at v5.18 146 kB view raw
1/* SPDX-License-Identifier: GPL-2.0-or-later */ 2/* 3 * Definitions for the 'struct sk_buff' memory handlers. 4 * 5 * Authors: 6 * Alan Cox, <gw4pts@gw4pts.ampr.org> 7 * Florian La Roche, <rzsfl@rz.uni-sb.de> 8 */ 9 10#ifndef _LINUX_SKBUFF_H 11#define _LINUX_SKBUFF_H 12 13#include <linux/kernel.h> 14#include <linux/compiler.h> 15#include <linux/time.h> 16#include <linux/bug.h> 17#include <linux/bvec.h> 18#include <linux/cache.h> 19#include <linux/rbtree.h> 20#include <linux/socket.h> 21#include <linux/refcount.h> 22 23#include <linux/atomic.h> 24#include <asm/types.h> 25#include <linux/spinlock.h> 26#include <linux/net.h> 27#include <linux/textsearch.h> 28#include <net/checksum.h> 29#include <linux/rcupdate.h> 30#include <linux/hrtimer.h> 31#include <linux/dma-mapping.h> 32#include <linux/netdev_features.h> 33#include <linux/sched.h> 34#include <linux/sched/clock.h> 35#include <net/flow_dissector.h> 36#include <linux/splice.h> 37#include <linux/in6.h> 38#include <linux/if_packet.h> 39#include <linux/llist.h> 40#include <net/flow.h> 41#include <net/page_pool.h> 42#if IS_ENABLED(CONFIG_NF_CONNTRACK) 43#include <linux/netfilter/nf_conntrack_common.h> 44#endif 45 46/* The interface for checksum offload between the stack and networking drivers 47 * is as follows... 48 * 49 * A. IP checksum related features 50 * 51 * Drivers advertise checksum offload capabilities in the features of a device. 52 * From the stack's point of view these are capabilities offered by the driver. 53 * A driver typically only advertises features that it is capable of offloading 54 * to its device. 55 * 56 * The checksum related features are: 57 * 58 * NETIF_F_HW_CSUM - The driver (or its device) is able to compute one 59 * IP (one's complement) checksum for any combination 60 * of protocols or protocol layering. The checksum is 61 * computed and set in a packet per the CHECKSUM_PARTIAL 62 * interface (see below). 63 * 64 * NETIF_F_IP_CSUM - Driver (device) is only able to checksum plain 65 * TCP or UDP packets over IPv4. These are specifically 66 * unencapsulated packets of the form IPv4|TCP or 67 * IPv4|UDP where the Protocol field in the IPv4 header 68 * is TCP or UDP. The IPv4 header may contain IP options. 69 * This feature cannot be set in features for a device 70 * with NETIF_F_HW_CSUM also set. This feature is being 71 * DEPRECATED (see below). 72 * 73 * NETIF_F_IPV6_CSUM - Driver (device) is only able to checksum plain 74 * TCP or UDP packets over IPv6. These are specifically 75 * unencapsulated packets of the form IPv6|TCP or 76 * IPv6|UDP where the Next Header field in the IPv6 77 * header is either TCP or UDP. IPv6 extension headers 78 * are not supported with this feature. This feature 79 * cannot be set in features for a device with 80 * NETIF_F_HW_CSUM also set. This feature is being 81 * DEPRECATED (see below). 82 * 83 * NETIF_F_RXCSUM - Driver (device) performs receive checksum offload. 84 * This flag is only used to disable the RX checksum 85 * feature for a device. The stack will accept receive 86 * checksum indication in packets received on a device 87 * regardless of whether NETIF_F_RXCSUM is set. 88 * 89 * B. Checksumming of received packets by device. Indication of checksum 90 * verification is set in skb->ip_summed. Possible values are: 91 * 92 * CHECKSUM_NONE: 93 * 94 * Device did not checksum this packet e.g. due to lack of capabilities. 95 * The packet contains full (though not verified) checksum in packet but 96 * not in skb->csum. Thus, skb->csum is undefined in this case. 97 * 98 * CHECKSUM_UNNECESSARY: 99 * 100 * The hardware you're dealing with doesn't calculate the full checksum 101 * (as in CHECKSUM_COMPLETE), but it does parse headers and verify checksums 102 * for specific protocols. For such packets it will set CHECKSUM_UNNECESSARY 103 * if their checksums are okay. skb->csum is still undefined in this case 104 * though. A driver or device must never modify the checksum field in the 105 * packet even if checksum is verified. 106 * 107 * CHECKSUM_UNNECESSARY is applicable to following protocols: 108 * TCP: IPv6 and IPv4. 109 * UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 110 * zero UDP checksum for either IPv4 or IPv6, the networking stack 111 * may perform further validation in this case. 112 * GRE: only if the checksum is present in the header. 113 * SCTP: indicates the CRC in SCTP header has been validated. 114 * FCOE: indicates the CRC in FC frame has been validated. 115 * 116 * skb->csum_level indicates the number of consecutive checksums found in 117 * the packet minus one that have been verified as CHECKSUM_UNNECESSARY. 118 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 119 * and a device is able to verify the checksums for UDP (possibly zero), 120 * GRE (checksum flag is set) and TCP, skb->csum_level would be set to 121 * two. If the device were only able to verify the UDP checksum and not 122 * GRE, either because it doesn't support GRE checksum or because GRE 123 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 124 * not considered in this case). 125 * 126 * CHECKSUM_COMPLETE: 127 * 128 * This is the most generic way. The device supplied checksum of the _whole_ 129 * packet as seen by netif_rx() and fills in skb->csum. This means the 130 * hardware doesn't need to parse L3/L4 headers to implement this. 131 * 132 * Notes: 133 * - Even if device supports only some protocols, but is able to produce 134 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 135 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 136 * 137 * CHECKSUM_PARTIAL: 138 * 139 * A checksum is set up to be offloaded to a device as described in the 140 * output description for CHECKSUM_PARTIAL. This may occur on a packet 141 * received directly from another Linux OS, e.g., a virtualized Linux kernel 142 * on the same host, or it may be set in the input path in GRO or remote 143 * checksum offload. For the purposes of checksum verification, the checksum 144 * referred to by skb->csum_start + skb->csum_offset and any preceding 145 * checksums in the packet are considered verified. Any checksums in the 146 * packet that are after the checksum being offloaded are not considered to 147 * be verified. 148 * 149 * C. Checksumming on transmit for non-GSO. The stack requests checksum offload 150 * in the skb->ip_summed for a packet. Values are: 151 * 152 * CHECKSUM_PARTIAL: 153 * 154 * The driver is required to checksum the packet as seen by hard_start_xmit() 155 * from skb->csum_start up to the end, and to record/write the checksum at 156 * offset skb->csum_start + skb->csum_offset. A driver may verify that the 157 * csum_start and csum_offset values are valid values given the length and 158 * offset of the packet, but it should not attempt to validate that the 159 * checksum refers to a legitimate transport layer checksum -- it is the 160 * purview of the stack to validate that csum_start and csum_offset are set 161 * correctly. 162 * 163 * When the stack requests checksum offload for a packet, the driver MUST 164 * ensure that the checksum is set correctly. A driver can either offload the 165 * checksum calculation to the device, or call skb_checksum_help (in the case 166 * that the device does not support offload for a particular checksum). 167 * 168 * NETIF_F_IP_CSUM and NETIF_F_IPV6_CSUM are being deprecated in favor of 169 * NETIF_F_HW_CSUM. New devices should use NETIF_F_HW_CSUM to indicate 170 * checksum offload capability. 171 * skb_csum_hwoffload_help() can be called to resolve CHECKSUM_PARTIAL based 172 * on network device checksumming capabilities: if a packet does not match 173 * them, skb_checksum_help or skb_crc32c_help (depending on the value of 174 * csum_not_inet, see item D.) is called to resolve the checksum. 175 * 176 * CHECKSUM_NONE: 177 * 178 * The skb was already checksummed by the protocol, or a checksum is not 179 * required. 180 * 181 * CHECKSUM_UNNECESSARY: 182 * 183 * This has the same meaning as CHECKSUM_NONE for checksum offload on 184 * output. 185 * 186 * CHECKSUM_COMPLETE: 187 * Not used in checksum output. If a driver observes a packet with this value 188 * set in skbuff, it should treat the packet as if CHECKSUM_NONE were set. 189 * 190 * D. Non-IP checksum (CRC) offloads 191 * 192 * NETIF_F_SCTP_CRC - This feature indicates that a device is capable of 193 * offloading the SCTP CRC in a packet. To perform this offload the stack 194 * will set csum_start and csum_offset accordingly, set ip_summed to 195 * CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication in 196 * the skbuff that the CHECKSUM_PARTIAL refers to CRC32c. 197 * A driver that supports both IP checksum offload and SCTP CRC32c offload 198 * must verify which offload is configured for a packet by testing the 199 * value of skb->csum_not_inet; skb_crc32c_csum_help is provided to resolve 200 * CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 201 * 202 * NETIF_F_FCOE_CRC - This feature indicates that a device is capable of 203 * offloading the FCOE CRC in a packet. To perform this offload the stack 204 * will set ip_summed to CHECKSUM_PARTIAL and set csum_start and csum_offset 205 * accordingly. Note that there is no indication in the skbuff that the 206 * CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports 207 * both IP checksum offload and FCOE CRC offload must verify which offload 208 * is configured for a packet, presumably by inspecting packet headers. 209 * 210 * E. Checksumming on output with GSO. 211 * 212 * In the case of a GSO packet (skb_is_gso(skb) is true), checksum offload 213 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 214 * gso_type is SKB_GSO_TCPV4 or SKB_GSO_TCPV6, TCP checksum offload as 215 * part of the GSO operation is implied. If a checksum is being offloaded 216 * with GSO then ip_summed is CHECKSUM_PARTIAL, and both csum_start and 217 * csum_offset are set to refer to the outermost checksum being offloaded 218 * (two offloaded checksums are possible with UDP encapsulation). 219 */ 220 221/* Don't change this without changing skb_csum_unnecessary! */ 222#define CHECKSUM_NONE 0 223#define CHECKSUM_UNNECESSARY 1 224#define CHECKSUM_COMPLETE 2 225#define CHECKSUM_PARTIAL 3 226 227/* Maximum value in skb->csum_level */ 228#define SKB_MAX_CSUM_LEVEL 3 229 230#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 231#define SKB_WITH_OVERHEAD(X) \ 232 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 233#define SKB_MAX_ORDER(X, ORDER) \ 234 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 235#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 236#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 237 238/* return minimum truesize of one skb containing X bytes of data */ 239#define SKB_TRUESIZE(X) ((X) + \ 240 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 241 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 242 243struct ahash_request; 244struct net_device; 245struct scatterlist; 246struct pipe_inode_info; 247struct iov_iter; 248struct napi_struct; 249struct bpf_prog; 250union bpf_attr; 251struct skb_ext; 252 253#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 254struct nf_bridge_info { 255 enum { 256 BRNF_PROTO_UNCHANGED, 257 BRNF_PROTO_8021Q, 258 BRNF_PROTO_PPPOE 259 } orig_proto:8; 260 u8 pkt_otherhost:1; 261 u8 in_prerouting:1; 262 u8 bridged_dnat:1; 263 __u16 frag_max_size; 264 struct net_device *physindev; 265 266 /* always valid & non-NULL from FORWARD on, for physdev match */ 267 struct net_device *physoutdev; 268 union { 269 /* prerouting: detect dnat in orig/reply direction */ 270 __be32 ipv4_daddr; 271 struct in6_addr ipv6_daddr; 272 273 /* after prerouting + nat detected: store original source 274 * mac since neigh resolution overwrites it, only used while 275 * skb is out in neigh layer. 276 */ 277 char neigh_header[8]; 278 }; 279}; 280#endif 281 282#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 283/* Chain in tc_skb_ext will be used to share the tc chain with 284 * ovs recirc_id. It will be set to the current chain by tc 285 * and read by ovs to recirc_id. 286 */ 287struct tc_skb_ext { 288 __u32 chain; 289 __u16 mru; 290 __u16 zone; 291 u8 post_ct:1; 292 u8 post_ct_snat:1; 293 u8 post_ct_dnat:1; 294}; 295#endif 296 297struct sk_buff_head { 298 /* These two members must be first to match sk_buff. */ 299 struct_group_tagged(sk_buff_list, list, 300 struct sk_buff *next; 301 struct sk_buff *prev; 302 ); 303 304 __u32 qlen; 305 spinlock_t lock; 306}; 307 308struct sk_buff; 309 310/* The reason of skb drop, which is used in kfree_skb_reason(). 311 * en...maybe they should be splited by group? 312 * 313 * Each item here should also be in 'TRACE_SKB_DROP_REASON', which is 314 * used to translate the reason to string. 315 */ 316enum skb_drop_reason { 317 SKB_NOT_DROPPED_YET = 0, 318 SKB_DROP_REASON_NOT_SPECIFIED, /* drop reason is not specified */ 319 SKB_DROP_REASON_NO_SOCKET, /* socket not found */ 320 SKB_DROP_REASON_PKT_TOO_SMALL, /* packet size is too small */ 321 SKB_DROP_REASON_TCP_CSUM, /* TCP checksum error */ 322 SKB_DROP_REASON_SOCKET_FILTER, /* dropped by socket filter */ 323 SKB_DROP_REASON_UDP_CSUM, /* UDP checksum error */ 324 SKB_DROP_REASON_NETFILTER_DROP, /* dropped by netfilter */ 325 SKB_DROP_REASON_OTHERHOST, /* packet don't belong to current 326 * host (interface is in promisc 327 * mode) 328 */ 329 SKB_DROP_REASON_IP_CSUM, /* IP checksum error */ 330 SKB_DROP_REASON_IP_INHDR, /* there is something wrong with 331 * IP header (see 332 * IPSTATS_MIB_INHDRERRORS) 333 */ 334 SKB_DROP_REASON_IP_RPFILTER, /* IP rpfilter validate failed. 335 * see the document for rp_filter 336 * in ip-sysctl.rst for more 337 * information 338 */ 339 SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST, /* destination address of L2 340 * is multicast, but L3 is 341 * unicast. 342 */ 343 SKB_DROP_REASON_XFRM_POLICY, /* xfrm policy check failed */ 344 SKB_DROP_REASON_IP_NOPROTO, /* no support for IP protocol */ 345 SKB_DROP_REASON_SOCKET_RCVBUFF, /* socket receive buff is full */ 346 SKB_DROP_REASON_PROTO_MEM, /* proto memory limition, such as 347 * udp packet drop out of 348 * udp_memory_allocated. 349 */ 350 SKB_DROP_REASON_TCP_MD5NOTFOUND, /* no MD5 hash and one 351 * expected, corresponding 352 * to LINUX_MIB_TCPMD5NOTFOUND 353 */ 354 SKB_DROP_REASON_TCP_MD5UNEXPECTED, /* MD5 hash and we're not 355 * expecting one, corresponding 356 * to LINUX_MIB_TCPMD5UNEXPECTED 357 */ 358 SKB_DROP_REASON_TCP_MD5FAILURE, /* MD5 hash and its wrong, 359 * corresponding to 360 * LINUX_MIB_TCPMD5FAILURE 361 */ 362 SKB_DROP_REASON_SOCKET_BACKLOG, /* failed to add skb to socket 363 * backlog (see 364 * LINUX_MIB_TCPBACKLOGDROP) 365 */ 366 SKB_DROP_REASON_TCP_FLAGS, /* TCP flags invalid */ 367 SKB_DROP_REASON_TCP_ZEROWINDOW, /* TCP receive window size is zero, 368 * see LINUX_MIB_TCPZEROWINDOWDROP 369 */ 370 SKB_DROP_REASON_TCP_OLD_DATA, /* the TCP data reveived is already 371 * received before (spurious retrans 372 * may happened), see 373 * LINUX_MIB_DELAYEDACKLOST 374 */ 375 SKB_DROP_REASON_TCP_OVERWINDOW, /* the TCP data is out of window, 376 * the seq of the first byte exceed 377 * the right edges of receive 378 * window 379 */ 380 SKB_DROP_REASON_TCP_OFOMERGE, /* the data of skb is already in 381 * the ofo queue, corresponding to 382 * LINUX_MIB_TCPOFOMERGE 383 */ 384 SKB_DROP_REASON_IP_OUTNOROUTES, /* route lookup failed */ 385 SKB_DROP_REASON_BPF_CGROUP_EGRESS, /* dropped by 386 * BPF_PROG_TYPE_CGROUP_SKB 387 * eBPF program 388 */ 389 SKB_DROP_REASON_IPV6DISABLED, /* IPv6 is disabled on the device */ 390 SKB_DROP_REASON_NEIGH_CREATEFAIL, /* failed to create neigh 391 * entry 392 */ 393 SKB_DROP_REASON_NEIGH_FAILED, /* neigh entry in failed state */ 394 SKB_DROP_REASON_NEIGH_QUEUEFULL, /* arp_queue for neigh 395 * entry is full 396 */ 397 SKB_DROP_REASON_NEIGH_DEAD, /* neigh entry is dead */ 398 SKB_DROP_REASON_TC_EGRESS, /* dropped in TC egress HOOK */ 399 SKB_DROP_REASON_QDISC_DROP, /* dropped by qdisc when packet 400 * outputting (failed to enqueue to 401 * current qdisc) 402 */ 403 SKB_DROP_REASON_CPU_BACKLOG, /* failed to enqueue the skb to 404 * the per CPU backlog queue. This 405 * can be caused by backlog queue 406 * full (see netdev_max_backlog in 407 * net.rst) or RPS flow limit 408 */ 409 SKB_DROP_REASON_XDP, /* dropped by XDP in input path */ 410 SKB_DROP_REASON_TC_INGRESS, /* dropped in TC ingress HOOK */ 411 SKB_DROP_REASON_PTYPE_ABSENT, /* not packet_type found to handle 412 * the skb. For an etner packet, 413 * this means that L3 protocol is 414 * not supported 415 */ 416 SKB_DROP_REASON_SKB_CSUM, /* sk_buff checksum computation 417 * error 418 */ 419 SKB_DROP_REASON_SKB_GSO_SEG, /* gso segmentation error */ 420 SKB_DROP_REASON_SKB_UCOPY_FAULT, /* failed to copy data from 421 * user space, e.g., via 422 * zerocopy_sg_from_iter() 423 * or skb_orphan_frags_rx() 424 */ 425 SKB_DROP_REASON_DEV_HDR, /* device driver specific 426 * header/metadata is invalid 427 */ 428 /* the device is not ready to xmit/recv due to any of its data 429 * structure that is not up/ready/initialized, e.g., the IFF_UP is 430 * not set, or driver specific tun->tfiles[txq] is not initialized 431 */ 432 SKB_DROP_REASON_DEV_READY, 433 SKB_DROP_REASON_FULL_RING, /* ring buffer is full */ 434 SKB_DROP_REASON_NOMEM, /* error due to OOM */ 435 SKB_DROP_REASON_HDR_TRUNC, /* failed to trunc/extract the header 436 * from networking data, e.g., failed 437 * to pull the protocol header from 438 * frags via pskb_may_pull() 439 */ 440 SKB_DROP_REASON_TAP_FILTER, /* dropped by (ebpf) filter directly 441 * attached to tun/tap, e.g., via 442 * TUNSETFILTEREBPF 443 */ 444 SKB_DROP_REASON_TAP_TXFILTER, /* dropped by tx filter implemented 445 * at tun/tap, e.g., check_filter() 446 */ 447 SKB_DROP_REASON_MAX, 448}; 449 450/* To allow 64K frame to be packed as single skb without frag_list we 451 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 452 * buffers which do not start on a page boundary. 453 * 454 * Since GRO uses frags we allocate at least 16 regardless of page 455 * size. 456 */ 457#if (65536/PAGE_SIZE + 1) < 16 458#define MAX_SKB_FRAGS 16UL 459#else 460#define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 461#endif 462extern int sysctl_max_skb_frags; 463 464/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 465 * segment using its current segmentation instead. 466 */ 467#define GSO_BY_FRAGS 0xFFFF 468 469typedef struct bio_vec skb_frag_t; 470 471/** 472 * skb_frag_size() - Returns the size of a skb fragment 473 * @frag: skb fragment 474 */ 475static inline unsigned int skb_frag_size(const skb_frag_t *frag) 476{ 477 return frag->bv_len; 478} 479 480/** 481 * skb_frag_size_set() - Sets the size of a skb fragment 482 * @frag: skb fragment 483 * @size: size of fragment 484 */ 485static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 486{ 487 frag->bv_len = size; 488} 489 490/** 491 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 492 * @frag: skb fragment 493 * @delta: value to add 494 */ 495static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 496{ 497 frag->bv_len += delta; 498} 499 500/** 501 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 502 * @frag: skb fragment 503 * @delta: value to subtract 504 */ 505static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 506{ 507 frag->bv_len -= delta; 508} 509 510/** 511 * skb_frag_must_loop - Test if %p is a high memory page 512 * @p: fragment's page 513 */ 514static inline bool skb_frag_must_loop(struct page *p) 515{ 516#if defined(CONFIG_HIGHMEM) 517 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 518 return true; 519#endif 520 return false; 521} 522 523/** 524 * skb_frag_foreach_page - loop over pages in a fragment 525 * 526 * @f: skb frag to operate on 527 * @f_off: offset from start of f->bv_page 528 * @f_len: length from f_off to loop over 529 * @p: (temp var) current page 530 * @p_off: (temp var) offset from start of current page, 531 * non-zero only on first page. 532 * @p_len: (temp var) length in current page, 533 * < PAGE_SIZE only on first and last page. 534 * @copied: (temp var) length so far, excluding current p_len. 535 * 536 * A fragment can hold a compound page, in which case per-page 537 * operations, notably kmap_atomic, must be called for each 538 * regular page. 539 */ 540#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 541 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 542 p_off = (f_off) & (PAGE_SIZE - 1), \ 543 p_len = skb_frag_must_loop(p) ? \ 544 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 545 copied = 0; \ 546 copied < f_len; \ 547 copied += p_len, p++, p_off = 0, \ 548 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 549 550#define HAVE_HW_TIME_STAMP 551 552/** 553 * struct skb_shared_hwtstamps - hardware time stamps 554 * @hwtstamp: hardware time stamp transformed into duration 555 * since arbitrary point in time 556 * 557 * Software time stamps generated by ktime_get_real() are stored in 558 * skb->tstamp. 559 * 560 * hwtstamps can only be compared against other hwtstamps from 561 * the same device. 562 * 563 * This structure is attached to packets as part of the 564 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 565 */ 566struct skb_shared_hwtstamps { 567 ktime_t hwtstamp; 568}; 569 570/* Definitions for tx_flags in struct skb_shared_info */ 571enum { 572 /* generate hardware time stamp */ 573 SKBTX_HW_TSTAMP = 1 << 0, 574 575 /* generate software time stamp when queueing packet to NIC */ 576 SKBTX_SW_TSTAMP = 1 << 1, 577 578 /* device driver is going to provide hardware time stamp */ 579 SKBTX_IN_PROGRESS = 1 << 2, 580 581 /* generate wifi status information (where possible) */ 582 SKBTX_WIFI_STATUS = 1 << 4, 583 584 /* generate software time stamp when entering packet scheduling */ 585 SKBTX_SCHED_TSTAMP = 1 << 6, 586}; 587 588#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 589 SKBTX_SCHED_TSTAMP) 590#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | SKBTX_ANY_SW_TSTAMP) 591 592/* Definitions for flags in struct skb_shared_info */ 593enum { 594 /* use zcopy routines */ 595 SKBFL_ZEROCOPY_ENABLE = BIT(0), 596 597 /* This indicates at least one fragment might be overwritten 598 * (as in vmsplice(), sendfile() ...) 599 * If we need to compute a TX checksum, we'll need to copy 600 * all frags to avoid possible bad checksum 601 */ 602 SKBFL_SHARED_FRAG = BIT(1), 603 604 /* segment contains only zerocopy data and should not be 605 * charged to the kernel memory. 606 */ 607 SKBFL_PURE_ZEROCOPY = BIT(2), 608}; 609 610#define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 611#define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY) 612 613/* 614 * The callback notifies userspace to release buffers when skb DMA is done in 615 * lower device, the skb last reference should be 0 when calling this. 616 * The zerocopy_success argument is true if zero copy transmit occurred, 617 * false on data copy or out of memory error caused by data copy attempt. 618 * The ctx field is used to track device context. 619 * The desc field is used to track userspace buffer index. 620 */ 621struct ubuf_info { 622 void (*callback)(struct sk_buff *, struct ubuf_info *, 623 bool zerocopy_success); 624 union { 625 struct { 626 unsigned long desc; 627 void *ctx; 628 }; 629 struct { 630 u32 id; 631 u16 len; 632 u16 zerocopy:1; 633 u32 bytelen; 634 }; 635 }; 636 refcount_t refcnt; 637 u8 flags; 638 639 struct mmpin { 640 struct user_struct *user; 641 unsigned int num_pg; 642 } mmp; 643}; 644 645#define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 646 647int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 648void mm_unaccount_pinned_pages(struct mmpin *mmp); 649 650struct ubuf_info *msg_zerocopy_alloc(struct sock *sk, size_t size); 651struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 652 struct ubuf_info *uarg); 653 654void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 655 656void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 657 bool success); 658 659int skb_zerocopy_iter_dgram(struct sk_buff *skb, struct msghdr *msg, int len); 660int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 661 struct msghdr *msg, int len, 662 struct ubuf_info *uarg); 663 664/* This data is invariant across clones and lives at 665 * the end of the header data, ie. at skb->end. 666 */ 667struct skb_shared_info { 668 __u8 flags; 669 __u8 meta_len; 670 __u8 nr_frags; 671 __u8 tx_flags; 672 unsigned short gso_size; 673 /* Warning: this field is not always filled in (UFO)! */ 674 unsigned short gso_segs; 675 struct sk_buff *frag_list; 676 struct skb_shared_hwtstamps hwtstamps; 677 unsigned int gso_type; 678 u32 tskey; 679 680 /* 681 * Warning : all fields before dataref are cleared in __alloc_skb() 682 */ 683 atomic_t dataref; 684 unsigned int xdp_frags_size; 685 686 /* Intermediate layers must ensure that destructor_arg 687 * remains valid until skb destructor */ 688 void * destructor_arg; 689 690 /* must be last field, see pskb_expand_head() */ 691 skb_frag_t frags[MAX_SKB_FRAGS]; 692}; 693 694/* We divide dataref into two halves. The higher 16 bits hold references 695 * to the payload part of skb->data. The lower 16 bits hold references to 696 * the entire skb->data. A clone of a headerless skb holds the length of 697 * the header in skb->hdr_len. 698 * 699 * All users must obey the rule that the skb->data reference count must be 700 * greater than or equal to the payload reference count. 701 * 702 * Holding a reference to the payload part means that the user does not 703 * care about modifications to the header part of skb->data. 704 */ 705#define SKB_DATAREF_SHIFT 16 706#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 707 708 709enum { 710 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 711 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 712 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 713}; 714 715enum { 716 SKB_GSO_TCPV4 = 1 << 0, 717 718 /* This indicates the skb is from an untrusted source. */ 719 SKB_GSO_DODGY = 1 << 1, 720 721 /* This indicates the tcp segment has CWR set. */ 722 SKB_GSO_TCP_ECN = 1 << 2, 723 724 SKB_GSO_TCP_FIXEDID = 1 << 3, 725 726 SKB_GSO_TCPV6 = 1 << 4, 727 728 SKB_GSO_FCOE = 1 << 5, 729 730 SKB_GSO_GRE = 1 << 6, 731 732 SKB_GSO_GRE_CSUM = 1 << 7, 733 734 SKB_GSO_IPXIP4 = 1 << 8, 735 736 SKB_GSO_IPXIP6 = 1 << 9, 737 738 SKB_GSO_UDP_TUNNEL = 1 << 10, 739 740 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 741 742 SKB_GSO_PARTIAL = 1 << 12, 743 744 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 745 746 SKB_GSO_SCTP = 1 << 14, 747 748 SKB_GSO_ESP = 1 << 15, 749 750 SKB_GSO_UDP = 1 << 16, 751 752 SKB_GSO_UDP_L4 = 1 << 17, 753 754 SKB_GSO_FRAGLIST = 1 << 18, 755}; 756 757#if BITS_PER_LONG > 32 758#define NET_SKBUFF_DATA_USES_OFFSET 1 759#endif 760 761#ifdef NET_SKBUFF_DATA_USES_OFFSET 762typedef unsigned int sk_buff_data_t; 763#else 764typedef unsigned char *sk_buff_data_t; 765#endif 766 767/** 768 * struct sk_buff - socket buffer 769 * @next: Next buffer in list 770 * @prev: Previous buffer in list 771 * @tstamp: Time we arrived/left 772 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 773 * for retransmit timer 774 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 775 * @list: queue head 776 * @ll_node: anchor in an llist (eg socket defer_list) 777 * @sk: Socket we are owned by 778 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in 779 * fragmentation management 780 * @dev: Device we arrived on/are leaving by 781 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 782 * @cb: Control buffer. Free for use by every layer. Put private vars here 783 * @_skb_refdst: destination entry (with norefcount bit) 784 * @sp: the security path, used for xfrm 785 * @len: Length of actual data 786 * @data_len: Data length 787 * @mac_len: Length of link layer header 788 * @hdr_len: writable header length of cloned skb 789 * @csum: Checksum (must include start/offset pair) 790 * @csum_start: Offset from skb->head where checksumming should start 791 * @csum_offset: Offset from csum_start where checksum should be stored 792 * @priority: Packet queueing priority 793 * @ignore_df: allow local fragmentation 794 * @cloned: Head may be cloned (check refcnt to be sure) 795 * @ip_summed: Driver fed us an IP checksum 796 * @nohdr: Payload reference only, must not modify header 797 * @pkt_type: Packet class 798 * @fclone: skbuff clone status 799 * @ipvs_property: skbuff is owned by ipvs 800 * @inner_protocol_type: whether the inner protocol is 801 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 802 * @remcsum_offload: remote checksum offload is enabled 803 * @offload_fwd_mark: Packet was L2-forwarded in hardware 804 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 805 * @tc_skip_classify: do not classify packet. set by IFB device 806 * @tc_at_ingress: used within tc_classify to distinguish in/egress 807 * @redirected: packet was redirected by packet classifier 808 * @from_ingress: packet was redirected from the ingress path 809 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 810 * @peeked: this packet has been seen already, so stats have been 811 * done for it, don't do them again 812 * @nf_trace: netfilter packet trace flag 813 * @protocol: Packet protocol from driver 814 * @destructor: Destruct function 815 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 816 * @_sk_redir: socket redirection information for skmsg 817 * @_nfct: Associated connection, if any (with nfctinfo bits) 818 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 819 * @skb_iif: ifindex of device we arrived on 820 * @tc_index: Traffic control index 821 * @hash: the packet hash 822 * @queue_mapping: Queue mapping for multiqueue devices 823 * @head_frag: skb was allocated from page fragments, 824 * not allocated by kmalloc() or vmalloc(). 825 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 826 * @pp_recycle: mark the packet for recycling instead of freeing (implies 827 * page_pool support on driver) 828 * @active_extensions: active extensions (skb_ext_id types) 829 * @ndisc_nodetype: router type (from link layer) 830 * @ooo_okay: allow the mapping of a socket to a queue to be changed 831 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 832 * ports. 833 * @sw_hash: indicates hash was computed in software stack 834 * @wifi_acked_valid: wifi_acked was set 835 * @wifi_acked: whether frame was acked on wifi or not 836 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 837 * @encapsulation: indicates the inner headers in the skbuff are valid 838 * @encap_hdr_csum: software checksum is needed 839 * @csum_valid: checksum is already valid 840 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 841 * @csum_complete_sw: checksum was completed by software 842 * @csum_level: indicates the number of consecutive checksums found in 843 * the packet minus one that have been verified as 844 * CHECKSUM_UNNECESSARY (max 3) 845 * @dst_pending_confirm: need to confirm neighbour 846 * @decrypted: Decrypted SKB 847 * @slow_gro: state present at GRO time, slower prepare step required 848 * @mono_delivery_time: When set, skb->tstamp has the 849 * delivery_time in mono clock base (i.e. EDT). Otherwise, the 850 * skb->tstamp has the (rcv) timestamp at ingress and 851 * delivery_time at egress. 852 * @napi_id: id of the NAPI struct this skb came from 853 * @sender_cpu: (aka @napi_id) source CPU in XPS 854 * @secmark: security marking 855 * @mark: Generic packet mark 856 * @reserved_tailroom: (aka @mark) number of bytes of free space available 857 * at the tail of an sk_buff 858 * @vlan_present: VLAN tag is present 859 * @vlan_proto: vlan encapsulation protocol 860 * @vlan_tci: vlan tag control information 861 * @inner_protocol: Protocol (encapsulation) 862 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 863 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 864 * @inner_transport_header: Inner transport layer header (encapsulation) 865 * @inner_network_header: Network layer header (encapsulation) 866 * @inner_mac_header: Link layer header (encapsulation) 867 * @transport_header: Transport layer header 868 * @network_header: Network layer header 869 * @mac_header: Link layer header 870 * @kcov_handle: KCOV remote handle for remote coverage collection 871 * @tail: Tail pointer 872 * @end: End pointer 873 * @head: Head of buffer 874 * @data: Data head pointer 875 * @truesize: Buffer size 876 * @users: User count - see {datagram,tcp}.c 877 * @extensions: allocated extensions, valid if active_extensions is nonzero 878 */ 879 880struct sk_buff { 881 union { 882 struct { 883 /* These two members must be first to match sk_buff_head. */ 884 struct sk_buff *next; 885 struct sk_buff *prev; 886 887 union { 888 struct net_device *dev; 889 /* Some protocols might use this space to store information, 890 * while device pointer would be NULL. 891 * UDP receive path is one user. 892 */ 893 unsigned long dev_scratch; 894 }; 895 }; 896 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 897 struct list_head list; 898 struct llist_node ll_node; 899 }; 900 901 union { 902 struct sock *sk; 903 int ip_defrag_offset; 904 }; 905 906 union { 907 ktime_t tstamp; 908 u64 skb_mstamp_ns; /* earliest departure time */ 909 }; 910 /* 911 * This is the control buffer. It is free to use for every 912 * layer. Please put your private variables there. If you 913 * want to keep them across layers you have to do a skb_clone() 914 * first. This is owned by whoever has the skb queued ATM. 915 */ 916 char cb[48] __aligned(8); 917 918 union { 919 struct { 920 unsigned long _skb_refdst; 921 void (*destructor)(struct sk_buff *skb); 922 }; 923 struct list_head tcp_tsorted_anchor; 924#ifdef CONFIG_NET_SOCK_MSG 925 unsigned long _sk_redir; 926#endif 927 }; 928 929#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 930 unsigned long _nfct; 931#endif 932 unsigned int len, 933 data_len; 934 __u16 mac_len, 935 hdr_len; 936 937 /* Following fields are _not_ copied in __copy_skb_header() 938 * Note that queue_mapping is here mostly to fill a hole. 939 */ 940 __u16 queue_mapping; 941 942/* if you move cloned around you also must adapt those constants */ 943#ifdef __BIG_ENDIAN_BITFIELD 944#define CLONED_MASK (1 << 7) 945#else 946#define CLONED_MASK 1 947#endif 948#define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 949 950 /* private: */ 951 __u8 __cloned_offset[0]; 952 /* public: */ 953 __u8 cloned:1, 954 nohdr:1, 955 fclone:2, 956 peeked:1, 957 head_frag:1, 958 pfmemalloc:1, 959 pp_recycle:1; /* page_pool recycle indicator */ 960#ifdef CONFIG_SKB_EXTENSIONS 961 __u8 active_extensions; 962#endif 963 964 /* Fields enclosed in headers group are copied 965 * using a single memcpy() in __copy_skb_header() 966 */ 967 struct_group(headers, 968 969 /* private: */ 970 __u8 __pkt_type_offset[0]; 971 /* public: */ 972 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 973 __u8 ignore_df:1; 974 __u8 nf_trace:1; 975 __u8 ip_summed:2; 976 __u8 ooo_okay:1; 977 978 __u8 l4_hash:1; 979 __u8 sw_hash:1; 980 __u8 wifi_acked_valid:1; 981 __u8 wifi_acked:1; 982 __u8 no_fcs:1; 983 /* Indicates the inner headers are valid in the skbuff. */ 984 __u8 encapsulation:1; 985 __u8 encap_hdr_csum:1; 986 __u8 csum_valid:1; 987 988 /* private: */ 989 __u8 __pkt_vlan_present_offset[0]; 990 /* public: */ 991 __u8 vlan_present:1; /* See PKT_VLAN_PRESENT_BIT */ 992 __u8 csum_complete_sw:1; 993 __u8 csum_level:2; 994 __u8 dst_pending_confirm:1; 995 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */ 996#ifdef CONFIG_NET_CLS_ACT 997 __u8 tc_skip_classify:1; 998 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 999#endif 1000#ifdef CONFIG_IPV6_NDISC_NODETYPE 1001 __u8 ndisc_nodetype:2; 1002#endif 1003 1004 __u8 ipvs_property:1; 1005 __u8 inner_protocol_type:1; 1006 __u8 remcsum_offload:1; 1007#ifdef CONFIG_NET_SWITCHDEV 1008 __u8 offload_fwd_mark:1; 1009 __u8 offload_l3_fwd_mark:1; 1010#endif 1011 __u8 redirected:1; 1012#ifdef CONFIG_NET_REDIRECT 1013 __u8 from_ingress:1; 1014#endif 1015#ifdef CONFIG_NETFILTER_SKIP_EGRESS 1016 __u8 nf_skip_egress:1; 1017#endif 1018#ifdef CONFIG_TLS_DEVICE 1019 __u8 decrypted:1; 1020#endif 1021 __u8 slow_gro:1; 1022 __u8 csum_not_inet:1; 1023 1024#ifdef CONFIG_NET_SCHED 1025 __u16 tc_index; /* traffic control index */ 1026#endif 1027 1028 union { 1029 __wsum csum; 1030 struct { 1031 __u16 csum_start; 1032 __u16 csum_offset; 1033 }; 1034 }; 1035 __u32 priority; 1036 int skb_iif; 1037 __u32 hash; 1038 __be16 vlan_proto; 1039 __u16 vlan_tci; 1040#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1041 union { 1042 unsigned int napi_id; 1043 unsigned int sender_cpu; 1044 }; 1045#endif 1046#ifdef CONFIG_NETWORK_SECMARK 1047 __u32 secmark; 1048#endif 1049 1050 union { 1051 __u32 mark; 1052 __u32 reserved_tailroom; 1053 }; 1054 1055 union { 1056 __be16 inner_protocol; 1057 __u8 inner_ipproto; 1058 }; 1059 1060 __u16 inner_transport_header; 1061 __u16 inner_network_header; 1062 __u16 inner_mac_header; 1063 1064 __be16 protocol; 1065 __u16 transport_header; 1066 __u16 network_header; 1067 __u16 mac_header; 1068 1069#ifdef CONFIG_KCOV 1070 u64 kcov_handle; 1071#endif 1072 1073 ); /* end headers group */ 1074 1075 /* These elements must be at the end, see alloc_skb() for details. */ 1076 sk_buff_data_t tail; 1077 sk_buff_data_t end; 1078 unsigned char *head, 1079 *data; 1080 unsigned int truesize; 1081 refcount_t users; 1082 1083#ifdef CONFIG_SKB_EXTENSIONS 1084 /* only useable after checking ->active_extensions != 0 */ 1085 struct skb_ext *extensions; 1086#endif 1087}; 1088 1089/* if you move pkt_type around you also must adapt those constants */ 1090#ifdef __BIG_ENDIAN_BITFIELD 1091#define PKT_TYPE_MAX (7 << 5) 1092#else 1093#define PKT_TYPE_MAX 7 1094#endif 1095#define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1096 1097/* if you move pkt_vlan_present, tc_at_ingress, or mono_delivery_time 1098 * around, you also must adapt these constants. 1099 */ 1100#ifdef __BIG_ENDIAN_BITFIELD 1101#define PKT_VLAN_PRESENT_BIT 7 1102#define TC_AT_INGRESS_MASK (1 << 0) 1103#define SKB_MONO_DELIVERY_TIME_MASK (1 << 2) 1104#else 1105#define PKT_VLAN_PRESENT_BIT 0 1106#define TC_AT_INGRESS_MASK (1 << 7) 1107#define SKB_MONO_DELIVERY_TIME_MASK (1 << 5) 1108#endif 1109#define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset) 1110 1111#ifdef __KERNEL__ 1112/* 1113 * Handling routines are only of interest to the kernel 1114 */ 1115 1116#define SKB_ALLOC_FCLONE 0x01 1117#define SKB_ALLOC_RX 0x02 1118#define SKB_ALLOC_NAPI 0x04 1119 1120/** 1121 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1122 * @skb: buffer 1123 */ 1124static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1125{ 1126 return unlikely(skb->pfmemalloc); 1127} 1128 1129/* 1130 * skb might have a dst pointer attached, refcounted or not. 1131 * _skb_refdst low order bit is set if refcount was _not_ taken 1132 */ 1133#define SKB_DST_NOREF 1UL 1134#define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1135 1136/** 1137 * skb_dst - returns skb dst_entry 1138 * @skb: buffer 1139 * 1140 * Returns skb dst_entry, regardless of reference taken or not. 1141 */ 1142static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1143{ 1144 /* If refdst was not refcounted, check we still are in a 1145 * rcu_read_lock section 1146 */ 1147 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1148 !rcu_read_lock_held() && 1149 !rcu_read_lock_bh_held()); 1150 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1151} 1152 1153/** 1154 * skb_dst_set - sets skb dst 1155 * @skb: buffer 1156 * @dst: dst entry 1157 * 1158 * Sets skb dst, assuming a reference was taken on dst and should 1159 * be released by skb_dst_drop() 1160 */ 1161static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1162{ 1163 skb->slow_gro |= !!dst; 1164 skb->_skb_refdst = (unsigned long)dst; 1165} 1166 1167/** 1168 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1169 * @skb: buffer 1170 * @dst: dst entry 1171 * 1172 * Sets skb dst, assuming a reference was not taken on dst. 1173 * If dst entry is cached, we do not take reference and dst_release 1174 * will be avoided by refdst_drop. If dst entry is not cached, we take 1175 * reference, so that last dst_release can destroy the dst immediately. 1176 */ 1177static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1178{ 1179 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1180 skb->slow_gro |= !!dst; 1181 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1182} 1183 1184/** 1185 * skb_dst_is_noref - Test if skb dst isn't refcounted 1186 * @skb: buffer 1187 */ 1188static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1189{ 1190 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1191} 1192 1193/** 1194 * skb_rtable - Returns the skb &rtable 1195 * @skb: buffer 1196 */ 1197static inline struct rtable *skb_rtable(const struct sk_buff *skb) 1198{ 1199 return (struct rtable *)skb_dst(skb); 1200} 1201 1202/* For mangling skb->pkt_type from user space side from applications 1203 * such as nft, tc, etc, we only allow a conservative subset of 1204 * possible pkt_types to be set. 1205*/ 1206static inline bool skb_pkt_type_ok(u32 ptype) 1207{ 1208 return ptype <= PACKET_OTHERHOST; 1209} 1210 1211/** 1212 * skb_napi_id - Returns the skb's NAPI id 1213 * @skb: buffer 1214 */ 1215static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1216{ 1217#ifdef CONFIG_NET_RX_BUSY_POLL 1218 return skb->napi_id; 1219#else 1220 return 0; 1221#endif 1222} 1223 1224/** 1225 * skb_unref - decrement the skb's reference count 1226 * @skb: buffer 1227 * 1228 * Returns true if we can free the skb. 1229 */ 1230static inline bool skb_unref(struct sk_buff *skb) 1231{ 1232 if (unlikely(!skb)) 1233 return false; 1234 if (likely(refcount_read(&skb->users) == 1)) 1235 smp_rmb(); 1236 else if (likely(!refcount_dec_and_test(&skb->users))) 1237 return false; 1238 1239 return true; 1240} 1241 1242void kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason); 1243 1244/** 1245 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1246 * @skb: buffer to free 1247 */ 1248static inline void kfree_skb(struct sk_buff *skb) 1249{ 1250 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1251} 1252 1253void skb_release_head_state(struct sk_buff *skb); 1254void kfree_skb_list_reason(struct sk_buff *segs, 1255 enum skb_drop_reason reason); 1256void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1257void skb_tx_error(struct sk_buff *skb); 1258 1259static inline void kfree_skb_list(struct sk_buff *segs) 1260{ 1261 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1262} 1263 1264#ifdef CONFIG_TRACEPOINTS 1265void consume_skb(struct sk_buff *skb); 1266#else 1267static inline void consume_skb(struct sk_buff *skb) 1268{ 1269 return kfree_skb(skb); 1270} 1271#endif 1272 1273void __consume_stateless_skb(struct sk_buff *skb); 1274void __kfree_skb(struct sk_buff *skb); 1275extern struct kmem_cache *skbuff_head_cache; 1276 1277void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1278bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1279 bool *fragstolen, int *delta_truesize); 1280 1281struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1282 int node); 1283struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1284struct sk_buff *build_skb(void *data, unsigned int frag_size); 1285struct sk_buff *build_skb_around(struct sk_buff *skb, 1286 void *data, unsigned int frag_size); 1287 1288struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1289 1290/** 1291 * alloc_skb - allocate a network buffer 1292 * @size: size to allocate 1293 * @priority: allocation mask 1294 * 1295 * This function is a convenient wrapper around __alloc_skb(). 1296 */ 1297static inline struct sk_buff *alloc_skb(unsigned int size, 1298 gfp_t priority) 1299{ 1300 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1301} 1302 1303struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1304 unsigned long data_len, 1305 int max_page_order, 1306 int *errcode, 1307 gfp_t gfp_mask); 1308struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1309 1310/* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1311struct sk_buff_fclones { 1312 struct sk_buff skb1; 1313 1314 struct sk_buff skb2; 1315 1316 refcount_t fclone_ref; 1317}; 1318 1319/** 1320 * skb_fclone_busy - check if fclone is busy 1321 * @sk: socket 1322 * @skb: buffer 1323 * 1324 * Returns true if skb is a fast clone, and its clone is not freed. 1325 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1326 * so we also check that this didnt happen. 1327 */ 1328static inline bool skb_fclone_busy(const struct sock *sk, 1329 const struct sk_buff *skb) 1330{ 1331 const struct sk_buff_fclones *fclones; 1332 1333 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1334 1335 return skb->fclone == SKB_FCLONE_ORIG && 1336 refcount_read(&fclones->fclone_ref) > 1 && 1337 READ_ONCE(fclones->skb2.sk) == sk; 1338} 1339 1340/** 1341 * alloc_skb_fclone - allocate a network buffer from fclone cache 1342 * @size: size to allocate 1343 * @priority: allocation mask 1344 * 1345 * This function is a convenient wrapper around __alloc_skb(). 1346 */ 1347static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1348 gfp_t priority) 1349{ 1350 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1351} 1352 1353struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1354void skb_headers_offset_update(struct sk_buff *skb, int off); 1355int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1356struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1357void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1358struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1359struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1360 gfp_t gfp_mask, bool fclone); 1361static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1362 gfp_t gfp_mask) 1363{ 1364 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1365} 1366 1367int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1368struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1369 unsigned int headroom); 1370struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1371struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1372 int newtailroom, gfp_t priority); 1373int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1374 int offset, int len); 1375int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1376 int offset, int len); 1377int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1378int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1379 1380/** 1381 * skb_pad - zero pad the tail of an skb 1382 * @skb: buffer to pad 1383 * @pad: space to pad 1384 * 1385 * Ensure that a buffer is followed by a padding area that is zero 1386 * filled. Used by network drivers which may DMA or transfer data 1387 * beyond the buffer end onto the wire. 1388 * 1389 * May return error in out of memory cases. The skb is freed on error. 1390 */ 1391static inline int skb_pad(struct sk_buff *skb, int pad) 1392{ 1393 return __skb_pad(skb, pad, true); 1394} 1395#define dev_kfree_skb(a) consume_skb(a) 1396 1397int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1398 int offset, size_t size); 1399 1400struct skb_seq_state { 1401 __u32 lower_offset; 1402 __u32 upper_offset; 1403 __u32 frag_idx; 1404 __u32 stepped_offset; 1405 struct sk_buff *root_skb; 1406 struct sk_buff *cur_skb; 1407 __u8 *frag_data; 1408 __u32 frag_off; 1409}; 1410 1411void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1412 unsigned int to, struct skb_seq_state *st); 1413unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1414 struct skb_seq_state *st); 1415void skb_abort_seq_read(struct skb_seq_state *st); 1416 1417unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1418 unsigned int to, struct ts_config *config); 1419 1420/* 1421 * Packet hash types specify the type of hash in skb_set_hash. 1422 * 1423 * Hash types refer to the protocol layer addresses which are used to 1424 * construct a packet's hash. The hashes are used to differentiate or identify 1425 * flows of the protocol layer for the hash type. Hash types are either 1426 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1427 * 1428 * Properties of hashes: 1429 * 1430 * 1) Two packets in different flows have different hash values 1431 * 2) Two packets in the same flow should have the same hash value 1432 * 1433 * A hash at a higher layer is considered to be more specific. A driver should 1434 * set the most specific hash possible. 1435 * 1436 * A driver cannot indicate a more specific hash than the layer at which a hash 1437 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1438 * 1439 * A driver may indicate a hash level which is less specific than the 1440 * actual layer the hash was computed on. For instance, a hash computed 1441 * at L4 may be considered an L3 hash. This should only be done if the 1442 * driver can't unambiguously determine that the HW computed the hash at 1443 * the higher layer. Note that the "should" in the second property above 1444 * permits this. 1445 */ 1446enum pkt_hash_types { 1447 PKT_HASH_TYPE_NONE, /* Undefined type */ 1448 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1449 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1450 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1451}; 1452 1453static inline void skb_clear_hash(struct sk_buff *skb) 1454{ 1455 skb->hash = 0; 1456 skb->sw_hash = 0; 1457 skb->l4_hash = 0; 1458} 1459 1460static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1461{ 1462 if (!skb->l4_hash) 1463 skb_clear_hash(skb); 1464} 1465 1466static inline void 1467__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1468{ 1469 skb->l4_hash = is_l4; 1470 skb->sw_hash = is_sw; 1471 skb->hash = hash; 1472} 1473 1474static inline void 1475skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1476{ 1477 /* Used by drivers to set hash from HW */ 1478 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1479} 1480 1481static inline void 1482__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1483{ 1484 __skb_set_hash(skb, hash, true, is_l4); 1485} 1486 1487void __skb_get_hash(struct sk_buff *skb); 1488u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1489u32 skb_get_poff(const struct sk_buff *skb); 1490u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1491 const struct flow_keys_basic *keys, int hlen); 1492__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1493 const void *data, int hlen_proto); 1494 1495static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1496 int thoff, u8 ip_proto) 1497{ 1498 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1499} 1500 1501void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1502 const struct flow_dissector_key *key, 1503 unsigned int key_count); 1504 1505struct bpf_flow_dissector; 1506bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1507 __be16 proto, int nhoff, int hlen, unsigned int flags); 1508 1509bool __skb_flow_dissect(const struct net *net, 1510 const struct sk_buff *skb, 1511 struct flow_dissector *flow_dissector, 1512 void *target_container, const void *data, 1513 __be16 proto, int nhoff, int hlen, unsigned int flags); 1514 1515static inline bool skb_flow_dissect(const struct sk_buff *skb, 1516 struct flow_dissector *flow_dissector, 1517 void *target_container, unsigned int flags) 1518{ 1519 return __skb_flow_dissect(NULL, skb, flow_dissector, 1520 target_container, NULL, 0, 0, 0, flags); 1521} 1522 1523static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1524 struct flow_keys *flow, 1525 unsigned int flags) 1526{ 1527 memset(flow, 0, sizeof(*flow)); 1528 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1529 flow, NULL, 0, 0, 0, flags); 1530} 1531 1532static inline bool 1533skb_flow_dissect_flow_keys_basic(const struct net *net, 1534 const struct sk_buff *skb, 1535 struct flow_keys_basic *flow, 1536 const void *data, __be16 proto, 1537 int nhoff, int hlen, unsigned int flags) 1538{ 1539 memset(flow, 0, sizeof(*flow)); 1540 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1541 data, proto, nhoff, hlen, flags); 1542} 1543 1544void skb_flow_dissect_meta(const struct sk_buff *skb, 1545 struct flow_dissector *flow_dissector, 1546 void *target_container); 1547 1548/* Gets a skb connection tracking info, ctinfo map should be a 1549 * map of mapsize to translate enum ip_conntrack_info states 1550 * to user states. 1551 */ 1552void 1553skb_flow_dissect_ct(const struct sk_buff *skb, 1554 struct flow_dissector *flow_dissector, 1555 void *target_container, 1556 u16 *ctinfo_map, size_t mapsize, 1557 bool post_ct, u16 zone); 1558void 1559skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1560 struct flow_dissector *flow_dissector, 1561 void *target_container); 1562 1563void skb_flow_dissect_hash(const struct sk_buff *skb, 1564 struct flow_dissector *flow_dissector, 1565 void *target_container); 1566 1567static inline __u32 skb_get_hash(struct sk_buff *skb) 1568{ 1569 if (!skb->l4_hash && !skb->sw_hash) 1570 __skb_get_hash(skb); 1571 1572 return skb->hash; 1573} 1574 1575static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1576{ 1577 if (!skb->l4_hash && !skb->sw_hash) { 1578 struct flow_keys keys; 1579 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1580 1581 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1582 } 1583 1584 return skb->hash; 1585} 1586 1587__u32 skb_get_hash_perturb(const struct sk_buff *skb, 1588 const siphash_key_t *perturb); 1589 1590static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1591{ 1592 return skb->hash; 1593} 1594 1595static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1596{ 1597 to->hash = from->hash; 1598 to->sw_hash = from->sw_hash; 1599 to->l4_hash = from->l4_hash; 1600}; 1601 1602static inline void skb_copy_decrypted(struct sk_buff *to, 1603 const struct sk_buff *from) 1604{ 1605#ifdef CONFIG_TLS_DEVICE 1606 to->decrypted = from->decrypted; 1607#endif 1608} 1609 1610#ifdef NET_SKBUFF_DATA_USES_OFFSET 1611static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1612{ 1613 return skb->head + skb->end; 1614} 1615 1616static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1617{ 1618 return skb->end; 1619} 1620 1621static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1622{ 1623 skb->end = offset; 1624} 1625#else 1626static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1627{ 1628 return skb->end; 1629} 1630 1631static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1632{ 1633 return skb->end - skb->head; 1634} 1635 1636static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1637{ 1638 skb->end = skb->head + offset; 1639} 1640#endif 1641 1642/* Internal */ 1643#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1644 1645static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1646{ 1647 return &skb_shinfo(skb)->hwtstamps; 1648} 1649 1650static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1651{ 1652 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1653 1654 return is_zcopy ? skb_uarg(skb) : NULL; 1655} 1656 1657static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1658{ 1659 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1660} 1661 1662static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1663 const struct sk_buff *skb2) 1664{ 1665 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1666} 1667 1668static inline void net_zcopy_get(struct ubuf_info *uarg) 1669{ 1670 refcount_inc(&uarg->refcnt); 1671} 1672 1673static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1674{ 1675 skb_shinfo(skb)->destructor_arg = uarg; 1676 skb_shinfo(skb)->flags |= uarg->flags; 1677} 1678 1679static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1680 bool *have_ref) 1681{ 1682 if (skb && uarg && !skb_zcopy(skb)) { 1683 if (unlikely(have_ref && *have_ref)) 1684 *have_ref = false; 1685 else 1686 net_zcopy_get(uarg); 1687 skb_zcopy_init(skb, uarg); 1688 } 1689} 1690 1691static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1692{ 1693 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1694 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1695} 1696 1697static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1698{ 1699 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1700} 1701 1702static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1703{ 1704 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1705} 1706 1707static inline void net_zcopy_put(struct ubuf_info *uarg) 1708{ 1709 if (uarg) 1710 uarg->callback(NULL, uarg, true); 1711} 1712 1713static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1714{ 1715 if (uarg) { 1716 if (uarg->callback == msg_zerocopy_callback) 1717 msg_zerocopy_put_abort(uarg, have_uref); 1718 else if (have_uref) 1719 net_zcopy_put(uarg); 1720 } 1721} 1722 1723/* Release a reference on a zerocopy structure */ 1724static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1725{ 1726 struct ubuf_info *uarg = skb_zcopy(skb); 1727 1728 if (uarg) { 1729 if (!skb_zcopy_is_nouarg(skb)) 1730 uarg->callback(skb, uarg, zerocopy_success); 1731 1732 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1733 } 1734} 1735 1736static inline void skb_mark_not_on_list(struct sk_buff *skb) 1737{ 1738 skb->next = NULL; 1739} 1740 1741/* Iterate through singly-linked GSO fragments of an skb. */ 1742#define skb_list_walk_safe(first, skb, next_skb) \ 1743 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1744 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1745 1746static inline void skb_list_del_init(struct sk_buff *skb) 1747{ 1748 __list_del_entry(&skb->list); 1749 skb_mark_not_on_list(skb); 1750} 1751 1752/** 1753 * skb_queue_empty - check if a queue is empty 1754 * @list: queue head 1755 * 1756 * Returns true if the queue is empty, false otherwise. 1757 */ 1758static inline int skb_queue_empty(const struct sk_buff_head *list) 1759{ 1760 return list->next == (const struct sk_buff *) list; 1761} 1762 1763/** 1764 * skb_queue_empty_lockless - check if a queue is empty 1765 * @list: queue head 1766 * 1767 * Returns true if the queue is empty, false otherwise. 1768 * This variant can be used in lockless contexts. 1769 */ 1770static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1771{ 1772 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1773} 1774 1775 1776/** 1777 * skb_queue_is_last - check if skb is the last entry in the queue 1778 * @list: queue head 1779 * @skb: buffer 1780 * 1781 * Returns true if @skb is the last buffer on the list. 1782 */ 1783static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1784 const struct sk_buff *skb) 1785{ 1786 return skb->next == (const struct sk_buff *) list; 1787} 1788 1789/** 1790 * skb_queue_is_first - check if skb is the first entry in the queue 1791 * @list: queue head 1792 * @skb: buffer 1793 * 1794 * Returns true if @skb is the first buffer on the list. 1795 */ 1796static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1797 const struct sk_buff *skb) 1798{ 1799 return skb->prev == (const struct sk_buff *) list; 1800} 1801 1802/** 1803 * skb_queue_next - return the next packet in the queue 1804 * @list: queue head 1805 * @skb: current buffer 1806 * 1807 * Return the next packet in @list after @skb. It is only valid to 1808 * call this if skb_queue_is_last() evaluates to false. 1809 */ 1810static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1811 const struct sk_buff *skb) 1812{ 1813 /* This BUG_ON may seem severe, but if we just return then we 1814 * are going to dereference garbage. 1815 */ 1816 BUG_ON(skb_queue_is_last(list, skb)); 1817 return skb->next; 1818} 1819 1820/** 1821 * skb_queue_prev - return the prev packet in the queue 1822 * @list: queue head 1823 * @skb: current buffer 1824 * 1825 * Return the prev packet in @list before @skb. It is only valid to 1826 * call this if skb_queue_is_first() evaluates to false. 1827 */ 1828static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1829 const struct sk_buff *skb) 1830{ 1831 /* This BUG_ON may seem severe, but if we just return then we 1832 * are going to dereference garbage. 1833 */ 1834 BUG_ON(skb_queue_is_first(list, skb)); 1835 return skb->prev; 1836} 1837 1838/** 1839 * skb_get - reference buffer 1840 * @skb: buffer to reference 1841 * 1842 * Makes another reference to a socket buffer and returns a pointer 1843 * to the buffer. 1844 */ 1845static inline struct sk_buff *skb_get(struct sk_buff *skb) 1846{ 1847 refcount_inc(&skb->users); 1848 return skb; 1849} 1850 1851/* 1852 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1853 */ 1854 1855/** 1856 * skb_cloned - is the buffer a clone 1857 * @skb: buffer to check 1858 * 1859 * Returns true if the buffer was generated with skb_clone() and is 1860 * one of multiple shared copies of the buffer. Cloned buffers are 1861 * shared data so must not be written to under normal circumstances. 1862 */ 1863static inline int skb_cloned(const struct sk_buff *skb) 1864{ 1865 return skb->cloned && 1866 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1867} 1868 1869static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1870{ 1871 might_sleep_if(gfpflags_allow_blocking(pri)); 1872 1873 if (skb_cloned(skb)) 1874 return pskb_expand_head(skb, 0, 0, pri); 1875 1876 return 0; 1877} 1878 1879/* This variant of skb_unclone() makes sure skb->truesize 1880 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 1881 * 1882 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 1883 * when various debugging features are in place. 1884 */ 1885int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 1886static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 1887{ 1888 might_sleep_if(gfpflags_allow_blocking(pri)); 1889 1890 if (skb_cloned(skb)) 1891 return __skb_unclone_keeptruesize(skb, pri); 1892 return 0; 1893} 1894 1895/** 1896 * skb_header_cloned - is the header a clone 1897 * @skb: buffer to check 1898 * 1899 * Returns true if modifying the header part of the buffer requires 1900 * the data to be copied. 1901 */ 1902static inline int skb_header_cloned(const struct sk_buff *skb) 1903{ 1904 int dataref; 1905 1906 if (!skb->cloned) 1907 return 0; 1908 1909 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1910 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1911 return dataref != 1; 1912} 1913 1914static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1915{ 1916 might_sleep_if(gfpflags_allow_blocking(pri)); 1917 1918 if (skb_header_cloned(skb)) 1919 return pskb_expand_head(skb, 0, 0, pri); 1920 1921 return 0; 1922} 1923 1924/** 1925 * __skb_header_release - release reference to header 1926 * @skb: buffer to operate on 1927 */ 1928static inline void __skb_header_release(struct sk_buff *skb) 1929{ 1930 skb->nohdr = 1; 1931 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1932} 1933 1934 1935/** 1936 * skb_shared - is the buffer shared 1937 * @skb: buffer to check 1938 * 1939 * Returns true if more than one person has a reference to this 1940 * buffer. 1941 */ 1942static inline int skb_shared(const struct sk_buff *skb) 1943{ 1944 return refcount_read(&skb->users) != 1; 1945} 1946 1947/** 1948 * skb_share_check - check if buffer is shared and if so clone it 1949 * @skb: buffer to check 1950 * @pri: priority for memory allocation 1951 * 1952 * If the buffer is shared the buffer is cloned and the old copy 1953 * drops a reference. A new clone with a single reference is returned. 1954 * If the buffer is not shared the original buffer is returned. When 1955 * being called from interrupt status or with spinlocks held pri must 1956 * be GFP_ATOMIC. 1957 * 1958 * NULL is returned on a memory allocation failure. 1959 */ 1960static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 1961{ 1962 might_sleep_if(gfpflags_allow_blocking(pri)); 1963 if (skb_shared(skb)) { 1964 struct sk_buff *nskb = skb_clone(skb, pri); 1965 1966 if (likely(nskb)) 1967 consume_skb(skb); 1968 else 1969 kfree_skb(skb); 1970 skb = nskb; 1971 } 1972 return skb; 1973} 1974 1975/* 1976 * Copy shared buffers into a new sk_buff. We effectively do COW on 1977 * packets to handle cases where we have a local reader and forward 1978 * and a couple of other messy ones. The normal one is tcpdumping 1979 * a packet thats being forwarded. 1980 */ 1981 1982/** 1983 * skb_unshare - make a copy of a shared buffer 1984 * @skb: buffer to check 1985 * @pri: priority for memory allocation 1986 * 1987 * If the socket buffer is a clone then this function creates a new 1988 * copy of the data, drops a reference count on the old copy and returns 1989 * the new copy with the reference count at 1. If the buffer is not a clone 1990 * the original buffer is returned. When called with a spinlock held or 1991 * from interrupt state @pri must be %GFP_ATOMIC 1992 * 1993 * %NULL is returned on a memory allocation failure. 1994 */ 1995static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 1996 gfp_t pri) 1997{ 1998 might_sleep_if(gfpflags_allow_blocking(pri)); 1999 if (skb_cloned(skb)) { 2000 struct sk_buff *nskb = skb_copy(skb, pri); 2001 2002 /* Free our shared copy */ 2003 if (likely(nskb)) 2004 consume_skb(skb); 2005 else 2006 kfree_skb(skb); 2007 skb = nskb; 2008 } 2009 return skb; 2010} 2011 2012/** 2013 * skb_peek - peek at the head of an &sk_buff_head 2014 * @list_: list to peek at 2015 * 2016 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2017 * be careful with this one. A peek leaves the buffer on the 2018 * list and someone else may run off with it. You must hold 2019 * the appropriate locks or have a private queue to do this. 2020 * 2021 * Returns %NULL for an empty list or a pointer to the head element. 2022 * The reference count is not incremented and the reference is therefore 2023 * volatile. Use with caution. 2024 */ 2025static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2026{ 2027 struct sk_buff *skb = list_->next; 2028 2029 if (skb == (struct sk_buff *)list_) 2030 skb = NULL; 2031 return skb; 2032} 2033 2034/** 2035 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2036 * @list_: list to peek at 2037 * 2038 * Like skb_peek(), but the caller knows that the list is not empty. 2039 */ 2040static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2041{ 2042 return list_->next; 2043} 2044 2045/** 2046 * skb_peek_next - peek skb following the given one from a queue 2047 * @skb: skb to start from 2048 * @list_: list to peek at 2049 * 2050 * Returns %NULL when the end of the list is met or a pointer to the 2051 * next element. The reference count is not incremented and the 2052 * reference is therefore volatile. Use with caution. 2053 */ 2054static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2055 const struct sk_buff_head *list_) 2056{ 2057 struct sk_buff *next = skb->next; 2058 2059 if (next == (struct sk_buff *)list_) 2060 next = NULL; 2061 return next; 2062} 2063 2064/** 2065 * skb_peek_tail - peek at the tail of an &sk_buff_head 2066 * @list_: list to peek at 2067 * 2068 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2069 * be careful with this one. A peek leaves the buffer on the 2070 * list and someone else may run off with it. You must hold 2071 * the appropriate locks or have a private queue to do this. 2072 * 2073 * Returns %NULL for an empty list or a pointer to the tail element. 2074 * The reference count is not incremented and the reference is therefore 2075 * volatile. Use with caution. 2076 */ 2077static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2078{ 2079 struct sk_buff *skb = READ_ONCE(list_->prev); 2080 2081 if (skb == (struct sk_buff *)list_) 2082 skb = NULL; 2083 return skb; 2084 2085} 2086 2087/** 2088 * skb_queue_len - get queue length 2089 * @list_: list to measure 2090 * 2091 * Return the length of an &sk_buff queue. 2092 */ 2093static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2094{ 2095 return list_->qlen; 2096} 2097 2098/** 2099 * skb_queue_len_lockless - get queue length 2100 * @list_: list to measure 2101 * 2102 * Return the length of an &sk_buff queue. 2103 * This variant can be used in lockless contexts. 2104 */ 2105static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2106{ 2107 return READ_ONCE(list_->qlen); 2108} 2109 2110/** 2111 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2112 * @list: queue to initialize 2113 * 2114 * This initializes only the list and queue length aspects of 2115 * an sk_buff_head object. This allows to initialize the list 2116 * aspects of an sk_buff_head without reinitializing things like 2117 * the spinlock. It can also be used for on-stack sk_buff_head 2118 * objects where the spinlock is known to not be used. 2119 */ 2120static inline void __skb_queue_head_init(struct sk_buff_head *list) 2121{ 2122 list->prev = list->next = (struct sk_buff *)list; 2123 list->qlen = 0; 2124} 2125 2126/* 2127 * This function creates a split out lock class for each invocation; 2128 * this is needed for now since a whole lot of users of the skb-queue 2129 * infrastructure in drivers have different locking usage (in hardirq) 2130 * than the networking core (in softirq only). In the long run either the 2131 * network layer or drivers should need annotation to consolidate the 2132 * main types of usage into 3 classes. 2133 */ 2134static inline void skb_queue_head_init(struct sk_buff_head *list) 2135{ 2136 spin_lock_init(&list->lock); 2137 __skb_queue_head_init(list); 2138} 2139 2140static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2141 struct lock_class_key *class) 2142{ 2143 skb_queue_head_init(list); 2144 lockdep_set_class(&list->lock, class); 2145} 2146 2147/* 2148 * Insert an sk_buff on a list. 2149 * 2150 * The "__skb_xxxx()" functions are the non-atomic ones that 2151 * can only be called with interrupts disabled. 2152 */ 2153static inline void __skb_insert(struct sk_buff *newsk, 2154 struct sk_buff *prev, struct sk_buff *next, 2155 struct sk_buff_head *list) 2156{ 2157 /* See skb_queue_empty_lockless() and skb_peek_tail() 2158 * for the opposite READ_ONCE() 2159 */ 2160 WRITE_ONCE(newsk->next, next); 2161 WRITE_ONCE(newsk->prev, prev); 2162 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2163 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2164 WRITE_ONCE(list->qlen, list->qlen + 1); 2165} 2166 2167static inline void __skb_queue_splice(const struct sk_buff_head *list, 2168 struct sk_buff *prev, 2169 struct sk_buff *next) 2170{ 2171 struct sk_buff *first = list->next; 2172 struct sk_buff *last = list->prev; 2173 2174 WRITE_ONCE(first->prev, prev); 2175 WRITE_ONCE(prev->next, first); 2176 2177 WRITE_ONCE(last->next, next); 2178 WRITE_ONCE(next->prev, last); 2179} 2180 2181/** 2182 * skb_queue_splice - join two skb lists, this is designed for stacks 2183 * @list: the new list to add 2184 * @head: the place to add it in the first list 2185 */ 2186static inline void skb_queue_splice(const struct sk_buff_head *list, 2187 struct sk_buff_head *head) 2188{ 2189 if (!skb_queue_empty(list)) { 2190 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2191 head->qlen += list->qlen; 2192 } 2193} 2194 2195/** 2196 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2197 * @list: the new list to add 2198 * @head: the place to add it in the first list 2199 * 2200 * The list at @list is reinitialised 2201 */ 2202static inline void skb_queue_splice_init(struct sk_buff_head *list, 2203 struct sk_buff_head *head) 2204{ 2205 if (!skb_queue_empty(list)) { 2206 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2207 head->qlen += list->qlen; 2208 __skb_queue_head_init(list); 2209 } 2210} 2211 2212/** 2213 * skb_queue_splice_tail - join two skb lists, each list being a queue 2214 * @list: the new list to add 2215 * @head: the place to add it in the first list 2216 */ 2217static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2218 struct sk_buff_head *head) 2219{ 2220 if (!skb_queue_empty(list)) { 2221 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2222 head->qlen += list->qlen; 2223 } 2224} 2225 2226/** 2227 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2228 * @list: the new list to add 2229 * @head: the place to add it in the first list 2230 * 2231 * Each of the lists is a queue. 2232 * The list at @list is reinitialised 2233 */ 2234static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2235 struct sk_buff_head *head) 2236{ 2237 if (!skb_queue_empty(list)) { 2238 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2239 head->qlen += list->qlen; 2240 __skb_queue_head_init(list); 2241 } 2242} 2243 2244/** 2245 * __skb_queue_after - queue a buffer at the list head 2246 * @list: list to use 2247 * @prev: place after this buffer 2248 * @newsk: buffer to queue 2249 * 2250 * Queue a buffer int the middle of a list. This function takes no locks 2251 * and you must therefore hold required locks before calling it. 2252 * 2253 * A buffer cannot be placed on two lists at the same time. 2254 */ 2255static inline void __skb_queue_after(struct sk_buff_head *list, 2256 struct sk_buff *prev, 2257 struct sk_buff *newsk) 2258{ 2259 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2260} 2261 2262void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2263 struct sk_buff_head *list); 2264 2265static inline void __skb_queue_before(struct sk_buff_head *list, 2266 struct sk_buff *next, 2267 struct sk_buff *newsk) 2268{ 2269 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2270} 2271 2272/** 2273 * __skb_queue_head - queue a buffer at the list head 2274 * @list: list to use 2275 * @newsk: buffer to queue 2276 * 2277 * Queue a buffer at the start of a list. This function takes no locks 2278 * and you must therefore hold required locks before calling it. 2279 * 2280 * A buffer cannot be placed on two lists at the same time. 2281 */ 2282static inline void __skb_queue_head(struct sk_buff_head *list, 2283 struct sk_buff *newsk) 2284{ 2285 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2286} 2287void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2288 2289/** 2290 * __skb_queue_tail - queue a buffer at the list tail 2291 * @list: list to use 2292 * @newsk: buffer to queue 2293 * 2294 * Queue a buffer at the end of a list. This function takes no locks 2295 * and you must therefore hold required locks before calling it. 2296 * 2297 * A buffer cannot be placed on two lists at the same time. 2298 */ 2299static inline void __skb_queue_tail(struct sk_buff_head *list, 2300 struct sk_buff *newsk) 2301{ 2302 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2303} 2304void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2305 2306/* 2307 * remove sk_buff from list. _Must_ be called atomically, and with 2308 * the list known.. 2309 */ 2310void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2311static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2312{ 2313 struct sk_buff *next, *prev; 2314 2315 WRITE_ONCE(list->qlen, list->qlen - 1); 2316 next = skb->next; 2317 prev = skb->prev; 2318 skb->next = skb->prev = NULL; 2319 WRITE_ONCE(next->prev, prev); 2320 WRITE_ONCE(prev->next, next); 2321} 2322 2323/** 2324 * __skb_dequeue - remove from the head of the queue 2325 * @list: list to dequeue from 2326 * 2327 * Remove the head of the list. This function does not take any locks 2328 * so must be used with appropriate locks held only. The head item is 2329 * returned or %NULL if the list is empty. 2330 */ 2331static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2332{ 2333 struct sk_buff *skb = skb_peek(list); 2334 if (skb) 2335 __skb_unlink(skb, list); 2336 return skb; 2337} 2338struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2339 2340/** 2341 * __skb_dequeue_tail - remove from the tail of the queue 2342 * @list: list to dequeue from 2343 * 2344 * Remove the tail of the list. This function does not take any locks 2345 * so must be used with appropriate locks held only. The tail item is 2346 * returned or %NULL if the list is empty. 2347 */ 2348static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2349{ 2350 struct sk_buff *skb = skb_peek_tail(list); 2351 if (skb) 2352 __skb_unlink(skb, list); 2353 return skb; 2354} 2355struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2356 2357 2358static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2359{ 2360 return skb->data_len; 2361} 2362 2363static inline unsigned int skb_headlen(const struct sk_buff *skb) 2364{ 2365 return skb->len - skb->data_len; 2366} 2367 2368static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2369{ 2370 unsigned int i, len = 0; 2371 2372 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2373 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2374 return len; 2375} 2376 2377static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2378{ 2379 return skb_headlen(skb) + __skb_pagelen(skb); 2380} 2381 2382/** 2383 * __skb_fill_page_desc - initialise a paged fragment in an skb 2384 * @skb: buffer containing fragment to be initialised 2385 * @i: paged fragment index to initialise 2386 * @page: the page to use for this fragment 2387 * @off: the offset to the data with @page 2388 * @size: the length of the data 2389 * 2390 * Initialises the @i'th fragment of @skb to point to &size bytes at 2391 * offset @off within @page. 2392 * 2393 * Does not take any additional reference on the fragment. 2394 */ 2395static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2396 struct page *page, int off, int size) 2397{ 2398 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2399 2400 /* 2401 * Propagate page pfmemalloc to the skb if we can. The problem is 2402 * that not all callers have unique ownership of the page but rely 2403 * on page_is_pfmemalloc doing the right thing(tm). 2404 */ 2405 frag->bv_page = page; 2406 frag->bv_offset = off; 2407 skb_frag_size_set(frag, size); 2408 2409 page = compound_head(page); 2410 if (page_is_pfmemalloc(page)) 2411 skb->pfmemalloc = true; 2412} 2413 2414/** 2415 * skb_fill_page_desc - initialise a paged fragment in an skb 2416 * @skb: buffer containing fragment to be initialised 2417 * @i: paged fragment index to initialise 2418 * @page: the page to use for this fragment 2419 * @off: the offset to the data with @page 2420 * @size: the length of the data 2421 * 2422 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2423 * @skb to point to @size bytes at offset @off within @page. In 2424 * addition updates @skb such that @i is the last fragment. 2425 * 2426 * Does not take any additional reference on the fragment. 2427 */ 2428static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2429 struct page *page, int off, int size) 2430{ 2431 __skb_fill_page_desc(skb, i, page, off, size); 2432 skb_shinfo(skb)->nr_frags = i + 1; 2433} 2434 2435void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 2436 int size, unsigned int truesize); 2437 2438void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2439 unsigned int truesize); 2440 2441#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2442 2443#ifdef NET_SKBUFF_DATA_USES_OFFSET 2444static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2445{ 2446 return skb->head + skb->tail; 2447} 2448 2449static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2450{ 2451 skb->tail = skb->data - skb->head; 2452} 2453 2454static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2455{ 2456 skb_reset_tail_pointer(skb); 2457 skb->tail += offset; 2458} 2459 2460#else /* NET_SKBUFF_DATA_USES_OFFSET */ 2461static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2462{ 2463 return skb->tail; 2464} 2465 2466static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2467{ 2468 skb->tail = skb->data; 2469} 2470 2471static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2472{ 2473 skb->tail = skb->data + offset; 2474} 2475 2476#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2477 2478/* 2479 * Add data to an sk_buff 2480 */ 2481void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2482void *skb_put(struct sk_buff *skb, unsigned int len); 2483static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2484{ 2485 void *tmp = skb_tail_pointer(skb); 2486 SKB_LINEAR_ASSERT(skb); 2487 skb->tail += len; 2488 skb->len += len; 2489 return tmp; 2490} 2491 2492static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2493{ 2494 void *tmp = __skb_put(skb, len); 2495 2496 memset(tmp, 0, len); 2497 return tmp; 2498} 2499 2500static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2501 unsigned int len) 2502{ 2503 void *tmp = __skb_put(skb, len); 2504 2505 memcpy(tmp, data, len); 2506 return tmp; 2507} 2508 2509static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2510{ 2511 *(u8 *)__skb_put(skb, 1) = val; 2512} 2513 2514static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2515{ 2516 void *tmp = skb_put(skb, len); 2517 2518 memset(tmp, 0, len); 2519 2520 return tmp; 2521} 2522 2523static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2524 unsigned int len) 2525{ 2526 void *tmp = skb_put(skb, len); 2527 2528 memcpy(tmp, data, len); 2529 2530 return tmp; 2531} 2532 2533static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2534{ 2535 *(u8 *)skb_put(skb, 1) = val; 2536} 2537 2538void *skb_push(struct sk_buff *skb, unsigned int len); 2539static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2540{ 2541 skb->data -= len; 2542 skb->len += len; 2543 return skb->data; 2544} 2545 2546void *skb_pull(struct sk_buff *skb, unsigned int len); 2547static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2548{ 2549 skb->len -= len; 2550 BUG_ON(skb->len < skb->data_len); 2551 return skb->data += len; 2552} 2553 2554static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2555{ 2556 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2557} 2558 2559void *skb_pull_data(struct sk_buff *skb, size_t len); 2560 2561void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2562 2563static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len) 2564{ 2565 if (len > skb_headlen(skb) && 2566 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 2567 return NULL; 2568 skb->len -= len; 2569 return skb->data += len; 2570} 2571 2572static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2573{ 2574 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 2575} 2576 2577static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2578{ 2579 if (likely(len <= skb_headlen(skb))) 2580 return true; 2581 if (unlikely(len > skb->len)) 2582 return false; 2583 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2584} 2585 2586void skb_condense(struct sk_buff *skb); 2587 2588/** 2589 * skb_headroom - bytes at buffer head 2590 * @skb: buffer to check 2591 * 2592 * Return the number of bytes of free space at the head of an &sk_buff. 2593 */ 2594static inline unsigned int skb_headroom(const struct sk_buff *skb) 2595{ 2596 return skb->data - skb->head; 2597} 2598 2599/** 2600 * skb_tailroom - bytes at buffer end 2601 * @skb: buffer to check 2602 * 2603 * Return the number of bytes of free space at the tail of an sk_buff 2604 */ 2605static inline int skb_tailroom(const struct sk_buff *skb) 2606{ 2607 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2608} 2609 2610/** 2611 * skb_availroom - bytes at buffer end 2612 * @skb: buffer to check 2613 * 2614 * Return the number of bytes of free space at the tail of an sk_buff 2615 * allocated by sk_stream_alloc() 2616 */ 2617static inline int skb_availroom(const struct sk_buff *skb) 2618{ 2619 if (skb_is_nonlinear(skb)) 2620 return 0; 2621 2622 return skb->end - skb->tail - skb->reserved_tailroom; 2623} 2624 2625/** 2626 * skb_reserve - adjust headroom 2627 * @skb: buffer to alter 2628 * @len: bytes to move 2629 * 2630 * Increase the headroom of an empty &sk_buff by reducing the tail 2631 * room. This is only allowed for an empty buffer. 2632 */ 2633static inline void skb_reserve(struct sk_buff *skb, int len) 2634{ 2635 skb->data += len; 2636 skb->tail += len; 2637} 2638 2639/** 2640 * skb_tailroom_reserve - adjust reserved_tailroom 2641 * @skb: buffer to alter 2642 * @mtu: maximum amount of headlen permitted 2643 * @needed_tailroom: minimum amount of reserved_tailroom 2644 * 2645 * Set reserved_tailroom so that headlen can be as large as possible but 2646 * not larger than mtu and tailroom cannot be smaller than 2647 * needed_tailroom. 2648 * The required headroom should already have been reserved before using 2649 * this function. 2650 */ 2651static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2652 unsigned int needed_tailroom) 2653{ 2654 SKB_LINEAR_ASSERT(skb); 2655 if (mtu < skb_tailroom(skb) - needed_tailroom) 2656 /* use at most mtu */ 2657 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2658 else 2659 /* use up to all available space */ 2660 skb->reserved_tailroom = needed_tailroom; 2661} 2662 2663#define ENCAP_TYPE_ETHER 0 2664#define ENCAP_TYPE_IPPROTO 1 2665 2666static inline void skb_set_inner_protocol(struct sk_buff *skb, 2667 __be16 protocol) 2668{ 2669 skb->inner_protocol = protocol; 2670 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2671} 2672 2673static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2674 __u8 ipproto) 2675{ 2676 skb->inner_ipproto = ipproto; 2677 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2678} 2679 2680static inline void skb_reset_inner_headers(struct sk_buff *skb) 2681{ 2682 skb->inner_mac_header = skb->mac_header; 2683 skb->inner_network_header = skb->network_header; 2684 skb->inner_transport_header = skb->transport_header; 2685} 2686 2687static inline void skb_reset_mac_len(struct sk_buff *skb) 2688{ 2689 skb->mac_len = skb->network_header - skb->mac_header; 2690} 2691 2692static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2693 *skb) 2694{ 2695 return skb->head + skb->inner_transport_header; 2696} 2697 2698static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2699{ 2700 return skb_inner_transport_header(skb) - skb->data; 2701} 2702 2703static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2704{ 2705 skb->inner_transport_header = skb->data - skb->head; 2706} 2707 2708static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2709 const int offset) 2710{ 2711 skb_reset_inner_transport_header(skb); 2712 skb->inner_transport_header += offset; 2713} 2714 2715static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2716{ 2717 return skb->head + skb->inner_network_header; 2718} 2719 2720static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2721{ 2722 skb->inner_network_header = skb->data - skb->head; 2723} 2724 2725static inline void skb_set_inner_network_header(struct sk_buff *skb, 2726 const int offset) 2727{ 2728 skb_reset_inner_network_header(skb); 2729 skb->inner_network_header += offset; 2730} 2731 2732static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2733{ 2734 return skb->head + skb->inner_mac_header; 2735} 2736 2737static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2738{ 2739 skb->inner_mac_header = skb->data - skb->head; 2740} 2741 2742static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2743 const int offset) 2744{ 2745 skb_reset_inner_mac_header(skb); 2746 skb->inner_mac_header += offset; 2747} 2748static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2749{ 2750 return skb->transport_header != (typeof(skb->transport_header))~0U; 2751} 2752 2753static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2754{ 2755 return skb->head + skb->transport_header; 2756} 2757 2758static inline void skb_reset_transport_header(struct sk_buff *skb) 2759{ 2760 skb->transport_header = skb->data - skb->head; 2761} 2762 2763static inline void skb_set_transport_header(struct sk_buff *skb, 2764 const int offset) 2765{ 2766 skb_reset_transport_header(skb); 2767 skb->transport_header += offset; 2768} 2769 2770static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2771{ 2772 return skb->head + skb->network_header; 2773} 2774 2775static inline void skb_reset_network_header(struct sk_buff *skb) 2776{ 2777 skb->network_header = skb->data - skb->head; 2778} 2779 2780static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2781{ 2782 skb_reset_network_header(skb); 2783 skb->network_header += offset; 2784} 2785 2786static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2787{ 2788 return skb->head + skb->mac_header; 2789} 2790 2791static inline int skb_mac_offset(const struct sk_buff *skb) 2792{ 2793 return skb_mac_header(skb) - skb->data; 2794} 2795 2796static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2797{ 2798 return skb->network_header - skb->mac_header; 2799} 2800 2801static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2802{ 2803 return skb->mac_header != (typeof(skb->mac_header))~0U; 2804} 2805 2806static inline void skb_unset_mac_header(struct sk_buff *skb) 2807{ 2808 skb->mac_header = (typeof(skb->mac_header))~0U; 2809} 2810 2811static inline void skb_reset_mac_header(struct sk_buff *skb) 2812{ 2813 skb->mac_header = skb->data - skb->head; 2814} 2815 2816static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2817{ 2818 skb_reset_mac_header(skb); 2819 skb->mac_header += offset; 2820} 2821 2822static inline void skb_pop_mac_header(struct sk_buff *skb) 2823{ 2824 skb->mac_header = skb->network_header; 2825} 2826 2827static inline void skb_probe_transport_header(struct sk_buff *skb) 2828{ 2829 struct flow_keys_basic keys; 2830 2831 if (skb_transport_header_was_set(skb)) 2832 return; 2833 2834 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 2835 NULL, 0, 0, 0, 0)) 2836 skb_set_transport_header(skb, keys.control.thoff); 2837} 2838 2839static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2840{ 2841 if (skb_mac_header_was_set(skb)) { 2842 const unsigned char *old_mac = skb_mac_header(skb); 2843 2844 skb_set_mac_header(skb, -skb->mac_len); 2845 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2846 } 2847} 2848 2849static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2850{ 2851 return skb->csum_start - skb_headroom(skb); 2852} 2853 2854static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2855{ 2856 return skb->head + skb->csum_start; 2857} 2858 2859static inline int skb_transport_offset(const struct sk_buff *skb) 2860{ 2861 return skb_transport_header(skb) - skb->data; 2862} 2863 2864static inline u32 skb_network_header_len(const struct sk_buff *skb) 2865{ 2866 return skb->transport_header - skb->network_header; 2867} 2868 2869static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2870{ 2871 return skb->inner_transport_header - skb->inner_network_header; 2872} 2873 2874static inline int skb_network_offset(const struct sk_buff *skb) 2875{ 2876 return skb_network_header(skb) - skb->data; 2877} 2878 2879static inline int skb_inner_network_offset(const struct sk_buff *skb) 2880{ 2881 return skb_inner_network_header(skb) - skb->data; 2882} 2883 2884static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2885{ 2886 return pskb_may_pull(skb, skb_network_offset(skb) + len); 2887} 2888 2889/* 2890 * CPUs often take a performance hit when accessing unaligned memory 2891 * locations. The actual performance hit varies, it can be small if the 2892 * hardware handles it or large if we have to take an exception and fix it 2893 * in software. 2894 * 2895 * Since an ethernet header is 14 bytes network drivers often end up with 2896 * the IP header at an unaligned offset. The IP header can be aligned by 2897 * shifting the start of the packet by 2 bytes. Drivers should do this 2898 * with: 2899 * 2900 * skb_reserve(skb, NET_IP_ALIGN); 2901 * 2902 * The downside to this alignment of the IP header is that the DMA is now 2903 * unaligned. On some architectures the cost of an unaligned DMA is high 2904 * and this cost outweighs the gains made by aligning the IP header. 2905 * 2906 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 2907 * to be overridden. 2908 */ 2909#ifndef NET_IP_ALIGN 2910#define NET_IP_ALIGN 2 2911#endif 2912 2913/* 2914 * The networking layer reserves some headroom in skb data (via 2915 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 2916 * the header has to grow. In the default case, if the header has to grow 2917 * 32 bytes or less we avoid the reallocation. 2918 * 2919 * Unfortunately this headroom changes the DMA alignment of the resulting 2920 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 2921 * on some architectures. An architecture can override this value, 2922 * perhaps setting it to a cacheline in size (since that will maintain 2923 * cacheline alignment of the DMA). It must be a power of 2. 2924 * 2925 * Various parts of the networking layer expect at least 32 bytes of 2926 * headroom, you should not reduce this. 2927 * 2928 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 2929 * to reduce average number of cache lines per packet. 2930 * get_rps_cpu() for example only access one 64 bytes aligned block : 2931 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 2932 */ 2933#ifndef NET_SKB_PAD 2934#define NET_SKB_PAD max(32, L1_CACHE_BYTES) 2935#endif 2936 2937int ___pskb_trim(struct sk_buff *skb, unsigned int len); 2938 2939static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 2940{ 2941 if (WARN_ON(skb_is_nonlinear(skb))) 2942 return; 2943 skb->len = len; 2944 skb_set_tail_pointer(skb, len); 2945} 2946 2947static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 2948{ 2949 __skb_set_length(skb, len); 2950} 2951 2952void skb_trim(struct sk_buff *skb, unsigned int len); 2953 2954static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 2955{ 2956 if (skb->data_len) 2957 return ___pskb_trim(skb, len); 2958 __skb_trim(skb, len); 2959 return 0; 2960} 2961 2962static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 2963{ 2964 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 2965} 2966 2967/** 2968 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 2969 * @skb: buffer to alter 2970 * @len: new length 2971 * 2972 * This is identical to pskb_trim except that the caller knows that 2973 * the skb is not cloned so we should never get an error due to out- 2974 * of-memory. 2975 */ 2976static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 2977{ 2978 int err = pskb_trim(skb, len); 2979 BUG_ON(err); 2980} 2981 2982static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 2983{ 2984 unsigned int diff = len - skb->len; 2985 2986 if (skb_tailroom(skb) < diff) { 2987 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 2988 GFP_ATOMIC); 2989 if (ret) 2990 return ret; 2991 } 2992 __skb_set_length(skb, len); 2993 return 0; 2994} 2995 2996/** 2997 * skb_orphan - orphan a buffer 2998 * @skb: buffer to orphan 2999 * 3000 * If a buffer currently has an owner then we call the owner's 3001 * destructor function and make the @skb unowned. The buffer continues 3002 * to exist but is no longer charged to its former owner. 3003 */ 3004static inline void skb_orphan(struct sk_buff *skb) 3005{ 3006 if (skb->destructor) { 3007 skb->destructor(skb); 3008 skb->destructor = NULL; 3009 skb->sk = NULL; 3010 } else { 3011 BUG_ON(skb->sk); 3012 } 3013} 3014 3015/** 3016 * skb_orphan_frags - orphan the frags contained in a buffer 3017 * @skb: buffer to orphan frags from 3018 * @gfp_mask: allocation mask for replacement pages 3019 * 3020 * For each frag in the SKB which needs a destructor (i.e. has an 3021 * owner) create a copy of that frag and release the original 3022 * page by calling the destructor. 3023 */ 3024static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3025{ 3026 if (likely(!skb_zcopy(skb))) 3027 return 0; 3028 if (!skb_zcopy_is_nouarg(skb) && 3029 skb_uarg(skb)->callback == msg_zerocopy_callback) 3030 return 0; 3031 return skb_copy_ubufs(skb, gfp_mask); 3032} 3033 3034/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3035static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3036{ 3037 if (likely(!skb_zcopy(skb))) 3038 return 0; 3039 return skb_copy_ubufs(skb, gfp_mask); 3040} 3041 3042/** 3043 * __skb_queue_purge - empty a list 3044 * @list: list to empty 3045 * 3046 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3047 * the list and one reference dropped. This function does not take the 3048 * list lock and the caller must hold the relevant locks to use it. 3049 */ 3050static inline void __skb_queue_purge(struct sk_buff_head *list) 3051{ 3052 struct sk_buff *skb; 3053 while ((skb = __skb_dequeue(list)) != NULL) 3054 kfree_skb(skb); 3055} 3056void skb_queue_purge(struct sk_buff_head *list); 3057 3058unsigned int skb_rbtree_purge(struct rb_root *root); 3059 3060void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3061 3062/** 3063 * netdev_alloc_frag - allocate a page fragment 3064 * @fragsz: fragment size 3065 * 3066 * Allocates a frag from a page for receive buffer. 3067 * Uses GFP_ATOMIC allocations. 3068 */ 3069static inline void *netdev_alloc_frag(unsigned int fragsz) 3070{ 3071 return __netdev_alloc_frag_align(fragsz, ~0u); 3072} 3073 3074static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3075 unsigned int align) 3076{ 3077 WARN_ON_ONCE(!is_power_of_2(align)); 3078 return __netdev_alloc_frag_align(fragsz, -align); 3079} 3080 3081struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3082 gfp_t gfp_mask); 3083 3084/** 3085 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3086 * @dev: network device to receive on 3087 * @length: length to allocate 3088 * 3089 * Allocate a new &sk_buff and assign it a usage count of one. The 3090 * buffer has unspecified headroom built in. Users should allocate 3091 * the headroom they think they need without accounting for the 3092 * built in space. The built in space is used for optimisations. 3093 * 3094 * %NULL is returned if there is no free memory. Although this function 3095 * allocates memory it can be called from an interrupt. 3096 */ 3097static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3098 unsigned int length) 3099{ 3100 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3101} 3102 3103/* legacy helper around __netdev_alloc_skb() */ 3104static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3105 gfp_t gfp_mask) 3106{ 3107 return __netdev_alloc_skb(NULL, length, gfp_mask); 3108} 3109 3110/* legacy helper around netdev_alloc_skb() */ 3111static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3112{ 3113 return netdev_alloc_skb(NULL, length); 3114} 3115 3116 3117static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3118 unsigned int length, gfp_t gfp) 3119{ 3120 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3121 3122 if (NET_IP_ALIGN && skb) 3123 skb_reserve(skb, NET_IP_ALIGN); 3124 return skb; 3125} 3126 3127static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3128 unsigned int length) 3129{ 3130 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3131} 3132 3133static inline void skb_free_frag(void *addr) 3134{ 3135 page_frag_free(addr); 3136} 3137 3138void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3139 3140static inline void *napi_alloc_frag(unsigned int fragsz) 3141{ 3142 return __napi_alloc_frag_align(fragsz, ~0u); 3143} 3144 3145static inline void *napi_alloc_frag_align(unsigned int fragsz, 3146 unsigned int align) 3147{ 3148 WARN_ON_ONCE(!is_power_of_2(align)); 3149 return __napi_alloc_frag_align(fragsz, -align); 3150} 3151 3152struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 3153 unsigned int length, gfp_t gfp_mask); 3154static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 3155 unsigned int length) 3156{ 3157 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 3158} 3159void napi_consume_skb(struct sk_buff *skb, int budget); 3160 3161void napi_skb_free_stolen_head(struct sk_buff *skb); 3162void __kfree_skb_defer(struct sk_buff *skb); 3163 3164/** 3165 * __dev_alloc_pages - allocate page for network Rx 3166 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3167 * @order: size of the allocation 3168 * 3169 * Allocate a new page. 3170 * 3171 * %NULL is returned if there is no free memory. 3172*/ 3173static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 3174 unsigned int order) 3175{ 3176 /* This piece of code contains several assumptions. 3177 * 1. This is for device Rx, therefor a cold page is preferred. 3178 * 2. The expectation is the user wants a compound page. 3179 * 3. If requesting a order 0 page it will not be compound 3180 * due to the check to see if order has a value in prep_new_page 3181 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3182 * code in gfp_to_alloc_flags that should be enforcing this. 3183 */ 3184 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3185 3186 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 3187} 3188 3189static inline struct page *dev_alloc_pages(unsigned int order) 3190{ 3191 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 3192} 3193 3194/** 3195 * __dev_alloc_page - allocate a page for network Rx 3196 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3197 * 3198 * Allocate a new page. 3199 * 3200 * %NULL is returned if there is no free memory. 3201 */ 3202static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 3203{ 3204 return __dev_alloc_pages(gfp_mask, 0); 3205} 3206 3207static inline struct page *dev_alloc_page(void) 3208{ 3209 return dev_alloc_pages(0); 3210} 3211 3212/** 3213 * dev_page_is_reusable - check whether a page can be reused for network Rx 3214 * @page: the page to test 3215 * 3216 * A page shouldn't be considered for reusing/recycling if it was allocated 3217 * under memory pressure or at a distant memory node. 3218 * 3219 * Returns false if this page should be returned to page allocator, true 3220 * otherwise. 3221 */ 3222static inline bool dev_page_is_reusable(const struct page *page) 3223{ 3224 return likely(page_to_nid(page) == numa_mem_id() && 3225 !page_is_pfmemalloc(page)); 3226} 3227 3228/** 3229 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3230 * @page: The page that was allocated from skb_alloc_page 3231 * @skb: The skb that may need pfmemalloc set 3232 */ 3233static inline void skb_propagate_pfmemalloc(const struct page *page, 3234 struct sk_buff *skb) 3235{ 3236 if (page_is_pfmemalloc(page)) 3237 skb->pfmemalloc = true; 3238} 3239 3240/** 3241 * skb_frag_off() - Returns the offset of a skb fragment 3242 * @frag: the paged fragment 3243 */ 3244static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3245{ 3246 return frag->bv_offset; 3247} 3248 3249/** 3250 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3251 * @frag: skb fragment 3252 * @delta: value to add 3253 */ 3254static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3255{ 3256 frag->bv_offset += delta; 3257} 3258 3259/** 3260 * skb_frag_off_set() - Sets the offset of a skb fragment 3261 * @frag: skb fragment 3262 * @offset: offset of fragment 3263 */ 3264static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3265{ 3266 frag->bv_offset = offset; 3267} 3268 3269/** 3270 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3271 * @fragto: skb fragment where offset is set 3272 * @fragfrom: skb fragment offset is copied from 3273 */ 3274static inline void skb_frag_off_copy(skb_frag_t *fragto, 3275 const skb_frag_t *fragfrom) 3276{ 3277 fragto->bv_offset = fragfrom->bv_offset; 3278} 3279 3280/** 3281 * skb_frag_page - retrieve the page referred to by a paged fragment 3282 * @frag: the paged fragment 3283 * 3284 * Returns the &struct page associated with @frag. 3285 */ 3286static inline struct page *skb_frag_page(const skb_frag_t *frag) 3287{ 3288 return frag->bv_page; 3289} 3290 3291/** 3292 * __skb_frag_ref - take an addition reference on a paged fragment. 3293 * @frag: the paged fragment 3294 * 3295 * Takes an additional reference on the paged fragment @frag. 3296 */ 3297static inline void __skb_frag_ref(skb_frag_t *frag) 3298{ 3299 get_page(skb_frag_page(frag)); 3300} 3301 3302/** 3303 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 3304 * @skb: the buffer 3305 * @f: the fragment offset. 3306 * 3307 * Takes an additional reference on the @f'th paged fragment of @skb. 3308 */ 3309static inline void skb_frag_ref(struct sk_buff *skb, int f) 3310{ 3311 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 3312} 3313 3314/** 3315 * __skb_frag_unref - release a reference on a paged fragment. 3316 * @frag: the paged fragment 3317 * @recycle: recycle the page if allocated via page_pool 3318 * 3319 * Releases a reference on the paged fragment @frag 3320 * or recycles the page via the page_pool API. 3321 */ 3322static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle) 3323{ 3324 struct page *page = skb_frag_page(frag); 3325 3326#ifdef CONFIG_PAGE_POOL 3327 if (recycle && page_pool_return_skb_page(page)) 3328 return; 3329#endif 3330 put_page(page); 3331} 3332 3333/** 3334 * skb_frag_unref - release a reference on a paged fragment of an skb. 3335 * @skb: the buffer 3336 * @f: the fragment offset 3337 * 3338 * Releases a reference on the @f'th paged fragment of @skb. 3339 */ 3340static inline void skb_frag_unref(struct sk_buff *skb, int f) 3341{ 3342 __skb_frag_unref(&skb_shinfo(skb)->frags[f], skb->pp_recycle); 3343} 3344 3345/** 3346 * skb_frag_address - gets the address of the data contained in a paged fragment 3347 * @frag: the paged fragment buffer 3348 * 3349 * Returns the address of the data within @frag. The page must already 3350 * be mapped. 3351 */ 3352static inline void *skb_frag_address(const skb_frag_t *frag) 3353{ 3354 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3355} 3356 3357/** 3358 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3359 * @frag: the paged fragment buffer 3360 * 3361 * Returns the address of the data within @frag. Checks that the page 3362 * is mapped and returns %NULL otherwise. 3363 */ 3364static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3365{ 3366 void *ptr = page_address(skb_frag_page(frag)); 3367 if (unlikely(!ptr)) 3368 return NULL; 3369 3370 return ptr + skb_frag_off(frag); 3371} 3372 3373/** 3374 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3375 * @fragto: skb fragment where page is set 3376 * @fragfrom: skb fragment page is copied from 3377 */ 3378static inline void skb_frag_page_copy(skb_frag_t *fragto, 3379 const skb_frag_t *fragfrom) 3380{ 3381 fragto->bv_page = fragfrom->bv_page; 3382} 3383 3384/** 3385 * __skb_frag_set_page - sets the page contained in a paged fragment 3386 * @frag: the paged fragment 3387 * @page: the page to set 3388 * 3389 * Sets the fragment @frag to contain @page. 3390 */ 3391static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 3392{ 3393 frag->bv_page = page; 3394} 3395 3396/** 3397 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 3398 * @skb: the buffer 3399 * @f: the fragment offset 3400 * @page: the page to set 3401 * 3402 * Sets the @f'th fragment of @skb to contain @page. 3403 */ 3404static inline void skb_frag_set_page(struct sk_buff *skb, int f, 3405 struct page *page) 3406{ 3407 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 3408} 3409 3410bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3411 3412/** 3413 * skb_frag_dma_map - maps a paged fragment via the DMA API 3414 * @dev: the device to map the fragment to 3415 * @frag: the paged fragment to map 3416 * @offset: the offset within the fragment (starting at the 3417 * fragment's own offset) 3418 * @size: the number of bytes to map 3419 * @dir: the direction of the mapping (``PCI_DMA_*``) 3420 * 3421 * Maps the page associated with @frag to @device. 3422 */ 3423static inline dma_addr_t skb_frag_dma_map(struct device *dev, 3424 const skb_frag_t *frag, 3425 size_t offset, size_t size, 3426 enum dma_data_direction dir) 3427{ 3428 return dma_map_page(dev, skb_frag_page(frag), 3429 skb_frag_off(frag) + offset, size, dir); 3430} 3431 3432static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3433 gfp_t gfp_mask) 3434{ 3435 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3436} 3437 3438 3439static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3440 gfp_t gfp_mask) 3441{ 3442 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3443} 3444 3445 3446/** 3447 * skb_clone_writable - is the header of a clone writable 3448 * @skb: buffer to check 3449 * @len: length up to which to write 3450 * 3451 * Returns true if modifying the header part of the cloned buffer 3452 * does not requires the data to be copied. 3453 */ 3454static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3455{ 3456 return !skb_header_cloned(skb) && 3457 skb_headroom(skb) + len <= skb->hdr_len; 3458} 3459 3460static inline int skb_try_make_writable(struct sk_buff *skb, 3461 unsigned int write_len) 3462{ 3463 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3464 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3465} 3466 3467static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3468 int cloned) 3469{ 3470 int delta = 0; 3471 3472 if (headroom > skb_headroom(skb)) 3473 delta = headroom - skb_headroom(skb); 3474 3475 if (delta || cloned) 3476 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3477 GFP_ATOMIC); 3478 return 0; 3479} 3480 3481/** 3482 * skb_cow - copy header of skb when it is required 3483 * @skb: buffer to cow 3484 * @headroom: needed headroom 3485 * 3486 * If the skb passed lacks sufficient headroom or its data part 3487 * is shared, data is reallocated. If reallocation fails, an error 3488 * is returned and original skb is not changed. 3489 * 3490 * The result is skb with writable area skb->head...skb->tail 3491 * and at least @headroom of space at head. 3492 */ 3493static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3494{ 3495 return __skb_cow(skb, headroom, skb_cloned(skb)); 3496} 3497 3498/** 3499 * skb_cow_head - skb_cow but only making the head writable 3500 * @skb: buffer to cow 3501 * @headroom: needed headroom 3502 * 3503 * This function is identical to skb_cow except that we replace the 3504 * skb_cloned check by skb_header_cloned. It should be used when 3505 * you only need to push on some header and do not need to modify 3506 * the data. 3507 */ 3508static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3509{ 3510 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3511} 3512 3513/** 3514 * skb_padto - pad an skbuff up to a minimal size 3515 * @skb: buffer to pad 3516 * @len: minimal length 3517 * 3518 * Pads up a buffer to ensure the trailing bytes exist and are 3519 * blanked. If the buffer already contains sufficient data it 3520 * is untouched. Otherwise it is extended. Returns zero on 3521 * success. The skb is freed on error. 3522 */ 3523static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3524{ 3525 unsigned int size = skb->len; 3526 if (likely(size >= len)) 3527 return 0; 3528 return skb_pad(skb, len - size); 3529} 3530 3531/** 3532 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3533 * @skb: buffer to pad 3534 * @len: minimal length 3535 * @free_on_error: free buffer on error 3536 * 3537 * Pads up a buffer to ensure the trailing bytes exist and are 3538 * blanked. If the buffer already contains sufficient data it 3539 * is untouched. Otherwise it is extended. Returns zero on 3540 * success. The skb is freed on error if @free_on_error is true. 3541 */ 3542static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3543 unsigned int len, 3544 bool free_on_error) 3545{ 3546 unsigned int size = skb->len; 3547 3548 if (unlikely(size < len)) { 3549 len -= size; 3550 if (__skb_pad(skb, len, free_on_error)) 3551 return -ENOMEM; 3552 __skb_put(skb, len); 3553 } 3554 return 0; 3555} 3556 3557/** 3558 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3559 * @skb: buffer to pad 3560 * @len: minimal length 3561 * 3562 * Pads up a buffer to ensure the trailing bytes exist and are 3563 * blanked. If the buffer already contains sufficient data it 3564 * is untouched. Otherwise it is extended. Returns zero on 3565 * success. The skb is freed on error. 3566 */ 3567static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3568{ 3569 return __skb_put_padto(skb, len, true); 3570} 3571 3572static inline int skb_add_data(struct sk_buff *skb, 3573 struct iov_iter *from, int copy) 3574{ 3575 const int off = skb->len; 3576 3577 if (skb->ip_summed == CHECKSUM_NONE) { 3578 __wsum csum = 0; 3579 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3580 &csum, from)) { 3581 skb->csum = csum_block_add(skb->csum, csum, off); 3582 return 0; 3583 } 3584 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3585 return 0; 3586 3587 __skb_trim(skb, off); 3588 return -EFAULT; 3589} 3590 3591static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3592 const struct page *page, int off) 3593{ 3594 if (skb_zcopy(skb)) 3595 return false; 3596 if (i) { 3597 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3598 3599 return page == skb_frag_page(frag) && 3600 off == skb_frag_off(frag) + skb_frag_size(frag); 3601 } 3602 return false; 3603} 3604 3605static inline int __skb_linearize(struct sk_buff *skb) 3606{ 3607 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3608} 3609 3610/** 3611 * skb_linearize - convert paged skb to linear one 3612 * @skb: buffer to linarize 3613 * 3614 * If there is no free memory -ENOMEM is returned, otherwise zero 3615 * is returned and the old skb data released. 3616 */ 3617static inline int skb_linearize(struct sk_buff *skb) 3618{ 3619 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3620} 3621 3622/** 3623 * skb_has_shared_frag - can any frag be overwritten 3624 * @skb: buffer to test 3625 * 3626 * Return true if the skb has at least one frag that might be modified 3627 * by an external entity (as in vmsplice()/sendfile()) 3628 */ 3629static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3630{ 3631 return skb_is_nonlinear(skb) && 3632 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3633} 3634 3635/** 3636 * skb_linearize_cow - make sure skb is linear and writable 3637 * @skb: buffer to process 3638 * 3639 * If there is no free memory -ENOMEM is returned, otherwise zero 3640 * is returned and the old skb data released. 3641 */ 3642static inline int skb_linearize_cow(struct sk_buff *skb) 3643{ 3644 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3645 __skb_linearize(skb) : 0; 3646} 3647 3648static __always_inline void 3649__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3650 unsigned int off) 3651{ 3652 if (skb->ip_summed == CHECKSUM_COMPLETE) 3653 skb->csum = csum_block_sub(skb->csum, 3654 csum_partial(start, len, 0), off); 3655 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3656 skb_checksum_start_offset(skb) < 0) 3657 skb->ip_summed = CHECKSUM_NONE; 3658} 3659 3660/** 3661 * skb_postpull_rcsum - update checksum for received skb after pull 3662 * @skb: buffer to update 3663 * @start: start of data before pull 3664 * @len: length of data pulled 3665 * 3666 * After doing a pull on a received packet, you need to call this to 3667 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3668 * CHECKSUM_NONE so that it can be recomputed from scratch. 3669 */ 3670static inline void skb_postpull_rcsum(struct sk_buff *skb, 3671 const void *start, unsigned int len) 3672{ 3673 if (skb->ip_summed == CHECKSUM_COMPLETE) 3674 skb->csum = wsum_negate(csum_partial(start, len, 3675 wsum_negate(skb->csum))); 3676 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3677 skb_checksum_start_offset(skb) < 0) 3678 skb->ip_summed = CHECKSUM_NONE; 3679} 3680 3681static __always_inline void 3682__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3683 unsigned int off) 3684{ 3685 if (skb->ip_summed == CHECKSUM_COMPLETE) 3686 skb->csum = csum_block_add(skb->csum, 3687 csum_partial(start, len, 0), off); 3688} 3689 3690/** 3691 * skb_postpush_rcsum - update checksum for received skb after push 3692 * @skb: buffer to update 3693 * @start: start of data after push 3694 * @len: length of data pushed 3695 * 3696 * After doing a push on a received packet, you need to call this to 3697 * update the CHECKSUM_COMPLETE checksum. 3698 */ 3699static inline void skb_postpush_rcsum(struct sk_buff *skb, 3700 const void *start, unsigned int len) 3701{ 3702 __skb_postpush_rcsum(skb, start, len, 0); 3703} 3704 3705void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3706 3707/** 3708 * skb_push_rcsum - push skb and update receive checksum 3709 * @skb: buffer to update 3710 * @len: length of data pulled 3711 * 3712 * This function performs an skb_push on the packet and updates 3713 * the CHECKSUM_COMPLETE checksum. It should be used on 3714 * receive path processing instead of skb_push unless you know 3715 * that the checksum difference is zero (e.g., a valid IP header) 3716 * or you are setting ip_summed to CHECKSUM_NONE. 3717 */ 3718static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3719{ 3720 skb_push(skb, len); 3721 skb_postpush_rcsum(skb, skb->data, len); 3722 return skb->data; 3723} 3724 3725int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3726/** 3727 * pskb_trim_rcsum - trim received skb and update checksum 3728 * @skb: buffer to trim 3729 * @len: new length 3730 * 3731 * This is exactly the same as pskb_trim except that it ensures the 3732 * checksum of received packets are still valid after the operation. 3733 * It can change skb pointers. 3734 */ 3735 3736static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3737{ 3738 if (likely(len >= skb->len)) 3739 return 0; 3740 return pskb_trim_rcsum_slow(skb, len); 3741} 3742 3743static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3744{ 3745 if (skb->ip_summed == CHECKSUM_COMPLETE) 3746 skb->ip_summed = CHECKSUM_NONE; 3747 __skb_trim(skb, len); 3748 return 0; 3749} 3750 3751static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3752{ 3753 if (skb->ip_summed == CHECKSUM_COMPLETE) 3754 skb->ip_summed = CHECKSUM_NONE; 3755 return __skb_grow(skb, len); 3756} 3757 3758#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3759#define skb_rb_first(root) rb_to_skb(rb_first(root)) 3760#define skb_rb_last(root) rb_to_skb(rb_last(root)) 3761#define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3762#define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3763 3764#define skb_queue_walk(queue, skb) \ 3765 for (skb = (queue)->next; \ 3766 skb != (struct sk_buff *)(queue); \ 3767 skb = skb->next) 3768 3769#define skb_queue_walk_safe(queue, skb, tmp) \ 3770 for (skb = (queue)->next, tmp = skb->next; \ 3771 skb != (struct sk_buff *)(queue); \ 3772 skb = tmp, tmp = skb->next) 3773 3774#define skb_queue_walk_from(queue, skb) \ 3775 for (; skb != (struct sk_buff *)(queue); \ 3776 skb = skb->next) 3777 3778#define skb_rbtree_walk(skb, root) \ 3779 for (skb = skb_rb_first(root); skb != NULL; \ 3780 skb = skb_rb_next(skb)) 3781 3782#define skb_rbtree_walk_from(skb) \ 3783 for (; skb != NULL; \ 3784 skb = skb_rb_next(skb)) 3785 3786#define skb_rbtree_walk_from_safe(skb, tmp) \ 3787 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3788 skb = tmp) 3789 3790#define skb_queue_walk_from_safe(queue, skb, tmp) \ 3791 for (tmp = skb->next; \ 3792 skb != (struct sk_buff *)(queue); \ 3793 skb = tmp, tmp = skb->next) 3794 3795#define skb_queue_reverse_walk(queue, skb) \ 3796 for (skb = (queue)->prev; \ 3797 skb != (struct sk_buff *)(queue); \ 3798 skb = skb->prev) 3799 3800#define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3801 for (skb = (queue)->prev, tmp = skb->prev; \ 3802 skb != (struct sk_buff *)(queue); \ 3803 skb = tmp, tmp = skb->prev) 3804 3805#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3806 for (tmp = skb->prev; \ 3807 skb != (struct sk_buff *)(queue); \ 3808 skb = tmp, tmp = skb->prev) 3809 3810static inline bool skb_has_frag_list(const struct sk_buff *skb) 3811{ 3812 return skb_shinfo(skb)->frag_list != NULL; 3813} 3814 3815static inline void skb_frag_list_init(struct sk_buff *skb) 3816{ 3817 skb_shinfo(skb)->frag_list = NULL; 3818} 3819 3820#define skb_walk_frags(skb, iter) \ 3821 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3822 3823 3824int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 3825 int *err, long *timeo_p, 3826 const struct sk_buff *skb); 3827struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3828 struct sk_buff_head *queue, 3829 unsigned int flags, 3830 int *off, int *err, 3831 struct sk_buff **last); 3832struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 3833 struct sk_buff_head *queue, 3834 unsigned int flags, int *off, int *err, 3835 struct sk_buff **last); 3836struct sk_buff *__skb_recv_datagram(struct sock *sk, 3837 struct sk_buff_head *sk_queue, 3838 unsigned int flags, int *off, int *err); 3839struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned flags, int noblock, 3840 int *err); 3841__poll_t datagram_poll(struct file *file, struct socket *sock, 3842 struct poll_table_struct *wait); 3843int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3844 struct iov_iter *to, int size); 3845static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3846 struct msghdr *msg, int size) 3847{ 3848 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3849} 3850int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3851 struct msghdr *msg); 3852int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 3853 struct iov_iter *to, int len, 3854 struct ahash_request *hash); 3855int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3856 struct iov_iter *from, int len); 3857int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3858void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3859void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3860static inline void skb_free_datagram_locked(struct sock *sk, 3861 struct sk_buff *skb) 3862{ 3863 __skb_free_datagram_locked(sk, skb, 0); 3864} 3865int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3866int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3867int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3868__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3869 int len); 3870int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3871 struct pipe_inode_info *pipe, unsigned int len, 3872 unsigned int flags); 3873int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3874 int len); 3875int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 3876void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3877unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3878int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3879 int len, int hlen); 3880void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3881int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3882void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3883bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu); 3884bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len); 3885struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3886struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 3887 unsigned int offset); 3888struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3889int skb_ensure_writable(struct sk_buff *skb, int write_len); 3890int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3891int skb_vlan_pop(struct sk_buff *skb); 3892int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3893int skb_eth_pop(struct sk_buff *skb); 3894int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 3895 const unsigned char *src); 3896int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 3897 int mac_len, bool ethernet); 3898int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 3899 bool ethernet); 3900int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 3901int skb_mpls_dec_ttl(struct sk_buff *skb); 3902struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3903 gfp_t gfp); 3904 3905static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3906{ 3907 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 3908} 3909 3910static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 3911{ 3912 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 3913} 3914 3915struct skb_checksum_ops { 3916 __wsum (*update)(const void *mem, int len, __wsum wsum); 3917 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 3918}; 3919 3920extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 3921 3922__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 3923 __wsum csum, const struct skb_checksum_ops *ops); 3924__wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 3925 __wsum csum); 3926 3927static inline void * __must_check 3928__skb_header_pointer(const struct sk_buff *skb, int offset, int len, 3929 const void *data, int hlen, void *buffer) 3930{ 3931 if (likely(hlen - offset >= len)) 3932 return (void *)data + offset; 3933 3934 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 3935 return NULL; 3936 3937 return buffer; 3938} 3939 3940static inline void * __must_check 3941skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 3942{ 3943 return __skb_header_pointer(skb, offset, len, skb->data, 3944 skb_headlen(skb), buffer); 3945} 3946 3947/** 3948 * skb_needs_linearize - check if we need to linearize a given skb 3949 * depending on the given device features. 3950 * @skb: socket buffer to check 3951 * @features: net device features 3952 * 3953 * Returns true if either: 3954 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 3955 * 2. skb is fragmented and the device does not support SG. 3956 */ 3957static inline bool skb_needs_linearize(struct sk_buff *skb, 3958 netdev_features_t features) 3959{ 3960 return skb_is_nonlinear(skb) && 3961 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 3962 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 3963} 3964 3965static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 3966 void *to, 3967 const unsigned int len) 3968{ 3969 memcpy(to, skb->data, len); 3970} 3971 3972static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 3973 const int offset, void *to, 3974 const unsigned int len) 3975{ 3976 memcpy(to, skb->data + offset, len); 3977} 3978 3979static inline void skb_copy_to_linear_data(struct sk_buff *skb, 3980 const void *from, 3981 const unsigned int len) 3982{ 3983 memcpy(skb->data, from, len); 3984} 3985 3986static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 3987 const int offset, 3988 const void *from, 3989 const unsigned int len) 3990{ 3991 memcpy(skb->data + offset, from, len); 3992} 3993 3994void skb_init(void); 3995 3996static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 3997{ 3998 return skb->tstamp; 3999} 4000 4001/** 4002 * skb_get_timestamp - get timestamp from a skb 4003 * @skb: skb to get stamp from 4004 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4005 * 4006 * Timestamps are stored in the skb as offsets to a base timestamp. 4007 * This function converts the offset back to a struct timeval and stores 4008 * it in stamp. 4009 */ 4010static inline void skb_get_timestamp(const struct sk_buff *skb, 4011 struct __kernel_old_timeval *stamp) 4012{ 4013 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4014} 4015 4016static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4017 struct __kernel_sock_timeval *stamp) 4018{ 4019 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4020 4021 stamp->tv_sec = ts.tv_sec; 4022 stamp->tv_usec = ts.tv_nsec / 1000; 4023} 4024 4025static inline void skb_get_timestampns(const struct sk_buff *skb, 4026 struct __kernel_old_timespec *stamp) 4027{ 4028 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4029 4030 stamp->tv_sec = ts.tv_sec; 4031 stamp->tv_nsec = ts.tv_nsec; 4032} 4033 4034static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4035 struct __kernel_timespec *stamp) 4036{ 4037 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4038 4039 stamp->tv_sec = ts.tv_sec; 4040 stamp->tv_nsec = ts.tv_nsec; 4041} 4042 4043static inline void __net_timestamp(struct sk_buff *skb) 4044{ 4045 skb->tstamp = ktime_get_real(); 4046 skb->mono_delivery_time = 0; 4047} 4048 4049static inline ktime_t net_timedelta(ktime_t t) 4050{ 4051 return ktime_sub(ktime_get_real(), t); 4052} 4053 4054static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4055 bool mono) 4056{ 4057 skb->tstamp = kt; 4058 skb->mono_delivery_time = kt && mono; 4059} 4060 4061DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4062 4063/* It is used in the ingress path to clear the delivery_time. 4064 * If needed, set the skb->tstamp to the (rcv) timestamp. 4065 */ 4066static inline void skb_clear_delivery_time(struct sk_buff *skb) 4067{ 4068 if (skb->mono_delivery_time) { 4069 skb->mono_delivery_time = 0; 4070 if (static_branch_unlikely(&netstamp_needed_key)) 4071 skb->tstamp = ktime_get_real(); 4072 else 4073 skb->tstamp = 0; 4074 } 4075} 4076 4077static inline void skb_clear_tstamp(struct sk_buff *skb) 4078{ 4079 if (skb->mono_delivery_time) 4080 return; 4081 4082 skb->tstamp = 0; 4083} 4084 4085static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4086{ 4087 if (skb->mono_delivery_time) 4088 return 0; 4089 4090 return skb->tstamp; 4091} 4092 4093static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4094{ 4095 if (!skb->mono_delivery_time && skb->tstamp) 4096 return skb->tstamp; 4097 4098 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4099 return ktime_get_real(); 4100 4101 return 0; 4102} 4103 4104static inline u8 skb_metadata_len(const struct sk_buff *skb) 4105{ 4106 return skb_shinfo(skb)->meta_len; 4107} 4108 4109static inline void *skb_metadata_end(const struct sk_buff *skb) 4110{ 4111 return skb_mac_header(skb); 4112} 4113 4114static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4115 const struct sk_buff *skb_b, 4116 u8 meta_len) 4117{ 4118 const void *a = skb_metadata_end(skb_a); 4119 const void *b = skb_metadata_end(skb_b); 4120 /* Using more efficient varaiant than plain call to memcmp(). */ 4121#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64 4122 u64 diffs = 0; 4123 4124 switch (meta_len) { 4125#define __it(x, op) (x -= sizeof(u##op)) 4126#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4127 case 32: diffs |= __it_diff(a, b, 64); 4128 fallthrough; 4129 case 24: diffs |= __it_diff(a, b, 64); 4130 fallthrough; 4131 case 16: diffs |= __it_diff(a, b, 64); 4132 fallthrough; 4133 case 8: diffs |= __it_diff(a, b, 64); 4134 break; 4135 case 28: diffs |= __it_diff(a, b, 64); 4136 fallthrough; 4137 case 20: diffs |= __it_diff(a, b, 64); 4138 fallthrough; 4139 case 12: diffs |= __it_diff(a, b, 64); 4140 fallthrough; 4141 case 4: diffs |= __it_diff(a, b, 32); 4142 break; 4143 } 4144 return diffs; 4145#else 4146 return memcmp(a - meta_len, b - meta_len, meta_len); 4147#endif 4148} 4149 4150static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4151 const struct sk_buff *skb_b) 4152{ 4153 u8 len_a = skb_metadata_len(skb_a); 4154 u8 len_b = skb_metadata_len(skb_b); 4155 4156 if (!(len_a | len_b)) 4157 return false; 4158 4159 return len_a != len_b ? 4160 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4161} 4162 4163static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4164{ 4165 skb_shinfo(skb)->meta_len = meta_len; 4166} 4167 4168static inline void skb_metadata_clear(struct sk_buff *skb) 4169{ 4170 skb_metadata_set(skb, 0); 4171} 4172 4173struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4174 4175#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4176 4177void skb_clone_tx_timestamp(struct sk_buff *skb); 4178bool skb_defer_rx_timestamp(struct sk_buff *skb); 4179 4180#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4181 4182static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4183{ 4184} 4185 4186static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4187{ 4188 return false; 4189} 4190 4191#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4192 4193/** 4194 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4195 * 4196 * PHY drivers may accept clones of transmitted packets for 4197 * timestamping via their phy_driver.txtstamp method. These drivers 4198 * must call this function to return the skb back to the stack with a 4199 * timestamp. 4200 * 4201 * @skb: clone of the original outgoing packet 4202 * @hwtstamps: hardware time stamps 4203 * 4204 */ 4205void skb_complete_tx_timestamp(struct sk_buff *skb, 4206 struct skb_shared_hwtstamps *hwtstamps); 4207 4208void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4209 struct skb_shared_hwtstamps *hwtstamps, 4210 struct sock *sk, int tstype); 4211 4212/** 4213 * skb_tstamp_tx - queue clone of skb with send time stamps 4214 * @orig_skb: the original outgoing packet 4215 * @hwtstamps: hardware time stamps, may be NULL if not available 4216 * 4217 * If the skb has a socket associated, then this function clones the 4218 * skb (thus sharing the actual data and optional structures), stores 4219 * the optional hardware time stamping information (if non NULL) or 4220 * generates a software time stamp (otherwise), then queues the clone 4221 * to the error queue of the socket. Errors are silently ignored. 4222 */ 4223void skb_tstamp_tx(struct sk_buff *orig_skb, 4224 struct skb_shared_hwtstamps *hwtstamps); 4225 4226/** 4227 * skb_tx_timestamp() - Driver hook for transmit timestamping 4228 * 4229 * Ethernet MAC Drivers should call this function in their hard_xmit() 4230 * function immediately before giving the sk_buff to the MAC hardware. 4231 * 4232 * Specifically, one should make absolutely sure that this function is 4233 * called before TX completion of this packet can trigger. Otherwise 4234 * the packet could potentially already be freed. 4235 * 4236 * @skb: A socket buffer. 4237 */ 4238static inline void skb_tx_timestamp(struct sk_buff *skb) 4239{ 4240 skb_clone_tx_timestamp(skb); 4241 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 4242 skb_tstamp_tx(skb, NULL); 4243} 4244 4245/** 4246 * skb_complete_wifi_ack - deliver skb with wifi status 4247 * 4248 * @skb: the original outgoing packet 4249 * @acked: ack status 4250 * 4251 */ 4252void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4253 4254__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4255__sum16 __skb_checksum_complete(struct sk_buff *skb); 4256 4257static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4258{ 4259 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4260 skb->csum_valid || 4261 (skb->ip_summed == CHECKSUM_PARTIAL && 4262 skb_checksum_start_offset(skb) >= 0)); 4263} 4264 4265/** 4266 * skb_checksum_complete - Calculate checksum of an entire packet 4267 * @skb: packet to process 4268 * 4269 * This function calculates the checksum over the entire packet plus 4270 * the value of skb->csum. The latter can be used to supply the 4271 * checksum of a pseudo header as used by TCP/UDP. It returns the 4272 * checksum. 4273 * 4274 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4275 * this function can be used to verify that checksum on received 4276 * packets. In that case the function should return zero if the 4277 * checksum is correct. In particular, this function will return zero 4278 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4279 * hardware has already verified the correctness of the checksum. 4280 */ 4281static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4282{ 4283 return skb_csum_unnecessary(skb) ? 4284 0 : __skb_checksum_complete(skb); 4285} 4286 4287static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4288{ 4289 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4290 if (skb->csum_level == 0) 4291 skb->ip_summed = CHECKSUM_NONE; 4292 else 4293 skb->csum_level--; 4294 } 4295} 4296 4297static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4298{ 4299 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4300 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4301 skb->csum_level++; 4302 } else if (skb->ip_summed == CHECKSUM_NONE) { 4303 skb->ip_summed = CHECKSUM_UNNECESSARY; 4304 skb->csum_level = 0; 4305 } 4306} 4307 4308static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4309{ 4310 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4311 skb->ip_summed = CHECKSUM_NONE; 4312 skb->csum_level = 0; 4313 } 4314} 4315 4316/* Check if we need to perform checksum complete validation. 4317 * 4318 * Returns true if checksum complete is needed, false otherwise 4319 * (either checksum is unnecessary or zero checksum is allowed). 4320 */ 4321static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4322 bool zero_okay, 4323 __sum16 check) 4324{ 4325 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4326 skb->csum_valid = 1; 4327 __skb_decr_checksum_unnecessary(skb); 4328 return false; 4329 } 4330 4331 return true; 4332} 4333 4334/* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4335 * in checksum_init. 4336 */ 4337#define CHECKSUM_BREAK 76 4338 4339/* Unset checksum-complete 4340 * 4341 * Unset checksum complete can be done when packet is being modified 4342 * (uncompressed for instance) and checksum-complete value is 4343 * invalidated. 4344 */ 4345static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4346{ 4347 if (skb->ip_summed == CHECKSUM_COMPLETE) 4348 skb->ip_summed = CHECKSUM_NONE; 4349} 4350 4351/* Validate (init) checksum based on checksum complete. 4352 * 4353 * Return values: 4354 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4355 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4356 * checksum is stored in skb->csum for use in __skb_checksum_complete 4357 * non-zero: value of invalid checksum 4358 * 4359 */ 4360static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4361 bool complete, 4362 __wsum psum) 4363{ 4364 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4365 if (!csum_fold(csum_add(psum, skb->csum))) { 4366 skb->csum_valid = 1; 4367 return 0; 4368 } 4369 } 4370 4371 skb->csum = psum; 4372 4373 if (complete || skb->len <= CHECKSUM_BREAK) { 4374 __sum16 csum; 4375 4376 csum = __skb_checksum_complete(skb); 4377 skb->csum_valid = !csum; 4378 return csum; 4379 } 4380 4381 return 0; 4382} 4383 4384static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4385{ 4386 return 0; 4387} 4388 4389/* Perform checksum validate (init). Note that this is a macro since we only 4390 * want to calculate the pseudo header which is an input function if necessary. 4391 * First we try to validate without any computation (checksum unnecessary) and 4392 * then calculate based on checksum complete calling the function to compute 4393 * pseudo header. 4394 * 4395 * Return values: 4396 * 0: checksum is validated or try to in skb_checksum_complete 4397 * non-zero: value of invalid checksum 4398 */ 4399#define __skb_checksum_validate(skb, proto, complete, \ 4400 zero_okay, check, compute_pseudo) \ 4401({ \ 4402 __sum16 __ret = 0; \ 4403 skb->csum_valid = 0; \ 4404 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4405 __ret = __skb_checksum_validate_complete(skb, \ 4406 complete, compute_pseudo(skb, proto)); \ 4407 __ret; \ 4408}) 4409 4410#define skb_checksum_init(skb, proto, compute_pseudo) \ 4411 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4412 4413#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4414 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4415 4416#define skb_checksum_validate(skb, proto, compute_pseudo) \ 4417 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4418 4419#define skb_checksum_validate_zero_check(skb, proto, check, \ 4420 compute_pseudo) \ 4421 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4422 4423#define skb_checksum_simple_validate(skb) \ 4424 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4425 4426static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4427{ 4428 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4429} 4430 4431static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4432{ 4433 skb->csum = ~pseudo; 4434 skb->ip_summed = CHECKSUM_COMPLETE; 4435} 4436 4437#define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4438do { \ 4439 if (__skb_checksum_convert_check(skb)) \ 4440 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4441} while (0) 4442 4443static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4444 u16 start, u16 offset) 4445{ 4446 skb->ip_summed = CHECKSUM_PARTIAL; 4447 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4448 skb->csum_offset = offset - start; 4449} 4450 4451/* Update skbuf and packet to reflect the remote checksum offload operation. 4452 * When called, ptr indicates the starting point for skb->csum when 4453 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4454 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4455 */ 4456static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4457 int start, int offset, bool nopartial) 4458{ 4459 __wsum delta; 4460 4461 if (!nopartial) { 4462 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4463 return; 4464 } 4465 4466 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4467 __skb_checksum_complete(skb); 4468 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4469 } 4470 4471 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4472 4473 /* Adjust skb->csum since we changed the packet */ 4474 skb->csum = csum_add(skb->csum, delta); 4475} 4476 4477static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4478{ 4479#if IS_ENABLED(CONFIG_NF_CONNTRACK) 4480 return (void *)(skb->_nfct & NFCT_PTRMASK); 4481#else 4482 return NULL; 4483#endif 4484} 4485 4486static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4487{ 4488#if IS_ENABLED(CONFIG_NF_CONNTRACK) 4489 return skb->_nfct; 4490#else 4491 return 0UL; 4492#endif 4493} 4494 4495static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4496{ 4497#if IS_ENABLED(CONFIG_NF_CONNTRACK) 4498 skb->slow_gro |= !!nfct; 4499 skb->_nfct = nfct; 4500#endif 4501} 4502 4503#ifdef CONFIG_SKB_EXTENSIONS 4504enum skb_ext_id { 4505#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4506 SKB_EXT_BRIDGE_NF, 4507#endif 4508#ifdef CONFIG_XFRM 4509 SKB_EXT_SEC_PATH, 4510#endif 4511#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4512 TC_SKB_EXT, 4513#endif 4514#if IS_ENABLED(CONFIG_MPTCP) 4515 SKB_EXT_MPTCP, 4516#endif 4517#if IS_ENABLED(CONFIG_MCTP_FLOWS) 4518 SKB_EXT_MCTP, 4519#endif 4520 SKB_EXT_NUM, /* must be last */ 4521}; 4522 4523/** 4524 * struct skb_ext - sk_buff extensions 4525 * @refcnt: 1 on allocation, deallocated on 0 4526 * @offset: offset to add to @data to obtain extension address 4527 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4528 * @data: start of extension data, variable sized 4529 * 4530 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4531 * to use 'u8' types while allowing up to 2kb worth of extension data. 4532 */ 4533struct skb_ext { 4534 refcount_t refcnt; 4535 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4536 u8 chunks; /* same */ 4537 char data[] __aligned(8); 4538}; 4539 4540struct skb_ext *__skb_ext_alloc(gfp_t flags); 4541void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4542 struct skb_ext *ext); 4543void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4544void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4545void __skb_ext_put(struct skb_ext *ext); 4546 4547static inline void skb_ext_put(struct sk_buff *skb) 4548{ 4549 if (skb->active_extensions) 4550 __skb_ext_put(skb->extensions); 4551} 4552 4553static inline void __skb_ext_copy(struct sk_buff *dst, 4554 const struct sk_buff *src) 4555{ 4556 dst->active_extensions = src->active_extensions; 4557 4558 if (src->active_extensions) { 4559 struct skb_ext *ext = src->extensions; 4560 4561 refcount_inc(&ext->refcnt); 4562 dst->extensions = ext; 4563 } 4564} 4565 4566static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4567{ 4568 skb_ext_put(dst); 4569 __skb_ext_copy(dst, src); 4570} 4571 4572static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4573{ 4574 return !!ext->offset[i]; 4575} 4576 4577static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4578{ 4579 return skb->active_extensions & (1 << id); 4580} 4581 4582static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4583{ 4584 if (skb_ext_exist(skb, id)) 4585 __skb_ext_del(skb, id); 4586} 4587 4588static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4589{ 4590 if (skb_ext_exist(skb, id)) { 4591 struct skb_ext *ext = skb->extensions; 4592 4593 return (void *)ext + (ext->offset[id] << 3); 4594 } 4595 4596 return NULL; 4597} 4598 4599static inline void skb_ext_reset(struct sk_buff *skb) 4600{ 4601 if (unlikely(skb->active_extensions)) { 4602 __skb_ext_put(skb->extensions); 4603 skb->active_extensions = 0; 4604 } 4605} 4606 4607static inline bool skb_has_extensions(struct sk_buff *skb) 4608{ 4609 return unlikely(skb->active_extensions); 4610} 4611#else 4612static inline void skb_ext_put(struct sk_buff *skb) {} 4613static inline void skb_ext_reset(struct sk_buff *skb) {} 4614static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4615static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4616static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4617static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4618#endif /* CONFIG_SKB_EXTENSIONS */ 4619 4620static inline void nf_reset_ct(struct sk_buff *skb) 4621{ 4622#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4623 nf_conntrack_put(skb_nfct(skb)); 4624 skb->_nfct = 0; 4625#endif 4626} 4627 4628static inline void nf_reset_trace(struct sk_buff *skb) 4629{ 4630#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4631 skb->nf_trace = 0; 4632#endif 4633} 4634 4635static inline void ipvs_reset(struct sk_buff *skb) 4636{ 4637#if IS_ENABLED(CONFIG_IP_VS) 4638 skb->ipvs_property = 0; 4639#endif 4640} 4641 4642/* Note: This doesn't put any conntrack info in dst. */ 4643static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4644 bool copy) 4645{ 4646#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4647 dst->_nfct = src->_nfct; 4648 nf_conntrack_get(skb_nfct(src)); 4649#endif 4650#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4651 if (copy) 4652 dst->nf_trace = src->nf_trace; 4653#endif 4654} 4655 4656static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4657{ 4658#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4659 nf_conntrack_put(skb_nfct(dst)); 4660#endif 4661 dst->slow_gro = src->slow_gro; 4662 __nf_copy(dst, src, true); 4663} 4664 4665#ifdef CONFIG_NETWORK_SECMARK 4666static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4667{ 4668 to->secmark = from->secmark; 4669} 4670 4671static inline void skb_init_secmark(struct sk_buff *skb) 4672{ 4673 skb->secmark = 0; 4674} 4675#else 4676static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4677{ } 4678 4679static inline void skb_init_secmark(struct sk_buff *skb) 4680{ } 4681#endif 4682 4683static inline int secpath_exists(const struct sk_buff *skb) 4684{ 4685#ifdef CONFIG_XFRM 4686 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4687#else 4688 return 0; 4689#endif 4690} 4691 4692static inline bool skb_irq_freeable(const struct sk_buff *skb) 4693{ 4694 return !skb->destructor && 4695 !secpath_exists(skb) && 4696 !skb_nfct(skb) && 4697 !skb->_skb_refdst && 4698 !skb_has_frag_list(skb); 4699} 4700 4701static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 4702{ 4703 skb->queue_mapping = queue_mapping; 4704} 4705 4706static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 4707{ 4708 return skb->queue_mapping; 4709} 4710 4711static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 4712{ 4713 to->queue_mapping = from->queue_mapping; 4714} 4715 4716static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 4717{ 4718 skb->queue_mapping = rx_queue + 1; 4719} 4720 4721static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 4722{ 4723 return skb->queue_mapping - 1; 4724} 4725 4726static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 4727{ 4728 return skb->queue_mapping != 0; 4729} 4730 4731static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 4732{ 4733 skb->dst_pending_confirm = val; 4734} 4735 4736static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 4737{ 4738 return skb->dst_pending_confirm != 0; 4739} 4740 4741static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 4742{ 4743#ifdef CONFIG_XFRM 4744 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 4745#else 4746 return NULL; 4747#endif 4748} 4749 4750/* Keeps track of mac header offset relative to skb->head. 4751 * It is useful for TSO of Tunneling protocol. e.g. GRE. 4752 * For non-tunnel skb it points to skb_mac_header() and for 4753 * tunnel skb it points to outer mac header. 4754 * Keeps track of level of encapsulation of network headers. 4755 */ 4756struct skb_gso_cb { 4757 union { 4758 int mac_offset; 4759 int data_offset; 4760 }; 4761 int encap_level; 4762 __wsum csum; 4763 __u16 csum_start; 4764}; 4765#define SKB_GSO_CB_OFFSET 32 4766#define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET)) 4767 4768static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 4769{ 4770 return (skb_mac_header(inner_skb) - inner_skb->head) - 4771 SKB_GSO_CB(inner_skb)->mac_offset; 4772} 4773 4774static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 4775{ 4776 int new_headroom, headroom; 4777 int ret; 4778 4779 headroom = skb_headroom(skb); 4780 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 4781 if (ret) 4782 return ret; 4783 4784 new_headroom = skb_headroom(skb); 4785 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 4786 return 0; 4787} 4788 4789static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 4790{ 4791 /* Do not update partial checksums if remote checksum is enabled. */ 4792 if (skb->remcsum_offload) 4793 return; 4794 4795 SKB_GSO_CB(skb)->csum = res; 4796 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 4797} 4798 4799/* Compute the checksum for a gso segment. First compute the checksum value 4800 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 4801 * then add in skb->csum (checksum from csum_start to end of packet). 4802 * skb->csum and csum_start are then updated to reflect the checksum of the 4803 * resultant packet starting from the transport header-- the resultant checksum 4804 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 4805 * header. 4806 */ 4807static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 4808{ 4809 unsigned char *csum_start = skb_transport_header(skb); 4810 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 4811 __wsum partial = SKB_GSO_CB(skb)->csum; 4812 4813 SKB_GSO_CB(skb)->csum = res; 4814 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 4815 4816 return csum_fold(csum_partial(csum_start, plen, partial)); 4817} 4818 4819static inline bool skb_is_gso(const struct sk_buff *skb) 4820{ 4821 return skb_shinfo(skb)->gso_size; 4822} 4823 4824/* Note: Should be called only if skb_is_gso(skb) is true */ 4825static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4826{ 4827 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4828} 4829 4830/* Note: Should be called only if skb_is_gso(skb) is true */ 4831static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4832{ 4833 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4834} 4835 4836/* Note: Should be called only if skb_is_gso(skb) is true */ 4837static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 4838{ 4839 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 4840} 4841 4842static inline void skb_gso_reset(struct sk_buff *skb) 4843{ 4844 skb_shinfo(skb)->gso_size = 0; 4845 skb_shinfo(skb)->gso_segs = 0; 4846 skb_shinfo(skb)->gso_type = 0; 4847} 4848 4849static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4850 u16 increment) 4851{ 4852 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4853 return; 4854 shinfo->gso_size += increment; 4855} 4856 4857static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4858 u16 decrement) 4859{ 4860 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4861 return; 4862 shinfo->gso_size -= decrement; 4863} 4864 4865void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4866 4867static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4868{ 4869 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4870 * wanted then gso_type will be set. */ 4871 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4872 4873 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4874 unlikely(shinfo->gso_type == 0)) { 4875 __skb_warn_lro_forwarding(skb); 4876 return true; 4877 } 4878 return false; 4879} 4880 4881static inline void skb_forward_csum(struct sk_buff *skb) 4882{ 4883 /* Unfortunately we don't support this one. Any brave souls? */ 4884 if (skb->ip_summed == CHECKSUM_COMPLETE) 4885 skb->ip_summed = CHECKSUM_NONE; 4886} 4887 4888/** 4889 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4890 * @skb: skb to check 4891 * 4892 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4893 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4894 * use this helper, to document places where we make this assertion. 4895 */ 4896static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4897{ 4898#ifdef DEBUG 4899 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 4900#endif 4901} 4902 4903bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4904 4905int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4906struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4907 unsigned int transport_len, 4908 __sum16(*skb_chkf)(struct sk_buff *skb)); 4909 4910/** 4911 * skb_head_is_locked - Determine if the skb->head is locked down 4912 * @skb: skb to check 4913 * 4914 * The head on skbs build around a head frag can be removed if they are 4915 * not cloned. This function returns true if the skb head is locked down 4916 * due to either being allocated via kmalloc, or by being a clone with 4917 * multiple references to the head. 4918 */ 4919static inline bool skb_head_is_locked(const struct sk_buff *skb) 4920{ 4921 return !skb->head_frag || skb_cloned(skb); 4922} 4923 4924/* Local Checksum Offload. 4925 * Compute outer checksum based on the assumption that the 4926 * inner checksum will be offloaded later. 4927 * See Documentation/networking/checksum-offloads.rst for 4928 * explanation of how this works. 4929 * Fill in outer checksum adjustment (e.g. with sum of outer 4930 * pseudo-header) before calling. 4931 * Also ensure that inner checksum is in linear data area. 4932 */ 4933static inline __wsum lco_csum(struct sk_buff *skb) 4934{ 4935 unsigned char *csum_start = skb_checksum_start(skb); 4936 unsigned char *l4_hdr = skb_transport_header(skb); 4937 __wsum partial; 4938 4939 /* Start with complement of inner checksum adjustment */ 4940 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 4941 skb->csum_offset)); 4942 4943 /* Add in checksum of our headers (incl. outer checksum 4944 * adjustment filled in by caller) and return result. 4945 */ 4946 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 4947} 4948 4949static inline bool skb_is_redirected(const struct sk_buff *skb) 4950{ 4951 return skb->redirected; 4952} 4953 4954static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 4955{ 4956 skb->redirected = 1; 4957#ifdef CONFIG_NET_REDIRECT 4958 skb->from_ingress = from_ingress; 4959 if (skb->from_ingress) 4960 skb_clear_tstamp(skb); 4961#endif 4962} 4963 4964static inline void skb_reset_redirect(struct sk_buff *skb) 4965{ 4966 skb->redirected = 0; 4967} 4968 4969static inline bool skb_csum_is_sctp(struct sk_buff *skb) 4970{ 4971 return skb->csum_not_inet; 4972} 4973 4974static inline void skb_set_kcov_handle(struct sk_buff *skb, 4975 const u64 kcov_handle) 4976{ 4977#ifdef CONFIG_KCOV 4978 skb->kcov_handle = kcov_handle; 4979#endif 4980} 4981 4982static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 4983{ 4984#ifdef CONFIG_KCOV 4985 return skb->kcov_handle; 4986#else 4987 return 0; 4988#endif 4989} 4990 4991#ifdef CONFIG_PAGE_POOL 4992static inline void skb_mark_for_recycle(struct sk_buff *skb) 4993{ 4994 skb->pp_recycle = 1; 4995} 4996#endif 4997 4998static inline bool skb_pp_recycle(struct sk_buff *skb, void *data) 4999{ 5000 if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle) 5001 return false; 5002 return page_pool_return_skb_page(virt_to_page(data)); 5003} 5004 5005#endif /* __KERNEL__ */ 5006#endif /* _LINUX_SKBUFF_H */