at v6.5 145 kB view raw
1/* SPDX-License-Identifier: GPL-2.0-or-later */ 2/* 3 * Definitions for the 'struct sk_buff' memory handlers. 4 * 5 * Authors: 6 * Alan Cox, <gw4pts@gw4pts.ampr.org> 7 * Florian La Roche, <rzsfl@rz.uni-sb.de> 8 */ 9 10#ifndef _LINUX_SKBUFF_H 11#define _LINUX_SKBUFF_H 12 13#include <linux/kernel.h> 14#include <linux/compiler.h> 15#include <linux/time.h> 16#include <linux/bug.h> 17#include <linux/bvec.h> 18#include <linux/cache.h> 19#include <linux/rbtree.h> 20#include <linux/socket.h> 21#include <linux/refcount.h> 22 23#include <linux/atomic.h> 24#include <asm/types.h> 25#include <linux/spinlock.h> 26#include <net/checksum.h> 27#include <linux/rcupdate.h> 28#include <linux/dma-mapping.h> 29#include <linux/netdev_features.h> 30#include <net/flow_dissector.h> 31#include <linux/in6.h> 32#include <linux/if_packet.h> 33#include <linux/llist.h> 34#include <net/flow.h> 35#include <net/page_pool.h> 36#if IS_ENABLED(CONFIG_NF_CONNTRACK) 37#include <linux/netfilter/nf_conntrack_common.h> 38#endif 39#include <net/net_debug.h> 40#include <net/dropreason-core.h> 41 42/** 43 * DOC: skb checksums 44 * 45 * The interface for checksum offload between the stack and networking drivers 46 * is as follows... 47 * 48 * IP checksum related features 49 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 50 * 51 * Drivers advertise checksum offload capabilities in the features of a device. 52 * From the stack's point of view these are capabilities offered by the driver. 53 * A driver typically only advertises features that it is capable of offloading 54 * to its device. 55 * 56 * .. flat-table:: Checksum related device features 57 * :widths: 1 10 58 * 59 * * - %NETIF_F_HW_CSUM 60 * - The driver (or its device) is able to compute one 61 * IP (one's complement) checksum for any combination 62 * of protocols or protocol layering. The checksum is 63 * computed and set in a packet per the CHECKSUM_PARTIAL 64 * interface (see below). 65 * 66 * * - %NETIF_F_IP_CSUM 67 * - Driver (device) is only able to checksum plain 68 * TCP or UDP packets over IPv4. These are specifically 69 * unencapsulated packets of the form IPv4|TCP or 70 * IPv4|UDP where the Protocol field in the IPv4 header 71 * is TCP or UDP. The IPv4 header may contain IP options. 72 * This feature cannot be set in features for a device 73 * with NETIF_F_HW_CSUM also set. This feature is being 74 * DEPRECATED (see below). 75 * 76 * * - %NETIF_F_IPV6_CSUM 77 * - Driver (device) is only able to checksum plain 78 * TCP or UDP packets over IPv6. These are specifically 79 * unencapsulated packets of the form IPv6|TCP or 80 * IPv6|UDP where the Next Header field in the IPv6 81 * header is either TCP or UDP. IPv6 extension headers 82 * are not supported with this feature. This feature 83 * cannot be set in features for a device with 84 * NETIF_F_HW_CSUM also set. This feature is being 85 * DEPRECATED (see below). 86 * 87 * * - %NETIF_F_RXCSUM 88 * - Driver (device) performs receive checksum offload. 89 * This flag is only used to disable the RX checksum 90 * feature for a device. The stack will accept receive 91 * checksum indication in packets received on a device 92 * regardless of whether NETIF_F_RXCSUM is set. 93 * 94 * Checksumming of received packets by device 95 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 96 * 97 * Indication of checksum verification is set in &sk_buff.ip_summed. 98 * Possible values are: 99 * 100 * - %CHECKSUM_NONE 101 * 102 * Device did not checksum this packet e.g. due to lack of capabilities. 103 * The packet contains full (though not verified) checksum in packet but 104 * not in skb->csum. Thus, skb->csum is undefined in this case. 105 * 106 * - %CHECKSUM_UNNECESSARY 107 * 108 * The hardware you're dealing with doesn't calculate the full checksum 109 * (as in %CHECKSUM_COMPLETE), but it does parse headers and verify checksums 110 * for specific protocols. For such packets it will set %CHECKSUM_UNNECESSARY 111 * if their checksums are okay. &sk_buff.csum is still undefined in this case 112 * though. A driver or device must never modify the checksum field in the 113 * packet even if checksum is verified. 114 * 115 * %CHECKSUM_UNNECESSARY is applicable to following protocols: 116 * 117 * - TCP: IPv6 and IPv4. 118 * - UDP: IPv4 and IPv6. A device may apply CHECKSUM_UNNECESSARY to a 119 * zero UDP checksum for either IPv4 or IPv6, the networking stack 120 * may perform further validation in this case. 121 * - GRE: only if the checksum is present in the header. 122 * - SCTP: indicates the CRC in SCTP header has been validated. 123 * - FCOE: indicates the CRC in FC frame has been validated. 124 * 125 * &sk_buff.csum_level indicates the number of consecutive checksums found in 126 * the packet minus one that have been verified as %CHECKSUM_UNNECESSARY. 127 * For instance if a device receives an IPv6->UDP->GRE->IPv4->TCP packet 128 * and a device is able to verify the checksums for UDP (possibly zero), 129 * GRE (checksum flag is set) and TCP, &sk_buff.csum_level would be set to 130 * two. If the device were only able to verify the UDP checksum and not 131 * GRE, either because it doesn't support GRE checksum or because GRE 132 * checksum is bad, skb->csum_level would be set to zero (TCP checksum is 133 * not considered in this case). 134 * 135 * - %CHECKSUM_COMPLETE 136 * 137 * This is the most generic way. The device supplied checksum of the _whole_ 138 * packet as seen by netif_rx() and fills in &sk_buff.csum. This means the 139 * hardware doesn't need to parse L3/L4 headers to implement this. 140 * 141 * Notes: 142 * 143 * - Even if device supports only some protocols, but is able to produce 144 * skb->csum, it MUST use CHECKSUM_COMPLETE, not CHECKSUM_UNNECESSARY. 145 * - CHECKSUM_COMPLETE is not applicable to SCTP and FCoE protocols. 146 * 147 * - %CHECKSUM_PARTIAL 148 * 149 * A checksum is set up to be offloaded to a device as described in the 150 * output description for CHECKSUM_PARTIAL. This may occur on a packet 151 * received directly from another Linux OS, e.g., a virtualized Linux kernel 152 * on the same host, or it may be set in the input path in GRO or remote 153 * checksum offload. For the purposes of checksum verification, the checksum 154 * referred to by skb->csum_start + skb->csum_offset and any preceding 155 * checksums in the packet are considered verified. Any checksums in the 156 * packet that are after the checksum being offloaded are not considered to 157 * be verified. 158 * 159 * Checksumming on transmit for non-GSO 160 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 161 * 162 * The stack requests checksum offload in the &sk_buff.ip_summed for a packet. 163 * Values are: 164 * 165 * - %CHECKSUM_PARTIAL 166 * 167 * The driver is required to checksum the packet as seen by hard_start_xmit() 168 * from &sk_buff.csum_start up to the end, and to record/write the checksum at 169 * offset &sk_buff.csum_start + &sk_buff.csum_offset. 170 * A driver may verify that the 171 * csum_start and csum_offset values are valid values given the length and 172 * offset of the packet, but it should not attempt to validate that the 173 * checksum refers to a legitimate transport layer checksum -- it is the 174 * purview of the stack to validate that csum_start and csum_offset are set 175 * correctly. 176 * 177 * When the stack requests checksum offload for a packet, the driver MUST 178 * ensure that the checksum is set correctly. A driver can either offload the 179 * checksum calculation to the device, or call skb_checksum_help (in the case 180 * that the device does not support offload for a particular checksum). 181 * 182 * %NETIF_F_IP_CSUM and %NETIF_F_IPV6_CSUM are being deprecated in favor of 183 * %NETIF_F_HW_CSUM. New devices should use %NETIF_F_HW_CSUM to indicate 184 * checksum offload capability. 185 * skb_csum_hwoffload_help() can be called to resolve %CHECKSUM_PARTIAL based 186 * on network device checksumming capabilities: if a packet does not match 187 * them, skb_checksum_help() or skb_crc32c_help() (depending on the value of 188 * &sk_buff.csum_not_inet, see :ref:`crc`) 189 * is called to resolve the checksum. 190 * 191 * - %CHECKSUM_NONE 192 * 193 * The skb was already checksummed by the protocol, or a checksum is not 194 * required. 195 * 196 * - %CHECKSUM_UNNECESSARY 197 * 198 * This has the same meaning as CHECKSUM_NONE for checksum offload on 199 * output. 200 * 201 * - %CHECKSUM_COMPLETE 202 * 203 * Not used in checksum output. If a driver observes a packet with this value 204 * set in skbuff, it should treat the packet as if %CHECKSUM_NONE were set. 205 * 206 * .. _crc: 207 * 208 * Non-IP checksum (CRC) offloads 209 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 210 * 211 * .. flat-table:: 212 * :widths: 1 10 213 * 214 * * - %NETIF_F_SCTP_CRC 215 * - This feature indicates that a device is capable of 216 * offloading the SCTP CRC in a packet. To perform this offload the stack 217 * will set csum_start and csum_offset accordingly, set ip_summed to 218 * %CHECKSUM_PARTIAL and set csum_not_inet to 1, to provide an indication 219 * in the skbuff that the %CHECKSUM_PARTIAL refers to CRC32c. 220 * A driver that supports both IP checksum offload and SCTP CRC32c offload 221 * must verify which offload is configured for a packet by testing the 222 * value of &sk_buff.csum_not_inet; skb_crc32c_csum_help() is provided to 223 * resolve %CHECKSUM_PARTIAL on skbs where csum_not_inet is set to 1. 224 * 225 * * - %NETIF_F_FCOE_CRC 226 * - This feature indicates that a device is capable of offloading the FCOE 227 * CRC in a packet. To perform this offload the stack will set ip_summed 228 * to %CHECKSUM_PARTIAL and set csum_start and csum_offset 229 * accordingly. Note that there is no indication in the skbuff that the 230 * %CHECKSUM_PARTIAL refers to an FCOE checksum, so a driver that supports 231 * both IP checksum offload and FCOE CRC offload must verify which offload 232 * is configured for a packet, presumably by inspecting packet headers. 233 * 234 * Checksumming on output with GSO 235 * ~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~ 236 * 237 * In the case of a GSO packet (skb_is_gso() is true), checksum offload 238 * is implied by the SKB_GSO_* flags in gso_type. Most obviously, if the 239 * gso_type is %SKB_GSO_TCPV4 or %SKB_GSO_TCPV6, TCP checksum offload as 240 * part of the GSO operation is implied. If a checksum is being offloaded 241 * with GSO then ip_summed is %CHECKSUM_PARTIAL, and both csum_start and 242 * csum_offset are set to refer to the outermost checksum being offloaded 243 * (two offloaded checksums are possible with UDP encapsulation). 244 */ 245 246/* Don't change this without changing skb_csum_unnecessary! */ 247#define CHECKSUM_NONE 0 248#define CHECKSUM_UNNECESSARY 1 249#define CHECKSUM_COMPLETE 2 250#define CHECKSUM_PARTIAL 3 251 252/* Maximum value in skb->csum_level */ 253#define SKB_MAX_CSUM_LEVEL 3 254 255#define SKB_DATA_ALIGN(X) ALIGN(X, SMP_CACHE_BYTES) 256#define SKB_WITH_OVERHEAD(X) \ 257 ((X) - SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 258 259/* For X bytes available in skb->head, what is the minimal 260 * allocation needed, knowing struct skb_shared_info needs 261 * to be aligned. 262 */ 263#define SKB_HEAD_ALIGN(X) (SKB_DATA_ALIGN(X) + \ 264 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 265 266#define SKB_MAX_ORDER(X, ORDER) \ 267 SKB_WITH_OVERHEAD((PAGE_SIZE << (ORDER)) - (X)) 268#define SKB_MAX_HEAD(X) (SKB_MAX_ORDER((X), 0)) 269#define SKB_MAX_ALLOC (SKB_MAX_ORDER(0, 2)) 270 271/* return minimum truesize of one skb containing X bytes of data */ 272#define SKB_TRUESIZE(X) ((X) + \ 273 SKB_DATA_ALIGN(sizeof(struct sk_buff)) + \ 274 SKB_DATA_ALIGN(sizeof(struct skb_shared_info))) 275 276struct ahash_request; 277struct net_device; 278struct scatterlist; 279struct pipe_inode_info; 280struct iov_iter; 281struct napi_struct; 282struct bpf_prog; 283union bpf_attr; 284struct skb_ext; 285struct ts_config; 286 287#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 288struct nf_bridge_info { 289 enum { 290 BRNF_PROTO_UNCHANGED, 291 BRNF_PROTO_8021Q, 292 BRNF_PROTO_PPPOE 293 } orig_proto:8; 294 u8 pkt_otherhost:1; 295 u8 in_prerouting:1; 296 u8 bridged_dnat:1; 297 u8 sabotage_in_done:1; 298 __u16 frag_max_size; 299 struct net_device *physindev; 300 301 /* always valid & non-NULL from FORWARD on, for physdev match */ 302 struct net_device *physoutdev; 303 union { 304 /* prerouting: detect dnat in orig/reply direction */ 305 __be32 ipv4_daddr; 306 struct in6_addr ipv6_daddr; 307 308 /* after prerouting + nat detected: store original source 309 * mac since neigh resolution overwrites it, only used while 310 * skb is out in neigh layer. 311 */ 312 char neigh_header[8]; 313 }; 314}; 315#endif 316 317#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 318/* Chain in tc_skb_ext will be used to share the tc chain with 319 * ovs recirc_id. It will be set to the current chain by tc 320 * and read by ovs to recirc_id. 321 */ 322struct tc_skb_ext { 323 union { 324 u64 act_miss_cookie; 325 __u32 chain; 326 }; 327 __u16 mru; 328 __u16 zone; 329 u8 post_ct:1; 330 u8 post_ct_snat:1; 331 u8 post_ct_dnat:1; 332 u8 act_miss:1; /* Set if act_miss_cookie is used */ 333 u8 l2_miss:1; /* Set by bridge upon FDB or MDB miss */ 334}; 335#endif 336 337struct sk_buff_head { 338 /* These two members must be first to match sk_buff. */ 339 struct_group_tagged(sk_buff_list, list, 340 struct sk_buff *next; 341 struct sk_buff *prev; 342 ); 343 344 __u32 qlen; 345 spinlock_t lock; 346}; 347 348struct sk_buff; 349 350#ifndef CONFIG_MAX_SKB_FRAGS 351# define CONFIG_MAX_SKB_FRAGS 17 352#endif 353 354#define MAX_SKB_FRAGS CONFIG_MAX_SKB_FRAGS 355 356extern int sysctl_max_skb_frags; 357 358/* Set skb_shinfo(skb)->gso_size to this in case you want skb_segment to 359 * segment using its current segmentation instead. 360 */ 361#define GSO_BY_FRAGS 0xFFFF 362 363typedef struct bio_vec skb_frag_t; 364 365/** 366 * skb_frag_size() - Returns the size of a skb fragment 367 * @frag: skb fragment 368 */ 369static inline unsigned int skb_frag_size(const skb_frag_t *frag) 370{ 371 return frag->bv_len; 372} 373 374/** 375 * skb_frag_size_set() - Sets the size of a skb fragment 376 * @frag: skb fragment 377 * @size: size of fragment 378 */ 379static inline void skb_frag_size_set(skb_frag_t *frag, unsigned int size) 380{ 381 frag->bv_len = size; 382} 383 384/** 385 * skb_frag_size_add() - Increments the size of a skb fragment by @delta 386 * @frag: skb fragment 387 * @delta: value to add 388 */ 389static inline void skb_frag_size_add(skb_frag_t *frag, int delta) 390{ 391 frag->bv_len += delta; 392} 393 394/** 395 * skb_frag_size_sub() - Decrements the size of a skb fragment by @delta 396 * @frag: skb fragment 397 * @delta: value to subtract 398 */ 399static inline void skb_frag_size_sub(skb_frag_t *frag, int delta) 400{ 401 frag->bv_len -= delta; 402} 403 404/** 405 * skb_frag_must_loop - Test if %p is a high memory page 406 * @p: fragment's page 407 */ 408static inline bool skb_frag_must_loop(struct page *p) 409{ 410#if defined(CONFIG_HIGHMEM) 411 if (IS_ENABLED(CONFIG_DEBUG_KMAP_LOCAL_FORCE_MAP) || PageHighMem(p)) 412 return true; 413#endif 414 return false; 415} 416 417/** 418 * skb_frag_foreach_page - loop over pages in a fragment 419 * 420 * @f: skb frag to operate on 421 * @f_off: offset from start of f->bv_page 422 * @f_len: length from f_off to loop over 423 * @p: (temp var) current page 424 * @p_off: (temp var) offset from start of current page, 425 * non-zero only on first page. 426 * @p_len: (temp var) length in current page, 427 * < PAGE_SIZE only on first and last page. 428 * @copied: (temp var) length so far, excluding current p_len. 429 * 430 * A fragment can hold a compound page, in which case per-page 431 * operations, notably kmap_atomic, must be called for each 432 * regular page. 433 */ 434#define skb_frag_foreach_page(f, f_off, f_len, p, p_off, p_len, copied) \ 435 for (p = skb_frag_page(f) + ((f_off) >> PAGE_SHIFT), \ 436 p_off = (f_off) & (PAGE_SIZE - 1), \ 437 p_len = skb_frag_must_loop(p) ? \ 438 min_t(u32, f_len, PAGE_SIZE - p_off) : f_len, \ 439 copied = 0; \ 440 copied < f_len; \ 441 copied += p_len, p++, p_off = 0, \ 442 p_len = min_t(u32, f_len - copied, PAGE_SIZE)) \ 443 444#define HAVE_HW_TIME_STAMP 445 446/** 447 * struct skb_shared_hwtstamps - hardware time stamps 448 * @hwtstamp: hardware time stamp transformed into duration 449 * since arbitrary point in time 450 * @netdev_data: address/cookie of network device driver used as 451 * reference to actual hardware time stamp 452 * 453 * Software time stamps generated by ktime_get_real() are stored in 454 * skb->tstamp. 455 * 456 * hwtstamps can only be compared against other hwtstamps from 457 * the same device. 458 * 459 * This structure is attached to packets as part of the 460 * &skb_shared_info. Use skb_hwtstamps() to get a pointer. 461 */ 462struct skb_shared_hwtstamps { 463 union { 464 ktime_t hwtstamp; 465 void *netdev_data; 466 }; 467}; 468 469/* Definitions for tx_flags in struct skb_shared_info */ 470enum { 471 /* generate hardware time stamp */ 472 SKBTX_HW_TSTAMP = 1 << 0, 473 474 /* generate software time stamp when queueing packet to NIC */ 475 SKBTX_SW_TSTAMP = 1 << 1, 476 477 /* device driver is going to provide hardware time stamp */ 478 SKBTX_IN_PROGRESS = 1 << 2, 479 480 /* generate hardware time stamp based on cycles if supported */ 481 SKBTX_HW_TSTAMP_USE_CYCLES = 1 << 3, 482 483 /* generate wifi status information (where possible) */ 484 SKBTX_WIFI_STATUS = 1 << 4, 485 486 /* determine hardware time stamp based on time or cycles */ 487 SKBTX_HW_TSTAMP_NETDEV = 1 << 5, 488 489 /* generate software time stamp when entering packet scheduling */ 490 SKBTX_SCHED_TSTAMP = 1 << 6, 491}; 492 493#define SKBTX_ANY_SW_TSTAMP (SKBTX_SW_TSTAMP | \ 494 SKBTX_SCHED_TSTAMP) 495#define SKBTX_ANY_TSTAMP (SKBTX_HW_TSTAMP | \ 496 SKBTX_HW_TSTAMP_USE_CYCLES | \ 497 SKBTX_ANY_SW_TSTAMP) 498 499/* Definitions for flags in struct skb_shared_info */ 500enum { 501 /* use zcopy routines */ 502 SKBFL_ZEROCOPY_ENABLE = BIT(0), 503 504 /* This indicates at least one fragment might be overwritten 505 * (as in vmsplice(), sendfile() ...) 506 * If we need to compute a TX checksum, we'll need to copy 507 * all frags to avoid possible bad checksum 508 */ 509 SKBFL_SHARED_FRAG = BIT(1), 510 511 /* segment contains only zerocopy data and should not be 512 * charged to the kernel memory. 513 */ 514 SKBFL_PURE_ZEROCOPY = BIT(2), 515 516 SKBFL_DONT_ORPHAN = BIT(3), 517 518 /* page references are managed by the ubuf_info, so it's safe to 519 * use frags only up until ubuf_info is released 520 */ 521 SKBFL_MANAGED_FRAG_REFS = BIT(4), 522}; 523 524#define SKBFL_ZEROCOPY_FRAG (SKBFL_ZEROCOPY_ENABLE | SKBFL_SHARED_FRAG) 525#define SKBFL_ALL_ZEROCOPY (SKBFL_ZEROCOPY_FRAG | SKBFL_PURE_ZEROCOPY | \ 526 SKBFL_DONT_ORPHAN | SKBFL_MANAGED_FRAG_REFS) 527 528/* 529 * The callback notifies userspace to release buffers when skb DMA is done in 530 * lower device, the skb last reference should be 0 when calling this. 531 * The zerocopy_success argument is true if zero copy transmit occurred, 532 * false on data copy or out of memory error caused by data copy attempt. 533 * The ctx field is used to track device context. 534 * The desc field is used to track userspace buffer index. 535 */ 536struct ubuf_info { 537 void (*callback)(struct sk_buff *, struct ubuf_info *, 538 bool zerocopy_success); 539 refcount_t refcnt; 540 u8 flags; 541}; 542 543struct ubuf_info_msgzc { 544 struct ubuf_info ubuf; 545 546 union { 547 struct { 548 unsigned long desc; 549 void *ctx; 550 }; 551 struct { 552 u32 id; 553 u16 len; 554 u16 zerocopy:1; 555 u32 bytelen; 556 }; 557 }; 558 559 struct mmpin { 560 struct user_struct *user; 561 unsigned int num_pg; 562 } mmp; 563}; 564 565#define skb_uarg(SKB) ((struct ubuf_info *)(skb_shinfo(SKB)->destructor_arg)) 566#define uarg_to_msgzc(ubuf_ptr) container_of((ubuf_ptr), struct ubuf_info_msgzc, \ 567 ubuf) 568 569int mm_account_pinned_pages(struct mmpin *mmp, size_t size); 570void mm_unaccount_pinned_pages(struct mmpin *mmp); 571 572/* This data is invariant across clones and lives at 573 * the end of the header data, ie. at skb->end. 574 */ 575struct skb_shared_info { 576 __u8 flags; 577 __u8 meta_len; 578 __u8 nr_frags; 579 __u8 tx_flags; 580 unsigned short gso_size; 581 /* Warning: this field is not always filled in (UFO)! */ 582 unsigned short gso_segs; 583 struct sk_buff *frag_list; 584 struct skb_shared_hwtstamps hwtstamps; 585 unsigned int gso_type; 586 u32 tskey; 587 588 /* 589 * Warning : all fields before dataref are cleared in __alloc_skb() 590 */ 591 atomic_t dataref; 592 unsigned int xdp_frags_size; 593 594 /* Intermediate layers must ensure that destructor_arg 595 * remains valid until skb destructor */ 596 void * destructor_arg; 597 598 /* must be last field, see pskb_expand_head() */ 599 skb_frag_t frags[MAX_SKB_FRAGS]; 600}; 601 602/** 603 * DOC: dataref and headerless skbs 604 * 605 * Transport layers send out clones of payload skbs they hold for 606 * retransmissions. To allow lower layers of the stack to prepend their headers 607 * we split &skb_shared_info.dataref into two halves. 608 * The lower 16 bits count the overall number of references. 609 * The higher 16 bits indicate how many of the references are payload-only. 610 * skb_header_cloned() checks if skb is allowed to add / write the headers. 611 * 612 * The creator of the skb (e.g. TCP) marks its skb as &sk_buff.nohdr 613 * (via __skb_header_release()). Any clone created from marked skb will get 614 * &sk_buff.hdr_len populated with the available headroom. 615 * If there's the only clone in existence it's able to modify the headroom 616 * at will. The sequence of calls inside the transport layer is:: 617 * 618 * <alloc skb> 619 * skb_reserve() 620 * __skb_header_release() 621 * skb_clone() 622 * // send the clone down the stack 623 * 624 * This is not a very generic construct and it depends on the transport layers 625 * doing the right thing. In practice there's usually only one payload-only skb. 626 * Having multiple payload-only skbs with different lengths of hdr_len is not 627 * possible. The payload-only skbs should never leave their owner. 628 */ 629#define SKB_DATAREF_SHIFT 16 630#define SKB_DATAREF_MASK ((1 << SKB_DATAREF_SHIFT) - 1) 631 632 633enum { 634 SKB_FCLONE_UNAVAILABLE, /* skb has no fclone (from head_cache) */ 635 SKB_FCLONE_ORIG, /* orig skb (from fclone_cache) */ 636 SKB_FCLONE_CLONE, /* companion fclone skb (from fclone_cache) */ 637}; 638 639enum { 640 SKB_GSO_TCPV4 = 1 << 0, 641 642 /* This indicates the skb is from an untrusted source. */ 643 SKB_GSO_DODGY = 1 << 1, 644 645 /* This indicates the tcp segment has CWR set. */ 646 SKB_GSO_TCP_ECN = 1 << 2, 647 648 SKB_GSO_TCP_FIXEDID = 1 << 3, 649 650 SKB_GSO_TCPV6 = 1 << 4, 651 652 SKB_GSO_FCOE = 1 << 5, 653 654 SKB_GSO_GRE = 1 << 6, 655 656 SKB_GSO_GRE_CSUM = 1 << 7, 657 658 SKB_GSO_IPXIP4 = 1 << 8, 659 660 SKB_GSO_IPXIP6 = 1 << 9, 661 662 SKB_GSO_UDP_TUNNEL = 1 << 10, 663 664 SKB_GSO_UDP_TUNNEL_CSUM = 1 << 11, 665 666 SKB_GSO_PARTIAL = 1 << 12, 667 668 SKB_GSO_TUNNEL_REMCSUM = 1 << 13, 669 670 SKB_GSO_SCTP = 1 << 14, 671 672 SKB_GSO_ESP = 1 << 15, 673 674 SKB_GSO_UDP = 1 << 16, 675 676 SKB_GSO_UDP_L4 = 1 << 17, 677 678 SKB_GSO_FRAGLIST = 1 << 18, 679}; 680 681#if BITS_PER_LONG > 32 682#define NET_SKBUFF_DATA_USES_OFFSET 1 683#endif 684 685#ifdef NET_SKBUFF_DATA_USES_OFFSET 686typedef unsigned int sk_buff_data_t; 687#else 688typedef unsigned char *sk_buff_data_t; 689#endif 690 691/** 692 * DOC: Basic sk_buff geometry 693 * 694 * struct sk_buff itself is a metadata structure and does not hold any packet 695 * data. All the data is held in associated buffers. 696 * 697 * &sk_buff.head points to the main "head" buffer. The head buffer is divided 698 * into two parts: 699 * 700 * - data buffer, containing headers and sometimes payload; 701 * this is the part of the skb operated on by the common helpers 702 * such as skb_put() or skb_pull(); 703 * - shared info (struct skb_shared_info) which holds an array of pointers 704 * to read-only data in the (page, offset, length) format. 705 * 706 * Optionally &skb_shared_info.frag_list may point to another skb. 707 * 708 * Basic diagram may look like this:: 709 * 710 * --------------- 711 * | sk_buff | 712 * --------------- 713 * ,--------------------------- + head 714 * / ,----------------- + data 715 * / / ,----------- + tail 716 * | | | , + end 717 * | | | | 718 * v v v v 719 * ----------------------------------------------- 720 * | headroom | data | tailroom | skb_shared_info | 721 * ----------------------------------------------- 722 * + [page frag] 723 * + [page frag] 724 * + [page frag] 725 * + [page frag] --------- 726 * + frag_list --> | sk_buff | 727 * --------- 728 * 729 */ 730 731/** 732 * struct sk_buff - socket buffer 733 * @next: Next buffer in list 734 * @prev: Previous buffer in list 735 * @tstamp: Time we arrived/left 736 * @skb_mstamp_ns: (aka @tstamp) earliest departure time; start point 737 * for retransmit timer 738 * @rbnode: RB tree node, alternative to next/prev for netem/tcp 739 * @list: queue head 740 * @ll_node: anchor in an llist (eg socket defer_list) 741 * @sk: Socket we are owned by 742 * @ip_defrag_offset: (aka @sk) alternate use of @sk, used in 743 * fragmentation management 744 * @dev: Device we arrived on/are leaving by 745 * @dev_scratch: (aka @dev) alternate use of @dev when @dev would be %NULL 746 * @cb: Control buffer. Free for use by every layer. Put private vars here 747 * @_skb_refdst: destination entry (with norefcount bit) 748 * @sp: the security path, used for xfrm 749 * @len: Length of actual data 750 * @data_len: Data length 751 * @mac_len: Length of link layer header 752 * @hdr_len: writable header length of cloned skb 753 * @csum: Checksum (must include start/offset pair) 754 * @csum_start: Offset from skb->head where checksumming should start 755 * @csum_offset: Offset from csum_start where checksum should be stored 756 * @priority: Packet queueing priority 757 * @ignore_df: allow local fragmentation 758 * @cloned: Head may be cloned (check refcnt to be sure) 759 * @ip_summed: Driver fed us an IP checksum 760 * @nohdr: Payload reference only, must not modify header 761 * @pkt_type: Packet class 762 * @fclone: skbuff clone status 763 * @ipvs_property: skbuff is owned by ipvs 764 * @inner_protocol_type: whether the inner protocol is 765 * ENCAP_TYPE_ETHER or ENCAP_TYPE_IPPROTO 766 * @remcsum_offload: remote checksum offload is enabled 767 * @offload_fwd_mark: Packet was L2-forwarded in hardware 768 * @offload_l3_fwd_mark: Packet was L3-forwarded in hardware 769 * @tc_skip_classify: do not classify packet. set by IFB device 770 * @tc_at_ingress: used within tc_classify to distinguish in/egress 771 * @redirected: packet was redirected by packet classifier 772 * @from_ingress: packet was redirected from the ingress path 773 * @nf_skip_egress: packet shall skip nf egress - see netfilter_netdev.h 774 * @peeked: this packet has been seen already, so stats have been 775 * done for it, don't do them again 776 * @nf_trace: netfilter packet trace flag 777 * @protocol: Packet protocol from driver 778 * @destructor: Destruct function 779 * @tcp_tsorted_anchor: list structure for TCP (tp->tsorted_sent_queue) 780 * @_sk_redir: socket redirection information for skmsg 781 * @_nfct: Associated connection, if any (with nfctinfo bits) 782 * @nf_bridge: Saved data about a bridged frame - see br_netfilter.c 783 * @skb_iif: ifindex of device we arrived on 784 * @tc_index: Traffic control index 785 * @hash: the packet hash 786 * @queue_mapping: Queue mapping for multiqueue devices 787 * @head_frag: skb was allocated from page fragments, 788 * not allocated by kmalloc() or vmalloc(). 789 * @pfmemalloc: skbuff was allocated from PFMEMALLOC reserves 790 * @pp_recycle: mark the packet for recycling instead of freeing (implies 791 * page_pool support on driver) 792 * @active_extensions: active extensions (skb_ext_id types) 793 * @ndisc_nodetype: router type (from link layer) 794 * @ooo_okay: allow the mapping of a socket to a queue to be changed 795 * @l4_hash: indicate hash is a canonical 4-tuple hash over transport 796 * ports. 797 * @sw_hash: indicates hash was computed in software stack 798 * @wifi_acked_valid: wifi_acked was set 799 * @wifi_acked: whether frame was acked on wifi or not 800 * @no_fcs: Request NIC to treat last 4 bytes as Ethernet FCS 801 * @encapsulation: indicates the inner headers in the skbuff are valid 802 * @encap_hdr_csum: software checksum is needed 803 * @csum_valid: checksum is already valid 804 * @csum_not_inet: use CRC32c to resolve CHECKSUM_PARTIAL 805 * @csum_complete_sw: checksum was completed by software 806 * @csum_level: indicates the number of consecutive checksums found in 807 * the packet minus one that have been verified as 808 * CHECKSUM_UNNECESSARY (max 3) 809 * @dst_pending_confirm: need to confirm neighbour 810 * @decrypted: Decrypted SKB 811 * @slow_gro: state present at GRO time, slower prepare step required 812 * @mono_delivery_time: When set, skb->tstamp has the 813 * delivery_time in mono clock base (i.e. EDT). Otherwise, the 814 * skb->tstamp has the (rcv) timestamp at ingress and 815 * delivery_time at egress. 816 * @napi_id: id of the NAPI struct this skb came from 817 * @sender_cpu: (aka @napi_id) source CPU in XPS 818 * @alloc_cpu: CPU which did the skb allocation. 819 * @secmark: security marking 820 * @mark: Generic packet mark 821 * @reserved_tailroom: (aka @mark) number of bytes of free space available 822 * at the tail of an sk_buff 823 * @vlan_all: vlan fields (proto & tci) 824 * @vlan_proto: vlan encapsulation protocol 825 * @vlan_tci: vlan tag control information 826 * @inner_protocol: Protocol (encapsulation) 827 * @inner_ipproto: (aka @inner_protocol) stores ipproto when 828 * skb->inner_protocol_type == ENCAP_TYPE_IPPROTO; 829 * @inner_transport_header: Inner transport layer header (encapsulation) 830 * @inner_network_header: Network layer header (encapsulation) 831 * @inner_mac_header: Link layer header (encapsulation) 832 * @transport_header: Transport layer header 833 * @network_header: Network layer header 834 * @mac_header: Link layer header 835 * @kcov_handle: KCOV remote handle for remote coverage collection 836 * @tail: Tail pointer 837 * @end: End pointer 838 * @head: Head of buffer 839 * @data: Data head pointer 840 * @truesize: Buffer size 841 * @users: User count - see {datagram,tcp}.c 842 * @extensions: allocated extensions, valid if active_extensions is nonzero 843 */ 844 845struct sk_buff { 846 union { 847 struct { 848 /* These two members must be first to match sk_buff_head. */ 849 struct sk_buff *next; 850 struct sk_buff *prev; 851 852 union { 853 struct net_device *dev; 854 /* Some protocols might use this space to store information, 855 * while device pointer would be NULL. 856 * UDP receive path is one user. 857 */ 858 unsigned long dev_scratch; 859 }; 860 }; 861 struct rb_node rbnode; /* used in netem, ip4 defrag, and tcp stack */ 862 struct list_head list; 863 struct llist_node ll_node; 864 }; 865 866 union { 867 struct sock *sk; 868 int ip_defrag_offset; 869 }; 870 871 union { 872 ktime_t tstamp; 873 u64 skb_mstamp_ns; /* earliest departure time */ 874 }; 875 /* 876 * This is the control buffer. It is free to use for every 877 * layer. Please put your private variables there. If you 878 * want to keep them across layers you have to do a skb_clone() 879 * first. This is owned by whoever has the skb queued ATM. 880 */ 881 char cb[48] __aligned(8); 882 883 union { 884 struct { 885 unsigned long _skb_refdst; 886 void (*destructor)(struct sk_buff *skb); 887 }; 888 struct list_head tcp_tsorted_anchor; 889#ifdef CONFIG_NET_SOCK_MSG 890 unsigned long _sk_redir; 891#endif 892 }; 893 894#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 895 unsigned long _nfct; 896#endif 897 unsigned int len, 898 data_len; 899 __u16 mac_len, 900 hdr_len; 901 902 /* Following fields are _not_ copied in __copy_skb_header() 903 * Note that queue_mapping is here mostly to fill a hole. 904 */ 905 __u16 queue_mapping; 906 907/* if you move cloned around you also must adapt those constants */ 908#ifdef __BIG_ENDIAN_BITFIELD 909#define CLONED_MASK (1 << 7) 910#else 911#define CLONED_MASK 1 912#endif 913#define CLONED_OFFSET offsetof(struct sk_buff, __cloned_offset) 914 915 /* private: */ 916 __u8 __cloned_offset[0]; 917 /* public: */ 918 __u8 cloned:1, 919 nohdr:1, 920 fclone:2, 921 peeked:1, 922 head_frag:1, 923 pfmemalloc:1, 924 pp_recycle:1; /* page_pool recycle indicator */ 925#ifdef CONFIG_SKB_EXTENSIONS 926 __u8 active_extensions; 927#endif 928 929 /* Fields enclosed in headers group are copied 930 * using a single memcpy() in __copy_skb_header() 931 */ 932 struct_group(headers, 933 934 /* private: */ 935 __u8 __pkt_type_offset[0]; 936 /* public: */ 937 __u8 pkt_type:3; /* see PKT_TYPE_MAX */ 938 __u8 ignore_df:1; 939 __u8 dst_pending_confirm:1; 940 __u8 ip_summed:2; 941 __u8 ooo_okay:1; 942 943 /* private: */ 944 __u8 __mono_tc_offset[0]; 945 /* public: */ 946 __u8 mono_delivery_time:1; /* See SKB_MONO_DELIVERY_TIME_MASK */ 947#ifdef CONFIG_NET_CLS_ACT 948 __u8 tc_at_ingress:1; /* See TC_AT_INGRESS_MASK */ 949 __u8 tc_skip_classify:1; 950#endif 951 __u8 remcsum_offload:1; 952 __u8 csum_complete_sw:1; 953 __u8 csum_level:2; 954 __u8 inner_protocol_type:1; 955 956 __u8 l4_hash:1; 957 __u8 sw_hash:1; 958#ifdef CONFIG_WIRELESS 959 __u8 wifi_acked_valid:1; 960 __u8 wifi_acked:1; 961#endif 962 __u8 no_fcs:1; 963 /* Indicates the inner headers are valid in the skbuff. */ 964 __u8 encapsulation:1; 965 __u8 encap_hdr_csum:1; 966 __u8 csum_valid:1; 967#ifdef CONFIG_IPV6_NDISC_NODETYPE 968 __u8 ndisc_nodetype:2; 969#endif 970 971#if IS_ENABLED(CONFIG_IP_VS) 972 __u8 ipvs_property:1; 973#endif 974#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 975 __u8 nf_trace:1; 976#endif 977#ifdef CONFIG_NET_SWITCHDEV 978 __u8 offload_fwd_mark:1; 979 __u8 offload_l3_fwd_mark:1; 980#endif 981 __u8 redirected:1; 982#ifdef CONFIG_NET_REDIRECT 983 __u8 from_ingress:1; 984#endif 985#ifdef CONFIG_NETFILTER_SKIP_EGRESS 986 __u8 nf_skip_egress:1; 987#endif 988#ifdef CONFIG_TLS_DEVICE 989 __u8 decrypted:1; 990#endif 991 __u8 slow_gro:1; 992#if IS_ENABLED(CONFIG_IP_SCTP) 993 __u8 csum_not_inet:1; 994#endif 995 996#ifdef CONFIG_NET_SCHED 997 __u16 tc_index; /* traffic control index */ 998#endif 999 1000 u16 alloc_cpu; 1001 1002 union { 1003 __wsum csum; 1004 struct { 1005 __u16 csum_start; 1006 __u16 csum_offset; 1007 }; 1008 }; 1009 __u32 priority; 1010 int skb_iif; 1011 __u32 hash; 1012 union { 1013 u32 vlan_all; 1014 struct { 1015 __be16 vlan_proto; 1016 __u16 vlan_tci; 1017 }; 1018 }; 1019#if defined(CONFIG_NET_RX_BUSY_POLL) || defined(CONFIG_XPS) 1020 union { 1021 unsigned int napi_id; 1022 unsigned int sender_cpu; 1023 }; 1024#endif 1025#ifdef CONFIG_NETWORK_SECMARK 1026 __u32 secmark; 1027#endif 1028 1029 union { 1030 __u32 mark; 1031 __u32 reserved_tailroom; 1032 }; 1033 1034 union { 1035 __be16 inner_protocol; 1036 __u8 inner_ipproto; 1037 }; 1038 1039 __u16 inner_transport_header; 1040 __u16 inner_network_header; 1041 __u16 inner_mac_header; 1042 1043 __be16 protocol; 1044 __u16 transport_header; 1045 __u16 network_header; 1046 __u16 mac_header; 1047 1048#ifdef CONFIG_KCOV 1049 u64 kcov_handle; 1050#endif 1051 1052 ); /* end headers group */ 1053 1054 /* These elements must be at the end, see alloc_skb() for details. */ 1055 sk_buff_data_t tail; 1056 sk_buff_data_t end; 1057 unsigned char *head, 1058 *data; 1059 unsigned int truesize; 1060 refcount_t users; 1061 1062#ifdef CONFIG_SKB_EXTENSIONS 1063 /* only useable after checking ->active_extensions != 0 */ 1064 struct skb_ext *extensions; 1065#endif 1066}; 1067 1068/* if you move pkt_type around you also must adapt those constants */ 1069#ifdef __BIG_ENDIAN_BITFIELD 1070#define PKT_TYPE_MAX (7 << 5) 1071#else 1072#define PKT_TYPE_MAX 7 1073#endif 1074#define PKT_TYPE_OFFSET offsetof(struct sk_buff, __pkt_type_offset) 1075 1076/* if you move tc_at_ingress or mono_delivery_time 1077 * around, you also must adapt these constants. 1078 */ 1079#ifdef __BIG_ENDIAN_BITFIELD 1080#define SKB_MONO_DELIVERY_TIME_MASK (1 << 7) 1081#define TC_AT_INGRESS_MASK (1 << 6) 1082#else 1083#define SKB_MONO_DELIVERY_TIME_MASK (1 << 0) 1084#define TC_AT_INGRESS_MASK (1 << 1) 1085#endif 1086#define SKB_BF_MONO_TC_OFFSET offsetof(struct sk_buff, __mono_tc_offset) 1087 1088#ifdef __KERNEL__ 1089/* 1090 * Handling routines are only of interest to the kernel 1091 */ 1092 1093#define SKB_ALLOC_FCLONE 0x01 1094#define SKB_ALLOC_RX 0x02 1095#define SKB_ALLOC_NAPI 0x04 1096 1097/** 1098 * skb_pfmemalloc - Test if the skb was allocated from PFMEMALLOC reserves 1099 * @skb: buffer 1100 */ 1101static inline bool skb_pfmemalloc(const struct sk_buff *skb) 1102{ 1103 return unlikely(skb->pfmemalloc); 1104} 1105 1106/* 1107 * skb might have a dst pointer attached, refcounted or not. 1108 * _skb_refdst low order bit is set if refcount was _not_ taken 1109 */ 1110#define SKB_DST_NOREF 1UL 1111#define SKB_DST_PTRMASK ~(SKB_DST_NOREF) 1112 1113/** 1114 * skb_dst - returns skb dst_entry 1115 * @skb: buffer 1116 * 1117 * Returns skb dst_entry, regardless of reference taken or not. 1118 */ 1119static inline struct dst_entry *skb_dst(const struct sk_buff *skb) 1120{ 1121 /* If refdst was not refcounted, check we still are in a 1122 * rcu_read_lock section 1123 */ 1124 WARN_ON((skb->_skb_refdst & SKB_DST_NOREF) && 1125 !rcu_read_lock_held() && 1126 !rcu_read_lock_bh_held()); 1127 return (struct dst_entry *)(skb->_skb_refdst & SKB_DST_PTRMASK); 1128} 1129 1130/** 1131 * skb_dst_set - sets skb dst 1132 * @skb: buffer 1133 * @dst: dst entry 1134 * 1135 * Sets skb dst, assuming a reference was taken on dst and should 1136 * be released by skb_dst_drop() 1137 */ 1138static inline void skb_dst_set(struct sk_buff *skb, struct dst_entry *dst) 1139{ 1140 skb->slow_gro |= !!dst; 1141 skb->_skb_refdst = (unsigned long)dst; 1142} 1143 1144/** 1145 * skb_dst_set_noref - sets skb dst, hopefully, without taking reference 1146 * @skb: buffer 1147 * @dst: dst entry 1148 * 1149 * Sets skb dst, assuming a reference was not taken on dst. 1150 * If dst entry is cached, we do not take reference and dst_release 1151 * will be avoided by refdst_drop. If dst entry is not cached, we take 1152 * reference, so that last dst_release can destroy the dst immediately. 1153 */ 1154static inline void skb_dst_set_noref(struct sk_buff *skb, struct dst_entry *dst) 1155{ 1156 WARN_ON(!rcu_read_lock_held() && !rcu_read_lock_bh_held()); 1157 skb->slow_gro |= !!dst; 1158 skb->_skb_refdst = (unsigned long)dst | SKB_DST_NOREF; 1159} 1160 1161/** 1162 * skb_dst_is_noref - Test if skb dst isn't refcounted 1163 * @skb: buffer 1164 */ 1165static inline bool skb_dst_is_noref(const struct sk_buff *skb) 1166{ 1167 return (skb->_skb_refdst & SKB_DST_NOREF) && skb_dst(skb); 1168} 1169 1170/** 1171 * skb_rtable - Returns the skb &rtable 1172 * @skb: buffer 1173 */ 1174static inline struct rtable *skb_rtable(const struct sk_buff *skb) 1175{ 1176 return (struct rtable *)skb_dst(skb); 1177} 1178 1179/* For mangling skb->pkt_type from user space side from applications 1180 * such as nft, tc, etc, we only allow a conservative subset of 1181 * possible pkt_types to be set. 1182*/ 1183static inline bool skb_pkt_type_ok(u32 ptype) 1184{ 1185 return ptype <= PACKET_OTHERHOST; 1186} 1187 1188/** 1189 * skb_napi_id - Returns the skb's NAPI id 1190 * @skb: buffer 1191 */ 1192static inline unsigned int skb_napi_id(const struct sk_buff *skb) 1193{ 1194#ifdef CONFIG_NET_RX_BUSY_POLL 1195 return skb->napi_id; 1196#else 1197 return 0; 1198#endif 1199} 1200 1201static inline bool skb_wifi_acked_valid(const struct sk_buff *skb) 1202{ 1203#ifdef CONFIG_WIRELESS 1204 return skb->wifi_acked_valid; 1205#else 1206 return 0; 1207#endif 1208} 1209 1210/** 1211 * skb_unref - decrement the skb's reference count 1212 * @skb: buffer 1213 * 1214 * Returns true if we can free the skb. 1215 */ 1216static inline bool skb_unref(struct sk_buff *skb) 1217{ 1218 if (unlikely(!skb)) 1219 return false; 1220 if (likely(refcount_read(&skb->users) == 1)) 1221 smp_rmb(); 1222 else if (likely(!refcount_dec_and_test(&skb->users))) 1223 return false; 1224 1225 return true; 1226} 1227 1228void __fix_address 1229kfree_skb_reason(struct sk_buff *skb, enum skb_drop_reason reason); 1230 1231/** 1232 * kfree_skb - free an sk_buff with 'NOT_SPECIFIED' reason 1233 * @skb: buffer to free 1234 */ 1235static inline void kfree_skb(struct sk_buff *skb) 1236{ 1237 kfree_skb_reason(skb, SKB_DROP_REASON_NOT_SPECIFIED); 1238} 1239 1240void skb_release_head_state(struct sk_buff *skb); 1241void kfree_skb_list_reason(struct sk_buff *segs, 1242 enum skb_drop_reason reason); 1243void skb_dump(const char *level, const struct sk_buff *skb, bool full_pkt); 1244void skb_tx_error(struct sk_buff *skb); 1245 1246static inline void kfree_skb_list(struct sk_buff *segs) 1247{ 1248 kfree_skb_list_reason(segs, SKB_DROP_REASON_NOT_SPECIFIED); 1249} 1250 1251#ifdef CONFIG_TRACEPOINTS 1252void consume_skb(struct sk_buff *skb); 1253#else 1254static inline void consume_skb(struct sk_buff *skb) 1255{ 1256 return kfree_skb(skb); 1257} 1258#endif 1259 1260void __consume_stateless_skb(struct sk_buff *skb); 1261void __kfree_skb(struct sk_buff *skb); 1262extern struct kmem_cache *skbuff_cache; 1263 1264void kfree_skb_partial(struct sk_buff *skb, bool head_stolen); 1265bool skb_try_coalesce(struct sk_buff *to, struct sk_buff *from, 1266 bool *fragstolen, int *delta_truesize); 1267 1268struct sk_buff *__alloc_skb(unsigned int size, gfp_t priority, int flags, 1269 int node); 1270struct sk_buff *__build_skb(void *data, unsigned int frag_size); 1271struct sk_buff *build_skb(void *data, unsigned int frag_size); 1272struct sk_buff *build_skb_around(struct sk_buff *skb, 1273 void *data, unsigned int frag_size); 1274void skb_attempt_defer_free(struct sk_buff *skb); 1275 1276struct sk_buff *napi_build_skb(void *data, unsigned int frag_size); 1277struct sk_buff *slab_build_skb(void *data); 1278 1279/** 1280 * alloc_skb - allocate a network buffer 1281 * @size: size to allocate 1282 * @priority: allocation mask 1283 * 1284 * This function is a convenient wrapper around __alloc_skb(). 1285 */ 1286static inline struct sk_buff *alloc_skb(unsigned int size, 1287 gfp_t priority) 1288{ 1289 return __alloc_skb(size, priority, 0, NUMA_NO_NODE); 1290} 1291 1292struct sk_buff *alloc_skb_with_frags(unsigned long header_len, 1293 unsigned long data_len, 1294 int max_page_order, 1295 int *errcode, 1296 gfp_t gfp_mask); 1297struct sk_buff *alloc_skb_for_msg(struct sk_buff *first); 1298 1299/* Layout of fast clones : [skb1][skb2][fclone_ref] */ 1300struct sk_buff_fclones { 1301 struct sk_buff skb1; 1302 1303 struct sk_buff skb2; 1304 1305 refcount_t fclone_ref; 1306}; 1307 1308/** 1309 * skb_fclone_busy - check if fclone is busy 1310 * @sk: socket 1311 * @skb: buffer 1312 * 1313 * Returns true if skb is a fast clone, and its clone is not freed. 1314 * Some drivers call skb_orphan() in their ndo_start_xmit(), 1315 * so we also check that this didnt happen. 1316 */ 1317static inline bool skb_fclone_busy(const struct sock *sk, 1318 const struct sk_buff *skb) 1319{ 1320 const struct sk_buff_fclones *fclones; 1321 1322 fclones = container_of(skb, struct sk_buff_fclones, skb1); 1323 1324 return skb->fclone == SKB_FCLONE_ORIG && 1325 refcount_read(&fclones->fclone_ref) > 1 && 1326 READ_ONCE(fclones->skb2.sk) == sk; 1327} 1328 1329/** 1330 * alloc_skb_fclone - allocate a network buffer from fclone cache 1331 * @size: size to allocate 1332 * @priority: allocation mask 1333 * 1334 * This function is a convenient wrapper around __alloc_skb(). 1335 */ 1336static inline struct sk_buff *alloc_skb_fclone(unsigned int size, 1337 gfp_t priority) 1338{ 1339 return __alloc_skb(size, priority, SKB_ALLOC_FCLONE, NUMA_NO_NODE); 1340} 1341 1342struct sk_buff *skb_morph(struct sk_buff *dst, struct sk_buff *src); 1343void skb_headers_offset_update(struct sk_buff *skb, int off); 1344int skb_copy_ubufs(struct sk_buff *skb, gfp_t gfp_mask); 1345struct sk_buff *skb_clone(struct sk_buff *skb, gfp_t priority); 1346void skb_copy_header(struct sk_buff *new, const struct sk_buff *old); 1347struct sk_buff *skb_copy(const struct sk_buff *skb, gfp_t priority); 1348struct sk_buff *__pskb_copy_fclone(struct sk_buff *skb, int headroom, 1349 gfp_t gfp_mask, bool fclone); 1350static inline struct sk_buff *__pskb_copy(struct sk_buff *skb, int headroom, 1351 gfp_t gfp_mask) 1352{ 1353 return __pskb_copy_fclone(skb, headroom, gfp_mask, false); 1354} 1355 1356int pskb_expand_head(struct sk_buff *skb, int nhead, int ntail, gfp_t gfp_mask); 1357struct sk_buff *skb_realloc_headroom(struct sk_buff *skb, 1358 unsigned int headroom); 1359struct sk_buff *skb_expand_head(struct sk_buff *skb, unsigned int headroom); 1360struct sk_buff *skb_copy_expand(const struct sk_buff *skb, int newheadroom, 1361 int newtailroom, gfp_t priority); 1362int __must_check skb_to_sgvec_nomark(struct sk_buff *skb, struct scatterlist *sg, 1363 int offset, int len); 1364int __must_check skb_to_sgvec(struct sk_buff *skb, struct scatterlist *sg, 1365 int offset, int len); 1366int skb_cow_data(struct sk_buff *skb, int tailbits, struct sk_buff **trailer); 1367int __skb_pad(struct sk_buff *skb, int pad, bool free_on_error); 1368 1369/** 1370 * skb_pad - zero pad the tail of an skb 1371 * @skb: buffer to pad 1372 * @pad: space to pad 1373 * 1374 * Ensure that a buffer is followed by a padding area that is zero 1375 * filled. Used by network drivers which may DMA or transfer data 1376 * beyond the buffer end onto the wire. 1377 * 1378 * May return error in out of memory cases. The skb is freed on error. 1379 */ 1380static inline int skb_pad(struct sk_buff *skb, int pad) 1381{ 1382 return __skb_pad(skb, pad, true); 1383} 1384#define dev_kfree_skb(a) consume_skb(a) 1385 1386int skb_append_pagefrags(struct sk_buff *skb, struct page *page, 1387 int offset, size_t size, size_t max_frags); 1388 1389struct skb_seq_state { 1390 __u32 lower_offset; 1391 __u32 upper_offset; 1392 __u32 frag_idx; 1393 __u32 stepped_offset; 1394 struct sk_buff *root_skb; 1395 struct sk_buff *cur_skb; 1396 __u8 *frag_data; 1397 __u32 frag_off; 1398}; 1399 1400void skb_prepare_seq_read(struct sk_buff *skb, unsigned int from, 1401 unsigned int to, struct skb_seq_state *st); 1402unsigned int skb_seq_read(unsigned int consumed, const u8 **data, 1403 struct skb_seq_state *st); 1404void skb_abort_seq_read(struct skb_seq_state *st); 1405 1406unsigned int skb_find_text(struct sk_buff *skb, unsigned int from, 1407 unsigned int to, struct ts_config *config); 1408 1409/* 1410 * Packet hash types specify the type of hash in skb_set_hash. 1411 * 1412 * Hash types refer to the protocol layer addresses which are used to 1413 * construct a packet's hash. The hashes are used to differentiate or identify 1414 * flows of the protocol layer for the hash type. Hash types are either 1415 * layer-2 (L2), layer-3 (L3), or layer-4 (L4). 1416 * 1417 * Properties of hashes: 1418 * 1419 * 1) Two packets in different flows have different hash values 1420 * 2) Two packets in the same flow should have the same hash value 1421 * 1422 * A hash at a higher layer is considered to be more specific. A driver should 1423 * set the most specific hash possible. 1424 * 1425 * A driver cannot indicate a more specific hash than the layer at which a hash 1426 * was computed. For instance an L3 hash cannot be set as an L4 hash. 1427 * 1428 * A driver may indicate a hash level which is less specific than the 1429 * actual layer the hash was computed on. For instance, a hash computed 1430 * at L4 may be considered an L3 hash. This should only be done if the 1431 * driver can't unambiguously determine that the HW computed the hash at 1432 * the higher layer. Note that the "should" in the second property above 1433 * permits this. 1434 */ 1435enum pkt_hash_types { 1436 PKT_HASH_TYPE_NONE, /* Undefined type */ 1437 PKT_HASH_TYPE_L2, /* Input: src_MAC, dest_MAC */ 1438 PKT_HASH_TYPE_L3, /* Input: src_IP, dst_IP */ 1439 PKT_HASH_TYPE_L4, /* Input: src_IP, dst_IP, src_port, dst_port */ 1440}; 1441 1442static inline void skb_clear_hash(struct sk_buff *skb) 1443{ 1444 skb->hash = 0; 1445 skb->sw_hash = 0; 1446 skb->l4_hash = 0; 1447} 1448 1449static inline void skb_clear_hash_if_not_l4(struct sk_buff *skb) 1450{ 1451 if (!skb->l4_hash) 1452 skb_clear_hash(skb); 1453} 1454 1455static inline void 1456__skb_set_hash(struct sk_buff *skb, __u32 hash, bool is_sw, bool is_l4) 1457{ 1458 skb->l4_hash = is_l4; 1459 skb->sw_hash = is_sw; 1460 skb->hash = hash; 1461} 1462 1463static inline void 1464skb_set_hash(struct sk_buff *skb, __u32 hash, enum pkt_hash_types type) 1465{ 1466 /* Used by drivers to set hash from HW */ 1467 __skb_set_hash(skb, hash, false, type == PKT_HASH_TYPE_L4); 1468} 1469 1470static inline void 1471__skb_set_sw_hash(struct sk_buff *skb, __u32 hash, bool is_l4) 1472{ 1473 __skb_set_hash(skb, hash, true, is_l4); 1474} 1475 1476void __skb_get_hash(struct sk_buff *skb); 1477u32 __skb_get_hash_symmetric(const struct sk_buff *skb); 1478u32 skb_get_poff(const struct sk_buff *skb); 1479u32 __skb_get_poff(const struct sk_buff *skb, const void *data, 1480 const struct flow_keys_basic *keys, int hlen); 1481__be32 __skb_flow_get_ports(const struct sk_buff *skb, int thoff, u8 ip_proto, 1482 const void *data, int hlen_proto); 1483 1484static inline __be32 skb_flow_get_ports(const struct sk_buff *skb, 1485 int thoff, u8 ip_proto) 1486{ 1487 return __skb_flow_get_ports(skb, thoff, ip_proto, NULL, 0); 1488} 1489 1490void skb_flow_dissector_init(struct flow_dissector *flow_dissector, 1491 const struct flow_dissector_key *key, 1492 unsigned int key_count); 1493 1494struct bpf_flow_dissector; 1495u32 bpf_flow_dissect(struct bpf_prog *prog, struct bpf_flow_dissector *ctx, 1496 __be16 proto, int nhoff, int hlen, unsigned int flags); 1497 1498bool __skb_flow_dissect(const struct net *net, 1499 const struct sk_buff *skb, 1500 struct flow_dissector *flow_dissector, 1501 void *target_container, const void *data, 1502 __be16 proto, int nhoff, int hlen, unsigned int flags); 1503 1504static inline bool skb_flow_dissect(const struct sk_buff *skb, 1505 struct flow_dissector *flow_dissector, 1506 void *target_container, unsigned int flags) 1507{ 1508 return __skb_flow_dissect(NULL, skb, flow_dissector, 1509 target_container, NULL, 0, 0, 0, flags); 1510} 1511 1512static inline bool skb_flow_dissect_flow_keys(const struct sk_buff *skb, 1513 struct flow_keys *flow, 1514 unsigned int flags) 1515{ 1516 memset(flow, 0, sizeof(*flow)); 1517 return __skb_flow_dissect(NULL, skb, &flow_keys_dissector, 1518 flow, NULL, 0, 0, 0, flags); 1519} 1520 1521static inline bool 1522skb_flow_dissect_flow_keys_basic(const struct net *net, 1523 const struct sk_buff *skb, 1524 struct flow_keys_basic *flow, 1525 const void *data, __be16 proto, 1526 int nhoff, int hlen, unsigned int flags) 1527{ 1528 memset(flow, 0, sizeof(*flow)); 1529 return __skb_flow_dissect(net, skb, &flow_keys_basic_dissector, flow, 1530 data, proto, nhoff, hlen, flags); 1531} 1532 1533void skb_flow_dissect_meta(const struct sk_buff *skb, 1534 struct flow_dissector *flow_dissector, 1535 void *target_container); 1536 1537/* Gets a skb connection tracking info, ctinfo map should be a 1538 * map of mapsize to translate enum ip_conntrack_info states 1539 * to user states. 1540 */ 1541void 1542skb_flow_dissect_ct(const struct sk_buff *skb, 1543 struct flow_dissector *flow_dissector, 1544 void *target_container, 1545 u16 *ctinfo_map, size_t mapsize, 1546 bool post_ct, u16 zone); 1547void 1548skb_flow_dissect_tunnel_info(const struct sk_buff *skb, 1549 struct flow_dissector *flow_dissector, 1550 void *target_container); 1551 1552void skb_flow_dissect_hash(const struct sk_buff *skb, 1553 struct flow_dissector *flow_dissector, 1554 void *target_container); 1555 1556static inline __u32 skb_get_hash(struct sk_buff *skb) 1557{ 1558 if (!skb->l4_hash && !skb->sw_hash) 1559 __skb_get_hash(skb); 1560 1561 return skb->hash; 1562} 1563 1564static inline __u32 skb_get_hash_flowi6(struct sk_buff *skb, const struct flowi6 *fl6) 1565{ 1566 if (!skb->l4_hash && !skb->sw_hash) { 1567 struct flow_keys keys; 1568 __u32 hash = __get_hash_from_flowi6(fl6, &keys); 1569 1570 __skb_set_sw_hash(skb, hash, flow_keys_have_l4(&keys)); 1571 } 1572 1573 return skb->hash; 1574} 1575 1576__u32 skb_get_hash_perturb(const struct sk_buff *skb, 1577 const siphash_key_t *perturb); 1578 1579static inline __u32 skb_get_hash_raw(const struct sk_buff *skb) 1580{ 1581 return skb->hash; 1582} 1583 1584static inline void skb_copy_hash(struct sk_buff *to, const struct sk_buff *from) 1585{ 1586 to->hash = from->hash; 1587 to->sw_hash = from->sw_hash; 1588 to->l4_hash = from->l4_hash; 1589}; 1590 1591static inline int skb_cmp_decrypted(const struct sk_buff *skb1, 1592 const struct sk_buff *skb2) 1593{ 1594#ifdef CONFIG_TLS_DEVICE 1595 return skb2->decrypted - skb1->decrypted; 1596#else 1597 return 0; 1598#endif 1599} 1600 1601static inline void skb_copy_decrypted(struct sk_buff *to, 1602 const struct sk_buff *from) 1603{ 1604#ifdef CONFIG_TLS_DEVICE 1605 to->decrypted = from->decrypted; 1606#endif 1607} 1608 1609#ifdef NET_SKBUFF_DATA_USES_OFFSET 1610static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1611{ 1612 return skb->head + skb->end; 1613} 1614 1615static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1616{ 1617 return skb->end; 1618} 1619 1620static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1621{ 1622 skb->end = offset; 1623} 1624#else 1625static inline unsigned char *skb_end_pointer(const struct sk_buff *skb) 1626{ 1627 return skb->end; 1628} 1629 1630static inline unsigned int skb_end_offset(const struct sk_buff *skb) 1631{ 1632 return skb->end - skb->head; 1633} 1634 1635static inline void skb_set_end_offset(struct sk_buff *skb, unsigned int offset) 1636{ 1637 skb->end = skb->head + offset; 1638} 1639#endif 1640 1641struct ubuf_info *msg_zerocopy_realloc(struct sock *sk, size_t size, 1642 struct ubuf_info *uarg); 1643 1644void msg_zerocopy_put_abort(struct ubuf_info *uarg, bool have_uref); 1645 1646void msg_zerocopy_callback(struct sk_buff *skb, struct ubuf_info *uarg, 1647 bool success); 1648 1649int __zerocopy_sg_from_iter(struct msghdr *msg, struct sock *sk, 1650 struct sk_buff *skb, struct iov_iter *from, 1651 size_t length); 1652 1653static inline int skb_zerocopy_iter_dgram(struct sk_buff *skb, 1654 struct msghdr *msg, int len) 1655{ 1656 return __zerocopy_sg_from_iter(msg, skb->sk, skb, &msg->msg_iter, len); 1657} 1658 1659int skb_zerocopy_iter_stream(struct sock *sk, struct sk_buff *skb, 1660 struct msghdr *msg, int len, 1661 struct ubuf_info *uarg); 1662 1663/* Internal */ 1664#define skb_shinfo(SKB) ((struct skb_shared_info *)(skb_end_pointer(SKB))) 1665 1666static inline struct skb_shared_hwtstamps *skb_hwtstamps(struct sk_buff *skb) 1667{ 1668 return &skb_shinfo(skb)->hwtstamps; 1669} 1670 1671static inline struct ubuf_info *skb_zcopy(struct sk_buff *skb) 1672{ 1673 bool is_zcopy = skb && skb_shinfo(skb)->flags & SKBFL_ZEROCOPY_ENABLE; 1674 1675 return is_zcopy ? skb_uarg(skb) : NULL; 1676} 1677 1678static inline bool skb_zcopy_pure(const struct sk_buff *skb) 1679{ 1680 return skb_shinfo(skb)->flags & SKBFL_PURE_ZEROCOPY; 1681} 1682 1683static inline bool skb_zcopy_managed(const struct sk_buff *skb) 1684{ 1685 return skb_shinfo(skb)->flags & SKBFL_MANAGED_FRAG_REFS; 1686} 1687 1688static inline bool skb_pure_zcopy_same(const struct sk_buff *skb1, 1689 const struct sk_buff *skb2) 1690{ 1691 return skb_zcopy_pure(skb1) == skb_zcopy_pure(skb2); 1692} 1693 1694static inline void net_zcopy_get(struct ubuf_info *uarg) 1695{ 1696 refcount_inc(&uarg->refcnt); 1697} 1698 1699static inline void skb_zcopy_init(struct sk_buff *skb, struct ubuf_info *uarg) 1700{ 1701 skb_shinfo(skb)->destructor_arg = uarg; 1702 skb_shinfo(skb)->flags |= uarg->flags; 1703} 1704 1705static inline void skb_zcopy_set(struct sk_buff *skb, struct ubuf_info *uarg, 1706 bool *have_ref) 1707{ 1708 if (skb && uarg && !skb_zcopy(skb)) { 1709 if (unlikely(have_ref && *have_ref)) 1710 *have_ref = false; 1711 else 1712 net_zcopy_get(uarg); 1713 skb_zcopy_init(skb, uarg); 1714 } 1715} 1716 1717static inline void skb_zcopy_set_nouarg(struct sk_buff *skb, void *val) 1718{ 1719 skb_shinfo(skb)->destructor_arg = (void *)((uintptr_t) val | 0x1UL); 1720 skb_shinfo(skb)->flags |= SKBFL_ZEROCOPY_FRAG; 1721} 1722 1723static inline bool skb_zcopy_is_nouarg(struct sk_buff *skb) 1724{ 1725 return (uintptr_t) skb_shinfo(skb)->destructor_arg & 0x1UL; 1726} 1727 1728static inline void *skb_zcopy_get_nouarg(struct sk_buff *skb) 1729{ 1730 return (void *)((uintptr_t) skb_shinfo(skb)->destructor_arg & ~0x1UL); 1731} 1732 1733static inline void net_zcopy_put(struct ubuf_info *uarg) 1734{ 1735 if (uarg) 1736 uarg->callback(NULL, uarg, true); 1737} 1738 1739static inline void net_zcopy_put_abort(struct ubuf_info *uarg, bool have_uref) 1740{ 1741 if (uarg) { 1742 if (uarg->callback == msg_zerocopy_callback) 1743 msg_zerocopy_put_abort(uarg, have_uref); 1744 else if (have_uref) 1745 net_zcopy_put(uarg); 1746 } 1747} 1748 1749/* Release a reference on a zerocopy structure */ 1750static inline void skb_zcopy_clear(struct sk_buff *skb, bool zerocopy_success) 1751{ 1752 struct ubuf_info *uarg = skb_zcopy(skb); 1753 1754 if (uarg) { 1755 if (!skb_zcopy_is_nouarg(skb)) 1756 uarg->callback(skb, uarg, zerocopy_success); 1757 1758 skb_shinfo(skb)->flags &= ~SKBFL_ALL_ZEROCOPY; 1759 } 1760} 1761 1762void __skb_zcopy_downgrade_managed(struct sk_buff *skb); 1763 1764static inline void skb_zcopy_downgrade_managed(struct sk_buff *skb) 1765{ 1766 if (unlikely(skb_zcopy_managed(skb))) 1767 __skb_zcopy_downgrade_managed(skb); 1768} 1769 1770static inline void skb_mark_not_on_list(struct sk_buff *skb) 1771{ 1772 skb->next = NULL; 1773} 1774 1775static inline void skb_poison_list(struct sk_buff *skb) 1776{ 1777#ifdef CONFIG_DEBUG_NET 1778 skb->next = SKB_LIST_POISON_NEXT; 1779#endif 1780} 1781 1782/* Iterate through singly-linked GSO fragments of an skb. */ 1783#define skb_list_walk_safe(first, skb, next_skb) \ 1784 for ((skb) = (first), (next_skb) = (skb) ? (skb)->next : NULL; (skb); \ 1785 (skb) = (next_skb), (next_skb) = (skb) ? (skb)->next : NULL) 1786 1787static inline void skb_list_del_init(struct sk_buff *skb) 1788{ 1789 __list_del_entry(&skb->list); 1790 skb_mark_not_on_list(skb); 1791} 1792 1793/** 1794 * skb_queue_empty - check if a queue is empty 1795 * @list: queue head 1796 * 1797 * Returns true if the queue is empty, false otherwise. 1798 */ 1799static inline int skb_queue_empty(const struct sk_buff_head *list) 1800{ 1801 return list->next == (const struct sk_buff *) list; 1802} 1803 1804/** 1805 * skb_queue_empty_lockless - check if a queue is empty 1806 * @list: queue head 1807 * 1808 * Returns true if the queue is empty, false otherwise. 1809 * This variant can be used in lockless contexts. 1810 */ 1811static inline bool skb_queue_empty_lockless(const struct sk_buff_head *list) 1812{ 1813 return READ_ONCE(list->next) == (const struct sk_buff *) list; 1814} 1815 1816 1817/** 1818 * skb_queue_is_last - check if skb is the last entry in the queue 1819 * @list: queue head 1820 * @skb: buffer 1821 * 1822 * Returns true if @skb is the last buffer on the list. 1823 */ 1824static inline bool skb_queue_is_last(const struct sk_buff_head *list, 1825 const struct sk_buff *skb) 1826{ 1827 return skb->next == (const struct sk_buff *) list; 1828} 1829 1830/** 1831 * skb_queue_is_first - check if skb is the first entry in the queue 1832 * @list: queue head 1833 * @skb: buffer 1834 * 1835 * Returns true if @skb is the first buffer on the list. 1836 */ 1837static inline bool skb_queue_is_first(const struct sk_buff_head *list, 1838 const struct sk_buff *skb) 1839{ 1840 return skb->prev == (const struct sk_buff *) list; 1841} 1842 1843/** 1844 * skb_queue_next - return the next packet in the queue 1845 * @list: queue head 1846 * @skb: current buffer 1847 * 1848 * Return the next packet in @list after @skb. It is only valid to 1849 * call this if skb_queue_is_last() evaluates to false. 1850 */ 1851static inline struct sk_buff *skb_queue_next(const struct sk_buff_head *list, 1852 const struct sk_buff *skb) 1853{ 1854 /* This BUG_ON may seem severe, but if we just return then we 1855 * are going to dereference garbage. 1856 */ 1857 BUG_ON(skb_queue_is_last(list, skb)); 1858 return skb->next; 1859} 1860 1861/** 1862 * skb_queue_prev - return the prev packet in the queue 1863 * @list: queue head 1864 * @skb: current buffer 1865 * 1866 * Return the prev packet in @list before @skb. It is only valid to 1867 * call this if skb_queue_is_first() evaluates to false. 1868 */ 1869static inline struct sk_buff *skb_queue_prev(const struct sk_buff_head *list, 1870 const struct sk_buff *skb) 1871{ 1872 /* This BUG_ON may seem severe, but if we just return then we 1873 * are going to dereference garbage. 1874 */ 1875 BUG_ON(skb_queue_is_first(list, skb)); 1876 return skb->prev; 1877} 1878 1879/** 1880 * skb_get - reference buffer 1881 * @skb: buffer to reference 1882 * 1883 * Makes another reference to a socket buffer and returns a pointer 1884 * to the buffer. 1885 */ 1886static inline struct sk_buff *skb_get(struct sk_buff *skb) 1887{ 1888 refcount_inc(&skb->users); 1889 return skb; 1890} 1891 1892/* 1893 * If users == 1, we are the only owner and can avoid redundant atomic changes. 1894 */ 1895 1896/** 1897 * skb_cloned - is the buffer a clone 1898 * @skb: buffer to check 1899 * 1900 * Returns true if the buffer was generated with skb_clone() and is 1901 * one of multiple shared copies of the buffer. Cloned buffers are 1902 * shared data so must not be written to under normal circumstances. 1903 */ 1904static inline int skb_cloned(const struct sk_buff *skb) 1905{ 1906 return skb->cloned && 1907 (atomic_read(&skb_shinfo(skb)->dataref) & SKB_DATAREF_MASK) != 1; 1908} 1909 1910static inline int skb_unclone(struct sk_buff *skb, gfp_t pri) 1911{ 1912 might_sleep_if(gfpflags_allow_blocking(pri)); 1913 1914 if (skb_cloned(skb)) 1915 return pskb_expand_head(skb, 0, 0, pri); 1916 1917 return 0; 1918} 1919 1920/* This variant of skb_unclone() makes sure skb->truesize 1921 * and skb_end_offset() are not changed, whenever a new skb->head is needed. 1922 * 1923 * Indeed there is no guarantee that ksize(kmalloc(X)) == ksize(kmalloc(X)) 1924 * when various debugging features are in place. 1925 */ 1926int __skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri); 1927static inline int skb_unclone_keeptruesize(struct sk_buff *skb, gfp_t pri) 1928{ 1929 might_sleep_if(gfpflags_allow_blocking(pri)); 1930 1931 if (skb_cloned(skb)) 1932 return __skb_unclone_keeptruesize(skb, pri); 1933 return 0; 1934} 1935 1936/** 1937 * skb_header_cloned - is the header a clone 1938 * @skb: buffer to check 1939 * 1940 * Returns true if modifying the header part of the buffer requires 1941 * the data to be copied. 1942 */ 1943static inline int skb_header_cloned(const struct sk_buff *skb) 1944{ 1945 int dataref; 1946 1947 if (!skb->cloned) 1948 return 0; 1949 1950 dataref = atomic_read(&skb_shinfo(skb)->dataref); 1951 dataref = (dataref & SKB_DATAREF_MASK) - (dataref >> SKB_DATAREF_SHIFT); 1952 return dataref != 1; 1953} 1954 1955static inline int skb_header_unclone(struct sk_buff *skb, gfp_t pri) 1956{ 1957 might_sleep_if(gfpflags_allow_blocking(pri)); 1958 1959 if (skb_header_cloned(skb)) 1960 return pskb_expand_head(skb, 0, 0, pri); 1961 1962 return 0; 1963} 1964 1965/** 1966 * __skb_header_release() - allow clones to use the headroom 1967 * @skb: buffer to operate on 1968 * 1969 * See "DOC: dataref and headerless skbs". 1970 */ 1971static inline void __skb_header_release(struct sk_buff *skb) 1972{ 1973 skb->nohdr = 1; 1974 atomic_set(&skb_shinfo(skb)->dataref, 1 + (1 << SKB_DATAREF_SHIFT)); 1975} 1976 1977 1978/** 1979 * skb_shared - is the buffer shared 1980 * @skb: buffer to check 1981 * 1982 * Returns true if more than one person has a reference to this 1983 * buffer. 1984 */ 1985static inline int skb_shared(const struct sk_buff *skb) 1986{ 1987 return refcount_read(&skb->users) != 1; 1988} 1989 1990/** 1991 * skb_share_check - check if buffer is shared and if so clone it 1992 * @skb: buffer to check 1993 * @pri: priority for memory allocation 1994 * 1995 * If the buffer is shared the buffer is cloned and the old copy 1996 * drops a reference. A new clone with a single reference is returned. 1997 * If the buffer is not shared the original buffer is returned. When 1998 * being called from interrupt status or with spinlocks held pri must 1999 * be GFP_ATOMIC. 2000 * 2001 * NULL is returned on a memory allocation failure. 2002 */ 2003static inline struct sk_buff *skb_share_check(struct sk_buff *skb, gfp_t pri) 2004{ 2005 might_sleep_if(gfpflags_allow_blocking(pri)); 2006 if (skb_shared(skb)) { 2007 struct sk_buff *nskb = skb_clone(skb, pri); 2008 2009 if (likely(nskb)) 2010 consume_skb(skb); 2011 else 2012 kfree_skb(skb); 2013 skb = nskb; 2014 } 2015 return skb; 2016} 2017 2018/* 2019 * Copy shared buffers into a new sk_buff. We effectively do COW on 2020 * packets to handle cases where we have a local reader and forward 2021 * and a couple of other messy ones. The normal one is tcpdumping 2022 * a packet thats being forwarded. 2023 */ 2024 2025/** 2026 * skb_unshare - make a copy of a shared buffer 2027 * @skb: buffer to check 2028 * @pri: priority for memory allocation 2029 * 2030 * If the socket buffer is a clone then this function creates a new 2031 * copy of the data, drops a reference count on the old copy and returns 2032 * the new copy with the reference count at 1. If the buffer is not a clone 2033 * the original buffer is returned. When called with a spinlock held or 2034 * from interrupt state @pri must be %GFP_ATOMIC 2035 * 2036 * %NULL is returned on a memory allocation failure. 2037 */ 2038static inline struct sk_buff *skb_unshare(struct sk_buff *skb, 2039 gfp_t pri) 2040{ 2041 might_sleep_if(gfpflags_allow_blocking(pri)); 2042 if (skb_cloned(skb)) { 2043 struct sk_buff *nskb = skb_copy(skb, pri); 2044 2045 /* Free our shared copy */ 2046 if (likely(nskb)) 2047 consume_skb(skb); 2048 else 2049 kfree_skb(skb); 2050 skb = nskb; 2051 } 2052 return skb; 2053} 2054 2055/** 2056 * skb_peek - peek at the head of an &sk_buff_head 2057 * @list_: list to peek at 2058 * 2059 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2060 * be careful with this one. A peek leaves the buffer on the 2061 * list and someone else may run off with it. You must hold 2062 * the appropriate locks or have a private queue to do this. 2063 * 2064 * Returns %NULL for an empty list or a pointer to the head element. 2065 * The reference count is not incremented and the reference is therefore 2066 * volatile. Use with caution. 2067 */ 2068static inline struct sk_buff *skb_peek(const struct sk_buff_head *list_) 2069{ 2070 struct sk_buff *skb = list_->next; 2071 2072 if (skb == (struct sk_buff *)list_) 2073 skb = NULL; 2074 return skb; 2075} 2076 2077/** 2078 * __skb_peek - peek at the head of a non-empty &sk_buff_head 2079 * @list_: list to peek at 2080 * 2081 * Like skb_peek(), but the caller knows that the list is not empty. 2082 */ 2083static inline struct sk_buff *__skb_peek(const struct sk_buff_head *list_) 2084{ 2085 return list_->next; 2086} 2087 2088/** 2089 * skb_peek_next - peek skb following the given one from a queue 2090 * @skb: skb to start from 2091 * @list_: list to peek at 2092 * 2093 * Returns %NULL when the end of the list is met or a pointer to the 2094 * next element. The reference count is not incremented and the 2095 * reference is therefore volatile. Use with caution. 2096 */ 2097static inline struct sk_buff *skb_peek_next(struct sk_buff *skb, 2098 const struct sk_buff_head *list_) 2099{ 2100 struct sk_buff *next = skb->next; 2101 2102 if (next == (struct sk_buff *)list_) 2103 next = NULL; 2104 return next; 2105} 2106 2107/** 2108 * skb_peek_tail - peek at the tail of an &sk_buff_head 2109 * @list_: list to peek at 2110 * 2111 * Peek an &sk_buff. Unlike most other operations you _MUST_ 2112 * be careful with this one. A peek leaves the buffer on the 2113 * list and someone else may run off with it. You must hold 2114 * the appropriate locks or have a private queue to do this. 2115 * 2116 * Returns %NULL for an empty list or a pointer to the tail element. 2117 * The reference count is not incremented and the reference is therefore 2118 * volatile. Use with caution. 2119 */ 2120static inline struct sk_buff *skb_peek_tail(const struct sk_buff_head *list_) 2121{ 2122 struct sk_buff *skb = READ_ONCE(list_->prev); 2123 2124 if (skb == (struct sk_buff *)list_) 2125 skb = NULL; 2126 return skb; 2127 2128} 2129 2130/** 2131 * skb_queue_len - get queue length 2132 * @list_: list to measure 2133 * 2134 * Return the length of an &sk_buff queue. 2135 */ 2136static inline __u32 skb_queue_len(const struct sk_buff_head *list_) 2137{ 2138 return list_->qlen; 2139} 2140 2141/** 2142 * skb_queue_len_lockless - get queue length 2143 * @list_: list to measure 2144 * 2145 * Return the length of an &sk_buff queue. 2146 * This variant can be used in lockless contexts. 2147 */ 2148static inline __u32 skb_queue_len_lockless(const struct sk_buff_head *list_) 2149{ 2150 return READ_ONCE(list_->qlen); 2151} 2152 2153/** 2154 * __skb_queue_head_init - initialize non-spinlock portions of sk_buff_head 2155 * @list: queue to initialize 2156 * 2157 * This initializes only the list and queue length aspects of 2158 * an sk_buff_head object. This allows to initialize the list 2159 * aspects of an sk_buff_head without reinitializing things like 2160 * the spinlock. It can also be used for on-stack sk_buff_head 2161 * objects where the spinlock is known to not be used. 2162 */ 2163static inline void __skb_queue_head_init(struct sk_buff_head *list) 2164{ 2165 list->prev = list->next = (struct sk_buff *)list; 2166 list->qlen = 0; 2167} 2168 2169/* 2170 * This function creates a split out lock class for each invocation; 2171 * this is needed for now since a whole lot of users of the skb-queue 2172 * infrastructure in drivers have different locking usage (in hardirq) 2173 * than the networking core (in softirq only). In the long run either the 2174 * network layer or drivers should need annotation to consolidate the 2175 * main types of usage into 3 classes. 2176 */ 2177static inline void skb_queue_head_init(struct sk_buff_head *list) 2178{ 2179 spin_lock_init(&list->lock); 2180 __skb_queue_head_init(list); 2181} 2182 2183static inline void skb_queue_head_init_class(struct sk_buff_head *list, 2184 struct lock_class_key *class) 2185{ 2186 skb_queue_head_init(list); 2187 lockdep_set_class(&list->lock, class); 2188} 2189 2190/* 2191 * Insert an sk_buff on a list. 2192 * 2193 * The "__skb_xxxx()" functions are the non-atomic ones that 2194 * can only be called with interrupts disabled. 2195 */ 2196static inline void __skb_insert(struct sk_buff *newsk, 2197 struct sk_buff *prev, struct sk_buff *next, 2198 struct sk_buff_head *list) 2199{ 2200 /* See skb_queue_empty_lockless() and skb_peek_tail() 2201 * for the opposite READ_ONCE() 2202 */ 2203 WRITE_ONCE(newsk->next, next); 2204 WRITE_ONCE(newsk->prev, prev); 2205 WRITE_ONCE(((struct sk_buff_list *)next)->prev, newsk); 2206 WRITE_ONCE(((struct sk_buff_list *)prev)->next, newsk); 2207 WRITE_ONCE(list->qlen, list->qlen + 1); 2208} 2209 2210static inline void __skb_queue_splice(const struct sk_buff_head *list, 2211 struct sk_buff *prev, 2212 struct sk_buff *next) 2213{ 2214 struct sk_buff *first = list->next; 2215 struct sk_buff *last = list->prev; 2216 2217 WRITE_ONCE(first->prev, prev); 2218 WRITE_ONCE(prev->next, first); 2219 2220 WRITE_ONCE(last->next, next); 2221 WRITE_ONCE(next->prev, last); 2222} 2223 2224/** 2225 * skb_queue_splice - join two skb lists, this is designed for stacks 2226 * @list: the new list to add 2227 * @head: the place to add it in the first list 2228 */ 2229static inline void skb_queue_splice(const struct sk_buff_head *list, 2230 struct sk_buff_head *head) 2231{ 2232 if (!skb_queue_empty(list)) { 2233 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2234 head->qlen += list->qlen; 2235 } 2236} 2237 2238/** 2239 * skb_queue_splice_init - join two skb lists and reinitialise the emptied list 2240 * @list: the new list to add 2241 * @head: the place to add it in the first list 2242 * 2243 * The list at @list is reinitialised 2244 */ 2245static inline void skb_queue_splice_init(struct sk_buff_head *list, 2246 struct sk_buff_head *head) 2247{ 2248 if (!skb_queue_empty(list)) { 2249 __skb_queue_splice(list, (struct sk_buff *) head, head->next); 2250 head->qlen += list->qlen; 2251 __skb_queue_head_init(list); 2252 } 2253} 2254 2255/** 2256 * skb_queue_splice_tail - join two skb lists, each list being a queue 2257 * @list: the new list to add 2258 * @head: the place to add it in the first list 2259 */ 2260static inline void skb_queue_splice_tail(const struct sk_buff_head *list, 2261 struct sk_buff_head *head) 2262{ 2263 if (!skb_queue_empty(list)) { 2264 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2265 head->qlen += list->qlen; 2266 } 2267} 2268 2269/** 2270 * skb_queue_splice_tail_init - join two skb lists and reinitialise the emptied list 2271 * @list: the new list to add 2272 * @head: the place to add it in the first list 2273 * 2274 * Each of the lists is a queue. 2275 * The list at @list is reinitialised 2276 */ 2277static inline void skb_queue_splice_tail_init(struct sk_buff_head *list, 2278 struct sk_buff_head *head) 2279{ 2280 if (!skb_queue_empty(list)) { 2281 __skb_queue_splice(list, head->prev, (struct sk_buff *) head); 2282 head->qlen += list->qlen; 2283 __skb_queue_head_init(list); 2284 } 2285} 2286 2287/** 2288 * __skb_queue_after - queue a buffer at the list head 2289 * @list: list to use 2290 * @prev: place after this buffer 2291 * @newsk: buffer to queue 2292 * 2293 * Queue a buffer int the middle of a list. This function takes no locks 2294 * and you must therefore hold required locks before calling it. 2295 * 2296 * A buffer cannot be placed on two lists at the same time. 2297 */ 2298static inline void __skb_queue_after(struct sk_buff_head *list, 2299 struct sk_buff *prev, 2300 struct sk_buff *newsk) 2301{ 2302 __skb_insert(newsk, prev, ((struct sk_buff_list *)prev)->next, list); 2303} 2304 2305void skb_append(struct sk_buff *old, struct sk_buff *newsk, 2306 struct sk_buff_head *list); 2307 2308static inline void __skb_queue_before(struct sk_buff_head *list, 2309 struct sk_buff *next, 2310 struct sk_buff *newsk) 2311{ 2312 __skb_insert(newsk, ((struct sk_buff_list *)next)->prev, next, list); 2313} 2314 2315/** 2316 * __skb_queue_head - queue a buffer at the list head 2317 * @list: list to use 2318 * @newsk: buffer to queue 2319 * 2320 * Queue a buffer at the start of a list. This function takes no locks 2321 * and you must therefore hold required locks before calling it. 2322 * 2323 * A buffer cannot be placed on two lists at the same time. 2324 */ 2325static inline void __skb_queue_head(struct sk_buff_head *list, 2326 struct sk_buff *newsk) 2327{ 2328 __skb_queue_after(list, (struct sk_buff *)list, newsk); 2329} 2330void skb_queue_head(struct sk_buff_head *list, struct sk_buff *newsk); 2331 2332/** 2333 * __skb_queue_tail - queue a buffer at the list tail 2334 * @list: list to use 2335 * @newsk: buffer to queue 2336 * 2337 * Queue a buffer at the end of a list. This function takes no locks 2338 * and you must therefore hold required locks before calling it. 2339 * 2340 * A buffer cannot be placed on two lists at the same time. 2341 */ 2342static inline void __skb_queue_tail(struct sk_buff_head *list, 2343 struct sk_buff *newsk) 2344{ 2345 __skb_queue_before(list, (struct sk_buff *)list, newsk); 2346} 2347void skb_queue_tail(struct sk_buff_head *list, struct sk_buff *newsk); 2348 2349/* 2350 * remove sk_buff from list. _Must_ be called atomically, and with 2351 * the list known.. 2352 */ 2353void skb_unlink(struct sk_buff *skb, struct sk_buff_head *list); 2354static inline void __skb_unlink(struct sk_buff *skb, struct sk_buff_head *list) 2355{ 2356 struct sk_buff *next, *prev; 2357 2358 WRITE_ONCE(list->qlen, list->qlen - 1); 2359 next = skb->next; 2360 prev = skb->prev; 2361 skb->next = skb->prev = NULL; 2362 WRITE_ONCE(next->prev, prev); 2363 WRITE_ONCE(prev->next, next); 2364} 2365 2366/** 2367 * __skb_dequeue - remove from the head of the queue 2368 * @list: list to dequeue from 2369 * 2370 * Remove the head of the list. This function does not take any locks 2371 * so must be used with appropriate locks held only. The head item is 2372 * returned or %NULL if the list is empty. 2373 */ 2374static inline struct sk_buff *__skb_dequeue(struct sk_buff_head *list) 2375{ 2376 struct sk_buff *skb = skb_peek(list); 2377 if (skb) 2378 __skb_unlink(skb, list); 2379 return skb; 2380} 2381struct sk_buff *skb_dequeue(struct sk_buff_head *list); 2382 2383/** 2384 * __skb_dequeue_tail - remove from the tail of the queue 2385 * @list: list to dequeue from 2386 * 2387 * Remove the tail of the list. This function does not take any locks 2388 * so must be used with appropriate locks held only. The tail item is 2389 * returned or %NULL if the list is empty. 2390 */ 2391static inline struct sk_buff *__skb_dequeue_tail(struct sk_buff_head *list) 2392{ 2393 struct sk_buff *skb = skb_peek_tail(list); 2394 if (skb) 2395 __skb_unlink(skb, list); 2396 return skb; 2397} 2398struct sk_buff *skb_dequeue_tail(struct sk_buff_head *list); 2399 2400 2401static inline bool skb_is_nonlinear(const struct sk_buff *skb) 2402{ 2403 return skb->data_len; 2404} 2405 2406static inline unsigned int skb_headlen(const struct sk_buff *skb) 2407{ 2408 return skb->len - skb->data_len; 2409} 2410 2411static inline unsigned int __skb_pagelen(const struct sk_buff *skb) 2412{ 2413 unsigned int i, len = 0; 2414 2415 for (i = skb_shinfo(skb)->nr_frags - 1; (int)i >= 0; i--) 2416 len += skb_frag_size(&skb_shinfo(skb)->frags[i]); 2417 return len; 2418} 2419 2420static inline unsigned int skb_pagelen(const struct sk_buff *skb) 2421{ 2422 return skb_headlen(skb) + __skb_pagelen(skb); 2423} 2424 2425static inline void skb_frag_fill_page_desc(skb_frag_t *frag, 2426 struct page *page, 2427 int off, int size) 2428{ 2429 frag->bv_page = page; 2430 frag->bv_offset = off; 2431 skb_frag_size_set(frag, size); 2432} 2433 2434static inline void __skb_fill_page_desc_noacc(struct skb_shared_info *shinfo, 2435 int i, struct page *page, 2436 int off, int size) 2437{ 2438 skb_frag_t *frag = &shinfo->frags[i]; 2439 2440 skb_frag_fill_page_desc(frag, page, off, size); 2441} 2442 2443/** 2444 * skb_len_add - adds a number to len fields of skb 2445 * @skb: buffer to add len to 2446 * @delta: number of bytes to add 2447 */ 2448static inline void skb_len_add(struct sk_buff *skb, int delta) 2449{ 2450 skb->len += delta; 2451 skb->data_len += delta; 2452 skb->truesize += delta; 2453} 2454 2455/** 2456 * __skb_fill_page_desc - initialise a paged fragment in an skb 2457 * @skb: buffer containing fragment to be initialised 2458 * @i: paged fragment index to initialise 2459 * @page: the page to use for this fragment 2460 * @off: the offset to the data with @page 2461 * @size: the length of the data 2462 * 2463 * Initialises the @i'th fragment of @skb to point to &size bytes at 2464 * offset @off within @page. 2465 * 2466 * Does not take any additional reference on the fragment. 2467 */ 2468static inline void __skb_fill_page_desc(struct sk_buff *skb, int i, 2469 struct page *page, int off, int size) 2470{ 2471 __skb_fill_page_desc_noacc(skb_shinfo(skb), i, page, off, size); 2472 2473 /* Propagate page pfmemalloc to the skb if we can. The problem is 2474 * that not all callers have unique ownership of the page but rely 2475 * on page_is_pfmemalloc doing the right thing(tm). 2476 */ 2477 page = compound_head(page); 2478 if (page_is_pfmemalloc(page)) 2479 skb->pfmemalloc = true; 2480} 2481 2482/** 2483 * skb_fill_page_desc - initialise a paged fragment in an skb 2484 * @skb: buffer containing fragment to be initialised 2485 * @i: paged fragment index to initialise 2486 * @page: the page to use for this fragment 2487 * @off: the offset to the data with @page 2488 * @size: the length of the data 2489 * 2490 * As per __skb_fill_page_desc() -- initialises the @i'th fragment of 2491 * @skb to point to @size bytes at offset @off within @page. In 2492 * addition updates @skb such that @i is the last fragment. 2493 * 2494 * Does not take any additional reference on the fragment. 2495 */ 2496static inline void skb_fill_page_desc(struct sk_buff *skb, int i, 2497 struct page *page, int off, int size) 2498{ 2499 __skb_fill_page_desc(skb, i, page, off, size); 2500 skb_shinfo(skb)->nr_frags = i + 1; 2501} 2502 2503/** 2504 * skb_fill_page_desc_noacc - initialise a paged fragment in an skb 2505 * @skb: buffer containing fragment to be initialised 2506 * @i: paged fragment index to initialise 2507 * @page: the page to use for this fragment 2508 * @off: the offset to the data with @page 2509 * @size: the length of the data 2510 * 2511 * Variant of skb_fill_page_desc() which does not deal with 2512 * pfmemalloc, if page is not owned by us. 2513 */ 2514static inline void skb_fill_page_desc_noacc(struct sk_buff *skb, int i, 2515 struct page *page, int off, 2516 int size) 2517{ 2518 struct skb_shared_info *shinfo = skb_shinfo(skb); 2519 2520 __skb_fill_page_desc_noacc(shinfo, i, page, off, size); 2521 shinfo->nr_frags = i + 1; 2522} 2523 2524void skb_add_rx_frag(struct sk_buff *skb, int i, struct page *page, int off, 2525 int size, unsigned int truesize); 2526 2527void skb_coalesce_rx_frag(struct sk_buff *skb, int i, int size, 2528 unsigned int truesize); 2529 2530#define SKB_LINEAR_ASSERT(skb) BUG_ON(skb_is_nonlinear(skb)) 2531 2532#ifdef NET_SKBUFF_DATA_USES_OFFSET 2533static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2534{ 2535 return skb->head + skb->tail; 2536} 2537 2538static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2539{ 2540 skb->tail = skb->data - skb->head; 2541} 2542 2543static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2544{ 2545 skb_reset_tail_pointer(skb); 2546 skb->tail += offset; 2547} 2548 2549#else /* NET_SKBUFF_DATA_USES_OFFSET */ 2550static inline unsigned char *skb_tail_pointer(const struct sk_buff *skb) 2551{ 2552 return skb->tail; 2553} 2554 2555static inline void skb_reset_tail_pointer(struct sk_buff *skb) 2556{ 2557 skb->tail = skb->data; 2558} 2559 2560static inline void skb_set_tail_pointer(struct sk_buff *skb, const int offset) 2561{ 2562 skb->tail = skb->data + offset; 2563} 2564 2565#endif /* NET_SKBUFF_DATA_USES_OFFSET */ 2566 2567static inline void skb_assert_len(struct sk_buff *skb) 2568{ 2569#ifdef CONFIG_DEBUG_NET 2570 if (WARN_ONCE(!skb->len, "%s\n", __func__)) 2571 DO_ONCE_LITE(skb_dump, KERN_ERR, skb, false); 2572#endif /* CONFIG_DEBUG_NET */ 2573} 2574 2575/* 2576 * Add data to an sk_buff 2577 */ 2578void *pskb_put(struct sk_buff *skb, struct sk_buff *tail, int len); 2579void *skb_put(struct sk_buff *skb, unsigned int len); 2580static inline void *__skb_put(struct sk_buff *skb, unsigned int len) 2581{ 2582 void *tmp = skb_tail_pointer(skb); 2583 SKB_LINEAR_ASSERT(skb); 2584 skb->tail += len; 2585 skb->len += len; 2586 return tmp; 2587} 2588 2589static inline void *__skb_put_zero(struct sk_buff *skb, unsigned int len) 2590{ 2591 void *tmp = __skb_put(skb, len); 2592 2593 memset(tmp, 0, len); 2594 return tmp; 2595} 2596 2597static inline void *__skb_put_data(struct sk_buff *skb, const void *data, 2598 unsigned int len) 2599{ 2600 void *tmp = __skb_put(skb, len); 2601 2602 memcpy(tmp, data, len); 2603 return tmp; 2604} 2605 2606static inline void __skb_put_u8(struct sk_buff *skb, u8 val) 2607{ 2608 *(u8 *)__skb_put(skb, 1) = val; 2609} 2610 2611static inline void *skb_put_zero(struct sk_buff *skb, unsigned int len) 2612{ 2613 void *tmp = skb_put(skb, len); 2614 2615 memset(tmp, 0, len); 2616 2617 return tmp; 2618} 2619 2620static inline void *skb_put_data(struct sk_buff *skb, const void *data, 2621 unsigned int len) 2622{ 2623 void *tmp = skb_put(skb, len); 2624 2625 memcpy(tmp, data, len); 2626 2627 return tmp; 2628} 2629 2630static inline void skb_put_u8(struct sk_buff *skb, u8 val) 2631{ 2632 *(u8 *)skb_put(skb, 1) = val; 2633} 2634 2635void *skb_push(struct sk_buff *skb, unsigned int len); 2636static inline void *__skb_push(struct sk_buff *skb, unsigned int len) 2637{ 2638 skb->data -= len; 2639 skb->len += len; 2640 return skb->data; 2641} 2642 2643void *skb_pull(struct sk_buff *skb, unsigned int len); 2644static inline void *__skb_pull(struct sk_buff *skb, unsigned int len) 2645{ 2646 skb->len -= len; 2647 if (unlikely(skb->len < skb->data_len)) { 2648#if defined(CONFIG_DEBUG_NET) 2649 skb->len += len; 2650 pr_err("__skb_pull(len=%u)\n", len); 2651 skb_dump(KERN_ERR, skb, false); 2652#endif 2653 BUG(); 2654 } 2655 return skb->data += len; 2656} 2657 2658static inline void *skb_pull_inline(struct sk_buff *skb, unsigned int len) 2659{ 2660 return unlikely(len > skb->len) ? NULL : __skb_pull(skb, len); 2661} 2662 2663void *skb_pull_data(struct sk_buff *skb, size_t len); 2664 2665void *__pskb_pull_tail(struct sk_buff *skb, int delta); 2666 2667static inline enum skb_drop_reason 2668pskb_may_pull_reason(struct sk_buff *skb, unsigned int len) 2669{ 2670 if (likely(len <= skb_headlen(skb))) 2671 return SKB_NOT_DROPPED_YET; 2672 2673 if (unlikely(len > skb->len)) 2674 return SKB_DROP_REASON_PKT_TOO_SMALL; 2675 2676 if (unlikely(!__pskb_pull_tail(skb, len - skb_headlen(skb)))) 2677 return SKB_DROP_REASON_NOMEM; 2678 2679 return SKB_NOT_DROPPED_YET; 2680} 2681 2682static inline bool pskb_may_pull(struct sk_buff *skb, unsigned int len) 2683{ 2684 return pskb_may_pull_reason(skb, len) == SKB_NOT_DROPPED_YET; 2685} 2686 2687static inline void *pskb_pull(struct sk_buff *skb, unsigned int len) 2688{ 2689 if (!pskb_may_pull(skb, len)) 2690 return NULL; 2691 2692 skb->len -= len; 2693 return skb->data += len; 2694} 2695 2696void skb_condense(struct sk_buff *skb); 2697 2698/** 2699 * skb_headroom - bytes at buffer head 2700 * @skb: buffer to check 2701 * 2702 * Return the number of bytes of free space at the head of an &sk_buff. 2703 */ 2704static inline unsigned int skb_headroom(const struct sk_buff *skb) 2705{ 2706 return skb->data - skb->head; 2707} 2708 2709/** 2710 * skb_tailroom - bytes at buffer end 2711 * @skb: buffer to check 2712 * 2713 * Return the number of bytes of free space at the tail of an sk_buff 2714 */ 2715static inline int skb_tailroom(const struct sk_buff *skb) 2716{ 2717 return skb_is_nonlinear(skb) ? 0 : skb->end - skb->tail; 2718} 2719 2720/** 2721 * skb_availroom - bytes at buffer end 2722 * @skb: buffer to check 2723 * 2724 * Return the number of bytes of free space at the tail of an sk_buff 2725 * allocated by sk_stream_alloc() 2726 */ 2727static inline int skb_availroom(const struct sk_buff *skb) 2728{ 2729 if (skb_is_nonlinear(skb)) 2730 return 0; 2731 2732 return skb->end - skb->tail - skb->reserved_tailroom; 2733} 2734 2735/** 2736 * skb_reserve - adjust headroom 2737 * @skb: buffer to alter 2738 * @len: bytes to move 2739 * 2740 * Increase the headroom of an empty &sk_buff by reducing the tail 2741 * room. This is only allowed for an empty buffer. 2742 */ 2743static inline void skb_reserve(struct sk_buff *skb, int len) 2744{ 2745 skb->data += len; 2746 skb->tail += len; 2747} 2748 2749/** 2750 * skb_tailroom_reserve - adjust reserved_tailroom 2751 * @skb: buffer to alter 2752 * @mtu: maximum amount of headlen permitted 2753 * @needed_tailroom: minimum amount of reserved_tailroom 2754 * 2755 * Set reserved_tailroom so that headlen can be as large as possible but 2756 * not larger than mtu and tailroom cannot be smaller than 2757 * needed_tailroom. 2758 * The required headroom should already have been reserved before using 2759 * this function. 2760 */ 2761static inline void skb_tailroom_reserve(struct sk_buff *skb, unsigned int mtu, 2762 unsigned int needed_tailroom) 2763{ 2764 SKB_LINEAR_ASSERT(skb); 2765 if (mtu < skb_tailroom(skb) - needed_tailroom) 2766 /* use at most mtu */ 2767 skb->reserved_tailroom = skb_tailroom(skb) - mtu; 2768 else 2769 /* use up to all available space */ 2770 skb->reserved_tailroom = needed_tailroom; 2771} 2772 2773#define ENCAP_TYPE_ETHER 0 2774#define ENCAP_TYPE_IPPROTO 1 2775 2776static inline void skb_set_inner_protocol(struct sk_buff *skb, 2777 __be16 protocol) 2778{ 2779 skb->inner_protocol = protocol; 2780 skb->inner_protocol_type = ENCAP_TYPE_ETHER; 2781} 2782 2783static inline void skb_set_inner_ipproto(struct sk_buff *skb, 2784 __u8 ipproto) 2785{ 2786 skb->inner_ipproto = ipproto; 2787 skb->inner_protocol_type = ENCAP_TYPE_IPPROTO; 2788} 2789 2790static inline void skb_reset_inner_headers(struct sk_buff *skb) 2791{ 2792 skb->inner_mac_header = skb->mac_header; 2793 skb->inner_network_header = skb->network_header; 2794 skb->inner_transport_header = skb->transport_header; 2795} 2796 2797static inline void skb_reset_mac_len(struct sk_buff *skb) 2798{ 2799 skb->mac_len = skb->network_header - skb->mac_header; 2800} 2801 2802static inline unsigned char *skb_inner_transport_header(const struct sk_buff 2803 *skb) 2804{ 2805 return skb->head + skb->inner_transport_header; 2806} 2807 2808static inline int skb_inner_transport_offset(const struct sk_buff *skb) 2809{ 2810 return skb_inner_transport_header(skb) - skb->data; 2811} 2812 2813static inline void skb_reset_inner_transport_header(struct sk_buff *skb) 2814{ 2815 skb->inner_transport_header = skb->data - skb->head; 2816} 2817 2818static inline void skb_set_inner_transport_header(struct sk_buff *skb, 2819 const int offset) 2820{ 2821 skb_reset_inner_transport_header(skb); 2822 skb->inner_transport_header += offset; 2823} 2824 2825static inline unsigned char *skb_inner_network_header(const struct sk_buff *skb) 2826{ 2827 return skb->head + skb->inner_network_header; 2828} 2829 2830static inline void skb_reset_inner_network_header(struct sk_buff *skb) 2831{ 2832 skb->inner_network_header = skb->data - skb->head; 2833} 2834 2835static inline void skb_set_inner_network_header(struct sk_buff *skb, 2836 const int offset) 2837{ 2838 skb_reset_inner_network_header(skb); 2839 skb->inner_network_header += offset; 2840} 2841 2842static inline unsigned char *skb_inner_mac_header(const struct sk_buff *skb) 2843{ 2844 return skb->head + skb->inner_mac_header; 2845} 2846 2847static inline void skb_reset_inner_mac_header(struct sk_buff *skb) 2848{ 2849 skb->inner_mac_header = skb->data - skb->head; 2850} 2851 2852static inline void skb_set_inner_mac_header(struct sk_buff *skb, 2853 const int offset) 2854{ 2855 skb_reset_inner_mac_header(skb); 2856 skb->inner_mac_header += offset; 2857} 2858static inline bool skb_transport_header_was_set(const struct sk_buff *skb) 2859{ 2860 return skb->transport_header != (typeof(skb->transport_header))~0U; 2861} 2862 2863static inline unsigned char *skb_transport_header(const struct sk_buff *skb) 2864{ 2865 DEBUG_NET_WARN_ON_ONCE(!skb_transport_header_was_set(skb)); 2866 return skb->head + skb->transport_header; 2867} 2868 2869static inline void skb_reset_transport_header(struct sk_buff *skb) 2870{ 2871 skb->transport_header = skb->data - skb->head; 2872} 2873 2874static inline void skb_set_transport_header(struct sk_buff *skb, 2875 const int offset) 2876{ 2877 skb_reset_transport_header(skb); 2878 skb->transport_header += offset; 2879} 2880 2881static inline unsigned char *skb_network_header(const struct sk_buff *skb) 2882{ 2883 return skb->head + skb->network_header; 2884} 2885 2886static inline void skb_reset_network_header(struct sk_buff *skb) 2887{ 2888 skb->network_header = skb->data - skb->head; 2889} 2890 2891static inline void skb_set_network_header(struct sk_buff *skb, const int offset) 2892{ 2893 skb_reset_network_header(skb); 2894 skb->network_header += offset; 2895} 2896 2897static inline int skb_mac_header_was_set(const struct sk_buff *skb) 2898{ 2899 return skb->mac_header != (typeof(skb->mac_header))~0U; 2900} 2901 2902static inline unsigned char *skb_mac_header(const struct sk_buff *skb) 2903{ 2904 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2905 return skb->head + skb->mac_header; 2906} 2907 2908static inline int skb_mac_offset(const struct sk_buff *skb) 2909{ 2910 return skb_mac_header(skb) - skb->data; 2911} 2912 2913static inline u32 skb_mac_header_len(const struct sk_buff *skb) 2914{ 2915 DEBUG_NET_WARN_ON_ONCE(!skb_mac_header_was_set(skb)); 2916 return skb->network_header - skb->mac_header; 2917} 2918 2919static inline void skb_unset_mac_header(struct sk_buff *skb) 2920{ 2921 skb->mac_header = (typeof(skb->mac_header))~0U; 2922} 2923 2924static inline void skb_reset_mac_header(struct sk_buff *skb) 2925{ 2926 skb->mac_header = skb->data - skb->head; 2927} 2928 2929static inline void skb_set_mac_header(struct sk_buff *skb, const int offset) 2930{ 2931 skb_reset_mac_header(skb); 2932 skb->mac_header += offset; 2933} 2934 2935static inline void skb_pop_mac_header(struct sk_buff *skb) 2936{ 2937 skb->mac_header = skb->network_header; 2938} 2939 2940static inline void skb_probe_transport_header(struct sk_buff *skb) 2941{ 2942 struct flow_keys_basic keys; 2943 2944 if (skb_transport_header_was_set(skb)) 2945 return; 2946 2947 if (skb_flow_dissect_flow_keys_basic(NULL, skb, &keys, 2948 NULL, 0, 0, 0, 0)) 2949 skb_set_transport_header(skb, keys.control.thoff); 2950} 2951 2952static inline void skb_mac_header_rebuild(struct sk_buff *skb) 2953{ 2954 if (skb_mac_header_was_set(skb)) { 2955 const unsigned char *old_mac = skb_mac_header(skb); 2956 2957 skb_set_mac_header(skb, -skb->mac_len); 2958 memmove(skb_mac_header(skb), old_mac, skb->mac_len); 2959 } 2960} 2961 2962static inline int skb_checksum_start_offset(const struct sk_buff *skb) 2963{ 2964 return skb->csum_start - skb_headroom(skb); 2965} 2966 2967static inline unsigned char *skb_checksum_start(const struct sk_buff *skb) 2968{ 2969 return skb->head + skb->csum_start; 2970} 2971 2972static inline int skb_transport_offset(const struct sk_buff *skb) 2973{ 2974 return skb_transport_header(skb) - skb->data; 2975} 2976 2977static inline u32 skb_network_header_len(const struct sk_buff *skb) 2978{ 2979 return skb->transport_header - skb->network_header; 2980} 2981 2982static inline u32 skb_inner_network_header_len(const struct sk_buff *skb) 2983{ 2984 return skb->inner_transport_header - skb->inner_network_header; 2985} 2986 2987static inline int skb_network_offset(const struct sk_buff *skb) 2988{ 2989 return skb_network_header(skb) - skb->data; 2990} 2991 2992static inline int skb_inner_network_offset(const struct sk_buff *skb) 2993{ 2994 return skb_inner_network_header(skb) - skb->data; 2995} 2996 2997static inline int pskb_network_may_pull(struct sk_buff *skb, unsigned int len) 2998{ 2999 return pskb_may_pull(skb, skb_network_offset(skb) + len); 3000} 3001 3002/* 3003 * CPUs often take a performance hit when accessing unaligned memory 3004 * locations. The actual performance hit varies, it can be small if the 3005 * hardware handles it or large if we have to take an exception and fix it 3006 * in software. 3007 * 3008 * Since an ethernet header is 14 bytes network drivers often end up with 3009 * the IP header at an unaligned offset. The IP header can be aligned by 3010 * shifting the start of the packet by 2 bytes. Drivers should do this 3011 * with: 3012 * 3013 * skb_reserve(skb, NET_IP_ALIGN); 3014 * 3015 * The downside to this alignment of the IP header is that the DMA is now 3016 * unaligned. On some architectures the cost of an unaligned DMA is high 3017 * and this cost outweighs the gains made by aligning the IP header. 3018 * 3019 * Since this trade off varies between architectures, we allow NET_IP_ALIGN 3020 * to be overridden. 3021 */ 3022#ifndef NET_IP_ALIGN 3023#define NET_IP_ALIGN 2 3024#endif 3025 3026/* 3027 * The networking layer reserves some headroom in skb data (via 3028 * dev_alloc_skb). This is used to avoid having to reallocate skb data when 3029 * the header has to grow. In the default case, if the header has to grow 3030 * 32 bytes or less we avoid the reallocation. 3031 * 3032 * Unfortunately this headroom changes the DMA alignment of the resulting 3033 * network packet. As for NET_IP_ALIGN, this unaligned DMA is expensive 3034 * on some architectures. An architecture can override this value, 3035 * perhaps setting it to a cacheline in size (since that will maintain 3036 * cacheline alignment of the DMA). It must be a power of 2. 3037 * 3038 * Various parts of the networking layer expect at least 32 bytes of 3039 * headroom, you should not reduce this. 3040 * 3041 * Using max(32, L1_CACHE_BYTES) makes sense (especially with RPS) 3042 * to reduce average number of cache lines per packet. 3043 * get_rps_cpu() for example only access one 64 bytes aligned block : 3044 * NET_IP_ALIGN(2) + ethernet_header(14) + IP_header(20/40) + ports(8) 3045 */ 3046#ifndef NET_SKB_PAD 3047#define NET_SKB_PAD max(32, L1_CACHE_BYTES) 3048#endif 3049 3050int ___pskb_trim(struct sk_buff *skb, unsigned int len); 3051 3052static inline void __skb_set_length(struct sk_buff *skb, unsigned int len) 3053{ 3054 if (WARN_ON(skb_is_nonlinear(skb))) 3055 return; 3056 skb->len = len; 3057 skb_set_tail_pointer(skb, len); 3058} 3059 3060static inline void __skb_trim(struct sk_buff *skb, unsigned int len) 3061{ 3062 __skb_set_length(skb, len); 3063} 3064 3065void skb_trim(struct sk_buff *skb, unsigned int len); 3066 3067static inline int __pskb_trim(struct sk_buff *skb, unsigned int len) 3068{ 3069 if (skb->data_len) 3070 return ___pskb_trim(skb, len); 3071 __skb_trim(skb, len); 3072 return 0; 3073} 3074 3075static inline int pskb_trim(struct sk_buff *skb, unsigned int len) 3076{ 3077 return (len < skb->len) ? __pskb_trim(skb, len) : 0; 3078} 3079 3080/** 3081 * pskb_trim_unique - remove end from a paged unique (not cloned) buffer 3082 * @skb: buffer to alter 3083 * @len: new length 3084 * 3085 * This is identical to pskb_trim except that the caller knows that 3086 * the skb is not cloned so we should never get an error due to out- 3087 * of-memory. 3088 */ 3089static inline void pskb_trim_unique(struct sk_buff *skb, unsigned int len) 3090{ 3091 int err = pskb_trim(skb, len); 3092 BUG_ON(err); 3093} 3094 3095static inline int __skb_grow(struct sk_buff *skb, unsigned int len) 3096{ 3097 unsigned int diff = len - skb->len; 3098 3099 if (skb_tailroom(skb) < diff) { 3100 int ret = pskb_expand_head(skb, 0, diff - skb_tailroom(skb), 3101 GFP_ATOMIC); 3102 if (ret) 3103 return ret; 3104 } 3105 __skb_set_length(skb, len); 3106 return 0; 3107} 3108 3109/** 3110 * skb_orphan - orphan a buffer 3111 * @skb: buffer to orphan 3112 * 3113 * If a buffer currently has an owner then we call the owner's 3114 * destructor function and make the @skb unowned. The buffer continues 3115 * to exist but is no longer charged to its former owner. 3116 */ 3117static inline void skb_orphan(struct sk_buff *skb) 3118{ 3119 if (skb->destructor) { 3120 skb->destructor(skb); 3121 skb->destructor = NULL; 3122 skb->sk = NULL; 3123 } else { 3124 BUG_ON(skb->sk); 3125 } 3126} 3127 3128/** 3129 * skb_orphan_frags - orphan the frags contained in a buffer 3130 * @skb: buffer to orphan frags from 3131 * @gfp_mask: allocation mask for replacement pages 3132 * 3133 * For each frag in the SKB which needs a destructor (i.e. has an 3134 * owner) create a copy of that frag and release the original 3135 * page by calling the destructor. 3136 */ 3137static inline int skb_orphan_frags(struct sk_buff *skb, gfp_t gfp_mask) 3138{ 3139 if (likely(!skb_zcopy(skb))) 3140 return 0; 3141 if (skb_shinfo(skb)->flags & SKBFL_DONT_ORPHAN) 3142 return 0; 3143 return skb_copy_ubufs(skb, gfp_mask); 3144} 3145 3146/* Frags must be orphaned, even if refcounted, if skb might loop to rx path */ 3147static inline int skb_orphan_frags_rx(struct sk_buff *skb, gfp_t gfp_mask) 3148{ 3149 if (likely(!skb_zcopy(skb))) 3150 return 0; 3151 return skb_copy_ubufs(skb, gfp_mask); 3152} 3153 3154/** 3155 * __skb_queue_purge - empty a list 3156 * @list: list to empty 3157 * 3158 * Delete all buffers on an &sk_buff list. Each buffer is removed from 3159 * the list and one reference dropped. This function does not take the 3160 * list lock and the caller must hold the relevant locks to use it. 3161 */ 3162static inline void __skb_queue_purge(struct sk_buff_head *list) 3163{ 3164 struct sk_buff *skb; 3165 while ((skb = __skb_dequeue(list)) != NULL) 3166 kfree_skb(skb); 3167} 3168void skb_queue_purge(struct sk_buff_head *list); 3169 3170unsigned int skb_rbtree_purge(struct rb_root *root); 3171 3172void *__netdev_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3173 3174/** 3175 * netdev_alloc_frag - allocate a page fragment 3176 * @fragsz: fragment size 3177 * 3178 * Allocates a frag from a page for receive buffer. 3179 * Uses GFP_ATOMIC allocations. 3180 */ 3181static inline void *netdev_alloc_frag(unsigned int fragsz) 3182{ 3183 return __netdev_alloc_frag_align(fragsz, ~0u); 3184} 3185 3186static inline void *netdev_alloc_frag_align(unsigned int fragsz, 3187 unsigned int align) 3188{ 3189 WARN_ON_ONCE(!is_power_of_2(align)); 3190 return __netdev_alloc_frag_align(fragsz, -align); 3191} 3192 3193struct sk_buff *__netdev_alloc_skb(struct net_device *dev, unsigned int length, 3194 gfp_t gfp_mask); 3195 3196/** 3197 * netdev_alloc_skb - allocate an skbuff for rx on a specific device 3198 * @dev: network device to receive on 3199 * @length: length to allocate 3200 * 3201 * Allocate a new &sk_buff and assign it a usage count of one. The 3202 * buffer has unspecified headroom built in. Users should allocate 3203 * the headroom they think they need without accounting for the 3204 * built in space. The built in space is used for optimisations. 3205 * 3206 * %NULL is returned if there is no free memory. Although this function 3207 * allocates memory it can be called from an interrupt. 3208 */ 3209static inline struct sk_buff *netdev_alloc_skb(struct net_device *dev, 3210 unsigned int length) 3211{ 3212 return __netdev_alloc_skb(dev, length, GFP_ATOMIC); 3213} 3214 3215/* legacy helper around __netdev_alloc_skb() */ 3216static inline struct sk_buff *__dev_alloc_skb(unsigned int length, 3217 gfp_t gfp_mask) 3218{ 3219 return __netdev_alloc_skb(NULL, length, gfp_mask); 3220} 3221 3222/* legacy helper around netdev_alloc_skb() */ 3223static inline struct sk_buff *dev_alloc_skb(unsigned int length) 3224{ 3225 return netdev_alloc_skb(NULL, length); 3226} 3227 3228 3229static inline struct sk_buff *__netdev_alloc_skb_ip_align(struct net_device *dev, 3230 unsigned int length, gfp_t gfp) 3231{ 3232 struct sk_buff *skb = __netdev_alloc_skb(dev, length + NET_IP_ALIGN, gfp); 3233 3234 if (NET_IP_ALIGN && skb) 3235 skb_reserve(skb, NET_IP_ALIGN); 3236 return skb; 3237} 3238 3239static inline struct sk_buff *netdev_alloc_skb_ip_align(struct net_device *dev, 3240 unsigned int length) 3241{ 3242 return __netdev_alloc_skb_ip_align(dev, length, GFP_ATOMIC); 3243} 3244 3245static inline void skb_free_frag(void *addr) 3246{ 3247 page_frag_free(addr); 3248} 3249 3250void *__napi_alloc_frag_align(unsigned int fragsz, unsigned int align_mask); 3251 3252static inline void *napi_alloc_frag(unsigned int fragsz) 3253{ 3254 return __napi_alloc_frag_align(fragsz, ~0u); 3255} 3256 3257static inline void *napi_alloc_frag_align(unsigned int fragsz, 3258 unsigned int align) 3259{ 3260 WARN_ON_ONCE(!is_power_of_2(align)); 3261 return __napi_alloc_frag_align(fragsz, -align); 3262} 3263 3264struct sk_buff *__napi_alloc_skb(struct napi_struct *napi, 3265 unsigned int length, gfp_t gfp_mask); 3266static inline struct sk_buff *napi_alloc_skb(struct napi_struct *napi, 3267 unsigned int length) 3268{ 3269 return __napi_alloc_skb(napi, length, GFP_ATOMIC); 3270} 3271void napi_consume_skb(struct sk_buff *skb, int budget); 3272 3273void napi_skb_free_stolen_head(struct sk_buff *skb); 3274void __napi_kfree_skb(struct sk_buff *skb, enum skb_drop_reason reason); 3275 3276/** 3277 * __dev_alloc_pages - allocate page for network Rx 3278 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3279 * @order: size of the allocation 3280 * 3281 * Allocate a new page. 3282 * 3283 * %NULL is returned if there is no free memory. 3284*/ 3285static inline struct page *__dev_alloc_pages(gfp_t gfp_mask, 3286 unsigned int order) 3287{ 3288 /* This piece of code contains several assumptions. 3289 * 1. This is for device Rx, therefor a cold page is preferred. 3290 * 2. The expectation is the user wants a compound page. 3291 * 3. If requesting a order 0 page it will not be compound 3292 * due to the check to see if order has a value in prep_new_page 3293 * 4. __GFP_MEMALLOC is ignored if __GFP_NOMEMALLOC is set due to 3294 * code in gfp_to_alloc_flags that should be enforcing this. 3295 */ 3296 gfp_mask |= __GFP_COMP | __GFP_MEMALLOC; 3297 3298 return alloc_pages_node(NUMA_NO_NODE, gfp_mask, order); 3299} 3300 3301static inline struct page *dev_alloc_pages(unsigned int order) 3302{ 3303 return __dev_alloc_pages(GFP_ATOMIC | __GFP_NOWARN, order); 3304} 3305 3306/** 3307 * __dev_alloc_page - allocate a page for network Rx 3308 * @gfp_mask: allocation priority. Set __GFP_NOMEMALLOC if not for network Rx 3309 * 3310 * Allocate a new page. 3311 * 3312 * %NULL is returned if there is no free memory. 3313 */ 3314static inline struct page *__dev_alloc_page(gfp_t gfp_mask) 3315{ 3316 return __dev_alloc_pages(gfp_mask, 0); 3317} 3318 3319static inline struct page *dev_alloc_page(void) 3320{ 3321 return dev_alloc_pages(0); 3322} 3323 3324/** 3325 * dev_page_is_reusable - check whether a page can be reused for network Rx 3326 * @page: the page to test 3327 * 3328 * A page shouldn't be considered for reusing/recycling if it was allocated 3329 * under memory pressure or at a distant memory node. 3330 * 3331 * Returns false if this page should be returned to page allocator, true 3332 * otherwise. 3333 */ 3334static inline bool dev_page_is_reusable(const struct page *page) 3335{ 3336 return likely(page_to_nid(page) == numa_mem_id() && 3337 !page_is_pfmemalloc(page)); 3338} 3339 3340/** 3341 * skb_propagate_pfmemalloc - Propagate pfmemalloc if skb is allocated after RX page 3342 * @page: The page that was allocated from skb_alloc_page 3343 * @skb: The skb that may need pfmemalloc set 3344 */ 3345static inline void skb_propagate_pfmemalloc(const struct page *page, 3346 struct sk_buff *skb) 3347{ 3348 if (page_is_pfmemalloc(page)) 3349 skb->pfmemalloc = true; 3350} 3351 3352/** 3353 * skb_frag_off() - Returns the offset of a skb fragment 3354 * @frag: the paged fragment 3355 */ 3356static inline unsigned int skb_frag_off(const skb_frag_t *frag) 3357{ 3358 return frag->bv_offset; 3359} 3360 3361/** 3362 * skb_frag_off_add() - Increments the offset of a skb fragment by @delta 3363 * @frag: skb fragment 3364 * @delta: value to add 3365 */ 3366static inline void skb_frag_off_add(skb_frag_t *frag, int delta) 3367{ 3368 frag->bv_offset += delta; 3369} 3370 3371/** 3372 * skb_frag_off_set() - Sets the offset of a skb fragment 3373 * @frag: skb fragment 3374 * @offset: offset of fragment 3375 */ 3376static inline void skb_frag_off_set(skb_frag_t *frag, unsigned int offset) 3377{ 3378 frag->bv_offset = offset; 3379} 3380 3381/** 3382 * skb_frag_off_copy() - Sets the offset of a skb fragment from another fragment 3383 * @fragto: skb fragment where offset is set 3384 * @fragfrom: skb fragment offset is copied from 3385 */ 3386static inline void skb_frag_off_copy(skb_frag_t *fragto, 3387 const skb_frag_t *fragfrom) 3388{ 3389 fragto->bv_offset = fragfrom->bv_offset; 3390} 3391 3392/** 3393 * skb_frag_page - retrieve the page referred to by a paged fragment 3394 * @frag: the paged fragment 3395 * 3396 * Returns the &struct page associated with @frag. 3397 */ 3398static inline struct page *skb_frag_page(const skb_frag_t *frag) 3399{ 3400 return frag->bv_page; 3401} 3402 3403/** 3404 * __skb_frag_ref - take an addition reference on a paged fragment. 3405 * @frag: the paged fragment 3406 * 3407 * Takes an additional reference on the paged fragment @frag. 3408 */ 3409static inline void __skb_frag_ref(skb_frag_t *frag) 3410{ 3411 get_page(skb_frag_page(frag)); 3412} 3413 3414/** 3415 * skb_frag_ref - take an addition reference on a paged fragment of an skb. 3416 * @skb: the buffer 3417 * @f: the fragment offset. 3418 * 3419 * Takes an additional reference on the @f'th paged fragment of @skb. 3420 */ 3421static inline void skb_frag_ref(struct sk_buff *skb, int f) 3422{ 3423 __skb_frag_ref(&skb_shinfo(skb)->frags[f]); 3424} 3425 3426static inline void 3427napi_frag_unref(skb_frag_t *frag, bool recycle, bool napi_safe) 3428{ 3429 struct page *page = skb_frag_page(frag); 3430 3431#ifdef CONFIG_PAGE_POOL 3432 if (recycle && page_pool_return_skb_page(page, napi_safe)) 3433 return; 3434#endif 3435 put_page(page); 3436} 3437 3438/** 3439 * __skb_frag_unref - release a reference on a paged fragment. 3440 * @frag: the paged fragment 3441 * @recycle: recycle the page if allocated via page_pool 3442 * 3443 * Releases a reference on the paged fragment @frag 3444 * or recycles the page via the page_pool API. 3445 */ 3446static inline void __skb_frag_unref(skb_frag_t *frag, bool recycle) 3447{ 3448 napi_frag_unref(frag, recycle, false); 3449} 3450 3451/** 3452 * skb_frag_unref - release a reference on a paged fragment of an skb. 3453 * @skb: the buffer 3454 * @f: the fragment offset 3455 * 3456 * Releases a reference on the @f'th paged fragment of @skb. 3457 */ 3458static inline void skb_frag_unref(struct sk_buff *skb, int f) 3459{ 3460 struct skb_shared_info *shinfo = skb_shinfo(skb); 3461 3462 if (!skb_zcopy_managed(skb)) 3463 __skb_frag_unref(&shinfo->frags[f], skb->pp_recycle); 3464} 3465 3466/** 3467 * skb_frag_address - gets the address of the data contained in a paged fragment 3468 * @frag: the paged fragment buffer 3469 * 3470 * Returns the address of the data within @frag. The page must already 3471 * be mapped. 3472 */ 3473static inline void *skb_frag_address(const skb_frag_t *frag) 3474{ 3475 return page_address(skb_frag_page(frag)) + skb_frag_off(frag); 3476} 3477 3478/** 3479 * skb_frag_address_safe - gets the address of the data contained in a paged fragment 3480 * @frag: the paged fragment buffer 3481 * 3482 * Returns the address of the data within @frag. Checks that the page 3483 * is mapped and returns %NULL otherwise. 3484 */ 3485static inline void *skb_frag_address_safe(const skb_frag_t *frag) 3486{ 3487 void *ptr = page_address(skb_frag_page(frag)); 3488 if (unlikely(!ptr)) 3489 return NULL; 3490 3491 return ptr + skb_frag_off(frag); 3492} 3493 3494/** 3495 * skb_frag_page_copy() - sets the page in a fragment from another fragment 3496 * @fragto: skb fragment where page is set 3497 * @fragfrom: skb fragment page is copied from 3498 */ 3499static inline void skb_frag_page_copy(skb_frag_t *fragto, 3500 const skb_frag_t *fragfrom) 3501{ 3502 fragto->bv_page = fragfrom->bv_page; 3503} 3504 3505bool skb_page_frag_refill(unsigned int sz, struct page_frag *pfrag, gfp_t prio); 3506 3507/** 3508 * skb_frag_dma_map - maps a paged fragment via the DMA API 3509 * @dev: the device to map the fragment to 3510 * @frag: the paged fragment to map 3511 * @offset: the offset within the fragment (starting at the 3512 * fragment's own offset) 3513 * @size: the number of bytes to map 3514 * @dir: the direction of the mapping (``PCI_DMA_*``) 3515 * 3516 * Maps the page associated with @frag to @device. 3517 */ 3518static inline dma_addr_t skb_frag_dma_map(struct device *dev, 3519 const skb_frag_t *frag, 3520 size_t offset, size_t size, 3521 enum dma_data_direction dir) 3522{ 3523 return dma_map_page(dev, skb_frag_page(frag), 3524 skb_frag_off(frag) + offset, size, dir); 3525} 3526 3527static inline struct sk_buff *pskb_copy(struct sk_buff *skb, 3528 gfp_t gfp_mask) 3529{ 3530 return __pskb_copy(skb, skb_headroom(skb), gfp_mask); 3531} 3532 3533 3534static inline struct sk_buff *pskb_copy_for_clone(struct sk_buff *skb, 3535 gfp_t gfp_mask) 3536{ 3537 return __pskb_copy_fclone(skb, skb_headroom(skb), gfp_mask, true); 3538} 3539 3540 3541/** 3542 * skb_clone_writable - is the header of a clone writable 3543 * @skb: buffer to check 3544 * @len: length up to which to write 3545 * 3546 * Returns true if modifying the header part of the cloned buffer 3547 * does not requires the data to be copied. 3548 */ 3549static inline int skb_clone_writable(const struct sk_buff *skb, unsigned int len) 3550{ 3551 return !skb_header_cloned(skb) && 3552 skb_headroom(skb) + len <= skb->hdr_len; 3553} 3554 3555static inline int skb_try_make_writable(struct sk_buff *skb, 3556 unsigned int write_len) 3557{ 3558 return skb_cloned(skb) && !skb_clone_writable(skb, write_len) && 3559 pskb_expand_head(skb, 0, 0, GFP_ATOMIC); 3560} 3561 3562static inline int __skb_cow(struct sk_buff *skb, unsigned int headroom, 3563 int cloned) 3564{ 3565 int delta = 0; 3566 3567 if (headroom > skb_headroom(skb)) 3568 delta = headroom - skb_headroom(skb); 3569 3570 if (delta || cloned) 3571 return pskb_expand_head(skb, ALIGN(delta, NET_SKB_PAD), 0, 3572 GFP_ATOMIC); 3573 return 0; 3574} 3575 3576/** 3577 * skb_cow - copy header of skb when it is required 3578 * @skb: buffer to cow 3579 * @headroom: needed headroom 3580 * 3581 * If the skb passed lacks sufficient headroom or its data part 3582 * is shared, data is reallocated. If reallocation fails, an error 3583 * is returned and original skb is not changed. 3584 * 3585 * The result is skb with writable area skb->head...skb->tail 3586 * and at least @headroom of space at head. 3587 */ 3588static inline int skb_cow(struct sk_buff *skb, unsigned int headroom) 3589{ 3590 return __skb_cow(skb, headroom, skb_cloned(skb)); 3591} 3592 3593/** 3594 * skb_cow_head - skb_cow but only making the head writable 3595 * @skb: buffer to cow 3596 * @headroom: needed headroom 3597 * 3598 * This function is identical to skb_cow except that we replace the 3599 * skb_cloned check by skb_header_cloned. It should be used when 3600 * you only need to push on some header and do not need to modify 3601 * the data. 3602 */ 3603static inline int skb_cow_head(struct sk_buff *skb, unsigned int headroom) 3604{ 3605 return __skb_cow(skb, headroom, skb_header_cloned(skb)); 3606} 3607 3608/** 3609 * skb_padto - pad an skbuff up to a minimal size 3610 * @skb: buffer to pad 3611 * @len: minimal length 3612 * 3613 * Pads up a buffer to ensure the trailing bytes exist and are 3614 * blanked. If the buffer already contains sufficient data it 3615 * is untouched. Otherwise it is extended. Returns zero on 3616 * success. The skb is freed on error. 3617 */ 3618static inline int skb_padto(struct sk_buff *skb, unsigned int len) 3619{ 3620 unsigned int size = skb->len; 3621 if (likely(size >= len)) 3622 return 0; 3623 return skb_pad(skb, len - size); 3624} 3625 3626/** 3627 * __skb_put_padto - increase size and pad an skbuff up to a minimal size 3628 * @skb: buffer to pad 3629 * @len: minimal length 3630 * @free_on_error: free buffer on error 3631 * 3632 * Pads up a buffer to ensure the trailing bytes exist and are 3633 * blanked. If the buffer already contains sufficient data it 3634 * is untouched. Otherwise it is extended. Returns zero on 3635 * success. The skb is freed on error if @free_on_error is true. 3636 */ 3637static inline int __must_check __skb_put_padto(struct sk_buff *skb, 3638 unsigned int len, 3639 bool free_on_error) 3640{ 3641 unsigned int size = skb->len; 3642 3643 if (unlikely(size < len)) { 3644 len -= size; 3645 if (__skb_pad(skb, len, free_on_error)) 3646 return -ENOMEM; 3647 __skb_put(skb, len); 3648 } 3649 return 0; 3650} 3651 3652/** 3653 * skb_put_padto - increase size and pad an skbuff up to a minimal size 3654 * @skb: buffer to pad 3655 * @len: minimal length 3656 * 3657 * Pads up a buffer to ensure the trailing bytes exist and are 3658 * blanked. If the buffer already contains sufficient data it 3659 * is untouched. Otherwise it is extended. Returns zero on 3660 * success. The skb is freed on error. 3661 */ 3662static inline int __must_check skb_put_padto(struct sk_buff *skb, unsigned int len) 3663{ 3664 return __skb_put_padto(skb, len, true); 3665} 3666 3667static inline int skb_add_data(struct sk_buff *skb, 3668 struct iov_iter *from, int copy) 3669{ 3670 const int off = skb->len; 3671 3672 if (skb->ip_summed == CHECKSUM_NONE) { 3673 __wsum csum = 0; 3674 if (csum_and_copy_from_iter_full(skb_put(skb, copy), copy, 3675 &csum, from)) { 3676 skb->csum = csum_block_add(skb->csum, csum, off); 3677 return 0; 3678 } 3679 } else if (copy_from_iter_full(skb_put(skb, copy), copy, from)) 3680 return 0; 3681 3682 __skb_trim(skb, off); 3683 return -EFAULT; 3684} 3685 3686static inline bool skb_can_coalesce(struct sk_buff *skb, int i, 3687 const struct page *page, int off) 3688{ 3689 if (skb_zcopy(skb)) 3690 return false; 3691 if (i) { 3692 const skb_frag_t *frag = &skb_shinfo(skb)->frags[i - 1]; 3693 3694 return page == skb_frag_page(frag) && 3695 off == skb_frag_off(frag) + skb_frag_size(frag); 3696 } 3697 return false; 3698} 3699 3700static inline int __skb_linearize(struct sk_buff *skb) 3701{ 3702 return __pskb_pull_tail(skb, skb->data_len) ? 0 : -ENOMEM; 3703} 3704 3705/** 3706 * skb_linearize - convert paged skb to linear one 3707 * @skb: buffer to linarize 3708 * 3709 * If there is no free memory -ENOMEM is returned, otherwise zero 3710 * is returned and the old skb data released. 3711 */ 3712static inline int skb_linearize(struct sk_buff *skb) 3713{ 3714 return skb_is_nonlinear(skb) ? __skb_linearize(skb) : 0; 3715} 3716 3717/** 3718 * skb_has_shared_frag - can any frag be overwritten 3719 * @skb: buffer to test 3720 * 3721 * Return true if the skb has at least one frag that might be modified 3722 * by an external entity (as in vmsplice()/sendfile()) 3723 */ 3724static inline bool skb_has_shared_frag(const struct sk_buff *skb) 3725{ 3726 return skb_is_nonlinear(skb) && 3727 skb_shinfo(skb)->flags & SKBFL_SHARED_FRAG; 3728} 3729 3730/** 3731 * skb_linearize_cow - make sure skb is linear and writable 3732 * @skb: buffer to process 3733 * 3734 * If there is no free memory -ENOMEM is returned, otherwise zero 3735 * is returned and the old skb data released. 3736 */ 3737static inline int skb_linearize_cow(struct sk_buff *skb) 3738{ 3739 return skb_is_nonlinear(skb) || skb_cloned(skb) ? 3740 __skb_linearize(skb) : 0; 3741} 3742 3743static __always_inline void 3744__skb_postpull_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3745 unsigned int off) 3746{ 3747 if (skb->ip_summed == CHECKSUM_COMPLETE) 3748 skb->csum = csum_block_sub(skb->csum, 3749 csum_partial(start, len, 0), off); 3750 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3751 skb_checksum_start_offset(skb) < 0) 3752 skb->ip_summed = CHECKSUM_NONE; 3753} 3754 3755/** 3756 * skb_postpull_rcsum - update checksum for received skb after pull 3757 * @skb: buffer to update 3758 * @start: start of data before pull 3759 * @len: length of data pulled 3760 * 3761 * After doing a pull on a received packet, you need to call this to 3762 * update the CHECKSUM_COMPLETE checksum, or set ip_summed to 3763 * CHECKSUM_NONE so that it can be recomputed from scratch. 3764 */ 3765static inline void skb_postpull_rcsum(struct sk_buff *skb, 3766 const void *start, unsigned int len) 3767{ 3768 if (skb->ip_summed == CHECKSUM_COMPLETE) 3769 skb->csum = wsum_negate(csum_partial(start, len, 3770 wsum_negate(skb->csum))); 3771 else if (skb->ip_summed == CHECKSUM_PARTIAL && 3772 skb_checksum_start_offset(skb) < 0) 3773 skb->ip_summed = CHECKSUM_NONE; 3774} 3775 3776static __always_inline void 3777__skb_postpush_rcsum(struct sk_buff *skb, const void *start, unsigned int len, 3778 unsigned int off) 3779{ 3780 if (skb->ip_summed == CHECKSUM_COMPLETE) 3781 skb->csum = csum_block_add(skb->csum, 3782 csum_partial(start, len, 0), off); 3783} 3784 3785/** 3786 * skb_postpush_rcsum - update checksum for received skb after push 3787 * @skb: buffer to update 3788 * @start: start of data after push 3789 * @len: length of data pushed 3790 * 3791 * After doing a push on a received packet, you need to call this to 3792 * update the CHECKSUM_COMPLETE checksum. 3793 */ 3794static inline void skb_postpush_rcsum(struct sk_buff *skb, 3795 const void *start, unsigned int len) 3796{ 3797 __skb_postpush_rcsum(skb, start, len, 0); 3798} 3799 3800void *skb_pull_rcsum(struct sk_buff *skb, unsigned int len); 3801 3802/** 3803 * skb_push_rcsum - push skb and update receive checksum 3804 * @skb: buffer to update 3805 * @len: length of data pulled 3806 * 3807 * This function performs an skb_push on the packet and updates 3808 * the CHECKSUM_COMPLETE checksum. It should be used on 3809 * receive path processing instead of skb_push unless you know 3810 * that the checksum difference is zero (e.g., a valid IP header) 3811 * or you are setting ip_summed to CHECKSUM_NONE. 3812 */ 3813static inline void *skb_push_rcsum(struct sk_buff *skb, unsigned int len) 3814{ 3815 skb_push(skb, len); 3816 skb_postpush_rcsum(skb, skb->data, len); 3817 return skb->data; 3818} 3819 3820int pskb_trim_rcsum_slow(struct sk_buff *skb, unsigned int len); 3821/** 3822 * pskb_trim_rcsum - trim received skb and update checksum 3823 * @skb: buffer to trim 3824 * @len: new length 3825 * 3826 * This is exactly the same as pskb_trim except that it ensures the 3827 * checksum of received packets are still valid after the operation. 3828 * It can change skb pointers. 3829 */ 3830 3831static inline int pskb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3832{ 3833 if (likely(len >= skb->len)) 3834 return 0; 3835 return pskb_trim_rcsum_slow(skb, len); 3836} 3837 3838static inline int __skb_trim_rcsum(struct sk_buff *skb, unsigned int len) 3839{ 3840 if (skb->ip_summed == CHECKSUM_COMPLETE) 3841 skb->ip_summed = CHECKSUM_NONE; 3842 __skb_trim(skb, len); 3843 return 0; 3844} 3845 3846static inline int __skb_grow_rcsum(struct sk_buff *skb, unsigned int len) 3847{ 3848 if (skb->ip_summed == CHECKSUM_COMPLETE) 3849 skb->ip_summed = CHECKSUM_NONE; 3850 return __skb_grow(skb, len); 3851} 3852 3853#define rb_to_skb(rb) rb_entry_safe(rb, struct sk_buff, rbnode) 3854#define skb_rb_first(root) rb_to_skb(rb_first(root)) 3855#define skb_rb_last(root) rb_to_skb(rb_last(root)) 3856#define skb_rb_next(skb) rb_to_skb(rb_next(&(skb)->rbnode)) 3857#define skb_rb_prev(skb) rb_to_skb(rb_prev(&(skb)->rbnode)) 3858 3859#define skb_queue_walk(queue, skb) \ 3860 for (skb = (queue)->next; \ 3861 skb != (struct sk_buff *)(queue); \ 3862 skb = skb->next) 3863 3864#define skb_queue_walk_safe(queue, skb, tmp) \ 3865 for (skb = (queue)->next, tmp = skb->next; \ 3866 skb != (struct sk_buff *)(queue); \ 3867 skb = tmp, tmp = skb->next) 3868 3869#define skb_queue_walk_from(queue, skb) \ 3870 for (; skb != (struct sk_buff *)(queue); \ 3871 skb = skb->next) 3872 3873#define skb_rbtree_walk(skb, root) \ 3874 for (skb = skb_rb_first(root); skb != NULL; \ 3875 skb = skb_rb_next(skb)) 3876 3877#define skb_rbtree_walk_from(skb) \ 3878 for (; skb != NULL; \ 3879 skb = skb_rb_next(skb)) 3880 3881#define skb_rbtree_walk_from_safe(skb, tmp) \ 3882 for (; tmp = skb ? skb_rb_next(skb) : NULL, (skb != NULL); \ 3883 skb = tmp) 3884 3885#define skb_queue_walk_from_safe(queue, skb, tmp) \ 3886 for (tmp = skb->next; \ 3887 skb != (struct sk_buff *)(queue); \ 3888 skb = tmp, tmp = skb->next) 3889 3890#define skb_queue_reverse_walk(queue, skb) \ 3891 for (skb = (queue)->prev; \ 3892 skb != (struct sk_buff *)(queue); \ 3893 skb = skb->prev) 3894 3895#define skb_queue_reverse_walk_safe(queue, skb, tmp) \ 3896 for (skb = (queue)->prev, tmp = skb->prev; \ 3897 skb != (struct sk_buff *)(queue); \ 3898 skb = tmp, tmp = skb->prev) 3899 3900#define skb_queue_reverse_walk_from_safe(queue, skb, tmp) \ 3901 for (tmp = skb->prev; \ 3902 skb != (struct sk_buff *)(queue); \ 3903 skb = tmp, tmp = skb->prev) 3904 3905static inline bool skb_has_frag_list(const struct sk_buff *skb) 3906{ 3907 return skb_shinfo(skb)->frag_list != NULL; 3908} 3909 3910static inline void skb_frag_list_init(struct sk_buff *skb) 3911{ 3912 skb_shinfo(skb)->frag_list = NULL; 3913} 3914 3915#define skb_walk_frags(skb, iter) \ 3916 for (iter = skb_shinfo(skb)->frag_list; iter; iter = iter->next) 3917 3918 3919int __skb_wait_for_more_packets(struct sock *sk, struct sk_buff_head *queue, 3920 int *err, long *timeo_p, 3921 const struct sk_buff *skb); 3922struct sk_buff *__skb_try_recv_from_queue(struct sock *sk, 3923 struct sk_buff_head *queue, 3924 unsigned int flags, 3925 int *off, int *err, 3926 struct sk_buff **last); 3927struct sk_buff *__skb_try_recv_datagram(struct sock *sk, 3928 struct sk_buff_head *queue, 3929 unsigned int flags, int *off, int *err, 3930 struct sk_buff **last); 3931struct sk_buff *__skb_recv_datagram(struct sock *sk, 3932 struct sk_buff_head *sk_queue, 3933 unsigned int flags, int *off, int *err); 3934struct sk_buff *skb_recv_datagram(struct sock *sk, unsigned int flags, int *err); 3935__poll_t datagram_poll(struct file *file, struct socket *sock, 3936 struct poll_table_struct *wait); 3937int skb_copy_datagram_iter(const struct sk_buff *from, int offset, 3938 struct iov_iter *to, int size); 3939static inline int skb_copy_datagram_msg(const struct sk_buff *from, int offset, 3940 struct msghdr *msg, int size) 3941{ 3942 return skb_copy_datagram_iter(from, offset, &msg->msg_iter, size); 3943} 3944int skb_copy_and_csum_datagram_msg(struct sk_buff *skb, int hlen, 3945 struct msghdr *msg); 3946int skb_copy_and_hash_datagram_iter(const struct sk_buff *skb, int offset, 3947 struct iov_iter *to, int len, 3948 struct ahash_request *hash); 3949int skb_copy_datagram_from_iter(struct sk_buff *skb, int offset, 3950 struct iov_iter *from, int len); 3951int zerocopy_sg_from_iter(struct sk_buff *skb, struct iov_iter *frm); 3952void skb_free_datagram(struct sock *sk, struct sk_buff *skb); 3953void __skb_free_datagram_locked(struct sock *sk, struct sk_buff *skb, int len); 3954static inline void skb_free_datagram_locked(struct sock *sk, 3955 struct sk_buff *skb) 3956{ 3957 __skb_free_datagram_locked(sk, skb, 0); 3958} 3959int skb_kill_datagram(struct sock *sk, struct sk_buff *skb, unsigned int flags); 3960int skb_copy_bits(const struct sk_buff *skb, int offset, void *to, int len); 3961int skb_store_bits(struct sk_buff *skb, int offset, const void *from, int len); 3962__wsum skb_copy_and_csum_bits(const struct sk_buff *skb, int offset, u8 *to, 3963 int len); 3964int skb_splice_bits(struct sk_buff *skb, struct sock *sk, unsigned int offset, 3965 struct pipe_inode_info *pipe, unsigned int len, 3966 unsigned int flags); 3967int skb_send_sock_locked(struct sock *sk, struct sk_buff *skb, int offset, 3968 int len); 3969int skb_send_sock(struct sock *sk, struct sk_buff *skb, int offset, int len); 3970void skb_copy_and_csum_dev(const struct sk_buff *skb, u8 *to); 3971unsigned int skb_zerocopy_headlen(const struct sk_buff *from); 3972int skb_zerocopy(struct sk_buff *to, struct sk_buff *from, 3973 int len, int hlen); 3974void skb_split(struct sk_buff *skb, struct sk_buff *skb1, const u32 len); 3975int skb_shift(struct sk_buff *tgt, struct sk_buff *skb, int shiftlen); 3976void skb_scrub_packet(struct sk_buff *skb, bool xnet); 3977struct sk_buff *skb_segment(struct sk_buff *skb, netdev_features_t features); 3978struct sk_buff *skb_segment_list(struct sk_buff *skb, netdev_features_t features, 3979 unsigned int offset); 3980struct sk_buff *skb_vlan_untag(struct sk_buff *skb); 3981int skb_ensure_writable(struct sk_buff *skb, unsigned int write_len); 3982int __skb_vlan_pop(struct sk_buff *skb, u16 *vlan_tci); 3983int skb_vlan_pop(struct sk_buff *skb); 3984int skb_vlan_push(struct sk_buff *skb, __be16 vlan_proto, u16 vlan_tci); 3985int skb_eth_pop(struct sk_buff *skb); 3986int skb_eth_push(struct sk_buff *skb, const unsigned char *dst, 3987 const unsigned char *src); 3988int skb_mpls_push(struct sk_buff *skb, __be32 mpls_lse, __be16 mpls_proto, 3989 int mac_len, bool ethernet); 3990int skb_mpls_pop(struct sk_buff *skb, __be16 next_proto, int mac_len, 3991 bool ethernet); 3992int skb_mpls_update_lse(struct sk_buff *skb, __be32 mpls_lse); 3993int skb_mpls_dec_ttl(struct sk_buff *skb); 3994struct sk_buff *pskb_extract(struct sk_buff *skb, int off, int to_copy, 3995 gfp_t gfp); 3996 3997static inline int memcpy_from_msg(void *data, struct msghdr *msg, int len) 3998{ 3999 return copy_from_iter_full(data, len, &msg->msg_iter) ? 0 : -EFAULT; 4000} 4001 4002static inline int memcpy_to_msg(struct msghdr *msg, void *data, int len) 4003{ 4004 return copy_to_iter(data, len, &msg->msg_iter) == len ? 0 : -EFAULT; 4005} 4006 4007struct skb_checksum_ops { 4008 __wsum (*update)(const void *mem, int len, __wsum wsum); 4009 __wsum (*combine)(__wsum csum, __wsum csum2, int offset, int len); 4010}; 4011 4012extern const struct skb_checksum_ops *crc32c_csum_stub __read_mostly; 4013 4014__wsum __skb_checksum(const struct sk_buff *skb, int offset, int len, 4015 __wsum csum, const struct skb_checksum_ops *ops); 4016__wsum skb_checksum(const struct sk_buff *skb, int offset, int len, 4017 __wsum csum); 4018 4019static inline void * __must_check 4020__skb_header_pointer(const struct sk_buff *skb, int offset, int len, 4021 const void *data, int hlen, void *buffer) 4022{ 4023 if (likely(hlen - offset >= len)) 4024 return (void *)data + offset; 4025 4026 if (!skb || !buffer || unlikely(skb_copy_bits(skb, offset, buffer, len) < 0)) 4027 return NULL; 4028 4029 return buffer; 4030} 4031 4032static inline void * __must_check 4033skb_header_pointer(const struct sk_buff *skb, int offset, int len, void *buffer) 4034{ 4035 return __skb_header_pointer(skb, offset, len, skb->data, 4036 skb_headlen(skb), buffer); 4037} 4038 4039/** 4040 * skb_needs_linearize - check if we need to linearize a given skb 4041 * depending on the given device features. 4042 * @skb: socket buffer to check 4043 * @features: net device features 4044 * 4045 * Returns true if either: 4046 * 1. skb has frag_list and the device doesn't support FRAGLIST, or 4047 * 2. skb is fragmented and the device does not support SG. 4048 */ 4049static inline bool skb_needs_linearize(struct sk_buff *skb, 4050 netdev_features_t features) 4051{ 4052 return skb_is_nonlinear(skb) && 4053 ((skb_has_frag_list(skb) && !(features & NETIF_F_FRAGLIST)) || 4054 (skb_shinfo(skb)->nr_frags && !(features & NETIF_F_SG))); 4055} 4056 4057static inline void skb_copy_from_linear_data(const struct sk_buff *skb, 4058 void *to, 4059 const unsigned int len) 4060{ 4061 memcpy(to, skb->data, len); 4062} 4063 4064static inline void skb_copy_from_linear_data_offset(const struct sk_buff *skb, 4065 const int offset, void *to, 4066 const unsigned int len) 4067{ 4068 memcpy(to, skb->data + offset, len); 4069} 4070 4071static inline void skb_copy_to_linear_data(struct sk_buff *skb, 4072 const void *from, 4073 const unsigned int len) 4074{ 4075 memcpy(skb->data, from, len); 4076} 4077 4078static inline void skb_copy_to_linear_data_offset(struct sk_buff *skb, 4079 const int offset, 4080 const void *from, 4081 const unsigned int len) 4082{ 4083 memcpy(skb->data + offset, from, len); 4084} 4085 4086void skb_init(void); 4087 4088static inline ktime_t skb_get_ktime(const struct sk_buff *skb) 4089{ 4090 return skb->tstamp; 4091} 4092 4093/** 4094 * skb_get_timestamp - get timestamp from a skb 4095 * @skb: skb to get stamp from 4096 * @stamp: pointer to struct __kernel_old_timeval to store stamp in 4097 * 4098 * Timestamps are stored in the skb as offsets to a base timestamp. 4099 * This function converts the offset back to a struct timeval and stores 4100 * it in stamp. 4101 */ 4102static inline void skb_get_timestamp(const struct sk_buff *skb, 4103 struct __kernel_old_timeval *stamp) 4104{ 4105 *stamp = ns_to_kernel_old_timeval(skb->tstamp); 4106} 4107 4108static inline void skb_get_new_timestamp(const struct sk_buff *skb, 4109 struct __kernel_sock_timeval *stamp) 4110{ 4111 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4112 4113 stamp->tv_sec = ts.tv_sec; 4114 stamp->tv_usec = ts.tv_nsec / 1000; 4115} 4116 4117static inline void skb_get_timestampns(const struct sk_buff *skb, 4118 struct __kernel_old_timespec *stamp) 4119{ 4120 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4121 4122 stamp->tv_sec = ts.tv_sec; 4123 stamp->tv_nsec = ts.tv_nsec; 4124} 4125 4126static inline void skb_get_new_timestampns(const struct sk_buff *skb, 4127 struct __kernel_timespec *stamp) 4128{ 4129 struct timespec64 ts = ktime_to_timespec64(skb->tstamp); 4130 4131 stamp->tv_sec = ts.tv_sec; 4132 stamp->tv_nsec = ts.tv_nsec; 4133} 4134 4135static inline void __net_timestamp(struct sk_buff *skb) 4136{ 4137 skb->tstamp = ktime_get_real(); 4138 skb->mono_delivery_time = 0; 4139} 4140 4141static inline ktime_t net_timedelta(ktime_t t) 4142{ 4143 return ktime_sub(ktime_get_real(), t); 4144} 4145 4146static inline void skb_set_delivery_time(struct sk_buff *skb, ktime_t kt, 4147 bool mono) 4148{ 4149 skb->tstamp = kt; 4150 skb->mono_delivery_time = kt && mono; 4151} 4152 4153DECLARE_STATIC_KEY_FALSE(netstamp_needed_key); 4154 4155/* It is used in the ingress path to clear the delivery_time. 4156 * If needed, set the skb->tstamp to the (rcv) timestamp. 4157 */ 4158static inline void skb_clear_delivery_time(struct sk_buff *skb) 4159{ 4160 if (skb->mono_delivery_time) { 4161 skb->mono_delivery_time = 0; 4162 if (static_branch_unlikely(&netstamp_needed_key)) 4163 skb->tstamp = ktime_get_real(); 4164 else 4165 skb->tstamp = 0; 4166 } 4167} 4168 4169static inline void skb_clear_tstamp(struct sk_buff *skb) 4170{ 4171 if (skb->mono_delivery_time) 4172 return; 4173 4174 skb->tstamp = 0; 4175} 4176 4177static inline ktime_t skb_tstamp(const struct sk_buff *skb) 4178{ 4179 if (skb->mono_delivery_time) 4180 return 0; 4181 4182 return skb->tstamp; 4183} 4184 4185static inline ktime_t skb_tstamp_cond(const struct sk_buff *skb, bool cond) 4186{ 4187 if (!skb->mono_delivery_time && skb->tstamp) 4188 return skb->tstamp; 4189 4190 if (static_branch_unlikely(&netstamp_needed_key) || cond) 4191 return ktime_get_real(); 4192 4193 return 0; 4194} 4195 4196static inline u8 skb_metadata_len(const struct sk_buff *skb) 4197{ 4198 return skb_shinfo(skb)->meta_len; 4199} 4200 4201static inline void *skb_metadata_end(const struct sk_buff *skb) 4202{ 4203 return skb_mac_header(skb); 4204} 4205 4206static inline bool __skb_metadata_differs(const struct sk_buff *skb_a, 4207 const struct sk_buff *skb_b, 4208 u8 meta_len) 4209{ 4210 const void *a = skb_metadata_end(skb_a); 4211 const void *b = skb_metadata_end(skb_b); 4212 /* Using more efficient varaiant than plain call to memcmp(). */ 4213#if defined(CONFIG_HAVE_EFFICIENT_UNALIGNED_ACCESS) && BITS_PER_LONG == 64 4214 u64 diffs = 0; 4215 4216 switch (meta_len) { 4217#define __it(x, op) (x -= sizeof(u##op)) 4218#define __it_diff(a, b, op) (*(u##op *)__it(a, op)) ^ (*(u##op *)__it(b, op)) 4219 case 32: diffs |= __it_diff(a, b, 64); 4220 fallthrough; 4221 case 24: diffs |= __it_diff(a, b, 64); 4222 fallthrough; 4223 case 16: diffs |= __it_diff(a, b, 64); 4224 fallthrough; 4225 case 8: diffs |= __it_diff(a, b, 64); 4226 break; 4227 case 28: diffs |= __it_diff(a, b, 64); 4228 fallthrough; 4229 case 20: diffs |= __it_diff(a, b, 64); 4230 fallthrough; 4231 case 12: diffs |= __it_diff(a, b, 64); 4232 fallthrough; 4233 case 4: diffs |= __it_diff(a, b, 32); 4234 break; 4235 } 4236 return diffs; 4237#else 4238 return memcmp(a - meta_len, b - meta_len, meta_len); 4239#endif 4240} 4241 4242static inline bool skb_metadata_differs(const struct sk_buff *skb_a, 4243 const struct sk_buff *skb_b) 4244{ 4245 u8 len_a = skb_metadata_len(skb_a); 4246 u8 len_b = skb_metadata_len(skb_b); 4247 4248 if (!(len_a | len_b)) 4249 return false; 4250 4251 return len_a != len_b ? 4252 true : __skb_metadata_differs(skb_a, skb_b, len_a); 4253} 4254 4255static inline void skb_metadata_set(struct sk_buff *skb, u8 meta_len) 4256{ 4257 skb_shinfo(skb)->meta_len = meta_len; 4258} 4259 4260static inline void skb_metadata_clear(struct sk_buff *skb) 4261{ 4262 skb_metadata_set(skb, 0); 4263} 4264 4265struct sk_buff *skb_clone_sk(struct sk_buff *skb); 4266 4267#ifdef CONFIG_NETWORK_PHY_TIMESTAMPING 4268 4269void skb_clone_tx_timestamp(struct sk_buff *skb); 4270bool skb_defer_rx_timestamp(struct sk_buff *skb); 4271 4272#else /* CONFIG_NETWORK_PHY_TIMESTAMPING */ 4273 4274static inline void skb_clone_tx_timestamp(struct sk_buff *skb) 4275{ 4276} 4277 4278static inline bool skb_defer_rx_timestamp(struct sk_buff *skb) 4279{ 4280 return false; 4281} 4282 4283#endif /* !CONFIG_NETWORK_PHY_TIMESTAMPING */ 4284 4285/** 4286 * skb_complete_tx_timestamp() - deliver cloned skb with tx timestamps 4287 * 4288 * PHY drivers may accept clones of transmitted packets for 4289 * timestamping via their phy_driver.txtstamp method. These drivers 4290 * must call this function to return the skb back to the stack with a 4291 * timestamp. 4292 * 4293 * @skb: clone of the original outgoing packet 4294 * @hwtstamps: hardware time stamps 4295 * 4296 */ 4297void skb_complete_tx_timestamp(struct sk_buff *skb, 4298 struct skb_shared_hwtstamps *hwtstamps); 4299 4300void __skb_tstamp_tx(struct sk_buff *orig_skb, const struct sk_buff *ack_skb, 4301 struct skb_shared_hwtstamps *hwtstamps, 4302 struct sock *sk, int tstype); 4303 4304/** 4305 * skb_tstamp_tx - queue clone of skb with send time stamps 4306 * @orig_skb: the original outgoing packet 4307 * @hwtstamps: hardware time stamps, may be NULL if not available 4308 * 4309 * If the skb has a socket associated, then this function clones the 4310 * skb (thus sharing the actual data and optional structures), stores 4311 * the optional hardware time stamping information (if non NULL) or 4312 * generates a software time stamp (otherwise), then queues the clone 4313 * to the error queue of the socket. Errors are silently ignored. 4314 */ 4315void skb_tstamp_tx(struct sk_buff *orig_skb, 4316 struct skb_shared_hwtstamps *hwtstamps); 4317 4318/** 4319 * skb_tx_timestamp() - Driver hook for transmit timestamping 4320 * 4321 * Ethernet MAC Drivers should call this function in their hard_xmit() 4322 * function immediately before giving the sk_buff to the MAC hardware. 4323 * 4324 * Specifically, one should make absolutely sure that this function is 4325 * called before TX completion of this packet can trigger. Otherwise 4326 * the packet could potentially already be freed. 4327 * 4328 * @skb: A socket buffer. 4329 */ 4330static inline void skb_tx_timestamp(struct sk_buff *skb) 4331{ 4332 skb_clone_tx_timestamp(skb); 4333 if (skb_shinfo(skb)->tx_flags & SKBTX_SW_TSTAMP) 4334 skb_tstamp_tx(skb, NULL); 4335} 4336 4337/** 4338 * skb_complete_wifi_ack - deliver skb with wifi status 4339 * 4340 * @skb: the original outgoing packet 4341 * @acked: ack status 4342 * 4343 */ 4344void skb_complete_wifi_ack(struct sk_buff *skb, bool acked); 4345 4346__sum16 __skb_checksum_complete_head(struct sk_buff *skb, int len); 4347__sum16 __skb_checksum_complete(struct sk_buff *skb); 4348 4349static inline int skb_csum_unnecessary(const struct sk_buff *skb) 4350{ 4351 return ((skb->ip_summed == CHECKSUM_UNNECESSARY) || 4352 skb->csum_valid || 4353 (skb->ip_summed == CHECKSUM_PARTIAL && 4354 skb_checksum_start_offset(skb) >= 0)); 4355} 4356 4357/** 4358 * skb_checksum_complete - Calculate checksum of an entire packet 4359 * @skb: packet to process 4360 * 4361 * This function calculates the checksum over the entire packet plus 4362 * the value of skb->csum. The latter can be used to supply the 4363 * checksum of a pseudo header as used by TCP/UDP. It returns the 4364 * checksum. 4365 * 4366 * For protocols that contain complete checksums such as ICMP/TCP/UDP, 4367 * this function can be used to verify that checksum on received 4368 * packets. In that case the function should return zero if the 4369 * checksum is correct. In particular, this function will return zero 4370 * if skb->ip_summed is CHECKSUM_UNNECESSARY which indicates that the 4371 * hardware has already verified the correctness of the checksum. 4372 */ 4373static inline __sum16 skb_checksum_complete(struct sk_buff *skb) 4374{ 4375 return skb_csum_unnecessary(skb) ? 4376 0 : __skb_checksum_complete(skb); 4377} 4378 4379static inline void __skb_decr_checksum_unnecessary(struct sk_buff *skb) 4380{ 4381 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4382 if (skb->csum_level == 0) 4383 skb->ip_summed = CHECKSUM_NONE; 4384 else 4385 skb->csum_level--; 4386 } 4387} 4388 4389static inline void __skb_incr_checksum_unnecessary(struct sk_buff *skb) 4390{ 4391 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4392 if (skb->csum_level < SKB_MAX_CSUM_LEVEL) 4393 skb->csum_level++; 4394 } else if (skb->ip_summed == CHECKSUM_NONE) { 4395 skb->ip_summed = CHECKSUM_UNNECESSARY; 4396 skb->csum_level = 0; 4397 } 4398} 4399 4400static inline void __skb_reset_checksum_unnecessary(struct sk_buff *skb) 4401{ 4402 if (skb->ip_summed == CHECKSUM_UNNECESSARY) { 4403 skb->ip_summed = CHECKSUM_NONE; 4404 skb->csum_level = 0; 4405 } 4406} 4407 4408/* Check if we need to perform checksum complete validation. 4409 * 4410 * Returns true if checksum complete is needed, false otherwise 4411 * (either checksum is unnecessary or zero checksum is allowed). 4412 */ 4413static inline bool __skb_checksum_validate_needed(struct sk_buff *skb, 4414 bool zero_okay, 4415 __sum16 check) 4416{ 4417 if (skb_csum_unnecessary(skb) || (zero_okay && !check)) { 4418 skb->csum_valid = 1; 4419 __skb_decr_checksum_unnecessary(skb); 4420 return false; 4421 } 4422 4423 return true; 4424} 4425 4426/* For small packets <= CHECKSUM_BREAK perform checksum complete directly 4427 * in checksum_init. 4428 */ 4429#define CHECKSUM_BREAK 76 4430 4431/* Unset checksum-complete 4432 * 4433 * Unset checksum complete can be done when packet is being modified 4434 * (uncompressed for instance) and checksum-complete value is 4435 * invalidated. 4436 */ 4437static inline void skb_checksum_complete_unset(struct sk_buff *skb) 4438{ 4439 if (skb->ip_summed == CHECKSUM_COMPLETE) 4440 skb->ip_summed = CHECKSUM_NONE; 4441} 4442 4443/* Validate (init) checksum based on checksum complete. 4444 * 4445 * Return values: 4446 * 0: checksum is validated or try to in skb_checksum_complete. In the latter 4447 * case the ip_summed will not be CHECKSUM_UNNECESSARY and the pseudo 4448 * checksum is stored in skb->csum for use in __skb_checksum_complete 4449 * non-zero: value of invalid checksum 4450 * 4451 */ 4452static inline __sum16 __skb_checksum_validate_complete(struct sk_buff *skb, 4453 bool complete, 4454 __wsum psum) 4455{ 4456 if (skb->ip_summed == CHECKSUM_COMPLETE) { 4457 if (!csum_fold(csum_add(psum, skb->csum))) { 4458 skb->csum_valid = 1; 4459 return 0; 4460 } 4461 } 4462 4463 skb->csum = psum; 4464 4465 if (complete || skb->len <= CHECKSUM_BREAK) { 4466 __sum16 csum; 4467 4468 csum = __skb_checksum_complete(skb); 4469 skb->csum_valid = !csum; 4470 return csum; 4471 } 4472 4473 return 0; 4474} 4475 4476static inline __wsum null_compute_pseudo(struct sk_buff *skb, int proto) 4477{ 4478 return 0; 4479} 4480 4481/* Perform checksum validate (init). Note that this is a macro since we only 4482 * want to calculate the pseudo header which is an input function if necessary. 4483 * First we try to validate without any computation (checksum unnecessary) and 4484 * then calculate based on checksum complete calling the function to compute 4485 * pseudo header. 4486 * 4487 * Return values: 4488 * 0: checksum is validated or try to in skb_checksum_complete 4489 * non-zero: value of invalid checksum 4490 */ 4491#define __skb_checksum_validate(skb, proto, complete, \ 4492 zero_okay, check, compute_pseudo) \ 4493({ \ 4494 __sum16 __ret = 0; \ 4495 skb->csum_valid = 0; \ 4496 if (__skb_checksum_validate_needed(skb, zero_okay, check)) \ 4497 __ret = __skb_checksum_validate_complete(skb, \ 4498 complete, compute_pseudo(skb, proto)); \ 4499 __ret; \ 4500}) 4501 4502#define skb_checksum_init(skb, proto, compute_pseudo) \ 4503 __skb_checksum_validate(skb, proto, false, false, 0, compute_pseudo) 4504 4505#define skb_checksum_init_zero_check(skb, proto, check, compute_pseudo) \ 4506 __skb_checksum_validate(skb, proto, false, true, check, compute_pseudo) 4507 4508#define skb_checksum_validate(skb, proto, compute_pseudo) \ 4509 __skb_checksum_validate(skb, proto, true, false, 0, compute_pseudo) 4510 4511#define skb_checksum_validate_zero_check(skb, proto, check, \ 4512 compute_pseudo) \ 4513 __skb_checksum_validate(skb, proto, true, true, check, compute_pseudo) 4514 4515#define skb_checksum_simple_validate(skb) \ 4516 __skb_checksum_validate(skb, 0, true, false, 0, null_compute_pseudo) 4517 4518static inline bool __skb_checksum_convert_check(struct sk_buff *skb) 4519{ 4520 return (skb->ip_summed == CHECKSUM_NONE && skb->csum_valid); 4521} 4522 4523static inline void __skb_checksum_convert(struct sk_buff *skb, __wsum pseudo) 4524{ 4525 skb->csum = ~pseudo; 4526 skb->ip_summed = CHECKSUM_COMPLETE; 4527} 4528 4529#define skb_checksum_try_convert(skb, proto, compute_pseudo) \ 4530do { \ 4531 if (__skb_checksum_convert_check(skb)) \ 4532 __skb_checksum_convert(skb, compute_pseudo(skb, proto)); \ 4533} while (0) 4534 4535static inline void skb_remcsum_adjust_partial(struct sk_buff *skb, void *ptr, 4536 u16 start, u16 offset) 4537{ 4538 skb->ip_summed = CHECKSUM_PARTIAL; 4539 skb->csum_start = ((unsigned char *)ptr + start) - skb->head; 4540 skb->csum_offset = offset - start; 4541} 4542 4543/* Update skbuf and packet to reflect the remote checksum offload operation. 4544 * When called, ptr indicates the starting point for skb->csum when 4545 * ip_summed is CHECKSUM_COMPLETE. If we need create checksum complete 4546 * here, skb_postpull_rcsum is done so skb->csum start is ptr. 4547 */ 4548static inline void skb_remcsum_process(struct sk_buff *skb, void *ptr, 4549 int start, int offset, bool nopartial) 4550{ 4551 __wsum delta; 4552 4553 if (!nopartial) { 4554 skb_remcsum_adjust_partial(skb, ptr, start, offset); 4555 return; 4556 } 4557 4558 if (unlikely(skb->ip_summed != CHECKSUM_COMPLETE)) { 4559 __skb_checksum_complete(skb); 4560 skb_postpull_rcsum(skb, skb->data, ptr - (void *)skb->data); 4561 } 4562 4563 delta = remcsum_adjust(ptr, skb->csum, start, offset); 4564 4565 /* Adjust skb->csum since we changed the packet */ 4566 skb->csum = csum_add(skb->csum, delta); 4567} 4568 4569static inline struct nf_conntrack *skb_nfct(const struct sk_buff *skb) 4570{ 4571#if IS_ENABLED(CONFIG_NF_CONNTRACK) 4572 return (void *)(skb->_nfct & NFCT_PTRMASK); 4573#else 4574 return NULL; 4575#endif 4576} 4577 4578static inline unsigned long skb_get_nfct(const struct sk_buff *skb) 4579{ 4580#if IS_ENABLED(CONFIG_NF_CONNTRACK) 4581 return skb->_nfct; 4582#else 4583 return 0UL; 4584#endif 4585} 4586 4587static inline void skb_set_nfct(struct sk_buff *skb, unsigned long nfct) 4588{ 4589#if IS_ENABLED(CONFIG_NF_CONNTRACK) 4590 skb->slow_gro |= !!nfct; 4591 skb->_nfct = nfct; 4592#endif 4593} 4594 4595#ifdef CONFIG_SKB_EXTENSIONS 4596enum skb_ext_id { 4597#if IS_ENABLED(CONFIG_BRIDGE_NETFILTER) 4598 SKB_EXT_BRIDGE_NF, 4599#endif 4600#ifdef CONFIG_XFRM 4601 SKB_EXT_SEC_PATH, 4602#endif 4603#if IS_ENABLED(CONFIG_NET_TC_SKB_EXT) 4604 TC_SKB_EXT, 4605#endif 4606#if IS_ENABLED(CONFIG_MPTCP) 4607 SKB_EXT_MPTCP, 4608#endif 4609#if IS_ENABLED(CONFIG_MCTP_FLOWS) 4610 SKB_EXT_MCTP, 4611#endif 4612 SKB_EXT_NUM, /* must be last */ 4613}; 4614 4615/** 4616 * struct skb_ext - sk_buff extensions 4617 * @refcnt: 1 on allocation, deallocated on 0 4618 * @offset: offset to add to @data to obtain extension address 4619 * @chunks: size currently allocated, stored in SKB_EXT_ALIGN_SHIFT units 4620 * @data: start of extension data, variable sized 4621 * 4622 * Note: offsets/lengths are stored in chunks of 8 bytes, this allows 4623 * to use 'u8' types while allowing up to 2kb worth of extension data. 4624 */ 4625struct skb_ext { 4626 refcount_t refcnt; 4627 u8 offset[SKB_EXT_NUM]; /* in chunks of 8 bytes */ 4628 u8 chunks; /* same */ 4629 char data[] __aligned(8); 4630}; 4631 4632struct skb_ext *__skb_ext_alloc(gfp_t flags); 4633void *__skb_ext_set(struct sk_buff *skb, enum skb_ext_id id, 4634 struct skb_ext *ext); 4635void *skb_ext_add(struct sk_buff *skb, enum skb_ext_id id); 4636void __skb_ext_del(struct sk_buff *skb, enum skb_ext_id id); 4637void __skb_ext_put(struct skb_ext *ext); 4638 4639static inline void skb_ext_put(struct sk_buff *skb) 4640{ 4641 if (skb->active_extensions) 4642 __skb_ext_put(skb->extensions); 4643} 4644 4645static inline void __skb_ext_copy(struct sk_buff *dst, 4646 const struct sk_buff *src) 4647{ 4648 dst->active_extensions = src->active_extensions; 4649 4650 if (src->active_extensions) { 4651 struct skb_ext *ext = src->extensions; 4652 4653 refcount_inc(&ext->refcnt); 4654 dst->extensions = ext; 4655 } 4656} 4657 4658static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *src) 4659{ 4660 skb_ext_put(dst); 4661 __skb_ext_copy(dst, src); 4662} 4663 4664static inline bool __skb_ext_exist(const struct skb_ext *ext, enum skb_ext_id i) 4665{ 4666 return !!ext->offset[i]; 4667} 4668 4669static inline bool skb_ext_exist(const struct sk_buff *skb, enum skb_ext_id id) 4670{ 4671 return skb->active_extensions & (1 << id); 4672} 4673 4674static inline void skb_ext_del(struct sk_buff *skb, enum skb_ext_id id) 4675{ 4676 if (skb_ext_exist(skb, id)) 4677 __skb_ext_del(skb, id); 4678} 4679 4680static inline void *skb_ext_find(const struct sk_buff *skb, enum skb_ext_id id) 4681{ 4682 if (skb_ext_exist(skb, id)) { 4683 struct skb_ext *ext = skb->extensions; 4684 4685 return (void *)ext + (ext->offset[id] << 3); 4686 } 4687 4688 return NULL; 4689} 4690 4691static inline void skb_ext_reset(struct sk_buff *skb) 4692{ 4693 if (unlikely(skb->active_extensions)) { 4694 __skb_ext_put(skb->extensions); 4695 skb->active_extensions = 0; 4696 } 4697} 4698 4699static inline bool skb_has_extensions(struct sk_buff *skb) 4700{ 4701 return unlikely(skb->active_extensions); 4702} 4703#else 4704static inline void skb_ext_put(struct sk_buff *skb) {} 4705static inline void skb_ext_reset(struct sk_buff *skb) {} 4706static inline void skb_ext_del(struct sk_buff *skb, int unused) {} 4707static inline void __skb_ext_copy(struct sk_buff *d, const struct sk_buff *s) {} 4708static inline void skb_ext_copy(struct sk_buff *dst, const struct sk_buff *s) {} 4709static inline bool skb_has_extensions(struct sk_buff *skb) { return false; } 4710#endif /* CONFIG_SKB_EXTENSIONS */ 4711 4712static inline void nf_reset_ct(struct sk_buff *skb) 4713{ 4714#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4715 nf_conntrack_put(skb_nfct(skb)); 4716 skb->_nfct = 0; 4717#endif 4718} 4719 4720static inline void nf_reset_trace(struct sk_buff *skb) 4721{ 4722#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4723 skb->nf_trace = 0; 4724#endif 4725} 4726 4727static inline void ipvs_reset(struct sk_buff *skb) 4728{ 4729#if IS_ENABLED(CONFIG_IP_VS) 4730 skb->ipvs_property = 0; 4731#endif 4732} 4733 4734/* Note: This doesn't put any conntrack info in dst. */ 4735static inline void __nf_copy(struct sk_buff *dst, const struct sk_buff *src, 4736 bool copy) 4737{ 4738#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4739 dst->_nfct = src->_nfct; 4740 nf_conntrack_get(skb_nfct(src)); 4741#endif 4742#if IS_ENABLED(CONFIG_NETFILTER_XT_TARGET_TRACE) || IS_ENABLED(CONFIG_NF_TABLES) 4743 if (copy) 4744 dst->nf_trace = src->nf_trace; 4745#endif 4746} 4747 4748static inline void nf_copy(struct sk_buff *dst, const struct sk_buff *src) 4749{ 4750#if defined(CONFIG_NF_CONNTRACK) || defined(CONFIG_NF_CONNTRACK_MODULE) 4751 nf_conntrack_put(skb_nfct(dst)); 4752#endif 4753 dst->slow_gro = src->slow_gro; 4754 __nf_copy(dst, src, true); 4755} 4756 4757#ifdef CONFIG_NETWORK_SECMARK 4758static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4759{ 4760 to->secmark = from->secmark; 4761} 4762 4763static inline void skb_init_secmark(struct sk_buff *skb) 4764{ 4765 skb->secmark = 0; 4766} 4767#else 4768static inline void skb_copy_secmark(struct sk_buff *to, const struct sk_buff *from) 4769{ } 4770 4771static inline void skb_init_secmark(struct sk_buff *skb) 4772{ } 4773#endif 4774 4775static inline int secpath_exists(const struct sk_buff *skb) 4776{ 4777#ifdef CONFIG_XFRM 4778 return skb_ext_exist(skb, SKB_EXT_SEC_PATH); 4779#else 4780 return 0; 4781#endif 4782} 4783 4784static inline bool skb_irq_freeable(const struct sk_buff *skb) 4785{ 4786 return !skb->destructor && 4787 !secpath_exists(skb) && 4788 !skb_nfct(skb) && 4789 !skb->_skb_refdst && 4790 !skb_has_frag_list(skb); 4791} 4792 4793static inline void skb_set_queue_mapping(struct sk_buff *skb, u16 queue_mapping) 4794{ 4795 skb->queue_mapping = queue_mapping; 4796} 4797 4798static inline u16 skb_get_queue_mapping(const struct sk_buff *skb) 4799{ 4800 return skb->queue_mapping; 4801} 4802 4803static inline void skb_copy_queue_mapping(struct sk_buff *to, const struct sk_buff *from) 4804{ 4805 to->queue_mapping = from->queue_mapping; 4806} 4807 4808static inline void skb_record_rx_queue(struct sk_buff *skb, u16 rx_queue) 4809{ 4810 skb->queue_mapping = rx_queue + 1; 4811} 4812 4813static inline u16 skb_get_rx_queue(const struct sk_buff *skb) 4814{ 4815 return skb->queue_mapping - 1; 4816} 4817 4818static inline bool skb_rx_queue_recorded(const struct sk_buff *skb) 4819{ 4820 return skb->queue_mapping != 0; 4821} 4822 4823static inline void skb_set_dst_pending_confirm(struct sk_buff *skb, u32 val) 4824{ 4825 skb->dst_pending_confirm = val; 4826} 4827 4828static inline bool skb_get_dst_pending_confirm(const struct sk_buff *skb) 4829{ 4830 return skb->dst_pending_confirm != 0; 4831} 4832 4833static inline struct sec_path *skb_sec_path(const struct sk_buff *skb) 4834{ 4835#ifdef CONFIG_XFRM 4836 return skb_ext_find(skb, SKB_EXT_SEC_PATH); 4837#else 4838 return NULL; 4839#endif 4840} 4841 4842static inline bool skb_is_gso(const struct sk_buff *skb) 4843{ 4844 return skb_shinfo(skb)->gso_size; 4845} 4846 4847/* Note: Should be called only if skb_is_gso(skb) is true */ 4848static inline bool skb_is_gso_v6(const struct sk_buff *skb) 4849{ 4850 return skb_shinfo(skb)->gso_type & SKB_GSO_TCPV6; 4851} 4852 4853/* Note: Should be called only if skb_is_gso(skb) is true */ 4854static inline bool skb_is_gso_sctp(const struct sk_buff *skb) 4855{ 4856 return skb_shinfo(skb)->gso_type & SKB_GSO_SCTP; 4857} 4858 4859/* Note: Should be called only if skb_is_gso(skb) is true */ 4860static inline bool skb_is_gso_tcp(const struct sk_buff *skb) 4861{ 4862 return skb_shinfo(skb)->gso_type & (SKB_GSO_TCPV4 | SKB_GSO_TCPV6); 4863} 4864 4865static inline void skb_gso_reset(struct sk_buff *skb) 4866{ 4867 skb_shinfo(skb)->gso_size = 0; 4868 skb_shinfo(skb)->gso_segs = 0; 4869 skb_shinfo(skb)->gso_type = 0; 4870} 4871 4872static inline void skb_increase_gso_size(struct skb_shared_info *shinfo, 4873 u16 increment) 4874{ 4875 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4876 return; 4877 shinfo->gso_size += increment; 4878} 4879 4880static inline void skb_decrease_gso_size(struct skb_shared_info *shinfo, 4881 u16 decrement) 4882{ 4883 if (WARN_ON_ONCE(shinfo->gso_size == GSO_BY_FRAGS)) 4884 return; 4885 shinfo->gso_size -= decrement; 4886} 4887 4888void __skb_warn_lro_forwarding(const struct sk_buff *skb); 4889 4890static inline bool skb_warn_if_lro(const struct sk_buff *skb) 4891{ 4892 /* LRO sets gso_size but not gso_type, whereas if GSO is really 4893 * wanted then gso_type will be set. */ 4894 const struct skb_shared_info *shinfo = skb_shinfo(skb); 4895 4896 if (skb_is_nonlinear(skb) && shinfo->gso_size != 0 && 4897 unlikely(shinfo->gso_type == 0)) { 4898 __skb_warn_lro_forwarding(skb); 4899 return true; 4900 } 4901 return false; 4902} 4903 4904static inline void skb_forward_csum(struct sk_buff *skb) 4905{ 4906 /* Unfortunately we don't support this one. Any brave souls? */ 4907 if (skb->ip_summed == CHECKSUM_COMPLETE) 4908 skb->ip_summed = CHECKSUM_NONE; 4909} 4910 4911/** 4912 * skb_checksum_none_assert - make sure skb ip_summed is CHECKSUM_NONE 4913 * @skb: skb to check 4914 * 4915 * fresh skbs have their ip_summed set to CHECKSUM_NONE. 4916 * Instead of forcing ip_summed to CHECKSUM_NONE, we can 4917 * use this helper, to document places where we make this assertion. 4918 */ 4919static inline void skb_checksum_none_assert(const struct sk_buff *skb) 4920{ 4921 DEBUG_NET_WARN_ON_ONCE(skb->ip_summed != CHECKSUM_NONE); 4922} 4923 4924bool skb_partial_csum_set(struct sk_buff *skb, u16 start, u16 off); 4925 4926int skb_checksum_setup(struct sk_buff *skb, bool recalculate); 4927struct sk_buff *skb_checksum_trimmed(struct sk_buff *skb, 4928 unsigned int transport_len, 4929 __sum16(*skb_chkf)(struct sk_buff *skb)); 4930 4931/** 4932 * skb_head_is_locked - Determine if the skb->head is locked down 4933 * @skb: skb to check 4934 * 4935 * The head on skbs build around a head frag can be removed if they are 4936 * not cloned. This function returns true if the skb head is locked down 4937 * due to either being allocated via kmalloc, or by being a clone with 4938 * multiple references to the head. 4939 */ 4940static inline bool skb_head_is_locked(const struct sk_buff *skb) 4941{ 4942 return !skb->head_frag || skb_cloned(skb); 4943} 4944 4945/* Local Checksum Offload. 4946 * Compute outer checksum based on the assumption that the 4947 * inner checksum will be offloaded later. 4948 * See Documentation/networking/checksum-offloads.rst for 4949 * explanation of how this works. 4950 * Fill in outer checksum adjustment (e.g. with sum of outer 4951 * pseudo-header) before calling. 4952 * Also ensure that inner checksum is in linear data area. 4953 */ 4954static inline __wsum lco_csum(struct sk_buff *skb) 4955{ 4956 unsigned char *csum_start = skb_checksum_start(skb); 4957 unsigned char *l4_hdr = skb_transport_header(skb); 4958 __wsum partial; 4959 4960 /* Start with complement of inner checksum adjustment */ 4961 partial = ~csum_unfold(*(__force __sum16 *)(csum_start + 4962 skb->csum_offset)); 4963 4964 /* Add in checksum of our headers (incl. outer checksum 4965 * adjustment filled in by caller) and return result. 4966 */ 4967 return csum_partial(l4_hdr, csum_start - l4_hdr, partial); 4968} 4969 4970static inline bool skb_is_redirected(const struct sk_buff *skb) 4971{ 4972 return skb->redirected; 4973} 4974 4975static inline void skb_set_redirected(struct sk_buff *skb, bool from_ingress) 4976{ 4977 skb->redirected = 1; 4978#ifdef CONFIG_NET_REDIRECT 4979 skb->from_ingress = from_ingress; 4980 if (skb->from_ingress) 4981 skb_clear_tstamp(skb); 4982#endif 4983} 4984 4985static inline void skb_reset_redirect(struct sk_buff *skb) 4986{ 4987 skb->redirected = 0; 4988} 4989 4990static inline void skb_set_redirected_noclear(struct sk_buff *skb, 4991 bool from_ingress) 4992{ 4993 skb->redirected = 1; 4994#ifdef CONFIG_NET_REDIRECT 4995 skb->from_ingress = from_ingress; 4996#endif 4997} 4998 4999static inline bool skb_csum_is_sctp(struct sk_buff *skb) 5000{ 5001#if IS_ENABLED(CONFIG_IP_SCTP) 5002 return skb->csum_not_inet; 5003#else 5004 return 0; 5005#endif 5006} 5007 5008static inline void skb_reset_csum_not_inet(struct sk_buff *skb) 5009{ 5010 skb->ip_summed = CHECKSUM_NONE; 5011#if IS_ENABLED(CONFIG_IP_SCTP) 5012 skb->csum_not_inet = 0; 5013#endif 5014} 5015 5016static inline void skb_set_kcov_handle(struct sk_buff *skb, 5017 const u64 kcov_handle) 5018{ 5019#ifdef CONFIG_KCOV 5020 skb->kcov_handle = kcov_handle; 5021#endif 5022} 5023 5024static inline u64 skb_get_kcov_handle(struct sk_buff *skb) 5025{ 5026#ifdef CONFIG_KCOV 5027 return skb->kcov_handle; 5028#else 5029 return 0; 5030#endif 5031} 5032 5033static inline void skb_mark_for_recycle(struct sk_buff *skb) 5034{ 5035#ifdef CONFIG_PAGE_POOL 5036 skb->pp_recycle = 1; 5037#endif 5038} 5039 5040ssize_t skb_splice_from_iter(struct sk_buff *skb, struct iov_iter *iter, 5041 ssize_t maxsize, gfp_t gfp); 5042 5043#endif /* __KERNEL__ */ 5044#endif /* _LINUX_SKBUFF_H */