb: Control buffer. Free for use by every layer. Put private vars here 575 * @_skb_refdst: destination entry (with norefcount bit) 576 * @sp: the security path, used for xfrm 577 * @len: Length of actual data 578 * @data_len: Data length 579 * @mac_len: Length of link layer header 580 * @hdr_len: writable header length of cloned skb 581 * @csum: Checksum (must include start/offset pair) 582 * @csum_start: Offset from skb->head where checksumming should start 583 * @csum_offset: Offset from csum_start where checksum should be stored 584 * @priority: Packet queueing priority 585 * @ignore_df: allow local fragmentation 586 * @cloned: Head may be cloned (check refcnt to be sure) 587 * @ip_summed: Driver fed us an IP checksum 588 * @nohdr: Payload reference only, must not modify header 589 * @pkt_type: Packet class 590 * @fclone: skbuff clone status 591 * @ipvs_property: skbuff is owned by ipvs 592 * @tc_skip_classify: do not classify packet. set by IFB device 593 * @tc_at_ingress: used within tc_classify to distinguish in/egress 594 * @tc_redirected: packet was redirected by a tc action 595 * @tc_from_ingress: if tc_redirected, tc_at_ingress at time of redirect 596 * @peeked: this packet has been seen already, so stats have been 597 * done for it, don't do them again 598 * @nf_trace: netfilter packet trace flag 599 * @protocol: Packet protocol from driver 600 * @destructor: Destruct function 601 * @_nfct: Associated connection, if any (with nfctinfo bits) 602 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 603 * @skb_iif: ifindex of device we arrived on 604 * @tc_index: Traffic control index 605 * @hash: the packet hash 606 * @queue_mapping: Queue mapping for multiqueue devices 607 * @xmit_more: More SKBs are pending for this queue 608 * @ndisc_nodetype: router type (from link layer) 609 * @ooo_okay: allow the mapping of a socket to a queue to be changed 610 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 611 * ports. 612 * @sw_hash: indicates hash was computed in software stack 613 * @wifi_acked_valid: wifi_acked was set 614 * @wifi_acked: whether frame was acked on wifi or not 615 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 616 * @dst_pending_confirm: need to confirm neighbour 617 * @napi_id: id of the NAPI struct this skb came from 618 * @secmark: security marking 619 * @mark: Generic packet mark 620 * @vlan_proto: vlan encapsulation protocol 621 * @vlan_tci: vlan tag control information 622 * @inner_protocol: Protocol (encapsulation) 623 * @inner_transport_header: Inner transport layer header (encapsulation) 624 * @inner_network_header: Network layer header (encapsulation) 625 * @inner_mac_header: Link layer header (encapsulation) 626 * @transport_header: Transport layer header 627 * @network_header: Network layer header 628 * @mac_header: Link layer header 629 * @tail: Tail pointer 630 * @end: End pointer 631 * @head: Head of buffer 632 * @data: Data head pointer 633 * @truesize: Buffer size 634 * @users: User count - see {datagram,tcp}.c 635 */ 636 637struct sk_buff { 638 union { 639 struct { 640 /* These two members must be first. */ 641 struct sk_buff *next; 642 struct sk_buff *prev; 643 644 union { 645 ktime_t tstamp; 646 struct skb_mstamp skb_mstamp; 647 }; 648 }; 649 struct rb_node rbnode; /* used in netem & tcp stack */ 650 }; 651 struct sock *sk; 652 653 union { 654 struct net_device *dev; 655 /* Some protocols might use this space to store information, 656 * while device pointer would be NULL. 657 * UDP receive path is one user. 658 */ 659 unsigned long dev_scratch; 660 }; 661 /* 662 * This is the control buffer. It is free to use for every 663 * layer. Please put your private variables there. If you 664 * want to keep them across layers you have to do a skb_clone() 665 * first. This is owned by whoever has the skb queued ATM. 666 */ 667 char cb[48] __aligned(8); 668 669 unsigned long _skb_refdst; 670 void (*destructor)(struct sk_buff *skb); 671#ifdef CONFIG_XFRM 672 struct sec_path *sp; 673#endif 674#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 675 unsigned long _nfct; 676#endif 677#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 678 struct nf_bridge_info *nf_bridge; 679#endif 680 unsigned int len, 681 data_len; 682 __u16 mac_len, 683 hdr_len; 684 685 /* Following fields are _not_ copied in __copy_skb_header() 686 * Note that queue_mapping is here mostly to fill a hole. 687 */ 688 kmemcheck_bitfield_begin(flags1); 689 __u16 queue_mapping; 690 691/* if you move cloned around you also must adapt those constants */ 692#ifdef __BIG_ENDIAN_BITFIELD 693#define CLONED_MASK (1 << 7) 694#else 695#define CLONED_MASK 1 696#endif 697#define CLONED_OFFSET() offsetof(struct sk_buff, __cloned_offset) 698 699 __u8 __cloned_offset[0]; 700 __u8 cloned:1, 701 nohdr:1, 702 fclone:2, 703 peeked:1, 704 head_frag:1, 705 xmit_more:1, 706 __unused:1; /* one bit hole */ 707 kmemcheck_bitfield_end(flags1); 708 709 /* fields enclosed in headers_start/headers_end are copied 710 * using a single memcpy() in __copy_skb_header() 711 */ 712 /* private: */ 713 __u32 headers_start[0]; 714 /* public: */ 715 716/* if you move pkt_type around you also must adapt those constants */ 717#ifdef __BIG_ENDIAN_BITFIELD 718#define PKT_TYPE_MAX (7 << 5) 719#else 720#define PKT_TYPE_MAX 7 721#endif 722#define PKT_TYPE_OFFSET() offsetof(struct sk_buff, __pkt_type_offset) 723 724 __u8 __pkt_type_offset[0]; 725 __u8 pkt_type:3; 726 __u8 pfmemalloc:1; 727 __u8 ignore_df:1; 728 729 __u8 nf_trace:1; 730 __u8 ip_summed:2; 731 __u8 ooo_okay:1; 732 __u8 l4_hash:1; 733 __u8 sw_hash:1; 734 __u8 wifi_acked_valid:1; 735 __u8 wifi_acked:1; 736 737 __u8 no_fcs:1; 738 /* Indicates the inner headers are valid in the skbuff. */ 739 __u8 encapsulation:1; 740 __u8 encap_hdr_csum:1; 741 __u8 csum_valid:1; 742 __u8 csum_complete_sw:1; 743 __u8 csum_level:2; 744 __u8 csum_bad:1; 745 746 __u8 dst_pending_confirm:1; 747#ifdef CONFIG_IPV6_NDISC_NODETYPE 748 __u8 ndisc_nodetype:2; 749#endif 750 __u8 ipvs_property:1; 751 __u8 inner_protocol_type:1; 752 __u8 remcsum_offload:1; 753#ifdef CONFIG_NET_SWITCHDEV 754 __u8 offload_fwd_mark:1; 755#endif 756#ifdef CONFIG_NET_CLS_ACT 757 __u8 tc_skip_classify:1; 758 __u8 tc_at_ingress:1; 759 __u8 tc_redirected:1; 760 __u8 tc_from_ingress:1; 761#endif 762 763#ifdef CONFIG_NET_SCHED 764 __u16 tc_index; /* traffic control index */ 765#endif 766 767 union { 768 __wsum csum; 769 struct { 770 __u16 csum_start; 771 __u16 csum_offset; 772 }; 773 }; 774 __u32 priority; 775 int skb_iif; 776 __u32 hash; 777 __be16 vlan_proto; 778 __u16 vlan_tci; 779#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 780 union { 781 unsigned int napi_id; 782 unsigned int sender_cpu; 783 }; 784#endif 785#ifdef CONFIG_NETWORK_SECMARK 786 __u32 secmark; 787#endif 788 789 union { 790 __u32 mark; 791 __u32 reserved_tailroom; 792 }; 793 794 union { 795 __be16 inner_protocol; 796 __u8 inner_ipproto; 797 }; 798 799 __u16 inner_transport_header; 800 __u16 inner_network_header; 801 __u16 inner_mac_header; 802 803 __be16 protocol; 804 __u16 transport_header; 805 __u16 network_header; 806 __u16 mac_header; 807 808 /* private: */ 809 __u32 headers_end[0]; 810 /* public: */ 811 812 /* These elements must be at the end, see alloc_skb() for details. */ 813 sk_buff_data_t tail; 814 sk_buff_data_t end; 815 unsigned char *head, 816 *data; 817 unsigned int truesize; 818 atomic_t users; 819}; 820 821#ifdef __KERNEL__ 822/* 823 * Handling routines are only of interest to the kernel 824 */ 825#include <linux/slab.h> 826 827 828#define SKB_ALLOC_FCLONE 0x01 829#define SKB_ALLOC_RX 0x02 830#define SKB_ALLOC_NAPI 0x04 831 832/* Returns true if the skb was allocated from PFMEMALLOC reserves */ 833static inline bool skb_pfmemalloc(const struct sk_buff *skb) 834{ 835 return unlikely(skb->pfmemalloc); 836} 837 838/* 839 * skb might have a dst pointer attached, refcounted or not. 840 * _skb_refdst low order bit is set if refcount was _not_ taken 841 */ 842#define SKB_DST_NOREF 1UL 843#define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 844 845#define SKB_NFCT_PTRMASK ~(7UL) 846/** 847 * skb_dst - returns skb dst_entry 848 * @skb: buffer 849 * 850 * Returns skb dst_entry, regardless of reference taken or not. 851 */ 852static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 853{ 854 /* If refdst was not refcounted, check we still are in a 855 * rcu_read_lock section 856 */ 857 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 858 !rcu_read_lock_held() && 859 !rcu_read_lock_bh_held()); 860 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 861} 862 863/** 864 * skb_dst_set - sets skb dst 865 * @skb: buffer 866 * @dst: dst entry 867 * 868 * Sets skb dst, assuming a reference was taken on dst and should 869 * be released by skb_dst_drop() 870 */ 871static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 872{ 873 skb->_skb_refdst = (unsigned long)dst; 874} 875 876/** 877 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 878 * @skb: buffer 879 * @dst: dst entry 880 * 881 * Sets skb dst, assuming a reference was not taken on dst. 882 * If dst entry is cached, we do not take reference and dst_release 883 * will be avoided by refdst_drop. If dst entry is not cached, we take 884 * reference, so that last dst_release can destroy the dst immediately. 885 */ 886static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 887{ 888 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 889 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 890} 891 892/** 893 * skb_dst_is_noref - Test if skb dst isn't refcounted 894 * @skb: buffer 895 */ 896static inline bool skb_dst_is_noref(const struct sk_buff *skb) 897{ 898 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 899} 900 901static inline struct rtable *skb_rtable(const struct sk_buff *skb) 902{ 903 return (struct rtable *)skb_dst(skb); 904} 905 906/* For mangling skb->pkt_type from user space side from applications 907 * such as nft, tc, etc, we only allow a conservative subset of 908 * possible pkt_types to be set. 909*/ 910static inline bool skb_pkt_type_ok(u32 ptype) 911{ 912 return ptype <= PACKET_OTHERHOST; 913} 914 915void kfree_skb(struct sk_buff *skb); 916void kfree_skb_list(struct sk_buff *segs); 917void skb_tx_error(struct sk_buff *skb); 918void consume_skb(struct sk_buff *skb); 919void __kfree_skb(struct sk_buff *skb); 920extern struct kmem_cache *skbuff_head_cache; 921 922void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 923bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 924 bool *fragstolen, int *delta_truesize); 925 926struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 927 int node); 928struct sk_buff *__build_skb(void *data, unsigned int frag_size); 929struct sk_buff *build_skb(void *data, unsigned int frag_size); 930static inline struct sk_buff *alloc_skb(unsigned int size, 931 gfp_t priority) 932{ 933 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 934} 935 936struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 937 unsigned long data_len, 938 int max_page_order, 939 int *errcode, 940 gfp_t gfp_mask); 941 942/* Layout of fast clones : [skb1][skb2][fclone_ref] */ 943struct sk_buff_fclones { 944 struct sk_buff skb1; 945 946 struct sk_buff skb2; 947 948 atomic_t fclone_ref; 949}; 950 951/** 952 * skb_fclone_busy - check if fclone is busy 953 * @sk: socket 954 * @skb: buffer 955 * 956 * Returns true if skb is a fast clone, and its clone is not freed. 957 * Some drivers call skb_orphan() in their ndo_start_xmit(), 958 * so we also check that this didnt happen. 959 */ 960static inline bool skb_fclone_busy(const struct sock *sk, 961 const struct sk_buff *skb) 962{ 963 const struct sk_buff_fclones *fclones; 964 965 fclones = container_of(skb, struct sk_buff_fclones, skb1); 966 967 return skb->fclone == SKB_FCLONE_ORIG && 968 atomic_read(&fclones->fclone_ref) > 1 && 969 fclones->skb2.sk == sk; 970} 971 972static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 973 gfp_t priority) 974{ 975 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 976} 977 978struct sk_buff *__alloc_skb_head(gfp_t priority, int node); 979static inline struct sk_buff *alloc_skb_head(gfp_t priority) 980{ 981 return __alloc_skb_head(priority, -1); 982} 983 984struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 985int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 986struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 987struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 988struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 989 gfp_t gfp_mask, bool fclone); 990static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 991 gfp_t gfp_mask) 992{ 993 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 994} 995 996int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 997struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 998 unsigned int headroom); 999struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1000 int newtailroom, gfp_t priority); 1001int skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1002 int offset, int len); 1003int skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, int offset, 1004 int len); 1005int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1006int skb_pad(struct sk_buff *skb, int pad); 1007#define dev_kfree_skb(a) consume_skb(a) 1008 1009int skb_append_datato_frags(struct sock *sk, struct sk_buff *skb, 1010 int getfrag(void *from, char *to, int offset, 1011 int len, int odd, struct sk_buff *skb), 1012 void *from, int length); 1013 1014int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1015 int offset, size_t size); 1016 1017struct skb_seq_state { 1018 __u32 lower_offset; 1019 __u32 upper_offset; 1020 __u32 frag_idx; 1021 __u32 stepped_offset; 1022 struct sk_buff *root_skb; 1023 struct sk_buff *cur_skb; 1024 __u8 *frag_data; 1025}; 1026 1027void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1028 unsigned int to, struct skb_seq_state *st); 1029unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1030 struct skb_seq_state *st); 1031void skb_abort_seq_read(struct skb_seq_state *st); 1032 1033unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1034 unsigned int to, struct ts_config *config); 1035 1036/* 1037 * Packet hash types specify the type of hash in skb_set_hash. 1038 * 1039 * Hash types refer to the protocol layer addresses which are used to 1040 * construct a packet's hash. The hashes are used to differentiate or identify 1041 * flows of the protocol layer for the hash type. Hash types are either 1042 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1043 * 1044 * Properties of hashes: 1045 * 1046 * 1) Two packets in different flows have different hash values 1047 * 2) Two packets in the same flow should have the same hash value 1048 * 1049 * A hash at a higher layer is considered to be more specific. A driver should 1050 * set the most specific hash possible. 1051 * 1052 * A driver cannot indicate a more specific hash than the layer at which a hash 1053 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1054 * 1055 * A driver may indicate a hash level which is less specific than the 1056 * actual layer the hash was computed on. For instance, a hash computed 1057 * at L4 may be considered an L3 hash. This should only be done if the 1058 * driver can't unambiguously determine that the HW computed the hash at 1059 * the higher layer. Note that the "should" in the second property above 1060 * permits this. 1061 */ 1062enum pkt_hash_types { 1063 PKT_HASH_TYPE_NONE, /* Undefined type */ 1064 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1065 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1066 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1067}; 1068 1069static inline void skb_clear_hash(struct sk_buff *skb) 1070{ 1071 skb->hash = 0; 1072 skb->sw_hash = 0; 1073 skb->l4_hash = 0; 1074} 1075 1076static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1077{ 1078 if (!skb->l4_hash) 1079 skb_clear_hash(skb); 1080} 1081 1082static inline void 1083__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1084{ 1085 skb->l4_hash = is_l4; 1086 skb->sw_hash = is_sw; 1087 skb->hash = hash; 1088} 1089 1090static inline void 1091skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1092{ 1093 /* Used by drivers to set hash from HW */ 1094 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1095} 1096 1097static inline void 1098__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1099{ 1100 __skb_set_hash(skb, hash, true, is_l4); 1101} 1102 1103void __skb_get_hash(struct sk_buff *skb); 1104u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1105u32 skb_get_poff(const struct sk_buff *skb); 1106u32 __skb_get_poff(const struct sk_buff *skb, void *data, 1107 const struct flow_keys *keys, int hlen); 1108__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1109 void *data, int hlen_proto); 1110 1111static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1112 int thoff, u8 ip_proto) 1113{ 1114 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1115} 1116 1117void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1118 const struct flow_dissector_key *key, 1119 unsigned int key_count); 1120 1121bool __skb_flow_dissect(const struct sk_buff *skb, 1122 struct flow_dissector *flow_dissector, 1123 void *target_container, 1124 void *data, __be16 proto, int nhoff, int hlen, 1125 unsigned int flags); 1126 1127static inline bool skb_flow_dissect(const struct sk_buff *skb, 1128 struct flow_dissector *flow_dissector, 1129 void *target_container, unsigned int flags) 1130{ 1131 return __skb_flow_dissect(skb, flow_dissector, target_container, 1132 NULL, 0, 0, 0, flags); 1133} 1134 1135static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1136 struct flow_keys *flow, 1137 unsigned int flags) 1138{ 1139 memset(flow, 0, sizeof(*flow)); 1140 return __skb_flow_dissect(skb, &flow_keys_dissector, flow, 1141 NULL, 0, 0, 0, flags); 1142} 1143 1144static inline bool skb_flow_dissect_flow_keys_buf(struct flow_keys *flow, 1145 void *data, __be16 proto, 1146 int nhoff, int hlen, 1147 unsigned int flags) 1148{ 1149 memset(flow, 0, sizeof(*flow)); 1150 return __skb_flow_dissect(NULL, &flow_keys_buf_dissector, flow, 1151 data, proto, nhoff, hlen, flags); 1152} 1153 1154static inline __u32 skb_get_hash(struct sk_buff *skb) 1155{ 1156 if (!skb->l4_hash && !skb->sw_hash) 1157 __skb_get_hash(skb); 1158 1159 return skb->hash; 1160} 1161 1162__u32 __skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6); 1163 1164static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1165{ 1166 if (!skb->l4_hash && !skb->sw_hash) { 1167 struct flow_keys keys; 1168 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1169 1170 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1171 } 1172 1173 return skb->hash; 1174} 1175 1176__u32 __skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl); 1177 1178static inline __u32 skb_get_hash_flowi4(struct sk_buff *skb, const struct flowi4 *fl4) 1179{ 1180 if (!skb->l4_hash && !skb->sw_hash) { 1181 struct flow_keys keys; 1182 __u32 hash = __get_hash_from_flowi4(fl4, &keys); 1183 1184 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1185 } 1186 1187 return skb->hash; 1188} 1189 1190__u32 skb_get_hash_perturb(const struct sk_buff *skb, u32 perturb); 1191 1192static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1193{ 1194 return skb->hash; 1195} 1196 1197static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1198{ 1199 to->hash = from->hash; 1200 to->sw_hash = from->sw_hash; 1201 to->l4_hash = from->l4_hash; 1202}; 1203 1204#ifdef NET_SKBUFF_DATA_USES_OFFSET 1205static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1206{ 1207 return skb->head + skb->end; 1208} 1209 1210static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1211{ 1212 return skb->end; 1213} 1214#else 1215static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1216{ 1217 return skb->end; 1218} 1219 1220static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1221{ 1222 return skb->end - skb->head; 1223} 1224#endif 1225 1226/* Internal */ 1227#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1228 1229static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1230{ 1231 return &skb_shinfo(skb)->hwtstamps; 1232} 1233 1234/** 1235 * skb_queue_empty - check if a queue is empty 1236 * @list: queue head 1237 * 1238 * Returns true if the queue is empty, false otherwise. 1239 */ 1240static inline int skb_queue_empty(const struct sk_buff_head *list) 1241{ 1242 return list->next == (const struct sk_buff *) list; 1243} 1244 1245/** 1246 * skb_queue_is_last - check if skb is the last entry in the queue 1247 * @list: queue head 1248 * @skb: buffer 1249 * 1250 * Returns true if @skb is the last buffer on the list. 1251 */ 1252static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1253 const struct sk_buff *skb) 1254{ 1255 return skb->next == (const struct sk_buff *) list; 1256} 1257 1258/** 1259 * skb_queue_is_first - check if skb is the first entry in the queue 1260 * @list: queue head 1261 * @skb: buffer 1262 * 1263 * Returns true if @skb is the first buffer on the list. 1264 */ 1265static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1266 const struct sk_buff *skb) 1267{ 1268 return skb->prev == (const struct sk_buff *) list; 1269} 1270 1271/** 1272 * skb_queue_next - return the next packet in the queue 1273 * @list: queue head 1274 * @skb: current buffer 1275 * 1276 * Return the next packet in @list after @skb. It is only valid to 1277 * call this if skb_queue_is_last() evaluates to false. 1278 */ 1279static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1280 const struct sk_buff *skb) 1281{ 1282 /* This BUG_ON may seem severe, but if we just return then we 1283 * are going to dereference garbage. 1284 */ 1285 BUG_ON(skb_queue_is_last(list, skb)); 1286 return skb->next; 1287} 1288 1289/** 1290 * skb_queue_prev - return the prev packet in the queue 1291 * @list: queue head 1292 * @skb: current buffer 1293 * 1294 * Return the prev packet in @list before @skb. It is only valid to 1295 * call this if skb_queue_is_first() evaluates to false. 1296 */ 1297static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1298 const struct sk_buff *skb) 1299{ 1300 /* This BUG_ON may seem severe, but if we just return then we 1301 * are going to dereference garbage. 1302 */ 1303 BUG_ON(skb_queue_is_first(list, skb)); 1304 return skb->prev; 1305} 1306 1307/** 1308 * skb_get - reference buffer 1309 * @skb: buffer to reference 1310 * 1311 * Makes another reference to a socket buffer and returns a pointer 1312 * to the buffer. 1313 */ 1314static inline struct sk_buff *skb_get(struct sk_buff *skb) 1315{ 1316 atomic_inc(&skb->users); 1317 return skb; 1318} 1319 1320/* 1321 * If users == 1, we are the only owner and are can avoid redundant 1322 * atomic change. 1323 */ 1324 1325/** 1326 * skb_cloned - is the buffer a clone 1327 * @skb: buffer to check 1328 * 1329 * Returns true if the buffer was generated with skb_clone() and is 1330 * one of multiple shared copies of the buffer. Cloned buffers are 1331 * shared data so must not be written to under normal circumstances. 1332 */ 1333static inline int skb_cloned(const struct sk_buff *skb) 1334{ 1335 return skb->cloned && 1336 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1337} 1338 1339static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1340{ 1341 might_sleep_if(gfpflags_allow_blocking(pri)); 1342 1343 if (skb_cloned(skb)) 1344 return pskb_expand_head(skb, 0, 0, pri); 1345 1346 return 0; 1347} 1348 1349/** 1350 * skb_header_cloned - is the header a clone 1351 * @skb: buffer to check 1352 * 1353 * Returns true if modifying the header part of the buffer requires 1354 * the data to be copied. 1355 */ 1356static inline int skb_header_cloned(const struct sk_buff *skb) 1357{ 1358 int dataref; 1359 1360 if (!skb->cloned) 1361 return 0; 1362 1363 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1364 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1365 return dataref != 1; 1366} 1367 1368static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1369{ 1370 might_sleep_if(gfpflags_allow_blocking(pri)); 1371 1372 if (skb_header_cloned(skb)) 1373 return pskb_expand_head(skb, 0, 0, pri); 1374 1375 return 0; 1376} 1377 1378/** 1379 * skb_header_release - release reference to header 1380 * @skb: buffer to operate on 1381 * 1382 * Drop a reference to the header part of the buffer. This is done 1383 * by acquiring a payload reference. You must not read from the header 1384 * part of skb->data after this. 1385 * Note : Check if you can use __skb_header_release() instead. 1386 */ 1387static inline void skb_header_release(struct sk_buff *skb) 1388{ 1389 BUG_ON(skb->nohdr); 1390 skb->nohdr = 1; 1391 atomic_add(1 << SKB_DATAREF_SHIFT, &skb_shinfo(skb)->dataref); 1392} 1393 1394/** 1395 * __skb_header_release - release reference to header 1396 * @skb: buffer to operate on 1397 * 1398 * Variant of skb_header_release() assuming skb is private to caller. 1399 * We can avoid one atomic operation. 1400 */ 1401static inline void __skb_header_release(struct sk_buff *skb) 1402{ 1403 skb->nohdr = 1; 1404 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1405} 1406 1407 1408/** 1409 * skb_shared - is the buffer shared 1410 * @skb: buffer to check 1411 * 1412 * Returns true if more than one person has a reference to this 1413 * buffer. 1414 */ 1415static inline int skb_shared(const struct sk_buff *skb) 1416{ 1417 return atomic_read(&skb->users) != 1; 1418} 1419 1420/** 1421 * skb_share_check - check if buffer is shared and if so clone it 1422 * @skb: buffer to check 1423 * @pri: priority for memory allocation 1424 * 1425 * If the buffer is shared the buffer is cloned and the old copy 1426 * drops a reference. A new clone with a single reference is returned. 1427 * If the buffer is not shared the original buffer is returned. When 1428 * being called from interrupt status or with spinlocks held pri must 1429 * be GFP_ATOMIC. 1430 * 1431 * NULL is returned on a memory allocation failure. 1432 */ 1433static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1434{ 1435 might_sleep_if(gfpflags_allow_blocking(pri)); 1436 if (skb_shared(skb)) { 1437 struct sk_buff *nskb = skb_clone(skb, pri); 1438 1439 if (likely(nskb)) 1440 consume_skb(skb); 1441 else 1442 kfree_skb(skb); 1443 skb = nskb; 1444 } 1445 return skb; 1446} 1447 1448/* 1449 * Copy shared buffers into a new sk_buff. We effectively do COW on 1450 * packets to handle cases where we have a local reader and forward 1451 * and a couple of other messy ones. The normal one is tcpdumping 1452 * a packet thats being forwarded. 1453 */ 1454 1455/** 1456 * skb_unshare - make a copy of a shared buffer 1457 * @skb: buffer to check 1458 * @pri: priority for memory allocation 1459 * 1460 * If the socket buffer is a clone then this function creates a new 1461 * copy of the data, drops a reference count on the old copy and returns 1462 * the new copy with the reference count at 1. If the buffer is not a clone 1463 * the original buffer is returned. When called with a spinlock held or 1464 * from interrupt state @pri must be %GFP_ATOMIC 1465 * 1466 * %NULL is returned on a memory allocation failure. 1467 */ 1468static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1469 gfp_t pri) 1470{ 1471 might_sleep_if(gfpflags_allow_blocking(pri)); 1472 if (skb_cloned(skb)) { 1473 struct sk_buff *nskb = skb_copy(skb, pri); 1474 1475 /* Free our shared copy */ 1476 if (likely(nskb)) 1477 consume_skb(skb); 1478 else 1479 kfree_skb(skb); 1480 skb = nskb; 1481 } 1482 return skb; 1483} 1484 1485/** 1486 * skb_peek - peek at the head of an &sk_buff_head 1487 * @list_: list to peek at 1488 * 1489 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1490 * be careful with this one. A peek leaves the buffer on the 1491 * list and someone else may run off with it. You must hold 1492 * the appropriate locks or have a private queue to do this. 1493 * 1494 * Returns %NULL for an empty list or a pointer to the head element. 1495 * The reference count is not incremented and the reference is therefore 1496 * volatile. Use with caution. 1497 */ 1498static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 1499{ 1500 struct sk_buff *skb = list_->next; 1501 1502 if (skb == (struct sk_buff *)list_) 1503 skb = NULL; 1504 return skb; 1505} 1506 1507/** 1508 * skb_peek_next - peek skb following the given one from a queue 1509 * @skb: skb to start from 1510 * @list_: list to peek at 1511 * 1512 * Returns %NULL when the end of the list is met or a pointer to the 1513 * next element. The reference count is not incremented and the 1514 * reference is therefore volatile. Use with caution. 1515 */ 1516static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 1517 const struct sk_buff_head *list_) 1518{ 1519 struct sk_buff *next = skb->next; 1520 1521 if (next == (struct sk_buff *)list_) 1522 next = NULL; 1523 return next; 1524} 1525 1526/** 1527 * skb_peek_tail - peek at the tail of an &sk_buff_head 1528 * @list_: list to peek at 1529 * 1530 * Peek an &sk_buff. Unlike most other operations you _MUST_ 1531 * be careful with this one. A peek leaves the buffer on the 1532 * list and someone else may run off with it. You must hold 1533 * the appropriate locks or have a private queue to do this. 1534 * 1535 * Returns %NULL for an empty list or a pointer to the tail element. 1536 * The reference count is not incremented and the reference is therefore 1537 * volatile. Use with caution. 1538 */ 1539static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 1540{ 1541 struct sk_buff *skb = list_->prev; 1542 1543 if (skb == (struct sk_buff *)list_) 1544 skb = NULL; 1545 return skb; 1546 1547} 1548 1549/** 1550 * skb_queue_len - get queue length 1551 * @list_: list to measure 1552 * 1553 * Return the length of an &sk_buff queue. 1554 */ 1555static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 1556{ 1557 return list_->qlen; 1558} 1559 1560/** 1561 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 1562 * @list: queue to initialize 1563 * 1564 * This initializes only the list and queue length aspects of 1565 * an sk_buff_head object. This allows to initialize the list 1566 * aspects of an sk_buff_head without reinitializing things like 1567 * the spinlock. It can also be used for on-stack sk_buff_head 1568 * objects where the spinlock is known to not be used. 1569 */ 1570static inline void __skb_queue_head_init(struct sk_buff_head *list) 1571{ 1572 list->prev = list->next = (struct sk_buff *)list; 1573 list->qlen = 0; 1574} 1575 1576/* 1577 * This function creates a split out lock class for each invocation; 1578 * this is needed for now since a whole lot of users of the skb-queue 1579 * infrastructure in drivers have different locking usage (in hardirq) 1580 * than the networking core (in softirq only). In the long run either the 1581 * network layer or drivers should need annotation to consolidate the 1582 * main types of usage into 3 classes. 1583 */ 1584static inline void skb_queue_head_init(struct sk_buff_head *list) 1585{ 1586 spin_lock_init(&list->lock); 1587 __skb_queue_head_init(list); 1588} 1589 1590static inline void skb_queue_head_init_class(struct sk_buff_head *list, 1591 struct lock_class_key *class) 1592{ 1593 skb_queue_head_init(list); 1594 lockdep_set_class(&list->lock, class); 1595} 1596 1597/* 1598 * Insert an sk_buff on a list. 1599 * 1600 * The "__skb_xxxx()" functions are the non-atomic ones that 1601 * can only be called with interrupts disabled. 1602 */ 1603void skb_insert(struct sk_buff *old, struct sk_buff *newsk, 1604 struct sk_buff_head *list); 1605static inline void __skb_insert(struct sk_buff *newsk, 1606 struct sk_buff *prev, struct sk_buff *next, 1607 struct sk_buff_head *list) 1608{ 1609 newsk->next = next; 1610 newsk->prev = prev; 1611 next->prev = prev->next = newsk; 1612 list->qlen++; 1613} 1614 1615static inline void __skb_queue_splice(const struct sk_buff_head *list, 1616 struct sk_buff *prev, 1617 struct sk_buff *next) 1618{ 1619 struct sk_buff *first = list->next; 1620 struct sk_buff *last = list->prev; 1621 1622 first->prev = prev; 1623 prev->next = first; 1624 1625 last->next = next; 1626 next->prev = last; 1627} 1628 1629/** 1630 * skb_queue_splice - join two skb lists, this is designed for stacks 1631 * @list: the new list to add 1632 * @head: the place to add it in the first list 1633 */ 1634static inline void skb_queue_splice(const struct sk_buff_head *list, 1635 struct sk_buff_head *head) 1636{ 1637 if (!skb_queue_empty(list)) { 1638 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1639 head->qlen += list->qlen; 1640 } 1641} 1642 1643/** 1644 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 1645 * @list: the new list to add 1646 * @head: the place to add it in the first list 1647 * 1648 * The list at @list is reinitialised 1649 */ 1650static inline void skb_queue_splice_init(struct sk_buff_head *list, 1651 struct sk_buff_head *head) 1652{ 1653 if (!skb_queue_empty(list)) { 1654 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 1655 head->qlen += list->qlen; 1656 __skb_queue_head_init(list); 1657 } 1658} 1659 1660/** 1661 * skb_queue_splice_tail - join two skb lists, each list being a queue 1662 * @list: the new list to add 1663 * @head: the place to add it in the first list 1664 */ 1665static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 1666 struct sk_buff_head *head) 1667{ 1668 if (!skb_queue_empty(list)) { 1669 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1670 head->qlen += list->qlen; 1671 } 1672} 1673 1674/** 1675 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 1676 * @list: the new list to add 1677 * @head: the place to add it in the first list 1678 * 1679 * Each of the lists is a queue. 1680 * The list at @list is reinitialised 1681 */ 1682static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 1683 struct sk_buff_head *head) 1684{ 1685 if (!skb_queue_empty(list)) { 1686 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 1687 head->qlen += list->qlen; 1688 __skb_queue_head_init(list); 1689 } 1690} 1691 1692/** 1693 * __skb_queue_after - queue a buffer at the list head 1694 * @list: list to use 1695 * @prev: place after this buffer 1696 * @newsk: buffer to queue 1697 * 1698 * Queue a buffer int the middle of a list. This function takes no locks 1699 * and you must therefore hold required locks before calling it. 1700 * 1701 * A buffer cannot be placed on two lists at the same time. 1702 */ 1703static inline void __skb_queue_after(struct sk_buff_head *list, 1704 struct sk_buff *prev, 1705 struct sk_buff *newsk) 1706{ 1707 __skb_insert(newsk, prev, prev->next, list); 1708} 1709 1710void skb_append(struct sk_buff *old, struct sk_buff *newsk, 1711 struct sk_buff_head *list); 1712 1713static inline void __skb_queue_before(struct sk_buff_head *list, 1714 struct sk_buff *next, 1715 struct sk_buff *newsk) 1716{ 1717 __skb_insert(newsk, next->prev, next, list); 1718} 1719 1720/** 1721 * __skb_queue_head - queue a buffer at the list head 1722 * @list: list to use 1723 * @newsk: buffer to queue 1724 * 1725 * Queue a buffer at the start of a list. This function takes no locks 1726 * and you must therefore hold required locks before calling it. 1727 * 1728 * A buffer cannot be placed on two lists at the same time. 1729 */ 1730void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 1731static inline void __skb_queue_head(struct sk_buff_head *list, 1732 struct sk_buff *newsk) 1733{ 1734 __skb_queue_after(list, (struct sk_buff *)list, newsk); 1735} 1736 1737/** 1738 * __skb_queue_tail - queue a buffer at the list tail 1739 * @list: list to use 1740 * @newsk: buffer to queue 1741 * 1742 * Queue a buffer at the end of a list. This function takes no locks 1743 * and you must therefore hold required locks before calling it. 1744 * 1745 * A buffer cannot be placed on two lists at the same time. 1746 */ 1747void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 1748static inline void __skb_queue_tail(struct sk_buff_head *list, 1749 struct sk_buff *newsk) 1750{ 1751 __skb_queue_before(list, (struct sk_buff *)list, newsk); 1752} 1753 1754/* 1755 * remove sk_buff from list. _Must_ be called atomically, and with 1756 * the list known.. 1757 */ 1758void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 1759static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 1760{ 1761 struct sk_buff *next, *prev; 1762 1763 list->qlen--; 1764 next = skb->next; 1765 prev = skb->prev; 1766 skb->next = skb->prev = NULL; 1767 next->prev = prev; 1768 prev->next = next; 1769} 1770 1771/** 1772 * __skb_dequeue - remove from the head of the queue 1773 * @list: list to dequeue from 1774 * 1775 * Remove the head of the list. This function does not take any locks 1776 * so must be used with appropriate locks held only. The head item is 1777 * returned or %NULL if the list is empty. 1778 */ 1779struct sk_buff *skb_dequeue(struct sk_buff_head *list); 1780static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 1781{ 1782 struct sk_buff *skb = skb_peek(list); 1783 if (skb) 1784 __skb_unlink(skb, list); 1785 return skb; 1786} 1787 1788/** 1789 * __skb_dequeue_tail - remove from the tail of the queue 1790 * @list: list to dequeue from 1791 * 1792 * Remove the tail of the list. This function does not take any locks 1793 * so must be used with appropriate locks held only. The tail item is 1794 * returned or %NULL if the list is empty. 1795 */ 1796struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 1797static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 1798{ 1799 struct sk_buff *skb = skb_peek_tail(list); 1800 if (skb) 1801 __skb_unlink(skb, list); 1802 return skb; 1803} 1804 1805 1806static inline bool skb_is_nonlinear(const struct sk_buff *skb) 1807{ 1808 return skb->data_len; 1809} 1810 1811static inline unsigned int skb_headlen(const struct sk_buff *skb) 1812{ 1813 return skb->len - skb->data_len; 1814} 1815 1816static inline unsigned int skb_pagelen(const struct sk_buff *skb) 1817{ 1818 unsigned int i, len = 0; 1819 1820 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 1821 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 1822 return len + skb_headlen(skb); 1823} 1824 1825/** 1826 * __skb_fill_page_desc - initialise a paged fragment in an skb 1827 * @skb: buffer containing fragment to be initialised 1828 * @i: paged fragment index to initialise 1829 * @page: the page to use for this fragment 1830 * @off: the offset to the data with @page 1831 * @size: the length of the data 1832 * 1833 * Initialises the @i'th fragment of @skb to point to &size bytes at 1834 * offset @off within @page. 1835 * 1836 * Does not take any additional reference on the fragment. 1837 */ 1838static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 1839 struct page *page, int off, int size) 1840{ 1841 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 1842 1843 /* 1844 * Propagate page pfmemalloc to the skb if we can. The problem is 1845 * that not all callers have unique ownership of the page but rely 1846 * on page_is_pfmemalloc doing the right thing(tm). 1847 */ 1848 frag->page.p = page; 1849 frag->page_offset = off; 1850 skb_frag_size_set(frag, size); 1851 1852 page = compound_head(page); 1853 if (page_is_pfmemalloc(page)) 1854 skb->pfmemalloc = true; 1855} 1856 1857/** 1858 * skb_fill_page_desc - initialise a paged fragment in an skb 1859 * @skb: buffer containing fragment to be initialised 1860 * @i: paged fragment index to initialise 1861 * @page: the page to use for this fragment 1862 * @off: the offset to the data with @page 1863 * @size: the length of the data 1864 * 1865 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 1866 * @skb to point to @size bytes at offset @off within @page. In 1867 * addition updates @skb such that @i is the last fragment. 1868 * 1869 * Does not take any additional reference on the fragment. 1870 */ 1871static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 1872 struct page *page, int off, int size) 1873{ 1874 __skb_fill_page_desc(skb, i, page, off, size); 1875 skb_shinfo(skb)->nr_frags = i + 1; 1876} 1877 1878void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 1879 int size, unsigned int truesize); 1880 1881void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 1882 unsigned int truesize); 1883 1884#define SKB_PAGE_ASSERT(skb) BUG_ON(skb_shinfo(skb)->nr_frags) 1885#define SKB_FRAG_ASSERT(skb) BUG_ON(skb_has_frag_list(skb)) 1886#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 1887 1888#ifdef NET_SKBUFF_DATA_USES_OFFSET 1889static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1890{ 1891 return skb->head + skb->tail; 1892} 1893 1894static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1895{ 1896 skb->tail = skb->data - skb->head; 1897} 1898 1899static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1900{ 1901 skb_reset_tail_pointer(skb); 1902 skb->tail += offset; 1903} 1904 1905#else /* NET_SKBUFF_DATA_USES_OFFSET */ 1906static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 1907{ 1908 return skb->tail; 1909} 1910 1911static inline void skb_reset_tail_pointer(struct sk_buff *skb) 1912{ 1913 skb->tail = skb->data; 1914} 1915 1916static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 1917{ 1918 skb->tail = skb->data + offset; 1919} 1920 1921#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 1922 1923/* 1924 * Add data to an sk_buff 1925 */ 1926unsigned char *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 1927unsigned char *skb_put(struct sk_buff *skb, unsigned int len); 1928static inline unsigned char *__skb_put(struct sk_buff *skb, unsigned int len) 1929{ 1930 unsigned char *tmp = skb_tail_pointer(skb); 1931 SKB_LINEAR_ASSERT(skb); 1932 skb->tail += len; 1933 skb->len += len; 1934 return tmp; 1935} 1936 1937unsigned char *skb_push(struct sk_buff *skb, unsigned int len); 1938static inline unsigned char *__skb_push(struct sk_buff *skb, unsigned int len) 1939{ 1940 skb->data -= len; 1941 skb->len += len; 1942 return skb->data; 1943} 1944 1945unsigned char *skb_pull(struct sk_buff *skb, unsigned int len); 1946static inline unsigned char *__skb_pull(struct sk_buff *skb, unsigned int len) 1947{ 1948 skb->len -= len; 1949 BUG_ON(skb->len < skb->data_len); 1950 return skb->data += len; 1951} 1952 1953static inline unsigned char *skb_pull_inline(struct sk_buff *skb, unsigned int len) 1954{ 1955 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 1956} 1957 1958unsigned char *__pskb_pull_tail(struct sk_buff *skb, int delta); 1959 1960static inline unsigned char *__pskb_pull(struct sk_buff *skb, unsigned int len) 1961{ 1962 if (len > skb_headlen(skb) && 1963 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 1964 return NULL; 1965 skb->len -= len; 1966 return skb->data += len; 1967} 1968 1969static inline unsigned char *pskb_pull(struct sk_buff *skb, unsigned int len) 1970{ 1971 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 1972} 1973 1974static inline int pskb_may_pull(struct sk_buff *skb, unsigned int len) 1975{ 1976 if (likely(len <= skb_headlen(skb))) 1977 return 1; 1978 if (unlikely(len > skb->len)) 1979 return 0; 1980 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 1981} 1982 1983void skb_condense(struct sk_buff *skb); 1984 1985/** 1986 * skb_headroom - bytes at buffer head 1987 * @skb: buffer to check 1988 * 1989 * Return the number of bytes of free space at the head of an &sk_buff. 1990 */ 1991static inline unsigned int skb_headroom(const struct sk_buff *skb) 1992{ 1993 return skb->data - skb->head; 1994} 1995 1996/** 1997 * skb_tailroom - bytes at buffer end 1998 * @skb: buffer to check 1999 * 2000 * Return the number of bytes of free space at the tail of an sk_buff 2001 */ 2002static inline int skb_tailroom(const struct sk_buff *skb) 2003{ 2004 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2005} 2006 2007/** 2008 * skb_availroom - bytes at buffer end 2009 * @skb: buffer to check 2010 * 2011 * Return the number of bytes of free space at the tail of an sk_buff 2012 * allocated by sk_stream_alloc() 2013 */ 2014static inline int skb_availroom(const struct sk_buff *skb) 2015{ 2016 if (skb_is_nonlinear(skb)) 2017 return 0; 2018 2019 return skb->end - skb->tail - skb->reserved_tailroom; 2020} 2021 2022/** 2023 * skb_reserve - adjust headroom 2024 * @skb: buffer to alter 2025 * @len: bytes to move 2026 * 2027 * Increase the headroom of an empty &sk_buff by reducing the tail 2028 * room. This is only allowed for an empty buffer. 2029 */ 2030static inline void skb_reserve(struct sk_buff *skb, int len) 2031{ 2032 skb->data += len; 2033 skb->tail += len; 2034} 2035 2036/** 2037 * skb_tailroom_reserve - adjust reserved_tailroom 2038 * @skb: buffer to alter 2039 * @mtu: maximum amount of headlen permitted 2040 * @needed_tailroom: minimum amount of reserved_tailroom 2041 * 2042 * Set reserved_tailroom so that headlen can be as large as possible but 2043 * not larger than mtu and tailroom cannot be smaller than 2044 * needed_tailroom. 2045 * The required headroom should already have been reserved before using 2046 * this function. 2047 */ 2048static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2049 unsigned int needed_tailroom) 2050{ 2051 SKB_LINEAR_ASSERT(skb); 2052 if (mtu < skb_tailroom(skb) - needed_tailroom) 2053 /* use at most mtu */ 2054 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2055 else 2056 /* use up to all available space */ 2057 skb->reserved_tailroom = needed_tailroom; 2058} 2059 2060#define ENCAP_TYPE_ETHER 0 2061#define ENCAP_TYPE_IPPROTO 1 2062 2063static inline void skb_set_inner_protocol(struct sk_buff *skb, 2064 __be16 protocol) 2065{ 2066 skb->inner_protocol = protocol; 2067 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2068} 2069 2070static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2071 __u8 ipproto) 2072{ 2073 skb->inner_ipproto = ipproto; 2074 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2075} 2076 2077static inline void skb_reset_inner_headers(struct sk_buff *skb) 2078{ 2079 skb->inner_mac_header = skb->mac_header; 2080 skb->inner_network_header = skb->network_header; 2081 skb->inner_transport_header = skb->transport_header; 2082} 2083 2084static inline void skb_reset_mac_len(struct sk_buff *skb) 2085{ 2086 skb->mac_len = skb->network_header - skb->mac_header; 2087} 2088 2089static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2090 *skb) 2091{ 2092 return skb->head + skb->inner_transport_header; 2093} 2094 2095static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2096{ 2097 return skb_inner_transport_header(skb) - skb->data; 2098} 2099 2100static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2101{ 2102 skb->inner_transport_header = skb->data - skb->head; 2103} 2104 2105static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2106 const int offset) 2107{ 2108 skb_reset_inner_transport_header(skb); 2109 skb->inner_transport_header += offset; 2110} 2111 2112static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2113{ 2114 return skb->head + skb->inner_network_header; 2115} 2116 2117static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2118{ 2119 skb->inner_network_header = skb->data - skb->head; 2120} 2121 2122static inline void skb_set_inner_network_header(struct sk_buff *skb, 2123 const int offset) 2124{ 2125 skb_reset_inner_network_header(skb); 2126 skb->inner_network_header += offset; 2127} 2128 2129static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2130{ 2131 return skb->head + skb->inner_mac_header; 2132} 2133 2134static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2135{ 2136 skb->inner_mac_header = skb->data - skb->head; 2137} 2138 2139static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2140 const int offset) 2141{ 2142 skb_reset_inner_mac_header(skb); 2143 skb->inner_mac_header += offset; 2144} 2145static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2146{ 2147 return skb->transport_header != (typeof(skb->transport_header))~0U; 2148} 2149 2150static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2151{ 2152 return skb->head + skb->transport_header; 2153} 2154 2155static inline void skb_reset_transport_header(struct sk_buff *skb) 2156{ 2157 skb->transport_header = skb->data - skb->head; 2158} 2159 2160static inline void skb_set_transport_header(struct sk_buff *skb, 2161 const int offset) 2162{ 2163 skb_reset_transport_header(skb); 2164 skb->transport_header += offset; 2165} 2166 2167static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2168{ 2169 return skb->head + skb->network_header; 2170} 2171 2172static inline void skb_reset_network_header(struct sk_buff *skb) 2173{ 2174 skb->network_header = skb->data - skb->head; 2175} 2176 2177static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2178{ 2179 skb_reset_network_header(skb); 2180 skb->network_header += offset; 2181} 2182 2183static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2184{ 2185 return skb->head + skb->mac_header; 2186} 2187 2188static inline int skb_mac_offset(const struct sk_buff *skb) 2189{ 2190 return skb_mac_header(skb) - skb->data; 2191} 2192 2193static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2194{ 2195 return skb->mac_header != (typeof(skb->mac_header))~0U; 2196} 2197 2198static inline void skb_reset_mac_header(struct sk_buff *skb) 2199{ 2200 skb->mac_header = skb->data - skb->head; 2201} 2202 2203static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2204{ 2205 skb_reset_mac_header(skb); 2206 skb->mac_header += offset; 2207} 2208 2209static inline void skb_pop_mac_header(struct sk_buff *skb) 2210{ 2211 skb->mac_header = skb->network_header; 2212} 2213 2214static inline void skb_probe_transport_header(struct sk_buff *skb, 2215 const int offset_hint) 2216{ 2217 struct flow_keys keys; 2218 2219 if (skb_transport_header_was_set(skb)) 2220 return; 2221 else if (skb_flow_dissect_flow_keys(skb, &keys, 0)) 2222 skb_set_transport_header(skb, keys.control.thoff); 2223 else 2224 skb_set_transport_header(skb, offset_hint); 2225} 2226 2227static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2228{ 2229 if (skb_mac_header_was_set(skb)) { 2230 const unsigned char *old_mac = skb_mac_header(skb); 2231 2232 skb_set_mac_header(skb, -skb->mac_len); 2233 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2234 } 2235} 2236 2237static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2238{ 2239 return skb->csum_start - skb_headroom(skb); 2240} 2241 2242static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2243{ 2244 return skb->head + skb->csum_start; 2245} 2246 2247static inline int skb_transport_offset(const struct sk_buff *skb) 2248{ 2249 return skb_transport_header(skb) - skb->data; 2250} 2251 2252static inline u32 skb_network_header_len(const struct sk_buff *skb) 2253{ 2254 return skb->transport_header - skb->network_header; 2255} 2256 2257static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2258{ 2259 return skb->inner_transport_header - skb->inner_network_header; 2260} 2261 2262static inline int skb_network_offset(const struct sk_buff *skb) 2263{ 2264 return skb_network_header(skb) - skb->data; 2265} 2266 2267static inline int skb_inner_network_offset(const struct sk_buff *skb) 2268{ 2269 return skb_inner_network_header(skb) - skb->data; 2270} 2271 2272static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2273{ 2274 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2275} 2276 2277/* 2278 * CPUs often take a performance hit when accessing unaligned memory 2279 * locations. The actual performance hit varies, it can be small if the 2280 * hardware handles it or large if we have to take an exception and fix it 2281 * in software. 2282 * 2283 * Since an ethernet header is 14 bytes network drivers often end up with 2284 * the IP header at an unaligned offset. The IP header can be aligned by 2285 * shifting the start of the packet by 2 bytes. Drivers should do this 2286 * with: 2287 * 2288 * skb_reserve(skb, NET_IP_ALIGN); 2289 * 2290 * The downside to this alignment of the IP header is that the DMA is now 2291 * unaligned. On some architectures the cost of an unaligned DMA is high 2292 * and this cost outweighs the gains made by aligning the IP header. 2293 * 2294 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2295 * to be overridden. 2296 */ 2297#ifndef NET_IP_ALIGN 2298#define NET_IP_ALIGN 2 2299#endif 2300 2301/* 2302 * The networking layer reserves some headroom in skb data (via 2303 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2304 * the header has to grow. In the default case, if the header has to grow 2305 * 32 bytes or less we avoid the reallocation. 2306 * 2307 * Unfortunately this headroom changes the DMA alignment of the resulting 2308 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2309 * on some architectures. An architecture can override this value, 2310 * perhaps setting it to a cacheline in size (since that will maintain 2311 * cacheline alignment of the DMA). It must be a power of 2. 2312 * 2313 * Various parts of the networking layer expect at least 32 bytes of 2314 * headroom, you should not reduce this. 2315 * 2316 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2317 * to reduce average number of cache lines per packet. 2318 * get_rps_cpus() for example only access one 64 bytes aligned block : 2319 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2320 */ 2321#ifndef NET_SKB_PAD 2322#define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2323#endif 2324 2325int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2326 2327static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2328{ 2329 if (unlikely(skb_is_nonlinear(skb))) { 2330 WARN_ON(1); 2331 return; 2332 } 2333 skb->len = len; 2334 skb_set_tail_pointer(skb, len); 2335} 2336 2337static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2338{ 2339 __skb_set_length(skb, len); 2340} 2341 2342void skb_trim(struct sk_buff *skb, unsigned int len); 2343 2344static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2345{ 2346 if (skb->data_len) 2347 return ___pskb_trim(skb, len); 2348 __skb_trim(skb, len); 2349 return 0; 2350} 2351 2352static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2353{ 2354 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2355} 2356 2357/** 2358 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2359 * @skb: buffer to alter 2360 * @len: new length 2361 * 2362 * This is identical to pskb_trim except that the caller knows that 2363 * the skb is not cloned so we should never get an error due to out- 2364 * of-memory. 2365 */ 2366static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2367{ 2368 int err = pskb_trim(skb, len); 2369 BUG_ON(err); 2370} 2371 2372static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2373{ 2374 unsigned int diff = len - skb->len; 2375 2376 if (skb_tailroom(skb) < diff) { 2377 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2378 GFP_ATOMIC); 2379 if (ret) 2380 return ret; 2381 } 2382 __skb_set_length(skb, len); 2383 return 0; 2384} 2385 2386/** 2387 * skb_orphan - orphan a buffer 2388 * @skb: buffer to orphan 2389 * 2390 * If a buffer currently has an owner then we call the owner's 2391 * destructor function and make the @skb unowned. The buffer continues 2392 * to exist but is no longer charged to its former owner. 2393 */ 2394static inline void skb_orphan(struct sk_buff *skb) 2395{ 2396 if (skb->destructor) { 2397 skb->destructor(skb); 2398 skb->destructor = NULL; 2399 skb->sk = NULL; 2400 } else { 2401 BUG_ON(skb->sk); 2402 } 2403} 2404 2405/** 2406 * skb_orphan_frags - orphan the frags contained in a buffer 2407 * @skb: buffer to orphan frags from 2408 * @gfp_mask: allocation mask for replacement pages 2409 * 2410 * For each frag in the SKB which needs a destructor (i.e. has an 2411 * owner) create a copy of that frag and release the original 2412 * page by calling the destructor. 2413 */ 2414static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 2415{ 2416 if (likely(!(skb_shinfo(skb)->tx_flags & SKBTX_DEV_ZEROCOPY))) 2417 return 0; 2418 return skb_copy_ubufs(skb, gfp_mask); 2419} 2420 2421/** 2422 * __skb_queue_purge - empty a list 2423 * @list: list to empty 2424 * 2425 * Delete all buffers on an &sk_buff list. Each buffer is removed from 2426 * the list and one reference dropped. This function does not take the 2427 * list lock and the caller must hold the relevant locks to use it. 2428 */ 2429void skb_queue_purge(struct sk_buff_head *list); 2430static inline void __skb_queue_purge(struct sk_buff_head *list) 2431{ 2432 struct sk_buff *skb; 2433 while ((skb = __skb_dequeue(list)) != NULL) 2434 kfree_skb(skb); 2435} 2436 2437void skb_rbtree_purge(struct rb_root *root); 2438 2439void *netdev_alloc_frag(unsigned int fragsz); 2440 2441struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 2442 gfp_t gfp_mask); 2443 2444/** 2445 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 2446 * @dev: network device to receive on 2447 * @length: length to allocate 2448 * 2449 * Allocate a new &sk_buff and assign it a usage count of one. The 2450 * buffer has unspecified headroom built in. Users should allocate 2451 * the headroom they think they need without accounting for the 2452 * built in space. The built in space is used for optimisations. 2453 * 2454 * %NULL is returned if there is no free memory. Although this function 2455 * allocates memory it can be called from an interrupt. 2456 */ 2457static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 2458 unsigned int length) 2459{ 2460 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 2461} 2462 2463/* legacy helper around __netdev_alloc_skb() */ 2464static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 2465 gfp_t gfp_mask) 2466{ 2467 return __netdev_alloc_skb(NULL, length, gfp_mask); 2468} 2469 2470/* legacy helper around netdev_alloc_skb() */ 2471static inline struct sk_buff *dev_alloc_skb(unsigned int length) 2472{ 2473 return netdev_alloc_skb(NULL, length); 2474} 2475 2476 2477static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 2478 unsigned int length, gfp_t gfp) 2479{ 2480 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 2481 2482 if (NET_IP_ALIGN && skb) 2483 skb_reserve(skb, NET_IP_ALIGN); 2484 return skb; 2485} 2486 2487static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 2488 unsigned int length) 2489{ 2490 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 2491} 2492 2493static inline void skb_free_frag(void *addr) 2494{ 2495 page_frag_free(addr); 2496} 2497 2498void *napi_alloc_frag(unsigned int fragsz); 2499struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 2500 unsigned int length, gfp_t gfp_mask); 2501static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 2502 unsigned int length) 2503{ 2504 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 2505} 2506void napi_consume_skb(struct sk_buff *skb, int budget); 2507 2508void __kfree_skb_flush(void); 2509void __kfree_skb_defer(struct sk_buff *skb); 2510 2511/** 2512 * __dev_alloc_pages - allocate page for network Rx 2513 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2514 * @order: size of the allocation 2515 * 2516 * Allocate a new page. 2517 * 2518 * %NULL is returned if there is no free memory. 2519*/ 2520static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 2521 unsigned int order) 2522{ 2523 /* This piece of code contains several assumptions. 2524 * 1. This is for device Rx, therefor a cold page is preferred. 2525 * 2. The expectation is the user wants a compound page. 2526 * 3. If requesting a order 0 page it will not be compound 2527 * due to the check to see if order has a value in prep_new_page 2528 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 2529 * code in gfp_to_alloc_flags that should be enforcing this. 2530 */ 2531 gfp_mask |= __GFP_COLD | __GFP_COMP | __GFP_MEMALLOC; 2532 2533 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 2534} 2535 2536static inline struct page *dev_alloc_pages(unsigned int order) 2537{ 2538 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 2539} 2540 2541/** 2542 * __dev_alloc_page - allocate a page for network Rx 2543 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 2544 * 2545 * Allocate a new page. 2546 * 2547 * %NULL is returned if there is no free memory. 2548 */ 2549static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 2550{ 2551 return __dev_alloc_pages(gfp_mask, 0); 2552} 2553 2554static inline struct page *dev_alloc_page(void) 2555{ 2556 return dev_alloc_pages(0); 2557} 2558 2559/** 2560 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 2561 * @page: The page that was allocated from skb_alloc_page 2562 * @skb: The skb that may need pfmemalloc set 2563 */ 2564static inline void skb_propagate_pfmemalloc(struct page *page, 2565 struct sk_buff *skb) 2566{ 2567 if (page_is_pfmemalloc(page)) 2568 skb->pfmemalloc = true; 2569} 2570 2571/** 2572 * skb_frag_page - retrieve the page referred to by a paged fragment 2573 * @frag: the paged fragment 2574 * 2575 * Returns the &struct page associated with @frag. 2576 */ 2577static inline struct page *skb_frag_page(const skb_frag_t *frag) 2578{ 2579 return frag->page.p; 2580} 2581 2582/** 2583 * __skb_frag_ref - take an addition reference on a paged fragment. 2584 * @frag: the paged fragment 2585 * 2586 * Takes an additional reference on the paged fragment @frag. 2587 */ 2588static inline void __skb_frag_ref(skb_frag_t *frag) 2589{ 2590 get_page(skb_frag_page(frag)); 2591} 2592 2593/** 2594 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 2595 * @skb: the buffer 2596 * @f: the fragment offset. 2597 * 2598 * Takes an additional reference on the @f'th paged fragment of @skb. 2599 */ 2600static inline void skb_frag_ref(struct sk_buff *skb, int f) 2601{ 2602 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 2603} 2604 2605/** 2606 * __skb_frag_unref - release a reference on a paged fragment. 2607 * @frag: the paged fragment 2608 * 2609 * Releases a reference on the paged fragment @frag. 2610 */ 2611static inline void __skb_frag_unref(skb_frag_t *frag) 2612{ 2613 put_page(skb_frag_page(frag)); 2614} 2615 2616/** 2617 * skb_frag_unref - release a reference on a paged fragment of an skb. 2618 * @skb: the buffer 2619 * @f: the fragment offset 2620 * 2621 * Releases a reference on the @f'th paged fragment of @skb. 2622 */ 2623static inline void skb_frag_unref(struct sk_buff *skb, int f) 2624{ 2625 __skb_frag_unref(&skb_shinfo(skb)->frags[f]); 2626} 2627 2628/** 2629 * skb_frag_address - gets the address of the data contained in a paged fragment 2630 * @frag: the paged fragment buffer 2631 * 2632 * Returns the address of the data within @frag. The page must already 2633 * be mapped. 2634 */ 2635static inline void *skb_frag_address(const skb_frag_t *frag) 2636{ 2637 return page_address(skb_frag_page(frag)) + frag->page_offset; 2638} 2639 2640/** 2641 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 2642 * @frag: the paged fragment buffer 2643 * 2644 * Returns the address of the data within @frag. Checks that the page 2645 * is mapped and returns %NULL otherwise. 2646 */ 2647static inline void *skb_frag_address_safe(const skb_frag_t *frag) 2648{ 2649 void *ptr = page_address(skb_frag_page(frag)); 2650 if (unlikely(!ptr)) 2651 return NULL; 2652 2653 return ptr + frag->page_offset; 2654} 2655 2656/** 2657 * __skb_frag_set_page - sets the page contained in a paged fragment 2658 * @frag: the paged fragment 2659 * @page: the page to set 2660 * 2661 * Sets the fragment @frag to contain @page. 2662 */ 2663static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 2664{ 2665 frag->page.p = page; 2666} 2667 2668/** 2669 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 2670 * @skb: the buffer 2671 * @f: the fragment offset 2672 * @page: the page to set 2673 * 2674 * Sets the @f'th fragment of @skb to contain @page. 2675 */ 2676static inline void skb_frag_set_page(struct sk_buff *skb, int f, 2677 struct page *page) 2678{ 2679 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 2680} 2681 2682bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 2683 2684/** 2685 * skb_frag_dma_map - maps a paged fragment via the DMA API 2686 * @dev: the device to map the fragment to 2687 * @frag: the paged fragment to map 2688 * @offset: the offset within the fragment (starting at the 2689 * fragment's own offset) 2690 * @size: the number of bytes to map 2691 * @dir: the direction of the mapping (%PCI_DMA_*) 2692 * 2693 * Maps the page associated with @frag to @device. 2694 */ 2695static inline dma_addr_t skb_frag_dma_map(struct device *dev, 2696 const skb_frag_t *frag, 2697 size_t offset, size_t size, 2698 enum dma_data_direction dir) 2699{ 2700 return dma_map_page(dev, skb_frag_page(frag), 2701 frag->page_offset + offset, size, dir); 2702} 2703 2704static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 2705 gfp_t gfp_mask) 2706{ 2707 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 2708} 2709 2710 2711static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 2712 gfp_t gfp_mask) 2713{ 2714 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 2715} 2716 2717 2718/** 2719 * skb_clone_writable - is the header of a clone writable 2720 * @skb: buffer to check 2721 * @len: length up to which to write 2722 * 2723 * Returns true if modifying the header part of the cloned buffer 2724 * does not requires the data to be copied. 2725 */ 2726static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 2727{ 2728 return !skb_header_cloned(skb) && 2729 skb_headroom(skb) + len <= skb->hdr_len; 2730} 2731 2732static inline int skb_try_make_writable(struct sk_buff *skb, 2733 unsigned int write_len) 2734{ 2735 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 2736 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 2737} 2738 2739static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 2740 int cloned) 2741{ 2742 int delta = 0; 2743 2744 if (headroom > skb_headroom(skb)) 2745 delta = headroom - skb_headroom(skb); 2746 2747 if (delta || cloned) 2748 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 2749 GFP_ATOMIC); 2750 return 0; 2751} 2752 2753/** 2754 * skb_cow - copy header of skb when it is required 2755 * @skb: buffer to cow 2756 * @headroom: needed headroom 2757 * 2758 * If the skb passed lacks sufficient headroom or its data part 2759 * is shared, data is reallocated. If reallocation fails, an error 2760 * is returned and original skb is not changed. 2761 * 2762 * The result is skb with writable area skb->head...skb->tail 2763 * and at least @headroom of space at head. 2764 */ 2765static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 2766{ 2767 return __skb_cow(skb, headroom, skb_cloned(skb)); 2768} 2769 2770/** 2771 * skb_cow_head - skb_cow but only making the head writable 2772 * @skb: buffer to cow 2773 * @headroom: needed headroom 2774 * 2775 * This function is identical to skb_cow except that we replace the 2776 * skb_cloned check by skb_header_cloned. It should be used when 2777 * you only need to push on some header and do not need to modify 2778 * the data. 2779 */ 2780static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 2781{ 2782 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 2783} 2784 2785/** 2786 * skb_padto - pad an skbuff up to a minimal size 2787 * @skb: buffer to pad 2788 * @len: minimal length 2789 * 2790 * Pads up a buffer to ensure the trailing bytes exist and are 2791 * blanked. If the buffer already contains sufficient data it 2792 * is untouched. Otherwise it is extended. Returns zero on 2793 * success. The skb is freed on error. 2794 */ 2795static inline int skb_padto(struct sk_buff *skb, unsigned int len) 2796{ 2797 unsigned int size = skb->len; 2798 if (likely(size >= len)) 2799 return 0; 2800 return skb_pad(skb, len - size); 2801} 2802 2803/** 2804 * skb_put_padto - increase size and pad an skbuff up to a minimal size 2805 * @skb: buffer to pad 2806 * @len: minimal length 2807 * 2808 * Pads up a buffer to ensure the trailing bytes exist and are 2809 * blanked. If the buffer already contains sufficient data it 2810 * is untouched. Otherwise it is extended. Returns zero on 2811 * success. The skb is freed on error. 2812 */ 2813static inline int skb_put_padto(struct sk_buff *skb, unsigned int len) 2814{ 2815 unsigned int size = skb->len; 2816 2817 if (unlikely(size < len)) { 2818 len -= size; 2819 if (skb_pad(skb, len)) 2820 return -ENOMEM; 2821 __skb_put(skb, len); 2822 } 2823 return 0; 2824} 2825 2826static inline int skb_add_data(struct sk_buff *skb, 2827 struct iov_iter *from, int copy) 2828{ 2829 const int off = skb->len; 2830 2831 if (skb->ip_summed == CHECKSUM_NONE) { 2832 __wsum csum = 0; 2833 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 2834 &csum, from)) { 2835 skb->csum = csum_block_add(skb->csum, csum, off); 2836 return 0; 2837 } 2838 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 2839 return 0; 2840 2841 __skb_trim(skb, off); 2842 return -EFAULT; 2843} 2844 2845static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 2846 const struct page *page, int off) 2847{ 2848 if (i) { 2849 const struct skb_frag_struct *frag = &skb_shinfo(skb)->frags[i - 1]; 2850 2851 return page == skb_frag_page(frag) && 2852 off == frag->page_offset + skb_frag_size(frag); 2853 } 2854 return false; 2855} 2856 2857static inline int __skb_linearize(struct sk_buff *skb) 2858{ 2859 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 2860} 2861 2862/** 2863 * skb_linearize - convert paged skb to linear one 2864 * @skb: buffer to linarize 2865 * 2866 * If there is no free memory -ENOMEM is returned, otherwise zero 2867 * is returned and the old skb data released. 2868 */ 2869static inline int skb_linearize(struct sk_buff *skb) 2870{ 2871 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 2872} 2873 2874/** 2875 * skb_has_shared_frag - can any frag be overwritten 2876 * @skb: buffer to test 2877 * 2878 * Return true if the skb has at least one frag that might be modified 2879 * by an external entity (as in vmsplice()/sendfile()) 2880 */ 2881static inline bool skb_has_shared_frag(const struct sk_buff *skb) 2882{ 2883 return skb_is_nonlinear(skb) && 2884 skb_shinfo(skb)->tx_flags & SKBTX_SHARED_FRAG; 2885} 2886 2887/** 2888 * skb_linearize_cow - make sure skb is linear and writable 2889 * @skb: buffer to process 2890 * 2891 * If there is no free memory -ENOMEM is returned, otherwise zero 2892 * is returned and the old skb data released. 2893 */ 2894static inline int skb_linearize_cow(struct sk_buff *skb) 2895{ 2896 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 2897 __skb_linearize(skb) : 0; 2898} 2899 2900static __always_inline void 2901__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 2902 unsigned int off) 2903{ 2904 if (skb->ip_summed == CHECKSUM_COMPLETE) 2905 skb->csum = csum_block_sub(skb->csum, 2906 csum_partial(start, len, 0), off); 2907 else if (skb->ip_summed == CHECKSUM_PARTIAL && 2908 skb_checksum_start_offset(skb) < 0) 2909 skb->ip_summed = CHECKSUM_NONE; 2910} 2911 2912/** 2913 * skb_postpull_rcsum - update checksum for received skb after pull 2914 * @skb: buffer to update 2915 * @start: start of data before pull 2916 * @len: length of data pulled 2917 * 2918 * After doing a pull on a received packet, you need to call this to 2919 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 2920 * CHECKSUM_NONE so that it can be recomputed from scratch. 2921 */ 2922static inline void skb_postpull_rcsum(struct sk_buff *skb, 2923 const void *start, unsigned int len) 2924{ 2925 __skb_postpull_rcsum(skb, start, len, 0); 2926} 2927 2928static __always_inline void 2929__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 2930 unsigned int off) 2931{ 2932 if (skb->ip_summed == CHECKSUM_COMPLETE) 2933 skb->csum = csum_block_add(skb->csum, 2934 csum_partial(start, len, 0), off); 2935} 2936 2937/** 2938 * skb_postpush_rcsum - update checksum for received skb after push 2939 * @skb: buffer to update 2940 * @start: start of data after push 2941 * @len: length of data pushed 2942 * 2943 * After doing a push on a received packet, you need to call this to 2944 * update the CHECKSUM_COMPLETE checksum. 2945 */ 2946static inline void skb_postpush_rcsum(struct sk_buff *skb, 2947 const void *start, unsigned int len) 2948{ 2949 __skb_postpush_rcsum(skb, start, len, 0); 2950} 2951 2952unsigned char *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 2953 2954/** 2955 * skb_push_rcsum - push skb and update receive checksum 2956 * @skb: buffer to update 2957 * @len: length of data pulled 2958 * 2959 * This function performs an skb_push on the packet and updates 2960 * the CHECKSUM_COMPLETE checksum. It should be used on 2961 * receive path processing instead of skb_push unless you know 2962 * that the checksum difference is zero (e.g., a valid IP header) 2963 * or you are setting ip_summed to CHECKSUM_NONE. 2964 */ 2965static inline unsigned char *skb_push_rcsum(struct sk_buff *skb, 2966 unsigned int len) 2967{ 2968 skb_push(skb, len); 2969 skb_postpush_rcsum(skb, skb->data, len); 2970 return skb->data; 2971} 2972 2973/** 2974 * pskb_trim_rcsum - trim received skb and update checksum 2975 * @skb: buffer to trim 2976 * @len: new length 2977 * 2978 * This is exactly the same as pskb_trim except that it ensures the 2979 * checksum of received packets are still valid after the operation. 2980 */ 2981 2982static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2983{ 2984 if (likely(len >= skb->len)) 2985 return 0; 2986 if (skb->ip_summed == CHECKSUM_COMPLETE) 2987 skb->ip_summed = CHECKSUM_NONE; 2988 return __pskb_trim(skb, len); 2989} 2990 2991static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 2992{ 2993 if (skb->ip_summed == CHECKSUM_COMPLETE) 2994 skb->ip_summed = CHECKSUM_NONE; 2995 __skb_trim(skb, len); 2996 return 0; 2997} 2998 2999static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3000{ 3001 if (skb->ip_summed == CHECKSUM_COMPLETE) 3002 skb->ip_summed = CHECKSUM_NONE; 3003 return __skb_grow(skb, len); 3004} 3005 3006#define skb_queue_walk(queue, skb) \ 3007 for (skb = (queue)->next; \ 3008 skb != (struct sk_buff *)(queue); \ 3009 skb = skb->next) 3010 3011#define skb_queue_walk_safe(queue, skb, tmp) \ 3012 for (skb = (queue)->next, tmp = skb->next; \ 3013 skb != (struct sk_buff *)(queue); \ 3014 skb = tmp, tmp = skb->next) 3015 3016#define skb_queue_walk_from(queue, skb) \ 3017 for (; skb != (struct sk_buff *)(queue); \ 3018 skb = skb->next) 3019 3020#define skb_queue_walk_from_safe(queue, skb, tmp) \ 3021 for (tmp = skb->next; \ 3022 skb != (struct sk_buff *)(queue); \ 3023 skb = tmp, tmp = skb->next) 3024 3025#define skb_queue_reverse_walk(queue, skb) \ 3026 for (skb = (queue)->prev; \ 3027 skb != (struct sk_buff *)(queue); \ 3028 skb = skb->prev) 3029 3030#define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3031 for (skb = (queue)->prev, tmp = skb->prev; \ 3032 skb != (struct sk_buff *)(queue); \ 3033 skb = tmp, tmp = skb->prev) 3034 3035#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3036 for (tmp = skb->prev; \ 3037 skb != (struct sk_buff *)(queue); \ 3038 skb = tmp, tmp = skb->prev) 3039 3040static inline bool skb_has_frag_list(const struct sk_buff *skb) 3041{ 3042 return skb_shinfo(skb)->frag_list != NULL; 3043} 3044 3045static inline void skb_frag_list_init(struct sk_buff *skb) 3046{ 3047 skb_shinfo(skb)->frag_list = NULL; 3048} 3049 3050#define skb_walk_frags(skb, iter) \ 3051 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3052 3053 3054int __skb_wait_for_more_packets(struct sock *sk, int *err, long *timeo_p, 3055 const struct sk_buff *skb); 3056struct sk_buff *__skb_try_recv_datagram(struct sock *sk, unsigned flags, 3057 void (*destructor)(struct sock *sk, 3058 struct sk_buff *skb), 3059 int *peeked, int *off, int *err, 3060 struct sk_buff **last); 3061struct sk_buff *__skb_recv_datagram(struct sock *sk, unsigned flags, 3062 void (*destructor)(struct sock *sk, 3063 struct sk_buff *skb), 3064 int *peeked, int *off, int *err); 3065struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 3066 int *err); 3067unsigned int datagram_poll(struct file *file, struct socket *sock, 3068 struct poll_table_struct *wait); 3069int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3070 struct iov_iter *to, int size); 3071static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3072 struct msghdr *msg, int size) 3073{ 3074 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3075} 3076int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3077 struct msghdr *msg); 3078int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3079 struct iov_iter *from, int len); 3080int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3081void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3082void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3083static inline void skb_free_datagram_locked(struct sock *sk, 3084 struct sk_buff *skb) 3085{ 3086 __skb_free_datagram_locked(sk, skb, 0); 3087} 3088int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3089int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3090int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3091__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3092 int len, __wsum csum); 3093int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3094 struct pipe_inode_info *pipe, unsigned int len, 3095 unsigned int flags); 3096void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3097unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3098int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3099 int len, int hlen); 3100void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3101int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3102void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3103unsigned int skb_gso_transport_seglen(const struct sk_buff *skb); 3104bool skb_gso_validate_mtu(const struct sk_buff *skb, unsigned int mtu); 3105struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3106struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3107int skb_ensure_writable(struct sk_buff *skb, int write_len); 3108int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3109int skb_vlan_pop(struct sk_buff *skb); 3110int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3111struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3112 gfp_t gfp); 3113 3114static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3115{ 3116 return copy_from_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3117} 3118 3119static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3120{ 3121 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3122} 3123 3124struct skb_checksum_ops { 3125 __wsum (*update)(const void *mem, int len, __wsum wsum); 3126 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3127}; 3128 3129__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3130 __wsum csum, const struct skb_checksum_ops *ops); 3131__wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3132 __wsum csum); 3133 3134static inline void * __must_check 3135__skb_header_pointer(const struct sk_buff *skb, int offset, 3136 int len, void *data, int hlen, void *buffer) 3137{ 3138 if (hlen - offset >= len) 3139 return data + offset; 3140 3141 if (!skb || 3142 skb_copy_bits(skb, offset, buffer, len) < 0) 3143 return NULL; 3144 3145 return buffer; 3146} 3147 3148static inline void * __must_check 3149skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3150{ 3151 return __skb_header_pointer(skb, offset, len, skb->data, 3152 skb_headlen(skb), buffer); 3153} 3154 3155/** 3156 * skb_needs_linearize - check if we need to linearize a given skb 3157 * depending on the given device features. 3158 * @skb: socket buffer to check 3159 * @features: net device features 3160 * 3161 * Returns true if either: 3162 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3163 * 2. skb is fragmented and the device does not support SG. 3164 */ 3165static inline bool skb_needs_linearize(struct sk_buff *skb, 3166 netdev_features_t features) 3167{ 3168 return skb_is_nonlinear(skb) && 3169 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3170 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3171} 3172 3173static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3174 void *to, 3175 const unsigned int len) 3176{ 3177 memcpy(to, skb->data, len); 3178} 3179 3180static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3181 const int offset, void *to, 3182 const unsigned int len) 3183{ 3184 memcpy(to, skb->data + offset, len); 3185} 3186 3187static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3188 const void *from, 3189 const unsigned int len) 3190{ 3191 memcpy(skb->data, from, len); 3192} 3193 3194static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3195 const int offset, 3196 const void *from, 3197 const unsigned int len) 3198{ 3199 memcpy(skb->data + offset, from, len); 3200} 3201 3202void skb_init(void); 3203 3204static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3205{ 3206 return skb->tstamp; 3207} 3208 3209/** 3210 * skb_get_timestamp - get timestamp from a skb 3211 * @skb: skb to get stamp from 3212 * @stamp: pointer to struct timeval to store stamp in 3213 * 3214 * Timestamps are stored in the skb as offsets to a base timestamp. 3215 * This function converts the offset back to a struct timeval and stores 3216 * it in stamp. 3217 */ 3218static inline void skb_get_timestamp(const struct sk_buff *skb, 3219 struct timeval *stamp) 3220{ 3221 *stamp = ktime_to_timeval(skb->tstamp); 3222} 3223 3224static inline void skb_get_timestampns(const struct sk_buff *skb, 3225 struct timespec *stamp) 3226{ 3227 *stamp = ktime_to_timespec(skb->tstamp); 3228} 3229 3230static inline void __net_timestamp(struct sk_buff *skb) 3231{ 3232 skb->tstamp = ktime_get_real(); 3233} 3234 3235static inline ktime_t net_timedelta(ktime_t t) 3236{ 3237 return ktime_sub(ktime_get_real(), t); 3238} 3239 3240static inline ktime_t net_invalid_timestamp(void) 3241{ 3242 return 0; 3243} 3244 3245struct sk_buff *skb_clone_sk(struct sk_buff *skb); 3246 3247#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 3248 3249void skb_clone_tx_timestamp(struct sk_buff *skb); 3250bool skb_defer_rx_timestamp(struct sk_buff *skb); 3251 3252#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 3253 3254static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 3255{ 3256} 3257 3258static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 3259{ 3260 return false; 3261} 3262 3263#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 3264 3265/** 3266 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 3267 * 3268 * PHY drivers may accept clones of transmitted packets for 3269 * timestamping via their phy_driver.txtstamp method. These drivers 3270 * must call this function to return the skb back to the stack with a 3271 * timestamp. 3272 * 3273 * @skb: clone of the the original outgoing packet 3274 * @hwtstamps: hardware time stamps 3275 * 3276 */ 3277void skb_complete_tx_timestamp(struct sk_buff *skb, 3278 struct skb_shared_hwtstamps *hwtstamps); 3279 3280void __skb_tstamp_tx(struct sk_buff *orig_skb, 3281 struct skb_shared_hwtstamps *hwtstamps, 3282 struct sock *sk, int tstype); 3283 3284/** 3285 * skb_tstamp_tx - queue clone of skb with send time stamps 3286 * @orig_skb: the original outgoing packet 3287 * @hwtstamps: hardware time stamps, may be NULL if not available 3288 * 3289 * If the skb has a socket associated, then this function clones the 3290 * skb (thus sharing the actual data and optional structures), stores 3291 * the optional hardware time stamping information (if non NULL) or 3292 * generates a software time stamp (otherwise), then queues the clone 3293 * to the error queue of the socket. Errors are silently ignored. 3294 */ 3295void skb_tstamp_tx(struct sk_buff *orig_skb, 3296 struct skb_shared_hwtstamps *hwtstamps); 3297 3298static inline void sw_tx_timestamp(struct sk_buff *skb) 3299{ 3300 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP && 3301 !(skb_shinfo(skb)->tx_flags & SKBTX_IN_PROGRESS)) 3302 skb_tstamp_tx(skb, NULL); 3303} 3304 3305/** 3306 * skb_tx_timestamp() - Driver hook for transmit timestamping 3307 * 3308 * Ethernet MAC Drivers should call this function in their hard_xmit() 3309 * function immediately before giving the sk_buff to the MAC hardware. 3310 * 3311 * Specifically, one should make absolutely sure that this function is 3312 * called before TX completion of this packet can trigger. Otherwise 3313 * the packet could potentially already be freed. 3314 * 3315 * @skb: A socket buffer. 3316 */ 3317static inline void skb_tx_timestamp(struct sk_buff *skb) 3318{ 3319 skb_clone_tx_timestamp(skb); 3320 sw_tx_timestamp(skb); 3321} 3322 3323/** 3324 * skb_complete_wifi_ack - deliver skb with wifi status 3325 * 3326 * @skb: the original outgoing packet 3327 * @acked: ack status 3328 * 3329 */ 3330void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 3331 3332__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 3333__sum16 __skb_checksum_complete(struct sk_buff *skb); 3334 3335static inline int skb_csum_unnecessary(const struct sk_buff *skb) 3336{ 3337 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 3338 skb->csum_valid || 3339 (skb->ip_summed == CHECKSUM_PARTIAL && 3340 skb_checksum_start_offset(skb) >= 0)); 3341} 3342 3343/** 3344 * skb_checksum_complete - Calculate checksum of an entire packet 3345 * @skb: packet to process 3346 * 3347 * This function calculates the checksum over the entire packet plus 3348 * the value of skb->csum. The latter can be used to supply the 3349 * checksum of a pseudo header as used by TCP/UDP. It returns the 3350 * checksum. 3351 * 3352 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 3353 * this function can be used to verify that checksum on received 3354 * packets. In that case the function should return zero if the 3355 * checksum is correct. In particular, this function will return zero 3356 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 3357 * hardware has already verified the correctness of the checksum. 3358 */ 3359static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 3360{ 3361 return skb_csum_unnecessary(skb) ? 3362 0 : __skb_checksum_complete(skb); 3363} 3364 3365static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 3366{ 3367 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3368 if (skb->csum_level == 0) 3369 skb->ip_summed = CHECKSUM_NONE; 3370 else 3371 skb->csum_level--; 3372 } 3373} 3374 3375static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 3376{ 3377 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 3378 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 3379 skb->csum_level++; 3380 } else if (skb->ip_summed == CHECKSUM_NONE) { 3381 skb->ip_summed = CHECKSUM_UNNECESSARY; 3382 skb->csum_level = 0; 3383 } 3384} 3385 3386static inline void __skb_mark_checksum_bad(struct sk_buff *skb) 3387{ 3388 /* Mark current checksum as bad (typically called from GRO 3389 * path). In the case that ip_summed is CHECKSUM_NONE 3390 * this must be the first checksum encountered in the packet. 3391 * When ip_summed is CHECKSUM_UNNECESSARY, this is the first 3392 * checksum after the last one validated. For UDP, a zero 3393 * checksum can not be marked as bad. 3394 */ 3395 3396 if (skb->ip_summed == CHECKSUM_NONE || 3397 skb->ip_summed == CHECKSUM_UNNECESSARY) 3398 skb->csum_bad = 1; 3399} 3400 3401/* Check if we need to perform checksum complete validation. 3402 * 3403 * Returns true if checksum complete is needed, false otherwise 3404 * (either checksum is unnecessary or zero checksum is allowed). 3405 */ 3406static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 3407 bool zero_okay, 3408 __sum16 check) 3409{ 3410 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 3411 skb->csum_valid = 1; 3412 __skb_decr_checksum_unnecessary(skb); 3413 return false; 3414 } 3415 3416 return true; 3417} 3418 3419/* For small packets <= CHECKSUM_BREAK peform checksum complete directly 3420 * in checksum_init. 3421 */ 3422#define CHECKSUM_BREAK 76 3423 3424/* Unset checksum-complete 3425 * 3426 * Unset checksum complete can be done when packet is being modified 3427 * (uncompressed for instance) and checksum-complete value is 3428 * invalidated. 3429 */ 3430static inline void skb_checksum_complete_unset(struct sk_buff *skb) 3431{ 3432 if (skb->ip_summed == CHECKSUM_COMPLETE) 3433 skb->ip_summed = CHECKSUM_NONE; 3434} 3435 3436/* Validate (init) checksum based on checksum complete. 3437 * 3438 * Return values: 3439 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 3440 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 3441 * checksum is stored in skb->csum for use in __skb_checksum_complete 3442 * non-zero: value of invalid checksum 3443 * 3444 */ 3445static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 3446 bool complete, 3447 __wsum psum) 3448{ 3449 if (skb->ip_summed == CHECKSUM_COMPLETE) { 3450 if (!csum_fold(csum_add(psum, skb->csum))) { 3451 skb->csum_valid = 1; 3452 return 0; 3453 } 3454 } else if (skb->csum_bad) { 3455 /* ip_summed == CHECKSUM_NONE in this case */ 3456 return (__force __sum16)1; 3457 } 3458 3459 skb->csum = psum; 3460 3461 if (complete || skb->len <= CHECKSUM_BREAK) { 3462 __sum16 csum; 3463 3464 csum = __skb_checksum_complete(skb); 3465 skb->csum_valid = !csum; 3466 return csum; 3467 } 3468 3469 return 0; 3470} 3471 3472static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 3473{ 3474 return 0; 3475} 3476 3477/* Perform checksum validate (init). Note that this is a macro since we only 3478 * want to calculate the pseudo header which is an input function if necessary. 3479 * First we try to validate without any computation (checksum unnecessary) and 3480 * then calculate based on checksum complete calling the function to compute 3481 * pseudo header. 3482 * 3483 * Return values: 3484 * 0: checksum is validated or try to in skb_checksum_complete 3485 * non-zero: value of invalid checksum 3486 */ 3487#define __skb_checksum_validate(skb, proto, complete, \ 3488 zero_okay, check, compute_pseudo) \ 3489({ \ 3490 __sum16 __ret = 0; \ 3491 skb->csum_valid = 0; \ 3492 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 3493 __ret = __skb_checksum_validate_complete(skb, \ 3494 complete, compute_pseudo(skb, proto)); \ 3495 __ret; \ 3496}) 3497 3498#define skb_checksum_init(skb, proto, compute_pseudo) \ 3499 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 3500 3501#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 3502 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 3503 3504#define skb_checksum_validate(skb, proto, compute_pseudo) \ 3505 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 3506 3507#define skb_checksum_validate_zero_check(skb, proto, check, \ 3508 compute_pseudo) \ 3509 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 3510 3511#define skb_checksum_simple_validate(skb) \ 3512 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 3513 3514static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 3515{ 3516 return (skb->ip_summed == CHECKSUM_NONE && 3517 skb->csum_valid && !skb->csum_bad); 3518} 3519 3520static inline void __skb_checksum_convert(struct sk_buff *skb, 3521 __sum16 check, __wsum pseudo) 3522{ 3523 skb->csum = ~pseudo; 3524 skb->ip_summed = CHECKSUM_COMPLETE; 3525} 3526 3527#define skb_checksum_try_convert(skb, proto, check, compute_pseudo) \ 3528do { \ 3529 if (__skb_checksum_convert_check(skb)) \ 3530 __skb_checksum_convert(skb, check, \ 3531 compute_pseudo(skb, proto)); \ 3532} while (0) 3533 3534static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 3535 u16 start, u16 offset) 3536{ 3537 skb->ip_summed = CHECKSUM_PARTIAL; 3538 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 3539 skb->csum_offset = offset - start; 3540} 3541 3542/* Update skbuf and packet to reflect the remote checksum offload operation. 3543 * When called, ptr indicates the starting point for skb->csum when 3544 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 3545 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 3546 */ 3547static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 3548 int start, int offset, bool nopartial) 3549{ 3550 __wsum delta; 3551 3552 if (!nopartial) { 3553 skb_remcsum_adjust_partial(skb, ptr, start, offset); 3554 return; 3555 } 3556 3557 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 3558 __skb_checksum_complete(skb); 3559 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 3560 } 3561 3562 delta = remcsum_adjust(ptr, skb->csum, start, offset); 3563 3564 /* Adjust skb->csum since we changed the packet */ 3565 skb->csum = csum_add(skb->csum, delta); 3566} 3567 3568static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 3569{ 3570#if IS_ENABLED(CONFIG_NF_CONNTRACK) 3571 return (void *)(skb->_nfct & SKB_NFCT_PTRMASK); 3572#else 3573 return NULL; 3574#endif 3575} 3576 3577#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3578void nf_conntrack_destroy(struct nf_conntrack *nfct); 3579static inline void nf_conntrack_put(struct nf_conntrack *nfct) 3580{ 3581 if (nfct && atomic_dec_and_test(&nfct->use)) 3582 nf_conntrack_destroy(nfct); 3583} 3584static inline void nf_conntrack_get(struct nf_conntrack *nfct) 3585{ 3586 if (nfct) 3587 atomic_inc(&nfct->use); 3588} 3589#endif 3590#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3591static inline void nf_bridge_put(struct nf_bridge_info *nf_bridge) 3592{ 3593 if (nf_bridge && atomic_dec_and_test(&nf_bridge->use)) 3594 kfree(nf_bridge); 3595} 3596static inline void nf_bridge_get(struct nf_bridge_info *nf_bridge) 3597{ 3598 if (nf_bridge) 3599 atomic_inc(&nf_bridge->use); 3600} 3601#endif /* CONFIG_BRIDGE_NETFILTER */ 3602static inline void nf_reset(struct sk_buff *skb) 3603{ 3604#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3605 nf_conntrack_put(skb_nfct(skb)); 3606 skb->_nfct = 0; 3607#endif 3608#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3609 nf_bridge_put(skb->nf_bridge); 3610 skb->nf_bridge = NULL; 3611#endif 3612} 3613 3614static inline void nf_reset_trace(struct sk_buff *skb) 3615{ 3616#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3617 skb->nf_trace = 0; 3618#endif 3619} 3620 3621/* Note: This doesn't put any conntrack and bridge info in dst. */ 3622static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 3623 bool copy) 3624{ 3625#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3626 dst->_nfct = src->_nfct; 3627 nf_conntrack_get(skb_nfct(src)); 3628#endif 3629#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3630 dst->nf_bridge = src->nf_bridge; 3631 nf_bridge_get(src->nf_bridge); 3632#endif 3633#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 3634 if (copy) 3635 dst->nf_trace = src->nf_trace; 3636#endif 3637} 3638 3639static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 3640{ 3641#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 3642 nf_conntrack_put(skb_nfct(dst)); 3643#endif 3644#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 3645 nf_bridge_put(dst->nf_bridge); 3646#endif 3647 __nf_copy(dst, src, true); 3648} 3649 3650#ifdef CONFIG_NETWORK_SECMARK 3651static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3652{ 3653 to->secmark = from->secmark; 3654} 3655 3656static inline void skb_init_secmark(struct sk_buff *skb) 3657{ 3658 skb->secmark = 0; 3659} 3660#else 3661static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 3662{ } 3663 3664static inline void skb_init_secmark(struct sk_buff *skb) 3665{ } 3666#endif 3667 3668static inline bool skb_irq_freeable(const struct sk_buff *skb) 3669{ 3670 return !skb->destructor && 3671#if IS_ENABLED(CONFIG_XFRM) 3672 !skb->sp && 3673#endif 3674 !skb_nfct(skb) && 3675 !skb->_skb_refdst && 3676 !skb_has_frag_list(skb); 3677} 3678 3679static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 3680{ 3681 skb->queue_mapping = queue_mapping; 3682} 3683 3684static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 3685{ 3686 return skb->queue_mapping; 3687} 3688 3689static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 3690{ 3691 to->queue_mapping = from->queue_mapping; 3692} 3693 3694static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 3695{ 3696 skb->queue_mapping = rx_queue + 1; 3697} 3698 3699static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 3700{ 3701 return skb->queue_mapping - 1; 3702} 3703 3704static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 3705{ 3706 return skb->queue_mapping != 0; 3707} 3708 3709static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 3710{ 3711 skb->dst_pending_confirm = val; 3712} 3713 3714static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 3715{ 3716 return skb->dst_pending_confirm != 0; 3717} 3718 3719static inline struct sec_path *skb_sec_path(struct sk_buff *skb) 3720{ 3721#ifdef CONFIG_XFRM 3722 return skb->sp; 3723#else 3724 return NULL; 3725#endif 3726} 3727 3728/* Keeps track of mac header offset relative to skb->head. 3729 * It is useful for TSO of Tunneling protocol. e.g. GRE. 3730 * For non-tunnel skb it points to skb_mac_header() and for 3731 * tunnel skb it points to outer mac header. 3732 * Keeps track of level of encapsulation of network headers. 3733 */ 3734struct skb_gso_cb { 3735 union { 3736 int mac_offset; 3737 int data_offset; 3738 }; 3739 int encap_level; 3740 __wsum csum; 3741 __u16 csum_start; 3742}; 3743#define SKB_SGO_CB_OFFSET 32 3744#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_SGO_CB_OFFSET)) 3745 3746static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 3747{ 3748 return (skb_mac_header(inner_skb) - inner_skb->head) - 3749 SKB_GSO_CB(inner_skb)->mac_offset; 3750} 3751 3752static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 3753{ 3754 int new_headroom, headroom; 3755 int ret; 3756 3757 headroom = skb_headroom(skb); 3758 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 3759 if (ret) 3760 return ret; 3761 3762 new_headroom = skb_headroom(skb); 3763 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 3764 return 0; 3765} 3766 3767static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 3768{ 3769 /* Do not update partial checksums if remote checksum is enabled. */ 3770 if (skb->remcsum_offload) 3771 return; 3772 3773 SKB_GSO_CB(skb)->csum = res; 3774 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 3775} 3776 3777/* Compute the checksum for a gso segment. First compute the checksum value 3778 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 3779 * then add in skb->csum (checksum from csum_start to end of packet). 3780 * skb->csum and csum_start are then updated to reflect the checksum of the 3781 * resultant packet starting from the transport header-- the resultant checksum 3782 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 3783 * header. 3784 */ 3785static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 3786{ 3787 unsigned char *csum_start = skb_transport_header(skb); 3788 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 3789 __wsum partial = SKB_GSO_CB(skb)->csum; 3790 3791 SKB_GSO_CB(skb)->csum = res; 3792 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 3793 3794 return csum_fold(csum_partial(csum_start, plen, partial)); 3795} 3796 3797static inline bool skb_is_gso(const struct sk_buff *skb) 3798{ 3799 return skb_shinfo(skb)->gso_size; 3800} 3801 3802/* Note: Should be called only if skb_is_gso(skb) is true */ 3803static inline bool skb_is_gso_v6(const struct sk_buff *skb) 3804{ 3805 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 3806} 3807 3808static inline void skb_gso_reset(struct sk_buff *skb) 3809{ 3810 skb_shinfo(skb)->gso_size = 0; 3811 skb_shinfo(skb)->gso_segs = 0; 3812 skb_shinfo(skb)->gso_type = 0; 3813} 3814 3815void __skb_warn_lro_forwarding(const struct sk_buff *skb); 3816 3817static inline bool skb_warn_if_lro(const struct sk_buff *skb) 3818{ 3819 /* LRO sets gso_size but not gso_type, whereas if GSO is really 3820 * wanted then gso_type will be set. */ 3821 const struct skb_shared_info *shinfo = skb_shinfo(skb); 3822 3823 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 3824 unlikely(shinfo->gso_type == 0)) { 3825 __skb_warn_lro_forwarding(skb); 3826 return true; 3827 } 3828 return false; 3829} 3830 3831static inline void skb_forward_csum(struct sk_buff *skb) 3832{ 3833 /* Unfortunately we don't support this one. Any brave souls? */ 3834 if (skb->ip_summed == CHECKSUM_COMPLETE) 3835 skb->ip_summed = CHECKSUM_NONE; 3836} 3837 3838/** 3839 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 3840 * @skb: skb to check 3841 * 3842 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 3843 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 3844 * use this helper, to document places where we make this assertion. 3845 */ 3846static inline void skb_checksum_none_assert(const struct sk_buff *skb) 3847{ 3848#ifdef DEBUG 3849 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 3850#endif 3851} 3852 3853bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 3854 3855int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 3856struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 3857 unsigned int transport_len, 3858 __sum16(*skb_chkf)(struct sk_buff *skb)); 3859 3860/** 3861 * skb_head_is_locked - Determine if the skb->head is locked down 3862 * @skb: skb to check 3863 * 3864 * The head on skbs build around a head frag can be removed if they are 3865 * not cloned. This function returns true if the skb head is locked down 3866 * due to either being allocated via kmalloc, or by being a clone with 3867 * multiple references to the head. 3868 */ 3869static inline bool skb_head_is_locked(const struct sk_buff *skb) 3870{ 3871 return !skb->head_frag || skb_cloned(skb); 3872} 3873 3874/** 3875 * skb_gso_network_seglen - Return length of individual segments of a gso packet 3876 * 3877 * @skb: GSO skb 3878 * 3879 * skb_gso_network_seglen is used to determine the real size of the 3880 * individual segments, including Layer3 (IP, IPv6) and L4 headers (TCP/UDP). 3881 * 3882 * The MAC/L2 header is not accounted for. 3883 */ 3884static inline unsigned int skb_gso_network_seglen(const struct sk_buff *skb) 3885{ 3886 unsigned int hdr_len = skb_transport_header(skb) - 3887 skb_network_header(skb); 3888 return hdr_len + skb_gso_transport_seglen(skb); 3889} 3890 3891/* Local Checksum Offload. 3892 * Compute outer checksum based on the assumption that the 3893 * inner checksum will be offloaded later. 3894 * See Documentation/networking/checksum-offloads.txt for 3895 * explanation of how this works. 3896 * Fill in outer checksum adjustment (e.g. with sum of outer 3897 * pseudo-header) before calling. 3898 * Also ensure that inner checksum is in linear data area. 3899 */ 3900static inline __wsum lco_csum(struct sk_buff *skb) 3901{ 3902 unsigned char *csum_start = skb_checksum_start(skb); 3903 unsigned char *l4_hdr = skb_transport_header(skb); 3904 __wsum partial; 3905 3906 /* Start with complement of inner checksum adjustment */ 3907 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 3908 skb->csum_offset)); 3909 3910 /* Add in checksum of our headers (incl. outer checksum 3911 * adjustment filled in by caller) and return result. 3912 */ 3913 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 3914} 3915 3916#endif /* __KERNEL__ */ 3917#endif /* _LINUX_SKBUFF_H */