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, <[email protected]> 7 * Florian La Roche, <[email protected]> 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 #include <net/net_debug.h> 46 47 /** 48 * DOC: skb checksums 49 * 50 * The interface for checksum offload between the stack and networking drivers 51 * is as follows... 52 * 53 * IP checksum related features 54 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 55 * 56 * Drivers advertise checksum offload capabilities in the features of a device. 57 * From the stack's point of view these are capabilities offered by the driver. 58 * A driver typically only advertises features that it is capable of offloading 59 * to its device. 60 * 61 * .. flat-table:: Checksum related device features 62 * :widths: 1 10 63 * 64 * * - %NETIF_F_HW_CSUM 65 * - The driver (or its device) is able to compute one 66 * IP (one's complement) checksum for any combination 67 * of protocols or protocol layering. The checksum is 68 * computed and set in a packet per the CHECKSUM_PARTIAL 69 * interface (see below). 70 * 71 * * - %NETIF_F_IP_CSUM 72 * - Driver (device) is only able to checksum plain 73 * TCP or UDP packets over IPv4. These are specifically 74 * unencapsulated packets of the form IPv4|TCP or 75 * IPv4|UDP where the Protocol field in the IPv4 header 76 * is TCP or UDP. The IPv4 header may contain IP options. 77 * This feature cannot be set in features for a device 78 * with NETIF_F_HW_CSUM also set. This feature is being 79 * DEPRECATED (see below). 80 * 81 * * - %NETIF_F_IPV6_CSUM 82 * - Driver (device) is only able to checksum plain 83 * TCP or UDP packets over IPv6. These are specifically 84 * unencapsulated packets of the form IPv6|TCP or 85 * IPv6|UDP where the Next Header field in the IPv6 86 * header is either TCP or UDP. IPv6 extension headers 87 * are not supported with this feature. This feature 88 * cannot be set in features for a device with 89 * NETIF_F_HW_CSUM also set. This feature is being 90 * DEPRECATED (see below). 91 * 92 * * - %NETIF_F_RXCSUM 93 * - Driver (device) performs receive checksum offload. 94 * This flag is only used to disable the RX checksum 95 * feature for a device. The stack will accept receive 96 * checksum indication in packets received on a device 97 * regardless of whether NETIF_F_RXCSUM is set. 98 * 99 * Checksumming of received packets by device 100 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 101 * 102 * Indication of checksum verification is set in &sk_buff.ip_summed. 103 * Possible values are: 104 * 105 * - %CHECKSUM_NONE 106 * 107 * Device did not checksum this packet e.g. due to lack of capabilities. 108 * The packet contains full (though not verified) checksum in packet but 109 * not in skb->csum. Thus, skb->csum is undefined in this case. 110 * 111 * - %CHECKSUM_UNNECESSARY 112 * 113 * The hardware you're dealing with doesn't calculate the full checksum 114 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums 115 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY 116 * if their checksums are okay. &sk_buff.csum is still undefined in this case 117 * though. A driver or device must never modify the checksum field in the 118 * packet even if checksum is verified. 119 * 120 * %CHECKSUM_UNNECESSARY is applicable to following protocols: 121 * 122 * - TCP: IPv6 and IPv4. 123 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 124 * zero UDP checksum for either IPv4 or IPv6, the networking stack 125 * may perform further validation in this case. 126 * - GRE: only if the checksum is present in the header. 127 * - SCTP: indicates the CRC in SCTP header has been validated. 128 * - FCOE: indicates the CRC in FC frame has been validated. 129 * 130 * &sk_buff.csum_level indicates the number of consecutive checksums found in 131 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY. 132 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 133 * and a device is able to verify the checksums for UDP (possibly zero), 134 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to 135 * two. If the device were only able to verify the UDP checksum and not 136 * GRE, either because it doesn't support GRE checksum or because GRE 137 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 138 * not considered in this case). 139 * 140 * - %CHECKSUM_COMPLETE 141 * 142 * This is the most generic way. The device supplied checksum of the _whole_ 143 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the 144 * hardware doesn't need to parse L3/L4 headers to implement this. 145 * 146 * Notes: 147 * 148 * - Even if device supports only some protocols, but is able to produce 149 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 150 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 151 * 152 * - %CHECKSUM_PARTIAL 153 * 154 * A checksum is set up to be offloaded to a device as described in the 155 * output description for CHECKSUM_PARTIAL. This may occur on a packet 156 * received directly from another Linux OS, e.g., a virtualized Linux kernel 157 * on the same host, or it may be set in the input path in GRO or remote 158 * checksum offload. For the purposes of checksum verification, the checksum 159 * referred to by skb->csum_start + skb->csum_offset and any preceding 160 * checksums in the packet are considered verified. Any checksums in the 161 * packet that are after the checksum being offloaded are not considered to 162 * be verified. 163 * 164 * Checksumming on transmit for non-GSO 165 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 166 * 167 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet. 168 * Values are: 169 * 170 * - %CHECKSUM_PARTIAL 171 * 172 * The driver is required to checksum the packet as seen by hard_start_xmit() 173 * from &sk_buff.csum_start up to the end, and to record/write the checksum at 174 * offset &sk_buff.csum_start + &sk_buff.csum_offset. 175 * A driver may verify that the 176 * csum_start and csum_offset values are valid values given the length and 177 * offset of the packet, but it should not attempt to validate that the 178 * checksum refers to a legitimate transport layer checksum -- it is the 179 * purview of the stack to validate that csum_start and csum_offset are set 180 * correctly. 181 * 182 * When the stack requests checksum offload for a packet, the driver MUST 183 * ensure that the checksum is set correctly. A driver can either offload the 184 * checksum calculation to the device, or call skb_checksum_help (in the case 185 * that the device does not support offload for a particular checksum). 186 * 187 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of 188 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate 189 * checksum offload capability. 190 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based 191 * on network device checksumming capabilities: if a packet does not match 192 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of 193 * &sk_buff.csum_not_inet, see :ref:`crc`) 194 * is called to resolve the checksum. 195 * 196 * - %CHECKSUM_NONE 197 * 198 * The skb was already checksummed by the protocol, or a checksum is not 199 * required. 200 * 201 * - %CHECKSUM_UNNECESSARY 202 * 203 * This has the same meaning as CHECKSUM_NONE for checksum offload on 204 * output. 205 * 206 * - %CHECKSUM_COMPLETE 207 * 208 * Not used in checksum output. If a driver observes a packet with this value 209 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set. 210 * 211 * .. _crc: 212 * 213 * Non-IP checksum (CRC) offloads 214 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 215 * 216 * .. flat-table:: 217 * :widths: 1 10 218 * 219 * * - %NETIF_F_SCTP_CRC 220 * - This feature indicates that a device is capable of 221 * offloading the SCTP CRC in a packet. To perform this offload the stack 222 * will set csum_start and csum_offset accordingly, set ip_summed to 223 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication 224 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c. 225 * A driver that supports both IP checksum offload and SCTP CRC32c offload 226 * must verify which offload is configured for a packet by testing the 227 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to 228 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 229 * 230 * * - %NETIF_F_FCOE_CRC 231 * - This feature indicates that a device is capable of offloading the FCOE 232 * CRC in a packet. To perform this offload the stack will set ip_summed 233 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset 234 * accordingly. Note that there is no indication in the skbuff that the 235 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports 236 * both IP checksum offload and FCOE CRC offload must verify which offload 237 * is configured for a packet, presumably by inspecting packet headers. 238 * 239 * Checksumming on output with GSO 240 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 241 * 242 * In the case of a GSO packet (skb_is_gso() is true), checksum offload 243 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 244 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as 245 * part of the GSO operation is implied. If a checksum is being offloaded 246 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and 247 * csum_offset are set to refer to the outermost checksum being offloaded 248 * (two offloaded checksums are possible with UDP encapsulation). 249 */ 250 251 /* Don't change this without changing skb_csum_unnecessary! */ 252 #define CHECKSUM_NONE 0 253 #define CHECKSUM_UNNECESSARY 1 254 #define CHECKSUM_COMPLETE 2 255 #define CHECKSUM_PARTIAL 3 256 257 /* Maximum value in skb->csum_level */ 258 #define SKB_MAX_CSUM_LEVEL 3 259 260 #define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 261 #define SKB_WITH_OVERHEAD(X) \ 262 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 263 #define SKB_MAX_ORDER(X, ORDER) \ 264 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 265 #define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 266 #define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 267 268 /* return minimum truesize of one skb containing X bytes of data */ 269 #define SKB_TRUESIZE(X) ((X) + \ 270 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 271 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 272 273 struct ahash_request; 274 struct net_device; 275 struct scatterlist; 276 struct pipe_inode_info; 277 struct iov_iter; 278 struct napi_struct; 279 struct bpf_prog; 280 union bpf_attr; 281 struct skb_ext; 282 283 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 284 struct nf_bridge_info { 285 enum { 286 BRNF_PROTO_UNCHANGED, 287 BRNF_PROTO_8021Q, 288 BRNF_PROTO_PPPOE 289 } orig_proto:8; 290 u8 pkt_otherhost:1; 291 u8 in_prerouting:1; 292 u8 bridged_dnat:1; 293 __u16 frag_max_size; 294 struct net_device *physindev; 295 296 /* always valid & non-NULL from FORWARD on, for physdev match */ 297 struct net_device *physoutdev; 298 union { 299 /* prerouting: detect dnat in orig/reply direction */ 300 __be32 ipv4_daddr; 301 struct in6_addr ipv6_daddr; 302 303 /* after prerouting + nat detected: store original source 304 * mac since neigh resolution overwrites it, only used while 305 * skb is out in neigh layer. 306 */ 307 char neigh_header[8]; 308 }; 309 }; 310 #endif 311 312 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 313 /* Chain in tc_skb_ext will be used to share the tc chain with 314 * ovs recirc_id. It will be set to the current chain by tc 315 * and read by ovs to recirc_id. 316 */ 317 struct tc_skb_ext { 318 __u32 chain; 319 __u16 mru; 320 __u16 zone; 321 u8 post_ct:1; 322 u8 post_ct_snat:1; 323 u8 post_ct_dnat:1; 324 }; 325 #endif 326 327 struct sk_buff_head { 328 /* These two members must be first to match sk_buff. */ 329 struct_group_tagged(sk_buff_list, list, 330 struct sk_buff *next; 331 struct sk_buff *prev; 332 ); 333 334 __u32 qlen; 335 spinlock_t lock; 336 }; 337 338 struct sk_buff; 339 340 /* The reason of skb drop, which is used in kfree_skb_reason(). 341 * en...maybe they should be splited by group? 342 * 343 * Each item here should also be in 'TRACE_SKB_DROP_REASON', which is 344 * used to translate the reason to string. 345 */ 346 enum skb_drop_reason { 347 SKB_NOT_DROPPED_YET = 0, 348 SKB_DROP_REASON_NOT_SPECIFIED, /* drop reason is not specified */ 349 SKB_DROP_REASON_NO_SOCKET, /* socket not found */ 350 SKB_DROP_REASON_PKT_TOO_SMALL, /* packet size is too small */ 351 SKB_DROP_REASON_TCP_CSUM, /* TCP checksum error */ 352 SKB_DROP_REASON_SOCKET_FILTER, /* dropped by socket filter */ 353 SKB_DROP_REASON_UDP_CSUM, /* UDP checksum error */ 354 SKB_DROP_REASON_NETFILTER_DROP, /* dropped by netfilter */ 355 SKB_DROP_REASON_OTHERHOST, /* packet don't belong to current 356 * host (interface is in promisc 357 * mode) 358 */ 359 SKB_DROP_REASON_IP_CSUM, /* IP checksum error */ 360 SKB_DROP_REASON_IP_INHDR, /* there is something wrong with 361 * IP header (see 362 * IPSTATS_MIB_INHDRERRORS) 363 */ 364 SKB_DROP_REASON_IP_RPFILTER, /* IP rpfilter validate failed. 365 * see the document for rp_filter 366 * in ip-sysctl.rst for more 367 * information 368 */ 369 SKB_DROP_REASON_UNICAST_IN_L2_MULTICAST, /* destination address of L2 370 * is multicast, but L3 is 371 * unicast. 372 */ 373 SKB_DROP_REASON_XFRM_POLICY, /* xfrm policy check failed */ 374 SKB_DROP_REASON_IP_NOPROTO, /* no support for IP protocol */ 375 SKB_DROP_REASON_SOCKET_RCVBUFF, /* socket receive buff is full */ 376 SKB_DROP_REASON_PROTO_MEM, /* proto memory limition, such as 377 * udp packet drop out of 378 * udp_memory_allocated. 379 */ 380 SKB_DROP_REASON_TCP_MD5NOTFOUND, /* no MD5 hash and one 381 * expected, corresponding 382 * to LINUX_MIB_TCPMD5NOTFOUND 383 */ 384 SKB_DROP_REASON_TCP_MD5UNEXPECTED, /* MD5 hash and we're not 385 * expecting one, corresponding 386 * to LINUX_MIB_TCPMD5UNEXPECTED 387 */ 388 SKB_DROP_REASON_TCP_MD5FAILURE, /* MD5 hash and its wrong, 389 * corresponding to 390 * LINUX_MIB_TCPMD5FAILURE 391 */ 392 SKB_DROP_REASON_SOCKET_BACKLOG, /* failed to add skb to socket 393 * backlog (see 394 * LINUX_MIB_TCPBACKLOGDROP) 395 */ 396 SKB_DROP_REASON_TCP_FLAGS, /* TCP flags invalid */ 397 SKB_DROP_REASON_TCP_ZEROWINDOW, /* TCP receive window size is zero, 398 * see LINUX_MIB_TCPZEROWINDOWDROP 399 */ 400 SKB_DROP_REASON_TCP_OLD_DATA, /* the TCP data reveived is already 401 * received before (spurious retrans 402 * may happened), see 403 * LINUX_MIB_DELAYEDACKLOST 404 */ 405 SKB_DROP_REASON_TCP_OVERWINDOW, /* the TCP data is out of window, 406 * the seq of the first byte exceed 407 * the right edges of receive 408 * window 409 */ 410 SKB_DROP_REASON_TCP_OFOMERGE, /* the data of skb is already in 411 * the ofo queue, corresponding to 412 * LINUX_MIB_TCPOFOMERGE 413 */ 414 SKB_DROP_REASON_TCP_RFC7323_PAWS, /* PAWS check, corresponding to 415 * LINUX_MIB_PAWSESTABREJECTED 416 */ 417 SKB_DROP_REASON_TCP_INVALID_SEQUENCE, /* Not acceptable SEQ field */ 418 SKB_DROP_REASON_TCP_RESET, /* Invalid RST packet */ 419 SKB_DROP_REASON_TCP_INVALID_SYN, /* Incoming packet has unexpected SYN flag */ 420 SKB_DROP_REASON_TCP_CLOSE, /* TCP socket in CLOSE state */ 421 SKB_DROP_REASON_TCP_FASTOPEN, /* dropped by FASTOPEN request socket */ 422 SKB_DROP_REASON_TCP_OLD_ACK, /* TCP ACK is old, but in window */ 423 SKB_DROP_REASON_TCP_TOO_OLD_ACK, /* TCP ACK is too old */ 424 SKB_DROP_REASON_TCP_ACK_UNSENT_DATA, /* TCP ACK for data we haven't sent yet */ 425 SKB_DROP_REASON_TCP_OFO_QUEUE_PRUNE, /* pruned from TCP OFO queue */ 426 SKB_DROP_REASON_TCP_OFO_DROP, /* data already in receive queue */ 427 SKB_DROP_REASON_IP_OUTNOROUTES, /* route lookup failed */ 428 SKB_DROP_REASON_BPF_CGROUP_EGRESS, /* dropped by 429 * BPF_PROG_TYPE_CGROUP_SKB 430 * eBPF program 431 */ 432 SKB_DROP_REASON_IPV6DISABLED, /* IPv6 is disabled on the device */ 433 SKB_DROP_REASON_NEIGH_CREATEFAIL, /* failed to create neigh 434 * entry 435 */ 436 SKB_DROP_REASON_NEIGH_FAILED, /* neigh entry in failed state */ 437 SKB_DROP_REASON_NEIGH_QUEUEFULL, /* arp_queue for neigh 438 * entry is full 439 */ 440 SKB_DROP_REASON_NEIGH_DEAD, /* neigh entry is dead */ 441 SKB_DROP_REASON_TC_EGRESS, /* dropped in TC egress HOOK */ 442 SKB_DROP_REASON_QDISC_DROP, /* dropped by qdisc when packet 443 * outputting (failed to enqueue to 444 * current qdisc) 445 */ 446 SKB_DROP_REASON_CPU_BACKLOG, /* failed to enqueue the skb to 447 * the per CPU backlog queue. This 448 * can be caused by backlog queue 449 * full (see netdev_max_backlog in 450 * net.rst) or RPS flow limit 451 */ 452 SKB_DROP_REASON_XDP, /* dropped by XDP in input path */ 453 SKB_DROP_REASON_TC_INGRESS, /* dropped in TC ingress HOOK */ 454 SKB_DROP_REASON_UNHANDLED_PROTO, /* protocol not implemented 455 * or not supported 456 */ 457 SKB_DROP_REASON_SKB_CSUM, /* sk_buff checksum computation 458 * error 459 */ 460 SKB_DROP_REASON_SKB_GSO_SEG, /* gso segmentation error */ 461 SKB_DROP_REASON_SKB_UCOPY_FAULT, /* failed to copy data from 462 * user space, e.g., via 463 * zerocopy_sg_from_iter() 464 * or skb_orphan_frags_rx() 465 */ 466 SKB_DROP_REASON_DEV_HDR, /* device driver specific 467 * header/metadata is invalid 468 */ 469 /* the device is not ready to xmit/recv due to any of its data 470 * structure that is not up/ready/initialized, e.g., the IFF_UP is 471 * not set, or driver specific tun->tfiles[txq] is not initialized 472 */ 473 SKB_DROP_REASON_DEV_READY, 474 SKB_DROP_REASON_FULL_RING, /* ring buffer is full */ 475 SKB_DROP_REASON_NOMEM, /* error due to OOM */ 476 SKB_DROP_REASON_HDR_TRUNC, /* failed to trunc/extract the header 477 * from networking data, e.g., failed 478 * to pull the protocol header from 479 * frags via pskb_may_pull() 480 */ 481 SKB_DROP_REASON_TAP_FILTER, /* dropped by (ebpf) filter directly 482 * attached to tun/tap, e.g., via 483 * TUNSETFILTEREBPF 484 */ 485 SKB_DROP_REASON_TAP_TXFILTER, /* dropped by tx filter implemented 486 * at tun/tap, e.g., check_filter() 487 */ 488 SKB_DROP_REASON_ICMP_CSUM, /* ICMP checksum error */ 489 SKB_DROP_REASON_INVALID_PROTO, /* the packet doesn't follow RFC 490 * 2211, such as a broadcasts 491 * ICMP_TIMESTAMP 492 */ 493 SKB_DROP_REASON_IP_INADDRERRORS, /* host unreachable, corresponding 494 * to IPSTATS_MIB_INADDRERRORS 495 */ 496 SKB_DROP_REASON_IP_INNOROUTES, /* network unreachable, corresponding 497 * to IPSTATS_MIB_INADDRERRORS 498 */ 499 SKB_DROP_REASON_PKT_TOO_BIG, /* packet size is too big (maybe exceed 500 * the MTU) 501 */ 502 SKB_DROP_REASON_MAX, 503 }; 504 505 #define SKB_DR_INIT(name, reason) \ 506 enum skb_drop_reason name = SKB_DROP_REASON_##reason 507 #define SKB_DR(name) \ 508 SKB_DR_INIT(name, NOT_SPECIFIED) 509 #define SKB_DR_SET(name, reason) \ 510 (name = SKB_DROP_REASON_##reason) 511 #define SKB_DR_OR(name, reason) \ 512 do { \ 513 if (name == SKB_DROP_REASON_NOT_SPECIFIED) \ 514 SKB_DR_SET(name, reason); \ 515 } while (0) 516 517 /* To allow 64K frame to be packed as single skb without frag_list we 518 * require 64K/PAGE_SIZE pages plus 1 additional page to allow for 519 * buffers which do not start on a page boundary. 520 * 521 * Since GRO uses frags we allocate at least 16 regardless of page 522 * size. 523 */ 524 #if (65536/PAGE_SIZE + 1) < 16 525 #define MAX_SKB_FRAGS 16UL 526 #else 527 #define MAX_SKB_FRAGS (65536/PAGE_SIZE + 1) 528 #endif 529 extern int sysctl_max_skb_frags; 530 531 /* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 532 * segment using its current segmentation instead. 533 */ 534 #define GSO_BY_FRAGS 0xFFFF 535 536 typedef struct bio_vec skb_frag_t; 537 538 /** 539 * skb_frag_size() - Returns the size of a skb fragment 540 * @frag: skb fragment 541 */ 542 static inline unsigned int skb_frag_size(const skb_frag_t *frag) 543 { 544 return frag->bv_len; 545 } 546 547 /** 548 * skb_frag_size_set() - Sets the size of a skb fragment 549 * @frag: skb fragment 550 * @size: size of fragment 551 */ 552 static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 553 { 554 frag->bv_len = size; 555 } 556 557 /** 558 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 559 * @frag: skb fragment 560 * @delta: value to add 561 */ 562 static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 563 { 564 frag->bv_len += delta; 565 } 566 567 /** 568 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 569 * @frag: skb fragment 570 * @delta: value to subtract 571 */ 572 static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 573 { 574 frag->bv_len -= delta; 575 } 576 577 /** 578 * skb_frag_must_loop - Test if %p is a high memory page 579 * @p: fragment's page 580 */ 581 static inline bool skb_frag_must_loop(struct page *p) 582 { 583 #if defined(CONFIG_HIGHMEM) 584 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 585 return true; 586 #endif 587 return false; 588 } 589 590 /** 591 * skb_frag_foreach_page - loop over pages in a fragment 592 * 593 * @f: skb frag to operate on 594 * @f_off: offset from start of f->bv_page 595 * @f_len: length from f_off to loop over 596 * @p: (temp var) current page 597 * @p_off: (temp var) offset from start of current page, 598 * non-zero only on first page. 599 * @p_len: (temp var) length in current page, 600 * < PAGE_SIZE only on first and last page. 601 * @copied: (temp var) length so far, excluding current p_len. 602 * 603 * A fragment can hold a compound page, in which case per-page 604 * operations, notably kmap_atomic, must be called for each 605 * regular page. 606 */ 607 #define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 608 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 609 p_off = (f_off) & (PAGE_SIZE - 1), \ 610 p_len = skb_frag_must_loop(p) ? \ 611 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 612 copied = 0; \ 613 copied < f_len; \ 614 copied += p_len, p++, p_off = 0, \ 615 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 616 617 #define HAVE_HW_TIME_STAMP 618 619 /** 620 * struct skb_shared_hwtstamps - hardware time stamps 621 * @hwtstamp: hardware time stamp transformed into duration 622 * since arbitrary point in time 623 * @netdev_data: address/cookie of network device driver used as 624 * reference to actual hardware time stamp 625 * 626 * Software time stamps generated by ktime_get_real() are stored in 627 * skb->tstamp. 628 * 629 * hwtstamps can only be compared against other hwtstamps from 630 * the same device. 631 * 632 * This structure is attached to packets as part of the 633 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 634 */ 635 struct skb_shared_hwtstamps { 636 union { 637 ktime_t hwtstamp; 638 void *netdev_data; 639 }; 640 }; 641 642 /* Definitions for tx_flags in struct skb_shared_info */ 643 enum { 644 /* generate hardware time stamp */ 645 SKBTX_HW_TSTAMP = 1 << 0, 646 647 /* generate software time stamp when queueing packet to NIC */ 648 SKBTX_SW_TSTAMP = 1 << 1, 649 650 /* device driver is going to provide hardware time stamp */ 651 SKBTX_IN_PROGRESS = 1 << 2, 652 653 /* generate hardware time stamp based on cycles if supported */ 654 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3, 655 656 /* generate wifi status information (where possible) */ 657 SKBTX_WIFI_STATUS = 1 << 4, 658 659 /* determine hardware time stamp based on time or cycles */ 660 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 661 662 /* generate software time stamp when entering packet scheduling */ 663 SKBTX_SCHED_TSTAMP = 1 << 6, 664 }; 665 666 #define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 667 SKBTX_SCHED_TSTAMP) 668 #define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 669 SKBTX_HW_TSTAMP_USE_CYCLES | \ 670 SKBTX_ANY_SW_TSTAMP) 671 672 /* Definitions for flags in struct skb_shared_info */ 673 enum { 674 /* use zcopy routines */ 675 SKBFL_ZEROCOPY_ENABLE = BIT(0), 676 677 /* This indicates at least one fragment might be overwritten 678 * (as in vmsplice(), sendfile() ...) 679 * If we need to compute a TX checksum, we'll need to copy 680 * all frags to avoid possible bad checksum 681 */ 682 SKBFL_SHARED_FRAG = BIT(1), 683 684 /* segment contains only zerocopy data and should not be 685 * charged to the kernel memory. 686 */ 687 SKBFL_PURE_ZEROCOPY = BIT(2), 688 }; 689 690 #define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 691 #define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY) 692 693 /* 694 * The callback notifies userspace to release buffers when skb DMA is done in 695 * lower device, the skb last reference should be 0 when calling this. 696 * The zerocopy_success argument is true if zero copy transmit occurred, 697 * false on data copy or out of memory error caused by data copy attempt. 698 * The ctx field is used to track device context. 699 * The desc field is used to track userspace buffer index. 700 */ 701 struct ubuf_info { 702 void (*callback)(struct sk_buff *, struct ubuf_info *, 703 bool zerocopy_success); 704 union { 705 struct { 706 unsigned long desc; 707 void *ctx; 708 }; 709 struct { 710 u32 id; 711 u16 len; 712 u16 zerocopy:1; 713 u32 bytelen; 714 }; 715 }; 716 refcount_t refcnt; 717 u8 flags; 718 719 struct mmpin { 720 struct user_struct *user; 721 unsigned int num_pg; 722 } mmp; 723 }; 724 725 #define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 726 727 int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 728 void mm_unaccount_pinned_pages(struct mmpin *mmp); 729 730 /* This data is invariant across clones and lives at 731 * the end of the header data, ie. at skb->end. 732 */ 733 struct skb_shared_info { 734 __u8 flags; 735 __u8 meta_len; 736 __u8 nr_frags; 737 __u8 tx_flags; 738 unsigned short gso_size; 739 /* Warning: this field is not always filled in (UFO)! */ 740 unsigned short gso_segs; 741 struct sk_buff *frag_list; 742 struct skb_shared_hwtstamps hwtstamps; 743 unsigned int gso_type; 744 u32 tskey; 745 746 /* 747 * Warning : all fields before dataref are cleared in __alloc_skb() 748 */ 749 atomic_t dataref; 750 unsigned int xdp_frags_size; 751 752 /* Intermediate layers must ensure that destructor_arg 753 * remains valid until skb destructor */ 754 void * destructor_arg; 755 756 /* must be last field, see pskb_expand_head() */ 757 skb_frag_t frags[MAX_SKB_FRAGS]; 758 }; 759 760 /** 761 * DOC: dataref and headerless skbs 762 * 763 * Transport layers send out clones of payload skbs they hold for 764 * retransmissions. To allow lower layers of the stack to prepend their headers 765 * we split &skb_shared_info.dataref into two halves. 766 * The lower 16 bits count the overall number of references. 767 * The higher 16 bits indicate how many of the references are payload-only. 768 * skb_header_cloned() checks if skb is allowed to add / write the headers. 769 * 770 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 771 * (via __skb_header_release()). Any clone created from marked skb will get 772 * &sk_buff.hdr_len populated with the available headroom. 773 * If there's the only clone in existence it's able to modify the headroom 774 * at will. The sequence of calls inside the transport layer is:: 775 * 776 * <alloc skb> 777 * skb_reserve() 778 * __skb_header_release() 779 * skb_clone() 780 * // send the clone down the stack 781 * 782 * This is not a very generic construct and it depends on the transport layers 783 * doing the right thing. In practice there's usually only one payload-only skb. 784 * Having multiple payload-only skbs with different lengths of hdr_len is not 785 * possible. The payload-only skbs should never leave their owner. 786 */ 787 #define SKB_DATAREF_SHIFT 16 788 #define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 789 790 791 enum { 792 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 793 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 794 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 795 }; 796 797 enum { 798 SKB_GSO_TCPV4 = 1 << 0, 799 800 /* This indicates the skb is from an untrusted source. */ 801 SKB_GSO_DODGY = 1 << 1, 802 803 /* This indicates the tcp segment has CWR set. */ 804 SKB_GSO_TCP_ECN = 1 << 2, 805 806 SKB_GSO_TCP_FIXEDID = 1 << 3, 807 808 SKB_GSO_TCPV6 = 1 << 4, 809 810 SKB_GSO_FCOE = 1 << 5, 811 812 SKB_GSO_GRE = 1 << 6, 813 814 SKB_GSO_GRE_CSUM = 1 << 7, 815 816 SKB_GSO_IPXIP4 = 1 << 8, 817 818 SKB_GSO_IPXIP6 = 1 << 9, 819 820 SKB_GSO_UDP_TUNNEL = 1 << 10, 821 822 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 823 824 SKB_GSO_PARTIAL = 1 << 12, 825 826 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 827 828 SKB_GSO_SCTP = 1 << 14, 829 830 SKB_GSO_ESP = 1 << 15, 831 832 SKB_GSO_UDP = 1 << 16, 833 834 SKB_GSO_UDP_L4 = 1 << 17, 835 836 SKB_GSO_FRAGLIST = 1 << 18, 837 }; 838 839 #if BITS_PER_LONG > 32 840 #define NET_SKBUFF_DATA_USES_OFFSET 1 841 #endif 842 843 #ifdef NET_SKBUFF_DATA_USES_OFFSET 844 typedef unsigned int sk_buff_data_t; 845 #else 846 typedef unsigned char *sk_buff_data_t; 847 #endif 848 849 /** 850 * DOC: Basic sk_buff geometry 851 * 852 * struct sk_buff itself is a metadata structure and does not hold any packet 853 * data. All the data is held in associated buffers. 854 * 855 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 856 * into two parts: 857 * 858 * - data buffer, containing headers and sometimes payload; 859 * this is the part of the skb operated on by the common helpers 860 * such as skb_put() or skb_pull(); 861 * - shared info (struct skb_shared_info) which holds an array of pointers 862 * to read-only data in the (page, offset, length) format. 863 * 864 * Optionally &skb_shared_info.frag_list may point to another skb. 865 * 866 * Basic diagram may look like this:: 867 * 868 * --------------- 869 * | sk_buff | 870 * --------------- 871 * ,--------------------------- + head 872 * / ,----------------- + data 873 * / / ,----------- + tail 874 * | | | , + end 875 * | | | | 876 * v v v v 877 * ----------------------------------------------- 878 * | headroom | data | tailroom | skb_shared_info | 879 * ----------------------------------------------- 880 * + [page frag] 881 * + [page frag] 882 * + [page frag] 883 * + [page frag] --------- 884 * + frag_list --> | sk_buff | 885 * --------- 886 * 887 */ 888 889 /** 890 * struct sk_buff - socket buffer 891 * @next: Next buffer in list 892 * @prev: Previous buffer in list 893 * @tstamp: Time we arrived/left 894 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 895 * for retransmit timer 896 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 897 * @list: queue head 898 * @ll_node: anchor in an llist (eg socket defer_list) 899 * @sk: Socket we are owned by 900 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in 901 * fragmentation management 902 * @dev: Device we arrived on/are leaving by 903 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 904 * @cb: Control buffer. Free for use by every layer. Put private vars here 905 * @_skb_refdst: destination entry (with norefcount bit) 906 * @sp: the security path, used for xfrm 907 * @len: Length of actual data 908 * @data_len: Data length 909 * @mac_len: Length of link layer header 910 * @hdr_len: writable header length of cloned skb 911 * @csum: Checksum (must include start/offset pair) 912 * @csum_start: Offset from skb->head where checksumming should start 913 * @csum_offset: Offset from csum_start where checksum should be stored 914 * @priority: Packet queueing priority 915 * @ignore_df: allow local fragmentation 916 * @cloned: Head may be cloned (check refcnt to be sure) 917 * @ip_summed: Driver fed us an IP checksum 918 * @nohdr: Payload reference only, must not modify header 919 * @pkt_type: Packet class 920 * @fclone: skbuff clone status 921 * @ipvs_property: skbuff is owned by ipvs 922 * @inner_protocol_type: whether the inner protocol is 923 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 924 * @remcsum_offload: remote checksum offload is enabled 925 * @offload_fwd_mark: Packet was L2-forwarded in hardware 926 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 927 * @tc_skip_classify: do not classify packet. set by IFB device 928 * @tc_at_ingress: used within tc_classify to distinguish in/egress 929 * @redirected: packet was redirected by packet classifier 930 * @from_ingress: packet was redirected from the ingress path 931 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 932 * @peeked: this packet has been seen already, so stats have been 933 * done for it, don't do them again 934 * @nf_trace: netfilter packet trace flag 935 * @protocol: Packet protocol from driver 936 * @destructor: Destruct function 937 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 938 * @_sk_redir: socket redirection information for skmsg 939 * @_nfct: Associated connection, if any (with nfctinfo bits) 940 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 941 * @skb_iif: ifindex of device we arrived on 942 * @tc_index: Traffic control index 943 * @hash: the packet hash 944 * @queue_mapping: Queue mapping for multiqueue devices 945 * @head_frag: skb was allocated from page fragments, 946 * not allocated by kmalloc() or vmalloc(). 947 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 948 * @pp_recycle: mark the packet for recycling instead of freeing (implies 949 * page_pool support on driver) 950 * @active_extensions: active extensions (skb_ext_id types) 951 * @ndisc_nodetype: router type (from link layer) 952 * @ooo_okay: allow the mapping of a socket to a queue to be changed 953 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 954 * ports. 955 * @sw_hash: indicates hash was computed in software stack 956 * @wifi_acked_valid: wifi_acked was set 957 * @wifi_acked: whether frame was acked on wifi or not 958 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 959 * @encapsulation: indicates the inner headers in the skbuff are valid 960 * @encap_hdr_csum: software checksum is needed 961 * @csum_valid: checksum is already valid 962 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 963 * @csum_complete_sw: checksum was completed by software 964 * @csum_level: indicates the number of consecutive checksums found in 965 * the packet minus one that have been verified as 966 * CHECKSUM_UNNECESSARY (max 3) 967 * @dst_pending_confirm: need to confirm neighbour 968 * @decrypted: Decrypted SKB 969 * @slow_gro: state present at GRO time, slower prepare step required 970 * @mono_delivery_time: When set, skb->tstamp has the 971 * delivery_time in mono clock base (i.e. EDT). Otherwise, the 972 * skb->tstamp has the (rcv) timestamp at ingress and 973 * delivery_time at egress. 974 * @napi_id: id of the NAPI struct this skb came from 975 * @sender_cpu: (aka @napi_id) source CPU in XPS 976 * @alloc_cpu: CPU which did the skb allocation. 977 * @secmark: security marking 978 * @mark: Generic packet mark 979 * @reserved_tailroom: (aka @mark) number of bytes of free space available 980 * at the tail of an sk_buff 981 * @vlan_present: VLAN tag is present 982 * @vlan_proto: vlan encapsulation protocol 983 * @vlan_tci: vlan tag control information 984 * @inner_protocol: Protocol (encapsulation) 985 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 986 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 987 * @inner_transport_header: Inner transport layer header (encapsulation) 988 * @inner_network_header: Network layer header (encapsulation) 989 * @inner_mac_header: Link layer header (encapsulation) 990 * @transport_header: Transport layer header 991 * @network_header: Network layer header 992 * @mac_header: Link layer header 993 * @kcov_handle: KCOV remote handle for remote coverage collection 994 * @tail: Tail pointer 995 * @end: End pointer 996 * @head: Head of buffer 997 * @data: Data head pointer 998 * @truesize: Buffer size 999 * @users: User count - see {datagram,tcp}.c 1000 * @extensions: allocated extensions, valid if active_extensions is nonzero 1001 */ 1002 1003 struct sk_buff { 1004 union { 1005 struct { 1006 /* These two members must be first to match sk_buff_head. */ 1007 struct sk_buff *next; 1008 struct sk_buff *prev; 1009 1010 union { 1011 struct net_device *dev; 1012 /* Some protocols might use this space to store information, 1013 * while device pointer would be NULL. 1014 * UDP receive path is one user. 1015 */ 1016 unsigned long dev_scratch; 1017 }; 1018 }; 1019 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 1020 struct list_head list; 1021 struct llist_node ll_node; 1022 }; 1023 1024 union { 1025 struct sock *sk; 1026 int ip_defrag_offset; 1027 }; 1028 1029 union { 1030 ktime_t tstamp; 1031 u64 skb_mstamp_ns; /* earliest departure time */ 1032 }; 1033 /* 1034 * This is the control buffer. It is free to use for every 1035 * layer. Please put your private variables there. If you 1036 * want to keep them across layers you have to do a skb_clone() 1037 * first. This is owned by whoever has the skb queued ATM. 1038 */ 1039 char cb[48] __aligned(8); 1040 1041 union { 1042 struct { 1043 unsigned long _skb_refdst; 1044 void (*destructor)(struct sk_buff *skb); 1045 }; 1046 struct list_head tcp_tsorted_anchor; 1047 #ifdef CONFIG_NET_SOCK_MSG 1048 unsigned long _sk_redir; 1049 #endif 1050 }; 1051 1052 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 1053 unsigned long _nfct; 1054 #endif 1055 unsigned int len, 1056 data_len; 1057 __u16 mac_len, 1058 hdr_len; 1059 1060 /* Following fields are _not_ copied in __copy_skb_header() 1061 * Note that queue_mapping is here mostly to fill a hole. 1062 */ 1063 __u16 queue_mapping; 1064 1065 /* if you move cloned around you also must adapt those constants */ 1066 #ifdef __BIG_ENDIAN_BITFIELD 1067 #define CLONED_MASK (1 << 7) 1068 #else 1069 #define CLONED_MASK 1 1070 #endif 1071 #define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 1072 1073 /* private: */ 1074 __u8 __cloned_offset[0]; 1075 /* public: */ 1076 __u8 cloned:1, 1077 nohdr:1, 1078 fclone:2, 1079 peeked:1, 1080 head_frag:1, 1081 pfmemalloc:1, 1082 pp_recycle:1; /* page_pool recycle indicator */ 1083 #ifdef CONFIG_SKB_EXTENSIONS 1084 __u8 active_extensions; 1085 #endif 1086 1087 /* Fields enclosed in headers group are copied 1088 * using a single memcpy() in __copy_skb_header() 1089 */ 1090 struct_group(headers, 1091 1092 /* private: */ 1093 __u8 __pkt_type_offset[0]; 1094 /* public: */ 1095 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 1096 __u8 ignore_df:1; 1097 __u8 nf_trace:1; 1098 __u8 ip_summed:2; 1099 __u8 ooo_okay:1; 1100 1101 __u8 l4_hash:1; 1102 __u8 sw_hash:1; 1103 __u8 wifi_acked_valid:1; 1104 __u8 wifi_acked:1; 1105 __u8 no_fcs:1; 1106 /* Indicates the inner headers are valid in the skbuff. */ 1107 __u8 encapsulation:1; 1108 __u8 encap_hdr_csum:1; 1109 __u8 csum_valid:1; 1110 1111 /* private: */ 1112 __u8 __pkt_vlan_present_offset[0]; 1113 /* public: */ 1114 __u8 vlan_present:1; /* See PKT_VLAN_PRESENT_BIT */ 1115 __u8 csum_complete_sw:1; 1116 __u8 csum_level:2; 1117 __u8 dst_pending_confirm:1; 1118 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */ 1119 #ifdef CONFIG_NET_CLS_ACT 1120 __u8 tc_skip_classify:1; 1121 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 1122 #endif 1123 #ifdef CONFIG_IPV6_NDISC_NODETYPE 1124 __u8 ndisc_nodetype:2; 1125 #endif 1126 1127 __u8 ipvs_property:1; 1128 __u8 inner_protocol_type:1; 1129 __u8 remcsum_offload:1; 1130 #ifdef CONFIG_NET_SWITCHDEV 1131 __u8 offload_fwd_mark:1; 1132 __u8 offload_l3_fwd_mark:1; 1133 #endif 1134 __u8 redirected:1; 1135 #ifdef CONFIG_NET_REDIRECT 1136 __u8 from_ingress:1; 1137 #endif 1138 #ifdef CONFIG_NETFILTER_SKIP_EGRESS 1139 __u8 nf_skip_egress:1; 1140 #endif 1141 #ifdef CONFIG_TLS_DEVICE 1142 __u8 decrypted:1; 1143 #endif 1144 __u8 slow_gro:1; 1145 __u8 csum_not_inet:1; 1146 1147 #ifdef CONFIG_NET_SCHED 1148 __u16 tc_index; /* traffic control index */ 1149 #endif 1150 1151 union { 1152 __wsum csum; 1153 struct { 1154 __u16 csum_start; 1155 __u16 csum_offset; 1156 }; 1157 }; 1158 __u32 priority; 1159 int skb_iif; 1160 __u32 hash; 1161 __be16 vlan_proto; 1162 __u16 vlan_tci; 1163 #if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1164 union { 1165 unsigned int napi_id; 1166 unsigned int sender_cpu; 1167 }; 1168 #endif 1169 u16 alloc_cpu; 1170 #ifdef CONFIG_NETWORK_SECMARK 1171 __u32 secmark; 1172 #endif 1173 1174 union { 1175 __u32 mark; 1176 __u32 reserved_tailroom; 1177 }; 1178 1179 union { 1180 __be16 inner_protocol; 1181 __u8 inner_ipproto; 1182 }; 1183 1184 __u16 inner_transport_header; 1185 __u16 inner_network_header; 1186 __u16 inner_mac_header; 1187 1188 __be16 protocol; 1189 __u16 transport_header; 1190 __u16 network_header; 1191 __u16 mac_header; 1192 1193 #ifdef CONFIG_KCOV 1194 u64 kcov_handle; 1195 #endif 1196 1197 ); /* end headers group */ 1198 1199 /* These elements must be at the end, see alloc_skb() for details. */ 1200 sk_buff_data_t tail; 1201 sk_buff_data_t end; 1202 unsigned char *head, 1203 *data; 1204 unsigned int truesize; 1205 refcount_t users; 1206 1207 #ifdef CONFIG_SKB_EXTENSIONS 1208 /* only useable after checking ->active_extensions != 0 */ 1209 struct skb_ext *extensions; 1210 #endif 1211 }; 1212 1213 /* if you move pkt_type around you also must adapt those constants */ 1214 #ifdef __BIG_ENDIAN_BITFIELD 1215 #define PKT_TYPE_MAX (7 << 5) 1216 #else 1217 #define PKT_TYPE_MAX 7 1218 #endif 1219 #define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1220 1221 /* if you move pkt_vlan_present, tc_at_ingress, or mono_delivery_time 1222 * around, you also must adapt these constants. 1223 */ 1224 #ifdef __BIG_ENDIAN_BITFIELD 1225 #define PKT_VLAN_PRESENT_BIT 7 1226 #define TC_AT_INGRESS_MASK (1 << 0) 1227 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 2) 1228 #else 1229 #define PKT_VLAN_PRESENT_BIT 0 1230 #define TC_AT_INGRESS_MASK (1 << 7) 1231 #define SKB_MONO_DELIVERY_TIME_MASK (1 << 5) 1232 #endif 1233 #define PKT_VLAN_PRESENT_OFFSET offsetof(struct sk_buff, __pkt_vlan_present_offset) 1234 1235 #ifdef __KERNEL__ 1236 /* 1237 * Handling routines are only of interest to the kernel 1238 */ 1239 1240 #define SKB_ALLOC_FCLONE 0x01 1241 #define SKB_ALLOC_RX 0x02 1242 #define SKB_ALLOC_NAPI 0x04 1243 1244 /** 1245 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1246 * @skb: buffer 1247 */ 1248 static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1249 { 1250 return unlikely(skb->pfmemalloc); 1251 } 1252 1253 /* 1254 * skb might have a dst pointer attached, refcounted or not. 1255 * _skb_refdst low order bit is set if refcount was _not_ taken 1256 */ 1257 #define SKB_DST_NOREF 1UL 1258 #define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1259 1260 /** 1261 * skb_dst - returns skb dst_entry 1262 * @skb: buffer 1263 * 1264 * Returns skb dst_entry, regardless of reference taken or not. 1265 */ 1266 static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1267 { 1268 /* If refdst was not refcounted, check we still are in a 1269 * rcu_read_lock section 1270 */ 1271 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1272 !rcu_read_lock_held() && 1273 !rcu_read_lock_bh_held()); 1274 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1275 } 1276 1277 /** 1278 * skb_dst_set - sets skb dst 1279 * @skb: buffer 1280 * @dst: dst entry 1281 * 1282 * Sets skb dst, assuming a reference was taken on dst and should 1283 * be released by skb_dst_drop() 1284 */ 1285 static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1286 { 1287 skb->slow_gro |= !!dst; 1288 skb->_skb_refdst = (unsigned long)dst; 1289 } 1290 1291 /** 1292 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1293 * @skb: buffer 1294 * @dst: dst entry 1295 * 1296 * Sets skb dst, assuming a reference was not taken on dst. 1297 * If dst entry is cached, we do not take reference and dst_release 1298 * will be avoided by refdst_drop. If dst entry is not cached, we take 1299 * reference, so that last dst_release can destroy the dst immediately. 1300 */ 1301 static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1302 { 1303 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1304 skb->slow_gro |= !!dst; 1305 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1306 } 1307 1308 /** 1309 * skb_dst_is_noref - Test if skb dst isn't refcounted 1310 * @skb: buffer 1311 */ 1312 static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1313 { 1314 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1315 } 1316 1317 /** 1318 * skb_rtable - Returns the skb &rtable 1319 * @skb: buffer 1320 */ 1321 static inline struct rtable *skb_rtable(const struct sk_buff *skb) 1322 { 1323 return (struct rtable *)skb_dst(skb); 1324 } 1325 1326 /* For mangling skb->pkt_type from user space side from applications 1327 * such as nft, tc, etc, we only allow a conservative subset of 1328 * possible pkt_types to be set. 1329 */ 1330 static inline bool skb_pkt_type_ok(u32 ptype) 1331 { 1332 return ptype <= PACKET_OTHERHOST; 1333 } 1334 1335 /** 1336 * skb_napi_id - Returns the skb's NAPI id 1337 * @skb: buffer 1338 */ 1339 static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1340 { 1341 #ifdef CONFIG_NET_RX_BUSY_POLL 1342 return skb->napi_id; 1343 #else 1344 return 0; 1345 #endif 1346 } 1347 1348 /** 1349 * skb_unref - decrement the skb's reference count 1350 * @skb: buffer 1351 * 1352 * Returns true if we can free the skb. 1353 */ 1354 static inline bool skb_unref(struct sk_buff *skb) 1355 { 1356 if (unlikely(!skb)) 1357 return false; 1358 if (likely(refcount_read(&skb->users) == 1)) 1359 smp_rmb(); 1360 else if (likely(!refcount_dec_and_test(&skb->users))) 1361 return false; 1362 1363 return true; 1364 } 1365 1366 void kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason); 1367 1368 /** 1369 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1370 * @skb: buffer to free 1371 */ 1372 static inline void kfree_skb(struct sk_buff *skb) 1373 { 1374 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1375 } 1376 1377 void skb_release_head_state(struct sk_buff *skb); 1378 void kfree_skb_list_reason(struct sk_buff *segs, 1379 enum skb_drop_reason reason); 1380 void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1381 void skb_tx_error(struct sk_buff *skb); 1382 1383 static inline void kfree_skb_list(struct sk_buff *segs) 1384 { 1385 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1386 } 1387 1388 #ifdef CONFIG_TRACEPOINTS 1389 void consume_skb(struct sk_buff *skb); 1390 #else 1391 static inline void consume_skb(struct sk_buff *skb) 1392 { 1393 return kfree_skb(skb); 1394 } 1395 #endif 1396 1397 void __consume_stateless_skb(struct sk_buff *skb); 1398 void __kfree_skb(struct sk_buff *skb); 1399 extern struct kmem_cache *skbuff_head_cache; 1400 1401 void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1402 bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1403 bool *fragstolen, int *delta_truesize); 1404 1405 struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1406 int node); 1407 struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1408 struct sk_buff *build_skb(void *data, unsigned int frag_size); 1409 struct sk_buff *build_skb_around(struct sk_buff *skb, 1410 void *data, unsigned int frag_size); 1411 void skb_attempt_defer_free(struct sk_buff *skb); 1412 1413 struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1414 1415 /** 1416 * alloc_skb - allocate a network buffer 1417 * @size: size to allocate 1418 * @priority: allocation mask 1419 * 1420 * This function is a convenient wrapper around __alloc_skb(). 1421 */ 1422 static inline struct sk_buff *alloc_skb(unsigned int size, 1423 gfp_t priority) 1424 { 1425 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1426 } 1427 1428 struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1429 unsigned long data_len, 1430 int max_page_order, 1431 int *errcode, 1432 gfp_t gfp_mask); 1433 struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1434 1435 /* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1436 struct sk_buff_fclones { 1437 struct sk_buff skb1; 1438 1439 struct sk_buff skb2; 1440 1441 refcount_t fclone_ref; 1442 }; 1443 1444 /** 1445 * skb_fclone_busy - check if fclone is busy 1446 * @sk: socket 1447 * @skb: buffer 1448 * 1449 * Returns true if skb is a fast clone, and its clone is not freed. 1450 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1451 * so we also check that this didnt happen. 1452 */ 1453 static inline bool skb_fclone_busy(const struct sock *sk, 1454 const struct sk_buff *skb) 1455 { 1456 const struct sk_buff_fclones *fclones; 1457 1458 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1459 1460 return skb->fclone == SKB_FCLONE_ORIG && 1461 refcount_read(&fclones->fclone_ref) > 1 && 1462 READ_ONCE(fclones->skb2.sk) == sk; 1463 } 1464 1465 /** 1466 * alloc_skb_fclone - allocate a network buffer from fclone cache 1467 * @size: size to allocate 1468 * @priority: allocation mask 1469 * 1470 * This function is a convenient wrapper around __alloc_skb(). 1471 */ 1472 static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1473 gfp_t priority) 1474 { 1475 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1476 } 1477 1478 struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1479 void skb_headers_offset_update(struct sk_buff *skb, int off); 1480 int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1481 struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1482 void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1483 struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1484 struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1485 gfp_t gfp_mask, bool fclone); 1486 static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1487 gfp_t gfp_mask) 1488 { 1489 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1490 } 1491 1492 int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1493 struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1494 unsigned int headroom); 1495 struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1496 struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1497 int newtailroom, gfp_t priority); 1498 int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1499 int offset, int len); 1500 int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1501 int offset, int len); 1502 int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1503 int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1504 1505 /** 1506 * skb_pad - zero pad the tail of an skb 1507 * @skb: buffer to pad 1508 * @pad: space to pad 1509 * 1510 * Ensure that a buffer is followed by a padding area that is zero 1511 * filled. Used by network drivers which may DMA or transfer data 1512 * beyond the buffer end onto the wire. 1513 * 1514 * May return error in out of memory cases. The skb is freed on error. 1515 */ 1516 static inline int skb_pad(struct sk_buff *skb, int pad) 1517 { 1518 return __skb_pad(skb, pad, true); 1519 } 1520 #define dev_kfree_skb(a) consume_skb(a) 1521 1522 int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1523 int offset, size_t size); 1524 1525 struct skb_seq_state { 1526 __u32 lower_offset; 1527 __u32 upper_offset; 1528 __u32 frag_idx; 1529 __u32 stepped_offset; 1530 struct sk_buff *root_skb; 1531 struct sk_buff *cur_skb; 1532 __u8 *frag_data; 1533 __u32 frag_off; 1534 }; 1535 1536 void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1537 unsigned int to, struct skb_seq_state *st); 1538 unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1539 struct skb_seq_state *st); 1540 void skb_abort_seq_read(struct skb_seq_state *st); 1541 1542 unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1543 unsigned int to, struct ts_config *config); 1544 1545 /* 1546 * Packet hash types specify the type of hash in skb_set_hash. 1547 * 1548 * Hash types refer to the protocol layer addresses which are used to 1549 * construct a packet's hash. The hashes are used to differentiate or identify 1550 * flows of the protocol layer for the hash type. Hash types are either 1551 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1552 * 1553 * Properties of hashes: 1554 * 1555 * 1) Two packets in different flows have different hash values 1556 * 2) Two packets in the same flow should have the same hash value 1557 * 1558 * A hash at a higher layer is considered to be more specific. A driver should 1559 * set the most specific hash possible. 1560 * 1561 * A driver cannot indicate a more specific hash than the layer at which a hash 1562 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1563 * 1564 * A driver may indicate a hash level which is less specific than the 1565 * actual layer the hash was computed on. For instance, a hash computed 1566 * at L4 may be considered an L3 hash. This should only be done if the 1567 * driver can't unambiguously determine that the HW computed the hash at 1568 * the higher layer. Note that the "should" in the second property above 1569 * permits this. 1570 */ 1571 enum pkt_hash_types { 1572 PKT_HASH_TYPE_NONE, /* Undefined type */ 1573 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1574 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1575 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1576 }; 1577 1578 static inline void skb_clear_hash(struct sk_buff *skb) 1579 { 1580 skb->hash = 0; 1581 skb->sw_hash = 0; 1582 skb->l4_hash = 0; 1583 } 1584 1585 static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1586 { 1587 if (!skb->l4_hash) 1588 skb_clear_hash(skb); 1589 } 1590 1591 static inline void 1592 __skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1593 { 1594 skb->l4_hash = is_l4; 1595 skb->sw_hash = is_sw; 1596 skb->hash = hash; 1597 } 1598 1599 static inline void 1600 skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1601 { 1602 /* Used by drivers to set hash from HW */ 1603 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1604 } 1605 1606 static inline void 1607 __skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1608 { 1609 __skb_set_hash(skb, hash, true, is_l4); 1610 } 1611 1612 void __skb_get_hash(struct sk_buff *skb); 1613 u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1614 u32 skb_get_poff(const struct sk_buff *skb); 1615 u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1616 const struct flow_keys_basic *keys, int hlen); 1617 __be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1618 const void *data, int hlen_proto); 1619 1620 static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1621 int thoff, u8 ip_proto) 1622 { 1623 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1624 } 1625 1626 void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1627 const struct flow_dissector_key *key, 1628 unsigned int key_count); 1629 1630 struct bpf_flow_dissector; 1631 bool bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1632 __be16 proto, int nhoff, int hlen, unsigned int flags); 1633 1634 bool __skb_flow_dissect(const struct net *net, 1635 const struct sk_buff *skb, 1636 struct flow_dissector *flow_dissector, 1637 void *target_container, const void *data, 1638 __be16 proto, int nhoff, int hlen, unsigned int flags); 1639 1640 static inline bool skb_flow_dissect(const struct sk_buff *skb, 1641 struct flow_dissector *flow_dissector, 1642 void *target_container, unsigned int flags) 1643 { 1644 return __skb_flow_dissect(NULL, skb, flow_dissector, 1645 target_container, NULL, 0, 0, 0, flags); 1646 } 1647 1648 static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1649 struct flow_keys *flow, 1650 unsigned int flags) 1651 { 1652 memset(flow, 0, sizeof(*flow)); 1653 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1654 flow, NULL, 0, 0, 0, flags); 1655 } 1656 1657 static inline bool 1658 skb_flow_dissect_flow_keys_basic(const struct net *net, 1659 const struct sk_buff *skb, 1660 struct flow_keys_basic *flow, 1661 const void *data, __be16 proto, 1662 int nhoff, int hlen, unsigned int flags) 1663 { 1664 memset(flow, 0, sizeof(*flow)); 1665 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1666 data, proto, nhoff, hlen, flags); 1667 } 1668 1669 void skb_flow_dissect_meta(const struct sk_buff *skb, 1670 struct flow_dissector *flow_dissector, 1671 void *target_container); 1672 1673 /* Gets a skb connection tracking info, ctinfo map should be a 1674 * map of mapsize to translate enum ip_conntrack_info states 1675 * to user states. 1676 */ 1677 void 1678 skb_flow_dissect_ct(const struct sk_buff *skb, 1679 struct flow_dissector *flow_dissector, 1680 void *target_container, 1681 u16 *ctinfo_map, size_t mapsize, 1682 bool post_ct, u16 zone); 1683 void 1684 skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1685 struct flow_dissector *flow_dissector, 1686 void *target_container); 1687 1688 void skb_flow_dissect_hash(const struct sk_buff *skb, 1689 struct flow_dissector *flow_dissector, 1690 void *target_container); 1691 1692 static inline __u32 skb_get_hash(struct sk_buff *skb) 1693 { 1694 if (!skb->l4_hash && !skb->sw_hash) 1695 __skb_get_hash(skb); 1696 1697 return skb->hash; 1698 } 1699 1700 static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1701 { 1702 if (!skb->l4_hash && !skb->sw_hash) { 1703 struct flow_keys keys; 1704 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1705 1706 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1707 } 1708 1709 return skb->hash; 1710 } 1711 1712 __u32 skb_get_hash_perturb(const struct sk_buff *skb, 1713 const siphash_key_t *perturb); 1714 1715 static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1716 { 1717 return skb->hash; 1718 } 1719 1720 static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1721 { 1722 to->hash = from->hash; 1723 to->sw_hash = from->sw_hash; 1724 to->l4_hash = from->l4_hash; 1725 }; 1726 1727 static inline void skb_copy_decrypted(struct sk_buff *to, 1728 const struct sk_buff *from) 1729 { 1730 #ifdef CONFIG_TLS_DEVICE 1731 to->decrypted = from->decrypted; 1732 #endif 1733 } 1734 1735 #ifdef NET_SKBUFF_DATA_USES_OFFSET 1736 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1737 { 1738 return skb->head + skb->end; 1739 } 1740 1741 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1742 { 1743 return skb->end; 1744 } 1745 1746 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1747 { 1748 skb->end = offset; 1749 } 1750 #else 1751 static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1752 { 1753 return skb->end; 1754 } 1755 1756 static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1757 { 1758 return skb->end - skb->head; 1759 } 1760 1761 static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1762 { 1763 skb->end = skb->head + offset; 1764 } 1765 #endif 1766 1767 static inline unsigned int skb_data_area_size(struct sk_buff *skb) 1768 { 1769 return skb_end_pointer(skb) - skb->data; 1770 } 1771 1772 struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1773 struct ubuf_info *uarg); 1774 1775 void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1776 1777 void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 1778 bool success); 1779 1780 int __zerocopy_sg_from_iter(struct sock *sk, struct sk_buff *skb, 1781 struct iov_iter *from, size_t length); 1782 1783 static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1784 struct msghdr *msg, int len) 1785 { 1786 return __zerocopy_sg_from_iter(skb->sk, skb, &msg->msg_iter, len); 1787 } 1788 1789 int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1790 struct msghdr *msg, int len, 1791 struct ubuf_info *uarg); 1792 1793 /* Internal */ 1794 #define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1795 1796 static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1797 { 1798 return &skb_shinfo(skb)->hwtstamps; 1799 } 1800 1801 static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1802 { 1803 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1804 1805 return is_zcopy ? skb_uarg(skb) : NULL; 1806 } 1807 1808 static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1809 { 1810 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1811 } 1812 1813 static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1814 const struct sk_buff *skb2) 1815 { 1816 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1817 } 1818 1819 static inline void net_zcopy_get(struct ubuf_info *uarg) 1820 { 1821 refcount_inc(&uarg->refcnt); 1822 } 1823 1824 static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1825 { 1826 skb_shinfo(skb)->destructor_arg = uarg; 1827 skb_shinfo(skb)->flags |= uarg->flags; 1828 } 1829 1830 static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1831 bool *have_ref) 1832 { 1833 if (skb && uarg && !skb_zcopy(skb)) { 1834 if (unlikely(have_ref && *have_ref)) 1835 *have_ref = false; 1836 else 1837 net_zcopy_get(uarg); 1838 skb_zcopy_init(skb, uarg); 1839 } 1840 } 1841 1842 static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1843 { 1844 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1845 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1846 } 1847 1848 static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1849 { 1850 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1851 } 1852 1853 static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1854 { 1855 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1856 } 1857 1858 static inline void net_zcopy_put(struct ubuf_info *uarg) 1859 { 1860 if (uarg) 1861 uarg->callback(NULL, uarg, true); 1862 } 1863 1864 static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1865 { 1866 if (uarg) { 1867 if (uarg->callback == msg_zerocopy_callback) 1868 msg_zerocopy_put_abort(uarg, have_uref); 1869 else if (have_uref) 1870 net_zcopy_put(uarg); 1871 } 1872 } 1873 1874 /* Release a reference on a zerocopy structure */ 1875 static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1876 { 1877 struct ubuf_info *uarg = skb_zcopy(skb); 1878 1879 if (uarg) { 1880 if (!skb_zcopy_is_nouarg(skb)) 1881 uarg->callback(skb, uarg, zerocopy_success); 1882 1883 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1884 } 1885 } 1886 1887 static inline void skb_mark_not_on_list(struct sk_buff *skb) 1888 { 1889 skb->next = NULL; 1890 } 1891 1892 /* Iterate through singly-linked GSO fragments of an skb. */ 1893 #define skb_list_walk_safe(first, skb, next_skb) \ 1894 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1895 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1896 1897 static inline void skb_list_del_init(struct sk_buff *skb) 1898 { 1899 __list_del_entry(&skb->list); 1900 skb_mark_not_on_list(skb); 1901 } 1902 1903 /** 1904 * skb_queue_empty - check if a queue is empty 1905 * @list: queue head 1906 * 1907 * Returns true if the queue is empty, false otherwise. 1908 */ 1909 static inline int skb_queue_empty(const struct sk_buff_head *list) 1910 { 1911 return list->next == (const struct sk_buff *) list; 1912 } 1913 1914 /** 1915 * skb_queue_empty_lockless - check if a queue is empty 1916 * @list: queue head 1917 * 1918 * Returns true if the queue is empty, false otherwise. 1919 * This variant can be used in lockless contexts. 1920 */ 1921 static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1922 { 1923 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1924 } 1925 1926 1927 /** 1928 * skb_queue_is_last - check if skb is the last entry in the queue 1929 * @list: queue head 1930 * @skb: buffer 1931 * 1932 * Returns true if @skb is the last buffer on the list. 1933 */ 1934 static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1935 const struct sk_buff *skb) 1936 { 1937 return skb->next == (const struct sk_buff *) list; 1938 } 1939 1940 /** 1941 * skb_queue_is_first - check if skb is the first entry in the queue 1942 * @list: queue head 1943 * @skb: buffer 1944 * 1945 * Returns true if @skb is the first buffer on the list. 1946 */ 1947 static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1948 const struct sk_buff *skb) 1949 { 1950 return skb->prev == (const struct sk_buff *) list; 1951 } 1952 1953 /** 1954 * skb_queue_next - return the next packet in the queue 1955 * @list: queue head 1956 * @skb: current buffer 1957 * 1958 * Return the next packet in @list after @skb. It is only valid to 1959 * call this if skb_queue_is_last() evaluates to false. 1960 */ 1961 static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1962 const struct sk_buff *skb) 1963 { 1964 /* This BUG_ON may seem severe, but if we just return then we 1965 * are going to dereference garbage. 1966 */ 1967 BUG_ON(skb_queue_is_last(list, skb)); 1968 return skb->next; 1969 } 1970 1971 /** 1972 * skb_queue_prev - return the prev packet in the queue 1973 * @list: queue head 1974 * @skb: current buffer 1975 * 1976 * Return the prev packet in @list before @skb. It is only valid to 1977 * call this if skb_queue_is_first() evaluates to false. 1978 */ 1979 static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1980 const struct sk_buff *skb) 1981 { 1982 /* This BUG_ON may seem severe, but if we just return then we 1983 * are going to dereference garbage. 1984 */ 1985 BUG_ON(skb_queue_is_first(list, skb)); 1986 return skb->prev; 1987 } 1988 1989 /** 1990 * skb_get - reference buffer 1991 * @skb: buffer to reference 1992 * 1993 * Makes another reference to a socket buffer and returns a pointer 1994 * to the buffer. 1995 */ 1996 static inline struct sk_buff *skb_get(struct sk_buff *skb) 1997 { 1998 refcount_inc(&skb->users); 1999 return skb; 2000 } 2001 2002 /* 2003 * If users == 1, we are the only owner and can avoid redundant atomic changes. 2004 */ 2005 2006 /** 2007 * skb_cloned - is the buffer a clone 2008 * @skb: buffer to check 2009 * 2010 * Returns true if the buffer was generated with skb_clone() and is 2011 * one of multiple shared copies of the buffer. Cloned buffers are 2012 * shared data so must not be written to under normal circumstances. 2013 */ 2014 static inline int skb_cloned(const struct sk_buff *skb) 2015 { 2016 return skb->cloned && 2017 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 2018 } 2019 2020 static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 2021 { 2022 might_sleep_if(gfpflags_allow_blocking(pri)); 2023 2024 if (skb_cloned(skb)) 2025 return pskb_expand_head(skb, 0, 0, pri); 2026 2027 return 0; 2028 } 2029 2030 /* This variant of skb_unclone() makes sure skb->truesize 2031 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 2032 * 2033 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 2034 * when various debugging features are in place. 2035 */ 2036 int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 2037 static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 2038 { 2039 might_sleep_if(gfpflags_allow_blocking(pri)); 2040 2041 if (skb_cloned(skb)) 2042 return __skb_unclone_keeptruesize(skb, pri); 2043 return 0; 2044 } 2045 2046 /** 2047 * skb_header_cloned - is the header a clone 2048 * @skb: buffer to check 2049 * 2050 * Returns true if modifying the header part of the buffer requires 2051 * the data to be copied. 2052 */ 2053 static inline int skb_header_cloned(const struct sk_buff *skb) 2054 { 2055 int dataref; 2056 2057 if (!skb->cloned) 2058 return 0; 2059 2060 dataref = atomic_read(&skb_shinfo(skb)->dataref); 2061 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 2062 return dataref != 1; 2063 } 2064 2065 static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 2066 { 2067 might_sleep_if(gfpflags_allow_blocking(pri)); 2068 2069 if (skb_header_cloned(skb)) 2070 return pskb_expand_head(skb, 0, 0, pri); 2071 2072 return 0; 2073 } 2074 2075 /** 2076 * __skb_header_release() - allow clones to use the headroom 2077 * @skb: buffer to operate on 2078 * 2079 * See "DOC: dataref and headerless skbs". 2080 */ 2081 static inline void __skb_header_release(struct sk_buff *skb) 2082 { 2083 skb->nohdr = 1; 2084 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 2085 } 2086 2087 2088 /** 2089 * skb_shared - is the buffer shared 2090 * @skb: buffer to check 2091 * 2092 * Returns true if more than one person has a reference to this 2093 * buffer. 2094 */ 2095 static inline int skb_shared(const struct sk_buff *skb) 2096 { 2097 return refcount_read(&skb->users) != 1; 2098 } 2099 2100 /** 2101 * skb_share_check - check if buffer is shared and if so clone it 2102 * @skb: buffer to check 2103 * @pri: priority for memory allocation 2104 * 2105 * If the buffer is shared the buffer is cloned and the old copy 2106 * drops a reference. A new clone with a single reference is returned. 2107 * If the buffer is not shared the original buffer is returned. When 2108 * being called from interrupt status or with spinlocks held pri must 2109 * be GFP_ATOMIC. 2110 * 2111 * NULL is returned on a memory allocation failure. 2112 */ 2113 static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2114 { 2115 might_sleep_if(gfpflags_allow_blocking(pri)); 2116 if (skb_shared(skb)) { 2117 struct sk_buff *nskb = skb_clone(skb, pri); 2118 2119 if (likely(nskb)) 2120 consume_skb(skb); 2121 else 2122 kfree_skb(skb); 2123 skb = nskb; 2124 } 2125 return skb; 2126 } 2127 2128 /* 2129 * Copy shared buffers into a new sk_buff. We effectively do COW on 2130 * packets to handle cases where we have a local reader and forward 2131 * and a couple of other messy ones. The normal one is tcpdumping 2132 * a packet thats being forwarded. 2133 */ 2134 2135 /** 2136 * skb_unshare - make a copy of a shared buffer 2137 * @skb: buffer to check 2138 * @pri: priority for memory allocation 2139 * 2140 * If the socket buffer is a clone then this function creates a new 2141 * copy of the data, drops a reference count on the old copy and returns 2142 * the new copy with the reference count at 1. If the buffer is not a clone 2143 * the original buffer is returned. When called with a spinlock held or 2144 * from interrupt state @pri must be %GFP_ATOMIC 2145 * 2146 * %NULL is returned on a memory allocation failure. 2147 */ 2148 static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2149 gfp_t pri) 2150 { 2151 might_sleep_if(gfpflags_allow_blocking(pri)); 2152 if (skb_cloned(skb)) { 2153 struct sk_buff *nskb = skb_copy(skb, pri); 2154 2155 /* Free our shared copy */ 2156 if (likely(nskb)) 2157 consume_skb(skb); 2158 else 2159 kfree_skb(skb); 2160 skb = nskb; 2161 } 2162 return skb; 2163 } 2164 2165 /** 2166 * skb_peek - peek at the head of an &sk_buff_head 2167 * @list_: list to peek at 2168 * 2169 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2170 * be careful with this one. A peek leaves the buffer on the 2171 * list and someone else may run off with it. You must hold 2172 * the appropriate locks or have a private queue to do this. 2173 * 2174 * Returns %NULL for an empty list or a pointer to the head element. 2175 * The reference count is not incremented and the reference is therefore 2176 * volatile. Use with caution. 2177 */ 2178 static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2179 { 2180 struct sk_buff *skb = list_->next; 2181 2182 if (skb == (struct sk_buff *)list_) 2183 skb = NULL; 2184 return skb; 2185 } 2186 2187 /** 2188 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2189 * @list_: list to peek at 2190 * 2191 * Like skb_peek(), but the caller knows that the list is not empty. 2192 */ 2193 static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2194 { 2195 return list_->next; 2196 } 2197 2198 /** 2199 * skb_peek_next - peek skb following the given one from a queue 2200 * @skb: skb to start from 2201 * @list_: list to peek at 2202 * 2203 * Returns %NULL when the end of the list is met or a pointer to the 2204 * next element. The reference count is not incremented and the 2205 * reference is therefore volatile. Use with caution. 2206 */ 2207 static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2208 const struct sk_buff_head *list_) 2209 { 2210 struct sk_buff *next = skb->next; 2211 2212 if (next == (struct sk_buff *)list_) 2213 next = NULL; 2214 return next; 2215 } 2216 2217 /** 2218 * skb_peek_tail - peek at the tail of an &sk_buff_head 2219 * @list_: list to peek at 2220 * 2221 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2222 * be careful with this one. A peek leaves the buffer on the 2223 * list and someone else may run off with it. You must hold 2224 * the appropriate locks or have a private queue to do this. 2225 * 2226 * Returns %NULL for an empty list or a pointer to the tail element. 2227 * The reference count is not incremented and the reference is therefore 2228 * volatile. Use with caution. 2229 */ 2230 static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2231 { 2232 struct sk_buff *skb = READ_ONCE(list_->prev); 2233 2234 if (skb == (struct sk_buff *)list_) 2235 skb = NULL; 2236 return skb; 2237 2238 } 2239 2240 /** 2241 * skb_queue_len - get queue length 2242 * @list_: list to measure 2243 * 2244 * Return the length of an &sk_buff queue. 2245 */ 2246 static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2247 { 2248 return list_->qlen; 2249 } 2250 2251 /** 2252 * skb_queue_len_lockless - get queue length 2253 * @list_: list to measure 2254 * 2255 * Return the length of an &sk_buff queue. 2256 * This variant can be used in lockless contexts. 2257 */ 2258 static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2259 { 2260 return READ_ONCE(list_->qlen); 2261 } 2262 2263 /** 2264 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2265 * @list: queue to initialize 2266 * 2267 * This initializes only the list and queue length aspects of 2268 * an sk_buff_head object. This allows to initialize the list 2269 * aspects of an sk_buff_head without reinitializing things like 2270 * the spinlock. It can also be used for on-stack sk_buff_head 2271 * objects where the spinlock is known to not be used. 2272 */ 2273 static inline void __skb_queue_head_init(struct sk_buff_head *list) 2274 { 2275 list->prev = list->next = (struct sk_buff *)list; 2276 list->qlen = 0; 2277 } 2278 2279 /* 2280 * This function creates a split out lock class for each invocation; 2281 * this is needed for now since a whole lot of users of the skb-queue 2282 * infrastructure in drivers have different locking usage (in hardirq) 2283 * than the networking core (in softirq only). In the long run either the 2284 * network layer or drivers should need annotation to consolidate the 2285 * main types of usage into 3 classes. 2286 */ 2287 static inline void skb_queue_head_init(struct sk_buff_head *list) 2288 { 2289 spin_lock_init(&list->lock); 2290 __skb_queue_head_init(list); 2291 } 2292 2293 static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2294 struct lock_class_key *class) 2295 { 2296 skb_queue_head_init(list); 2297 lockdep_set_class(&list->lock, class); 2298 } 2299 2300 /* 2301 * Insert an sk_buff on a list. 2302 * 2303 * The "__skb_xxxx()" functions are the non-atomic ones that 2304 * can only be called with interrupts disabled. 2305 */ 2306 static inline void __skb_insert(struct sk_buff *newsk, 2307 struct sk_buff *prev, struct sk_buff *next, 2308 struct sk_buff_head *list) 2309 { 2310 /* See skb_queue_empty_lockless() and skb_peek_tail() 2311 * for the opposite READ_ONCE() 2312 */ 2313 WRITE_ONCE(newsk->next, next); 2314 WRITE_ONCE(newsk->prev, prev); 2315 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2316 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2317 WRITE_ONCE(list->qlen, list->qlen + 1); 2318 } 2319 2320 static inline void __skb_queue_splice(const struct sk_buff_head *list, 2321 struct sk_buff *prev, 2322 struct sk_buff *next) 2323 { 2324 struct sk_buff *first = list->next; 2325 struct sk_buff *last = list->prev; 2326 2327 WRITE_ONCE(first->prev, prev); 2328 WRITE_ONCE(prev->next, first); 2329 2330 WRITE_ONCE(last->next, next); 2331 WRITE_ONCE(next->prev, last); 2332 } 2333 2334 /** 2335 * skb_queue_splice - join two skb lists, this is designed for stacks 2336 * @list: the new list to add 2337 * @head: the place to add it in the first list 2338 */ 2339 static inline void skb_queue_splice(const struct sk_buff_head *list, 2340 struct sk_buff_head *head) 2341 { 2342 if (!skb_queue_empty(list)) { 2343 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2344 head->qlen += list->qlen; 2345 } 2346 } 2347 2348 /** 2349 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2350 * @list: the new list to add 2351 * @head: the place to add it in the first list 2352 * 2353 * The list at @list is reinitialised 2354 */ 2355 static inline void skb_queue_splice_init(struct sk_buff_head *list, 2356 struct sk_buff_head *head) 2357 { 2358 if (!skb_queue_empty(list)) { 2359 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2360 head->qlen += list->qlen; 2361 __skb_queue_head_init(list); 2362 } 2363 } 2364 2365 /** 2366 * skb_queue_splice_tail - join two skb lists, each list being a queue 2367 * @list: the new list to add 2368 * @head: the place to add it in the first list 2369 */ 2370 static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2371 struct sk_buff_head *head) 2372 { 2373 if (!skb_queue_empty(list)) { 2374 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2375 head->qlen += list->qlen; 2376 } 2377 } 2378 2379 /** 2380 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2381 * @list: the new list to add 2382 * @head: the place to add it in the first list 2383 * 2384 * Each of the lists is a queue. 2385 * The list at @list is reinitialised 2386 */ 2387 static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2388 struct sk_buff_head *head) 2389 { 2390 if (!skb_queue_empty(list)) { 2391 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2392 head->qlen += list->qlen; 2393 __skb_queue_head_init(list); 2394 } 2395 } 2396 2397 /** 2398 * __skb_queue_after - queue a buffer at the list head 2399 * @list: list to use 2400 * @prev: place after this buffer 2401 * @newsk: buffer to queue 2402 * 2403 * Queue a buffer int the middle of a list. This function takes no locks 2404 * and you must therefore hold required locks before calling it. 2405 * 2406 * A buffer cannot be placed on two lists at the same time. 2407 */ 2408 static inline void __skb_queue_after(struct sk_buff_head *list, 2409 struct sk_buff *prev, 2410 struct sk_buff *newsk) 2411 { 2412 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2413 } 2414 2415 void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2416 struct sk_buff_head *list); 2417 2418 static inline void __skb_queue_before(struct sk_buff_head *list, 2419 struct sk_buff *next, 2420 struct sk_buff *newsk) 2421 { 2422 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2423 } 2424 2425 /** 2426 * __skb_queue_head - queue a buffer at the list head 2427 * @list: list to use 2428 * @newsk: buffer to queue 2429 * 2430 * Queue a buffer at the start of a list. This function takes no locks 2431 * and you must therefore hold required locks before calling it. 2432 * 2433 * A buffer cannot be placed on two lists at the same time. 2434 */ 2435 static inline void __skb_queue_head(struct sk_buff_head *list, 2436 struct sk_buff *newsk) 2437 { 2438 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2439 } 2440 void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2441 2442 /** 2443 * __skb_queue_tail - queue a buffer at the list tail 2444 * @list: list to use 2445 * @newsk: buffer to queue 2446 * 2447 * Queue a buffer at the end of a list. This function takes no locks 2448 * and you must therefore hold required locks before calling it. 2449 * 2450 * A buffer cannot be placed on two lists at the same time. 2451 */ 2452 static inline void __skb_queue_tail(struct sk_buff_head *list, 2453 struct sk_buff *newsk) 2454 { 2455 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2456 } 2457 void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2458 2459 /* 2460 * remove sk_buff from list. _Must_ be called atomically, and with 2461 * the list known.. 2462 */ 2463 void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2464 static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2465 { 2466 struct sk_buff *next, *prev; 2467 2468 WRITE_ONCE(list->qlen, list->qlen - 1); 2469 next = skb->next; 2470 prev = skb->prev; 2471 skb->next = skb->prev = NULL; 2472 WRITE_ONCE(next->prev, prev); 2473 WRITE_ONCE(prev->next, next); 2474 } 2475 2476 /** 2477 * __skb_dequeue - remove from the head of the queue 2478 * @list: list to dequeue from 2479 * 2480 * Remove the head of the list. This function does not take any locks 2481 * so must be used with appropriate locks held only. The head item is 2482 * returned or %NULL if the list is empty. 2483 */ 2484 static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2485 { 2486 struct sk_buff *skb = skb_peek(list); 2487 if (skb) 2488 __skb_unlink(skb, list); 2489 return skb; 2490 } 2491 struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2492 2493 /** 2494 * __skb_dequeue_tail - remove from the tail of the queue 2495 * @list: list to dequeue from 2496 * 2497 * Remove the tail of the list. This function does not take any locks 2498 * so must be used with appropriate locks held only. The tail item is 2499 * returned or %NULL if the list is empty. 2500 */ 2501 static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2502 { 2503 struct sk_buff *skb = skb_peek_tail(list); 2504 if (skb) 2505 __skb_unlink(skb, list); 2506 return skb; 2507 } 2508 struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2509 2510 2511 static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2512 { 2513 return skb->data_len; 2514 } 2515 2516 static inline unsigned int skb_headlen(const struct sk_buff *skb) 2517 { 2518 return skb->len - skb->data_len; 2519 } 2520 2521 static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2522 { 2523 unsigned int i, len = 0; 2524 2525 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2526 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2527 return len; 2528 } 2529 2530 static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2531 { 2532 return skb_headlen(skb) + __skb_pagelen(skb); 2533 } 2534 2535 /** 2536 * __skb_fill_page_desc - initialise a paged fragment in an skb 2537 * @skb: buffer containing fragment to be initialised 2538 * @i: paged fragment index to initialise 2539 * @page: the page to use for this fragment 2540 * @off: the offset to the data with @page 2541 * @size: the length of the data 2542 * 2543 * Initialises the @i'th fragment of @skb to point to &size bytes at 2544 * offset @off within @page. 2545 * 2546 * Does not take any additional reference on the fragment. 2547 */ 2548 static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2549 struct page *page, int off, int size) 2550 { 2551 skb_frag_t *frag = &skb_shinfo(skb)->frags[i]; 2552 2553 /* 2554 * Propagate page pfmemalloc to the skb if we can. The problem is 2555 * that not all callers have unique ownership of the page but rely 2556 * on page_is_pfmemalloc doing the right thing(tm). 2557 */ 2558 frag->bv_page = page; 2559 frag->bv_offset = off; 2560 skb_frag_size_set(frag, size); 2561 2562 page = compound_head(page); 2563 if (page_is_pfmemalloc(page)) 2564 skb->pfmemalloc = true; 2565 } 2566 2567 /** 2568 * skb_fill_page_desc - initialise a paged fragment in an skb 2569 * @skb: buffer containing fragment to be initialised 2570 * @i: paged fragment index to initialise 2571 * @page: the page to use for this fragment 2572 * @off: the offset to the data with @page 2573 * @size: the length of the data 2574 * 2575 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2576 * @skb to point to @size bytes at offset @off within @page. In 2577 * addition updates @skb such that @i is the last fragment. 2578 * 2579 * Does not take any additional reference on the fragment. 2580 */ 2581 static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2582 struct page *page, int off, int size) 2583 { 2584 __skb_fill_page_desc(skb, i, page, off, size); 2585 skb_shinfo(skb)->nr_frags = i + 1; 2586 } 2587 2588 void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 2589 int size, unsigned int truesize); 2590 2591 void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2592 unsigned int truesize); 2593 2594 #define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2595 2596 #ifdef NET_SKBUFF_DATA_USES_OFFSET 2597 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2598 { 2599 return skb->head + skb->tail; 2600 } 2601 2602 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2603 { 2604 skb->tail = skb->data - skb->head; 2605 } 2606 2607 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2608 { 2609 skb_reset_tail_pointer(skb); 2610 skb->tail += offset; 2611 } 2612 2613 #else /* NET_SKBUFF_DATA_USES_OFFSET */ 2614 static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2615 { 2616 return skb->tail; 2617 } 2618 2619 static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2620 { 2621 skb->tail = skb->data; 2622 } 2623 2624 static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2625 { 2626 skb->tail = skb->data + offset; 2627 } 2628 2629 #endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2630 2631 /* 2632 * Add data to an sk_buff 2633 */ 2634 void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2635 void *skb_put(struct sk_buff *skb, unsigned int len); 2636 static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2637 { 2638 void *tmp = skb_tail_pointer(skb); 2639 SKB_LINEAR_ASSERT(skb); 2640 skb->tail += len; 2641 skb->len += len; 2642 return tmp; 2643 } 2644 2645 static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2646 { 2647 void *tmp = __skb_put(skb, len); 2648 2649 memset(tmp, 0, len); 2650 return tmp; 2651 } 2652 2653 static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2654 unsigned int len) 2655 { 2656 void *tmp = __skb_put(skb, len); 2657 2658 memcpy(tmp, data, len); 2659 return tmp; 2660 } 2661 2662 static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2663 { 2664 *(u8 *)__skb_put(skb, 1) = val; 2665 } 2666 2667 static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2668 { 2669 void *tmp = skb_put(skb, len); 2670 2671 memset(tmp, 0, len); 2672 2673 return tmp; 2674 } 2675 2676 static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2677 unsigned int len) 2678 { 2679 void *tmp = skb_put(skb, len); 2680 2681 memcpy(tmp, data, len); 2682 2683 return tmp; 2684 } 2685 2686 static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2687 { 2688 *(u8 *)skb_put(skb, 1) = val; 2689 } 2690 2691 void *skb_push(struct sk_buff *skb, unsigned int len); 2692 static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2693 { 2694 skb->data -= len; 2695 skb->len += len; 2696 return skb->data; 2697 } 2698 2699 void *skb_pull(struct sk_buff *skb, unsigned int len); 2700 static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2701 { 2702 skb->len -= len; 2703 BUG_ON(skb->len < skb->data_len); 2704 return skb->data += len; 2705 } 2706 2707 static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2708 { 2709 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2710 } 2711 2712 void *skb_pull_data(struct sk_buff *skb, size_t len); 2713 2714 void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2715 2716 static inline void *__pskb_pull(struct sk_buff *skb, unsigned int len) 2717 { 2718 if (len > skb_headlen(skb) && 2719 !__pskb_pull_tail(skb, len - skb_headlen(skb))) 2720 return NULL; 2721 skb->len -= len; 2722 return skb->data += len; 2723 } 2724 2725 static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2726 { 2727 return unlikely(len > skb->len) ? NULL : __pskb_pull(skb, len); 2728 } 2729 2730 static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2731 { 2732 if (likely(len <= skb_headlen(skb))) 2733 return true; 2734 if (unlikely(len > skb->len)) 2735 return false; 2736 return __pskb_pull_tail(skb, len - skb_headlen(skb)) != NULL; 2737 } 2738 2739 void skb_condense(struct sk_buff *skb); 2740 2741 /** 2742 * skb_headroom - bytes at buffer head 2743 * @skb: buffer to check 2744 * 2745 * Return the number of bytes of free space at the head of an &sk_buff. 2746 */ 2747 static inline unsigned int skb_headroom(const struct sk_buff *skb) 2748 { 2749 return skb->data - skb->head; 2750 } 2751 2752 /** 2753 * skb_tailroom - bytes at buffer end 2754 * @skb: buffer to check 2755 * 2756 * Return the number of bytes of free space at the tail of an sk_buff 2757 */ 2758 static inline int skb_tailroom(const struct sk_buff *skb) 2759 { 2760 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2761 } 2762 2763 /** 2764 * skb_availroom - bytes at buffer end 2765 * @skb: buffer to check 2766 * 2767 * Return the number of bytes of free space at the tail of an sk_buff 2768 * allocated by sk_stream_alloc() 2769 */ 2770 static inline int skb_availroom(const struct sk_buff *skb) 2771 { 2772 if (skb_is_nonlinear(skb)) 2773 return 0; 2774 2775 return skb->end - skb->tail - skb->reserved_tailroom; 2776 } 2777 2778 /** 2779 * skb_reserve - adjust headroom 2780 * @skb: buffer to alter 2781 * @len: bytes to move 2782 * 2783 * Increase the headroom of an empty &sk_buff by reducing the tail 2784 * room. This is only allowed for an empty buffer. 2785 */ 2786 static inline void skb_reserve(struct sk_buff *skb, int len) 2787 { 2788 skb->data += len; 2789 skb->tail += len; 2790 } 2791 2792 /** 2793 * skb_tailroom_reserve - adjust reserved_tailroom 2794 * @skb: buffer to alter 2795 * @mtu: maximum amount of headlen permitted 2796 * @needed_tailroom: minimum amount of reserved_tailroom 2797 * 2798 * Set reserved_tailroom so that headlen can be as large as possible but 2799 * not larger than mtu and tailroom cannot be smaller than 2800 * needed_tailroom. 2801 * The required headroom should already have been reserved before using 2802 * this function. 2803 */ 2804 static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2805 unsigned int needed_tailroom) 2806 { 2807 SKB_LINEAR_ASSERT(skb); 2808 if (mtu < skb_tailroom(skb) - needed_tailroom) 2809 /* use at most mtu */ 2810 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2811 else 2812 /* use up to all available space */ 2813 skb->reserved_tailroom = needed_tailroom; 2814 } 2815 2816 #define ENCAP_TYPE_ETHER 0 2817 #define ENCAP_TYPE_IPPROTO 1 2818 2819 static inline void skb_set_inner_protocol(struct sk_buff *skb, 2820 __be16 protocol) 2821 { 2822 skb->inner_protocol = protocol; 2823 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2824 } 2825 2826 static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2827 __u8 ipproto) 2828 { 2829 skb->inner_ipproto = ipproto; 2830 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2831 } 2832 2833 static inline void skb_reset_inner_headers(struct sk_buff *skb) 2834 { 2835 skb->inner_mac_header = skb->mac_header; 2836 skb->inner_network_header = skb->network_header; 2837 skb->inner_transport_header = skb->transport_header; 2838 } 2839 2840 static inline void skb_reset_mac_len(struct sk_buff *skb) 2841 { 2842 skb->mac_len = skb->network_header - skb->mac_header; 2843 } 2844 2845 static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2846 *skb) 2847 { 2848 return skb->head + skb->inner_transport_header; 2849 } 2850 2851 static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2852 { 2853 return skb_inner_transport_header(skb) - skb->data; 2854 } 2855 2856 static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2857 { 2858 skb->inner_transport_header = skb->data - skb->head; 2859 } 2860 2861 static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2862 const int offset) 2863 { 2864 skb_reset_inner_transport_header(skb); 2865 skb->inner_transport_header += offset; 2866 } 2867 2868 static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2869 { 2870 return skb->head + skb->inner_network_header; 2871 } 2872 2873 static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2874 { 2875 skb->inner_network_header = skb->data - skb->head; 2876 } 2877 2878 static inline void skb_set_inner_network_header(struct sk_buff *skb, 2879 const int offset) 2880 { 2881 skb_reset_inner_network_header(skb); 2882 skb->inner_network_header += offset; 2883 } 2884 2885 static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2886 { 2887 return skb->head + skb->inner_mac_header; 2888 } 2889 2890 static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2891 { 2892 skb->inner_mac_header = skb->data - skb->head; 2893 } 2894 2895 static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2896 const int offset) 2897 { 2898 skb_reset_inner_mac_header(skb); 2899 skb->inner_mac_header += offset; 2900 } 2901 static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2902 { 2903 return skb->transport_header != (typeof(skb->transport_header))~0U; 2904 } 2905 2906 static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2907 { 2908 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 2909 return skb->head + skb->transport_header; 2910 } 2911 2912 static inline void skb_reset_transport_header(struct sk_buff *skb) 2913 { 2914 skb->transport_header = skb->data - skb->head; 2915 } 2916 2917 static inline void skb_set_transport_header(struct sk_buff *skb, 2918 const int offset) 2919 { 2920 skb_reset_transport_header(skb); 2921 skb->transport_header += offset; 2922 } 2923 2924 static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2925 { 2926 return skb->head + skb->network_header; 2927 } 2928 2929 static inline void skb_reset_network_header(struct sk_buff *skb) 2930 { 2931 skb->network_header = skb->data - skb->head; 2932 } 2933 2934 static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2935 { 2936 skb_reset_network_header(skb); 2937 skb->network_header += offset; 2938 } 2939 2940 static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2941 { 2942 return skb->head + skb->mac_header; 2943 } 2944 2945 static inline int skb_mac_offset(const struct sk_buff *skb) 2946 { 2947 return skb_mac_header(skb) - skb->data; 2948 } 2949 2950 static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2951 { 2952 return skb->network_header - skb->mac_header; 2953 } 2954 2955 static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2956 { 2957 return skb->mac_header != (typeof(skb->mac_header))~0U; 2958 } 2959 2960 static inline void skb_unset_mac_header(struct sk_buff *skb) 2961 { 2962 skb->mac_header = (typeof(skb->mac_header))~0U; 2963 } 2964 2965 static inline void skb_reset_mac_header(struct sk_buff *skb) 2966 { 2967 skb->mac_header = skb->data - skb->head; 2968 } 2969 2970 static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2971 { 2972 skb_reset_mac_header(skb); 2973 skb->mac_header += offset; 2974 } 2975 2976 static inline void skb_pop_mac_header(struct sk_buff *skb) 2977 { 2978 skb->mac_header = skb->network_header; 2979 } 2980 2981 static inline void skb_probe_transport_header(struct sk_buff *skb) 2982 { 2983 struct flow_keys_basic keys; 2984 2985 if (skb_transport_header_was_set(skb)) 2986 return; 2987 2988 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 2989 NULL, 0, 0, 0, 0)) 2990 skb_set_transport_header(skb, keys.control.thoff); 2991 } 2992 2993 static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2994 { 2995 if (skb_mac_header_was_set(skb)) { 2996 const unsigned char *old_mac = skb_mac_header(skb); 2997 2998 skb_set_mac_header(skb, -skb->mac_len); 2999 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 3000 } 3001 } 3002 3003 static inline int skb_checksum_start_offset(const struct sk_buff *skb) 3004 { 3005 return skb->csum_start - skb_headroom(skb); 3006 } 3007 3008 static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 3009 { 3010 return skb->head + skb->csum_start; 3011 } 3012 3013 static inline int skb_transport_offset(const struct sk_buff *skb) 3014 { 3015 return skb_transport_header(skb) - skb->data; 3016 } 3017 3018 static inline u32 skb_network_header_len(const struct sk_buff *skb) 3019 { 3020 return skb->transport_header - skb->network_header; 3021 } 3022 3023 static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 3024 { 3025 return skb->inner_transport_header - skb->inner_network_header; 3026 } 3027 3028 static inline int skb_network_offset(const struct sk_buff *skb) 3029 { 3030 return skb_network_header(skb) - skb->data; 3031 } 3032 3033 static inline int skb_inner_network_offset(const struct sk_buff *skb) 3034 { 3035 return skb_inner_network_header(skb) - skb->data; 3036 } 3037 3038 static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 3039 { 3040 return pskb_may_pull(skb, skb_network_offset(skb) + len); 3041 } 3042 3043 /* 3044 * CPUs often take a performance hit when accessing unaligned memory 3045 * locations. The actual performance hit varies, it can be small if the 3046 * hardware handles it or large if we have to take an exception and fix it 3047 * in software. 3048 * 3049 * Since an ethernet header is 14 bytes network drivers often end up with 3050 * the IP header at an unaligned offset. The IP header can be aligned by 3051 * shifting the start of the packet by 2 bytes. Drivers should do this 3052 * with: 3053 * 3054 * skb_reserve(skb, NET_IP_ALIGN); 3055 * 3056 * The downside to this alignment of the IP header is that the DMA is now 3057 * unaligned. On some architectures the cost of an unaligned DMA is high 3058 * and this cost outweighs the gains made by aligning the IP header. 3059 * 3060 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3061 * to be overridden. 3062 */ 3063 #ifndef NET_IP_ALIGN 3064 #define NET_IP_ALIGN 2 3065 #endif 3066 3067 /* 3068 * The networking layer reserves some headroom in skb data (via 3069 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3070 * the header has to grow. In the default case, if the header has to grow 3071 * 32 bytes or less we avoid the reallocation. 3072 * 3073 * Unfortunately this headroom changes the DMA alignment of the resulting 3074 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3075 * on some architectures. An architecture can override this value, 3076 * perhaps setting it to a cacheline in size (since that will maintain 3077 * cacheline alignment of the DMA). It must be a power of 2. 3078 * 3079 * Various parts of the networking layer expect at least 32 bytes of 3080 * headroom, you should not reduce this. 3081 * 3082 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3083 * to reduce average number of cache lines per packet. 3084 * get_rps_cpu() for example only access one 64 bytes aligned block : 3085 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3086 */ 3087 #ifndef NET_SKB_PAD 3088 #define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3089 #endif 3090 3091 int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3092 3093 static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3094 { 3095 if (WARN_ON(skb_is_nonlinear(skb))) 3096 return; 3097 skb->len = len; 3098 skb_set_tail_pointer(skb, len); 3099 } 3100 3101 static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3102 { 3103 __skb_set_length(skb, len); 3104 } 3105 3106 void skb_trim(struct sk_buff *skb, unsigned int len); 3107 3108 static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3109 { 3110 if (skb->data_len) 3111 return ___pskb_trim(skb, len); 3112 __skb_trim(skb, len); 3113 return 0; 3114 } 3115 3116 static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3117 { 3118 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3119 } 3120 3121 /** 3122 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3123 * @skb: buffer to alter 3124 * @len: new length 3125 * 3126 * This is identical to pskb_trim except that the caller knows that 3127 * the skb is not cloned so we should never get an error due to out- 3128 * of-memory. 3129 */ 3130 static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3131 { 3132 int err = pskb_trim(skb, len); 3133 BUG_ON(err); 3134 } 3135 3136 static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3137 { 3138 unsigned int diff = len - skb->len; 3139 3140 if (skb_tailroom(skb) < diff) { 3141 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3142 GFP_ATOMIC); 3143 if (ret) 3144 return ret; 3145 } 3146 __skb_set_length(skb, len); 3147 return 0; 3148 } 3149 3150 /** 3151 * skb_orphan - orphan a buffer 3152 * @skb: buffer to orphan 3153 * 3154 * If a buffer currently has an owner then we call the owner's 3155 * destructor function and make the @skb unowned. The buffer continues 3156 * to exist but is no longer charged to its former owner. 3157 */ 3158 static inline void skb_orphan(struct sk_buff *skb) 3159 { 3160 if (skb->destructor) { 3161 skb->destructor(skb); 3162 skb->destructor = NULL; 3163 skb->sk = NULL; 3164 } else { 3165 BUG_ON(skb->sk); 3166 } 3167 } 3168 3169 /** 3170 * skb_orphan_frags - orphan the frags contained in a buffer 3171 * @skb: buffer to orphan frags from 3172 * @gfp_mask: allocation mask for replacement pages 3173 * 3174 * For each frag in the SKB which needs a destructor (i.e. has an 3175 * owner) create a copy of that frag and release the original 3176 * page by calling the destructor. 3177 */ 3178 static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3179 { 3180 if (likely(!skb_zcopy(skb))) 3181 return 0; 3182 if (!skb_zcopy_is_nouarg(skb) && 3183 skb_uarg(skb)->callback == msg_zerocopy_callback) 3184 return 0; 3185 return skb_copy_ubufs(skb, gfp_mask); 3186 } 3187 3188 /* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3189 static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3190 { 3191 if (likely(!skb_zcopy(skb))) 3192 return 0; 3193 return skb_copy_ubufs(skb, gfp_mask); 3194 } 3195 3196 /** 3197 * __skb_queue_purge - empty a list 3198 * @list: list to empty 3199 * 3200 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3201 * the list and one reference dropped. This function does not take the 3202 * list lock and the caller must hold the relevant locks to use it. 3203 */ 3204 static inline void __skb_queue_purge(struct sk_buff_head *list) 3205 { 3206 struct sk_buff *skb; 3207 while ((skb = __skb_dequeue(list)) != NULL) 3208 kfree_skb(skb); 3209 } 3210 void skb_queue_purge(struct sk_buff_head *list); 3211 3212 unsigned int skb_rbtree_purge(struct rb_root *root); 3213 3214 void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3215 3216 /** 3217 * netdev_alloc_frag - allocate a page fragment 3218 * @fragsz: fragment size 3219 * 3220 * Allocates a frag from a page for receive buffer. 3221 * Uses GFP_ATOMIC allocations. 3222 */ 3223 static inline void *netdev_alloc_frag(unsigned int fragsz) 3224 { 3225 return __netdev_alloc_frag_align(fragsz, ~0u); 3226 } 3227 3228 static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3229 unsigned int align) 3230 { 3231 WARN_ON_ONCE(!is_power_of_2(align)); 3232 return __netdev_alloc_frag_align(fragsz, -align); 3233 } 3234 3235 struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3236 gfp_t gfp_mask); 3237 3238 /** 3239 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3240 * @dev: network device to receive on 3241 * @length: length to allocate 3242 * 3243 * Allocate a new &sk_buff and assign it a usage count of one. The 3244 * buffer has unspecified headroom built in. Users should allocate 3245 * the headroom they think they need without accounting for the 3246 * built in space. The built in space is used for optimisations. 3247 * 3248 * %NULL is returned if there is no free memory. Although this function 3249 * allocates memory it can be called from an interrupt. 3250 */ 3251 static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3252 unsigned int length) 3253 { 3254 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3255 } 3256 3257 /* legacy helper around __netdev_alloc_skb() */ 3258 static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3259 gfp_t gfp_mask) 3260 { 3261 return __netdev_alloc_skb(NULL, length, gfp_mask); 3262 } 3263 3264 /* legacy helper around netdev_alloc_skb() */ 3265 static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3266 { 3267 return netdev_alloc_skb(NULL, length); 3268 } 3269 3270 3271 static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3272 unsigned int length, gfp_t gfp) 3273 { 3274 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3275 3276 if (NET_IP_ALIGN && skb) 3277 skb_reserve(skb, NET_IP_ALIGN); 3278 return skb; 3279 } 3280 3281 static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3282 unsigned int length) 3283 { 3284 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3285 } 3286 3287 static inline void skb_free_frag(void *addr) 3288 { 3289 page_frag_free(addr); 3290 } 3291 3292 void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3293 3294 static inline void *napi_alloc_frag(unsigned int fragsz) 3295 { 3296 return __napi_alloc_frag_align(fragsz, ~0u); 3297 } 3298 3299 static inline void *napi_alloc_frag_align(unsigned int fragsz, 3300 unsigned int align) 3301 { 3302 WARN_ON_ONCE(!is_power_of_2(align)); 3303 return __napi_alloc_frag_align(fragsz, -align); 3304 } 3305 3306 struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 3307 unsigned int length, gfp_t gfp_mask); 3308 static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 3309 unsigned int length) 3310 { 3311 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 3312 } 3313 void napi_consume_skb(struct sk_buff *skb, int budget); 3314 3315 void napi_skb_free_stolen_head(struct sk_buff *skb); 3316 void __kfree_skb_defer(struct sk_buff *skb); 3317 3318 /** 3319 * __dev_alloc_pages - allocate page for network Rx 3320 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3321 * @order: size of the allocation 3322 * 3323 * Allocate a new page. 3324 * 3325 * %NULL is returned if there is no free memory. 3326 */ 3327 static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 3328 unsigned int order) 3329 { 3330 /* This piece of code contains several assumptions. 3331 * 1. This is for device Rx, therefor a cold page is preferred. 3332 * 2. The expectation is the user wants a compound page. 3333 * 3. If requesting a order 0 page it will not be compound 3334 * due to the check to see if order has a value in prep_new_page 3335 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3336 * code in gfp_to_alloc_flags that should be enforcing this. 3337 */ 3338 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3339 3340 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 3341 } 3342 3343 static inline struct page *dev_alloc_pages(unsigned int order) 3344 { 3345 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 3346 } 3347 3348 /** 3349 * __dev_alloc_page - allocate a page for network Rx 3350 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3351 * 3352 * Allocate a new page. 3353 * 3354 * %NULL is returned if there is no free memory. 3355 */ 3356 static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 3357 { 3358 return __dev_alloc_pages(gfp_mask, 0); 3359 } 3360 3361 static inline struct page *dev_alloc_page(void) 3362 { 3363 return dev_alloc_pages(0); 3364 } 3365 3366 /** 3367 * dev_page_is_reusable - check whether a page can be reused for network Rx 3368 * @page: the page to test 3369 * 3370 * A page shouldn't be considered for reusing/recycling if it was allocated 3371 * under memory pressure or at a distant memory node. 3372 * 3373 * Returns false if this page should be returned to page allocator, true 3374 * otherwise. 3375 */ 3376 static inline bool dev_page_is_reusable(const struct page *page) 3377 { 3378 return likely(page_to_nid(page) == numa_mem_id() && 3379 !page_is_pfmemalloc(page)); 3380 } 3381 3382 /** 3383 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3384 * @page: The page that was allocated from skb_alloc_page 3385 * @skb: The skb that may need pfmemalloc set 3386 */ 3387 static inline void skb_propagate_pfmemalloc(const struct page *page, 3388 struct sk_buff *skb) 3389 { 3390 if (page_is_pfmemalloc(page)) 3391 skb->pfmemalloc = true; 3392 } 3393 3394 /** 3395 * skb_frag_off() - Returns the offset of a skb fragment 3396 * @frag: the paged fragment 3397 */ 3398 static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3399 { 3400 return frag->bv_offset; 3401 } 3402 3403 /** 3404 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3405 * @frag: skb fragment 3406 * @delta: value to add 3407 */ 3408 static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3409 { 3410 frag->bv_offset += delta; 3411 } 3412 3413 /** 3414 * skb_frag_off_set() - Sets the offset of a skb fragment 3415 * @frag: skb fragment 3416 * @offset: offset of fragment 3417 */ 3418 static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3419 { 3420 frag->bv_offset = offset; 3421 } 3422 3423 /** 3424 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3425 * @fragto: skb fragment where offset is set 3426 * @fragfrom: skb fragment offset is copied from 3427 */ 3428 static inline void skb_frag_off_copy(skb_frag_t *fragto, 3429 const skb_frag_t *fragfrom) 3430 { 3431 fragto->bv_offset = fragfrom->bv_offset; 3432 } 3433 3434 /** 3435 * skb_frag_page - retrieve the page referred to by a paged fragment 3436 * @frag: the paged fragment 3437 * 3438 * Returns the &struct page associated with @frag. 3439 */ 3440 static inline struct page *skb_frag_page(const skb_frag_t *frag) 3441 { 3442 return frag->bv_page; 3443 } 3444 3445 /** 3446 * __skb_frag_ref - take an addition reference on a paged fragment. 3447 * @frag: the paged fragment 3448 * 3449 * Takes an additional reference on the paged fragment @frag. 3450 */ 3451 static inline void __skb_frag_ref(skb_frag_t *frag) 3452 { 3453 get_page(skb_frag_page(frag)); 3454 } 3455 3456 /** 3457 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 3458 * @skb: the buffer 3459 * @f: the fragment offset. 3460 * 3461 * Takes an additional reference on the @f'th paged fragment of @skb. 3462 */ 3463 static inline void skb_frag_ref(struct sk_buff *skb, int f) 3464 { 3465 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 3466 } 3467 3468 /** 3469 * __skb_frag_unref - release a reference on a paged fragment. 3470 * @frag: the paged fragment 3471 * @recycle: recycle the page if allocated via page_pool 3472 * 3473 * Releases a reference on the paged fragment @frag 3474 * or recycles the page via the page_pool API. 3475 */ 3476 static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle) 3477 { 3478 struct page *page = skb_frag_page(frag); 3479 3480 #ifdef CONFIG_PAGE_POOL 3481 if (recycle && page_pool_return_skb_page(page)) 3482 return; 3483 #endif 3484 put_page(page); 3485 } 3486 3487 /** 3488 * skb_frag_unref - release a reference on a paged fragment of an skb. 3489 * @skb: the buffer 3490 * @f: the fragment offset 3491 * 3492 * Releases a reference on the @f'th paged fragment of @skb. 3493 */ 3494 static inline void skb_frag_unref(struct sk_buff *skb, int f) 3495 { 3496 __skb_frag_unref(&skb_shinfo(skb)->frags[f], skb->pp_recycle); 3497 } 3498 3499 /** 3500 * skb_frag_address - gets the address of the data contained in a paged fragment 3501 * @frag: the paged fragment buffer 3502 * 3503 * Returns the address of the data within @frag. The page must already 3504 * be mapped. 3505 */ 3506 static inline void *skb_frag_address(const skb_frag_t *frag) 3507 { 3508 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3509 } 3510 3511 /** 3512 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3513 * @frag: the paged fragment buffer 3514 * 3515 * Returns the address of the data within @frag. Checks that the page 3516 * is mapped and returns %NULL otherwise. 3517 */ 3518 static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3519 { 3520 void *ptr = page_address(skb_frag_page(frag)); 3521 if (unlikely(!ptr)) 3522 return NULL; 3523 3524 return ptr + skb_frag_off(frag); 3525 } 3526 3527 /** 3528 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3529 * @fragto: skb fragment where page is set 3530 * @fragfrom: skb fragment page is copied from 3531 */ 3532 static inline void skb_frag_page_copy(skb_frag_t *fragto, 3533 const skb_frag_t *fragfrom) 3534 { 3535 fragto->bv_page = fragfrom->bv_page; 3536 } 3537 3538 /** 3539 * __skb_frag_set_page - sets the page contained in a paged fragment 3540 * @frag: the paged fragment 3541 * @page: the page to set 3542 * 3543 * Sets the fragment @frag to contain @page. 3544 */ 3545 static inline void __skb_frag_set_page(skb_frag_t *frag, struct page *page) 3546 { 3547 frag->bv_page = page; 3548 } 3549 3550 /** 3551 * skb_frag_set_page - sets the page contained in a paged fragment of an skb 3552 * @skb: the buffer 3553 * @f: the fragment offset 3554 * @page: the page to set 3555 * 3556 * Sets the @f'th fragment of @skb to contain @page. 3557 */ 3558 static inline void skb_frag_set_page(struct sk_buff *skb, int f, 3559 struct page *page) 3560 { 3561 __skb_frag_set_page(&skb_shinfo(skb)->frags[f], page); 3562 } 3563 3564 bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3565 3566 /** 3567 * skb_frag_dma_map - maps a paged fragment via the DMA API 3568 * @dev: the device to map the fragment to 3569 * @frag: the paged fragment to map 3570 * @offset: the offset within the fragment (starting at the 3571 * fragment's own offset) 3572 * @size: the number of bytes to map 3573 * @dir: the direction of the mapping (``PCI_DMA_*``) 3574 * 3575 * Maps the page associated with @frag to @device. 3576 */ 3577 static inline dma_addr_t skb_frag_dma_map(struct device *dev, 3578 const skb_frag_t *frag, 3579 size_t offset, size_t size, 3580 enum dma_data_direction dir) 3581 { 3582 return dma_map_page(dev, skb_frag_page(frag), 3583 skb_frag_off(frag) + offset, size, dir); 3584 } 3585 3586 static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3587 gfp_t gfp_mask) 3588 { 3589 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3590 } 3591 3592 3593 static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3594 gfp_t gfp_mask) 3595 { 3596 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3597 } 3598 3599 3600 /** 3601 * skb_clone_writable - is the header of a clone writable 3602 * @skb: buffer to check 3603 * @len: length up to which to write 3604 * 3605 * Returns true if modifying the header part of the cloned buffer 3606 * does not requires the data to be copied. 3607 */ 3608 static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3609 { 3610 return !skb_header_cloned(skb) && 3611 skb_headroom(skb) + len <= skb->hdr_len; 3612 } 3613 3614 static inline int skb_try_make_writable(struct sk_buff *skb, 3615 unsigned int write_len) 3616 { 3617 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3618 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3619 } 3620 3621 static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3622 int cloned) 3623 { 3624 int delta = 0; 3625 3626 if (headroom > skb_headroom(skb)) 3627 delta = headroom - skb_headroom(skb); 3628 3629 if (delta || cloned) 3630 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3631 GFP_ATOMIC); 3632 return 0; 3633 } 3634 3635 /** 3636 * skb_cow - copy header of skb when it is required 3637 * @skb: buffer to cow 3638 * @headroom: needed headroom 3639 * 3640 * If the skb passed lacks sufficient headroom or its data part 3641 * is shared, data is reallocated. If reallocation fails, an error 3642 * is returned and original skb is not changed. 3643 * 3644 * The result is skb with writable area skb->head...skb->tail 3645 * and at least @headroom of space at head. 3646 */ 3647 static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3648 { 3649 return __skb_cow(skb, headroom, skb_cloned(skb)); 3650 } 3651 3652 /** 3653 * skb_cow_head - skb_cow but only making the head writable 3654 * @skb: buffer to cow 3655 * @headroom: needed headroom 3656 * 3657 * This function is identical to skb_cow except that we replace the 3658 * skb_cloned check by skb_header_cloned. It should be used when 3659 * you only need to push on some header and do not need to modify 3660 * the data. 3661 */ 3662 static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3663 { 3664 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3665 } 3666 3667 /** 3668 * skb_padto - pad an skbuff up to a minimal size 3669 * @skb: buffer to pad 3670 * @len: minimal length 3671 * 3672 * Pads up a buffer to ensure the trailing bytes exist and are 3673 * blanked. If the buffer already contains sufficient data it 3674 * is untouched. Otherwise it is extended. Returns zero on 3675 * success. The skb is freed on error. 3676 */ 3677 static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3678 { 3679 unsigned int size = skb->len; 3680 if (likely(size >= len)) 3681 return 0; 3682 return skb_pad(skb, len - size); 3683 } 3684 3685 /** 3686 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3687 * @skb: buffer to pad 3688 * @len: minimal length 3689 * @free_on_error: free buffer on error 3690 * 3691 * Pads up a buffer to ensure the trailing bytes exist and are 3692 * blanked. If the buffer already contains sufficient data it 3693 * is untouched. Otherwise it is extended. Returns zero on 3694 * success. The skb is freed on error if @free_on_error is true. 3695 */ 3696 static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3697 unsigned int len, 3698 bool free_on_error) 3699 { 3700 unsigned int size = skb->len; 3701 3702 if (unlikely(size < len)) { 3703 len -= size; 3704 if (__skb_pad(skb, len, free_on_error)) 3705 return -ENOMEM; 3706 __skb_put(skb, len); 3707 } 3708 return 0; 3709 } 3710 3711 /** 3712 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3713 * @skb: buffer to pad 3714 * @len: minimal length 3715 * 3716 * Pads up a buffer to ensure the trailing bytes exist and are 3717 * blanked. If the buffer already contains sufficient data it 3718 * is untouched. Otherwise it is extended. Returns zero on 3719 * success. The skb is freed on error. 3720 */ 3721 static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3722 { 3723 return __skb_put_padto(skb, len, true); 3724 } 3725 3726 static inline int skb_add_data(struct sk_buff *skb, 3727 struct iov_iter *from, int copy) 3728 { 3729 const int off = skb->len; 3730 3731 if (skb->ip_summed == CHECKSUM_NONE) { 3732 __wsum csum = 0; 3733 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3734 &csum, from)) { 3735 skb->csum = csum_block_add(skb->csum, csum, off); 3736 return 0; 3737 } 3738 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3739 return 0; 3740 3741 __skb_trim(skb, off); 3742 return -EFAULT; 3743 } 3744 3745 static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3746 const struct page *page, int off) 3747 { 3748 if (skb_zcopy(skb)) 3749 return false; 3750 if (i) { 3751 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3752 3753 return page == skb_frag_page(frag) && 3754 off == skb_frag_off(frag) + skb_frag_size(frag); 3755 } 3756 return false; 3757 } 3758 3759 static inline int __skb_linearize(struct sk_buff *skb) 3760 { 3761 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3762 } 3763 3764 /** 3765 * skb_linearize - convert paged skb to linear one 3766 * @skb: buffer to linarize 3767 * 3768 * If there is no free memory -ENOMEM is returned, otherwise zero 3769 * is returned and the old skb data released. 3770 */ 3771 static inline int skb_linearize(struct sk_buff *skb) 3772 { 3773 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3774 } 3775 3776 /** 3777 * skb_has_shared_frag - can any frag be overwritten 3778 * @skb: buffer to test 3779 * 3780 * Return true if the skb has at least one frag that might be modified 3781 * by an external entity (as in vmsplice()/sendfile()) 3782 */ 3783 static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3784 { 3785 return skb_is_nonlinear(skb) && 3786 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3787 } 3788 3789 /** 3790 * skb_linearize_cow - make sure skb is linear and writable 3791 * @skb: buffer to process 3792 * 3793 * If there is no free memory -ENOMEM is returned, otherwise zero 3794 * is returned and the old skb data released. 3795 */ 3796 static inline int skb_linearize_cow(struct sk_buff *skb) 3797 { 3798 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3799 __skb_linearize(skb) : 0; 3800 } 3801 3802 static __always_inline void 3803 __skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3804 unsigned int off) 3805 { 3806 if (skb->ip_summed == CHECKSUM_COMPLETE) 3807 skb->csum = csum_block_sub(skb->csum, 3808 csum_partial(start, len, 0), off); 3809 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3810 skb_checksum_start_offset(skb) < 0) 3811 skb->ip_summed = CHECKSUM_NONE; 3812 } 3813 3814 /** 3815 * skb_postpull_rcsum - update checksum for received skb after pull 3816 * @skb: buffer to update 3817 * @start: start of data before pull 3818 * @len: length of data pulled 3819 * 3820 * After doing a pull on a received packet, you need to call this to 3821 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3822 * CHECKSUM_NONE so that it can be recomputed from scratch. 3823 */ 3824 static inline void skb_postpull_rcsum(struct sk_buff *skb, 3825 const void *start, unsigned int len) 3826 { 3827 if (skb->ip_summed == CHECKSUM_COMPLETE) 3828 skb->csum = wsum_negate(csum_partial(start, len, 3829 wsum_negate(skb->csum))); 3830 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3831 skb_checksum_start_offset(skb) < 0) 3832 skb->ip_summed = CHECKSUM_NONE; 3833 } 3834 3835 static __always_inline void 3836 __skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3837 unsigned int off) 3838 { 3839 if (skb->ip_summed == CHECKSUM_COMPLETE) 3840 skb->csum = csum_block_add(skb->csum, 3841 csum_partial(start, len, 0), off); 3842 } 3843 3844 /** 3845 * skb_postpush_rcsum - update checksum for received skb after push 3846 * @skb: buffer to update 3847 * @start: start of data after push 3848 * @len: length of data pushed 3849 * 3850 * After doing a push on a received packet, you need to call this to 3851 * update the CHECKSUM_COMPLETE checksum. 3852 */ 3853 static inline void skb_postpush_rcsum(struct sk_buff *skb, 3854 const void *start, unsigned int len) 3855 { 3856 __skb_postpush_rcsum(skb, start, len, 0); 3857 } 3858 3859 void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3860 3861 /** 3862 * skb_push_rcsum - push skb and update receive checksum 3863 * @skb: buffer to update 3864 * @len: length of data pulled 3865 * 3866 * This function performs an skb_push on the packet and updates 3867 * the CHECKSUM_COMPLETE checksum. It should be used on 3868 * receive path processing instead of skb_push unless you know 3869 * that the checksum difference is zero (e.g., a valid IP header) 3870 * or you are setting ip_summed to CHECKSUM_NONE. 3871 */ 3872 static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3873 { 3874 skb_push(skb, len); 3875 skb_postpush_rcsum(skb, skb->data, len); 3876 return skb->data; 3877 } 3878 3879 int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3880 /** 3881 * pskb_trim_rcsum - trim received skb and update checksum 3882 * @skb: buffer to trim 3883 * @len: new length 3884 * 3885 * This is exactly the same as pskb_trim except that it ensures the 3886 * checksum of received packets are still valid after the operation. 3887 * It can change skb pointers. 3888 */ 3889 3890 static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3891 { 3892 if (likely(len >= skb->len)) 3893 return 0; 3894 return pskb_trim_rcsum_slow(skb, len); 3895 } 3896 3897 static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3898 { 3899 if (skb->ip_summed == CHECKSUM_COMPLETE) 3900 skb->ip_summed = CHECKSUM_NONE; 3901 __skb_trim(skb, len); 3902 return 0; 3903 } 3904 3905 static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3906 { 3907 if (skb->ip_summed == CHECKSUM_COMPLETE) 3908 skb->ip_summed = CHECKSUM_NONE; 3909 return __skb_grow(skb, len); 3910 } 3911 3912 #define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3913 #define skb_rb_first(root) rb_to_skb(rb_first(root)) 3914 #define skb_rb_last(root) rb_to_skb(rb_last(root)) 3915 #define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3916 #define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3917 3918 #define skb_queue_walk(queue, skb) \ 3919 for (skb = (queue)->next; \ 3920 skb != (struct sk_buff *)(queue); \ 3921 skb = skb->next) 3922 3923 #define skb_queue_walk_safe(queue, skb, tmp) \ 3924 for (skb = (queue)->next, tmp = skb->next; \ 3925 skb != (struct sk_buff *)(queue); \ 3926 skb = tmp, tmp = skb->next) 3927 3928 #define skb_queue_walk_from(queue, skb) \ 3929 for (; skb != (struct sk_buff *)(queue); \ 3930 skb = skb->next) 3931 3932 #define skb_rbtree_walk(skb, root) \ 3933 for (skb = skb_rb_first(root); skb != NULL; \ 3934 skb = skb_rb_next(skb)) 3935 3936 #define skb_rbtree_walk_from(skb) \ 3937 for (; skb != NULL; \ 3938 skb = skb_rb_next(skb)) 3939 3940 #define skb_rbtree_walk_from_safe(skb, tmp) \ 3941 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3942 skb = tmp) 3943 3944 #define skb_queue_walk_from_safe(queue, skb, tmp) \ 3945 for (tmp = skb->next; \ 3946 skb != (struct sk_buff *)(queue); \ 3947 skb = tmp, tmp = skb->next) 3948 3949 #define skb_queue_reverse_walk(queue, skb) \ 3950 for (skb = (queue)->prev; \ 3951 skb != (struct sk_buff *)(queue); \ 3952 skb = skb->prev) 3953 3954 #define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3955 for (skb = (queue)->prev, tmp = skb->prev; \ 3956 skb != (struct sk_buff *)(queue); \ 3957 skb = tmp, tmp = skb->prev) 3958 3959 #define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3960 for (tmp = skb->prev; \ 3961 skb != (struct sk_buff *)(queue); \ 3962 skb = tmp, tmp = skb->prev) 3963 3964 static inline bool skb_has_frag_list(const struct sk_buff *skb) 3965 { 3966 return skb_shinfo(skb)->frag_list != NULL; 3967 } 3968 3969 static inline void skb_frag_list_init(struct sk_buff *skb) 3970 { 3971 skb_shinfo(skb)->frag_list = NULL; 3972 } 3973 3974 #define skb_walk_frags(skb, iter) \ 3975 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3976 3977 3978 int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 3979 int *err, long *timeo_p, 3980 const struct sk_buff *skb); 3981 struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3982 struct sk_buff_head *queue, 3983 unsigned int flags, 3984 int *off, int *err, 3985 struct sk_buff **last); 3986 struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 3987 struct sk_buff_head *queue, 3988 unsigned int flags, int *off, int *err, 3989 struct sk_buff **last); 3990 struct sk_buff *__skb_recv_datagram(struct sock *sk, 3991 struct sk_buff_head *sk_queue, 3992 unsigned int flags, int *off, int *err); 3993 struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 3994 __poll_t datagram_poll(struct file *file, struct socket *sock, 3995 struct poll_table_struct *wait); 3996 int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3997 struct iov_iter *to, int size); 3998 static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3999 struct msghdr *msg, int size) 4000 { 4001 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 4002 } 4003 int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 4004 struct msghdr *msg); 4005 int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 4006 struct iov_iter *to, int len, 4007 struct ahash_request *hash); 4008 int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 4009 struct iov_iter *from, int len); 4010 int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 4011 void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 4012 void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 4013 static inline void skb_free_datagram_locked(struct sock *sk, 4014 struct sk_buff *skb) 4015 { 4016 __skb_free_datagram_locked(sk, skb, 0); 4017 } 4018 int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 4019 int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 4020 int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 4021 __wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 4022 int len); 4023 int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 4024 struct pipe_inode_info *pipe, unsigned int len, 4025 unsigned int flags); 4026 int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 4027 int len); 4028 int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 4029 void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 4030 unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 4031 int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 4032 int len, int hlen); 4033 void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 4034 int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 4035 void skb_scrub_packet(struct sk_buff *skb, bool xnet); 4036 bool skb_gso_validate_network_len(const struct sk_buff *skb, unsigned int mtu); 4037 bool skb_gso_validate_mac_len(const struct sk_buff *skb, unsigned int len); 4038 struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 4039 struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 4040 unsigned int offset); 4041 struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 4042 int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 4043 int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 4044 int skb_vlan_pop(struct sk_buff *skb); 4045 int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 4046 int skb_eth_pop(struct sk_buff *skb); 4047 int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 4048 const unsigned char *src); 4049 int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 4050 int mac_len, bool ethernet); 4051 int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 4052 bool ethernet); 4053 int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 4054 int skb_mpls_dec_ttl(struct sk_buff *skb); 4055 struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 4056 gfp_t gfp); 4057 4058 static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 4059 { 4060 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4061 } 4062 4063 static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4064 { 4065 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4066 } 4067 4068 struct skb_checksum_ops { 4069 __wsum (*update)(const void *mem, int len, __wsum wsum); 4070 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 4071 }; 4072 4073 extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 4074 4075 __wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 4076 __wsum csum, const struct skb_checksum_ops *ops); 4077 __wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4078 __wsum csum); 4079 4080 static inline void * __must_check 4081 __skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4082 const void *data, int hlen, void *buffer) 4083 { 4084 if (likely(hlen - offset >= len)) 4085 return (void *)data + offset; 4086 4087 if (!skb || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4088 return NULL; 4089 4090 return buffer; 4091 } 4092 4093 static inline void * __must_check 4094 skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4095 { 4096 return __skb_header_pointer(skb, offset, len, skb->data, 4097 skb_headlen(skb), buffer); 4098 } 4099 4100 /** 4101 * skb_needs_linearize - check if we need to linearize a given skb 4102 * depending on the given device features. 4103 * @skb: socket buffer to check 4104 * @features: net device features 4105 * 4106 * Returns true if either: 4107 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4108 * 2. skb is fragmented and the device does not support SG. 4109 */ 4110 static inline bool skb_needs_linearize(struct sk_buff *skb, 4111 netdev_features_t features) 4112 { 4113 return skb_is_nonlinear(skb) && 4114 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4115 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4116 } 4117 4118 static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4119 void *to, 4120 const unsigned int len) 4121 { 4122 memcpy(to, skb->data, len); 4123 } 4124 4125 static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4126 const int offset, void *to, 4127 const unsigned int len) 4128 { 4129 memcpy(to, skb->data + offset, len); 4130 } 4131 4132 static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4133 const void *from, 4134 const unsigned int len) 4135 { 4136 memcpy(skb->data, from, len); 4137 } 4138 4139 static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4140 const int offset, 4141 const void *from, 4142 const unsigned int len) 4143 { 4144 memcpy(skb->data + offset, from, len); 4145 } 4146 4147 void skb_init(void); 4148 4149 static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4150 { 4151 return skb->tstamp; 4152 } 4153 4154 /** 4155 * skb_get_timestamp - get timestamp from a skb 4156 * @skb: skb to get stamp from 4157 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4158 * 4159 * Timestamps are stored in the skb as offsets to a base timestamp. 4160 * This function converts the offset back to a struct timeval and stores 4161 * it in stamp. 4162 */ 4163 static inline void skb_get_timestamp(const struct sk_buff *skb, 4164 struct __kernel_old_timeval *stamp) 4165 { 4166 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4167 } 4168 4169 static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4170 struct __kernel_sock_timeval *stamp) 4171 { 4172 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4173 4174 stamp->tv_sec = ts.tv_sec; 4175 stamp->tv_usec = ts.tv_nsec / 1000; 4176 } 4177 4178 static inline void skb_get_timestampns(const struct sk_buff *skb, 4179 struct __kernel_old_timespec *stamp) 4180 { 4181 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4182 4183 stamp->tv_sec = ts.tv_sec; 4184 stamp->tv_nsec = ts.tv_nsec; 4185 } 4186 4187 static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4188 struct __kernel_timespec *stamp) 4189 { 4190 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4191 4192 stamp->tv_sec = ts.tv_sec; 4193 stamp->tv_nsec = ts.tv_nsec; 4194 } 4195 4196 static inline void __net_timestamp(struct sk_buff *skb) 4197 { 4198 skb->tstamp = ktime_get_real(); 4199 skb->mono_delivery_time = 0; 4200 } 4201 4202 static inline ktime_t net_timedelta(ktime_t t) 4203 { 4204 return ktime_sub(ktime_get_real(), t); 4205 } 4206 4207 static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4208 bool mono) 4209 { 4210 skb->tstamp = kt; 4211 skb->mono_delivery_time = kt && mono; 4212 } 4213 4214 DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4215 4216 /* It is used in the ingress path to clear the delivery_time. 4217 * If needed, set the skb->tstamp to the (rcv) timestamp. 4218 */ 4219 static inline void skb_clear_delivery_time(struct sk_buff *skb) 4220 { 4221 if (skb->mono_delivery_time) { 4222 skb->mono_delivery_time = 0; 4223 if (static_branch_unlikely(&netstamp_needed_key)) 4224 skb->tstamp = ktime_get_real(); 4225 else 4226 skb->tstamp = 0; 4227 } 4228 } 4229 4230 static inline void skb_clear_tstamp(struct sk_buff *skb) 4231 { 4232 if (skb->mono_delivery_time) 4233 return; 4234 4235 skb->tstamp = 0; 4236 } 4237 4238 static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4239 { 4240 if (skb->mono_delivery_time) 4241 return 0; 4242 4243 return skb->tstamp; 4244 } 4245 4246 static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4247 { 4248 if (!skb->mono_delivery_time && skb->tstamp) 4249 return skb->tstamp; 4250 4251 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4252 return ktime_get_real(); 4253 4254 return 0; 4255 } 4256 4257 static inline u8 skb_metadata_len(const struct sk_buff *skb) 4258 { 4259 return skb_shinfo(skb)->meta_len; 4260 } 4261 4262 static inline void *skb_metadata_end(const struct sk_buff *skb) 4263 { 4264 return skb_mac_header(skb); 4265 } 4266 4267 static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4268 const struct sk_buff *skb_b, 4269 u8 meta_len) 4270 { 4271 const void *a = skb_metadata_end(skb_a); 4272 const void *b = skb_metadata_end(skb_b); 4273 /* Using more efficient varaiant than plain call to memcmp(). */ 4274 #if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64 4275 u64 diffs = 0; 4276 4277 switch (meta_len) { 4278 #define __it(x, op) (x -= sizeof(u##op)) 4279 #define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4280 case 32: diffs |= __it_diff(a, b, 64); 4281 fallthrough; 4282 case 24: diffs |= __it_diff(a, b, 64); 4283 fallthrough; 4284 case 16: diffs |= __it_diff(a, b, 64); 4285 fallthrough; 4286 case 8: diffs |= __it_diff(a, b, 64); 4287 break; 4288 case 28: diffs |= __it_diff(a, b, 64); 4289 fallthrough; 4290 case 20: diffs |= __it_diff(a, b, 64); 4291 fallthrough; 4292 case 12: diffs |= __it_diff(a, b, 64); 4293 fallthrough; 4294 case 4: diffs |= __it_diff(a, b, 32); 4295 break; 4296 } 4297 return diffs; 4298 #else 4299 return memcmp(a - meta_len, b - meta_len, meta_len); 4300 #endif 4301 } 4302 4303 static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4304 const struct sk_buff *skb_b) 4305 { 4306 u8 len_a = skb_metadata_len(skb_a); 4307 u8 len_b = skb_metadata_len(skb_b); 4308 4309 if (!(len_a | len_b)) 4310 return false; 4311 4312 return len_a != len_b ? 4313 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4314 } 4315 4316 static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4317 { 4318 skb_shinfo(skb)->meta_len = meta_len; 4319 } 4320 4321 static inline void skb_metadata_clear(struct sk_buff *skb) 4322 { 4323 skb_metadata_set(skb, 0); 4324 } 4325 4326 struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4327 4328 #ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4329 4330 void skb_clone_tx_timestamp(struct sk_buff *skb); 4331 bool skb_defer_rx_timestamp(struct sk_buff *skb); 4332 4333 #else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4334 4335 static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4336 { 4337 } 4338 4339 static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4340 { 4341 return false; 4342 } 4343 4344 #endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4345 4346 /** 4347 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4348 * 4349 * PHY drivers may accept clones of transmitted packets for 4350 * timestamping via their phy_driver.txtstamp method. These drivers 4351 * must call this function to return the skb back to the stack with a 4352 * timestamp. 4353 * 4354 * @skb: clone of the original outgoing packet 4355 * @hwtstamps: hardware time stamps 4356 * 4357 */ 4358 void skb_complete_tx_timestamp(struct sk_buff *skb, 4359 struct skb_shared_hwtstamps *hwtstamps); 4360 4361 void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4362 struct skb_shared_hwtstamps *hwtstamps, 4363 struct sock *sk, int tstype); 4364 4365 /** 4366 * skb_tstamp_tx - queue clone of skb with send time stamps 4367 * @orig_skb: the original outgoing packet 4368 * @hwtstamps: hardware time stamps, may be NULL if not available 4369 * 4370 * If the skb has a socket associated, then this function clones the 4371 * skb (thus sharing the actual data and optional structures), stores 4372 * the optional hardware time stamping information (if non NULL) or 4373 * generates a software time stamp (otherwise), then queues the clone 4374 * to the error queue of the socket. Errors are silently ignored. 4375 */ 4376 void skb_tstamp_tx(struct sk_buff *orig_skb, 4377 struct skb_shared_hwtstamps *hwtstamps); 4378 4379 /** 4380 * skb_tx_timestamp() - Driver hook for transmit timestamping 4381 * 4382 * Ethernet MAC Drivers should call this function in their hard_xmit() 4383 * function immediately before giving the sk_buff to the MAC hardware. 4384 * 4385 * Specifically, one should make absolutely sure that this function is 4386 * called before TX completion of this packet can trigger. Otherwise 4387 * the packet could potentially already be freed. 4388 * 4389 * @skb: A socket buffer. 4390 */ 4391 static inline void skb_tx_timestamp(struct sk_buff *skb) 4392 { 4393 skb_clone_tx_timestamp(skb); 4394 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 4395 skb_tstamp_tx(skb, NULL); 4396 } 4397 4398 /** 4399 * skb_complete_wifi_ack - deliver skb with wifi status 4400 * 4401 * @skb: the original outgoing packet 4402 * @acked: ack status 4403 * 4404 */ 4405 void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4406 4407 __sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4408 __sum16 __skb_checksum_complete(struct sk_buff *skb); 4409 4410 static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4411 { 4412 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4413 skb->csum_valid || 4414 (skb->ip_summed == CHECKSUM_PARTIAL && 4415 skb_checksum_start_offset(skb) >= 0)); 4416 } 4417 4418 /** 4419 * skb_checksum_complete - Calculate checksum of an entire packet 4420 * @skb: packet to process 4421 * 4422 * This function calculates the checksum over the entire packet plus 4423 * the value of skb->csum. The latter can be used to supply the 4424 * checksum of a pseudo header as used by TCP/UDP. It returns the 4425 * checksum. 4426 * 4427 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4428 * this function can be used to verify that checksum on received 4429 * packets. In that case the function should return zero if the 4430 * checksum is correct. In particular, this function will return zero 4431 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4432 * hardware has already verified the correctness of the checksum. 4433 */ 4434 static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4435 { 4436 return skb_csum_unnecessary(skb) ? 4437 0 : __skb_checksum_complete(skb); 4438 } 4439 4440 static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4441 { 4442 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4443 if (skb->csum_level == 0) 4444 skb->ip_summed = CHECKSUM_NONE; 4445 else 4446 skb->csum_level--; 4447 } 4448 } 4449 4450 static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4451 { 4452 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4453 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4454 skb->csum_level++; 4455 } else if (skb->ip_summed == CHECKSUM_NONE) { 4456 skb->ip_summed = CHECKSUM_UNNECESSARY; 4457 skb->csum_level = 0; 4458 } 4459 } 4460 4461 static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4462 { 4463 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4464 skb->ip_summed = CHECKSUM_NONE; 4465 skb->csum_level = 0; 4466 } 4467 } 4468 4469 /* Check if we need to perform checksum complete validation. 4470 * 4471 * Returns true if checksum complete is needed, false otherwise 4472 * (either checksum is unnecessary or zero checksum is allowed). 4473 */ 4474 static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4475 bool zero_okay, 4476 __sum16 check) 4477 { 4478 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4479 skb->csum_valid = 1; 4480 __skb_decr_checksum_unnecessary(skb); 4481 return false; 4482 } 4483 4484 return true; 4485 } 4486 4487 /* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4488 * in checksum_init. 4489 */ 4490 #define CHECKSUM_BREAK 76 4491 4492 /* Unset checksum-complete 4493 * 4494 * Unset checksum complete can be done when packet is being modified 4495 * (uncompressed for instance) and checksum-complete value is 4496 * invalidated. 4497 */ 4498 static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4499 { 4500 if (skb->ip_summed == CHECKSUM_COMPLETE) 4501 skb->ip_summed = CHECKSUM_NONE; 4502 } 4503 4504 /* Validate (init) checksum based on checksum complete. 4505 * 4506 * Return values: 4507 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4508 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4509 * checksum is stored in skb->csum for use in __skb_checksum_complete 4510 * non-zero: value of invalid checksum 4511 * 4512 */ 4513 static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4514 bool complete, 4515 __wsum psum) 4516 { 4517 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4518 if (!csum_fold(csum_add(psum, skb->csum))) { 4519 skb->csum_valid = 1; 4520 return 0; 4521 } 4522 } 4523 4524 skb->csum = psum; 4525 4526 if (complete || skb->len <= CHECKSUM_BREAK) { 4527 __sum16 csum; 4528 4529 csum = __skb_checksum_complete(skb); 4530 skb->csum_valid = !csum; 4531 return csum; 4532 } 4533 4534 return 0; 4535 } 4536 4537 static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4538 { 4539 return 0; 4540 } 4541 4542 /* Perform checksum validate (init). Note that this is a macro since we only 4543 * want to calculate the pseudo header which is an input function if necessary. 4544 * First we try to validate without any computation (checksum unnecessary) and 4545 * then calculate based on checksum complete calling the function to compute 4546 * pseudo header. 4547 * 4548 * Return values: 4549 * 0: checksum is validated or try to in skb_checksum_complete 4550 * non-zero: value of invalid checksum 4551 */ 4552 #define __skb_checksum_validate(skb, proto, complete, \ 4553 zero_okay, check, compute_pseudo) \ 4554 ({ \ 4555 __sum16 __ret = 0; \ 4556 skb->csum_valid = 0; \ 4557 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4558 __ret = __skb_checksum_validate_complete(skb, \ 4559 complete, compute_pseudo(skb, proto)); \ 4560 __ret; \ 4561 }) 4562 4563 #define skb_checksum_init(skb, proto, compute_pseudo) \ 4564 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4565 4566 #define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4567 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4568 4569 #define skb_checksum_validate(skb, proto, compute_pseudo) \ 4570 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4571 4572 #define skb_checksum_validate_zero_check(skb, proto, check, \ 4573 compute_pseudo) \ 4574 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4575 4576 #define skb_checksum_simple_validate(skb) \ 4577 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4578 4579 static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4580 { 4581 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4582 } 4583 4584 static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4585 { 4586 skb->csum = ~pseudo; 4587 skb->ip_summed = CHECKSUM_COMPLETE; 4588 } 4589 4590 #define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4591 do { \ 4592 if (__skb_checksum_convert_check(skb)) \ 4593 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4594 } while (0) 4595 4596 static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4597 u16 start, u16 offset) 4598 { 4599 skb->ip_summed = CHECKSUM_PARTIAL; 4600 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4601 skb->csum_offset = offset - start; 4602 } 4603 4604 /* Update skbuf and packet to reflect the remote checksum offload operation. 4605 * When called, ptr indicates the starting point for skb->csum when 4606 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4607 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4608 */ 4609 static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4610 int start, int offset, bool nopartial) 4611 { 4612 __wsum delta; 4613 4614 if (!nopartial) { 4615 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4616 return; 4617 } 4618 4619 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4620 __skb_checksum_complete(skb); 4621 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4622 } 4623 4624 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4625 4626 /* Adjust skb->csum since we changed the packet */ 4627 skb->csum = csum_add(skb->csum, delta); 4628 } 4629 4630 static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4631 { 4632 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4633 return (void *)(skb->_nfct & NFCT_PTRMASK); 4634 #else 4635 return NULL; 4636 #endif 4637 } 4638 4639 static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4640 { 4641 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4642 return skb->_nfct; 4643 #else 4644 return 0UL; 4645 #endif 4646 } 4647 4648 static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4649 { 4650 #if IS_ENABLED(CONFIG_NF_CONNTRACK) 4651 skb->slow_gro |= !!nfct; 4652 skb->_nfct = nfct; 4653 #endif 4654 } 4655 4656 #ifdef CONFIG_SKB_EXTENSIONS 4657 enum skb_ext_id { 4658 #if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4659 SKB_EXT_BRIDGE_NF, 4660 #endif 4661 #ifdef CONFIG_XFRM 4662 SKB_EXT_SEC_PATH, 4663 #endif 4664 #if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4665 TC_SKB_EXT, 4666 #endif 4667 #if IS_ENABLED(CONFIG_MPTCP) 4668 SKB_EXT_MPTCP, 4669 #endif 4670 #if IS_ENABLED(CONFIG_MCTP_FLOWS) 4671 SKB_EXT_MCTP, 4672 #endif 4673 SKB_EXT_NUM, /* must be last */ 4674 }; 4675 4676 /** 4677 * struct skb_ext - sk_buff extensions 4678 * @refcnt: 1 on allocation, deallocated on 0 4679 * @offset: offset to add to @data to obtain extension address 4680 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4681 * @data: start of extension data, variable sized 4682 * 4683 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4684 * to use 'u8' types while allowing up to 2kb worth of extension data. 4685 */ 4686 struct skb_ext { 4687 refcount_t refcnt; 4688 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4689 u8 chunks; /* same */ 4690 char data[] __aligned(8); 4691 }; 4692 4693 struct skb_ext *__skb_ext_alloc(gfp_t flags); 4694 void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4695 struct skb_ext *ext); 4696 void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4697 void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4698 void __skb_ext_put(struct skb_ext *ext); 4699 4700 static inline void skb_ext_put(struct sk_buff *skb) 4701 { 4702 if (skb->active_extensions) 4703 __skb_ext_put(skb->extensions); 4704 } 4705 4706 static inline void __skb_ext_copy(struct sk_buff *dst, 4707 const struct sk_buff *src) 4708 { 4709 dst->active_extensions = src->active_extensions; 4710 4711 if (src->active_extensions) { 4712 struct skb_ext *ext = src->extensions; 4713 4714 refcount_inc(&ext->refcnt); 4715 dst->extensions = ext; 4716 } 4717 } 4718 4719 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4720 { 4721 skb_ext_put(dst); 4722 __skb_ext_copy(dst, src); 4723 } 4724 4725 static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4726 { 4727 return !!ext->offset[i]; 4728 } 4729 4730 static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4731 { 4732 return skb->active_extensions & (1 << id); 4733 } 4734 4735 static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4736 { 4737 if (skb_ext_exist(skb, id)) 4738 __skb_ext_del(skb, id); 4739 } 4740 4741 static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4742 { 4743 if (skb_ext_exist(skb, id)) { 4744 struct skb_ext *ext = skb->extensions; 4745 4746 return (void *)ext + (ext->offset[id] << 3); 4747 } 4748 4749 return NULL; 4750 } 4751 4752 static inline void skb_ext_reset(struct sk_buff *skb) 4753 { 4754 if (unlikely(skb->active_extensions)) { 4755 __skb_ext_put(skb->extensions); 4756 skb->active_extensions = 0; 4757 } 4758 } 4759 4760 static inline bool skb_has_extensions(struct sk_buff *skb) 4761 { 4762 return unlikely(skb->active_extensions); 4763 } 4764 #else 4765 static inline void skb_ext_put(struct sk_buff *skb) {} 4766 static inline void skb_ext_reset(struct sk_buff *skb) {} 4767 static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4768 static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4769 static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4770 static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4771 #endif /* CONFIG_SKB_EXTENSIONS */ 4772 4773 static inline void nf_reset_ct(struct sk_buff *skb) 4774 { 4775 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4776 nf_conntrack_put(skb_nfct(skb)); 4777 skb->_nfct = 0; 4778 #endif 4779 } 4780 4781 static inline void nf_reset_trace(struct sk_buff *skb) 4782 { 4783 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4784 skb->nf_trace = 0; 4785 #endif 4786 } 4787 4788 static inline void ipvs_reset(struct sk_buff *skb) 4789 { 4790 #if IS_ENABLED(CONFIG_IP_VS) 4791 skb->ipvs_property = 0; 4792 #endif 4793 } 4794 4795 /* Note: This doesn't put any conntrack info in dst. */ 4796 static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4797 bool copy) 4798 { 4799 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4800 dst->_nfct = src->_nfct; 4801 nf_conntrack_get(skb_nfct(src)); 4802 #endif 4803 #if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || defined(CONFIG_NF_TABLES) 4804 if (copy) 4805 dst->nf_trace = src->nf_trace; 4806 #endif 4807 } 4808 4809 static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4810 { 4811 #if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4812 nf_conntrack_put(skb_nfct(dst)); 4813 #endif 4814 dst->slow_gro = src->slow_gro; 4815 __nf_copy(dst, src, true); 4816 } 4817 4818 #ifdef CONFIG_NETWORK_SECMARK 4819 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4820 { 4821 to->secmark = from->secmark; 4822 } 4823 4824 static inline void skb_init_secmark(struct sk_buff *skb) 4825 { 4826 skb->secmark = 0; 4827 } 4828 #else 4829 static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4830 { } 4831 4832 static inline void skb_init_secmark(struct sk_buff *skb) 4833 { } 4834 #endif 4835 4836 static inline int secpath_exists(const struct sk_buff *skb) 4837 { 4838 #ifdef CONFIG_XFRM 4839 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4840 #else 4841 return 0; 4842 #endif 4843 } 4844 4845 static inline bool skb_irq_freeable(const struct sk_buff *skb) 4846 { 4847 return !skb->destructor && 4848 !secpath_exists(skb) && 4849 !skb_nfct(skb) && 4850 !skb->_skb_refdst && 4851 !skb_has_frag_list(skb); 4852 } 4853 4854 static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 4855 { 4856 skb->queue_mapping = queue_mapping; 4857 } 4858 4859 static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 4860 { 4861 return skb->queue_mapping; 4862 } 4863 4864 static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 4865 { 4866 to->queue_mapping = from->queue_mapping; 4867 } 4868 4869 static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 4870 { 4871 skb->queue_mapping = rx_queue + 1; 4872 } 4873 4874 static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 4875 { 4876 return skb->queue_mapping - 1; 4877 } 4878 4879 static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 4880 { 4881 return skb->queue_mapping != 0; 4882 } 4883 4884 static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 4885 { 4886 skb->dst_pending_confirm = val; 4887 } 4888 4889 static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 4890 { 4891 return skb->dst_pending_confirm != 0; 4892 } 4893 4894 static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 4895 { 4896 #ifdef CONFIG_XFRM 4897 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 4898 #else 4899 return NULL; 4900 #endif 4901 } 4902 4903 /* Keeps track of mac header offset relative to skb->head. 4904 * It is useful for TSO of Tunneling protocol. e.g. GRE. 4905 * For non-tunnel skb it points to skb_mac_header() and for 4906 * tunnel skb it points to outer mac header. 4907 * Keeps track of level of encapsulation of network headers. 4908 */ 4909 struct skb_gso_cb { 4910 union { 4911 int mac_offset; 4912 int data_offset; 4913 }; 4914 int encap_level; 4915 __wsum csum; 4916 __u16 csum_start; 4917 }; 4918 #define SKB_GSO_CB_OFFSET 32 4919 #define SKB_GSO_CB(skb) ((struct skb_gso_cb *)((skb)->cb + SKB_GSO_CB_OFFSET)) 4920 4921 static inline int skb_tnl_header_len(const struct sk_buff *inner_skb) 4922 { 4923 return (skb_mac_header(inner_skb) - inner_skb->head) - 4924 SKB_GSO_CB(inner_skb)->mac_offset; 4925 } 4926 4927 static inline int gso_pskb_expand_head(struct sk_buff *skb, int extra) 4928 { 4929 int new_headroom, headroom; 4930 int ret; 4931 4932 headroom = skb_headroom(skb); 4933 ret = pskb_expand_head(skb, extra, 0, GFP_ATOMIC); 4934 if (ret) 4935 return ret; 4936 4937 new_headroom = skb_headroom(skb); 4938 SKB_GSO_CB(skb)->mac_offset += (new_headroom - headroom); 4939 return 0; 4940 } 4941 4942 static inline void gso_reset_checksum(struct sk_buff *skb, __wsum res) 4943 { 4944 /* Do not update partial checksums if remote checksum is enabled. */ 4945 if (skb->remcsum_offload) 4946 return; 4947 4948 SKB_GSO_CB(skb)->csum = res; 4949 SKB_GSO_CB(skb)->csum_start = skb_checksum_start(skb) - skb->head; 4950 } 4951 4952 /* Compute the checksum for a gso segment. First compute the checksum value 4953 * from the start of transport header to SKB_GSO_CB(skb)->csum_start, and 4954 * then add in skb->csum (checksum from csum_start to end of packet). 4955 * skb->csum and csum_start are then updated to reflect the checksum of the 4956 * resultant packet starting from the transport header-- the resultant checksum 4957 * is in the res argument (i.e. normally zero or ~ of checksum of a pseudo 4958 * header. 4959 */ 4960 static inline __sum16 gso_make_checksum(struct sk_buff *skb, __wsum res) 4961 { 4962 unsigned char *csum_start = skb_transport_header(skb); 4963 int plen = (skb->head + SKB_GSO_CB(skb)->csum_start) - csum_start; 4964 __wsum partial = SKB_GSO_CB(skb)->csum; 4965 4966 SKB_GSO_CB(skb)->csum = res; 4967 SKB_GSO_CB(skb)->csum_start = csum_start - skb->head; 4968 4969 return csum_fold(csum_partial(csum_start, plen, partial)); 4970 } 4971 4972 static inline bool skb_is_gso(const struct sk_buff *skb) 4973 { 4974 return skb_shinfo(skb)->gso_size; 4975 } 4976 4977 /* Note: Should be called only if skb_is_gso(skb) is true */ 4978 static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4979 { 4980 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4981 } 4982 4983 /* Note: Should be called only if skb_is_gso(skb) is true */ 4984 static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4985 { 4986 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4987 } 4988 4989 /* Note: Should be called only if skb_is_gso(skb) is true */ 4990 static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 4991 { 4992 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 4993 } 4994 4995 static inline void skb_gso_reset(struct sk_buff *skb) 4996 { 4997 skb_shinfo(skb)->gso_size = 0; 4998 skb_shinfo(skb)->gso_segs = 0; 4999 skb_shinfo(skb)->gso_type = 0; 5000 } 5001 5002 static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 5003 u16 increment) 5004 { 5005 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5006 return; 5007 shinfo->gso_size += increment; 5008 } 5009 5010 static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 5011 u16 decrement) 5012 { 5013 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 5014 return; 5015 shinfo->gso_size -= decrement; 5016 } 5017 5018 void __skb_warn_lro_forwarding(const struct sk_buff *skb); 5019 5020 static inline bool skb_warn_if_lro(const struct sk_buff *skb) 5021 { 5022 /* LRO sets gso_size but not gso_type, whereas if GSO is really 5023 * wanted then gso_type will be set. */ 5024 const struct skb_shared_info *shinfo = skb_shinfo(skb); 5025 5026 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 5027 unlikely(shinfo->gso_type == 0)) { 5028 __skb_warn_lro_forwarding(skb); 5029 return true; 5030 } 5031 return false; 5032 } 5033 5034 static inline void skb_forward_csum(struct sk_buff *skb) 5035 { 5036 /* Unfortunately we don't support this one. Any brave souls? */ 5037 if (skb->ip_summed == CHECKSUM_COMPLETE) 5038 skb->ip_summed = CHECKSUM_NONE; 5039 } 5040 5041 /** 5042 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 5043 * @skb: skb to check 5044 * 5045 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 5046 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 5047 * use this helper, to document places where we make this assertion. 5048 */ 5049 static inline void skb_checksum_none_assert(const struct sk_buff *skb) 5050 { 5051 #ifdef DEBUG 5052 BUG_ON(skb->ip_summed != CHECKSUM_NONE); 5053 #endif 5054 } 5055 5056 bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 5057 5058 int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 5059 struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 5060 unsigned int transport_len, 5061 __sum16(*skb_chkf)(struct sk_buff *skb)); 5062 5063 /** 5064 * skb_head_is_locked - Determine if the skb->head is locked down 5065 * @skb: skb to check 5066 * 5067 * The head on skbs build around a head frag can be removed if they are 5068 * not cloned. This function returns true if the skb head is locked down 5069 * due to either being allocated via kmalloc, or by being a clone with 5070 * multiple references to the head. 5071 */ 5072 static inline bool skb_head_is_locked(const struct sk_buff *skb) 5073 { 5074 return !skb->head_frag || skb_cloned(skb); 5075 } 5076 5077 /* Local Checksum Offload. 5078 * Compute outer checksum based on the assumption that the 5079 * inner checksum will be offloaded later. 5080 * See Documentation/networking/checksum-offloads.rst for 5081 * explanation of how this works. 5082 * Fill in outer checksum adjustment (e.g. with sum of outer 5083 * pseudo-header) before calling. 5084 * Also ensure that inner checksum is in linear data area. 5085 */ 5086 static inline __wsum lco_csum(struct sk_buff *skb) 5087 { 5088 unsigned char *csum_start = skb_checksum_start(skb); 5089 unsigned char *l4_hdr = skb_transport_header(skb); 5090 __wsum partial; 5091 5092 /* Start with complement of inner checksum adjustment */ 5093 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 5094 skb->csum_offset)); 5095 5096 /* Add in checksum of our headers (incl. outer checksum 5097 * adjustment filled in by caller) and return result. 5098 */ 5099 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 5100 } 5101 5102 static inline bool skb_is_redirected(const struct sk_buff *skb) 5103 { 5104 return skb->redirected; 5105 } 5106 5107 static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 5108 { 5109 skb->redirected = 1; 5110 #ifdef CONFIG_NET_REDIRECT 5111 skb->from_ingress = from_ingress; 5112 if (skb->from_ingress) 5113 skb_clear_tstamp(skb); 5114 #endif 5115 } 5116 5117 static inline void skb_reset_redirect(struct sk_buff *skb) 5118 { 5119 skb->redirected = 0; 5120 } 5121 5122 static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5123 { 5124 return skb->csum_not_inet; 5125 } 5126 5127 static inline void skb_set_kcov_handle(struct sk_buff *skb, 5128 const u64 kcov_handle) 5129 { 5130 #ifdef CONFIG_KCOV 5131 skb->kcov_handle = kcov_handle; 5132 #endif 5133 } 5134 5135 static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5136 { 5137 #ifdef CONFIG_KCOV 5138 return skb->kcov_handle; 5139 #else 5140 return 0; 5141 #endif 5142 } 5143 5144 #ifdef CONFIG_PAGE_POOL 5145 static inline void skb_mark_for_recycle(struct sk_buff *skb) 5146 { 5147 skb->pp_recycle = 1; 5148 } 5149 #endif 5150 5151 static inline bool skb_pp_recycle(struct sk_buff *skb, void *data) 5152 { 5153 if (!IS_ENABLED(CONFIG_PAGE_POOL) || !skb->pp_recycle) 5154 return false; 5155 return page_pool_return_skb_page(virt_to_page(data)); 5156 } 5157 5158 #endif /* __KERNEL__ */ 5159 #endif /* _LINUX_SKBUFF_H */ 5160